restructure play-adpcm example code, stream-wav can now play stereo adpcm wavs

correct case of several other cx16 kernal calls.

corrected to upper case: cx16 kernal calls CLOSE_ALL, LKUPLA, LKUPSA, JSRFAR, PRIMM, MACPTR.

.github/workflows/all-ci.yml

@@ -10,13 +10,18 @@ jobs:
 
     steps:
       - name: Checkout code
-        uses: actions/checkout@v2
+        uses: actions/checkout@v3
 
-      - name: Install 64tass
-        run: sudo apt-get update -y && sudo apt-get install -y 64tass
+      - name: build and install recent 64tass
+        run: |
+          sudo apt-get install -y make build-essential
+          git clone --depth=1 https://github.com/irmen/64tass
+          cd 64tass
+          make -j4
+          sudo make install
 
       - name: Set up JDK 11
-        uses: actions/setup-java@v2
+        uses: actions/setup-java@v3
         with:
           java-version: 11
           distribution: adopt

.gitignore

@@ -15,6 +15,7 @@ out/
 parser/**/*.interp
 parser/**/*.tokens
 parser/**/*.java
+compiler/src/prog8/buildversion/*
 *.py[cod]
 *.egg
 *.egg-info
@@ -29,6 +30,8 @@ parsetab.py
 compiler/lib/
 
 .gradle
+**/BuildVersion.kt
 /prog8compiler.jar
 sd*.img
 *.d64
+

.idea/kotlinc.xml

@@ -0,0 +1,9 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<project version="4">
+  <component name="Kotlin2JvmCompilerArguments">
+    <option name="jvmTarget" value="11" />
+  </component>
+  <component name="KotlinJpsPluginSettings">
+    <option name="version" value="1.9.0-release-358" />
+  </component>
+</project>

.idea/libraries/antlr_antlr4.xml

@@ -1,17 +1,16 @@
 <component name="libraryTable">
   <library name="antlr.antlr4" type="repository">
-    <properties maven-id="org.antlr:antlr4:4.11.1">
+    <properties maven-id="org.antlr:antlr4:4.13.1">
       <exclude>
         <dependency maven-id="com.ibm.icu:icu4j" />
       </exclude>
     </properties>
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4/4.11.1/antlr4-4.11.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4-runtime/4.11.1/antlr4-runtime-4.11.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4/4.13.1/antlr4-4.13.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr4-runtime/4.13.1/antlr4-runtime-4.13.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/antlr/antlr-runtime/3.5.3/antlr-runtime-3.5.3.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/antlr/ST4/4.3.4/ST4-4.3.4.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/abego/treelayout/org.abego.treelayout.core/1.0.3/org.abego.treelayout.core-1.0.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/glassfish/javax.json/1.1.4/javax.json-1.1.4.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/github_hypfvieh_dbus_java.xml

@@ -1,23 +1,23 @@
 <component name="libraryTable">
   <library name="github.hypfvieh.dbus.java" type="repository">
-    <properties maven-id="com.github.hypfvieh:dbus-java:3.3.1" />
+    <properties maven-id="com.github.hypfvieh:dbus-java:3.3.2" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/hypfvieh/dbus-java/3.3.1/dbus-java-3.3.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-unixsocket/0.38.6/jnr-unixsocket-0.38.6.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-ffi/2.2.2/jnr-ffi-2.2.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jffi/1.3.1/jffi-1.3.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jffi/1.3.1/jffi-1.3.1-native.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm/9.1/asm-9.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-commons/9.1/asm-commons-9.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-analysis/9.1/asm-analysis-9.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-tree/9.1/asm-tree-9.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-util/9.1/asm-util-9.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/hypfvieh/dbus-java/3.3.2/dbus-java-3.3.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-unixsocket/0.38.17/jnr-unixsocket-0.38.17.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-ffi/2.2.11/jnr-ffi-2.2.11.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jffi/1.3.9/jffi-1.3.9.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jffi/1.3.9/jffi-1.3.9-native.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm/9.2/asm-9.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-commons/9.2/asm-commons-9.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-analysis/9.2/asm-analysis-9.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-tree/9.2/asm-tree-9.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/ow2/asm/asm-util/9.2/asm-util-9.2.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-a64asm/1.0.0/jnr-a64asm-1.0.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-x86asm/1.0.2/jnr-x86asm-1.0.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-constants/0.10.1/jnr-constants-0.10.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-enxio/0.32.4/jnr-enxio-0.32.4.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-posix/3.1.5/jnr-posix-3.1.5.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-api/1.7.30/slf4j-api-1.7.30.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-constants/0.10.3/jnr-constants-0.10.3.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-enxio/0.32.13/jnr-enxio-0.32.13.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/github/jnr/jnr-posix/3.1.15/jnr-posix-3.1.15.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-api/1.7.36/slf4j-api-1.7.36.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/io_kotest_assertions_core_jvm.xml

@@ -1,21 +1,21 @@
 <component name="libraryTable">
   <library name="io.kotest.assertions.core.jvm" type="repository">
-    <properties maven-id="io.kotest:kotest-assertions-core-jvm:5.3.2" />
+    <properties maven-id="io.kotest:kotest-assertions-core-jvm:5.6.2" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.3.2/kotest-assertions-core-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.6.21/kotlin-stdlib-jdk8-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.6.21/kotlin-stdlib-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.6.21/kotlin-stdlib-jdk7-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.3.2/kotest-assertions-shared-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.11/java-diff-utils-4.11.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.6.2/kotest-assertions-core-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.8.10/kotlin-stdlib-jdk8-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.8.10/kotlin-stdlib-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.8.10/kotlin-stdlib-jdk7-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.6.2/kotest-assertions-shared-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.12/java-diff-utils-4.12.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/opentest4j/opentest4j/1.2.0/opentest4j-1.2.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.6.1/kotlinx-coroutines-jdk8-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.6.21/kotlin-reflect-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.6.1/kotlinx-coroutines-core-jvm-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.6.0/kotlin-stdlib-common-1.6.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.3.2/kotest-common-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.3.2/kotest-assertions-api-jvm-5.3.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.8.10/kotlin-stdlib-common-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.7.0/kotlinx-coroutines-jdk8-1.7.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.8.10/kotlin-reflect-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.6.2/kotest-common-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.6.2/kotest-assertions-api-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.7.0/kotlinx-coroutines-core-jvm-1.7.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/23.0.0/annotations-23.0.0.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/io_kotest_property_jvm.xml

@@ -1,24 +0,0 @@
-<component name="libraryTable">
-  <library name="io.kotest.property.jvm" type="repository">
-    <properties maven-id="io.kotest:kotest-property-jvm:5.3.2" />
-    <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-property-jvm/5.3.2/kotest-property-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/github/curious-odd-man/rgxgen/1.3/rgxgen-1.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.6.21/kotlin-stdlib-jdk8-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.6.21/kotlin-stdlib-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.6.21/kotlin-stdlib-jdk7-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.3.2/kotest-common-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.3.2/kotest-assertions-shared-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.3.2/kotest-assertions-api-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.6.1/kotlinx-coroutines-jdk8-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/opentest4j/opentest4j/1.2.0/opentest4j-1.2.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.11/java-diff-utils-4.11.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.6.21/kotlin-reflect-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.6.1/kotlinx-coroutines-core-jvm-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.6.0/kotlin-stdlib-common-1.6.0.jar!/" />
-    </CLASSES>
-    <JAVADOC />
-    <SOURCES />
-  </library>
-</component>

.idea/libraries/io_kotest_runner_junit5_jvm.xml

@@ -1,51 +1,42 @@
 <component name="libraryTable">
   <library name="io.kotest.runner.junit5.jvm" type="repository">
-    <properties maven-id="io.kotest:kotest-runner-junit5-jvm:5.3.2" />
+    <properties maven-id="io.kotest:kotest-runner-junit5-jvm:5.6.2" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-runner-junit5-jvm/5.3.2/kotest-runner-junit5-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-api-jvm/5.3.2/kotest-framework-api-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.3.2/kotest-assertions-shared-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.11/java-diff-utils-4.11.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-test-jvm/1.6.1/kotlinx-coroutines-test-jvm-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.3.2/kotest-common-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-engine-jvm/5.3.2/kotest-framework-engine-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/github/classgraph/classgraph/4.8.146/classgraph-4.8.146.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-runner-junit5-jvm/5.6.2/kotest-runner-junit5-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-api-jvm/5.6.2/kotest-framework-api-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-shared-jvm/5.6.2/kotest-assertions-shared-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/github/java-diff-utils/java-diff-utils/4.12/java-diff-utils-4.12.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-test-jvm/1.7.0/kotlinx-coroutines-test-jvm-1.7.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-common-jvm/5.6.2/kotest-common-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-engine-jvm/5.6.2/kotest-framework-engine-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/github/classgraph/classgraph/4.8.157/classgraph-4.8.157.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/ajalt/mordant/1.2.1/mordant-1.2.1.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/com/github/ajalt/colormath/1.2.0/colormath-1.2.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/opentest4j/opentest4j/1.2.0/opentest4j-1.2.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-debug/1.6.1/kotlinx-coroutines-debug-1.6.1.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-debug/1.7.0/kotlinx-coroutines-debug-1.7.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/net/java/dev/jna/jna/5.9.0/jna-5.9.0.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/net/java/dev/jna/jna-platform/5.9.0/jna-platform-5.9.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-discovery-jvm/5.3.2/kotest-framework-discovery-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.3.2/kotest-assertions-core-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.6.1/kotlinx-coroutines-jdk8-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.3.2/kotest-assertions-api-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-extensions-jvm/5.3.2/kotest-extensions-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/commons-io/commons-io/2.11.0/commons-io-2.11.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk/1.12.3/mockk-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-common/1.12.3/mockk-common-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-dsl/1.12.3/mockk-dsl-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-dsl-jvm/1.12.3/mockk-dsl-jvm-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-agent-jvm/1.12.3/mockk-agent-jvm-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-agent-api/1.12.3/mockk-agent-api-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/mockk/mockk-agent-common/1.12.3/mockk-agent-common-1.12.3.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/objenesis/objenesis/3.1/objenesis-3.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy/1.12.6/byte-buddy-1.12.6.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy-agent/1.12.6/byte-buddy-agent-1.12.6.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-concurrency-jvm/5.3.2/kotest-framework-concurrency-jvm-5.3.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.6.1/kotlinx-coroutines-core-jvm-1.6.1.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.6.0/kotlin-stdlib-common-1.6.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-engine/1.7.2/junit-platform-engine-1.7.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/apiguardian/apiguardian-api/1.1.0/apiguardian-api-1.1.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-commons/1.7.2/junit-platform-commons-1.7.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-suite-api/1.7.2/junit-platform-suite-api-1.7.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-launcher/1.7.2/junit-platform-launcher-1.7.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/junit/jupiter/junit-jupiter-api/5.7.2/junit-jupiter-api-5.7.2.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.6.21/kotlin-stdlib-jdk8-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.6.21/kotlin-stdlib-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.6.21/kotlin-stdlib-jdk7-1.6.21.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.6.21/kotlin-reflect-1.6.21.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy/1.10.9/byte-buddy-1.10.9.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/net/bytebuddy/byte-buddy-agent/1.10.9/byte-buddy-agent-1.10.9.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-discovery-jvm/5.6.2/kotest-framework-discovery-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-core-jvm/5.6.2/kotest-assertions-core-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-jdk8/1.7.0/kotlinx-coroutines-jdk8-1.7.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-assertions-api-jvm/5.6.2/kotest-assertions-api-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-extensions-jvm/5.6.2/kotest-extensions-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/io/kotest/kotest-framework-concurrency-jvm/5.6.2/kotest-framework-concurrency-jvm-5.6.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-coroutines-core-jvm/1.7.0/kotlinx-coroutines-core-jvm-1.7.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/23.0.0/annotations-23.0.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-engine/1.8.2/junit-platform-engine-1.8.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-commons/1.8.2/junit-platform-commons-1.8.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/apiguardian/apiguardian-api/1.1.2/apiguardian-api-1.1.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-suite-api/1.8.2/junit-platform-suite-api-1.8.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/junit/platform/junit-platform-launcher/1.8.2/junit-platform-launcher-1.8.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/junit/jupiter/junit-jupiter-api/5.8.2/junit-jupiter-api-5.8.2.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.8.10/kotlin-stdlib-jdk8-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.8.10/kotlin-stdlib-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.8.10/kotlin-stdlib-jdk7-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.8.10/kotlin-stdlib-common-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-reflect/1.8.10/kotlin-reflect-1.8.10.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/jetbrains_kotlinx_cli_jvm.xml

@@ -1,8 +1,8 @@
 <component name="libraryTable">
   <library name="jetbrains.kotlinx.cli.jvm" type="repository">
-    <properties include-transitive-deps="false" maven-id="org.jetbrains.kotlinx:kotlinx-cli-jvm:0.3.4" />
+    <properties include-transitive-deps="false" maven-id="org.jetbrains.kotlinx:kotlinx-cli-jvm:0.3.6" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-cli-jvm/0.3.4/kotlinx-cli-jvm-0.3.4.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlinx/kotlinx-cli-jvm/0.3.6/kotlinx-cli-jvm-0.3.6.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/michael_bull_kotlin_result_jvm.xml

@@ -1,13 +1,13 @@
 <component name="libraryTable">
   <library name="michael.bull.kotlin.result.jvm" type="repository">
-    <properties maven-id="com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16" />
+    <properties maven-id="com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/com/michael-bull/kotlin-result/kotlin-result-jvm/1.1.16/kotlin-result-jvm-1.1.16.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.6.20/kotlin-stdlib-common-1.6.20.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.6.20/kotlin-stdlib-jdk8-1.6.20.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.6.20/kotlin-stdlib-1.6.20.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/michael-bull/kotlin-result/kotlin-result-jvm/1.1.18/kotlin-result-jvm-1.1.18.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk8/1.8.10/kotlin-stdlib-jdk8-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib/1.8.10/kotlin-stdlib-1.8.10.jar!/" />
       <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/annotations/13.0/annotations-13.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.6.20/kotlin-stdlib-jdk7-1.6.20.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-jdk7/1.8.10/kotlin-stdlib-jdk7-1.8.10.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/jetbrains/kotlin/kotlin-stdlib-common/1.8.10/kotlin-stdlib-common-1.8.10.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/slf4j_simple.xml

@@ -1,9 +1,9 @@
 <component name="libraryTable">
   <library name="slf4j.simple" type="repository">
-    <properties maven-id="org.slf4j:slf4j-simple:1.7.36" />
+    <properties maven-id="org.slf4j:slf4j-simple:2.0.9" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-simple/1.7.36/slf4j-simple-1.7.36.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-api/1.7.36/slf4j-api-1.7.36.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-simple/2.0.9/slf4j-simple-2.0.9.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-api/2.0.9/slf4j-api-2.0.9.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/libraries/takes.xml

@@ -1,15 +1,16 @@
 <component name="libraryTable">
   <library name="takes" type="repository">
-    <properties maven-id="org.takes:takes:1.20" />
+    <properties maven-id="org.takes:takes:1.24.4" />
     <CLASSES>
-      <root url="jar://$MAVEN_REPOSITORY$/org/takes/takes/1.20/takes-1.20.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/cactoos/cactoos/0.50/cactoos-0.50.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/javax/xml/bind/jaxb-api/2.3.0/jaxb-api-2.3.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/slf4j/slf4j-api/1.7.25/slf4j-api-1.7.25.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/apache/commons/commons-text/1.4/commons-text-1.4.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/org/apache/commons/commons-lang3/3.7/commons-lang3-3.7.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/sun/xml/bind/jaxb-core/2.3.0/jaxb-core-2.3.0.jar!/" />
-      <root url="jar://$MAVEN_REPOSITORY$/com/sun/xml/bind/jaxb-impl/2.3.0/jaxb-impl-2.3.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/takes/takes/1.24.4/takes-1.24.4.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/cactoos/cactoos/0.54.0/cactoos-0.54.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/javax/xml/bind/jaxb-api/2.4.0-b180830.0359/jaxb-api-2.4.0-b180830.0359.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/javax/activation/javax.activation-api/1.2.0/javax.activation-api-1.2.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/sun/xml/bind/jaxb-core/4.0.0/jaxb-core-4.0.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/jakarta/xml/bind/jakarta.xml.bind-api/4.0.0/jakarta.xml.bind-api-4.0.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/jakarta/activation/jakarta.activation-api/2.1.0/jakarta.activation-api-2.1.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/org/eclipse/angus/angus-activation/1.0.0/angus-activation-1.0.0.jar!/" />
+      <root url="jar://$MAVEN_REPOSITORY$/com/sun/xml/bind/jaxb-impl/4.0.0/jaxb-impl-4.0.0.jar!/" />
     </CLASSES>
     <JAVADOC />
     <SOURCES />

.idea/misc.xml

@@ -16,6 +16,9 @@
       </list>
     </option>
   </component>
+  <component name="Black">
+    <option name="sdkName" value="Python 3.11" />
+  </component>
   <component name="FrameworkDetectionExcludesConfiguration">
     <type id="Python" />
   </component>

.readthedocs.yaml

@@ -7,13 +7,9 @@ version: 2
 
 # Set the version of Python and other tools you might need
 build:
-  os: ubuntu-20.04
+  os: ubuntu-22.04
   tools:
-    python: "3.9"
-    # You can also specify other tool versions:
-    # nodejs: "16"
-    # rust: "1.55"
-    # golang: "1.17"
+    python: "3.11"
 
 # Build documentation in the docs/ directory with Sphinx
 sphinx:

README.md

@@ -14,6 +14,21 @@ Documentation
 Full documentation (syntax reference, how to use the language and the compiler, etc.) can be found at:
 https://prog8.readthedocs.io/
 
+How to get it/build it
+----------------------
+
+- Download the latest [official release](https://github.com/irmen/prog8/releases) from github.
+- Or, if you want/need a bleeding edge development version, you can:
+  - download a build artifact zipfile from a recent [github action build](https://github.com/irmen/prog8/actions).
+  - you can also compile it yourself from source. [Instructions here](https://prog8.readthedocs.io/en/latest/compiling.html).
+
+Community
+---------
+Most of the development on Prog8 and the use of it is currently centered around 
+the [Commander X16](https://www.commanderx16.com/) retro computer. Their [discord server](https://discord.gg/nS2PqEC) contains a small channel
+dedicated to Prog8. Other than that, use the issue tracker on github.
+
+
 Software license
 ----------------
 GNU GPL 3.0 (see file LICENSE), with exception for generated code:
@@ -59,6 +74,8 @@ What does Prog8 provide?
 - "c64": Commodore-64  (6502 like CPU)
 - "c128": Commodore-128  (6502 like CPU - the Z80 cpu mode is not supported)
 - "cx16": [CommanderX16](https://www.commanderx16.com)  (65c02 CPU)
+- "pet32": Commodore PET (experimental)
+- "atari": Atari 8 bit such as 800XL (experimental)
 - If you only use standard kernal and prog8 library routines, it is possible to compile the *exact same program* for different machines (just change the compiler target flag)
 
 
@@ -76,7 +93,9 @@ IntelliJ IDEA with the Kotlin plugin).
 
 It's handy to have an emulator (or a real machine perhaps!) to run the programs on. The compiler assumes the presence
 of the [Vice emulator](http://vice-emu.sourceforge.net/)  for the C64 target,
-and the [x16emu emulator](https://github.com/commanderx16/x16-emulator) for the CommanderX16 target.
+and a recent emulator version (R42 or newer) for the CommanderX16, such as [x16emu](https://cx16forum.com/forum/viewforum.php?f=30) 
+(preferred, this is the official emulator. If required, source code is [here](https://github.com/X16Community/x16-emulator/)).
+There is also [Box16](https://github.com/indigodarkwolf/box16) which has powerful debugging features.
 
 **Syntax highlighting:** for a few different editors, syntax highlighting definition files are provided.
 Look in the [syntax-files](https://github.com/irmen/prog8/tree/master/syntax-files) directory in the github repository to find them.

codeCore/build.gradle

@@ -26,16 +26,16 @@ compileTestKotlin {
 dependencies {
     // should have no dependencies to other modules
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
-    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16"
+    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18"
 }
 
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
 }

codeCore/src/prog8/code/SymbolTable.kt

@@ -15,14 +15,26 @@ class SymbolTable(astProgram: PtProgram) : StNode(astProgram.name, StNodeType.GL
      * This gives the fastest lookup possible (no need to traverse tree nodes)
      */
 
-    val flat: Map<String, StNode> by lazy {
+    private var cachedFlat: Map<String, StNode>? = null
+
+    val flat: Map<String, StNode> get()  {
+        if(cachedFlat!=null)
+            return cachedFlat!!
+
         val result = mutableMapOf<String, StNode>()
-        fun flatten(node: StNode) {
-            result[node.scopedName] = node
-            node.children.values.forEach { flatten(it) }
+        fun collect(node: StNode) {
+            for(child in node.children) {
+                result[child.value.scopedName] = child.value
+                collect(child.value)
+            }
         }
-        children.values.forEach { flatten(it) }
-        result
+        collect(this)
+        cachedFlat = result
+        return result
+    }
+
+    fun resetCachedFlat() {
+        cachedFlat = null
     }
 
     val allVariables: Collection<StStaticVariable> by lazy {
@@ -68,6 +80,16 @@ class SymbolTable(astProgram: PtProgram) : StNode(astProgram.name, StNodeType.GL
     }
 
     override fun lookup(scopedName: String) = flat[scopedName]
+
+    fun getLength(name: String): Int? {
+        val node = flat[name]
+        return when(node) {
+            is StMemVar -> node.length
+            is StMemorySlab -> node.size.toInt()
+            is StStaticVariable -> node.length
+            else -> null
+        }
+    }
 }
 
 
@@ -133,7 +155,7 @@ open class StNode(val name: String,
     }
 
     private val scopedNameList: List<String> by lazy {
-        if(type== StNodeType.GLOBAL)
+        if(type==StNodeType.GLOBAL)
             emptyList()
         else
             parent.scopedNameList + name
@@ -142,7 +164,7 @@ open class StNode(val name: String,
     private fun lookup(scopedName: List<String>): StNode? {
         // a scoped name refers to a name in another namespace, and always stars from the root.
         var node = this
-        while(node.type!= StNodeType.GLOBAL)
+        while(node.type!=StNodeType.GLOBAL)
             node = node.parent
 
         for(name in scopedName) {
@@ -174,13 +196,10 @@ class StStaticVariable(name: String,
         }
         if(onetimeInitializationNumericValue!=null) {
             require(dt in NumericDatatypes)
-            require(onetimeInitializationNumericValue!=0.0) { "zero as init value should just remain uninitialized"}
         }
         if(onetimeInitializationArrayValue!=null) {
             require(dt in ArrayDatatypes)
-            if(onetimeInitializationArrayValue.all { it.number!=null} ) {
-                require(onetimeInitializationArrayValue.any { it.number != 0.0 }) { "array of all zerors as init value should just remain uninitialized" }
-            }
+            require(length==onetimeInitializationArrayValue.size)
         }
         if(onetimeInitializationStringValue!=null) {
             require(dt == DataType.STR)
@@ -191,8 +210,7 @@ class StStaticVariable(name: String,
 
 
 class StConstant(name: String, val dt: DataType, val value: Double, astNode: PtNode) :
-    StNode(name, StNodeType.CONSTANT, astNode) {
-}
+    StNode(name, StNodeType.CONSTANT, astNode)
 
 
 class StMemVar(name: String,
@@ -201,6 +219,11 @@ class StMemVar(name: String,
                val length: Int?,             // for arrays: the number of elements, for strings: number of characters *including* the terminating 0-byte
                astNode: PtNode) :
     StNode(name, StNodeType.MEMVAR, astNode) {
+
+    init{
+        if(dt in ArrayDatatypes || dt == DataType.STR)
+            require(length!=null) { "memory mapped array or string must have known length" }
+    }
 }
 
 class StMemorySlab(
@@ -209,16 +232,15 @@ class StMemorySlab(
     val align: UInt,
     astNode: PtNode
 ):
-    StNode(name, StNodeType.MEMORYSLAB, astNode) {
-}
+    StNode(name, StNodeType.MEMORYSLAB, astNode)
+
 
 class StSub(name: String, val parameters: List<StSubroutineParameter>, val returnType: DataType?, astNode: PtNode) :
-        StNode(name, StNodeType.SUBROUTINE, astNode) {
-}
+        StNode(name, StNodeType.SUBROUTINE, astNode)
 
 
 class StRomSub(name: String,
-               val address: UInt,
+               val address: UInt?,      // null in case of asmsub, specified in case of romsub
                val parameters: List<StRomSubParameter>,
                val returns: List<StRomSubParameter>,
                astNode: PtNode) :

codeCore/src/prog8/code/SymbolTableMaker.kt

@@ -26,11 +26,9 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
                 PtMemMapped("P8ZP_SCRATCH_REG", DataType.UBYTE, options.compTarget.machine.zeropage.SCRATCH_REG, null, Position.DUMMY),
                 PtMemMapped("P8ZP_SCRATCH_W1", DataType.UWORD, options.compTarget.machine.zeropage.SCRATCH_W1, null, Position.DUMMY),
                 PtMemMapped("P8ZP_SCRATCH_W2", DataType.UWORD, options.compTarget.machine.zeropage.SCRATCH_W2, null, Position.DUMMY),
-                PtMemMapped("P8ESTACK_LO", DataType.ARRAY_UB, options.compTarget.machine.ESTACK_LO, 256u, Position.DUMMY),
-                PtMemMapped("P8ESTACK_HI", DataType.ARRAY_UB, options.compTarget.machine.ESTACK_HI, 256u, Position.DUMMY)
             ).forEach {
                 it.parent = program
-                st.add(StMemVar(it.name, it.type, it.address, null, it))
+                st.add(StMemVar(it.name, it.type, it.address, it.arraySize?.toInt(), it))
             }
         }
 
@@ -40,14 +38,9 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
     private fun addToSt(node: PtNode, scope: Stack<StNode>) {
         val stNode = when(node) {
             is PtAsmSub -> {
-                if(node.address==null) {
-                    val params = node.parameters.map { StSubroutineParameter(it.second.name, it.second.type) }
-                    StSub(node.name, params, node.returns.singleOrNull()?.second, node)
-                } else {
-                    val parameters = node.parameters.map { StRomSubParameter(it.first, it.second.type) }
-                    val returns = node.returns.map { StRomSubParameter(it.first, it.second) }
-                    StRomSub(node.name, node.address, parameters, returns, node)
-                }
+                val parameters = node.parameters.map { StRomSubParameter(it.first, it.second.type) }
+                val returns = node.returns.map { StRomSubParameter(it.first, it.second) }
+                StRomSub(node.name, node.address, parameters, returns, node)
             }
             is PtBlock -> {
                 StNode(node.name, StNodeType.BLOCK, node)
@@ -81,10 +74,9 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
                             numElements = value.value.length + 1   // include the terminating 0-byte
                         }
                         is PtArray -> {
-                            val array = makeInitialArray(value)
-                            initialArray = if(array.all { it.number==0.0 }) null else array            // all 0 as init value -> just uninitialized
+                            initialArray = makeInitialArray(value)
                             initialString = null
-                            numElements = array.size
+                            numElements = initialArray.size
                             require(node.arraySize?.toInt()==numElements)
                         }
                         else -> {
@@ -99,6 +91,9 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
                     initialString = null
                     numElements = node.arraySize?.toInt()
                 }
+//                if(node.type in SplitWordArrayTypes) {
+//                    ... split array also add _lsb and _msb to symboltable?
+//                }
                 StStaticVariable(node.name, node.type, initialNumeric, initialString, initialArray, numElements, node.zeropage, node)
             }
             is PtBuiltinFunctionCall -> {
@@ -130,7 +125,11 @@ class SymbolTableMaker(private val program: PtProgram, private val options: Comp
     private fun makeInitialArray(value: PtArray): List<StArrayElement> {
         return value.children.map {
             when(it) {
-                is PtAddressOf -> StArrayElement(null, it.identifier.name)
+                is PtAddressOf -> {
+                    if(it.isFromArrayElement)
+                        TODO("address-of array element $it in initial array value")
+                    StArrayElement(null, it.identifier.name)
+                }
                 is PtIdentifier -> StArrayElement(null, it.name)
                 is PtNumber -> StArrayElement(it.number, null)
                 else -> throw AssemblyError("invalid array element $it")

codeCore/src/prog8/code/ast/AstBase.kt

@@ -37,16 +37,17 @@ class PtNodeGroup : PtNode(Position.DUMMY)
 sealed class PtNamedNode(var name: String, position: Position): PtNode(position) {
     // Note that as an exception, the 'name' is not read-only
     // but a var. This is to allow for cheap node renames.
-    val scopedName: String by lazy {
-        var namedParent: PtNode = this.parent
-        if(namedParent is PtProgram)
-            name
-        else {
-            while (namedParent !is PtNamedNode)
-                namedParent = namedParent.parent
-            namedParent.scopedName + "." + name
+    val scopedName: String
+        get() {
+            var namedParent: PtNode = this.parent
+            return if(namedParent is PtProgram)
+                name
+            else {
+                while (namedParent !is PtNamedNode)
+                    namedParent = namedParent.parent
+                namedParent.scopedName + "." + name
+            }
         }
-    }
 }
 
 
@@ -63,7 +64,9 @@ class PtProgram(
         children.asSequence().filterIsInstance<PtBlock>()
 
     fun entrypoint(): PtSub? =
-        allBlocks().firstOrNull { it.name == "main" }?.children?.firstOrNull { it is PtSub && it.name == "start" } as PtSub?
+        allBlocks().firstOrNull { it.name == "main" || it.name=="p8_main" }
+            ?.children
+            ?.firstOrNull { it is PtSub && (it.name == "start" || it.name=="main.start" || it.name=="p8_start" || it.name=="p8_main.p8_start") } as PtSub?
 }
 
 
@@ -71,6 +74,8 @@ class PtBlock(name: String,
               val address: UInt?,
               val library: Boolean,
               val forceOutput: Boolean,
+              val noSymbolPrefixing: Boolean,
+              val veraFxMuls: Boolean,
               val alignment: BlockAlignment,
               val source: SourceCode,       // taken from the module the block is defined in.
               position: Position

codeCore/src/prog8/code/ast/AstExpressions.kt

@@ -24,8 +24,18 @@ sealed class PtExpression(val type: DataType, position: Position) : PtNode(posit
 
     infix fun isSameAs(other: PtExpression): Boolean {
         return when(this) {
-            is PtAddressOf -> other is PtAddressOf && other.type==type && other.identifier isSameAs identifier
-            is PtArrayIndexer -> other is PtArrayIndexer && other.type==type && other.variable isSameAs variable && other.index isSameAs index
+            is PtAddressOf -> {
+                if(other !is PtAddressOf)
+                    return false
+                if (other.type!==type || !(other.identifier isSameAs identifier))
+                    return false
+                if(other.children.size!=children.size)
+                    return false
+                if(children.size==1)
+                    return true
+                return arrayIndexExpr!! isSameAs other.arrayIndexExpr!!
+            }
+            is PtArrayIndexer -> other is PtArrayIndexer && other.type==type && other.variable isSameAs variable && other.index isSameAs index && other.splitWords==splitWords
             is PtBinaryExpression -> other is PtBinaryExpression && other.left isSameAs left && other.right isSameAs right
             is PtContainmentCheck -> other is PtContainmentCheck && other.type==type && other.element isSameAs element && other.iterable isSameAs iterable
             is PtIdentifier -> other is PtIdentifier && other.type==type && other.name==name
@@ -48,7 +58,7 @@ sealed class PtExpression(val type: DataType, position: Position) : PtNode(posit
                 this.name == target.identifier!!.name
             }
             target.array != null && this is PtArrayIndexer -> {
-                this.variable.name == target.array!!.variable.name && this.index isSameAs target.array!!.index
+                this.variable.name == target.array!!.variable.name && this.index isSameAs target.array!!.index && this.splitWords==target.array!!.splitWords
             }
             else -> false
         }
@@ -105,7 +115,12 @@ sealed class PtExpression(val type: DataType, position: Position) : PtNode(posit
 
 class PtAddressOf(position: Position) : PtExpression(DataType.UWORD, position) {
     val identifier: PtIdentifier
-        get() = children.single() as PtIdentifier
+        get() = children[0] as PtIdentifier
+    val arrayIndexExpr: PtExpression?
+        get() = if(children.size==2) children[1] as PtExpression else null
+
+    val isFromArrayElement: Boolean
+        get() = children.size==2
 }
 
 
@@ -115,6 +130,12 @@ class PtArrayIndexer(elementType: DataType, position: Position): PtExpression(el
     val index: PtExpression
         get() = children[1] as PtExpression
 
+    val splitWords: Boolean
+        get() = variable.type in SplitWordArrayTypes
+
+    val usesPointerVariable: Boolean
+        get() = variable.type==DataType.UWORD
+
     init {
         require(elementType in NumericDatatypes)
     }
@@ -180,7 +201,13 @@ class PtFunctionCall(val name: String,
 }
 
 
-class PtIdentifier(val name: String, type: DataType, position: Position) : PtExpression(type, position)
+class PtIdentifier(val name: String, type: DataType, position: Position) : PtExpression(type, position) {
+    override fun toString(): String {
+        return "[PtIdentifier:$name $type $position]"
+    }
+
+    fun copy() = PtIdentifier(name, type, position)
+}
 
 
 class PtMemoryByte(position: Position) : PtExpression(DataType.UBYTE, position) {
@@ -215,6 +242,8 @@ class PtNumber(type: DataType, val number: Double, position: Position) : PtExpre
     }
 
     operator fun compareTo(other: PtNumber): Int = number.compareTo(other.number)
+
+    override fun toString() = "PtNumber:$type:$number"
 }
 
 

codeCore/src/prog8/code/ast/AstPrinter.kt

@@ -6,7 +6,7 @@ import prog8.code.core.*
  * Produces readable text from a [PtNode] (AST node, usually starting with PtProgram as root),
  * passing it as a String to the specified receiver function.
  */
-fun printAst(root: PtNode, output: (text: String) -> Unit) {
+fun printAst(root: PtNode, skipLibraries: Boolean, output: (text: String) -> Unit) {
     fun type(dt: DataType) = "!${dt.name.lowercase()}!"
     fun txt(node: PtNode): String {
         return when(node) {
@@ -17,7 +17,7 @@ fun printAst(root: PtNode, output: (text: String) -> Unit) {
             is PtConditionalBranch -> "if_${node.condition.name.lowercase()}"
             is PtAddressOf -> "&"
             is PtArray -> "array len=${node.children.size} ${type(node.type)}"
-            is PtArrayIndexer -> "<arrayindexer> ${type(node.type)}"
+            is PtArrayIndexer -> "<arrayindexer> ${type(node.type)} ${if(node.splitWords) "[splitwords]" else ""}"
             is PtBinaryExpression -> "<expr> ${node.operator} ${type(node.type)}"
             is PtBuiltinFunctionCall -> {
                 val str = if(node.void) "void " else ""
@@ -96,13 +96,14 @@ fun printAst(root: PtNode, output: (text: String) -> Unit) {
                 str
             }
             is PtVariable -> {
+                val split = if(node.type in SplitWordArrayTypes) "@split" else ""
                 val str = if(node.arraySize!=null) {
                     val eltType = ArrayToElementTypes.getValue(node.type)
-                    "${eltType.name.lowercase()}[${node.arraySize}] ${node.name}"
+                    "${eltType.name.lowercase()}[${node.arraySize}] $split ${node.name}"
                 }
                 else if(node.type in ArrayDatatypes) {
                     val eltType = ArrayToElementTypes.getValue(node.type)
-                    "${eltType.name.lowercase()}[] ${node.name}"
+                    "${eltType.name.lowercase()}[] $split ${node.name}"
                 }
                 else
                     "${node.type.name.lowercase()} ${node.name}"
@@ -125,7 +126,6 @@ fun printAst(root: PtNode, output: (text: String) -> Unit) {
                 else
                     "->"
             }
-            else -> throw InternalCompilerException("unrecognised ast node $node")
         }
     }
 
@@ -134,16 +134,22 @@ fun printAst(root: PtNode, output: (text: String) -> Unit) {
         root.children.forEach {
             walkAst(it) { node, depth ->
                 val txt = txt(node)
-                if(txt.isNotEmpty())
-                    output("    ".repeat(depth) + txt(node))
+                val library = if(node is PtBlock) node.library else node.definingBlock()?.library==true
+                if(!library || !skipLibraries) {
+                    if (txt.isNotEmpty())
+                        output("    ".repeat(depth) + txt(node))
+                }
             }
         }
         println()
     } else {
         walkAst(root) { node, depth ->
             val txt = txt(node)
-            if(txt.isNotEmpty())
-                output("    ".repeat(depth) + txt(node))
+            val library = if(node is PtBlock) node.library else node.definingBlock()?.library==true
+            if(!library || !skipLibraries) {
+                if (txt.isNotEmpty())
+                    output("    ".repeat(depth) + txt(node))
+            }
         }
     }
 }

codeCore/src/prog8/code/ast/AstStatements.kt

@@ -38,25 +38,17 @@ class PtSub(
 class PtSubroutineParameter(name: String, val type: DataType, position: Position): PtNamedNode(name, position)
 
 
-class PtAssignment(position: Position) : PtNode(position) {
+sealed interface IPtAssignment {
+    val children: MutableList<PtNode>
     val target: PtAssignTarget
         get() = children[0] as PtAssignTarget
     val value: PtExpression
         get() = children[1] as PtExpression
 }
 
+class PtAssignment(position: Position) : PtNode(position), IPtAssignment
 
-class PtAugmentedAssign(val operator: String, position: Position) : PtNode(position) {
-    val target: PtAssignTarget
-        get() = children[0] as PtAssignTarget
-    val value: PtExpression
-        get() = children[1] as PtExpression
-    init {
-        require(operator.endsWith('=') || operator in PrefixOperators) {
-            "invalid augmented assign operator $operator"
-        }
-    }
-}
+class PtAugmentedAssign(val operator: String, position: Position) : PtNode(position), IPtAssignment
 
 
 class PtAssignTarget(position: Position) : PtNode(position) {
@@ -100,8 +92,8 @@ class PtForLoop(position: Position) : PtNode(position) {
 
 
 class PtIfElse(position: Position) : PtNode(position) {
-    val condition: PtBinaryExpression
-        get() = children[0] as PtBinaryExpression
+    val condition: PtExpression
+        get() = children[0] as PtExpression
     val ifScope: PtNodeGroup
         get() = children[1] as PtNodeGroup
     val elseScope: PtNodeGroup

codeCore/src/prog8/code/core/BuiltinFunctions.kt

@@ -68,9 +68,10 @@ class FSignature(val pure: Boolean,      // does it have side effects?
     }
 }
 
-
 val BuiltinFunctions: Map<String, FSignature> = mapOf(
     // this set of function have no return value and operate in-place:
+    "setlsb"    to FSignature(false, listOf(FParam("variable", arrayOf(DataType.WORD, DataType.UWORD)), FParam("value", arrayOf(DataType.BYTE, DataType.UBYTE))), null),
+    "setmsb"    to FSignature(false, listOf(FParam("variable", arrayOf(DataType.WORD, DataType.UWORD)), FParam("value", arrayOf(DataType.BYTE, DataType.UBYTE))), null),
     "rol"       to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
     "ror"       to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
     "rol2"      to FSignature(false, listOf(FParam("item", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
@@ -79,17 +80,44 @@ val BuiltinFunctions: Map<String, FSignature> = mapOf(
     "reverse"   to FSignature(false, listOf(FParam("array", ArrayDatatypes)), null),
     // cmp returns a status in the carry flag, but not a proper return value
     "cmp"       to FSignature(false, listOf(FParam("value1", IntegerDatatypesNoBool), FParam("value2", NumericDatatypesNoBool)), null),
-    "abs"       to FSignature(true, listOf(FParam("value", IntegerDatatypesNoBool)), DataType.UWORD),
+    "prog8_lib_stringcompare"     to FSignature(true, listOf(FParam("str1", arrayOf(DataType.STR)), FParam("str2", arrayOf(DataType.STR))), DataType.BYTE),
+    "prog8_lib_square_byte"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE, DataType.UBYTE))), DataType.UBYTE),
+    "prog8_lib_square_word"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.WORD, DataType.UWORD))), DataType.UWORD),
+    "abs"       to FSignature(true, listOf(FParam("value", NumericDatatypesNoBool)), null),
+    "abs__byte"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE))), DataType.BYTE),
+    "abs__word"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.WORD))), DataType.WORD),
+    "abs__float" to FSignature(true, listOf(FParam("value", arrayOf(DataType.FLOAT))), DataType.FLOAT),
     "len"       to FSignature(true, listOf(FParam("values", IterableDatatypes)), DataType.UWORD),
     // normal functions follow:
-    "sizeof"    to FSignature(true, listOf(FParam("object", DataType.values())), DataType.UBYTE),
+    "sizeof"    to FSignature(true, listOf(FParam("object", DataType.entries.toTypedArray())), DataType.UBYTE),
     "sgn"       to FSignature(true, listOf(FParam("value", NumericDatatypesNoBool)), DataType.BYTE),
-    "sqrt16"    to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD))), DataType.UBYTE),
+    "sqrt"        to FSignature(true, listOf(FParam("value", NumericDatatypesNoBool)), null),
+    "sqrt__ubyte" to FSignature(true, listOf(FParam("value", arrayOf(DataType.UBYTE))), DataType.UBYTE),
+    "sqrt__uword" to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD))), DataType.UBYTE),
+    "sqrt__float" to FSignature(true, listOf(FParam("value", arrayOf(DataType.FLOAT))), DataType.FLOAT),
+    "divmod"           to FSignature(false, listOf(FParam("number", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("divident", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("division", arrayOf(DataType.UBYTE, DataType.UWORD)), FParam("remainder", arrayOf(DataType.UBYTE, DataType.UWORD))), null),
+    "divmod__ubyte"    to FSignature(false, listOf(FParam("number", arrayOf(DataType.UBYTE)), FParam("divident", arrayOf(DataType.UBYTE)), FParam("division", arrayOf(DataType.UBYTE)), FParam("remainder", arrayOf(DataType.UBYTE))), null),
+    "divmod__uword"    to FSignature(false, listOf(FParam("number", arrayOf(DataType.UWORD)), FParam("divident", arrayOf(DataType.UWORD)), FParam("division", arrayOf(DataType.UWORD)), FParam("remainder", arrayOf(DataType.UWORD))), null),
     "any"       to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE),
     "all"       to FSignature(true, listOf(FParam("values", ArrayDatatypes)), DataType.UBYTE),
     "lsb"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE),
     "msb"       to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD, DataType.WORD))), DataType.UBYTE),
     "mkword"    to FSignature(true, listOf(FParam("msb", arrayOf(DataType.UBYTE)), FParam("lsb", arrayOf(DataType.UBYTE))), DataType.UWORD),
+    "clamp"     to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE)), FParam("minimum", arrayOf(DataType.BYTE)), FParam("maximum", arrayOf(DataType.BYTE))), null),
+    "clamp__byte"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.BYTE)), FParam("minimum", arrayOf(DataType.BYTE)), FParam("maximum", arrayOf(DataType.BYTE))), DataType.BYTE),
+    "clamp__ubyte" to FSignature(true, listOf(FParam("value", arrayOf(DataType.UBYTE)), FParam("minimum", arrayOf(DataType.UBYTE)), FParam("maximum", arrayOf(DataType.UBYTE))), DataType.UBYTE),
+    "clamp__word"  to FSignature(true, listOf(FParam("value", arrayOf(DataType.WORD)), FParam("minimum", arrayOf(DataType.WORD)), FParam("maximum", arrayOf(DataType.WORD))), DataType.WORD),
+    "clamp__uword" to FSignature(true, listOf(FParam("value", arrayOf(DataType.UWORD)), FParam("minimum", arrayOf(DataType.UWORD)), FParam("maximum", arrayOf(DataType.UWORD))), DataType.UWORD),
+    "min"        to FSignature(true, listOf(FParam("val1", arrayOf(DataType.BYTE)), FParam("val2", arrayOf(DataType.BYTE))), null),
+    "min__byte"  to FSignature(true, listOf(FParam("val1", arrayOf(DataType.BYTE)), FParam("val2", arrayOf(DataType.BYTE))), DataType.BYTE),
+    "min__ubyte" to FSignature(true, listOf(FParam("val1", arrayOf(DataType.UBYTE)), FParam("val2", arrayOf(DataType.UBYTE))), DataType.UBYTE),
+    "min__word"  to FSignature(true, listOf(FParam("val1", arrayOf(DataType.WORD)), FParam("val2", arrayOf(DataType.WORD))), DataType.WORD),
+    "min__uword" to FSignature(true, listOf(FParam("val1", arrayOf(DataType.UWORD)), FParam("val2", arrayOf(DataType.UWORD))), DataType.UWORD),
+    "max"        to FSignature(true, listOf(FParam("val1", arrayOf(DataType.BYTE)), FParam("val2", arrayOf(DataType.BYTE))), null),
+    "max__byte"  to FSignature(true, listOf(FParam("val1", arrayOf(DataType.BYTE)), FParam("val2", arrayOf(DataType.BYTE))), DataType.BYTE),
+    "max__ubyte" to FSignature(true, listOf(FParam("val1", arrayOf(DataType.UBYTE)), FParam("val2", arrayOf(DataType.UBYTE))), DataType.UBYTE),
+    "max__word"  to FSignature(true, listOf(FParam("val1", arrayOf(DataType.WORD)), FParam("val2", arrayOf(DataType.WORD))), DataType.WORD),
+    "max__uword" to FSignature(true, listOf(FParam("val1", arrayOf(DataType.UWORD)), FParam("val2", arrayOf(DataType.UWORD))), DataType.UWORD),
     "peek"      to FSignature(true, listOf(FParam("address", arrayOf(DataType.UWORD))), DataType.UBYTE),
     "peekw"     to FSignature(true, listOf(FParam("address", arrayOf(DataType.UWORD))), DataType.UWORD),
     "poke"      to FSignature(false, listOf(FParam("address", arrayOf(DataType.UWORD)), FParam("value", arrayOf(DataType.UBYTE))), null),
@@ -100,11 +128,9 @@ val BuiltinFunctions: Map<String, FSignature> = mapOf(
     "push"      to FSignature(false, listOf(FParam("value", ByteDatatypes)), null),
     "pushw"     to FSignature(false, listOf(FParam("value", WordDatatypes)), null),
     "rsave"     to FSignature(false, emptyList(), null),
-    "rsavex"    to FSignature(false, emptyList(), null),
     "rrestore"  to FSignature(false, emptyList(), null),
-    "rrestorex" to FSignature(false, emptyList(), null),
     "memory"    to FSignature(true, listOf(FParam("name", arrayOf(DataType.STR)), FParam("size", arrayOf(DataType.UWORD)), FParam("alignment", arrayOf(DataType.UWORD))), DataType.UWORD),
     "callfar"   to FSignature(false, listOf(FParam("bank", arrayOf(DataType.UBYTE)), FParam("address", arrayOf(DataType.UWORD)), FParam("arg", arrayOf(DataType.UWORD))), DataType.UWORD),
 )
 
-val InplaceModifyingBuiltinFunctions = setOf("rol", "ror", "rol2", "ror2", "sort", "reverse")
+val InplaceModifyingBuiltinFunctions = setOf("setlsb", "setmsb", "rol", "ror", "rol2", "ror2", "sort", "reverse")

codeCore/src/prog8/code/core/CompilationOptions.kt

@@ -13,14 +13,14 @@ class CompilationOptions(val output: OutputType,
                          val compTarget: ICompilationTarget,
                          // these are set later, based on command line arguments or options in the source code:
                          var loadAddress: UInt,
-                         var slowCodegenWarnings: Boolean = false,
+                         var warnSymbolShadowing: Boolean = false,
                          var optimize: Boolean = false,
-                         var optimizeFloatExpressions: Boolean = false,
                          var asmQuiet: Boolean = false,
                          var asmListfile: Boolean = false,
+                         var includeSourcelines: Boolean = false,
                          var experimentalCodegen: Boolean = false,
-                         var varsHigh: Boolean = false,
-                         var evalStackBaseAddress: UInt? = null,
+                         var varsHighBank: Int? = null,
+                         var splitWordArrays: Boolean = false,
                          var outputDir: Path = Path(""),
                          var symbolDefs: Map<String, String> = emptyMap()
 ) {

codeCore/src/prog8/code/core/Enumerations.kt

@@ -11,7 +11,9 @@ enum class DataType {
     ARRAY_UB,           // pass by reference
     ARRAY_B,            // pass by reference
     ARRAY_UW,           // pass by reference
+    ARRAY_UW_SPLIT,     // pass by reference, lo/hi byte split
     ARRAY_W,            // pass by reference
+    ARRAY_W_SPLIT,      // pass by reference, lo/hi byte split
     ARRAY_F,            // pass by reference
     ARRAY_BOOL,         // pass by reference
     UNDEFINED;
@@ -119,12 +121,14 @@ val IntegerDatatypesNoBool = arrayOf(DataType.UBYTE, DataType.BYTE, DataType.UWO
 val NumericDatatypes = arrayOf(DataType.UBYTE, DataType.BYTE, DataType.UWORD, DataType.WORD, DataType.FLOAT, DataType.BOOL)
 val NumericDatatypesNoBool = arrayOf(DataType.UBYTE, DataType.BYTE, DataType.UWORD, DataType.WORD, DataType.FLOAT)
 val SignedDatatypes =  arrayOf(DataType.BYTE, DataType.WORD, DataType.FLOAT)
-val ArrayDatatypes = arrayOf(DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_W, DataType.ARRAY_F, DataType.ARRAY_BOOL)
+val ArrayDatatypes = arrayOf(DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_UW_SPLIT, DataType.ARRAY_W, DataType.ARRAY_W_SPLIT, DataType.ARRAY_F, DataType.ARRAY_BOOL)
 val StringlyDatatypes = arrayOf(DataType.STR, DataType.ARRAY_UB, DataType.ARRAY_B, DataType.UWORD)
+val SplitWordArrayTypes = arrayOf(DataType.ARRAY_UW_SPLIT, DataType.ARRAY_W_SPLIT)
 val IterableDatatypes = arrayOf(
     DataType.STR,
     DataType.ARRAY_UB, DataType.ARRAY_B,
     DataType.ARRAY_UW, DataType.ARRAY_W,
+    DataType.ARRAY_UW_SPLIT, DataType.ARRAY_W_SPLIT,
     DataType.ARRAY_F, DataType.ARRAY_BOOL
 )
 val PassByValueDatatypes = NumericDatatypes
@@ -135,6 +139,8 @@ val ArrayToElementTypes = mapOf(
     DataType.ARRAY_UB to DataType.UBYTE,
     DataType.ARRAY_W to DataType.WORD,
     DataType.ARRAY_UW to DataType.UWORD,
+    DataType.ARRAY_W_SPLIT to DataType.WORD,
+    DataType.ARRAY_UW_SPLIT to DataType.UWORD,
     DataType.ARRAY_F to DataType.FLOAT,
     DataType.ARRAY_BOOL to DataType.BOOL
 )

codeCore/src/prog8/code/core/ICodeGeneratorBackend.kt

@@ -13,7 +13,5 @@ interface ICodeGeneratorBackend {
 
 interface IAssemblyProgram {
     val name: String
-    fun assemble(options: CompilationOptions): Boolean
+    fun assemble(options: CompilationOptions, errors: IErrorReporter): Boolean
 }
-
-fun viceMonListName(baseFilename: String) = "$baseFilename.vice-mon-list"

codeCore/src/prog8/code/core/IErrorReporter.kt

@@ -3,6 +3,7 @@ package prog8.code.core
 interface IErrorReporter {
     fun err(msg: String, position: Position)
     fun warn(msg: String, position: Position)
+    fun undefined(symbol: List<String>, position: Position)
     fun noErrors(): Boolean
     fun report()
     fun finalizeNumErrors(numErrors: Int, numWarnings: Int) {

codeCore/src/prog8/code/core/IMachineDefinition.kt

@@ -13,8 +13,6 @@ interface IMachineDefinition {
     val FLOAT_MAX_NEGATIVE: Double
     val FLOAT_MAX_POSITIVE: Double
     val FLOAT_MEM_SIZE: Int
-    var ESTACK_LO: UInt
-    var ESTACK_HI: UInt
     val PROGRAM_LOAD_ADDRESS : UInt
     val BSSHIGHRAM_START: UInt
     val BSSHIGHRAM_END: UInt
@@ -29,11 +27,4 @@ interface IMachineDefinition {
     fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String>
     fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path)
     fun isIOAddress(address: UInt): Boolean
-    fun overrideEvalStack(evalStackBaseAddress: UInt) {
-        require(evalStackBaseAddress and 255u == 0u)
-        ESTACK_LO = evalStackBaseAddress
-        ESTACK_HI = evalStackBaseAddress + 256u
-        require(ESTACK_LO !in golden.region && ESTACK_HI !in golden.region) { "user-set ESTACK can't be in GOLDEN ram" }
-    }
-
 }

codeCore/src/prog8/code/core/MemoryRegions.kt

@@ -131,6 +131,7 @@ abstract class Zeropage(options: CompilationOptions): MemoryAllocator(options) {
 }
 
 
+// TODO: this class is not yet used
 class GoldenRam(options: CompilationOptions, val region: UIntRange): MemoryAllocator(options) {
     private var nextLocation: UInt = region.first
 

codeCore/src/prog8/code/core/Position.kt

@@ -1,5 +1,6 @@
 package prog8.code.core
 
+import prog8.code.core.SourceCode.Companion.LIBRARYFILEPREFIX
 import java.nio.file.InvalidPathException
 import kotlin.io.path.Path
 import kotlin.io.path.absolute
@@ -7,6 +8,10 @@ import kotlin.io.path.absolute
 data class Position(val file: String, val line: Int, val startCol: Int, val endCol: Int) {
     override fun toString(): String = "[$file: line $line col ${startCol+1}-${endCol+1}]"
     fun toClickableStr(): String {
+        if(this===DUMMY)
+            return ""
+        if(file.startsWith(LIBRARYFILEPREFIX))
+            return "$file:$line:$startCol:"
         return try {
             val path = Path(file).absolute().normalize().toString()
             "file://$path:$line:$startCol:"

codeCore/src/prog8/code/core/SourceCode.kt

@@ -54,12 +54,14 @@ sealed class SourceCode {
         /**
          * filename prefix to designate library files that will be retreived from internal resources rather than disk
          */
-        const val libraryFilePrefix = "library:"
-        const val stringSourcePrefix = "string:"
+        const val LIBRARYFILEPREFIX = "library:"
+        const val STRINGSOURCEPREFIX = "string:"
         val curdir: Path = Path(".").toAbsolutePath()
         fun relative(path: Path): Path = curdir.relativize(path.toAbsolutePath())
         fun isRegularFilesystemPath(pathString: String) =
-            !(pathString.startsWith(libraryFilePrefix) || pathString.startsWith(stringSourcePrefix))
+            !(pathString.startsWith(LIBRARYFILEPREFIX) || pathString.startsWith(STRINGSOURCEPREFIX))
+
+        fun isLibraryResource(path: String) = path.startsWith(LIBRARYFILEPREFIX)
     }
 
     /**
@@ -69,7 +71,7 @@ sealed class SourceCode {
     class Text(override val text: String): SourceCode() {
         override val isFromResources = false
         override val isFromFilesystem = false
-        override val origin = "$stringSourcePrefix${System.identityHashCode(text).toString(16)}"
+        override val origin = "$STRINGSOURCEPREFIX${System.identityHashCode(text).toString(16)}"
         override val name = "<unnamed-text>"
     }
 
@@ -110,7 +112,7 @@ sealed class SourceCode {
 
         override val isFromResources = true
         override val isFromFilesystem = false
-        override val origin = "$libraryFilePrefix$normalized"
+        override val origin = "$LIBRARYFILEPREFIX$normalized"
         override val text: String
         override val name: String
 
@@ -139,3 +141,33 @@ sealed class SourceCode {
         override val text: String = "<generated code node, no text representation>"
     }
 }
+
+
+object SourceLineCache {
+    private val cache = mutableMapOf<String, List<String>>()
+
+    private fun getCachedFile(file: String): List<String> {
+        val existing = cache[file]
+        if(existing!=null)
+            return existing
+        if (SourceCode.isRegularFilesystemPath(file)) {
+            val source = SourceCode.File(Path(file))
+            cache[file] = source.text.split('\n', '\r').map { it.trim() }
+            return cache.getValue(file)
+        } else if(file.startsWith(SourceCode.LIBRARYFILEPREFIX)) {
+            val source = SourceCode.Resource(file.drop(SourceCode.LIBRARYFILEPREFIX.length))
+            cache[file] = source.text.split('\n', '\r').map { it.trim()}
+            return cache.getValue(file)
+        }
+        return emptyList()
+    }
+
+    fun retrieveLine(position: Position): String? {
+        if (position.line>0) {
+            val lines = getCachedFile(position.file)
+            if(lines.isNotEmpty())
+                return lines[position.line-1]
+        }
+        return null
+    }
+}

codeCore/src/prog8/code/target/C64Target.kt

@@ -16,5 +16,7 @@ class C64Target: ICompilationTarget, IStringEncoding by Encoder, IMemSizer by Cb
 
     companion object {
         const val NAME = "c64"
+
+        fun viceMonListName(baseFilename: String) = "$baseFilename.vice-mon-list"
     }
 }

codeCore/src/prog8/code/target/PETTarget.kt

@@ -0,0 +1,20 @@
+package prog8.code.target
+
+import prog8.code.core.Encoding
+import prog8.code.core.ICompilationTarget
+import prog8.code.core.IMemSizer
+import prog8.code.core.IStringEncoding
+import prog8.code.target.cbm.CbmMemorySizer
+import prog8.code.target.pet.PETMachineDefinition
+
+
+class PETTarget: ICompilationTarget, IStringEncoding by Encoder, IMemSizer by CbmMemorySizer {
+    override val name = NAME
+    override val machine = PETMachineDefinition()
+    override val supportedEncodings = setOf(Encoding.PETSCII, Encoding.SCREENCODES)
+    override val defaultEncoding = Encoding.PETSCII
+
+    companion object {
+        const val NAME = "pet32"
+    }
+}

codeCore/src/prog8/code/target/atari/AtariMachineDefinition.kt

@@ -13,17 +13,13 @@ class AtariMachineDefinition: IMachineDefinition {
     override val FLOAT_MEM_SIZE = 6
     override val PROGRAM_LOAD_ADDRESS = 0x2000u
 
-    // the 2*128 byte evaluation stack (1 page, on which bytes, words, and even floats are stored during calculations)
-    override var ESTACK_LO = 0x1b00u     //  $1b00-$1b7f inclusive      // TODO
-    override var ESTACK_HI = 0x1b80u     //  $1b80-$1bff inclusive      // TODO
-
     override val BSSHIGHRAM_START = 0u    // TODO
     override val BSSHIGHRAM_END = 0u      // TODO
 
     override lateinit var zeropage: Zeropage
     override lateinit var golden: GoldenRam
 
-    override fun getFloatAsmBytes(num: Number) = TODO("float asm bytes from number")
+    override fun getFloatAsmBytes(num: Number) = TODO("atari float asm bytes from number")
 
     override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
         return if (compilerOptions.output == OutputType.XEX)

codeCore/src/prog8/code/target/atari/AtariZeropage.kt

@@ -13,6 +13,10 @@ class AtariZeropage(options: CompilationOptions) : Zeropage(options) {
     override val SCRATCH_W2 = 0xcfu      // temp storage 2 for a word  $cf+$d0        TODO is $d0 okay to use?
 
     init {
+        if (options.floats) {
+            throw InternalCompilerException("Atari target doesn't yet support floating point routines")
+        }
+
         if (options.floats && options.zeropage !in arrayOf(
                 ZeropageType.FLOATSAFE,
                 ZeropageType.BASICSAFE,
@@ -39,9 +43,9 @@ class AtariZeropage(options: CompilationOptions) : Zeropage(options) {
             }
         }
 
-        val distictFree = free.distinct()
+        val distinctFree = free.distinct()
         free.clear()
-        free.addAll(distictFree)
+        free.addAll(distinctFree)
 
         removeReservedFromFreePool()
     }

codeCore/src/prog8/code/target/c128/C128MachineDefinition.kt

@@ -1,6 +1,7 @@
 package prog8.code.target.c128
 
 import prog8.code.core.*
+import prog8.code.target.C64Target
 import prog8.code.target.cbm.Mflpt5
 import java.nio.file.Path
 
@@ -14,10 +15,6 @@ class C128MachineDefinition: IMachineDefinition {
     override val FLOAT_MEM_SIZE = Mflpt5.FLOAT_MEM_SIZE
     override val PROGRAM_LOAD_ADDRESS = 0x1c01u
 
-    // the 2*128 byte evaluation stack (1 page, on which bytes, words, and even floats are stored during calculations)
-    override var ESTACK_LO = 0x1b00u     //  $1b00-$1b7f inclusive
-    override var ESTACK_HI = 0x1b80u     //  $1b80-$1bff inclusive
-
     override val BSSHIGHRAM_START = 0u    // TODO
     override val BSSHIGHRAM_END = 0u      // TODO
 
@@ -40,7 +37,7 @@ class C128MachineDefinition: IMachineDefinition {
         }
 
         println("\nStarting C-128 emulator x128...")
-        val viceMonlist = viceMonListName(programNameWithPath.toString())
+        val viceMonlist = C64Target.viceMonListName(programNameWithPath.toString())
         val cmdline = listOf("x128", "-silent", "-moncommands", viceMonlist,
                 "-autostartprgmode", "1", "-autostart-warp", "-autostart", "${programNameWithPath}.prg")
         val processb = ProcessBuilder(cmdline).inheritIO()

codeCore/src/prog8/code/target/c128/C128Zeropage.kt

@@ -6,15 +6,22 @@ import prog8.code.core.Zeropage
 import prog8.code.core.ZeropageType
 
 
+// reference: "Mapping the C128" zero page chapter.
+
 class C128Zeropage(options: CompilationOptions) : Zeropage(options) {
 
-    override val SCRATCH_B1 = 0x9bu      // temp storage for a single byte
-    override val SCRATCH_REG = 0x9cu     // temp storage for a register, must be B1+1
+    override val SCRATCH_B1 = 0x74u      // temp storage for a single byte
+    override val SCRATCH_REG = 0x75u     // temp storage for a register, must be B1+1
     override val SCRATCH_W1 = 0xfbu      // temp storage 1 for a word  $fb+$fc
     override val SCRATCH_W2 = 0xfdu      // temp storage 2 for a word  $fd+$fe
 
-
     init {
+        if (options.floats) {
+            throw InternalCompilerException("C128 target doesn't yet support floating point routines")
+            // note: in git commit labeled 'c128: remove floats module' the floats.p8 and floats.asm files are removed,
+            //       they could be retrieved again at a later time if the compiler somehow *does* store the fp variables in bank1.
+        }
+
         if (options.floats && options.zeropage !in arrayOf(
                 ZeropageType.FLOATSAFE,
                 ZeropageType.BASICSAFE,
@@ -24,23 +31,42 @@ class C128Zeropage(options: CompilationOptions) : Zeropage(options) {
 
         when (options.zeropage) {
             ZeropageType.FULL -> {
-                // TODO all c128 usable zero page locations, except the ones used by the system's IRQ routine
-                free.addAll(0x0au..0xffu)       // TODO c128 what about $02-$09?
-                // TODO c128  free.removeAll(setOf(0xa0u, 0xa1u, 0xa2u, 0x91u, 0xc0u, 0xc5u, 0xcbu, 0xf5u, 0xf6u))        // these are updated by IRQ
+                // $00/$01 are data port IO registers, // $02-$09 are storage locations for JSRFAR and such
+                free.addAll(0x0au..0xffu)
+                free.removeAll(listOf(0x90u, 0x91u, 0xa0u, 0xa1u, 0xa2u, 0xc0u, 0xccu, 0xcdu, 0xd0u, 0xd1u, 0xd2u, 0xd3u, 0xd4u, 0xd5u, 0xf7u))        // these are updated/used by IRQ
+            }
+            ZeropageType.KERNALSAFE -> {
+                free.addAll(0x0au..0x8fu)       // BASIC variables
+                free.addAll(listOf(0x92u, 0x96u, 0x9bu, 0x9cu, 0x9eu, 0x9fu, 0xa4u, 0xa7u, 0xa8u, 0xa9u, 0xaau, 0xabu,
+                    0xb0u, 0xb1u, 0xb4u, 0xb5u, 0xb6u))
             }
-            ZeropageType.KERNALSAFE,
             ZeropageType.FLOATSAFE,
             ZeropageType.BASICSAFE -> {
-                free.clear()   // TODO c128 usable zero page addresses
+                free.addAll(listOf(0x0bu, 0x0cu, 0x0du, 0x0eu, 0x0fu, 0x10u, 0x11u, 0x12u, 0x16u, 0x17u, 0x18u, 0x19u, 0x1au))
+                free.addAll(0x1bu..0x23u)
+                free.addAll(listOf(0x3fu, 0x40u, 0x41u, 0x42u, 0x43u, 0x44u, 0x47u, 0x48u, 0x49u, 0x4au, 0x4bu, 0x4cu, 0x4fu,
+                    0x55u, 0x56u, 0x57u, 0x58u,
+                    0x74u, 0x75u, 0x78u, 0x80u, 0x83u, 0x87u, 0x88u, 0x89u, 0x8au, 0x8bu, 0x8cu, 0x8du, 0x8eu, 0x8fu,
+                    0x92u, 0x96u, 0x9bu, 0x9cu, 0x9eu, 0x9fu, 0xa4u, 0xa7u, 0xa8u, 0xa9u, 0xaau, 0xabu,
+                    0xb0u, 0xb1u, 0xb4u, 0xb5u, 0xb6u
+                ))
+
+                // if(options.zeropage==ZeropageType.BASICSAFE) {
+                    // can also clobber the FP locations (unconditionally, because the C128 target doesn't support floating point calculations in prog8 at this time0
+                    free.addAll(listOf(0x14u, 0x28u, 0x29u, 0x2au, 0x2bu, 0x2cu,
+                        0x50u, 0x51u, 0x52u, 0x53u, 0x54u, 0x59u, 0x5au, 0x5bu, 0x5cu, 0x5du, 0x5eu, 0x5fu, 0x60u, 0x61u, 0x62u,
+                        0x63u, 0x64u, 0x65u, 0x66u, 0x67u, 0x68u,
+                        0x6au, 0x6bu, 0x6cu, 0x6du, 0x6eu, 0x6fu, 0x71u))
+                // }
             }
             ZeropageType.DONTUSE -> {
                 free.clear()  // don't use zeropage at all
             }
         }
 
-        val distictFree = free.distinct()
+        val distinctFree = free.distinct()
         free.clear()
-        free.addAll(distictFree)
+        free.addAll(distinctFree)
 
         removeReservedFromFreePool()
     }

codeCore/src/prog8/code/target/c64/C64MachineDefinition.kt

@@ -1,6 +1,7 @@
 package prog8.code.target.c64
 
 import prog8.code.core.*
+import prog8.code.target.C64Target
 import prog8.code.target.cbm.Mflpt5
 import java.io.IOException
 import java.nio.file.Path
@@ -15,12 +16,8 @@ class C64MachineDefinition: IMachineDefinition {
     override val FLOAT_MEM_SIZE = Mflpt5.FLOAT_MEM_SIZE
     override val PROGRAM_LOAD_ADDRESS = 0x0801u
 
-    // the 2*128 byte evaluation stack (1 page, on which bytes, words, and even floats are stored during calculations)
-    override var ESTACK_LO = 0xcf00u     //  $cf00-$cf7f inclusive
-    override var ESTACK_HI = 0xcf80u     //  $cf80-$cfff inclusive
-
     override val BSSHIGHRAM_START = 0xc000u
-    override val BSSHIGHRAM_END = ESTACK_LO
+    override val BSSHIGHRAM_END = 0xd000u
 
     override lateinit var zeropage: Zeropage
     override lateinit var golden: GoldenRam
@@ -42,7 +39,7 @@ class C64MachineDefinition: IMachineDefinition {
 
         for(emulator in listOf("x64sc", "x64")) {
             println("\nStarting C-64 emulator $emulator...")
-            val viceMonlist = viceMonListName(programNameWithPath.toString())
+            val viceMonlist = C64Target.viceMonListName(programNameWithPath.toString())
             val cmdline = listOf(emulator, "-silent", "-moncommands", viceMonlist,
                     "-autostartprgmode", "1", "-autostart-warp", "-autostart", "${programNameWithPath}.prg")
             val processb = ProcessBuilder(cmdline).inheritIO()
@@ -61,7 +58,7 @@ class C64MachineDefinition: IMachineDefinition {
 
     override fun initializeMemoryAreas(compilerOptions: CompilationOptions) {
         zeropage = C64Zeropage(compilerOptions)
-        golden = GoldenRam(compilerOptions, 0xc000u until ESTACK_LO)
+        golden = GoldenRam(compilerOptions, 0xc000u until 0xd000u)
     }
 
 }

codeCore/src/prog8/code/target/c64/C64Zeropage.kt

@@ -44,7 +44,7 @@ class C64Zeropage(options: CompilationOptions) : Zeropage(options) {
             if (options.zeropage == ZeropageType.FLOATSAFE) {
                 // remove the zeropage locations used for floating point operations from the free list
                 free.removeAll(listOf(
-                        0x03, 0x04, 0x10, 0x11, 0x12,
+                        0x03, 0x04, 0x05, 0x06, 0x10, 0x11, 0x12,
                         0x22, 0x23, 0x24, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2a,
                         0x57, 0x58, 0x59, 0x5a, 0x5b, 0x5c, 0x5d, 0x5e, 0x5f, 0x60,
                         0x61, 0x62, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68,
@@ -53,11 +53,11 @@ class C64Zeropage(options: CompilationOptions) : Zeropage(options) {
                 ).map{it.toUInt()})
             }
 
-            if(options.zeropage!= ZeropageType.DONTUSE) {
+            if(options.zeropage != ZeropageType.DONTUSE) {
                 // add the free Zp addresses
                 // these are valid for the C-64 but allow BASIC to keep running fully *as long as you don't use tape I/O*
                 free.addAll(listOf(0x02, 0x03, 0x04, 0x05, 0x06, 0x0a, 0x0e,
-                        0x92, 0x96, 0x9b, 0x9c, 0x9e, 0x9f, 0xa5, 0xa6,
+                        0x92, 0x96, 0x9b, 0x9c, 0x9e, 0x9f, 0xa6,
                         0xb0, 0xb1, 0xbe, 0xbf, 0xf9).map{it.toUInt()})
             } else {
                 // don't use the zeropage at all
@@ -65,9 +65,9 @@ class C64Zeropage(options: CompilationOptions) : Zeropage(options) {
             }
         }
 
-        val distictFree = free.distinct()
+        val distinctFree = free.distinct()
         free.clear()
-        free.addAll(distictFree)
+        free.addAll(distinctFree)
 
         removeReservedFromFreePool()
 

codeCore/src/prog8/code/target/cbm/PetsciiEncoding.kt

@@ -24,7 +24,7 @@ object PetsciiEncoding {
         '\ufffe',    //       0x0A -> UNDEFINED
         '\ufffe',    //       0x0B -> UNDEFINED
         '\ufffe',    //       0x0C -> UNDEFINED
-        '\r'    ,    //       0x0D -> CARRIAGE RETURN
+        '\n'    ,    //       0x0D -> LINE FEED (RETURN)
         '\u000e',    //       0x0E -> SHIFT OUT
         '\ufffe',    //       0x0F -> UNDEFINED
         '\ufffe',    //       0x10 -> UNDEFINED
@@ -152,7 +152,7 @@ object PetsciiEncoding {
         '\uf113',    //      0x8A -> FUNCTION KEY 4 (CUS)
         '\uf115',    //      0x8B -> FUNCTION KEY 6 (CUS)
         '\uf117',    //      0x8C -> FUNCTION KEY 8 (CUS)
-        '\n'    ,    //       0x8D -> LINE FEED
+        '\r'    ,    //       0x8D -> CARRIAGE RETURN (SHIFT-RETURN)
         '\u000f',    //      0x8E -> SHIFT IN
         '\ufffe',    //       0x8F -> UNDEFINED
         '\uf105',    //       0x90 -> BLACK COLOR SWITCH (CUS)
@@ -283,7 +283,7 @@ object PetsciiEncoding {
         '\ufffe',    //       0x0A -> UNDEFINED
         '\ufffe',    //       0x0B -> UNDEFINED
         '\ufffe',    //       0x0C -> UNDEFINED
-        '\r'    ,    //       0x0D -> CARRIAGE RETURN
+        '\n'    ,    //       0x0D -> LINE FEED (RETURN)
         '\u000e',    //       0x0E -> SHIFT OUT
         '\ufffe',    //       0x0F -> UNDEFINED
         '\ufffe',    //       0x10 -> UNDEFINED
@@ -411,7 +411,7 @@ object PetsciiEncoding {
         '\uf113',    //       0x8A -> FUNCTION KEY 4 (CUS)
         '\uf115',    //       0x8B -> FUNCTION KEY 6 (CUS)
         '\uf117',    //       0x8C -> FUNCTION KEY 8 (CUS)
-        '\n'    ,    //       0x8D -> LINE FEED
+        '\r'    ,    //       0x8D -> CARRIAGE RETURN (SHIFT-RETURN)
         '\u000f',    //       0x8E -> SHIFT IN
         '\ufffe',    //       0x8F -> UNDEFINED
         '\uf105',    //       0x90 -> BLACK COLOR SWITCH (CUS)
@@ -1061,6 +1061,7 @@ object PetsciiEncoding {
             '}' -> '├'
             '|' -> '│'
             '\\' -> '╲'
+            '\r' -> '\n'        // to make \r (carriage returrn) equivalent to \n (line feed): RETURN ($0d)
             else -> chr
         }
 
@@ -1076,7 +1077,10 @@ object PetsciiEncoding {
                 }
                 else -> {
                     val case = if (lowercase) "lower" else "upper"
-                    throw CharConversionException("no ${case}Petscii character for '${chr}' (${chr.code})")
+                    if(chr.isISOControl())
+                        throw CharConversionException("no ${case}Petscii character for char #${chr.code}")
+                    else
+                        throw CharConversionException("no ${case}Petscii character for char #${chr.code} '${chr}'")
                 }
             }
         }
@@ -1119,7 +1123,10 @@ object PetsciiEncoding {
                 }
                 else -> {
                     val case = if (lowercase) "lower" else "upper"
-                    throw CharConversionException("no ${case}Screencode character for '${chr}' (${chr.code})")
+                    if(chr.isISOControl())
+                        throw CharConversionException("no ${case}Screencode character for char #${chr.code}")
+                    else
+                        throw CharConversionException("no ${case}Screencode character for char #${chr.code} '${chr}'")
                 }
             }
         }

codeCore/src/prog8/code/target/cx16/CX16MachineDefinition.kt

@@ -1,6 +1,7 @@
 package prog8.code.target.cx16
 
 import prog8.code.core.*
+import prog8.code.target.C64Target
 import prog8.code.target.cbm.Mflpt5
 import java.nio.file.Path
 
@@ -14,10 +15,6 @@ class CX16MachineDefinition: IMachineDefinition {
     override val FLOAT_MEM_SIZE = Mflpt5.FLOAT_MEM_SIZE
     override val PROGRAM_LOAD_ADDRESS = 0x0801u
 
-    // the 2*128 byte evaluation stack (1 page, on which bytes, words, and even floats are stored during calculations)
-    override var ESTACK_LO = 0x0700u        //  $0700-$077f inclusive
-    override var ESTACK_HI = 0x0780u        //  $0780-$07ff inclusive
-
     override val BSSHIGHRAM_START = 0xa000u     // hiram bank 1, 8Kb, assumed to be active
     override val BSSHIGHRAM_END = 0xc000u       // rom starts here.
 
@@ -43,7 +40,7 @@ class CX16MachineDefinition: IMachineDefinition {
             }
             2 -> {
                 emulator = "box16"
-                extraArgs = listOf("-sym", viceMonListName(programNameWithPath.toString()))
+                extraArgs = listOf("-sym", C64Target.viceMonListName(programNameWithPath.toString()))
             }
             else -> {
                 System.err.println("Cx16 target only supports x16emu and box16 emulators.")
@@ -63,7 +60,7 @@ class CX16MachineDefinition: IMachineDefinition {
 
     override fun initializeMemoryAreas(compilerOptions: CompilationOptions) {
         zeropage = CX16Zeropage(compilerOptions)
-        golden = GoldenRam(compilerOptions, 0x0600u until 0x0800u)
+        golden = GoldenRam(compilerOptions, 0x0400u until 0x0800u)
     }
 
 }

codeCore/src/prog8/code/target/cx16/CX16Zeropage.kt

@@ -40,12 +40,11 @@ class CX16Zeropage(options: CompilationOptions) : Zeropage(options) {
                 ZeropageType.DONTUSE -> {
                     free.clear() // don't use zeropage at all
                 }
-                else -> throw InternalCompilerException("for this machine target, zero page type 'floatsafe' is not available. ${options.zeropage}")
             }
 
-            val distictFree = free.distinct()
+            val distinctFree = free.distinct()
             free.clear()
-            free.addAll(distictFree)
+            free.addAll(distinctFree)
 
             removeReservedFromFreePool()
 

codeCore/src/prog8/code/target/pet/PETMachineDefinition.kt

@@ -0,0 +1,55 @@
+package prog8.code.target.pet
+
+import prog8.code.core.*
+import prog8.code.target.C64Target
+import prog8.code.target.cbm.Mflpt5
+import java.nio.file.Path
+
+
+class PETMachineDefinition: IMachineDefinition {
+
+    override val cpu = CpuType.CPU6502
+
+    override val FLOAT_MAX_POSITIVE = Mflpt5.FLOAT_MAX_POSITIVE
+    override val FLOAT_MAX_NEGATIVE = Mflpt5.FLOAT_MAX_NEGATIVE
+    override val FLOAT_MEM_SIZE = Mflpt5.FLOAT_MEM_SIZE
+    override val PROGRAM_LOAD_ADDRESS = 0x0401u
+
+    override val BSSHIGHRAM_START = 0xffffu
+    override val BSSHIGHRAM_END = 0xffffu
+
+    override lateinit var zeropage: Zeropage
+    override lateinit var golden: GoldenRam
+
+    override fun getFloatAsmBytes(num: Number) = Mflpt5.fromNumber(num).makeFloatFillAsm()
+
+    override fun importLibs(compilerOptions: CompilationOptions, compilationTargetName: String): List<String> {
+        return if (compilerOptions.launcher == CbmPrgLauncherType.BASIC || compilerOptions.output == OutputType.PRG)
+            listOf("syslib")
+        else
+            emptyList()
+    }
+
+    override fun launchEmulator(selectedEmulator: Int, programNameWithPath: Path) {
+        if(selectedEmulator!=1) {
+            System.err.println("The pet target only supports the main emulator (Vice).")
+            return
+        }
+
+        println("\nStarting PET emulator...")
+        val viceMonlist = C64Target.viceMonListName(programNameWithPath.toString())
+        val cmdline = listOf("xpet", "-model", "4032", "-ramsize", "32", "-videosize", "40", "-silent", "-moncommands", viceMonlist,
+                "-autostartprgmode", "1", "-autostart-warp", "-autostart", "${programNameWithPath}.prg")
+        val processb = ProcessBuilder(cmdline).inheritIO()
+        val process=processb.start()
+        process.waitFor()
+    }
+
+    override fun isIOAddress(address: UInt): Boolean = address in 0xe800u..0xe8ffu
+
+    override fun initializeMemoryAreas(compilerOptions: CompilationOptions) {
+        zeropage = PETZeropage(compilerOptions)
+        // there's no golden ram.
+    }
+
+}

codeCore/src/prog8/code/target/pet/PETZeropage.kt

@@ -0,0 +1,58 @@
+package prog8.code.target.pet
+
+import prog8.code.core.CompilationOptions
+import prog8.code.core.InternalCompilerException
+import prog8.code.core.Zeropage
+import prog8.code.core.ZeropageType
+
+
+// reference: http://www.zimmers.net/cbmpics/cbm/PETx/petmem.txt
+
+class PETZeropage(options: CompilationOptions) : Zeropage(options) {
+
+    override val SCRATCH_B1 = 0xb3u      // temp storage for a single byte
+    override val SCRATCH_REG = 0xb4u     // temp storage for a register, must be B1+1
+    override val SCRATCH_W1 = 0xb6u      // temp storage 1 for a word
+    override val SCRATCH_W2 = 0xb8u      // temp storage 2 for a word
+
+    init {
+        if (options.floats) {
+            throw InternalCompilerException("PET target doesn't yet support floating point routines")
+        }
+
+        if (options.floats && options.zeropage !in arrayOf(
+                ZeropageType.FLOATSAFE,
+                ZeropageType.BASICSAFE,
+                ZeropageType.DONTUSE
+            ))
+            throw InternalCompilerException("when floats are enabled, zero page type should be 'floatsafe' or 'basicsafe' or 'dontuse'")
+
+        when (options.zeropage) {
+            ZeropageType.FULL -> {
+                free.addAll(0x00u..0xffu)
+                free.removeAll(listOf(0x8du, 0x8eu, 0x8fu, 0x97u, 0x98u, 0x99u, 0x9au, 0x9bu, 0x9eu, 0xa7u, 0xa8u, 0xa9u, 0xaau))        // these are updated/used by IRQ
+            }
+            ZeropageType.KERNALSAFE -> {
+                free.addAll(0x00u..0xffu)
+                free.removeAll(listOf(0x8du, 0x8eu, 0x8fu, 0x97u, 0x98u, 0x99u, 0x9au, 0x9bu, 0x9eu, 0xa7u, 0xa8u, 0xa9u, 0xaau))        // these are updated/used by IRQ
+            }
+            ZeropageType.FLOATSAFE,
+            ZeropageType.BASICSAFE -> {
+                free.addAll(0xb3u..0xbau)       // TODO more?
+            }
+            ZeropageType.DONTUSE -> {
+                free.clear()  // don't use zeropage at all
+            }
+        }
+
+        val distinctFree = free.distinct()
+        free.clear()
+        free.addAll(distinctFree)
+
+        removeReservedFromFreePool()
+    }
+
+    override fun allocateCx16VirtualRegisters() {
+        TODO("Not known if PET can put the virtual regs in ZP")
+    }
+}

codeCore/src/prog8/code/target/virtual/VirtualMachineDefinition.kt

@@ -15,8 +15,6 @@ class VirtualMachineDefinition: IMachineDefinition {
     override val FLOAT_MEM_SIZE = 4             // 32-bits floating point
     override val PROGRAM_LOAD_ADDRESS = 0u      // not actually used
 
-    override var ESTACK_LO = 0u                 // not actually used
-    override var ESTACK_HI = 0u                 // not actually used
     override val BSSHIGHRAM_START = 0u          // not actually used
     override val BSSHIGHRAM_END = 0u            // not actually used
     override lateinit var zeropage: Zeropage    // not actually used

codeGenCpu6502/build.gradle

@@ -28,18 +28,18 @@ dependencies {
     implementation project(':codeCore')
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16"
+    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18"
 
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.3.2'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.6.2'
 }
 
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
     test {

codeGenCpu6502/src/prog8/codegen/cpu6502/AsmOptimizer.kt

@@ -15,21 +15,7 @@ internal fun optimizeAssembly(lines: MutableList<String>, machine: IMachineDefin
 
     var linesByFour = getLinesBy(lines, 4)
 
-    var mods = optimizeUselessStackByteWrites(linesByFour)
-    if(mods.isNotEmpty()) {
-        apply(mods, lines)
-        linesByFour = getLinesBy(lines, 4)
-        numberOfOptimizations++
-    }
-
-    mods = optimizeIncDec(linesByFour)
-    if(mods.isNotEmpty()) {
-        apply(mods, lines)
-        linesByFour = getLinesBy(lines, 4)
-        numberOfOptimizations++
-    }
-
-    mods = optimizeCmpSequence(linesByFour)
+    var mods = optimizeIncDec(linesByFour)
     if(mods.isNotEmpty()) {
         apply(mods, lines)
         linesByFour = getLinesBy(lines, 4)
@@ -43,7 +29,21 @@ internal fun optimizeAssembly(lines: MutableList<String>, machine: IMachineDefin
         numberOfOptimizations++
     }
 
-    mods= optimizeJsrRtsAndOtherCombinations(linesByFour)
+    mods = optimizeJsrRtsAndOtherCombinations(linesByFour)
+    if(mods.isNotEmpty()) {
+        apply(mods, lines)
+        linesByFour = getLinesBy(lines, 4)
+        numberOfOptimizations++
+    }
+
+    mods = optimizeUselessPushPopStack(linesByFour)
+    if(mods.isNotEmpty()) {
+        apply(mods, lines)
+        linesByFour = getLinesBy(lines, 4)
+        numberOfOptimizations++
+    }
+
+    mods = optimizeUnneededTempvarInAdd(linesByFour)
     if(mods.isNotEmpty()) {
         apply(mods, lines)
         linesByFour = getLinesBy(lines, 4)
@@ -90,44 +90,6 @@ private fun getLinesBy(lines: MutableList<String>, windowSize: Int) =
 // all lines (that aren't empty or comments) in sliding windows of certain size
         lines.withIndex().filter { it.value.isNotBlank() && !it.value.trimStart().startsWith(';') }.windowed(windowSize, partialWindows = false)
 
-private fun optimizeCmpSequence(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
-    // when statement (on bytes) generates a sequence of:
-    //	 lda  $ce01,x
-    //	 cmp  #$20
-    //	 beq  check_prog8_s72choice_32
-    //	 lda  $ce01,x
-    //	 cmp  #$21
-    //	 beq  check_prog8_s73choice_33
-    // the repeated lda can be removed
-    val mods = mutableListOf<Modification>()
-    for(lines in linesByFour) {
-        if(lines[0].value.trim()=="lda  P8ESTACK_LO+1,x" &&
-                lines[1].value.trim().startsWith("cmp ") &&
-                lines[2].value.trim().startsWith("beq ") &&
-                lines[3].value.trim()=="lda  P8ESTACK_LO+1,x") {
-            mods.add(Modification(lines[3].index, true, null)) // remove the second lda
-        }
-    }
-    return mods
-}
-
-private fun optimizeUselessStackByteWrites(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
-    // sta on stack, dex, inx, lda from stack -> eliminate this useless stack byte write
-    // this is a lot harder for word values because the instruction sequence varies.
-    val mods = mutableListOf<Modification>()
-    for(lines in linesByFour) {
-        if(lines[0].value.trim()=="sta  P8ESTACK_LO,x" &&
-                lines[1].value.trim()=="dex" &&
-                lines[2].value.trim()=="inx" &&
-                lines[3].value.trim()=="lda  P8ESTACK_LO,x") {
-            mods.add(Modification(lines[1].index, true, null))
-            mods.add(Modification(lines[2].index, true, null))
-            mods.add(Modification(lines[3].index, true, null))
-        }
-    }
-    return mods
-}
-
 private fun optimizeSameAssignments(
     linesByFourteen: List<List<IndexedValue<String>>>,
     machine: IMachineDefinition,
@@ -311,20 +273,9 @@ private fun optimizeSameAssignments(
         /*
         sta  A   ; or stz     double store, remove this first one
         sta  A   ; or stz
-         */
-        if(!overlappingMods && first.isStoreRegOrZero() && second.isStoreRegOrZero()) {
-            if(first[2]==second[2]) {
-                val firstvalue = first.substring(4)
-                val secondvalue = second.substring(4)
-                if(firstvalue==secondvalue) {
-                    val address = getAddressArg(first, symbolTable)
-                    if(address==null || !machine.isIOAddress(address)) {
-                        overlappingMods = true
-                        mods.add(Modification(lines[0].index, true, null))
-                    }
-                }
-            }
-        }
+        However, this cannot be done relyably because 'A' could be a constant symbol referring to an I/O address.
+        We can't see that here and would otherwise delete valid double stores.
+        */
     }
 
     return mods
@@ -528,3 +479,63 @@ private fun optimizeJsrRtsAndOtherCombinations(linesByFour: List<List<IndexedVal
     }
     return mods
 }
+
+private fun optimizeUselessPushPopStack(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
+    val mods = mutableListOf<Modification>()
+
+    fun optimize(register: Char, lines: List<IndexedValue<String>>) {
+        if(lines[0].value.trimStart().startsWith("ph$register")) {
+            if(lines[2].value.trimStart().startsWith("pl$register")) {
+                val second = lines[1].value.trimStart().take(6).lowercase()
+                if(register!in second
+                    && !second.startsWith("jsr")
+                    && !second.startsWith("pl")
+                    && !second.startsWith("ph")) {
+                    mods.add(Modification(lines[0].index, true, null))
+                    mods.add(Modification(lines[2].index, true, null))
+                }
+            }
+            else if (lines[3].value.trimStart().startsWith("pl$register")) {
+                val second = lines[1].value.trimStart().take(6).lowercase()
+                val third = lines[2].value.trimStart().take(6).lowercase()
+                if(register !in second && register !in third
+                    && !second.startsWith("jsr") && !third.startsWith("jsr")
+                    && !second.startsWith("pl") && !third.startsWith("pl")
+                    && !second.startsWith("ph") && !third.startsWith("ph")) {
+                    mods.add(Modification(lines[0].index, true, null))
+                    mods.add(Modification(lines[3].index, true, null))
+                }
+            }
+        }
+    }
+
+    for (lines in linesByFour) {
+        optimize('a', lines)
+        optimize('x', lines)
+        optimize('y', lines)
+    }
+    return mods
+}
+
+private fun optimizeUnneededTempvarInAdd(linesByFour: List<List<IndexedValue<String>>>): List<Modification> {
+    // sequence:  sta  P8ZP_SCRATCH_XX  / lda  something / clc / adc  P8ZP_SCRATCH_XX
+    // this can be performed without the scratch variable:  clc  /  adc  something
+    val mods = mutableListOf<Modification>()
+
+    for(lines in linesByFour) {
+        val first = lines[0].value.trimStart()
+        val second = lines[1].value.trimStart()
+        val third = lines[2].value.trimStart()
+        val fourth = lines[3].value.trimStart()
+        if(first.startsWith("sta  P8ZP_SCRATCH_") && second.startsWith("lda") && third.startsWith("clc") && fourth.startsWith("adc  P8ZP_SCRATCH_") ) {
+            if(fourth.substring(4)==first.substring(4)) {
+                mods.add(Modification(lines[0].index, false, "    clc"))
+                mods.add(Modification(lines[1].index, false, "    adc  ${second.substring(3).trimStart()}"))
+                mods.add(Modification(lines[2].index, true, null))
+                mods.add(Modification(lines[3].index, true, null))
+            }
+        }
+    }
+
+    return mods
+}

codeGenCpu6502/src/prog8/codegen/cpu6502/AssemblyProgram.kt

@@ -1,13 +1,8 @@
 package prog8.codegen.cpu6502
 
-import com.github.michaelbull.result.Ok
-import com.github.michaelbull.result.Result
-import com.github.michaelbull.result.mapError
 import prog8.code.core.*
-import java.io.File
+import prog8.code.target.C64Target
 import java.nio.file.Path
-import kotlin.io.path.Path
-import kotlin.io.path.isRegularFile
 
 
 internal class AssemblyProgram(
@@ -19,23 +14,28 @@ internal class AssemblyProgram(
     private val prgFile = outputDir.resolve("$name.prg")        // CBM prg executable program
     private val xexFile = outputDir.resolve("$name.xex")        // Atari xex executable program
     private val binFile = outputDir.resolve("$name.bin")
-    private val viceMonListFile = outputDir.resolve(viceMonListName(name))
+    private val viceMonListFile = outputDir.resolve(C64Target.viceMonListName(name))
     private val listFile = outputDir.resolve("$name.list")
 
-    override fun assemble(options: CompilationOptions): Boolean {
+    override fun assemble(options: CompilationOptions, errors: IErrorReporter): Boolean {
 
         val assemblerCommand: List<String>
 
         when (compTarget.name) {
-            in setOf("c64", "c128", "cx16") -> {
+            in setOf("c64", "c128", "cx16", "pet32") -> {
                 // CBM machines .prg generation.
 
                 // add "-Wlong-branch"  to see warnings about conversion of branch instructions to jumps (default = do this silently)
                 val command = mutableListOf("64tass", "--ascii", "--case-sensitive", "--long-branch",
-                    "-Wall", "-Wno-strict-bool", "-Wno-shadow", // "-Werror",
+                    "-Wall", // "-Wno-strict-bool", "-Werror",
                     "--dump-labels", "--vice-labels", "--labels=$viceMonListFile", "--no-monitor"
                 )
 
+                if(options.warnSymbolShadowing)
+                    command.add("-Wshadow")
+                else
+                    command.add("-Wno-shadow")
+
                 if(options.asmQuiet)
                     command.add("--quiet")
 
@@ -64,10 +64,15 @@ internal class AssemblyProgram(
 
                 // TODO are these options okay for atari?
                 val command = mutableListOf("64tass", "--ascii", "--case-sensitive", "--long-branch",
-                    "-Wall", "-Wno-strict-bool", "-Wno-shadow", // "-Werror",
+                    "-Wall", // "-Werror", "-Wno-strict-bool"
                     "--no-monitor"
                 )
 
+                if(options.warnSymbolShadowing)
+                    command.add("-Wshadow")
+                else
+                    command.add("-Wno-shadow")
+
                 if(options.asmQuiet)
                     command.add("--quiet")
 
@@ -129,18 +134,3 @@ internal class AssemblyProgram(
         viceMonListFile.toFile().appendText(breakpoints.joinToString("\n") + "\n")
     }
 }
-
-
-internal fun loadAsmIncludeFile(filename: String, source: SourceCode): Result<String, NoSuchFileException> {
-    return if (filename.startsWith(SourceCode.libraryFilePrefix)) {
-        return com.github.michaelbull.result.runCatching {
-            SourceCode.Resource("/prog8lib/${filename.substring(SourceCode.libraryFilePrefix.length)}").text
-        }.mapError { NoSuchFileException(File(filename)) }
-    } else {
-        val sib = Path(source.origin).resolveSibling(filename)
-        if (sib.isRegularFile())
-            Ok(SourceCode.File(sib).text)
-        else
-            Ok(SourceCode.File(Path(filename)).text)
-    }
-}

codeGenCpu6502/src/prog8/codegen/cpu6502/ExpressionsAsmGen.kt

@@ -1,907 +0,0 @@
-package prog8.codegen.cpu6502
-
-import prog8.code.ast.*
-import prog8.code.core.*
-import kotlin.math.absoluteValue
-
-internal class ExpressionsAsmGen(private val program: PtProgram,
-                                 private val asmgen: AsmGen6502Internal,
-                                 private val allocator: VariableAllocator) {
-
-    @Deprecated("avoid calling this as it generates slow evalstack based code")
-    internal fun translateExpression(expression: PtExpression) {
-        if (this.asmgen.options.slowCodegenWarnings) {
-            asmgen.errors.warn("slow stack evaluation used for expression", expression.position)
-        }
-        translateExpressionInternal(expression)
-    }
-
-
-    // the rest of the methods are all PRIVATE
-
-
-    private fun translateExpressionInternal(expression: PtExpression) {
-
-        when(expression) {
-            is PtPrefix -> translateExpression(expression)
-            is PtBinaryExpression -> translateExpression(expression)
-            is PtArrayIndexer -> translateExpression(expression)
-            is PtTypeCast -> translateExpression(expression)
-            is PtAddressOf -> translateExpression(expression)
-            is PtMemoryByte -> asmgen.translateDirectMemReadExpressionToRegAorStack(expression, true)
-            is PtNumber -> translateExpression(expression)
-            is PtIdentifier -> translateExpression(expression)
-            is PtFunctionCall -> translateFunctionCallResultOntoStack(expression)
-            is PtBuiltinFunctionCall -> asmgen.translateBuiltinFunctionCallExpression(expression, true, null)
-            is PtContainmentCheck -> throw AssemblyError("containment check as complex expression value is not supported")
-            is PtArray, is PtString -> throw AssemblyError("no asm gen for string/array literal value assignment - should have been replaced by a variable")
-            is PtRange -> throw AssemblyError("range expression should have been changed into array values")
-            is PtMachineRegister -> throw AssemblyError("machine register ast node should not occur in 6502 codegen it is for IR code")
-            else -> TODO("missing expression asmgen for $expression")
-        }
-    }
-
-    private fun translateFunctionCallResultOntoStack(call: PtFunctionCall) {
-        // only for use in nested expression evaluation
-
-        val symbol = asmgen.symbolTable.lookup(call.name)
-        val sub = symbol!!.astNode as IPtSubroutine
-        asmgen.saveXbeforeCall(call)
-        asmgen.translateFunctionCall(call)
-        if(sub.regXasResult()) {
-            // store the return value in X somewhere that we can access again below
-            asmgen.out("  stx  P8ZP_SCRATCH_REG")
-        }
-        asmgen.restoreXafterCall(call)
-
-        val returns: List<Pair<RegisterOrStatusflag, DataType>> = sub.returnsWhatWhere()
-        for ((reg, _) in returns) {
-            // result value is in cpu or status registers, put it on the stack instead (as we're evaluating an expression tree)
-            if (reg.registerOrPair != null) {
-                when (reg.registerOrPair!!) {
-                    RegisterOrPair.A -> asmgen.out("  sta  P8ESTACK_LO,x |  dex")
-                    RegisterOrPair.Y -> asmgen.out("  tya |  sta  P8ESTACK_LO,x |  dex")
-                    RegisterOrPair.AY -> asmgen.out("  sta  P8ESTACK_LO,x |  tya |  sta  P8ESTACK_HI,x |  dex")
-                    RegisterOrPair.X -> asmgen.out("  lda  P8ZP_SCRATCH_REG |  sta  P8ESTACK_LO,x |  dex")
-                    RegisterOrPair.AX -> asmgen.out("  sta  P8ESTACK_LO,x |  lda  P8ZP_SCRATCH_REG |  sta  P8ESTACK_HI,x |  dex")
-                    RegisterOrPair.XY -> asmgen.out("  tya |  sta  P8ESTACK_HI,x |  lda  P8ZP_SCRATCH_REG |  sta  P8ESTACK_LO,x |  dex")
-                    RegisterOrPair.FAC1 -> asmgen.out("  jsr  floats.push_fac1")
-                    RegisterOrPair.FAC2 -> asmgen.out("  jsr  floats.push_fac2")
-                    RegisterOrPair.R0,
-                    RegisterOrPair.R1,
-                    RegisterOrPair.R2,
-                    RegisterOrPair.R3,
-                    RegisterOrPair.R4,
-                    RegisterOrPair.R5,
-                    RegisterOrPair.R6,
-                    RegisterOrPair.R7,
-                    RegisterOrPair.R8,
-                    RegisterOrPair.R9,
-                    RegisterOrPair.R10,
-                    RegisterOrPair.R11,
-                    RegisterOrPair.R12,
-                    RegisterOrPair.R13,
-                    RegisterOrPair.R14,
-                    RegisterOrPair.R15 -> {
-                        asmgen.out(
-                            """
-                            lda  cx16.${reg.registerOrPair.toString().lowercase()}
-                            sta  P8ESTACK_LO,x
-                            lda  cx16.${reg.registerOrPair.toString().lowercase()}+1
-                            sta  P8ESTACK_HI,x
-                            dex
-                        """)
-                    }
-                }
-            } else when(reg.statusflag) {
-                Statusflag.Pc -> {
-                    asmgen.out("""
-                        lda  #0
-                        rol  a
-                        sta  P8ESTACK_LO,x
-                        dex""")
-                }
-                Statusflag.Pz -> {
-                    asmgen.out("""
-                        beq  +
-                        lda  #0
-                        beq  ++
-+                           lda  #1
-+                           sta  P8ESTACK_LO,x
-                        dex""")
-                }
-                Statusflag.Pv -> {
-                    asmgen.out("""
-                        bvs  +
-                        lda  #0
-                        beq  ++
-+                           lda  #1
-+                           sta  P8ESTACK_LO,x
-                        dex""")
-                }
-                Statusflag.Pn -> {
-                    asmgen.out("""
-                        bmi  +
-                        lda  #0
-                        beq  ++
-+                           lda  #1
-+                           sta  P8ESTACK_LO,x
-                        dex""")
-                }
-                null -> {}
-            }
-        }
-    }
-
-    private fun translateExpression(typecast: PtTypeCast) {
-        translateExpressionInternal(typecast.value)
-        when(typecast.value.type) {
-            DataType.UBYTE, DataType.BOOL -> {
-                when(typecast.type) {
-                    DataType.UBYTE, DataType.BYTE -> {}
-                    DataType.UWORD, DataType.WORD -> {
-                        if(asmgen.isTargetCpu(CpuType.CPU65c02))
-                            asmgen.out("  stz  P8ESTACK_HI+1,x")
-                        else
-                            asmgen.out("  lda  #0  |  sta  P8ESTACK_HI+1,x")
-                    }
-                    DataType.FLOAT -> asmgen.out(" jsr  floats.stack_ub2float")
-                    in PassByReferenceDatatypes -> throw AssemblyError("cannot cast to a pass-by-reference datatype")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            DataType.BYTE -> {
-                when(typecast.type) {
-                    DataType.UBYTE, DataType.BYTE -> {}
-                    DataType.UWORD, DataType.WORD -> asmgen.signExtendStackLsb(DataType.BYTE)
-                    DataType.FLOAT -> asmgen.out(" jsr  floats.stack_b2float")
-                    in PassByReferenceDatatypes -> throw AssemblyError("cannot cast to a pass-by-reference datatype")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            DataType.UWORD -> {
-                when(typecast.type) {
-                    DataType.BYTE, DataType.UBYTE -> {}
-                    DataType.WORD, DataType.UWORD -> {}
-                    DataType.FLOAT -> asmgen.out(" jsr  floats.stack_uw2float")
-                    in PassByReferenceDatatypes -> throw AssemblyError("cannot cast to a pass-by-reference datatype")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            DataType.WORD -> {
-                when(typecast.type) {
-                    DataType.BYTE, DataType.UBYTE -> {}
-                    DataType.WORD, DataType.UWORD -> {}
-                    DataType.FLOAT -> asmgen.out(" jsr  floats.stack_w2float")
-                    in PassByReferenceDatatypes -> throw AssemblyError("cannot cast to a pass-by-reference datatype")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            DataType.FLOAT -> {
-                when(typecast.type) {
-                    DataType.UBYTE -> asmgen.out(" jsr  floats.stack_float2uw")
-                    DataType.BYTE -> asmgen.out(" jsr  floats.stack_float2w")
-                    DataType.UWORD -> asmgen.out(" jsr  floats.stack_float2uw")
-                    DataType.WORD -> asmgen.out(" jsr  floats.stack_float2w")
-                    DataType.FLOAT -> {}
-                    in PassByReferenceDatatypes -> throw AssemblyError("cannot cast to a pass-by-reference datatype")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            DataType.STR -> {
-                if (typecast.type != DataType.UWORD && typecast.type == DataType.STR)
-                    throw AssemblyError("cannot typecast a string into another incompatitble type")
-            }
-            in PassByReferenceDatatypes -> throw AssemblyError("cannot cast pass-by-reference value into another type")
-            else -> throw AssemblyError("weird type")
-        }
-    }
-
-    private fun translateExpression(expr: PtAddressOf) {
-        val name = asmgen.asmVariableName(expr.identifier)
-        asmgen.out("  lda  #<$name |  sta  P8ESTACK_LO,x |  lda  #>$name  |  sta  P8ESTACK_HI,x  | dex")
-    }
-
-    private fun translateExpression(expr: PtNumber) {
-        when(expr.type) {
-            DataType.UBYTE, DataType.BYTE -> asmgen.out(" lda  #${expr.number.toHex()}  | sta  P8ESTACK_LO,x  | dex")
-            DataType.UWORD, DataType.WORD -> asmgen.out("""
-                lda  #<${expr.number.toHex()}
-                sta  P8ESTACK_LO,x
-                lda  #>${expr.number.toHex()}
-                sta  P8ESTACK_HI,x
-                dex
-            """)
-            DataType.FLOAT -> {
-                val floatConst = allocator.getFloatAsmConst(expr.number)
-                asmgen.out(" lda  #<$floatConst |  ldy  #>$floatConst |  jsr  floats.push_float")
-            }
-            else -> throw AssemblyError("weird type")
-        }
-    }
-
-    private fun translateExpression(expr: PtIdentifier) {
-        val varname = asmgen.asmVariableName(expr)
-        when(expr.type) {
-            DataType.UBYTE, DataType.BYTE -> {
-                asmgen.out("  lda  $varname  |  sta  P8ESTACK_LO,x  |  dex")
-            }
-            DataType.UWORD, DataType.WORD -> {
-                asmgen.out("  lda  $varname  |  sta  P8ESTACK_LO,x  |  lda  $varname+1 |  sta  P8ESTACK_HI,x |  dex")
-            }
-            DataType.FLOAT -> {
-                asmgen.out(" lda  #<$varname |  ldy  #>$varname|  jsr  floats.push_float")
-            }
-            in IterableDatatypes -> {
-                asmgen.out("  lda  #<$varname  |  sta  P8ESTACK_LO,x  |  lda  #>$varname |  sta  P8ESTACK_HI,x |  dex")
-            }
-            else -> throw AssemblyError("stack push weird variable type $expr")
-        }
-    }
-
-    private fun translateExpression(expr: PtBinaryExpression) {
-        // Uses evalstack to evaluate the given expression.  THIS IS SLOW AND SHOULD BE AVOIDED!
-        val leftDt = expr.left.type
-        val rightDt = expr.right.type
-        // see if we can apply some optimized routines still
-        when(expr.operator) {
-            "+" -> {
-                if(leftDt in IntegerDatatypes && rightDt in IntegerDatatypes) {
-                    val leftVal = expr.left.asConstInteger()
-                    val rightVal = expr.right.asConstInteger()
-                    if (leftVal!=null && leftVal in -4..4) {
-                        translateExpressionInternal(expr.right)
-                        if(rightDt in ByteDatatypes) {
-                            val incdec = if(leftVal<0) "dec" else "inc"
-                            repeat(leftVal.absoluteValue) {
-                                asmgen.out("  $incdec  P8ESTACK_LO+1,x")
-                            }
-                        } else {
-                            // word
-                            if(leftVal<0) {
-                                repeat(leftVal.absoluteValue) {
-                                    asmgen.out("""
-                                        lda  P8ESTACK_LO+1,x
-                                        bne  +
-                                        dec  P8ESTACK_HI+1,x
-+                                       dec  P8ESTACK_LO+1,x""")
-                                }
-                            } else {
-                                repeat(leftVal) {
-                                    asmgen.out("""
-                                        inc  P8ESTACK_LO+1,x
-                                        bne  +
-                                        inc  P8ESTACK_HI+1,x
-+""")
-                                }
-                            }
-                        }
-                        return
-                    }
-                    else if (rightVal!=null && rightVal in -4..4)
-                    {
-                        translateExpressionInternal(expr.left)
-                        if(leftDt in ByteDatatypes) {
-                            val incdec = if(rightVal<0) "dec" else "inc"
-                            repeat(rightVal.absoluteValue) {
-                                asmgen.out("  $incdec  P8ESTACK_LO+1,x")
-                            }
-                        } else {
-                            // word
-                            if(rightVal<0) {
-                                repeat(rightVal.absoluteValue) {
-                                    asmgen.out("""
-                                        lda  P8ESTACK_LO+1,x
-                                        bne  +
-                                        dec  P8ESTACK_HI+1,x
-+                                       dec  P8ESTACK_LO+1,x""")
-                                }
-                            } else {
-                                repeat(rightVal) {
-                                    asmgen.out("""
-                                        inc  P8ESTACK_LO+1,x
-                                        bne  +
-                                        inc  P8ESTACK_HI+1,x
-+""")
-                                }
-                            }
-                        }
-                        return
-                    }
-                }
-            }
-            "-" -> {
-                if(leftDt in IntegerDatatypes && rightDt in IntegerDatatypes) {
-                    val rightVal = expr.right.asConstInteger()
-                    if (rightVal!=null && rightVal in -4..4)
-                    {
-                        translateExpressionInternal(expr.left)
-                        if(leftDt in ByteDatatypes) {
-                            val incdec = if(rightVal<0) "inc" else "dec"
-                            repeat(rightVal.absoluteValue) {
-                                asmgen.out("  $incdec  P8ESTACK_LO+1,x")
-                            }
-                        } else {
-                            // word
-                            if(rightVal>0) {
-                                repeat(rightVal.absoluteValue) {
-                                    asmgen.out("""
-                                        lda  P8ESTACK_LO+1,x
-                                        bne  +
-                                        dec  P8ESTACK_HI+1,x
-+                                       dec  P8ESTACK_LO+1,x""")
-                                }
-                            } else {
-                                repeat(rightVal) {
-                                    asmgen.out("""
-                                        inc  P8ESTACK_LO+1,x
-                                        bne  +
-                                        inc  P8ESTACK_HI+1,x
-+""")
-                                }
-                            }
-                        }
-                        return
-                    }
-                }
-            }
-            ">>" -> {
-                val amount = expr.right.asConstInteger()
-                if(amount!=null) {
-                    translateExpressionInternal(expr.left)
-                    when (leftDt) {
-                        DataType.UBYTE -> {
-                            if (amount <= 2)
-                                repeat(amount) { asmgen.out(" lsr  P8ESTACK_LO+1,x") }
-                            else {
-                                asmgen.out(" lda  P8ESTACK_LO+1,x")
-                                repeat(amount) { asmgen.out(" lsr  a") }
-                                asmgen.out(" sta  P8ESTACK_LO+1,x")
-                            }
-                        }
-                        DataType.BYTE -> {
-                            if (amount <= 2)
-                                repeat(amount) { asmgen.out(" lda  P8ESTACK_LO+1,x |  asl  a |  ror  P8ESTACK_LO+1,x") }
-                            else {
-                                asmgen.out(" lda  P8ESTACK_LO+1,x |  sta  P8ZP_SCRATCH_B1")
-                                repeat(amount) { asmgen.out(" asl  a |  ror  P8ZP_SCRATCH_B1 |  lda  P8ZP_SCRATCH_B1") }
-                                asmgen.out(" sta  P8ESTACK_LO+1,x")
-                            }
-                        }
-                        DataType.UWORD -> {
-                            if(amount>=16) {
-                                if(asmgen.isTargetCpu(CpuType.CPU65c02))
-                                    asmgen.out("  stz  P8ESTACK_LO+1,x |  stz  P8ESTACK_HI+1,x")
-                                else
-                                    asmgen.out("  lda  #0 |  sta  P8ESTACK_LO+1,x |  sta  P8ESTACK_HI+1,x")
-                                return
-                            }
-                            var left = amount
-                            while (left >= 7) {
-                                asmgen.out(" jsr  math.shift_right_uw_7")
-                                left -= 7
-                            }
-                            if (left in 0..2)
-                                repeat(left) { asmgen.out(" lsr  P8ESTACK_HI+1,x |  ror  P8ESTACK_LO+1,x") }
-                            else
-                                asmgen.out(" jsr  math.shift_right_uw_$left")
-                        }
-                        DataType.WORD -> {
-                            if(amount>=16) {
-                                asmgen.out("""
-                                    lda  P8ESTACK_HI+1,x
-                                    bmi  +
-                                    lda  #0
-                                    sta  P8ESTACK_LO+1,x
-                                    sta  P8ESTACK_HI+1,x
-                                    beq  ++
-+                                   lda  #255
-                                    sta  P8ESTACK_LO+1,x
-                                    sta  P8ESTACK_HI+1,x
-+""")
-                                return
-                            }
-                            var left = amount
-                            while (left >= 7) {
-                                asmgen.out(" jsr  math.shift_right_w_7")
-                                left -= 7
-                            }
-                            if (left in 0..2)
-                                repeat(left) { asmgen.out(" lda  P8ESTACK_HI+1,x |  asl  a  |  ror  P8ESTACK_HI+1,x |  ror  P8ESTACK_LO+1,x") }
-                            else
-                                asmgen.out(" jsr  math.shift_right_w_$left")
-                        }
-                        else -> throw AssemblyError("weird type")
-                    }
-                    return
-                }
-            }
-            "<<" -> {
-                val amount = expr.right.asConstInteger()
-                if(amount!=null) {
-                    translateExpressionInternal(expr.left)
-                    if (leftDt in ByteDatatypes) {
-                        if (amount <= 2)
-                            repeat(amount) { asmgen.out("  asl  P8ESTACK_LO+1,x") }
-                        else {
-                            asmgen.out("  lda  P8ESTACK_LO+1,x")
-                            repeat(amount) { asmgen.out("  asl  a") }
-                            asmgen.out("  sta  P8ESTACK_LO+1,x")
-                        }
-                    } else {
-                        var left = amount
-                        while (left >= 7) {
-                            asmgen.out(" jsr  math.shift_left_w_7")
-                            left -= 7
-                        }
-                        if (left in 0..2)
-                            repeat(left) { asmgen.out("  asl  P8ESTACK_LO+1,x |  rol  P8ESTACK_HI+1,x") }
-                        else
-                            asmgen.out(" jsr  math.shift_left_w_$left")
-                    }
-                    return
-                }
-            }
-            "*" -> {
-                if(leftDt in IntegerDatatypes && rightDt in IntegerDatatypes) {
-                    val leftVar = expr.left as? PtIdentifier
-                    val rightVar = expr.right as? PtIdentifier
-                    if(leftVar!=null && rightVar!=null && leftVar==rightVar)
-                        return translateSquared(leftVar, leftDt)
-                }
-
-                val value = expr.right as? PtNumber
-                if(value!=null) {
-                    if(rightDt in IntegerDatatypes) {
-                        val amount = value.number.toInt()
-                        if(amount==2) {
-                            // optimize x*2 common case
-                            translateExpressionInternal(expr.left)
-                            if(leftDt in ByteDatatypes) {
-                                asmgen.out("  asl  P8ESTACK_LO+1,x")
-                            } else {
-                                asmgen.out("  asl  P8ESTACK_LO+1,x |  rol  P8ESTACK_HI+1,x")
-                            }
-                            return
-                        }
-                        when(rightDt) {
-                            DataType.UBYTE -> {
-                                if(amount in asmgen.optimizedByteMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  math.stack_mul_byte_$amount")
-                                    return
-                                }
-                            }
-                            DataType.BYTE -> {
-                                if(amount in asmgen.optimizedByteMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  math.stack_mul_byte_$amount")
-                                    return
-                                }
-                                if(amount.absoluteValue in asmgen.optimizedByteMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  prog8_lib.neg_b |  jsr  math.stack_mul_byte_${amount.absoluteValue}")
-                                    return
-                                }
-                            }
-                            DataType.UWORD -> {
-                                if(amount in asmgen.optimizedWordMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  math.stack_mul_word_$amount")
-                                    return
-                                }
-                            }
-                            DataType.WORD -> {
-                                if(amount in asmgen.optimizedWordMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  math.stack_mul_word_$amount")
-                                    return
-                                }
-                                if(amount.absoluteValue in asmgen.optimizedWordMultiplications) {
-                                    translateExpressionInternal(expr.left)
-                                    asmgen.out(" jsr  prog8_lib.neg_w |  jsr  math.stack_mul_word_${amount.absoluteValue}")
-                                    return
-                                }
-                            }
-                            else -> {}
-                        }
-                    }
-                }
-            }
-            "/" -> {
-                if(leftDt in IntegerDatatypes && rightDt in IntegerDatatypes) {
-                    val rightVal = expr.right.asConstInteger()
-                    if(rightVal!=null && rightVal==2) {
-                        translateExpressionInternal(expr.left)
-                        when (leftDt) {
-                            DataType.UBYTE -> {
-                                asmgen.out("  lsr  P8ESTACK_LO+1,x")
-                            }
-                            DataType.UWORD -> {
-                                asmgen.out("  lsr  P8ESTACK_HI+1,x |  ror  P8ESTACK_LO+1,x")
-                            }
-                            DataType.BYTE -> {
-                                // signed divide using shift needs adjusting of negative value to get correct rounding towards zero
-                                asmgen.out("""
-                                    lda  P8ESTACK_LO+1,x
-                                    bpl  +
-                                    inc  P8ESTACK_LO+1,x
-                                    lda  P8ESTACK_LO+1,x
-+                                   asl  a
-                                    ror  P8ESTACK_LO+1,x""")
-                            }
-                            DataType.WORD -> {
-                                // signed divide using shift needs adjusting of negative value to get correct rounding towards zero
-                                asmgen.out("""
-                                    lda  P8ESTACK_HI+1,x
-                                    bpl  ++
-                                    inc  P8ESTACK_LO+1,x
-                                    bne  +
-                                    inc  P8ESTACK_HI+1,x
-+                                   lda  P8ESTACK_HI+1,x
-+                                   asl  a
-                                    ror  P8ESTACK_HI+1,x
-                                    ror  P8ESTACK_LO+1,x""")
-                            }
-                            else -> throw AssemblyError("weird dt")
-                        }
-                        return
-                    }
-                }
-            }
-            in ComparisonOperators -> {
-                if(leftDt in NumericDatatypes && rightDt in NumericDatatypes) {
-                    val rightVal = expr.right.asConstInteger()
-                    if(rightVal==0)
-                        return translateComparisonWithZero(expr.left, leftDt, expr.operator)
-                }
-            }
-        }
-
-        if((leftDt in ByteDatatypes && rightDt !in ByteDatatypes)
-                || (leftDt in WordDatatypes && rightDt !in WordDatatypes))
-            throw AssemblyError("binary operator ${expr.operator} left/right dt not identical")
-
-        if(leftDt==DataType.STR && rightDt==DataType.STR && expr.operator in ComparisonOperators) {
-            translateCompareStrings(expr.left, expr.operator, expr.right)
-        }
-        else {
-            // the general, non-optimized cases
-            // TODO optimize more cases.... (or one day just don't use the evalstack at all anymore)
-            translateExpressionInternal(expr.left)
-            translateExpressionInternal(expr.right)
-            when (leftDt) {
-                in ByteDatatypes -> translateBinaryOperatorBytes(expr.operator, leftDt)
-                in WordDatatypes -> translateBinaryOperatorWords(expr.operator, leftDt)
-                DataType.FLOAT -> translateBinaryOperatorFloats(expr.operator)
-                else -> throw AssemblyError("non-numerical datatype")
-            }
-        }
-    }
-
-    private fun translateComparisonWithZero(expr: PtExpression, dt: DataType, operator: String) {
-        if(expr.isSimple()) {
-            if(operator=="!=") {
-                when (dt) {
-                    in ByteDatatypes -> {
-                        asmgen.assignExpressionToRegister(expr, RegisterOrPair.A, dt == DataType.BYTE)
-                        asmgen.out("""
-                            beq  +
-                            lda  #1
-+                           sta  P8ESTACK_LO,x
-                            dex""")
-                        return
-                    }
-                    in WordDatatypes -> {
-                        asmgen.assignExpressionToRegister(expr, RegisterOrPair.AY, dt == DataType.WORD)
-                        asmgen.out("""
-                            sty  P8ZP_SCRATCH_B1 
-                            ora  P8ZP_SCRATCH_B1
-                            beq  +
-                            lda  #1
-+                           sta  P8ESTACK_LO,x
-                            dex""")
-                        return
-                    }
-                    DataType.FLOAT -> {
-                        asmgen.assignExpressionToRegister(expr, RegisterOrPair.FAC1, true)
-                        asmgen.out("""
-                            jsr  floats.SIGN
-                            sta  P8ESTACK_LO,x
-                            dex""")
-                        return
-                    }
-                    else -> {}
-                }
-            }
-            /* operator == is not worth it to special case, the code is mostly larger */
-        }
-        translateExpressionInternal(expr)
-        when(operator) {
-            "==" -> {
-                when(dt) {
-                    DataType.UBYTE, DataType.BYTE -> asmgen.out("  jsr  prog8_lib.equalzero_b")
-                    DataType.UWORD, DataType.WORD -> asmgen.out("  jsr  prog8_lib.equalzero_w")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.equal_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            "!=" -> {
-                when(dt) {
-                    DataType.UBYTE, DataType.BYTE -> asmgen.out("  jsr  prog8_lib.notequalzero_b")
-                    DataType.UWORD, DataType.WORD -> asmgen.out("  jsr  prog8_lib.notequalzero_w")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.notequal_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            "<" -> {
-                if(dt==DataType.UBYTE || dt==DataType.UWORD)
-                    return translateExpressionInternal(PtNumber.fromBoolean(false, expr.position))
-                when(dt) {
-                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.lesszero_b")
-                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.lesszero_w")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.less_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            ">" -> {
-                when(dt) {
-                    DataType.UBYTE -> asmgen.out("  jsr  prog8_lib.greaterzero_ub")
-                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.greaterzero_sb")
-                    DataType.UWORD -> asmgen.out("  jsr  prog8_lib.greaterzero_uw")
-                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.greaterzero_sw")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.greater_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            "<=" -> {
-                when(dt) {
-                    DataType.UBYTE -> asmgen.out("  jsr  prog8_lib.equalzero_b")
-                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.lessequalzeros_b")
-                    DataType.UWORD -> asmgen.out("  jsr  prog8_lib.equalzero_w")
-                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.lessequalzero_sw")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.lessequal_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            ">=" -> {
-                if(dt==DataType.UBYTE || dt==DataType.UWORD)
-                    return translateExpressionInternal(PtNumber.fromBoolean(true, expr.position))
-                when(dt) {
-                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.greaterequalzero_sb")
-                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.greaterequalzero_sw")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.greaterequal_zero")
-                    else -> throw AssemblyError("wrong dt")
-                }
-            }
-            else -> throw AssemblyError("invalid comparison operator")
-        }
-    }
-
-    private fun translateSquared(variable: PtIdentifier, dt: DataType) {
-        val asmVar = asmgen.asmVariableName(variable)
-        when(dt) {
-            DataType.BYTE, DataType.UBYTE -> {
-                asmgen.out("  lda  $asmVar")
-                asmgen.signExtendAYlsb(dt)
-                asmgen.out("  jsr  math.square")
-            }
-            DataType.UWORD, DataType.WORD -> {
-                asmgen.out("  lda  $asmVar |  ldy  $asmVar+1 |  jsr  math.square")
-            }
-            else -> throw AssemblyError("require integer dt for square")
-        }
-        asmgen.out("  sta  P8ESTACK_LO,x |  tya |  sta  P8ESTACK_HI,x |  dex")
-    }
-
-    private fun translateExpression(expr: PtPrefix) {
-        translateExpressionInternal(expr.value)
-        when(expr.operator) {
-            "+" -> {}
-            "-" -> {
-                when(expr.type) {
-                    in ByteDatatypes -> asmgen.out("  jsr  prog8_lib.neg_b")
-                    in WordDatatypes -> asmgen.out("  jsr  prog8_lib.neg_w")
-                    DataType.FLOAT -> asmgen.out("  jsr  floats.neg_f")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            "~" -> {
-                when(expr.type) {
-                    in ByteDatatypes ->
-                        asmgen.out("""
-                            lda  P8ESTACK_LO+1,x
-                            eor  #255
-                            sta  P8ESTACK_LO+1,x
-                            """)
-                    in WordDatatypes -> asmgen.out("  jsr  prog8_lib.inv_word")
-                    else -> throw AssemblyError("weird type")
-                }
-            }
-            else -> throw AssemblyError("invalid prefix operator ${expr.operator}")
-        }
-    }
-
-    private fun translateExpression(arrayExpr: PtArrayIndexer) {
-        val elementDt = arrayExpr.type
-        val arrayVarName = asmgen.asmVariableName(arrayExpr.variable)
-
-        if(arrayExpr.variable.type==DataType.UWORD) {
-            // indexing a pointer var instead of a real array or string
-            if(elementDt !in ByteDatatypes)
-                throw AssemblyError("non-array var indexing requires bytes dt")
-            if(arrayExpr.index.type != DataType.UBYTE)
-                throw AssemblyError("non-array var indexing requires bytes index")
-            asmgen.loadScaledArrayIndexIntoRegister(arrayExpr, elementDt, CpuRegister.Y)
-            if(asmgen.isZpVar(arrayExpr.variable)) {
-                asmgen.out("  lda  ($arrayVarName),y")
-            } else {
-                asmgen.out("  lda  $arrayVarName |  sta  P8ZP_SCRATCH_W1 |  lda  $arrayVarName+1 |  sta  P8ZP_SCRATCH_W1+1")
-                asmgen.out("  lda  (P8ZP_SCRATCH_W1),y")
-            }
-            asmgen.out("  sta  P8ESTACK_LO,x |  dex")
-            return
-        }
-
-        val constIndexNum = arrayExpr.index.asConstInteger()
-        if(constIndexNum!=null) {
-            val indexValue = constIndexNum * program.memsizer.memorySize(elementDt)
-            when(elementDt) {
-                in ByteDatatypes -> {
-                    asmgen.out("  lda  $arrayVarName+$indexValue |  sta  P8ESTACK_LO,x |  dex")
-                }
-                in WordDatatypes -> {
-                    asmgen.out("  lda  $arrayVarName+$indexValue |  sta  P8ESTACK_LO,x |  lda  $arrayVarName+$indexValue+1 |  sta  P8ESTACK_HI,x |  dex")
-                }
-                DataType.FLOAT -> {
-                    asmgen.out("  lda  #<($arrayVarName+$indexValue) |  ldy  #>($arrayVarName+$indexValue) |  jsr  floats.push_float")
-                }
-                else -> throw AssemblyError("weird element type")
-            }
-        } else {
-            when(elementDt) {
-                in ByteDatatypes -> {
-                    asmgen.loadScaledArrayIndexIntoRegister(arrayExpr, elementDt, CpuRegister.Y)
-                    asmgen.out(" lda  $arrayVarName,y |  sta  P8ESTACK_LO,x |  dex")
-                }
-                in WordDatatypes -> {
-                    asmgen.loadScaledArrayIndexIntoRegister(arrayExpr, elementDt, CpuRegister.Y)
-                    asmgen.out(" lda  $arrayVarName,y |  sta  P8ESTACK_LO,x |  lda  $arrayVarName+1,y |  sta  P8ESTACK_HI,x |  dex")
-                }
-                DataType.FLOAT -> {
-                    asmgen.loadScaledArrayIndexIntoRegister(arrayExpr, elementDt, CpuRegister.A)
-                    asmgen.out("""
-                        ldy  #>$arrayVarName
-                        clc
-                        adc  #<$arrayVarName
-                        bcc  +
-                        iny
-+                       jsr  floats.push_float""")
-                }
-                else -> throw AssemblyError("weird dt")
-            }
-
-        }
-    }
-
-    private fun translateBinaryOperatorBytes(operator: String, types: DataType) {
-        when(operator) {
-            "*" -> asmgen.out("  jsr  prog8_lib.mul_byte")  //  the optimized routines should have been checked earlier
-            "/" -> asmgen.out(if(types==DataType.UBYTE) "  jsr  prog8_lib.idiv_ub" else "  jsr  prog8_lib.idiv_b")
-            "%" -> {
-                if(types==DataType.BYTE)
-                    throw AssemblyError("remainder of signed integers is not properly defined/implemented, use unsigned instead")
-                asmgen.out("  jsr prog8_lib.remainder_ub")
-            }
-            "+" -> asmgen.out("""
-                lda  P8ESTACK_LO+2,x
-                clc
-                adc  P8ESTACK_LO+1,x
-                inx
-                sta  P8ESTACK_LO+1,x
-                """)
-            "-" -> asmgen.out("""
-                lda  P8ESTACK_LO+2,x
-                sec
-                sbc  P8ESTACK_LO+1,x
-                inx
-                sta  P8ESTACK_LO+1,x
-                """)
-            "<<" -> asmgen.out("  jsr  prog8_lib.shiftleft_b")
-            ">>" -> asmgen.out("  jsr  prog8_lib.shiftright_b")
-            "<" -> asmgen.out(if(types==DataType.UBYTE) "  jsr  prog8_lib.less_ub" else "  jsr  prog8_lib.less_b")
-            ">" -> asmgen.out(if(types==DataType.UBYTE) "  jsr  prog8_lib.greater_ub" else "  jsr  prog8_lib.greater_b")
-            "<=" -> asmgen.out(if(types==DataType.UBYTE) "  jsr  prog8_lib.lesseq_ub" else "  jsr  prog8_lib.lesseq_b")
-            ">=" -> asmgen.out(if(types==DataType.UBYTE) "  jsr  prog8_lib.greatereq_ub" else "  jsr  prog8_lib.greatereq_b")
-            "==" -> asmgen.out("  jsr  prog8_lib.equal_b")
-            "!=" -> asmgen.out("  jsr  prog8_lib.notequal_b")
-            "&" -> asmgen.out("  jsr  prog8_lib.bitand_b")
-            "^" -> asmgen.out("  jsr  prog8_lib.bitxor_b")
-            "|" -> asmgen.out("  jsr  prog8_lib.bitor_b")
-            else -> throw AssemblyError("invalid operator $operator")
-        }
-    }
-
-    private fun translateBinaryOperatorWords(operator: String, dt: DataType) {
-        when(operator) {
-            "*" -> asmgen.out("  jsr  prog8_lib.mul_word")
-            "/" -> asmgen.out(if(dt==DataType.UWORD) "  jsr  prog8_lib.idiv_uw" else "  jsr  prog8_lib.idiv_w")
-            "%" -> {
-                if(dt==DataType.WORD)
-                    throw AssemblyError("remainder of signed integers is not properly defined/implemented, use unsigned instead")
-                asmgen.out("  jsr prog8_lib.remainder_uw")
-            }
-            "+" -> asmgen.out("  jsr  prog8_lib.add_w")
-            "-" -> asmgen.out("  jsr  prog8_lib.sub_w")
-            "<<" -> asmgen.out("  jsr  math.shift_left_w")
-            ">>" -> {
-                if(dt==DataType.UWORD)
-                    asmgen.out("  jsr  math.shift_right_uw")
-                else
-                    asmgen.out("  jsr  math.shift_right_w")
-            }
-            "<" -> asmgen.out(if(dt==DataType.UWORD) "  jsr  prog8_lib.less_uw" else "  jsr  prog8_lib.less_w")
-            ">" -> asmgen.out(if(dt==DataType.UWORD) "  jsr  prog8_lib.greater_uw" else "  jsr  prog8_lib.greater_w")
-            "<=" -> asmgen.out(if(dt==DataType.UWORD) "  jsr  prog8_lib.lesseq_uw" else "  jsr  prog8_lib.lesseq_w")
-            ">=" -> asmgen.out(if(dt==DataType.UWORD) "  jsr  prog8_lib.greatereq_uw" else "  jsr  prog8_lib.greatereq_w")
-            "==" -> asmgen.out("  jsr  prog8_lib.equal_w")
-            "!=" -> asmgen.out("  jsr  prog8_lib.notequal_w")
-            "&" -> asmgen.out("  jsr  prog8_lib.bitand_w")
-            "^" -> asmgen.out("  jsr  prog8_lib.bitxor_w")
-            "|" -> asmgen.out("  jsr  prog8_lib.bitor_w")
-            else -> throw AssemblyError("invalid operator $operator")
-        }
-    }
-
-    private fun translateBinaryOperatorFloats(operator: String) {
-        when(operator) {
-            "*" -> asmgen.out("  jsr  floats.mul_f")
-            "/" -> asmgen.out("  jsr  floats.div_f")
-            "+" -> asmgen.out("  jsr  floats.add_f")
-            "-" -> asmgen.out("  jsr  floats.sub_f")
-            "<" -> asmgen.out("  jsr  floats.less_f")
-            ">" -> asmgen.out("  jsr  floats.greater_f")
-            "<=" -> asmgen.out("  jsr  floats.lesseq_f")
-            ">=" -> asmgen.out("  jsr  floats.greatereq_f")
-            "==" -> asmgen.out("  jsr  floats.equal_f")
-            "!=" -> asmgen.out("  jsr  floats.notequal_f")
-            "%", "<<", ">>", "&", "^", "|" -> throw AssemblyError("requires integer datatype")
-            else -> throw AssemblyError("invalid operator $operator")
-        }
-    }
-
-    private fun translateCompareStrings(s1: PtExpression, operator: String, s2: PtExpression) {
-        asmgen.assignExpressionToVariable(s1, "prog8_lib.strcmp_expression._arg_s1", DataType.UWORD, null)
-        asmgen.assignExpressionToVariable(s2, "prog8_lib.strcmp_expression._arg_s2", DataType.UWORD, null)
-        asmgen.out(" jsr  prog8_lib.strcmp_expression")    // result  of compare is in A
-        when(operator) {
-            "==" -> asmgen.out(" and  #1 |  eor  #1 |  sta  P8ESTACK_LO,x")
-            "!=" -> asmgen.out(" and  #1 |  sta  P8ESTACK_LO,x")
-            "<=" -> asmgen.out("""
-                bpl  +
-                lda  #1
-                bne  ++
-+               lda  #0
-+               sta  P8ESTACK_LO,x""")
-            ">=" -> asmgen.out("""
-                bmi  +
-                lda  #1
-                bne  ++
-+               lda  #0
-+               sta  P8ESTACK_LO,x""")
-            "<" -> asmgen.out("""
-                bmi  +
-                lda  #0
-                beq  ++
-+               lda  #1
-+               sta  P8ESTACK_LO,x""")
-            ">" -> asmgen.out("""
-                bpl  +
-                lda  #0
-                beq  ++
-+               lda  #1
-+               sta  P8ESTACK_LO,x""")
-        }
-        asmgen.out("  dex")
-    }
-}

codeGenCpu6502/src/prog8/codegen/cpu6502/Extensions.kt

@@ -6,26 +6,6 @@ import prog8.code.ast.PtSub
 import prog8.code.core.*
 
 
-internal fun IPtSubroutine.regXasResult(): Boolean =
-    (this is PtAsmSub) && this.returns.any { it.first.registerOrPair in arrayOf(RegisterOrPair.X, RegisterOrPair.AX, RegisterOrPair.XY) }
-
-internal fun IPtSubroutine.shouldSaveX(): Boolean =
-    this.regXasResult() || (this is PtAsmSub && (CpuRegister.X in this.clobbers || regXasParam()))
-
-internal fun PtAsmSub.regXasParam(): Boolean =
-    parameters.any { it.first.registerOrPair in arrayOf(RegisterOrPair.X, RegisterOrPair.AX, RegisterOrPair.XY) }
-
-internal class KeepAresult(val saveOnEntry: Boolean, val saveOnReturn: Boolean)
-
-internal fun PtAsmSub.shouldKeepA(): KeepAresult {
-    // determine if A's value should be kept when preparing for calling the subroutine, and when returning from it
-
-    // it seems that we never have to save A when calling? will be loaded correctly after setup.
-    // but on return it depends on wether the routine returns something in A.
-    val saveAonReturn = returns.any { it.first.registerOrPair==RegisterOrPair.A || it.first.registerOrPair==RegisterOrPair.AY || it.first.registerOrPair==RegisterOrPair.AX }
-    return KeepAresult(false, saveAonReturn)
-}
-
 internal fun IPtSubroutine.returnsWhatWhere(): List<Pair<RegisterOrStatusflag, DataType>> {
     when(this) {
         is PtAsmSub -> {

codeGenCpu6502/src/prog8/codegen/cpu6502/ForLoopsAsmGen.kt

@@ -5,9 +5,10 @@ import prog8.code.ast.*
 import prog8.code.core.*
 import kotlin.math.absoluteValue
 
-internal class ForLoopsAsmGen(private val program: PtProgram,
-                              private val asmgen: AsmGen6502Internal,
-                              private val zeropage: Zeropage) {
+internal class ForLoopsAsmGen(
+    private val asmgen: AsmGen6502Internal,
+    private val zeropage: Zeropage
+) {
 
     internal fun translate(stmt: PtForLoop) {
         val iterableDt = stmt.iterable.type
@@ -50,8 +51,36 @@ internal class ForLoopsAsmGen(private val program: PtProgram,
                     val incdec = if(stepsize==1) "inc" else "dec"
                     // loop over byte range via loopvar
                     val varname = asmgen.asmVariableName(stmt.variable)
-                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt), null)
-                    asmgen.assignExpressionToVariable(range.to, "$modifiedLabel+1", ArrayToElementTypes.getValue(iterableDt), null)
+                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt))
+                    asmgen.assignExpressionToRegister(range.to, RegisterOrPair.A, false)
+                    // pre-check for end already reached
+                    if(iterableDt==DataType.ARRAY_B) {
+                        asmgen.out("  sta  $modifiedLabel+1")
+                        if(stepsize<0)
+                            asmgen.out("""
+                                clc
+                                sbc  $varname
+                                bvc  +
+                                eor  #${'$'}80
++                               bpl  $endLabel""")
+                        else
+                            asmgen.out("""
+                                sec
+                                sbc  $varname
+                                bvc  +
+                                eor  #${'$'}80
++                               bmi  $endLabel""")
+                    } else {
+                        if(stepsize<0)
+                            asmgen.out("""
+                                cmp  $varname
+                                beq  +
+                                bcs  $endLabel
++""")
+                        else
+                            asmgen.out("  cmp  $varname  |  bcc  $endLabel")
+                        asmgen.out("  sta  $modifiedLabel+1")
+                    }
                     asmgen.out(loopLabel)
                     asmgen.translate(stmt.statements)
                     asmgen.out("""
@@ -68,8 +97,36 @@ $modifiedLabel          cmp  #0         ; modified
 
                     // loop over byte range via loopvar
                     val varname = asmgen.asmVariableName(stmt.variable)
-                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt), null)
-                    asmgen.assignExpressionToVariable(range.to, "$modifiedLabel+1", ArrayToElementTypes.getValue(iterableDt), null)
+                    asmgen.assignExpressionToVariable(range.from, varname, ArrayToElementTypes.getValue(iterableDt))
+                    asmgen.assignExpressionToRegister(range.to, RegisterOrPair.A, false)
+                    // pre-check for end already reached
+                    if(iterableDt==DataType.ARRAY_B) {
+                        asmgen.out("  sta  $modifiedLabel+1")
+                        if(stepsize<0)
+                            asmgen.out("""
+                                clc
+                                sbc  $varname
+                                bvc  +
+                                eor  #${'$'}80
++                               bpl  $endLabel""")
+                        else
+                            asmgen.out("""
+                                sec
+                                sbc  $varname
+                                bvc  +
+                                eor  #${'$'}80
++                               bmi  $endLabel""")
+                    } else {
+                        if(stepsize<0)
+                            asmgen.out("""
+                                cmp  $varname
+                                beq  +
+                                bcs  $endLabel
++""")
+                        else
+                            asmgen.out("  cmp  $varname  |  bcc  $endLabel")
+                        asmgen.out("  sta  $modifiedLabel+1")
+                    }
                     asmgen.out(loopLabel)
                     asmgen.translate(stmt.statements)
                     if(stepsize>0) {
@@ -100,8 +157,9 @@ $modifiedLabel              cmp  #0     ; modified
 
                     stepsize == 1 || stepsize == -1 -> {
                         val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvar(stmt, range)
+                        assignLoopvarWord(stmt, range)
                         asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
                         asmgen.out("""
                             sty  $modifiedLabel+1
                             sta  $modifiedLabel2+1
@@ -134,8 +192,9 @@ $modifiedLabel2 cmp  #0    ; modified
 
                         // (u)words, step >= 2
                         val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvar(stmt, range)
+                        assignLoopvarWord(stmt, range)
                         asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
                         asmgen.out("""
                             sty  $modifiedLabel+1
                             sta  $modifiedLabel2+1
@@ -182,41 +241,25 @@ $endLabel""")
 
                         // (u)words, step <= -2
                         val varname = asmgen.asmVariableName(stmt.variable)
-                        assignLoopvar(stmt, range)
+                        assignLoopvarWord(stmt, range)
                         asmgen.assignExpressionToRegister(range.to, RegisterOrPair.AY)
+                        precheckFromToWord(iterableDt, stepsize, varname, endLabel)
                         asmgen.out("""
                             sty  $modifiedLabel+1
                             sta  $modifiedLabel2+1
 $loopLabel""")
                         asmgen.translate(stmt.statements)
 
-                        if(iterableDt==DataType.ARRAY_UW) {
                             asmgen.out("""
                 lda  $varname
                 sec
                 sbc  #<${stepsize.absoluteValue}
                 sta  $varname
+                tax
                 lda  $varname+1
                 sbc  #>${stepsize.absoluteValue}
                 sta  $varname+1
-$modifiedLabel  cmp  #0    ; modified                
-                bcc  $endLabel
-                bne  $loopLabel
-                lda  $varname
-$modifiedLabel2 cmp  #0    ; modified 
-                bcs  $loopLabel
-$endLabel""")
-                        } else {
-                            asmgen.out("""
-                lda  $varname
-                sec
-                sbc  #<${stepsize.absoluteValue}
-                sta  $varname
-                pha
-                lda  $varname+1
-                sbc  #>${stepsize.absoluteValue}
-                sta  $varname+1
-                pla
+                txa
 $modifiedLabel2 cmp  #0    ; modified 
                 lda  $varname+1
 $modifiedLabel  sbc  #0    ; modified
@@ -224,7 +267,6 @@ $modifiedLabel  sbc  #0    ; modified
                 eor  #$80
 +               bpl  $loopLabel                
 $endLabel""")
-                        }
                     }
                 }
             }
@@ -234,6 +276,56 @@ $endLabel""")
         asmgen.loopEndLabels.pop()
     }
 
+    private fun precheckFromToWord(iterableDt: DataType, stepsize: Int, fromVar: String, endLabel: String) {
+        // pre-check for end already reached.
+        // 'to' is in AY, do NOT clobber this!
+        if(iterableDt==DataType.ARRAY_W) {
+            if(stepsize<0)
+                asmgen.out("""
+                    sta  P8ZP_SCRATCH_W2        ; to
+                    sty  P8ZP_SCRATCH_W2+1      ; to
+                    lda  $fromVar
+	                cmp  P8ZP_SCRATCH_W2
+                    lda  $fromVar+1
+	                sbc  P8ZP_SCRATCH_W2+1
+	                bvc  +
+	                eor  #${'$'}80
++           		bmi  $endLabel
+                    lda  P8ZP_SCRATCH_W2
+                    ldy  P8ZP_SCRATCH_W2+1""")
+            else
+                asmgen.out("""
+                    sta  P8ZP_SCRATCH_REG
+	                cmp  $fromVar
+	                tya
+	                sbc  $fromVar+1
+	                bvc  +
+	                eor  #${'$'}80
++           		bmi  $endLabel
+                    lda  P8ZP_SCRATCH_REG""")
+        } else {
+            if(stepsize<0)
+                asmgen.out("""
+                    cpy  $fromVar+1
+                    beq  +
+                    bcc  ++
+                    bcs  $endLabel
++                   cmp  $fromVar
+                    bcc  +
+                    beq  +
+                    bne  $endLabel
++""")
+            else
+                asmgen.out("""
+                	cpy  $fromVar+1
+	                bcc  $endLabel
+	                bne  +
+	                cmp  $fromVar
+	                bcc  $endLabel
++""")
+        }
+    }
+
     private fun translateForOverIterableVar(stmt: PtForLoop, iterableDt: DataType, ident: PtIdentifier) {
         val loopLabel = asmgen.makeLabel("for_loop")
         val endLabel = asmgen.makeLabel("for_end")
@@ -291,7 +383,7 @@ $loopLabel          sty  $indexVar
                     // allocate index var on ZP if possible
                     val result = zeropage.allocate(indexVar, DataType.UBYTE, null, stmt.position, asmgen.errors)
                     result.fold(
-                        success = { (address,_)-> asmgen.out("""$indexVar = $address  ; auto zp UBYTE""") },
+                        success = { (address, _, _)-> asmgen.out("""$indexVar = $address  ; auto zp UBYTE""") },
                         failure = { asmgen.out("$indexVar    .byte  0") }
                     )
                 } else {
@@ -332,7 +424,46 @@ $loopLabel          sty  $indexVar
                     // allocate index var on ZP if possible
                     val result = zeropage.allocate(indexVar, DataType.UBYTE, null, stmt.position, asmgen.errors)
                     result.fold(
-                        success = { (address,_)-> asmgen.out("""$indexVar = $address  ; auto zp UBYTE""") },
+                        success = { (address,_,_)-> asmgen.out("""$indexVar = $address  ; auto zp UBYTE""") },
+                        failure = { asmgen.out("$indexVar    .byte  0") }
+                    )
+                } else {
+                    asmgen.out("$indexVar    .byte  0")
+                }
+                asmgen.out(endLabel)
+            }
+            DataType.ARRAY_UW_SPLIT, DataType.ARRAY_W_SPLIT -> {
+                numElements!!
+                val indexVar = asmgen.makeLabel("for_index")
+                val loopvarName = asmgen.asmVariableName(stmt.variable)
+                asmgen.out("""
+                    ldy  #0
+$loopLabel          sty  $indexVar
+                    lda  ${iterableName}_lsb,y
+                    sta  $loopvarName
+                    lda  ${iterableName}_msb,y
+                    sta  $loopvarName+1""")
+                asmgen.translate(stmt.statements)
+                if(numElements<=255u) {
+                    asmgen.out("""
+                        ldy  $indexVar
+                        iny
+                        cpy  #$numElements
+                        beq  $endLabel
+                        bne  $loopLabel""")
+                } else {
+                    // length is 256
+                    asmgen.out("""
+                        ldy  $indexVar
+                        iny
+                        bne  $loopLabel
+                        beq  $endLabel""")
+                }
+                if(numElements>=16u) {
+                    // allocate index var on ZP if possible
+                    val result = zeropage.allocate(indexVar, DataType.UBYTE, null, stmt.position, asmgen.errors)
+                    result.fold(
+                        success = { (address,_,_)-> asmgen.out("""$indexVar = $address  ; auto zp UBYTE""") },
                         failure = { asmgen.out("$indexVar    .byte  0") }
                     )
                 } else {
@@ -590,10 +721,9 @@ $loopLabel""")
         asmgen.loopEndLabels.pop()
     }
 
-    private fun assignLoopvar(stmt: PtForLoop, range: PtRange) =
+    private fun assignLoopvarWord(stmt: PtForLoop, range: PtRange) =
         asmgen.assignExpressionToVariable(
             range.from,
             asmgen.asmVariableName(stmt.variable),
-            stmt.variable.type,
-            stmt.definingISub())
+            stmt.variable.type)
 }

codeGenCpu6502/src/prog8/codegen/cpu6502/FunctionCallAsmGen.kt

@@ -11,42 +11,10 @@ import prog8.codegen.cpu6502.assignment.TargetStorageKind
 internal class FunctionCallAsmGen(private val program: PtProgram, private val asmgen: AsmGen6502Internal) {
 
     internal fun translateFunctionCallStatement(stmt: PtFunctionCall) {
-        saveXbeforeCall(stmt)
         translateFunctionCall(stmt)
-        restoreXafterCall(stmt)
         // just ignore any result values from the function call.
     }
 
-    internal fun saveXbeforeCall(stmt: PtFunctionCall) {
-        val symbol = asmgen.symbolTable.lookup(stmt.name)
-        val sub = symbol!!.astNode as IPtSubroutine
-        if(sub.shouldSaveX()) {
-            if(sub is PtAsmSub) {
-                val regSaveOnStack = sub.address == null       // rom-routines don't require registers to be saved on stack, normal subroutines do because they can contain nested calls
-                if (regSaveOnStack)
-                    asmgen.saveRegisterStack(CpuRegister.X, sub.shouldKeepA().saveOnEntry)
-                else
-                    asmgen.saveRegisterLocal(CpuRegister.X, stmt.definingISub()!!)
-            } else
-                asmgen.saveRegisterLocal(CpuRegister.X, stmt.definingISub()!!)
-        }
-    }
-
-    internal fun restoreXafterCall(stmt: PtFunctionCall) {
-        val symbol = asmgen.symbolTable.lookup(stmt.name)
-        val sub = symbol!!.astNode as IPtSubroutine
-        if(sub.shouldSaveX()) {
-            if(sub is PtAsmSub) {
-                val regSaveOnStack = sub.address == null       // rom-routines don't require registers to be saved on stack, normal subroutines do because they can contain nested calls
-                if (regSaveOnStack)
-                    asmgen.restoreRegisterStack(CpuRegister.X, sub.shouldKeepA().saveOnReturn)
-                else
-                    asmgen.restoreRegisterLocal(CpuRegister.X)
-            } else
-                asmgen.restoreRegisterLocal(CpuRegister.X)
-        }
-    }
-
     internal fun optimizeIntArgsViaRegisters(sub: PtSub) =
         (sub.parameters.size==1 && sub.parameters[0].type in IntegerDatatypes)
                 || (sub.parameters.size==2 && sub.parameters[0].type in ByteDatatypes && sub.parameters[1].type in ByteDatatypes)
@@ -63,8 +31,8 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
 
         if(sub is PtAsmSub) {
             argumentsViaRegisters(sub, call)
-            if (sub.inline && asmgen.options.optimize) {
-                // inline the subroutine.
+            if (sub.inline) {
+                // inline the subroutine. (regardless of optimization settings!)
                 // we do this by copying the subroutine's statements at the call site.
                 // NOTE: *if* there is a return statement, it will be the only one, and the very last statement of the subroutine
                 // (this condition has been enforced by an ast check earlier)
@@ -141,7 +109,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
             throw AssemblyError("argument type incompatible")
 
         val varName = asmgen.asmVariableName(sub.scopedName + "." + parameter.name)
-        asmgen.assignExpressionToVariable(value, varName, parameter.type, sub)
+        asmgen.assignExpressionToVariable(value, varName, parameter.type)
     }
 
     private fun argumentViaRegister(sub: IPtSubroutine, parameter: IndexedValue<PtSubroutineParameter>, value: PtExpression, registerOverride: RegisterOrPair? = null) {
@@ -164,7 +132,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
             if(requiredDt!=value.type)
                 throw AssemblyError("for statusflag, byte value is required")
             if (statusflag == Statusflag.Pc) {
-                // this param needs to be set last, right before the jsr
+                // this boolean param needs to be set last, right before the jsr
                 // for now, this is already enforced on the subroutine definition by the Ast Checker
                 when(value) {
                     is PtNumber -> {
@@ -173,6 +141,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                     }
                     is PtIdentifier -> {
                         val sourceName = asmgen.asmVariableName(value)
+                        // note: cannot use X register here to store A because it might be used for other arguments
                         asmgen.out("""
                             pha
                             clc
@@ -183,12 +152,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                     }
                     else -> {
                         asmgen.assignExpressionToRegister(value, RegisterOrPair.A)
-                        asmgen.out("""
-                            beq  +
-                            sec
-                            bcs  ++
-+                           clc
-+""")
+                        asmgen.out("  ror  a")
                     }
                 }
             } else throw AssemblyError("can only use Carry as status flag parameter")
@@ -198,16 +162,17 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
             register!!
             if(requiredDt largerThan value.type) {
                 // we need to sign extend the source, do this via temporary word variable
-                asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_W1", DataType.UBYTE, sub)
+                asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_W1", DataType.UBYTE)
                 asmgen.signExtendVariableLsb("P8ZP_SCRATCH_W1", value.type)
-                asmgen.assignVariableToRegister("P8ZP_SCRATCH_W1", register)
+                asmgen.assignVariableToRegister("P8ZP_SCRATCH_W1", register, null, Position.DUMMY)
             } else {
+                val scope = value.definingISub()
                 val target: AsmAssignTarget =
                     if(parameter.value.type in ByteDatatypes && (register==RegisterOrPair.AX || register == RegisterOrPair.AY || register==RegisterOrPair.XY || register in Cx16VirtualRegisters))
-                        AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, parameter.value.type, sub, register = register)
+                        AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, parameter.value.type, scope, value.position, register = register)
                     else {
                         val signed = parameter.value.type == DataType.BYTE || parameter.value.type == DataType.WORD
-                        AsmAssignTarget.fromRegisters(register, signed, sub, asmgen)
+                        AsmAssignTarget.fromRegisters(register, signed, value.position, scope, asmgen)
                     }
                 val src = if(value.type in PassByReferenceDatatypes) {
                     if(value is PtIdentifier) {
@@ -221,7 +186,7 @@ internal class FunctionCallAsmGen(private val program: PtProgram, private val as
                 } else {
                     AsmAssignSource.fromAstSource(value, program, asmgen).adjustSignedUnsigned(target)
                 }
-                asmgen.translateNormalAssignment(AsmAssignment(src, target, program.memsizer, Position.DUMMY))
+                asmgen.translateNormalAssignment(AsmAssignment(src, target, program.memsizer, Position.DUMMY), scope)
             }
         }
     }

codeGenCpu6502/src/prog8/codegen/cpu6502/PostIncrDecrAsmGen.kt

@@ -13,7 +13,6 @@ internal class PostIncrDecrAsmGen(private val program: PtProgram, private val as
         val targetIdent = stmt.target.identifier
         val targetMemory = stmt.target.memory
         val targetArrayIdx = stmt.target.array
-        val scope = stmt.definingISub()
         when {
             targetIdent!=null -> {
                 val what = asmgen.asmVariableName(targetIdent)
@@ -65,20 +64,59 @@ internal class PostIncrDecrAsmGen(private val program: PtProgram, private val as
                 val asmArrayvarname = asmgen.asmVariableName(targetArrayIdx.variable)
                 val elementDt = targetArrayIdx.type
                 val constIndex = targetArrayIdx.index.asConstInteger()
+                if(targetArrayIdx.splitWords) {
+                    if(constIndex!=null) {
+                        if(incr)
+                            asmgen.out(" inc  ${asmArrayvarname}_lsb+$constIndex |  bne  + |  inc  ${asmArrayvarname}_msb+$constIndex |+")
+                        else
+                            asmgen.out("""
+        lda  ${asmArrayvarname}_lsb+$constIndex
+        bne  +
+        dec  ${asmArrayvarname}_msb+$constIndex
++       dec  ${asmArrayvarname}_lsb+$constIndex""")
+                    } else {
+                        asmgen.loadScaledArrayIndexIntoRegister(targetArrayIdx, elementDt, CpuRegister.X)
+                        if(incr)
+                            asmgen.out(" inc  ${asmArrayvarname}_lsb,x |  bne  + |  inc  ${asmArrayvarname}_msb,x |+")
+                        else
+                            asmgen.out("""
+        lda  ${asmArrayvarname}_lsb,x
+        bne  +
+        dec  ${asmArrayvarname}_msb,x
++       dec  ${asmArrayvarname}_lsb,x""")
+                    }
+                    return
+                }
                 if(constIndex!=null) {
                     val indexValue = constIndex * program.memsizer.memorySize(elementDt)
                     when(elementDt) {
-                        in ByteDatatypes -> asmgen.out(if (incr) "  inc  $asmArrayvarname+$indexValue" else "  dec  $asmArrayvarname+$indexValue")
+                        in ByteDatatypes -> {
+                            if(targetArrayIdx.usesPointerVariable) {
+                                asmgen.out("""
+                                    lda  $asmArrayvarname
+                                    clc
+                                    adc  #$indexValue
+                                    sta  (+) +1
+                                    lda  $asmArrayvarname+1
+                                    adc  #0
+                                    sta  (+) +2""")
+                                if(incr)
+                                    asmgen.out("+\tinc  ${'$'}ffff\t; modified")
+                                else
+                                    asmgen.out("+\tdec  ${'$'}ffff\t; modified")
+                            } else {
+                                asmgen.out(if (incr) "  inc  $asmArrayvarname+$indexValue" else "  dec  $asmArrayvarname+$indexValue")
+                            }
+                        }
                         in WordDatatypes -> {
                             if(incr)
                                 asmgen.out(" inc  $asmArrayvarname+$indexValue |  bne  + |  inc  $asmArrayvarname+$indexValue+1 |+")
                             else
                                 asmgen.out("""
-    lda  $asmArrayvarname+$indexValue
-    bne  +
-    dec  $asmArrayvarname+$indexValue+1
-+       dec  $asmArrayvarname+$indexValue 
-""")
+        lda  $asmArrayvarname+$indexValue
+        bne  +
+        dec  $asmArrayvarname+$indexValue+1
++       dec  $asmArrayvarname+$indexValue""")
                         }
                         DataType.FLOAT -> {
                             asmgen.out("  lda  #<($asmArrayvarname+$indexValue) |  ldy  #>($asmArrayvarname+$indexValue)")
@@ -89,36 +127,47 @@ internal class PostIncrDecrAsmGen(private val program: PtProgram, private val as
                 }
                 else
                 {
-                    asmgen.loadScaledArrayIndexIntoRegister(targetArrayIdx, elementDt, CpuRegister.A)
-                    asmgen.saveRegisterLocal(CpuRegister.X, scope!!)
-                    asmgen.out("  tax")
+                    asmgen.loadScaledArrayIndexIntoRegister(targetArrayIdx, elementDt, CpuRegister.X)
                     when(elementDt) {
                         in ByteDatatypes -> {
-                            asmgen.out(if(incr) "  inc  $asmArrayvarname,x" else "  dec  $asmArrayvarname,x")
+                            if(targetArrayIdx.usesPointerVariable) {
+                                asmgen.out("""
+                                    txa
+                                    clc
+                                    adc  $asmArrayvarname
+                                    sta  (+) +1
+                                    lda  $asmArrayvarname+1
+                                    adc  #0
+                                    sta  (+) +2""")
+                                if(incr)
+                                    asmgen.out("+\tinc  ${'$'}ffff\t; modified")
+                                else
+                                    asmgen.out("+\tdec  ${'$'}ffff\t; modified")
+                            } else {
+                                asmgen.out(if (incr) "  inc  $asmArrayvarname,x" else "  dec  $asmArrayvarname,x")
+                            }
                         }
                         in WordDatatypes -> {
                             if(incr)
                                 asmgen.out(" inc  $asmArrayvarname,x |  bne  + |  inc  $asmArrayvarname+1,x |+")
                             else
                                 asmgen.out("""
-    lda  $asmArrayvarname,x
-    bne  +
-    dec  $asmArrayvarname+1,x
-+       dec  $asmArrayvarname,x 
-""")
+        lda  $asmArrayvarname,x
+        bne  +
+        dec  $asmArrayvarname+1,x
++       dec  $asmArrayvarname,x""")
                         }
                         DataType.FLOAT -> {
                             asmgen.out("""
-                    ldy  #>$asmArrayvarname
-                    clc
-                    adc  #<$asmArrayvarname
-                    bcc  +
-                    iny
-+                       jsr  floats.inc_var_f""")
+        ldy  #>$asmArrayvarname
+        clc
+        adc  #<$asmArrayvarname
+        bcc  +
+        iny
++       jsr  floats.inc_var_f""")
                         }
                         else -> throw AssemblyError("weird array elt dt")
                     }
-                    asmgen.restoreRegisterLocal(CpuRegister.X)
                 }
             }
         }

codeGenCpu6502/src/prog8/codegen/cpu6502/ProgramAndVarsGen.kt

@@ -33,8 +33,9 @@ internal class ProgramAndVarsGen(
     internal fun generate() {
         header()
         val allBlocks = program.allBlocks()
-        if(allBlocks.first().name != "main")
-            throw AssemblyError("first block should be 'main'")
+
+        if(allBlocks.first().name != "p8_main" && allBlocks.first().name != "main")
+            throw AssemblyError("first block should be 'main' or 'p8_main'")
 
         if(errors.noErrors())  {
             program.allBlocks().forEach { block2asm(it) }
@@ -74,8 +75,6 @@ internal class ProgramAndVarsGen(
         asmgen.out("P8ZP_SCRATCH_W1 = ${zp.SCRATCH_W1}    ; word")
         asmgen.out("P8ZP_SCRATCH_W2 = ${zp.SCRATCH_W2}    ; word")
         asmgen.out(".weak")   // hack to allow user to override the following two with command line redefinition (however, just use '-esa' command line option instead!)
-        asmgen.out("P8ESTACK_LO = ${compTarget.machine.ESTACK_LO.toHex()}")
-        asmgen.out("P8ESTACK_HI = ${compTarget.machine.ESTACK_HI.toHex()}")
         asmgen.out(".endweak")
 
         if(options.symbolDefs.isNotEmpty()) {
@@ -104,15 +103,15 @@ internal class ProgramAndVarsGen(
                         asmgen.out("+\t.word  0")
                         asmgen.out("prog8_entrypoint\t; assembly code starts here")
                         if(!options.noSysInit)
-                            asmgen.out("  jsr  ${compTarget.name}.init_system")
-                        asmgen.out("  jsr  ${compTarget.name}.init_system_phase2")
+                            asmgen.out("  jsr  sys.init_system")
+                        asmgen.out("  jsr  sys.init_system_phase2")
                     }
                     CbmPrgLauncherType.NONE -> {
                         asmgen.out("; ---- program without basic sys call ----")
                         asmgen.out("* = ${options.loadAddress.toHex()}")
                         if(!options.noSysInit)
-                            asmgen.out("  jsr  ${compTarget.name}.init_system")
-                        asmgen.out("  jsr  ${compTarget.name}.init_system_phase2")
+                            asmgen.out("  jsr  sys.init_system")
+                        asmgen.out("  jsr  sys.init_system_phase2")
                     }
                 }
             }
@@ -120,8 +119,8 @@ internal class ProgramAndVarsGen(
                 asmgen.out("; ---- atari xex program ----")
                 asmgen.out("* = ${options.loadAddress.toHex()}")
                 if(!options.noSysInit)
-                    asmgen.out("  jsr  ${compTarget.name}.init_system")
-                asmgen.out("  jsr  ${compTarget.name}.init_system_phase2")
+                    asmgen.out("  jsr  sys.init_system")
+                asmgen.out("  jsr  sys.init_system_phase2")
             }
         }
 
@@ -139,25 +138,24 @@ internal class ProgramAndVarsGen(
             "cx16" -> {
                 if(options.floats)
                     asmgen.out("  lda  #4 |  sta  $01")    // to use floats, make sure Basic rom is banked in
-                asmgen.out("  jsr  main.start")
-                asmgen.out("  jmp  ${compTarget.name}.cleanup_at_exit")
+                asmgen.out("  jsr  p8_main.p8_start")
+                asmgen.out("  jmp  sys.cleanup_at_exit")
             }
             "c64" -> {
-                asmgen.out("  jsr  main.start |  lda  #31 |  sta  $01")
+                asmgen.out("  jsr  p8_main.p8_start |  lda  #31 |  sta  $01")
                 if(!options.noSysInit)
-                    asmgen.out("  jmp  ${compTarget.name}.cleanup_at_exit")
+                    asmgen.out("  jmp  sys.cleanup_at_exit")
                 else
                     asmgen.out("  rts")
             }
             "c128" -> {
-                asmgen.out("  jsr  main.start")
-                // TODO c128: how to bank basic+kernal back in?
+                asmgen.out("  jsr  p8_main.p8_start |  lda  #0 |  sta ${"$"}ff00")
                 if(!options.noSysInit)
-                    asmgen.out("  jmp  ${compTarget.name}.cleanup_at_exit")
+                    asmgen.out("  jmp  sys.cleanup_at_exit")
                 else
                     asmgen.out("  rts")
             }
-            else -> asmgen.jmp("main.start")
+            else -> asmgen.jmp("p8_main.p8_start")
         }
     }
 
@@ -176,7 +174,8 @@ internal class ProgramAndVarsGen(
 
     private fun footer() {
         asmgen.out("; bss sections")
-        if(options.varsHigh) {
+        asmgen.out("PROG8_VARSHIGH_RAMBANK = ${options.varsHighBank ?: 1}")
+        if(options.varsHighBank!=null) {
             if(options.compTarget.machine.BSSHIGHRAM_START == 0u || options.compTarget.machine.BSSHIGHRAM_END==0u) {
                 throw AssemblyError("current compilation target hasn't got the high ram area properly defined")
             }
@@ -270,9 +269,7 @@ internal class ProgramAndVarsGen(
 
     internal fun translateAsmSubroutine(sub: PtAsmSub) {
         if(sub.inline) {
-            if(options.optimize) {
-                return
-            }
+            return      // subroutine gets inlined at call site.
         }
 
         asmgen.out("")
@@ -312,7 +309,9 @@ internal class ProgramAndVarsGen(
 
         asmgen.out("${sub.name}\t$asmStartScope")
 
-        val scope = symboltable.lookupOrElse(sub.scopedName) { throw AssemblyError("lookup") }
+        val scope = symboltable.lookupOrElse(sub.scopedName) {
+            throw AssemblyError("lookup ${sub.scopedName}")
+        }
         require(scope.type==StNodeType.SUBROUTINE)
         val varsInSubroutine = getVars(scope)
 
@@ -332,14 +331,14 @@ internal class ProgramAndVarsGen(
         asmsubs2asm(sub.children)
 
         // the main.start subroutine is the program's entrypoint and should perform some initialization logic
-        if(sub.name=="start" && sub.definingBlock()!!.name=="main")
+        if((sub.name=="start" || sub.name=="p8_start") && (sub.definingBlock()!!.name=="main" || sub.definingBlock()!!.name=="p8_main"))
             entrypointInitialization()
 
         if(functioncallAsmGen.optimizeIntArgsViaRegisters(sub)) {
             asmgen.out("; simple int arg(s) passed via register(s)")
             if(sub.parameters.size==1) {
                 val dt = sub.parameters[0].type
-                val target = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, dt, sub, variableAsmName = sub.parameters[0].name)
+                val target = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, dt, sub, sub.parameters[0].position, variableAsmName = sub.parameters[0].name)
                 if(dt in ByteDatatypes)
                     asmgen.assignRegister(RegisterOrPair.A, target)
                 else
@@ -347,8 +346,8 @@ internal class ProgramAndVarsGen(
             } else {
                 require(sub.parameters.size==2)
                 // 2 simple byte args, first in A, second in Y
-                val target1 = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, sub.parameters[0].type, sub, variableAsmName = sub.parameters[0].name)
-                val target2 = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, sub.parameters[1].type, sub, variableAsmName = sub.parameters[1].name)
+                val target1 = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, sub.parameters[0].type, sub, sub.parameters[0].position, variableAsmName = sub.parameters[0].name)
+                val target2 = AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, sub.parameters[1].type, sub, sub.parameters[1].position, variableAsmName = sub.parameters[1].name)
                 asmgen.assignRegister(RegisterOrPair.A, target1)
                 asmgen.assignRegister(RegisterOrPair.Y, target2)
             }
@@ -358,26 +357,23 @@ internal class ProgramAndVarsGen(
         sub.children.forEach { asmgen.translate(it) }
 
         asmgen.out("; variables")
+        asmgen.out("    .section BSS")
         val asmGenInfo = asmgen.subroutineExtra(sub)
         for((dt, name, addr) in asmGenInfo.extraVars) {
             if(addr!=null)
                 asmgen.out("$name = $addr")
             else when(dt) {
-                DataType.UBYTE -> asmgen.out("$name    .byte  0")
-                DataType.UWORD -> asmgen.out("$name    .word  0")
-                else -> throw AssemblyError("weird dt")
+                DataType.UBYTE -> asmgen.out("$name    .byte  ?")
+                DataType.UWORD -> asmgen.out("$name    .word  ?")
+                DataType.FLOAT -> asmgen.out("$name    .fill  ${options.compTarget.machine.FLOAT_MEM_SIZE}")
+                else -> throw AssemblyError("weird dt for extravar $dt")
             }
         }
-        if(asmGenInfo.usedRegsaveA)      // will probably never occur
-            asmgen.out("prog8_regsaveA     .byte  0")
-        if(asmGenInfo.usedRegsaveX)
-            asmgen.out("prog8_regsaveX     .byte  0")
-        if(asmGenInfo.usedRegsaveY)
-            asmgen.out("prog8_regsaveY     .byte  0")
         if(asmGenInfo.usedFloatEvalResultVar1)
-            asmgen.out("$subroutineFloatEvalResultVar1    .byte  0,0,0,0,0")
+            asmgen.out("$subroutineFloatEvalResultVar1    .fill  ${options.compTarget.machine.FLOAT_MEM_SIZE}")
         if(asmGenInfo.usedFloatEvalResultVar2)
-            asmgen.out("$subroutineFloatEvalResultVar2    .byte  0,0,0,0,0")
+            asmgen.out("$subroutineFloatEvalResultVar2    .fill  ${options.compTarget.machine.FLOAT_MEM_SIZE}")
+        asmgen.out("  .send BSS")
 
         // normal statically allocated variables
         val variables = varsInSubroutine
@@ -432,12 +428,12 @@ internal class ProgramAndVarsGen(
                 asmgen.out("""
                     lda  #<${name}_init_value
                     ldy  #>${name}_init_value
-                    sta  cx16.r0L
-                    sty  cx16.r0H
+                    sta  cx16.r0
+                    sty  cx16.r0+1
                     lda  #<${name}
                     ldy  #>${name}
-                    sta  cx16.r1L
-                    sty  cx16.r1H
+                    sta  cx16.r1
+                    sty  cx16.r1+1
                     lda  #<$size
                     ldy  #>$size
                     jsr  sys.memcopy""")
@@ -456,7 +452,6 @@ internal class ProgramAndVarsGen(
         }
 
         asmgen.out("""+        
-                    ldx  #127       ; init estack ptr (half page)
                     clv
                     clc""")
     }
@@ -478,8 +473,8 @@ internal class ProgramAndVarsGen(
         val vars = allocator.zeropageVars.filter { it.value.dt==DataType.STR }
         for (variable in vars) {
             val scopedName = variable.key
-            val svar = symboltable.flat.getValue(scopedName) as StStaticVariable
-            if(svar.onetimeInitializationStringValue!=null)
+            val svar = symboltable.lookup(scopedName) as? StStaticVariable
+            if(svar?.onetimeInitializationStringValue!=null)
                 result.add(ZpStringWithInitial(scopedName, variable.value, svar.onetimeInitializationStringValue!!))
         }
         return result
@@ -490,8 +485,8 @@ internal class ProgramAndVarsGen(
         val vars = allocator.zeropageVars.filter { it.value.dt in ArrayDatatypes }
         for (variable in vars) {
             val scopedName = variable.key
-            val svar = symboltable.flat.getValue(scopedName) as StStaticVariable
-            if(svar.onetimeInitializationArrayValue!=null)
+            val svar = symboltable.lookup(scopedName) as? StStaticVariable
+            if(svar?.onetimeInitializationArrayValue!=null)
                 result.add(ZpArrayWithInitial(scopedName, variable.value, svar.onetimeInitializationArrayValue!!))
         }
         return result
@@ -541,6 +536,11 @@ internal class ProgramAndVarsGen(
             DataType.UWORD -> asmgen.out("${variable.name}\t.word  ?")
             DataType.WORD -> asmgen.out("${variable.name}\t.sint  ?")
             DataType.FLOAT -> asmgen.out("${variable.name}\t.fill  ${compTarget.machine.FLOAT_MEM_SIZE}")
+            in SplitWordArrayTypes -> {
+                val numbytesPerHalf = compTarget.memorySize(variable.dt, variable.length!!) / 2
+                asmgen.out("${variable.name}_lsb\t.fill  $numbytesPerHalf")
+                asmgen.out("${variable.name}_msb\t.fill  $numbytesPerHalf")
+            }
             in ArrayDatatypes -> {
                 val numbytes = compTarget.memorySize(variable.dt, variable.length!!)
                 asmgen.out("${variable.name}\t.fill  $numbytes")
@@ -625,6 +625,18 @@ internal class ProgramAndVarsGen(
                         asmgen.out("  .sint  " + chunk.joinToString())
                 }
             }
+            DataType.ARRAY_UW_SPLIT -> {
+                val data = makeArrayFillDataUnsigned(dt, value, orNumberOfZeros)
+                asmgen.out("_array_$varname := ${data.joinToString()}")
+                asmgen.out("${varname}_lsb\t.byte <_array_$varname")
+                asmgen.out("${varname}_msb\t.byte >_array_$varname")
+            }
+            DataType.ARRAY_W_SPLIT -> {
+                val data = makeArrayFillDataSigned(dt, value, orNumberOfZeros)
+                asmgen.out("_array_$varname := ${data.joinToString()}")
+                asmgen.out("${varname}_lsb\t.byte <_array_$varname")
+                asmgen.out("${varname}_msb\t.byte >_array_$varname")
+            }
             DataType.ARRAY_F -> {
                 val array = value ?: zeroFilledArray(orNumberOfZeros!!)
                 val floatFills = array.map {
@@ -684,7 +696,7 @@ internal class ProgramAndVarsGen(
                     val number = it.number!!.toInt()
                     "$"+number.toString(16).padStart(2, '0')
                 }
-            DataType.ARRAY_UW -> array.map {
+            DataType.ARRAY_UW, DataType.ARRAY_UW_SPLIT -> array.map {
                 if(it.number!=null) {
                     "$" + it.number!!.toInt().toString(16).padStart(4, '0')
                 }
@@ -716,11 +728,11 @@ internal class ProgramAndVarsGen(
                     else
                         "-$$hexnum"
                 }
-            DataType.ARRAY_UW -> array.map {
+            DataType.ARRAY_UW, DataType.ARRAY_UW_SPLIT -> array.map {
                 val number = it.number!!.toInt()
                 "$" + number.toString(16).padStart(4, '0')
             }
-            DataType.ARRAY_W -> array.map {
+            DataType.ARRAY_W, DataType.ARRAY_W_SPLIT -> array.map {
                 val number = it.number!!.toInt()
                 val hexnum = number.absoluteValue.toString(16).padStart(4, '0')
                 if(number>=0)

codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AnyExprAsmGen.kt

@@ -0,0 +1,281 @@
+package prog8.codegen.cpu6502.assignment
+
+import prog8.code.ast.PtBinaryExpression
+import prog8.code.ast.PtExpression
+import prog8.code.core.*
+import prog8.code.target.C64Target
+import prog8.code.target.Cx16Target
+import prog8.codegen.cpu6502.AsmGen6502Internal
+
+//
+// This contains codegen for stack-based evaluation of binary expressions.
+// It uses the CPU stack so depth is limited.
+// It is called "as a last resort" if the optimized codegen path is unable
+// to come up with a special case of the expression.
+//
+internal class AnyExprAsmGen(
+    private val asmgen: AsmGen6502Internal
+) {
+    fun assignAnyExpressionUsingStack(expr: PtBinaryExpression, assign: AsmAssignment): Boolean {
+        when(expr.type) {
+            in ByteDatatypes -> {
+                if(expr.left.type in ByteDatatypes && expr.right.type in ByteDatatypes)
+                    return assignByteBinExpr(expr, assign)
+                if (expr.left.type in WordDatatypes && expr.right.type in WordDatatypes) {
+                    require(expr.operator in ComparisonOperators)
+                    TODO("words operands comparison -> byte")
+                }
+                if (expr.left.type==DataType.FLOAT && expr.right.type==DataType.FLOAT) {
+                    require(expr.operator in ComparisonOperators)
+                    return assignFloatBinExpr(expr, assign)
+                }
+                TODO("weird expr operand types")
+            }
+            in WordDatatypes -> {
+                require(expr.left.type in WordDatatypes && expr.right.type in WordDatatypes) {
+                    "both operands must be words"
+                }
+                return assignWordBinExpr(expr)
+            }
+            DataType.FLOAT -> {
+                require(expr.left.type==DataType.FLOAT && expr.right.type==DataType.FLOAT) {
+                    "both operands must be floats"
+                }
+                return assignFloatBinExpr(expr, assign)
+            }
+            else -> throw AssemblyError("weird expression type in assignment")
+        }
+    }
+
+    private fun assignWordBinExpr(expr: PtBinaryExpression): Boolean {
+        when(expr.operator) {
+            "+" -> {
+                TODO("word + at ${expr.position}")
+            }
+            "-" -> {
+                TODO("word - at ${expr.position}")
+            }
+            "*" -> {
+                TODO("word * at ${expr.position}")
+            }
+            "/" -> {
+                TODO("word / at ${expr.position}")
+            }
+            "<<" -> {
+                TODO("word << at ${expr.position}")
+            }
+            ">>" -> {
+                TODO("word >> at ${expr.position}")
+            }
+            "%" -> {
+                TODO("word % at ${expr.position}")
+            }
+            "&", "and" -> {
+                TODO("word and at ${expr.position}")
+            }
+            "|", "or" -> {
+                TODO("word or at ${expr.position}")
+            }
+            "^", "xor" -> {
+                TODO("word xor at ${expr.position}")
+            }
+            "==" -> {
+                TODO("word == at ${expr.position}")
+            }
+            "!=" -> {
+                TODO("word != at ${expr.position}")
+            }
+            "<" -> {
+                TODO("word < at ${expr.position}")
+            }
+            "<=" -> {
+                TODO("word <= at ${expr.position}")
+            }
+            ">" -> {
+                TODO("word > at ${expr.position}")
+            }
+            ">=" -> {
+                TODO("word >= at ${expr.position}")
+            }
+            else -> return false
+        }
+    }
+
+    private fun assignByteBinExpr(expr: PtBinaryExpression, assign: AsmAssignment): Boolean {
+        when(expr.operator) {
+            "+" -> {
+                asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
+                asmgen.out("  pha")
+                asmgen.assignExpressionToVariable(expr.right, "P8ZP_SCRATCH_B1", DataType.UBYTE)
+                asmgen.out("  pla |  clc |  adc  P8ZP_SCRATCH_B1")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "-" -> {
+                asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
+                asmgen.out("  pha")
+                asmgen.assignExpressionToVariable(expr.right, "P8ZP_SCRATCH_B1", DataType.UBYTE)
+                asmgen.out("  pla |  sec |  sbc  P8ZP_SCRATCH_B1")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "*" -> {
+                TODO("byte * at ${expr.position}")
+            }
+            "/" -> {
+                TODO("byte / at ${expr.position}")
+            }
+            "<<" -> {
+                TODO("byte << at ${expr.position}")
+            }
+            ">>" -> {
+                TODO("byte >> at ${expr.position}")
+            }
+            "%" -> {
+                TODO("byte % at ${expr.position}")
+            }
+            "&", "and" -> {
+                asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
+                asmgen.out("  pha")
+                asmgen.assignExpressionToVariable(expr.right, "P8ZP_SCRATCH_B1", DataType.UBYTE)
+                asmgen.out("  pla |  and  P8ZP_SCRATCH_B1")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "|", "or" -> {
+                asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
+                asmgen.out("  pha")
+                asmgen.assignExpressionToVariable(expr.right, "P8ZP_SCRATCH_B1", DataType.UBYTE)
+                asmgen.out("  pla |  ora  P8ZP_SCRATCH_B1")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "^", "xor" -> {
+                asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.A, false)
+                asmgen.out("  pha")
+                asmgen.assignExpressionToVariable(expr.right, "P8ZP_SCRATCH_B1", DataType.UBYTE)
+                asmgen.out("  pla |  eor  P8ZP_SCRATCH_B1")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "==" -> {
+                TODO("byte == at ${expr.position}")
+            }
+            "!=" -> {
+                TODO("byte != at ${expr.position}")
+            }
+            "<" -> {
+                TODO("byte < at ${expr.position}")
+            }
+            "<=" -> {
+                TODO("byte <= at ${expr.position}")
+            }
+            ">" -> {
+                TODO("byte > at ${expr.position}")
+            }
+            ">=" -> {
+                TODO("byte >= at ${expr.position}")
+            }
+            else -> return false
+        }
+    }
+
+    private fun assignFloatBinExpr(expr: PtBinaryExpression, assign: AsmAssignment): Boolean {
+        when(expr.operator) {
+            "+" -> {
+                assignFloatOperandsToFACandARG(expr.left, expr.right)
+                asmgen.out("  jsr  floats.FADDT")
+                asmgen.assignRegister(RegisterOrPair.FAC1, assign.target)
+                return true
+            }
+            "-" -> {
+                assignFloatOperandsToFACandARG(expr.right, expr.left)
+                asmgen.out("  jsr  floats.FSUBT")
+                asmgen.assignRegister(RegisterOrPair.FAC1, assign.target)
+                return true
+            }
+            "*" -> {
+                assignFloatOperandsToFACandARG(expr.left, expr.right)
+                asmgen.out("  jsr  floats.FMULTT")
+                asmgen.assignRegister(RegisterOrPair.FAC1, assign.target)
+                return true
+            }
+            "/" -> {
+                assignFloatOperandsToFACandARG(expr.right, expr.left)
+                asmgen.out("  jsr  floats.FDIVT")
+                asmgen.assignRegister(RegisterOrPair.FAC1, assign.target)
+                return true
+            }
+            "==" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_equal_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "!=" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_notequal_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "<" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_less_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            ">" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_greater_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            "<=" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_lesseq_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            ">=" -> {
+                setupFloatComparisonFAC1vsVarAY(expr)
+                asmgen.out("  jsr  floats.var_fac1_greatereq_f")
+                asmgen.assignRegister(RegisterOrPair.A, assign.target)
+                return true
+            }
+            else -> TODO("float expression operator ${expr.operator}")
+        }
+    }
+
+    private fun assignFloatOperandsToFACandARG(left: PtExpression, right: PtExpression) {
+        when(asmgen.options.compTarget.name) {
+            C64Target.NAME -> {
+                // c64 has a quirk: always make sure FAC2 is loaded last (done using CONUPK)  otherwise the result will be corrupt on C64
+                // this requires some more forced copying around of float values in certain cases
+                if (right.isSimple()) {
+                    asmgen.assignExpressionToRegister(left, RegisterOrPair.FAC1, true)
+                    asmgen.assignExpressionToRegister(right, RegisterOrPair.FAC2, true)
+                } else {
+                    asmgen.assignExpressionToRegister(right, RegisterOrPair.FAC1, true)
+                    asmgen.pushFAC1()
+                    asmgen.assignExpressionToRegister(left, RegisterOrPair.FAC1, true)
+                    asmgen.popFAC2()
+                }
+            }
+            Cx16Target.NAME -> {
+                asmgen.assignExpressionToRegister(left, RegisterOrPair.FAC1, true)
+                if (!right.isSimple()) asmgen.pushFAC1()
+                asmgen.assignExpressionToRegister(right, RegisterOrPair.FAC2, true)
+                if (!right.isSimple()) asmgen.popFAC1()
+            }
+            else -> TODO("don't know how to evaluate float expression for selected compilation target")
+        }
+    }
+
+    private fun setupFloatComparisonFAC1vsVarAY(expr: PtBinaryExpression) {
+        asmgen.assignExpressionToRegister(expr.left, RegisterOrPair.FAC1, true)
+        if(!expr.right.isSimple()) asmgen.pushFAC1()
+        asmgen.assignExpressionToVariable(expr.right, "floats.floats_temp_var", DataType.FLOAT)
+        if(!expr.right.isSimple()) asmgen.popFAC1()
+        asmgen.out("  lda  #<floats.floats_temp_var |  ldy  #>floats.floats_temp_var")
+    }
+}

codeGenCpu6502/src/prog8/codegen/cpu6502/assignment/AsmAssignment.kt

@@ -10,8 +10,7 @@ internal enum class TargetStorageKind {
     VARIABLE,
     ARRAY,
     MEMORY,
-    REGISTER,
-    STACK
+    REGISTER
 }
 
 internal enum class SourceStorageKind {
@@ -20,7 +19,6 @@ internal enum class SourceStorageKind {
     ARRAY,
     MEMORY,
     REGISTER,
-    STACK,              // value is already present on stack
     EXPRESSION,         // expression in ast-form, still to be evaluated
 }
 
@@ -28,6 +26,7 @@ internal class AsmAssignTarget(val kind: TargetStorageKind,
                                private val asmgen: AsmGen6502Internal,
                                val datatype: DataType,
                                val scope: IPtSubroutine?,
+                               val position: Position,
                                private val variableAsmName: String? = null,
                                val array: PtArrayIndexer? = null,
                                val memory: PtMemoryByte? = null,
@@ -43,8 +42,6 @@ internal class AsmAssignTarget(val kind: TargetStorageKind,
             asmgen.asmVariableName(array.variable)
     }
 
-    lateinit var origAssign: AsmAssignmentBase
-
     init {
         if(register!=null && datatype !in NumericDatatypes)
             throw AssemblyError("register must be integer or float type")
@@ -63,28 +60,28 @@ internal class AsmAssignTarget(val kind: TargetStorageKind,
                                 if(reg.statusflag!=null)
                                     throw AssemblyError("can't assign value to processor statusflag directly")
                                 else
-                                    return AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, type, definingSub, register=reg.registerOrPair, origAstTarget = this)
+                                    return AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, type, definingSub, target.position, register=reg.registerOrPair, origAstTarget = this)
                             }
                         }
-                        return AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, type, definingSub, variableAsmName = asmgen.asmVariableName(identifier!!), origAstTarget =  this)
+                        return AsmAssignTarget(TargetStorageKind.VARIABLE, asmgen, type, definingSub, target.position, variableAsmName = asmgen.asmVariableName(identifier!!), origAstTarget =  this)
                     }
-                    array != null -> return AsmAssignTarget(TargetStorageKind.ARRAY, asmgen, type, definingSub, array = array, origAstTarget =  this)
-                    memory != null -> return AsmAssignTarget(TargetStorageKind.MEMORY, asmgen, type, definingSub, memory =  memory, origAstTarget =  this)
+                    array != null -> return AsmAssignTarget(TargetStorageKind.ARRAY, asmgen, type, definingSub, target.position, array = array, origAstTarget =  this)
+                    memory != null -> return AsmAssignTarget(TargetStorageKind.MEMORY, asmgen, type, definingSub, target.position, memory =  memory, origAstTarget =  this)
                     else -> throw AssemblyError("weird target")
                 }
             }
         }
 
-        fun fromRegisters(registers: RegisterOrPair, signed: Boolean, scope: IPtSubroutine?, asmgen: AsmGen6502Internal): AsmAssignTarget =
+        fun fromRegisters(registers: RegisterOrPair, signed: Boolean, pos: Position, scope: IPtSubroutine?, asmgen: AsmGen6502Internal): AsmAssignTarget =
                 when(registers) {
                     RegisterOrPair.A,
                     RegisterOrPair.X,
-                    RegisterOrPair.Y -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.BYTE else DataType.UBYTE, scope, register = registers)
+                    RegisterOrPair.Y -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.BYTE else DataType.UBYTE, scope, pos, register = registers)
                     RegisterOrPair.AX,
                     RegisterOrPair.AY,
-                    RegisterOrPair.XY -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.WORD else DataType.UWORD, scope, register = registers)
+                    RegisterOrPair.XY -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.WORD else DataType.UWORD, scope, pos, register = registers)
                     RegisterOrPair.FAC1,
-                    RegisterOrPair.FAC2 -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, DataType.FLOAT, scope, register = registers)
+                    RegisterOrPair.FAC2 -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, DataType.FLOAT, scope, pos, register = registers)
                     RegisterOrPair.R0,
                     RegisterOrPair.R1,
                     RegisterOrPair.R2,
@@ -100,9 +97,31 @@ internal class AsmAssignTarget(val kind: TargetStorageKind,
                     RegisterOrPair.R12,
                     RegisterOrPair.R13,
                     RegisterOrPair.R14,
-                    RegisterOrPair.R15 -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.WORD else DataType.UWORD, scope, register = registers)
+                    RegisterOrPair.R15 -> AsmAssignTarget(TargetStorageKind.REGISTER, asmgen, if(signed) DataType.WORD else DataType.UWORD, scope, pos, register = registers)
                 }
     }
+
+    fun isSameAs(left: PtExpression): Boolean =
+        when(kind) {
+            TargetStorageKind.VARIABLE -> {
+                val scopedName: String = if('.' in asmVarname)
+                    asmVarname
+                else {
+                    val scopeName = (scope as? PtNamedNode)?.scopedName
+                    if (scopeName == null) asmVarname else "$scopeName.$asmVarname"
+                }
+                left is PtIdentifier && left.name==scopedName
+            }
+            TargetStorageKind.ARRAY -> {
+                left is PtArrayIndexer && left isSameAs array!! && left.splitWords==array.splitWords
+            }
+            TargetStorageKind.MEMORY -> {
+                left isSameAs memory!!
+            }
+            TargetStorageKind.REGISTER -> {
+                false
+            }
+        }
 }
 
 internal class AsmAssignSource(val kind: SourceStorageKind,
@@ -157,8 +176,8 @@ internal class AsmAssignSource(val kind: SourceStorageKind,
                     AsmAssignSource(SourceStorageKind.EXPRESSION, program, asmgen, value.type, expression = value)
                 }
                 is PtFunctionCall -> {
-                    val symbol = asmgen.symbolTable.lookup(value.name)
-                    val sub = symbol!!.astNode as IPtSubroutine
+                    val symbol = asmgen.symbolTable.lookup(value.name) ?: throw AssemblyError("lookup error ${value.name}")
+                    val sub = symbol.astNode as IPtSubroutine
                     val returnType = sub.returnsWhatWhere().firstOrNull { rr -> rr.first.registerOrPair != null || rr.first.statusflag!=null }?.second
                             ?: throw AssemblyError("can't translate zero return values in assignment")
 

codeGenCpu6502/test/Dummies.kt

@@ -34,11 +34,19 @@ internal class ErrorReporterForTests(private val throwExceptionAtReportIfErrors:
     val warnings = mutableListOf<String>()
 
     override fun err(msg: String, position: Position) {
-        errors.add("${position.toClickableStr()} $msg")
+        val text = "${position.toClickableStr()} $msg"
+        if(text !in errors)
+            errors.add(text)
     }
 
     override fun warn(msg: String, position: Position) {
-        warnings.add("${position.toClickableStr()} $msg")
+        val text = "${position.toClickableStr()} $msg"
+        if(text !in warnings)
+            warnings.add(text)
+    }
+
+    override fun undefined(symbol: List<String>, position: Position) {
+        err("undefined symbol: ${symbol.joinToString(".")}", position)
     }
 
     override fun noErrors(): Boolean  = errors.isEmpty()

codeGenCpu6502/test/TestCodegen.kt

@@ -1,6 +1,9 @@
 package prog8tests.codegencpu6502
 
+import io.kotest.assertions.throwables.shouldNotThrowAny
+import io.kotest.assertions.withClue
 import io.kotest.core.spec.style.FunSpec
+import io.kotest.matchers.ints.shouldBeGreaterThanOrEqual
 import io.kotest.matchers.shouldBe
 import prog8.code.SymbolTableMaker
 import prog8.code.ast.*
@@ -39,9 +42,9 @@ class TestCodegen: FunSpec({
 //        xx += cx16.r0
 //    }
 //}
-        val codegen = AsmGen6502()
+        val codegen = AsmGen6502(prefixSymbols = false)
         val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
-        val block = PtBlock("main", null, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
         val sub = PtSub("start", emptyList(), null, Position.DUMMY)
         sub.add(PtVariable("pi", DataType.UBYTE, ZeropageWish.DONTCARE, PtNumber(DataType.UBYTE, 0.0, Position.DUMMY), null, Position.DUMMY))
         sub.add(PtVariable("particleX", DataType.ARRAY_UB, ZeropageWish.DONTCARE, null, 3u, Position.DUMMY))
@@ -91,7 +94,7 @@ class TestCodegen: FunSpec({
         program.add(block)
 
         // define the "cx16.r0" virtual register
-        val cx16block = PtBlock("cx16", null, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val cx16block = PtBlock("cx16", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
         cx16block.add(PtMemMapped("r0", DataType.UWORD, 100u, null, Position.DUMMY))
         program.add(cx16block)
 
@@ -102,5 +105,24 @@ class TestCodegen: FunSpec({
         result.name shouldBe "test"
         Files.deleteIfExists(Path("${result.name}.asm"))
     }
+
+    test("64tass assembler available? - if this fails you need to install 64tass version 1.58 or newer in the path") {
+        val command = mutableListOf("64tass", "--version")
+        shouldNotThrowAny {
+            val proc = ProcessBuilder(command).start()
+            val output = String(proc.inputStream.readBytes())
+            val result = proc.waitFor()
+            result.shouldBe(0)
+            val (_, version) = output.split('V')
+            val (major, minor, _) = version.split('.')
+            val majorNum = major.toInt()
+            val minorNum = minor.toInt()
+            withClue("64tass version should be 1.58 or newer") {
+                majorNum shouldBeGreaterThanOrEqual 1
+                if (majorNum == 1)
+                    minorNum shouldBeGreaterThanOrEqual 58
+            }
+        }
+    }
 })
 

codeGenExperimental/build.gradle

@@ -29,17 +29,17 @@ dependencies {
     implementation project(':codeGenIntermediate')
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16"
+    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18"
 
 }
 
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
 }

codeGenExperimental/src/prog8/codegen/experimental/ExperiCodeGen.kt

@@ -15,7 +15,8 @@ class ExperiCodeGen: ICodeGeneratorBackend {
         symbolTable: SymbolTable,
         options: CompilationOptions,
         errors: IErrorReporter
-    ): IAssemblyProgram? {
+    ): IAssemblyProgram {
+
         // you could write a code generator directly on the PtProgram AST,
         // but you can also use the Intermediate Representation to build a codegen on:
         val irCodeGen = IRCodeGen(program, symbolTable, options, errors)
@@ -25,6 +26,15 @@ class ExperiCodeGen: ICodeGeneratorBackend {
         IRFileWriter(irProgram, null).write()
 
         println("** experimental codegen stub: no assembly generated **")
-        return null
+        return EmptyProgram
     }
-}
+}
+
+private object EmptyProgram : IAssemblyProgram {
+    override val name = "<Empty Program>"
+    override fun assemble(options: CompilationOptions, errors: IErrorReporter): Boolean {
+        println("** nothing assembled **")
+        return true
+    }
+
+}

codeGenIntermediate/build.gradle

@@ -28,18 +28,18 @@ dependencies {
     implementation project(':intermediate')
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16"
+    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18"
 
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.3.2'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.6.2'
 }
 
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
     test {

codeGenIntermediate/src/prog8/codegen/intermediate/AssignmentGen.kt

@@ -13,36 +13,36 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
         if(assignment.target.children.single() is PtMachineRegister)
             throw AssemblyError("assigning to a register should be done by just evaluating the expression into resultregister")
 
-        return translateRegularAssign(assignment)
+        val chunks = translateRegularAssign(assignment)
+        chunks.filterIsInstance<IRCodeChunk>().firstOrNull()?.appendSrcPosition(assignment.position)
+        return chunks
     }
 
-    internal fun translate(augmentedAssign: PtAugmentedAssign): IRCodeChunks {
-        if(augmentedAssign.target.children.single() is PtMachineRegister)
+    internal fun translate(augAssign: PtAugmentedAssign): IRCodeChunks {
+        if(augAssign.target.children.single() is PtMachineRegister)
             throw AssemblyError("assigning to a register should be done by just evaluating the expression into resultregister")
 
-        return translateInplaceAssign(augmentedAssign)
-    }
+        val ident = augAssign.target.identifier
+        val memory = augAssign.target.memory
+        val array = augAssign.target.array
 
-    private fun translateInplaceAssign(assignment: PtAugmentedAssign): IRCodeChunks {
-        val ident = assignment.target.identifier
-        val memory = assignment.target.memory
-        val array = assignment.target.array
-
-        return if(ident!=null) {
-            assignVarAugmented(ident.name, assignment)
+        val chunks = if(ident!=null) {
+            assignVarAugmented(ident.name, augAssign)
         } else if(memory != null) {
             if(memory.address is PtNumber)
-                assignMemoryAugmented((memory.address as PtNumber).number.toInt(), assignment)
+                assignMemoryAugmented((memory.address as PtNumber).number.toInt(), augAssign)
             else
-                fallbackAssign(assignment)
+                fallbackAssign(augAssign)
         } else if(array!=null) {
             // NOTE: naive fallback assignment here will sometimes generate code that loads the index value multiple times
             // in a register. It's way too much work to optimize that here - instead, we trust that the generated IL assembly
             // will be optimized later and have the double assignments removed.
-            fallbackAssign(assignment)
+            fallbackAssign(augAssign)
         } else {
-            fallbackAssign(assignment)
+            fallbackAssign(augAssign)
         }
+        chunks.filterIsInstance<IRCodeChunk>().firstOrNull()?.appendSrcPosition(augAssign.position)
+        return chunks
     }
 
     private fun assignMemoryAugmented(
@@ -50,45 +50,56 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
         assignment: PtAugmentedAssign
     ): IRCodeChunks {
         val value = assignment.value
-        val vmDt = codeGen.irType(value.type)
+        val vmDt = irType(value.type)
         return when(assignment.operator) {
-            "+" -> expressionEval.operatorPlusInplace(address, null, vmDt, value)
-            "-" -> expressionEval.operatorMinusInplace(address, null, vmDt, value)
-            "*" -> expressionEval.operatorMultiplyInplace(address, null, vmDt, value)
-            "/" -> expressionEval.operatorDivideInplace(address, null, vmDt, value.type in SignedDatatypes, value)
-            "|" -> expressionEval.operatorOrInplace(address, null, vmDt, value)
-            "&" -> expressionEval.operatorAndInplace(address, null, vmDt, value)
-            "^" -> expressionEval.operatorXorInplace(address, null, vmDt, value)
-            "<<" -> expressionEval.operatorShiftLeftInplace(address, null, vmDt, value)
-            ">>" -> expressionEval.operatorShiftRightInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            "+=" -> expressionEval.operatorPlusInplace(address, null, vmDt, value)
+            "-=" -> expressionEval.operatorMinusInplace(address, null, vmDt, value)
+            "*=" -> expressionEval.operatorMultiplyInplace(address, null, vmDt, value)
+            "/=" -> expressionEval.operatorDivideInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            "|=" -> expressionEval.operatorOrInplace(address, null, vmDt, value)
+            "&=" -> expressionEval.operatorAndInplace(address, null, vmDt, value)
+            "^=" -> expressionEval.operatorXorInplace(address, null, vmDt, value)
+            "<<=" -> expressionEval.operatorShiftLeftInplace(address, null, vmDt, value)
+            ">>=" -> expressionEval.operatorShiftRightInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            "%=" -> expressionEval.operatorModuloInplace(address, null, vmDt, value)
+            "==" -> expressionEval.operatorEqualsInplace(address, null, vmDt, value)
+            "!=" -> expressionEval.operatorNotEqualsInplace(address, null, vmDt, value)
+            "<" -> expressionEval.operatorLessInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            ">" -> expressionEval.operatorGreaterInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            "<=" -> expressionEval.operatorLessEqualInplace(address, null, vmDt, value.type in SignedDatatypes, value)
+            ">=" -> expressionEval.operatorGreaterEqualInplace(address, null, vmDt, value.type in SignedDatatypes, value)
             in PrefixOperators -> inplacePrefix(assignment.operator, vmDt, address, null)
+
             else -> throw AssemblyError("invalid augmented assign operator ${assignment.operator}")
         }
     }
 
     private fun assignVarAugmented(symbol: String, assignment: PtAugmentedAssign): IRCodeChunks {
         val value = assignment.value
-        val valueVmDt = codeGen.irType(value.type)
+        val targetDt = irType(assignment.target.type)
         return when (assignment.operator) {
-            "+=" -> expressionEval.operatorPlusInplace(null, symbol, valueVmDt, value)
-            "-=" -> expressionEval.operatorMinusInplace(null, symbol, valueVmDt, value)
-            "*=" -> expressionEval.operatorMultiplyInplace(null, symbol, valueVmDt, value)
-            "/=" -> expressionEval.operatorDivideInplace(null, symbol, valueVmDt, value.type in SignedDatatypes, value)
-            "|=" -> expressionEval.operatorOrInplace(null, symbol, valueVmDt, value)
-            "&=" -> expressionEval.operatorAndInplace(null, symbol, valueVmDt, value)
-            "^=" -> expressionEval.operatorXorInplace(null, symbol, valueVmDt, value)
-            "<<=" -> expressionEval.operatorShiftLeftInplace(null, symbol, valueVmDt, value)
-            ">>=" -> expressionEval.operatorShiftRightInplace(null, symbol, valueVmDt, value.type in SignedDatatypes, value)
-            in PrefixOperators -> inplacePrefix(assignment.operator, valueVmDt, null, symbol)
+            "+=" -> expressionEval.operatorPlusInplace(null, symbol, targetDt, value)
+            "-=" -> expressionEval.operatorMinusInplace(null, symbol, targetDt, value)
+            "*=" -> expressionEval.operatorMultiplyInplace(null, symbol, targetDt, value)
+            "/=" -> expressionEval.operatorDivideInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            "|=" -> expressionEval.operatorOrInplace(null, symbol, targetDt, value)
+            "&=" -> expressionEval.operatorAndInplace(null, symbol, targetDt, value)
+            "^=" -> expressionEval.operatorXorInplace(null, symbol, targetDt, value)
+            "<<=" -> expressionEval.operatorShiftLeftInplace(null, symbol, targetDt, value)
+            ">>=" -> expressionEval.operatorShiftRightInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            "%=" -> expressionEval.operatorModuloInplace(null, symbol, targetDt, value)
+            "==" -> expressionEval.operatorEqualsInplace(null, symbol, targetDt, value)
+            "!=" -> expressionEval.operatorNotEqualsInplace(null, symbol, targetDt, value)
+            "<" -> expressionEval.operatorLessInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            ">" -> expressionEval.operatorGreaterInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            "<=" -> expressionEval.operatorLessEqualInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            ">=" -> expressionEval.operatorGreaterEqualInplace(null, symbol, targetDt, value.type in SignedDatatypes, value)
+            in PrefixOperators -> inplacePrefix(assignment.operator, targetDt, null, symbol)
             else -> throw AssemblyError("invalid augmented assign operator ${assignment.operator}")
         }
     }
 
     private fun fallbackAssign(origAssign: PtAugmentedAssign): IRCodeChunks {
-        if (codeGen.options.slowCodegenWarnings)
-            codeGen.errors.warn("indirect code for in-place assignment", origAssign.position)
-        val normalAssign = PtAssignment(origAssign.position)
-        normalAssign.add(origAssign.target)
         val value: PtExpression
         if(origAssign.operator in PrefixOperators) {
             value = PtPrefix(origAssign.operator, origAssign.value.type, origAssign.value.position)
@@ -100,6 +111,8 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
             value.add(left)
             value.add(origAssign.value)
         }
+        val normalAssign = PtAssignment(origAssign.position)
+        normalAssign.add(origAssign.target)
         normalAssign.add(value)
         return translateRegularAssign(normalAssign)
     }
@@ -110,16 +123,16 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
             "+" -> { }
             "-" -> {
                 code += if(address!=null)
-                    IRInstruction(Opcode.NEGM, vmDt, value = address)
+                    IRInstruction(Opcode.NEGM, vmDt, address = address)
                 else
                     IRInstruction(Opcode.NEGM, vmDt, labelSymbol = symbol)
             }
             "~" -> {
                 val regMask = codeGen.registers.nextFree()
                 val mask = if(vmDt==IRDataType.BYTE) 0x00ff else 0xffff
-                code += IRInstruction(Opcode.LOAD, vmDt, reg1=regMask, value = mask)
+                code += IRInstruction(Opcode.LOAD, vmDt, reg1=regMask, immediate = mask)
                 code += if(address!=null)
-                    IRInstruction(Opcode.XORM, vmDt, reg1=regMask, value = address)
+                    IRInstruction(Opcode.XORM, vmDt, reg1=regMask, address = address)
                 else
                     IRInstruction(Opcode.XORM, vmDt, reg1=regMask, labelSymbol = symbol)
             }
@@ -130,119 +143,176 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
 
     private fun translateRegularAssign(assignment: PtAssignment): IRCodeChunks {
         // note: assigning array and string values is done via an explicit memcopy/stringcopy function call.
-        val ident = assignment.target.identifier
-        val memory = assignment.target.memory
-        val array = assignment.target.array
-        val vmDt = codeGen.irType(assignment.value.type)
+        val targetIdent = assignment.target.identifier
+        val targetMemory = assignment.target.memory
+        val targetArray = assignment.target.array
+        val valueDt = irType(assignment.value.type)
+        val targetDt = irType(assignment.target.type)
         val result = mutableListOf<IRCodeChunkBase>()
-        var resultRegister = -1
-        var resultFpRegister = -1
+
+        var valueRegister = -1
+        var valueFpRegister = -1
         val zero = codeGen.isZero(assignment.value)
         if(!zero) {
             // calculate the assignment value
-            if (vmDt == IRDataType.FLOAT) {
-                resultFpRegister = codeGen.registers.nextFreeFloat()
-                result += expressionEval.translateExpression(assignment.value, -1, resultFpRegister)
+            if (valueDt == IRDataType.FLOAT) {
+                val tr = expressionEval.translateExpression(assignment.value)
+                valueFpRegister = tr.resultFpReg
+                addToResult(result, tr, -1, valueFpRegister)
             } else {
-                resultRegister = if (assignment.value is PtMachineRegister) {
-                    (assignment.value as PtMachineRegister).register
+                val extendByteToWord = if(targetDt != valueDt) {
+                    // usually an error EXCEPT when a byte is assigned to a word.
+                    if(targetDt==IRDataType.WORD && valueDt==IRDataType.BYTE)
+                        true
+                    else
+                        throw AssemblyError("assignment value and target dt mismatch")
+                } else false
+                if (assignment.value is PtMachineRegister) {
+                    valueRegister = (assignment.value as PtMachineRegister).register
+                    if(extendByteToWord) {
+                        valueRegister = codeGen.registers.nextFree()
+                        addInstr(result, IRInstruction(Opcode.EXT, IRDataType.BYTE, reg1=valueRegister, reg2=(assignment.value as PtMachineRegister).register), null)
+                    }
                 } else {
-                    val reg = codeGen.registers.nextFree()
-                    result += expressionEval.translateExpression(assignment.value, reg, -1)
-                    reg
+                    val tr = expressionEval.translateExpression(assignment.value)
+                    valueRegister = tr.resultReg
+                    addToResult(result, tr, valueRegister, -1)
+                    if(extendByteToWord) {
+                        valueRegister = codeGen.registers.nextFree()
+                        val opcode = if(assignment.value.type in SignedDatatypes) Opcode.EXTS else Opcode.EXT
+                        addInstr(result, IRInstruction(opcode, IRDataType.BYTE, reg1=valueRegister, reg2=tr.resultReg), null)
+                    }
                 }
             }
         }
-        if(ident!=null) {
+
+        if(targetIdent!=null) {
             val instruction = if(zero) {
-                IRInstruction(Opcode.STOREZM, vmDt, labelSymbol = ident.name)
+                IRInstruction(Opcode.STOREZM, targetDt, labelSymbol = targetIdent.name)
             } else {
-                if (vmDt == IRDataType.FLOAT)
-                    IRInstruction(Opcode.STOREM, vmDt, fpReg1 = resultFpRegister, labelSymbol = ident.name)
+                if (targetDt == IRDataType.FLOAT)
+                    IRInstruction(Opcode.STOREM, targetDt, fpReg1 = valueFpRegister, labelSymbol = targetIdent.name)
                 else
-                    IRInstruction(Opcode.STOREM, vmDt, reg1 = resultRegister, labelSymbol = ident.name)
+                    IRInstruction(Opcode.STOREM, targetDt, reg1 = valueRegister, labelSymbol = targetIdent.name)
             }
             result += IRCodeChunk(null, null).also { it += instruction }
             return result
         }
-        else if(array!=null) {
-            val variable = array.variable.name
-            val itemsize = codeGen.program.memsizer.memorySize(array.type)
+        else if(targetArray!=null) {
+            val variable = targetArray.variable.name
+            val itemsize = codeGen.program.memsizer.memorySize(targetArray.type)
 
-            if(array.variable.type==DataType.UWORD) {
-                // indexing a pointer var instead of a real array or string
+            if(targetArray.usesPointerVariable) {
                 if(itemsize!=1)
                     throw AssemblyError("non-array var indexing requires bytes dt")
-                if(array.index.type!=DataType.UBYTE)
+                if(targetArray.index.type!=DataType.UBYTE)
                     throw AssemblyError("non-array var indexing requires bytes index")
-                val idxReg = codeGen.registers.nextFree()
-                result += expressionEval.translateExpression(array.index, idxReg, -1)
+                val tr = expressionEval.translateExpression(targetArray.index)
+                val idxReg = tr.resultReg
+                addToResult(result, tr, tr.resultReg, -1)
                 val code = IRCodeChunk(null, null)
                 if(zero) {
                     // there's no STOREZIX instruction
-                    resultRegister = codeGen.registers.nextFree()
-                    code += IRInstruction(Opcode.LOAD, vmDt, reg1=resultRegister, value=0)
+                    valueRegister = codeGen.registers.nextFree()
+                    code += IRInstruction(Opcode.LOAD, targetDt, reg1=valueRegister, immediate = 0)
                 }
-                code += IRInstruction(Opcode.STOREIX, vmDt, reg1=resultRegister, reg2=idxReg, labelSymbol = variable)
+                code += IRInstruction(Opcode.STOREIX, targetDt, reg1=valueRegister, reg2=idxReg, labelSymbol = variable)
                 result += code
                 return result
             }
 
-            val fixedIndex = constIntValue(array.index)
+            val fixedIndex = constIntValue(targetArray.index)
+            val arrayLength = codeGen.symbolTable.getLength(targetArray.variable.name)
             if(zero) {
                 if(fixedIndex!=null) {
-                    val offset = fixedIndex*itemsize
-                    val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZM, vmDt, labelSymbol = "$variable+$offset") }
+                    val chunk = IRCodeChunk(null, null).also {
+                        if(targetArray.splitWords) {
+                            it += IRInstruction(Opcode.STOREZM, IRDataType.BYTE, immediate = arrayLength, labelSymbol = "${variable}_lsb+$fixedIndex")
+                            it += IRInstruction(Opcode.STOREZM, IRDataType.BYTE, immediate = arrayLength, labelSymbol = "${variable}_msb+$fixedIndex")
+                        }
+                        else
+                            it += IRInstruction(Opcode.STOREZM, targetDt, labelSymbol = "$variable+${fixedIndex*itemsize}")
+                    }
                     result += chunk
                 } else {
-                    val indexReg = codeGen.registers.nextFree()
-                    result += loadIndexReg(array, itemsize, indexReg)
-                    result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZX, vmDt, reg1=indexReg, labelSymbol = variable) }
+                    val (code, indexReg) = loadIndexReg(targetArray, itemsize)
+                    result += code
+                    result += IRCodeChunk(null, null).also {
+                        if(targetArray.splitWords) {
+                            it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1 = indexReg, immediate = arrayLength, labelSymbol = variable+"_lsb")
+                            it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1 = indexReg, immediate = arrayLength, labelSymbol = variable+"_msb")
+                        }
+                        else
+                            it += IRInstruction(Opcode.STOREZX, targetDt, reg1=indexReg, labelSymbol = variable)
+                    }
                 }
             } else {
-                if(vmDt== IRDataType.FLOAT) {
+                if(targetDt== IRDataType.FLOAT) {
                     if(fixedIndex!=null) {
                         val offset = fixedIndex*itemsize
-                        val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREM, vmDt, fpReg1 = resultFpRegister, labelSymbol = "$variable+$offset") }
+                        val chunk = IRCodeChunk(null, null).also {
+                            it += IRInstruction(Opcode.STOREM, targetDt, fpReg1 = valueFpRegister, labelSymbol = "$variable+$offset")
+                        }
                         result += chunk
                     } else {
-                        val indexReg = codeGen.registers.nextFree()
-                        result += loadIndexReg(array, itemsize, indexReg)
-                        result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREX, vmDt, reg1 = indexReg, fpReg1 = resultFpRegister, labelSymbol = variable) }
+                        val (code, indexReg) = loadIndexReg(targetArray, itemsize)
+                        result += code
+                        result += IRCodeChunk(null, null).also {
+                            it += IRInstruction(Opcode.STOREX, targetDt, reg1 = indexReg, fpReg1 = valueFpRegister, labelSymbol = variable)
+                        }
                     }
                 } else {
                     if(fixedIndex!=null) {
-                        val offset = fixedIndex*itemsize
-                        val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREM, vmDt, reg1 = resultRegister, labelSymbol = "$variable+$offset") }
+                        val chunk = IRCodeChunk(null, null).also {
+                            if(targetArray.splitWords) {
+                                val msbReg = codeGen.registers.nextFree()
+                                it += IRInstruction(Opcode.STOREM, IRDataType.BYTE, reg1 = valueRegister, immediate = arrayLength, labelSymbol = "${variable}_lsb+$fixedIndex")
+                                it += IRInstruction(Opcode.MSIG, IRDataType.BYTE, reg1 = msbReg, reg2 = valueRegister)
+                                it += IRInstruction(Opcode.STOREM, IRDataType.BYTE, reg1 = msbReg, immediate = arrayLength, labelSymbol = "${variable}_msb+$fixedIndex")
+                            }
+                            else
+                                it += IRInstruction(Opcode.STOREM, targetDt, reg1 = valueRegister, labelSymbol = "$variable+${fixedIndex*itemsize}")
+                        }
                         result += chunk
                     } else {
-                        val indexReg = codeGen.registers.nextFree()
-                        result += loadIndexReg(array, itemsize, indexReg)
-                        result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREX, vmDt, reg1 = resultRegister, reg2=indexReg, labelSymbol = variable) }
+                        val (code, indexReg) = loadIndexReg(targetArray, itemsize)
+                        result += code
+                        result += IRCodeChunk(null, null).also {
+                            if(targetArray.splitWords) {
+                                val msbReg = codeGen.registers.nextFree()
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1 = valueRegister, reg2=indexReg, immediate = arrayLength, labelSymbol = "${variable}_lsb")
+                                it += IRInstruction(Opcode.MSIG, IRDataType.BYTE, reg1 = msbReg, reg2 = valueRegister)
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1 = msbReg, reg2=indexReg, immediate = arrayLength, labelSymbol = "${variable}_msb")
+                            }
+                            else
+                                it += IRInstruction(Opcode.STOREX, targetDt, reg1 = valueRegister, reg2=indexReg, labelSymbol = variable)
+                        }
                     }
                 }
             }
             return result
         }
-        else if(memory!=null) {
-            require(vmDt== IRDataType.BYTE) { "must be byte type ${memory.position}"}
+        else if(targetMemory!=null) {
+            require(targetDt == IRDataType.BYTE) { "must be byte type ${targetMemory.position}"}
             if(zero) {
-                if(memory.address is PtNumber) {
-                    val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZM, vmDt, value=(memory.address as PtNumber).number.toInt()) }
+                if(targetMemory.address is PtNumber) {
+                    val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZM, targetDt, address = (targetMemory.address as PtNumber).number.toInt()) }
                     result += chunk
                 } else {
-                    val addressReg = codeGen.registers.nextFree()
-                    result += expressionEval.translateExpression(memory.address, addressReg, -1)
-                    result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZI, vmDt, reg1=addressReg) }
+                    val tr = expressionEval.translateExpression(targetMemory.address)
+                    val addressReg = tr.resultReg
+                    addToResult(result, tr, tr.resultReg, -1)
+                    result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREZI, targetDt, reg1=addressReg) }
                 }
             } else {
-                if(memory.address is PtNumber) {
-                    val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREM, vmDt, reg1=resultRegister, value=(memory.address as PtNumber).number.toInt()) }
+                if(targetMemory.address is PtNumber) {
+                    val chunk = IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREM, targetDt, reg1=valueRegister, address=(targetMemory.address as PtNumber).number.toInt()) }
                     result += chunk
                 } else {
-                    val addressReg = codeGen.registers.nextFree()
-                    result += expressionEval.translateExpression(memory.address, addressReg, -1)
-                    result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREI, vmDt, reg1=resultRegister, reg2=addressReg) }
+                    val tr = expressionEval.translateExpression(targetMemory.address)
+                    val addressReg = tr.resultReg
+                    addToResult(result, tr, tr.resultReg, -1)
+                    result += IRCodeChunk(null, null).also { it += IRInstruction(Opcode.STOREI, targetDt, reg1=valueRegister, reg2=addressReg) }
                 }
             }
 
@@ -252,14 +322,22 @@ internal class AssignmentGen(private val codeGen: IRCodeGen, private val express
             throw AssemblyError("weird assigntarget")
     }
 
-    private fun loadIndexReg(array: PtArrayIndexer, itemsize: Int, indexReg: Int): IRCodeChunks {
-        return if(itemsize==1) {
-            expressionEval.translateExpression(array.index, indexReg, -1)
-        } else {
-            val mult = PtBinaryExpression("*", DataType.UBYTE, array.position)
-            mult.children += array.index
-            mult.children += PtNumber(DataType.UBYTE, itemsize.toDouble(), array.position)
-            expressionEval.translateExpression(mult, indexReg, -1)
+    private fun loadIndexReg(array: PtArrayIndexer, itemsize: Int): Pair<IRCodeChunks, Int> {
+        // returns the code to load the Index into the register, which is also returned.
+
+        val result = mutableListOf<IRCodeChunkBase>()
+        if(itemsize==1 || array.splitWords) {
+            val tr = expressionEval.translateExpression(array.index)
+            addToResult(result, tr, tr.resultReg, -1)
+            return Pair(result, tr.resultReg)
         }
+
+        val mult: PtExpression
+        mult = PtBinaryExpression("*", DataType.UBYTE, array.position)
+        mult.children += array.index
+        mult.children += PtNumber(DataType.UBYTE, itemsize.toDouble(), array.position)
+        val tr = expressionEval.translateExpression(mult)
+        addToResult(result, tr, tr.resultReg, -1)
+        return Pair(result, tr.resultReg)
     }
 }

codeGenIntermediate/src/prog8/codegen/intermediate/BuiltinFuncGen.kt

@@ -1,67 +1,145 @@
 package prog8.codegen.intermediate
 
-import prog8.code.StStaticVariable
 import prog8.code.ast.*
 import prog8.code.core.AssemblyError
 import prog8.code.core.DataType
+import prog8.code.core.SignedDatatypes
+import prog8.code.core.SplitWordArrayTypes
 import prog8.intermediate.*
 
 
 internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGen: ExpressionGen) {
 
-    fun translate(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    fun translate(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         return when(call.name) {
-            "any" -> funcAny(call, resultRegister)
-            "all" -> funcAll(call, resultRegister)
-            "abs" -> funcAbs(call, resultRegister)
+            "any" -> funcAny(call)
+            "all" -> funcAll(call)
+            "abs__byte", "abs__word", "abs__float" -> funcAbs(call)
             "cmp" -> funcCmp(call)
-            "sgn" -> funcSgn(call, resultRegister)
-            "sqrt16" -> funcSqrt16(call, resultRegister)
+            "sgn" -> funcSgn(call)
+            "sqrt__ubyte", "sqrt__uword", "sqrt__float" -> funcSqrt(call)
+            "divmod__ubyte" -> funcDivmod(call, IRDataType.BYTE)
+            "divmod__uword" -> funcDivmod(call, IRDataType.WORD)
             "pop" -> funcPop(call)
             "popw" -> funcPopw(call)
             "push" -> funcPush(call)
             "pushw" -> funcPushw(call)
-            "rsave",
-            "rsavex",
-            "rrestore",
-            "rrestorex" -> emptyList()  // vm doesn't have registers to save/restore
-            "callfar" -> throw AssemblyError("callfar() is for cx16 target only")
-            "msb" -> funcMsb(call, resultRegister)
-            "lsb" -> funcLsb(call, resultRegister)
-            "memory" -> funcMemory(call, resultRegister)
-            "peek" -> funcPeek(call, resultRegister)
-            "peekw" -> funcPeekW(call, resultRegister)
+            "rsave", "rrestore" -> ExpressionCodeResult.EMPTY  // vm doesn't have registers to save/restore
+            "callfar" -> funcCallfar(call)
+            "msb" -> funcMsb(call)
+            "lsb" -> funcLsb(call)
+            "memory" -> funcMemory(call)
+            "peek" -> funcPeek(call)
+            "peekw" -> funcPeekW(call)
             "poke" -> funcPoke(call)
             "pokew" -> funcPokeW(call)
-            "pokemon" -> emptyList()
-            "mkword" -> funcMkword(call, resultRegister)
+            "pokemon" -> ExpressionCodeResult.EMPTY     // easter egg function
+            "mkword" -> funcMkword(call)
+            "clamp__byte", "clamp__ubyte", "clamp__word", "clamp__uword" -> funcClamp(call)
+            "min__byte", "min__ubyte", "min__word", "min__uword" -> funcMin(call)
+            "max__byte", "max__ubyte", "max__word", "max__uword" -> funcMax(call)
             "sort" -> funcSort(call)
             "reverse" -> funcReverse(call)
-            "rol" -> funcRolRor(Opcode.ROXL, call, resultRegister)
-            "ror" -> funcRolRor(Opcode.ROXR, call, resultRegister)
-            "rol2" -> funcRolRor(Opcode.ROL, call, resultRegister)
-            "ror2" -> funcRolRor(Opcode.ROR, call, resultRegister)
+            "setlsb" -> funcSetLsbMsb(call, false)
+            "setmsb" -> funcSetLsbMsb(call, true)
+            "rol" -> funcRolRor(Opcode.ROXL, call)
+            "ror" -> funcRolRor(Opcode.ROXR, call)
+            "rol2" -> funcRolRor(Opcode.ROL, call)
+            "ror2" -> funcRolRor(Opcode.ROR, call)
+            "prog8_lib_stringcompare" -> funcStringCompare(call)
+            "prog8_lib_square_byte" -> funcSquare(call, IRDataType.BYTE)
+            "prog8_lib_square_word" -> funcSquare(call, IRDataType.WORD)
             else -> throw AssemblyError("missing builtinfunc for ${call.name}")
         }
     }
 
-    private fun funcCmp(call: PtBuiltinFunctionCall): IRCodeChunks {
-        val leftRegister = codeGen.registers.nextFree()
-        val rightRegister = codeGen.registers.nextFree()
+    private fun funcSquare(call: PtBuiltinFunctionCall, resultType: IRDataType): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], leftRegister, -1)
-        result += exprGen.translateExpression(call.args[1], rightRegister, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.CMP, codeGen.irType(call.args[0].type), reg1=leftRegister, reg2=rightRegister)
+        val valueTr = exprGen.translateExpression(call.args[0])
+        addToResult(result, valueTr, valueTr.resultReg, valueTr.resultFpReg)
+        return if(resultType==IRDataType.FLOAT) {
+            val resultFpReg = codeGen.registers.nextFreeFloat()
+            addInstr(result, IRInstruction(Opcode.SQUARE, resultType, fpReg1 = resultFpReg, fpReg2 = valueTr.resultFpReg), null)
+            ExpressionCodeResult(result, resultType, -1, resultFpReg)
+        }
+        else {
+            val resultReg = codeGen.registers.nextFree()
+            addInstr(result, IRInstruction(Opcode.SQUARE, resultType, reg1 = resultReg, reg2 = valueTr.resultReg), null)
+            ExpressionCodeResult(result, resultType, resultReg, -1)
         }
-        return result
     }
 
-    private fun funcAny(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcCallfar(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 3), null)
+        val bankTr = exprGen.translateExpression(call.args[0])
+        val addressTr = exprGen.translateExpression(call.args[1])
+        val argumentwordTr = exprGen.translateExpression(call.args[2])
+        addToResult(result, bankTr, bankTr.resultReg, -1)
+        addToResult(result, addressTr, addressTr.resultReg, -1)
+        addToResult(result, argumentwordTr, argumentwordTr.resultReg, -1)
+        result += codeGen.makeSyscall(IMSyscall.CALLFAR, listOf(IRDataType.BYTE to bankTr.resultReg, IRDataType.WORD to addressTr.resultReg, IRDataType.WORD to argumentwordTr.resultReg), IRDataType.WORD to argumentwordTr.resultReg)
+        return ExpressionCodeResult(result, IRDataType.WORD, argumentwordTr.resultReg, -1)
+    }
+
+    private fun funcDivmod(call: PtBuiltinFunctionCall, type: IRDataType): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        val number = call.args[0]
+        val divident = call.args[1]
+        val divisionReg: Int
+        val remainderReg: Int
+        if(divident is PtNumber) {
+            val tr = exprGen.translateExpression(number)
+            addToResult(result, tr, tr.resultReg, -1)
+            addInstr(result, IRInstruction(Opcode.DIVMOD, type, reg1 = tr.resultReg, immediate = divident.number.toInt()), null)
+            divisionReg = tr.resultReg
+            remainderReg = codeGen.registers.nextFree()
+        } else {
+            val numTr = exprGen.translateExpression(number)
+            addToResult(result, numTr, numTr.resultReg, -1)
+            val dividentTr = exprGen.translateExpression(divident)
+            addToResult(result, dividentTr, dividentTr.resultReg, -1)
+            addInstr(result, IRInstruction(Opcode.DIVMODR, type, reg1 = numTr.resultReg, reg2=dividentTr.resultReg), null)
+            divisionReg = numTr.resultReg
+            remainderReg = dividentTr.resultReg
+        }
+        // DIVMOD result convention: on value stack, division and remainder on top.
+        addInstr(result, IRInstruction(Opcode.POP, type, reg1=remainderReg), null)
+        addInstr(result, IRInstruction(Opcode.POP, type, reg1=divisionReg), null)
+        result += assignRegisterTo(call.args[2], divisionReg)
+        result += assignRegisterTo(call.args[3], remainderReg)
+        return ExpressionCodeResult(result, type, -1, -1)
+    }
+
+    private fun funcStringCompare(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
+        val left  = exprGen.translateExpression(call.args[0])
+        val right = exprGen.translateExpression(call.args[1])
+        addToResult(result, left, left.resultReg, -1)
+        addToResult(result, right, right.resultReg, -1)
+        result += codeGen.makeSyscall(IMSyscall.COMPARE_STRINGS, listOf(IRDataType.WORD to left.resultReg, IRDataType.WORD to right.resultReg), IRDataType.BYTE to left.resultReg)
+        return ExpressionCodeResult(result, IRDataType.BYTE, left.resultReg, -1)
+    }
+
+    private fun funcCmp(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        val leftTr = exprGen.translateExpression(call.args[0])
+        addToResult(result, leftTr, leftTr.resultReg, -1)
+        val rightTr = exprGen.translateExpression(call.args[1])
+        addToResult(result, rightTr, rightTr.resultReg, -1)
+        val dt = irType(call.args[0].type)
+        result += IRCodeChunk(null, null).also {
+            it += IRInstruction(Opcode.CMP, dt, reg1=leftTr.resultReg, reg2=rightTr.resultReg)
+        }
+        return ExpressionCodeResult(result, dt, leftTr.resultReg, -1)
+    }
+
+    private fun funcAny(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val arrayName = call.args[0] as PtIdentifier
-        val array = codeGen.symbolTable.flat.getValue(arrayName.name) as StStaticVariable
+        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
         val syscall =
-            when (array.dt) {
+            when (arrayName.type) {
                 DataType.ARRAY_UB,
                 DataType.ARRAY_B -> IMSyscall.ANY_BYTE
                 DataType.ARRAY_UW,
@@ -70,21 +148,20 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
                 else -> throw IllegalArgumentException("weird type")
             }
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], SyscallRegisterBase, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = SyscallRegisterBase+1, value = array.length)
-            it += IRInstruction(Opcode.SYSCALL, value = syscall.number)
-            if(resultRegister!=0)
-                it += IRInstruction(Opcode.LOADR, IRDataType.BYTE, reg1 = resultRegister, reg2 = 0)
-        }
-        return result
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
+        val lengthReg = codeGen.registers.nextFree()
+        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength), null)
+        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to tr.resultReg)
+        return ExpressionCodeResult(result, IRDataType.BYTE, tr.resultReg, -1)
     }
 
-    private fun funcAll(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcAll(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val arrayName = call.args[0] as PtIdentifier
-        val array = codeGen.symbolTable.flat.getValue(arrayName.name) as StStaticVariable
+        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
         val syscall =
-            when(array.dt) {
+            when(arrayName.type) {
                 DataType.ARRAY_UB,
                 DataType.ARRAY_B -> IMSyscall.ALL_BYTE
                 DataType.ARRAY_UW,
@@ -93,303 +170,537 @@ internal class BuiltinFuncGen(private val codeGen: IRCodeGen, private val exprGe
                 else -> throw IllegalArgumentException("weird type")
             }
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], SyscallRegisterBase, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = SyscallRegisterBase+1, value = array.length)
-            it += IRInstruction(Opcode.SYSCALL, value = syscall.number)
-            if(resultRegister!=0)
-                it += IRInstruction(Opcode.LOADR, IRDataType.BYTE, reg1 = resultRegister, reg2 = 0)
-        }
-        return result
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
+        val lengthReg = codeGen.registers.nextFree()
+        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = arrayLength), null)
+        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), IRDataType.BYTE to tr.resultReg)
+        return ExpressionCodeResult(result, IRDataType.BYTE, tr.resultReg, -1)
     }
 
-    private fun funcAbs(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcAbs(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val sourceDt = call.args.single().type
         val result = mutableListOf<IRCodeChunkBase>()
-        if(sourceDt!=DataType.UWORD) {
-            result += exprGen.translateExpression(call.args[0], resultRegister, -1)
-            when (sourceDt) {
-                DataType.UBYTE -> {
-                    result += IRCodeChunk(null, null).also {
-                        it += IRInstruction(Opcode.EXT, IRDataType.BYTE, reg1 = resultRegister)
-                    }
+        if(sourceDt==DataType.UWORD)
+            return ExpressionCodeResult.EMPTY
+
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
+        when (sourceDt) {
+            DataType.BYTE -> {
+                val notNegativeLabel = codeGen.createLabelName()
+                val compareReg = codeGen.registers.nextFree()
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.LOADR, IRDataType.BYTE, reg1=compareReg, reg2=tr.resultReg)
+                    it += IRInstruction(Opcode.BSTPOS, labelSymbol = notNegativeLabel)
+                    it += IRInstruction(Opcode.NEG, IRDataType.BYTE, reg1=tr.resultReg)
                 }
-                DataType.BYTE -> {
-                    val notNegativeLabel = codeGen.createLabelName()
-                    val compareReg = codeGen.registers.nextFree()
-                    result += IRCodeChunk(null, null).also {
-                        it += IRInstruction(Opcode.LOADR, IRDataType.BYTE, reg1=compareReg, reg2=resultRegister)
-                        it += IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=compareReg, value=0x80)
-                        it += IRInstruction(Opcode.BZ, IRDataType.BYTE, reg1=compareReg, labelSymbol = notNegativeLabel)
-                        it += IRInstruction(Opcode.NEG, IRDataType.BYTE, reg1=resultRegister)
-                        it += IRInstruction(Opcode.EXT, IRDataType.BYTE, reg1=resultRegister)
-                    }
-                    result += IRCodeChunk(notNegativeLabel, null)
-                }
-                DataType.WORD -> {
-                    val notNegativeLabel = codeGen.createLabelName()
-                    val compareReg = codeGen.registers.nextFree()
-                    result += IRCodeChunk(null, null).also {
-                        it += IRInstruction(Opcode.LOADR, IRDataType.WORD, reg1=compareReg, reg2=resultRegister)
-                        it += IRInstruction(Opcode.AND, IRDataType.WORD, reg1=compareReg, value=0x8000)
-                        it += IRInstruction(Opcode.BZ, IRDataType.WORD, reg1=compareReg, labelSymbol = notNegativeLabel)
-                        it += IRInstruction(Opcode.NEG, IRDataType.WORD, reg1=resultRegister)
-                    }
-                    result += IRCodeChunk(notNegativeLabel, null)
-                }
-                else -> throw AssemblyError("weird type")
+                result += IRCodeChunk(notNegativeLabel, null)
+                return ExpressionCodeResult(result, IRDataType.BYTE, tr.resultReg, -1)
             }
+            DataType.WORD -> {
+                val notNegativeLabel = codeGen.createLabelName()
+                val compareReg = codeGen.registers.nextFree()
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.LOADR, IRDataType.WORD, reg1=compareReg, reg2=tr.resultReg)
+                    it += IRInstruction(Opcode.BSTPOS, labelSymbol = notNegativeLabel)
+                    it += IRInstruction(Opcode.NEG, IRDataType.WORD, reg1=tr.resultReg)
+                }
+                result += IRCodeChunk(notNegativeLabel, null)
+                return ExpressionCodeResult(result, IRDataType.WORD, tr.resultReg, -1)
+            }
+            DataType.FLOAT -> {
+                val resultFpReg = codeGen.registers.nextFreeFloat()
+                addInstr(result, IRInstruction(Opcode.FABS, IRDataType.FLOAT, fpReg1 = resultFpReg, fpReg2 = tr.resultFpReg), null)
+                return ExpressionCodeResult(result, IRDataType.FLOAT, -1, resultFpReg)
+            }
+            else -> throw AssemblyError("weird dt")
         }
-        return result
     }
 
-    private fun funcSgn(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
-        val reg = codeGen.registers.nextFree()
+    private fun funcSgn(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args.single(), reg, -1)
+        val vmDt = irType(call.type)
+        val tr = exprGen.translateExpression(call.args.single())
+        addToResult(result, tr, tr.resultReg, -1)
+        val resultReg = codeGen.registers.nextFree()
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.SGN, codeGen.irType(call.type), reg1 = resultRegister, reg2 = reg)
+            it += IRInstruction(Opcode.SGN, vmDt, reg1 = resultReg, reg2 = tr.resultReg)
         }
-        return result
+        return ExpressionCodeResult(result, vmDt, resultReg, -1)
     }
 
-    private fun funcSqrt16(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
-        val reg = codeGen.registers.nextFree()
+    private fun funcSqrt(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args.single(), reg, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.SQRT, IRDataType.WORD, reg1=resultRegister, reg2=reg)
+        val tr = exprGen.translateExpression(call.args.single())
+        val dt = call.args[0].type
+        when(dt) {
+            DataType.UBYTE -> {
+                addToResult(result, tr, tr.resultReg, -1)
+                val resultReg = codeGen.registers.nextFree()
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.SQRT, IRDataType.BYTE, reg1=resultReg, reg2=tr.resultReg)
+                }
+                return ExpressionCodeResult(result, IRDataType.BYTE, resultReg, -1)
+            }
+            DataType.UWORD -> {
+                addToResult(result, tr, tr.resultReg, -1)
+                val resultReg = codeGen.registers.nextFree()
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.SQRT, IRDataType.WORD, reg1=resultReg, reg2=tr.resultReg)
+                }
+                return ExpressionCodeResult(result, IRDataType.WORD, resultReg, -1)
+            }
+            DataType.FLOAT -> {
+                addToResult(result, tr, -1, tr.resultFpReg)
+                val resultFpReg = codeGen.registers.nextFreeFloat()
+                result += IRCodeChunk(null, null).also {
+                    it += IRInstruction(Opcode.SQRT, IRDataType.FLOAT, fpReg1 = resultFpReg, fpReg2 = tr.resultFpReg)
+                }
+                return ExpressionCodeResult(result, IRDataType.FLOAT, -1, resultFpReg)
+            }
+            else -> throw AssemblyError("invalid dt for sqrt")
         }
-        return result
     }
 
-    private fun funcPop(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcPop(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val code = IRCodeChunk(null, null)
         val reg = codeGen.registers.nextFree()
         code += IRInstruction(Opcode.POP, IRDataType.BYTE, reg1=reg)
         val result = mutableListOf<IRCodeChunkBase>(code)
         result += assignRegisterTo(call.args.single(), reg)
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, reg, -1)
     }
 
-    private fun funcPopw(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcPopw(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val code = IRCodeChunk(null, null)
         val reg = codeGen.registers.nextFree()
         code += IRInstruction(Opcode.POP, IRDataType.WORD, reg1=reg)
         val result = mutableListOf<IRCodeChunkBase>(code)
         result += assignRegisterTo(call.args.single(), reg)
-        return result
+        return ExpressionCodeResult(result, IRDataType.WORD, reg, -1)
     }
 
-    private fun funcPush(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcPush(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        val reg = codeGen.registers.nextFree()
-        result += exprGen.translateExpression(call.args.single(), reg, -1)
+        val tr = exprGen.translateExpression(call.args.single())
+        addToResult(result, tr, tr.resultReg, -1)
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.PUSH, IRDataType.BYTE, reg1=reg)
+            it += IRInstruction(Opcode.PUSH, IRDataType.BYTE, reg1=tr.resultReg)
         }
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcPushw(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcPushw(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        val reg = codeGen.registers.nextFree()
-        result += exprGen.translateExpression(call.args.single(), reg, -1)
+        val tr = exprGen.translateExpression(call.args.single())
+        addToResult(result, tr, tr.resultReg, -1)
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.PUSH, IRDataType.WORD, reg1 = reg)
+            it += IRInstruction(Opcode.PUSH, IRDataType.WORD, reg1 = tr.resultReg)
         }
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcReverse(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcReverse(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val arrayName = call.args[0] as PtIdentifier
-        val array = codeGen.symbolTable.flat.getValue(arrayName.name) as StStaticVariable
+        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
         val syscall =
-            when(array.dt) {
+            when(arrayName.type) {
                 DataType.ARRAY_UB, DataType.ARRAY_B, DataType.STR -> IMSyscall.REVERSE_BYTES
                 DataType.ARRAY_UW, DataType.ARRAY_W -> IMSyscall.REVERSE_WORDS
                 DataType.ARRAY_F -> IMSyscall.REVERSE_FLOATS
+                in SplitWordArrayTypes -> TODO("split word reverse")
                 else -> throw IllegalArgumentException("weird type to reverse")
             }
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], SyscallRegisterBase, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = SyscallRegisterBase+1, value = array.length)
-            it += IRInstruction(Opcode.SYSCALL, value = syscall.number)
-        }
-        return result
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
+        val lengthReg = codeGen.registers.nextFree()
+        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
+        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), null)
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcSort(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcSort(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val arrayName = call.args[0] as PtIdentifier
-        val array = codeGen.symbolTable.flat.getValue(arrayName.name) as StStaticVariable
+        val arrayLength = codeGen.symbolTable.getLength(arrayName.name)
         val syscall =
-            when(array.dt) {
+            when(arrayName.type) {
                 DataType.ARRAY_UB -> IMSyscall.SORT_UBYTE
                 DataType.ARRAY_B -> IMSyscall.SORT_BYTE
                 DataType.ARRAY_UW -> IMSyscall.SORT_UWORD
                 DataType.ARRAY_W -> IMSyscall.SORT_WORD
                 DataType.STR -> IMSyscall.SORT_UBYTE
                 DataType.ARRAY_F -> throw IllegalArgumentException("sorting a floating point array is not supported")
+                in SplitWordArrayTypes -> TODO("split word sort")
                 else -> throw IllegalArgumentException("weird type to sort")
             }
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], SyscallRegisterBase, -1)
-        result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = SyscallRegisterBase+1, value = array.length)
-            it += IRInstruction(Opcode.SYSCALL, value = syscall.number)
-        }
-        return result
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 2), null)
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
+        val lengthReg = codeGen.registers.nextFree()
+        addInstr(result, IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1 = lengthReg, immediate = if(arrayName.type==DataType.STR) arrayLength!!-1 else arrayLength), null)
+        result += codeGen.makeSyscall(syscall, listOf(IRDataType.WORD to tr.resultReg, IRDataType.BYTE to lengthReg), null)
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcMkword(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
-        val msbReg = codeGen.registers.nextFree()
+    private fun funcMkword(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], msbReg, -1)
-        result += exprGen.translateExpression(call.args[1], resultRegister, -1)
+        val msbTr = exprGen.translateExpression(call.args[0])
+        addToResult(result, msbTr, msbTr.resultReg, -1)
+        val lsbTr = exprGen.translateExpression(call.args[1])
+        addToResult(result, lsbTr, lsbTr.resultReg, -1)
+        val resultReg = codeGen.registers.nextFree()
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.CONCAT, IRDataType.BYTE, reg1 = resultRegister, reg2 = msbReg)
+            it += IRInstruction(Opcode.CONCAT, IRDataType.BYTE, reg1=resultReg, reg2 = lsbTr.resultReg, reg3 = msbTr.resultReg)
         }
-        return result
+        return ExpressionCodeResult(result, IRDataType.WORD, resultReg, -1)
     }
 
-    private fun funcPokeW(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcClamp(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        addInstr(result, IRInstruction(Opcode.PREPARECALL, immediate = 3), null)
+        val type = irType(call.type)
+        val valueTr = exprGen.translateExpression(call.args[0])
+        val minimumTr = exprGen.translateExpression(call.args[1])
+        val maximumTr = exprGen.translateExpression(call.args[2])
+        result += valueTr.chunks
+        result += minimumTr.chunks
+        result += maximumTr.chunks
+        if(type==IRDataType.FLOAT) {
+            result += codeGen.makeSyscall(
+                IMSyscall.CLAMP_FLOAT, listOf(
+                    valueTr.dt to valueTr.resultFpReg,
+                    minimumTr.dt to minimumTr.resultFpReg,
+                    maximumTr.dt to maximumTr.resultFpReg,
+                ), type to valueTr.resultFpReg
+            )
+            return ExpressionCodeResult(result, type, -1, valueTr.resultFpReg)
+        } else {
+            val syscall = when(call.type) {
+                DataType.UBYTE -> IMSyscall.CLAMP_UBYTE
+                DataType.BYTE -> IMSyscall.CLAMP_BYTE
+                DataType.UWORD -> IMSyscall.CLAMP_UWORD
+                DataType.WORD -> IMSyscall.CLAMP_WORD
+                else -> throw AssemblyError("invalid dt")
+            }
+            result += codeGen.makeSyscall(syscall, listOf(
+                    valueTr.dt to valueTr.resultReg,
+                    minimumTr.dt to minimumTr.resultReg,
+                    maximumTr.dt to maximumTr.resultReg,
+                ), type to valueTr.resultReg
+            )
+            return ExpressionCodeResult(result, type, valueTr.resultReg, -1)
+        }
+    }
+
+    private fun funcMin(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val type = irType(call.type)
+        val result = mutableListOf<IRCodeChunkBase>()
+        val leftTr = exprGen.translateExpression(call.args[0])
+        addToResult(result, leftTr, leftTr.resultReg, -1)
+        val rightTr = exprGen.translateExpression(call.args[1])
+        addToResult(result, rightTr, rightTr.resultReg, -1)
+        val comparisonOpcode = if(call.type in SignedDatatypes) Opcode.BGTSR else Opcode.BGTR
+        val after = codeGen.createLabelName()
+        result += IRCodeChunk(null, null).also {
+            it += IRInstruction(comparisonOpcode, type, reg1 = rightTr.resultReg, reg2 = leftTr.resultReg, labelSymbol = after)
+            // right <= left, take right
+            it += IRInstruction(Opcode.LOADR, type, reg1=leftTr.resultReg, reg2=rightTr.resultReg)
+            it += IRInstruction(Opcode.JUMP, labelSymbol = after)
+        }
+        result += IRCodeChunk(after, null)
+        return ExpressionCodeResult(result, type, leftTr.resultReg, -1)
+    }
+
+    private fun funcMax(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val type = irType(call.type)
+        val result = mutableListOf<IRCodeChunkBase>()
+        val leftTr = exprGen.translateExpression(call.args[0])
+        addToResult(result, leftTr, leftTr.resultReg, -1)
+        val rightTr = exprGen.translateExpression(call.args[1])
+        addToResult(result, rightTr, rightTr.resultReg, -1)
+        val comparisonOpcode = if(call.type in SignedDatatypes) Opcode.BGTSR else Opcode.BGTR
+        val after = codeGen.createLabelName()
+        result += IRCodeChunk(null, null).also {
+            it += IRInstruction(comparisonOpcode, type, reg1 = leftTr.resultReg, reg2 = rightTr.resultReg, labelSymbol = after)
+            // right >= left, take right
+            it += IRInstruction(Opcode.LOADR, type, reg1=leftTr.resultReg, reg2=rightTr.resultReg)
+            it += IRInstruction(Opcode.JUMP, labelSymbol = after)
+        }
+        result += IRCodeChunk(after, null)
+        return ExpressionCodeResult(result, type, leftTr.resultReg, -1)
+    }
+
+    private fun funcPokeW(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
         if(codeGen.isZero(call.args[1])) {
             if (call.args[0] is PtNumber) {
                 val address = (call.args[0] as PtNumber).number.toInt()
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREZM, IRDataType.WORD, value = address)
+                    it += IRInstruction(Opcode.STOREZM, IRDataType.WORD, address = address)
                 }
             } else {
-                val addressReg = codeGen.registers.nextFree()
-                result += exprGen.translateExpression(call.args[0], addressReg, -1)
+                val tr = exprGen.translateExpression(call.args[0])
+                addToResult(result, tr, tr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREZI, IRDataType.WORD, reg2 = addressReg)
+                    it += IRInstruction(Opcode.STOREZI, IRDataType.WORD, reg1 = tr.resultReg)
                 }
             }
         } else {
-            val valueReg = codeGen.registers.nextFree()
             if (call.args[0] is PtNumber) {
                 val address = (call.args[0] as PtNumber).number.toInt()
-                result += exprGen.translateExpression(call.args[1], valueReg, -1)
+                val tr = exprGen.translateExpression(call.args[1])
+                addToResult(result, tr, tr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREM, IRDataType.WORD, reg1 = valueReg, value = address)
+                    it += IRInstruction(Opcode.STOREM, IRDataType.WORD, reg1 = tr.resultReg, address = address)
                 }
             } else {
-                val addressReg = codeGen.registers.nextFree()
-                result += exprGen.translateExpression(call.args[0], addressReg, -1)
-                result += exprGen.translateExpression(call.args[1], valueReg, -1)
+                val addressTr = exprGen.translateExpression(call.args[0])
+                addToResult(result, addressTr, addressTr.resultReg, -1)
+                val valueTr = exprGen.translateExpression(call.args[1])
+                addToResult(result, valueTr, valueTr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREI, IRDataType.WORD, reg1 = valueReg, reg2 = addressReg)
+                    it += IRInstruction(Opcode.STOREI, IRDataType.WORD, reg1 = valueTr.resultReg, reg2 = addressTr.resultReg)
                 }
             }
         }
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcPoke(call: PtBuiltinFunctionCall): IRCodeChunks {
+    private fun funcPoke(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
         if(codeGen.isZero(call.args[1])) {
             if (call.args[0] is PtNumber) {
                 val address = (call.args[0] as PtNumber).number.toInt()
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREZM, IRDataType.BYTE, value = address)
+                    it += IRInstruction(Opcode.STOREZM, IRDataType.BYTE, address = address)
                 }
             } else {
-                val addressReg = codeGen.registers.nextFree()
-                result += exprGen.translateExpression(call.args[0], addressReg, -1)
+                val tr = exprGen.translateExpression(call.args[0])
+                addToResult(result, tr, tr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREZI, IRDataType.BYTE, reg2 = addressReg)
+                    it += IRInstruction(Opcode.STOREZI, IRDataType.BYTE, reg1 = tr.resultReg)
                 }
             }
         } else {
-            val valueReg = codeGen.registers.nextFree()
             if (call.args[0] is PtNumber) {
                 val address = (call.args[0] as PtNumber).number.toInt()
-                result += exprGen.translateExpression(call.args[1], valueReg, -1)
+                val tr = exprGen.translateExpression(call.args[1])
+                addToResult(result, tr, tr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREM, IRDataType.BYTE, reg1 = valueReg, value = address)
+                    it += IRInstruction(Opcode.STOREM, IRDataType.BYTE, reg1 = tr.resultReg, address = address)
                 }
             } else {
-                val addressReg = codeGen.registers.nextFree()
-                result += exprGen.translateExpression(call.args[0], addressReg, -1)
-                result += exprGen.translateExpression(call.args[1], valueReg, -1)
+                val addressTr = exprGen.translateExpression(call.args[0])
+                addToResult(result, addressTr, addressTr.resultReg, -1)
+                val valueTr = exprGen.translateExpression(call.args[1])
+                addToResult(result, valueTr, valueTr.resultReg, -1)
                 result += IRCodeChunk(null, null).also {
-                    it += IRInstruction(Opcode.STOREI, IRDataType.BYTE, reg1 = valueReg, reg2 = addressReg)
+                    it += IRInstruction(Opcode.STOREI, IRDataType.BYTE, reg1 = valueTr.resultReg, reg2 = addressTr.resultReg)
                 }
             }
         }
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, -1, -1)
     }
 
-    private fun funcPeekW(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcPeekW(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        if(call.args[0] is PtNumber) {
+        return if(call.args[0] is PtNumber) {
+            val resultRegister = codeGen.registers.nextFree()
             val address = (call.args[0] as PtNumber).number.toInt()
             result += IRCodeChunk(null, null).also {
-                it += IRInstruction(Opcode.LOADM, IRDataType.WORD, reg1 = resultRegister, value = address)
+                it += IRInstruction(Opcode.LOADM, IRDataType.WORD, reg1 = resultRegister, address = address)
             }
+            ExpressionCodeResult(result, IRDataType.BYTE, resultRegister, -1)
         } else {
-            val addressReg = codeGen.registers.nextFree()
-            result += exprGen.translateExpression(call.args.single(), addressReg, -1)
+            val tr = exprGen.translateExpression(call.args.single())
+            addToResult(result, tr, tr.resultReg, -1)
+            val resultReg = codeGen.registers.nextFree()
             result += IRCodeChunk(null, null).also {
-                it += IRInstruction(Opcode.LOADI, IRDataType.WORD, reg1 = resultRegister, reg2 = addressReg)
+                it += IRInstruction(Opcode.LOADI, IRDataType.WORD, reg1 = resultReg, reg2 = tr.resultReg)
             }
+            ExpressionCodeResult(result, IRDataType.WORD, resultReg, -1)
         }
-        return result
     }
 
-    private fun funcPeek(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcPeek(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        if(call.args[0] is PtNumber) {
+        return if(call.args[0] is PtNumber) {
+            val resultRegister = codeGen.registers.nextFree()
             val address = (call.args[0] as PtNumber).number.toInt()
             result += IRCodeChunk(null, null).also {
-                it += IRInstruction(Opcode.LOADM, IRDataType.BYTE, reg1 = resultRegister, value = address)
+                it += IRInstruction(Opcode.LOADM, IRDataType.BYTE, reg1 = resultRegister, address = address)
             }
+            ExpressionCodeResult(result, IRDataType.BYTE, resultRegister, -1)
         } else {
-            val addressReg = codeGen.registers.nextFree()
-            result += exprGen.translateExpression(call.args.single(), addressReg, -1)
+            val tr = exprGen.translateExpression(call.args.single())
+            addToResult(result, tr, tr.resultReg, -1)
+            val resultReg = codeGen.registers.nextFree()
             result += IRCodeChunk(null, null).also {
-                it += IRInstruction(Opcode.LOADI, IRDataType.BYTE, reg1 = resultRegister, reg2 = addressReg)
+                it += IRInstruction(Opcode.LOADI, IRDataType.BYTE, reg1 = resultReg, reg2 = tr.resultReg)
             }
+            ExpressionCodeResult(result, IRDataType.BYTE, resultReg, -1)
         }
-        return result
     }
 
-    private fun funcMemory(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcMemory(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val name = (call.args[0] as PtString).value
         val code = IRCodeChunk(null, null)
-        code += IRInstruction(Opcode.LOAD, IRDataType.WORD, reg1=resultRegister, labelSymbol = "prog8_slabs.prog8_memoryslab_$name")
-        return listOf(code)
+        val resultReg = codeGen.registers.nextFree()
+        code += IRInstruction(Opcode.LOAD, IRDataType.WORD, reg1=resultReg, labelSymbol = "prog8_slabs.prog8_memoryslab_$name")
+        return ExpressionCodeResult(code, IRDataType.BYTE, resultReg, -1)
     }
 
-    private fun funcLsb(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
-        return exprGen.translateExpression(call.args.single(), resultRegister, -1)
+    private fun funcLsb(call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        return exprGen.translateExpression(call.args.single())
         // note: if a word result is needed, the upper byte is cleared by the typecast that follows. No need to do it here.
+        // TODO to be more strict, maybe we *should* introduce a new result register that is of type .b?
     }
 
-    private fun funcMsb(call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
+    private fun funcMsb(call: PtBuiltinFunctionCall): ExpressionCodeResult {
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args.single(), resultRegister, -1)
+        val tr = exprGen.translateExpression(call.args.single())
+        addToResult(result, tr, tr.resultReg, -1)
+        val resultReg = codeGen.registers.nextFree()
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(Opcode.MSIG, IRDataType.BYTE, reg1 = resultRegister, reg2 = resultRegister)
+            it += IRInstruction(Opcode.MSIG, IRDataType.BYTE, reg1 = resultReg, reg2 = tr.resultReg)
         }
         // note: if a word result is needed, the upper byte is cleared by the typecast that follows. No need to do it here.
-        return result
+        return ExpressionCodeResult(result, IRDataType.BYTE, resultReg, -1)
     }
 
-    private fun funcRolRor(opcode: Opcode, call: PtBuiltinFunctionCall, resultRegister: Int): IRCodeChunks {
-        val vmDt = codeGen.irType(call.args[0].type)
+    private fun funcRolRor(opcode: Opcode, call: PtBuiltinFunctionCall): ExpressionCodeResult {
+        val vmDt = irType(call.args[0].type)
         val result = mutableListOf<IRCodeChunkBase>()
-        result += exprGen.translateExpression(call.args[0], resultRegister, -1)
+        val tr = exprGen.translateExpression(call.args[0])
+        addToResult(result, tr, tr.resultReg, -1)
         result += IRCodeChunk(null, null).also {
-            it += IRInstruction(opcode, vmDt, reg1 = resultRegister)
+            it += IRInstruction(opcode, vmDt, reg1 = tr.resultReg)
         }
-        result += assignRegisterTo(call.args[0], resultRegister)
-        return result
+        result += assignRegisterTo(call.args[0], tr.resultReg)
+        return ExpressionCodeResult(result, vmDt, -1, -1)
     }
 
+    private fun funcSetLsbMsb(call: PtBuiltinFunctionCall, msb: Boolean): ExpressionCodeResult {
+        val result = mutableListOf<IRCodeChunkBase>()
+        val target = call.args[0]
+        val isConstZeroValue = call.args[1].asConstInteger()==0
+        when(target) {
+            is PtIdentifier -> {
+                if(isConstZeroValue) {
+                    result += IRCodeChunk(null, null).also {
+                        val pointerReg = codeGen.registers.nextFree()
+                        it += IRInstruction(Opcode.LOAD, IRDataType.WORD, reg1 = pointerReg, labelSymbol = target.name)
+                        if (msb)
+                            it += IRInstruction(Opcode.INC, IRDataType.WORD, reg1 = pointerReg)
+                        it += IRInstruction(Opcode.STOREZI, IRDataType.BYTE, reg1 = pointerReg)
+                    }
+                } else {
+                    val valueTr = exprGen.translateExpression(call.args[1])
+                    addToResult(result, valueTr, valueTr.resultReg, -1)
+                    result += IRCodeChunk(null, null).also {
+                        val pointerReg = codeGen.registers.nextFree()
+                        it += IRInstruction(Opcode.LOAD, IRDataType.WORD, reg1 = pointerReg, labelSymbol = target.name)
+                        if (msb)
+                            it += IRInstruction(Opcode.INC, IRDataType.WORD, reg1 = pointerReg)
+                        it += IRInstruction(Opcode.STOREI, IRDataType.BYTE, reg1 = valueTr.resultReg, reg2 = pointerReg)
+                    }
+                }
+            }
+            is PtArrayIndexer -> {
+                require(!target.usesPointerVariable)
+                if(target.splitWords) {
+                    // lsb/msb in split arrays, element index 'size' is always 1
+                    val constIndex = target.index.asConstInteger()
+                    val varName = target.variable.name + if(msb) "_msb" else "_lsb"
+                    if(isConstZeroValue) {
+                        if(constIndex!=null) {
+                            val offsetReg = codeGen.registers.nextFree()
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=offsetReg, immediate = constIndex)
+                                it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1=offsetReg, labelSymbol = varName)
+                            }
+                        } else {
+                            val indexTr = exprGen.translateExpression(target.index)
+                            addToResult(result, indexTr, indexTr.resultReg, -1)
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1=indexTr.resultReg, labelSymbol = varName)
+                            }
+                        }
+                    } else {
+                        val valueTr = exprGen.translateExpression(call.args[1])
+                        addToResult(result, valueTr, valueTr.resultReg, -1)
+                        if(constIndex!=null) {
+                            val offsetReg = codeGen.registers.nextFree()
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=offsetReg, immediate = constIndex)
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1=valueTr.resultReg, reg2=offsetReg, labelSymbol = varName)
+                            }
+                        } else {
+                            val indexTr = exprGen.translateExpression(target.index)
+                            addToResult(result, indexTr, indexTr.resultReg, -1)
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1=valueTr.resultReg, reg2=indexTr.resultReg, labelSymbol = varName)
+                            }
+                        }
+                    }
+                }
+                else {
+                    val eltSize = codeGen.program.memsizer.memorySize(target.type)
+                    val constIndex = target.index.asConstInteger()
+                    if(isConstZeroValue) {
+                        if(constIndex!=null) {
+                            val offsetReg = codeGen.registers.nextFree()
+                            val offset = eltSize*constIndex + if(msb) 1 else 0
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=offsetReg, immediate = offset)
+                                it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1=offsetReg, labelSymbol = target.variable.name)
+                            }
+                        } else {
+                            val indexTr = exprGen.translateExpression(target.index)
+                            addToResult(result, indexTr, indexTr.resultReg, -1)
+                            result += IRCodeChunk(null, null).also {
+                                if(eltSize>1)
+                                    it += codeGen.multiplyByConst(IRDataType.BYTE, indexTr.resultReg, eltSize)
+                                if(msb)
+                                    it += IRInstruction(Opcode.INC, IRDataType.BYTE, reg1=indexTr.resultReg)
+                                it += IRInstruction(Opcode.STOREZX, IRDataType.BYTE, reg1=indexTr.resultReg, labelSymbol = target.variable.name)
+                            }
+                        }
+                    } else {
+                        val valueTr = exprGen.translateExpression(call.args[1])
+                        addToResult(result, valueTr, valueTr.resultReg, -1)
+                        if(constIndex!=null) {
+                            val offsetReg = codeGen.registers.nextFree()
+                            val offset = eltSize*constIndex + if(msb) 1 else 0
+                            result += IRCodeChunk(null, null).also {
+                                it += IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=offsetReg, immediate = offset)
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1=valueTr.resultReg, reg2=offsetReg, labelSymbol = target.variable.name)
+                            }
+                        } else {
+                            val indexTr = exprGen.translateExpression(target.index)
+                            addToResult(result, indexTr, indexTr.resultReg, -1)
+                            result += IRCodeChunk(null, null).also {
+                                if(eltSize>1)
+                                    it += codeGen.multiplyByConst(IRDataType.BYTE, indexTr.resultReg, eltSize)
+                                if(msb)
+                                    it += IRInstruction(Opcode.INC, IRDataType.BYTE, reg1=indexTr.resultReg)
+                                it += IRInstruction(Opcode.STOREX, IRDataType.BYTE, reg1=valueTr.resultReg, reg2=indexTr.resultReg, labelSymbol = target.variable.name)
+                            }
+                        }
+                    }
+                }
+            }
+            else -> throw AssemblyError("weird target for setlsb/setmsb: $target")
+        }
+        return ExpressionCodeResult(result, IRDataType.WORD, -1, -1)
+    }
+
+
     private fun assignRegisterTo(target: PtExpression, register: Int): IRCodeChunks {
         val assignment = PtAssignment(target.position)
         val assignTarget = PtAssignTarget(target.position)

codeGenIntermediate/src/prog8/codegen/intermediate/IRPeepholeOptimizer.kt

@@ -1,12 +1,42 @@
 package prog8.codegen.intermediate
 
+import prog8.code.core.IErrorReporter
 import prog8.intermediate.*
 
-internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
-    fun optimize() {
-        irprog.blocks.asSequence().flatMap { it.children.filterIsInstance<IRSubroutine>() }.forEach { sub ->
+class IRPeepholeOptimizer(private val irprog: IRProgram) {
+    fun optimize(optimizationsEnabled: Boolean, errors: IErrorReporter) {
+        if(!optimizationsEnabled)
+            return optimizeOnlyJoinChunks()
+
+        peepholeOptimize()
+        val remover = IRUnusedCodeRemover(irprog, errors)
+        var totalRemovals = 0
+        do {
+            val numRemoved = remover.optimize()
+            totalRemovals += numRemoved
+        } while(numRemoved>0 && errors.noErrors())
+        errors.report()
+
+        if(totalRemovals>0) {
+            irprog.linkChunks()  // re-link again.
+        }
+    }
+
+    private fun optimizeOnlyJoinChunks() {
+        irprog.foreachSub { sub ->
+            joinChunks(sub)
             removeEmptyChunks(sub)
             joinChunks(sub)
+        }
+        irprog.linkChunks() // re-link
+    }
+
+    private fun peepholeOptimize() {
+        irprog.foreachSub { sub ->
+            joinChunks(sub)
+            removeEmptyChunks(sub)
+            joinChunks(sub)
+
             sub.chunks.withIndex().forEach { (index, chunk1) ->
                 // we don't optimize Inline Asm chunks here.
                 val chunk2 = if(index<sub.chunks.size-1) sub.chunks[index+1] else null
@@ -17,18 +47,20 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
                         val changed = removeNops(chunk1, indexedInstructions)
                                 || removeDoubleLoadsAndStores(chunk1, indexedInstructions)       // TODO not yet implemented
                                 || removeUselessArithmetic(chunk1, indexedInstructions)
+                                || removeNeedlessCompares(chunk1, indexedInstructions)
                                 || removeWeirdBranches(chunk1, chunk2, indexedInstructions)
                                 || removeDoubleSecClc(chunk1, indexedInstructions)
                                 || cleanupPushPop(chunk1, indexedInstructions)
-                        // TODO other optimizations:
-                        //  more complex optimizations such as unused registers
+                        // TODO other optimizations
                     } while (changed)
                 }
             }
             removeEmptyChunks(sub)
         }
 
-        irprog.linkChunks() // re-link
+        // TODO also do register optimization step here at the end?
+
+        irprog.linkChunks()  // re-link
     }
 
     private fun removeEmptyChunks(sub: IRSubroutine) {
@@ -38,7 +70,7 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         /*
         Empty Code chunk with label ->
             If next chunk has no label -> move label to next chunk, remove original
-            If next chunk has label -> label name should be the same, remove original. Otherwise FOR NOW leave it in place. (TODO: consolidate labels into 1)
+            If next chunk has label -> label name should be the same, remove original. Otherwise FOR NOW leave it in place. (TODO: merge both labels into 1)
             If is last chunk -> keep chunk in place because of the label.
         Empty Code chunk without label ->
             should not have been generated! ERROR.
@@ -63,7 +95,7 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
                             if (chunk.label == nextchunk.label)
                                 removeChunks += index
                             else {
-                                // TODO: consolidate labels on same chunk
+                                // TODO: merge labels on same chunk
                             }
                         }
                     }
@@ -73,7 +105,10 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
 
         relabelChunks.forEach { (index, label) ->
             val chunk = IRCodeChunk(label, null)
-            chunk.instructions += sub.chunks[index].instructions
+            val subChunk = sub.chunks[index]
+            chunk.instructions += subChunk.instructions
+            if(subChunk is IRCodeChunk)
+                chunk.appendSrcPositions(subChunk.sourceLinesPositions)
             sub.chunks[index] = chunk
         }
         removeChunks.reversed().forEach { index -> sub.chunks.removeAt(index) }
@@ -85,7 +120,7 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         if(sub.chunks.isEmpty())
             return
 
-        fun mayJoin(previous: IRCodeChunkBase, chunk: IRCodeChunkBase): Boolean {
+        fun mayJoinCodeChunks(previous: IRCodeChunkBase, chunk: IRCodeChunkBase): Boolean {
             if(chunk.label!=null)
                 return false
             if(previous is IRCodeChunk && chunk is IRCodeChunk) {
@@ -102,12 +137,41 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         chunks += sub.chunks[0]
         for(ix in 1 until sub.chunks.size) {
             val lastChunk = chunks.last()
-            if(mayJoin(lastChunk, sub.chunks[ix])) {
-                lastChunk.instructions += sub.chunks[ix].instructions
-                lastChunk.next = sub.chunks[ix].next
+            val candidate = sub.chunks[ix]
+            when(candidate) {
+                is IRCodeChunk -> {
+                    if(mayJoinCodeChunks(lastChunk, candidate)) {
+                        lastChunk.instructions += candidate.instructions
+                        lastChunk.next = candidate.next
+                        if(lastChunk is IRCodeChunk)
+                            lastChunk.appendSrcPositions(candidate.sourceLinesPositions)
+                    }
+                    else
+                        chunks += candidate
+                }
+                is IRInlineAsmChunk -> {
+                    if(candidate.label!=null)
+                        chunks += candidate
+                    else if(lastChunk.isEmpty()) {
+                        val label = lastChunk.label
+                        if(label!=null)
+                            chunks += IRInlineAsmChunk(label, candidate.assembly, candidate.isIR, candidate.next)
+                        else
+                            chunks += candidate
+                    }
+                }
+                is IRInlineBinaryChunk -> {
+                    if(candidate.label!=null)
+                        chunks += candidate
+                    else if(lastChunk.isEmpty()) {
+                        val label = lastChunk.label
+                        if(label!=null)
+                            chunks += IRInlineBinaryChunk(label, candidate.data, candidate.next)
+                        else
+                            chunks += candidate
+                    }
+                }
             }
-            else
-                chunks += sub.chunks[ix]
         }
         sub.chunks.clear()
         sub.chunks += chunks
@@ -119,8 +183,8 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         indexedInstructions.reversed().forEach { (idx, ins) ->
             if(ins.opcode== Opcode.PUSH) {
                 if(idx < chunk.instructions.size-1) {
-                    val insAfter = chunk.instructions[idx+1] as? IRInstruction
-                    if(insAfter!=null && insAfter.opcode == Opcode.POP) {
+                    val insAfter = chunk.instructions[idx+1]
+                    if(insAfter.opcode == Opcode.POP) {
                         if(ins.reg1==insAfter.reg1) {
                             chunk.instructions.removeAt(idx)
                             chunk.instructions.removeAt(idx)
@@ -143,16 +207,16 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         indexedInstructions.reversed().forEach { (idx, ins) ->
             if(ins.opcode== Opcode.SEC || ins.opcode== Opcode.CLC) {
                 if(idx < chunk.instructions.size-1) {
-                    val insAfter = chunk.instructions[idx+1] as? IRInstruction
-                    if(insAfter?.opcode == ins.opcode) {
+                    val insAfter = chunk.instructions[idx+1]
+                    if(insAfter.opcode == ins.opcode) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
-                    else if(ins.opcode== Opcode.SEC && insAfter?.opcode== Opcode.CLC) {
+                    else if(ins.opcode== Opcode.SEC && insAfter.opcode== Opcode.CLC) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
-                    else if(ins.opcode== Opcode.CLC && insAfter?.opcode== Opcode.SEC) {
+                    else if(ins.opcode== Opcode.CLC && insAfter.opcode== Opcode.SEC) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
@@ -174,14 +238,58 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
                     changed = true
                 }
             }
+
             // remove useless RETURN
-            if(ins.opcode == Opcode.RETURN && idx>0) {
-                val previous = chunk.instructions[idx-1] as? IRInstruction
-                if(previous?.opcode in OpcodesThatJump) {
+            if(idx>0 && (ins.opcode == Opcode.RETURN || ins.opcode==Opcode.RETURNR)) {
+                val previous = chunk.instructions[idx-1]
+                if(previous.opcode in OpcodesThatJump) {
                     chunk.instructions.removeAt(idx)
                     changed = true
                 }
             }
+
+            // replace subsequent opcodes that jump by just the first
+            if(idx>0 && (ins.opcode in OpcodesThatJump)) {
+                val previous = chunk.instructions[idx-1]
+                if(previous.opcode in OpcodesThatJump) {
+                    chunk.instructions.removeAt(idx)
+                    changed = true
+                }
+            }
+
+            // replace call + return --> jump
+            // This can no longer be done here on the IR level, with the current CALL opcode that encodes the full subroutine call setup.
+            // If machine code is ever generated from this IR, *that* should possibly optimize the JSR + RTS into a JMP.
+//            if(idx>0 && ins.opcode==Opcode.RETURN) {
+//                val previous = chunk.instructions[idx-1]
+//                if(previous.opcode==Opcode.CALL) {
+//                    chunk.instructions[idx-1] = IRInstruction(Opcode.JUMP, address = previous.address, labelSymbol = previous.labelSymbol, branchTarget = previous.branchTarget)
+//                    chunk.instructions.removeAt(idx)
+//                    changed = true
+//                }
+//            }
+        }
+        return changed
+    }
+
+    private fun removeNeedlessCompares(chunk: IRCodeChunk, indexedInstructions: List<IndexedValue<IRInstruction>>): Boolean {
+        // a CMPI with 0, after an instruction like LOAD that already sets the status bits, can be removed.
+        // but only if the instruction after it is not using the Carry bit because that won't be set by a LOAD instruction etc.
+        var changed = false
+        indexedInstructions.reversed().forEach { (idx, ins) ->
+            if(idx>0 && idx<(indexedInstructions.size-1) && ins.opcode==Opcode.CMPI && ins.immediate==0) {
+                val previous = indexedInstructions[idx-1].value
+                if(previous.opcode in OpcodesThatSetStatusbitsIncludingCarry) {
+                    chunk.instructions.removeAt(idx)
+                    changed = true
+                } else if(previous.opcode in OpcodesThatSetStatusbitsButNotCarry) {
+                    val next = indexedInstructions[idx+1].value
+                    if(next.opcode !in arrayOf(Opcode.BSTCC, Opcode.BSTCS, Opcode.BSTPOS, Opcode.BSTNEG)) {
+                        chunk.instructions.removeAt(idx)
+                        changed = true
+                    }
+                }
+            }
         }
         return changed
     }
@@ -192,47 +300,47 @@ internal class IRPeepholeOptimizer(private val irprog: IRProgram) {
         indexedInstructions.reversed().forEach { (idx, ins) ->
             when (ins.opcode) {
                 Opcode.DIV, Opcode.DIVS, Opcode.MUL, Opcode.MOD -> {
-                    if (ins.value == 1) {
+                    if (ins.immediate == 1) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
                 }
                 Opcode.ADD, Opcode.SUB -> {
-                    if (ins.value == 1) {
+                    if (ins.immediate == 1) {
                         chunk.instructions[idx] = IRInstruction(
                             if (ins.opcode == Opcode.ADD) Opcode.INC else Opcode.DEC,
                             ins.type,
                             ins.reg1
                         )
                         changed = true
-                    } else if (ins.value == 0) {
+                    } else if (ins.immediate == 0) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
                 }
                 Opcode.AND -> {
-                    if (ins.value == 0) {
-                        chunk.instructions[idx] = IRInstruction(Opcode.LOAD, ins.type, reg1 = ins.reg1, value = 0)
+                    if (ins.immediate == 0) {
+                        chunk.instructions[idx] = IRInstruction(Opcode.LOAD, ins.type, reg1 = ins.reg1, immediate = 0)
                         changed = true
-                    } else if (ins.value == 255 && ins.type == IRDataType.BYTE) {
+                    } else if (ins.immediate == 255 && ins.type == IRDataType.BYTE) {
                         chunk.instructions.removeAt(idx)
                         changed = true
-                    } else if (ins.value == 65535 && ins.type == IRDataType.WORD) {
+                    } else if (ins.immediate == 65535 && ins.type == IRDataType.WORD) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }
                 }
                 Opcode.OR -> {
-                    if (ins.value == 0) {
+                    if (ins.immediate == 0) {
                         chunk.instructions.removeAt(idx)
                         changed = true
-                    } else if ((ins.value == 255 && ins.type == IRDataType.BYTE) || (ins.value == 65535 && ins.type == IRDataType.WORD)) {
-                        chunk.instructions[idx] = IRInstruction(Opcode.LOAD, ins.type, reg1 = ins.reg1, value = ins.value)
+                    } else if ((ins.immediate == 255 && ins.type == IRDataType.BYTE) || (ins.immediate == 65535 && ins.type == IRDataType.WORD)) {
+                        chunk.instructions[idx] = IRInstruction(Opcode.LOAD, ins.type, reg1 = ins.reg1, immediate = ins.immediate)
                         changed = true
                     }
                 }
                 Opcode.XOR -> {
-                    if (ins.value == 0) {
+                    if (ins.immediate == 0) {
                         chunk.instructions.removeAt(idx)
                         changed = true
                     }

codeGenIntermediate/src/prog8/codegen/intermediate/IRRegisterOptimizer.kt

@@ -0,0 +1,103 @@
+package prog8.codegen.intermediate
+
+import prog8.intermediate.IRProgram
+
+
+class IRRegisterOptimizer(private val irProg: IRProgram) {
+    fun optimize() {
+        // reuseRegisters()
+    }
+
+/*
+    TODO: this register re-use renumbering isn't going to work like this,
+    because subroutines will be clobbering the registers that the subroutine
+    which is calling them might be using...
+
+
+    private fun reuseRegisters() {
+
+        fun addToUsage(usage: MutableMap<Pair<Int, IRDataType>, MutableSet<IRCodeChunkBase>>,
+                       regnum: Int,
+                       dt: IRDataType,
+                       chunk: IRCodeChunkBase) {
+            val key = regnum to dt
+            val chunks = usage[key] ?: mutableSetOf()
+            chunks.add(chunk)
+            usage[key] = chunks
+        }
+
+        val usage: MutableMap<Pair<Int, IRDataType>, MutableSet<IRCodeChunkBase>> = mutableMapOf()
+
+        irProg.foreachCodeChunk { chunk ->
+            chunk.usedRegisters().regsTypes.forEach { (regNum, types) ->
+                types.forEach { dt ->
+                    addToUsage(usage, regNum, dt, chunk)
+                }
+            }
+        }
+
+        val registerReplacements = usage.asSequence()
+            .filter { it.value.size==1 }
+            .map { it.key to it.value.iterator().next() }
+            .groupBy({ it.second }, {it.first})
+            .asSequence()
+            .associate { (chunk, registers) ->
+                chunk to registers.withIndex().associate { (index, reg) -> reg to 50000+index }
+            }
+
+        registerReplacements.forEach { replaceRegisters(it.key, it.value) }
+    }
+
+    private fun replaceRegisters(chunk: IRCodeChunkBase, replacements: Map<Pair<Int, IRDataType>, Int>) {
+        val (rF, rI) = replacements.asSequence().partition { it.key.second==IRDataType.FLOAT }
+        val replacementsInt = rI.associate { it.key.first to it.value }
+        val replacementsFloat = rF.associate { it.key.first to it.value }
+
+        fun replaceRegs(fcallArgs: FunctionCallArgs?): FunctionCallArgs? {
+            if(fcallArgs==null)
+                return null
+            val args = if(fcallArgs.arguments.isEmpty()) fcallArgs.arguments else {
+                fcallArgs.arguments.map {
+                    FunctionCallArgs.ArgumentSpec(
+                        it.name,
+                        it.address,
+                        FunctionCallArgs.RegSpec(
+                            it.reg.dt,
+                            if(it.reg.dt==IRDataType.FLOAT)
+                                replacementsFloat.getOrDefault(it.reg.registerNum, it.reg.registerNum)
+                            else
+                                replacementsInt.getOrDefault(it.reg.registerNum, it.reg.registerNum),
+                            it.reg.cpuRegister
+                        )
+                    )
+                }
+            }
+            val rt = fcallArgs.returns
+            val returns = if(rt==null) null else {
+                FunctionCallArgs.RegSpec(
+                    rt.dt,
+                    if(rt.dt==IRDataType.FLOAT)
+                        replacementsFloat.getOrDefault(rt.registerNum, rt.registerNum)
+                    else
+                        replacementsInt.getOrDefault(rt.registerNum, rt.registerNum),
+                    rt.cpuRegister
+                )
+            }
+            return FunctionCallArgs(args, returns)
+        }
+
+        fun replaceRegs(instruction: IRInstruction): IRInstruction {
+            val reg1 = replacementsInt.getOrDefault(instruction.reg1, instruction.reg1)
+            val reg2 = replacementsInt.getOrDefault(instruction.reg2, instruction.reg2)
+            val fpReg1 = replacementsFloat.getOrDefault(instruction.fpReg1, instruction.fpReg1)
+            val fpReg2 = replacementsFloat.getOrDefault(instruction.fpReg2, instruction.fpReg2)
+            return instruction.copy(reg1 = reg1, reg2 = reg2, fpReg1 = fpReg1, fpReg2 = fpReg2, fcallArgs = replaceRegs(instruction.fcallArgs))
+        }
+        val newInstructions = chunk.instructions.map {
+            replaceRegs(it)
+        }
+        chunk.instructions.clear()
+        chunk.instructions.addAll(newInstructions)
+    }
+*/
+}

codeGenIntermediate/src/prog8/codegen/intermediate/IRUnusedCodeRemover.kt

@@ -1,38 +1,22 @@
 package prog8.codegen.intermediate
 
 import prog8.code.core.IErrorReporter
-import prog8.code.core.SourceCode.Companion.libraryFilePrefix
+import prog8.code.core.SourceCode.Companion.LIBRARYFILEPREFIX
 import prog8.intermediate.*
 
 
-internal class IRUnusedCodeRemover(private val irprog: IRProgram, private val errors: IErrorReporter) {
+class IRUnusedCodeRemover(
+    private val irprog: IRProgram,
+    private val errors: IErrorReporter
+) {
     fun optimize(): Int {
-        val allLabeledChunks = mutableMapOf<String, IRCodeChunkBase>()
-
-        irprog.blocks.asSequence().flatMap { it.children.filterIsInstance<IRSubroutine>() }.forEach { sub ->
-            sub.chunks.forEach { chunk ->
-                chunk.label?.let { allLabeledChunks[it] = chunk }
-            }
-        }
-
-        var numRemoved = removeSimpleUnlinked(allLabeledChunks) + removeUnreachable(allLabeledChunks)
-
-        // remove empty subs
-        irprog.blocks.forEach { block ->
-            block.children.filterIsInstance<IRSubroutine>().reversed().forEach { sub ->
-                if(sub.isEmpty()) {
-                    if(!sub.position.file.startsWith(libraryFilePrefix))
-                        errors.warn("unused subroutine ${sub.label}", sub.position)
-                    block.children.remove(sub)
-                    numRemoved++
-                }
-            }
-        }
+        var numRemoved = removeUnusedSubroutines() + removeUnusedAsmSubroutines()
 
         // remove empty blocks
         irprog.blocks.reversed().forEach { block ->
             if(block.isEmpty()) {
                 irprog.blocks.remove(block)
+                pruneSymboltable(block.label)
                 numRemoved++
             }
         }
@@ -40,10 +24,144 @@ internal class IRUnusedCodeRemover(private val irprog: IRProgram, private val er
         return numRemoved
     }
 
+    private fun pruneSymboltable(blockLabel: String) {
+        // we could clean up the SymbolTable as well, but ONLY if these symbols aren't referenced somewhere still in an instruction or variable initializer value
+        val prefix = "$blockLabel."
+        val blockVars = irprog.st.allVariables().filter { it.name.startsWith(prefix) }
+        blockVars.forEach { stVar ->
+            irprog.allSubs().flatMap { it.chunks }.forEach { chunk ->
+                chunk.instructions.forEach { ins ->
+                    if(ins.labelSymbol == stVar.name) {
+                        return  // symbol occurs in an instruction
+                    }
+                }
+            }
+
+            irprog.st.allVariables().forEach { stVar->
+                val initValue = stVar.onetimeInitializationArrayValue
+                if(initValue!=null && !initValue.isEmpty()) {
+                    if(initValue.any {
+                        it.addressOfSymbol?.startsWith(blockLabel)==true
+                    })
+                        return   // symbol occurs in an initializer value (address-of this symbol)_
+                }
+            }
+        }
+
+        irprog.st.removeTree(blockLabel)
+    }
+
+    private fun removeUnusedSubroutines(): Int {
+        val allLabeledChunks = mutableMapOf<String, IRCodeChunkBase>()
+        irprog.foreachCodeChunk { chunk ->
+            chunk.label?.let { allLabeledChunks[it] = chunk }
+        }
+
+        var numRemoved = removeSimpleUnlinked(allLabeledChunks) + removeUnreachable(allLabeledChunks)
+        irprog.blocks.forEach { block ->
+            block.children.filterIsInstance<IRSubroutine>().reversed().forEach { sub ->
+                if(sub.isEmpty()) {
+                    if(!sub.position.file.startsWith(LIBRARYFILEPREFIX)) {
+                        errors.warn("unused subroutine ${sub.label}", sub.position)
+                    }
+                    block.children.remove(sub)
+                    irprog.st.removeTree(sub.label)
+                    numRemoved++
+                }
+            }
+        }
+
+        return numRemoved
+    }
+
+    private fun removeUnusedAsmSubroutines(): Int {
+        val allLabeledAsmsubs = irprog.blocks.asSequence().flatMap { it.children.filterIsInstance<IRAsmSubroutine>() }
+            .associateBy { it.label }
+
+        var numRemoved = removeSimpleUnlinkedAsmsubs(allLabeledAsmsubs)
+        irprog.blocks.forEach { block ->
+            block.children.filterIsInstance<IRAsmSubroutine>().reversed().forEach { sub ->
+                if(sub.isEmpty()) {
+                    if(!sub.position.file.startsWith(LIBRARYFILEPREFIX)) {
+                        errors.warn("unused asmsubroutine ${sub.label}", sub.position)
+                    }
+                    block.children.remove(sub)
+                    irprog.st.removeTree(sub.label)
+                    numRemoved++
+                }
+            }
+        }
+
+        return numRemoved
+    }
+
+    private fun removeSimpleUnlinkedAsmsubs(allSubs: Map<String, IRAsmSubroutine>): Int {
+        val linkedAsmSubs = mutableSetOf<IRAsmSubroutine>()
+
+        // TODO: asmsubs in library modules are never removed, we can't really tell here if they're actually being called or not...
+
+        // check if asmsub is called from another asmsub
+        irprog.blocks.asSequence().forEach { block ->
+            block.children.filterIsInstance<IRAsmSubroutine>().forEach { sub ->
+                if (block.forceOutput || block.library)
+                    linkedAsmSubs += sub
+                if (sub.asmChunk.isNotEmpty()) {
+                    allSubs.forEach { (label, asmsub) ->
+                        if (sub.asmChunk.assembly.contains(label))
+                            linkedAsmSubs += asmsub
+                    }
+                }
+                val inlineAsm = sub.asmChunk.next as? IRInlineAsmChunk
+                if(inlineAsm!=null) {
+                    allSubs.forEach { (label, asmsub) ->
+                        if (inlineAsm.assembly.contains(label))
+                            linkedAsmSubs += asmsub
+                    }
+                }
+            }
+        }
+
+        // check if asmsub is linked or called from another regular subroutine
+        irprog.foreachCodeChunk { chunk ->
+            chunk.instructions.forEach {
+                it.labelSymbol?.let { label -> allSubs[label]?.let { cc -> linkedAsmSubs += cc } }
+                // note: branchTarget can't yet point to another IRAsmSubroutine, so do nothing when it's set
+            }
+        }
+
+        return removeUnlinkedAsmsubs(linkedAsmSubs)
+    }
+
+    private fun removeUnlinkedAsmsubs(linkedAsmSubs: Set<IRAsmSubroutine>): Int {
+        var numRemoved = 0
+        irprog.blocks.asSequence().forEach { block ->
+            block.children.withIndex().reversed().forEach { (index, child) ->
+                if(child is IRAsmSubroutine && child !in linkedAsmSubs) {
+                    block.children.removeAt(index)
+                    numRemoved++
+                }
+            }
+        }
+        return numRemoved
+    }
+
     private fun removeUnreachable(allLabeledChunks: MutableMap<String, IRCodeChunkBase>): Int {
-        val entrypointSub = irprog.blocks.single { it.name=="main" }.children.single { it is IRSubroutine && it.label=="main.start" }
+        val entrypointSub = irprog.blocks.single { it.label=="main" }
+            .children.single { it is IRSubroutine && it.label=="main.start" }
         val reachable = mutableSetOf((entrypointSub as IRSubroutine).chunks.first())
 
+        // all chunks referenced in array initializer values are also 'reachable':
+        irprog.st.allVariables()
+            .filter { !it.uninitialized }
+            .forEach {
+                it.onetimeInitializationArrayValue?.let { array ->
+                    array.forEach {elt ->
+                        if(elt.addressOfSymbol!=null && irprog.st.lookup(elt.addressOfSymbol!!)==null)
+                            reachable.add(irprog.getChunkWithLabel(elt.addressOfSymbol!!))
+                    }
+                }
+            }
+
         fun grow() {
             val new = mutableSetOf<IRCodeChunkBase>()
             reachable.forEach {
@@ -72,29 +190,37 @@ internal class IRUnusedCodeRemover(private val irprog: IRProgram, private val er
     private fun removeSimpleUnlinked(allLabeledChunks: Map<String, IRCodeChunkBase>): Int {
         val linkedChunks = mutableSetOf<IRCodeChunkBase>()
 
-        irprog.blocks.asSequence().flatMap { it.children.filterIsInstance<IRSubroutine>() }.forEach { sub ->
-            sub.chunks.forEach { chunk ->
-                chunk.next?.let { next -> linkedChunks += next }
-                chunk.instructions.forEach {
-                    if(it.branchTarget==null) {
-                        it.labelSymbol?.let { label -> allLabeledChunks[label]?.let { cc -> linkedChunks += cc } }
-                    } else {
-                        linkedChunks += it.branchTarget!!
+        // all chunks referenced in array initializer values are linked as well!:
+        irprog.st.allVariables()
+            .filter { !it.uninitialized }
+            .forEach {
+                it.onetimeInitializationArrayValue?.let { array ->
+                    array.forEach {elt ->
+                        if(elt.addressOfSymbol!=null && irprog.st.lookup(elt.addressOfSymbol!!)==null)
+                            linkedChunks += irprog.getChunkWithLabel(elt.addressOfSymbol!!)
                     }
                 }
-                if (chunk.label == "main.start")
-                    linkedChunks += chunk
             }
+
+        irprog.foreachCodeChunk { chunk ->
+            chunk.next?.let { next -> linkedChunks += next }
+            chunk.instructions.forEach {
+                if(it.branchTarget==null) {
+                    it.labelSymbol?.let { label -> allLabeledChunks[label]?.let { cc -> linkedChunks += cc } }
+                } else {
+                    linkedChunks += it.branchTarget!!
+                }
+            }
+            if (chunk.label == "main.start")
+                linkedChunks += chunk
         }
 
         return removeUnlinkedChunks(linkedChunks)
     }
 
-    private fun removeUnlinkedChunks(
-        linkedChunks: MutableSet<IRCodeChunkBase>
-    ): Int {
+    private fun removeUnlinkedChunks(linkedChunks: Set<IRCodeChunkBase>): Int {
         var numRemoved = 0
-        irprog.blocks.asSequence().flatMap { it.children.filterIsInstance<IRSubroutine>() }.forEach { sub ->
+        irprog.foreachSub { sub ->
             sub.chunks.withIndex().reversed().forEach { (index, chunk) ->
                 if (chunk !in linkedChunks) {
                     if (chunk === sub.chunks[0]) {

codeGenIntermediate/src/prog8/codegen/intermediate/RegisterPool.kt

@@ -1,10 +1,7 @@
 package prog8.codegen.intermediate
 
-import prog8.code.core.AssemblyError
-import prog8.intermediate.SyscallRegisterBase
-
 internal class RegisterPool {
-    // reserve 0,1,2 for return values of subroutine calls and syscalls
+    // reserve 0,1,2 for return values of subroutine calls and syscalls in IR assembly code
     private var firstFree: Int=3
     private var firstFreeFloat: Int=3
 
@@ -14,16 +11,12 @@ internal class RegisterPool {
     fun nextFree(): Int {
         val result = firstFree
         firstFree++
-        if(firstFree >= SyscallRegisterBase)
-            throw AssemblyError("out of virtual registers (int)")
         return result
     }
 
     fun nextFreeFloat(): Int {
         val result = firstFreeFloat
         firstFreeFloat++
-        if(firstFreeFloat >= SyscallRegisterBase)
-            throw AssemblyError("out of virtual registers (fp)")
         return result
     }
 }

codeGenIntermediate/src/prog8/codegen/vm/VmCodeGen.kt

@@ -16,7 +16,7 @@ class VmCodeGen: ICodeGeneratorBackend {
         symbolTable: SymbolTable,
         options: CompilationOptions,
         errors: IErrorReporter
-    ): IAssemblyProgram? {
+    ): IAssemblyProgram {
         val irCodeGen = IRCodeGen(program, symbolTable, options, errors)
         val irProgram = irCodeGen.generate()
         return VmAssemblyProgram(irProgram.name, irProgram)
@@ -26,7 +26,7 @@ class VmCodeGen: ICodeGeneratorBackend {
 
 internal class VmAssemblyProgram(override val name: String, internal val irProgram: IRProgram): IAssemblyProgram {
 
-    override fun assemble(options: CompilationOptions): Boolean {
+    override fun assemble(options: CompilationOptions, errors: IErrorReporter): Boolean {
         // the VM reads the IR file from disk.
         IRFileWriter(irProgram, null).write()
         return true

codeGenIntermediate/test/Dummies.kt

@@ -32,11 +32,19 @@ internal class ErrorReporterForTests(private val throwExceptionAtReportIfErrors:
     val warnings = mutableListOf<String>()
 
     override fun err(msg: String, position: Position) {
-        errors.add("${position.toClickableStr()} $msg")
+        val text = "${position.toClickableStr()} $msg"
+        if(text !in errors)
+            errors.add(text)
     }
 
     override fun warn(msg: String, position: Position) {
-        warnings.add("${position.toClickableStr()} $msg")
+        val text = "${position.toClickableStr()} $msg"
+        if(text !in warnings)
+            warnings.add(text)
+    }
+
+    override fun undefined(symbol: List<String>, position: Position) {
+        err("undefined symbol: ${symbol.joinToString(".")}", position)
     }
 
     override fun noErrors(): Boolean  = errors.isEmpty()

codeGenIntermediate/test/TestIRPeepholeOpt.kt

@@ -8,7 +8,7 @@ import prog8.intermediate.*
 class TestIRPeepholeOpt: FunSpec({
     fun makeIRProgram(chunks: List<IRCodeChunkBase>): IRProgram {
         require(chunks.first().label=="main.start")
-        val block = IRBlock("main", null, IRBlock.BlockAlignment.NONE, Position.DUMMY)
+        val block = IRBlock("main", null, false, false, IRBlock.BlockAlignment.NONE, Position.DUMMY)
         val sub = IRSubroutine("main.start", emptyList(), null, Position.DUMMY)
         chunks.forEach { sub += it }
         block += sub
@@ -36,17 +36,17 @@ class TestIRPeepholeOpt: FunSpec({
         return makeIRProgram(listOf(chunk))
     }
 
-    fun IRProgram.chunks(): List<IRCodeChunkBase> = this.blocks.flatMap { it.children.filterIsInstance<IRSubroutine>() }.flatMap { it.chunks }
+    fun IRProgram.chunks(): List<IRCodeChunkBase> = this.allSubs().flatMap { it.chunks }.toList()
 
     test("remove nops") {
         val irProg = makeIRProgram(listOf(
-            IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=1, value=42),
+            IRInstruction(Opcode.LOAD, IRDataType.BYTE, reg1=1, immediate=42),
             IRInstruction(Opcode.NOP),
             IRInstruction(Opcode.NOP)
         ))
         irProg.chunks().single().instructions.size shouldBe 3
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         irProg.chunks().single().instructions.size shouldBe 1
     }
 
@@ -65,7 +65,7 @@ class TestIRPeepholeOpt: FunSpec({
         irProg.chunks().size shouldBe 4
         irProg.chunks().flatMap { it.instructions }.size shouldBe 5
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         irProg.chunks().size shouldBe 4
         irProg.chunks()[0].label shouldBe "main.start"
         irProg.chunks()[1].label shouldBe "label"
@@ -92,7 +92,7 @@ class TestIRPeepholeOpt: FunSpec({
         ))
         irProg.chunks().single().instructions.size shouldBe 6
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         val instr = irProg.chunks().single().instructions
         instr.size shouldBe 1
         instr[0].opcode shouldBe Opcode.CLC
@@ -107,7 +107,7 @@ class TestIRPeepholeOpt: FunSpec({
         ))
         irProg.chunks().single().instructions.size shouldBe 4
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         val instr = irProg.chunks().single().instructions
         instr.size shouldBe 1
         instr[0].opcode shouldBe Opcode.LOADR
@@ -117,31 +117,31 @@ class TestIRPeepholeOpt: FunSpec({
 
     test("remove useless div/mul, add/sub") {
         val irProg = makeIRProgram(listOf(
-            IRInstruction(Opcode.DIV, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.DIVS, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.MUL, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.MOD, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.DIV, IRDataType.BYTE, reg1=42, value = 2),
-            IRInstruction(Opcode.DIVS, IRDataType.BYTE, reg1=42, value = 2),
-            IRInstruction(Opcode.MUL, IRDataType.BYTE, reg1=42, value = 2),
-            IRInstruction(Opcode.MOD, IRDataType.BYTE, reg1=42, value = 2),
-            IRInstruction(Opcode.ADD, IRDataType.BYTE, reg1=42, value = 0),
-            IRInstruction(Opcode.SUB, IRDataType.BYTE, reg1=42, value = 0)
+            IRInstruction(Opcode.DIV, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.DIVS, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.MUL, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.MOD, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.DIV, IRDataType.BYTE, reg1=42, immediate = 2),
+            IRInstruction(Opcode.DIVS, IRDataType.BYTE, reg1=42, immediate = 2),
+            IRInstruction(Opcode.MUL, IRDataType.BYTE, reg1=42, immediate = 2),
+            IRInstruction(Opcode.MOD, IRDataType.BYTE, reg1=42, immediate = 2),
+            IRInstruction(Opcode.ADD, IRDataType.BYTE, reg1=42, immediate = 0),
+            IRInstruction(Opcode.SUB, IRDataType.BYTE, reg1=42, immediate = 0)
         ))
         irProg.chunks().single().instructions.size shouldBe 10
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         irProg.chunks().single().instructions.size shouldBe 4
     }
 
     test("replace add/sub 1 by inc/dec") {
         val irProg = makeIRProgram(listOf(
-            IRInstruction(Opcode.ADD, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.SUB, IRDataType.BYTE, reg1=42, value = 1)
+            IRInstruction(Opcode.ADD, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.SUB, IRDataType.BYTE, reg1=42, immediate = 1)
         ))
         irProg.chunks().single().instructions.size shouldBe 2
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         val instr = irProg.chunks().single().instructions
         instr.size shouldBe 2
         instr[0].opcode shouldBe Opcode.INC
@@ -150,40 +150,40 @@ class TestIRPeepholeOpt: FunSpec({
 
     test("remove useless and/or/xor") {
         val irProg = makeIRProgram(listOf(
-            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, value = 255),
-            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, value = 65535),
-            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, value = 0),
-            IRInstruction(Opcode.XOR, IRDataType.BYTE, reg1=42, value = 0),
-            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, value = 200),
-            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, value = 60000),
-            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, value = 1),
-            IRInstruction(Opcode.XOR, IRDataType.BYTE, reg1=42, value = 1)
+            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, immediate = 255),
+            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, immediate = 65535),
+            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, immediate = 0),
+            IRInstruction(Opcode.XOR, IRDataType.BYTE, reg1=42, immediate = 0),
+            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, immediate = 200),
+            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, immediate = 60000),
+            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, immediate = 1),
+            IRInstruction(Opcode.XOR, IRDataType.BYTE, reg1=42, immediate = 1)
         ))
         irProg.chunks().single().instructions.size shouldBe 8
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         irProg.chunks().single().instructions.size shouldBe 4
     }
 
     test("replace and/or/xor by constant number") {
         val irProg = makeIRProgram(listOf(
-            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, value = 0),
-            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, value = 0),
-            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, value = 255),
-            IRInstruction(Opcode.OR, IRDataType.WORD, reg1=42, value = 65535)
+            IRInstruction(Opcode.AND, IRDataType.BYTE, reg1=42, immediate = 0),
+            IRInstruction(Opcode.AND, IRDataType.WORD, reg1=42, immediate = 0),
+            IRInstruction(Opcode.OR, IRDataType.BYTE, reg1=42, immediate = 255),
+            IRInstruction(Opcode.OR, IRDataType.WORD, reg1=42, immediate = 65535)
         ))
         irProg.chunks().single().instructions.size shouldBe 4
         val opt = IRPeepholeOptimizer(irProg)
-        opt.optimize()
+        opt.optimize(true, ErrorReporterForTests())
         val instr = irProg.chunks().single().instructions
         instr.size shouldBe 4
         instr[0].opcode shouldBe Opcode.LOAD
         instr[1].opcode shouldBe Opcode.LOAD
         instr[2].opcode shouldBe Opcode.LOAD
         instr[3].opcode shouldBe Opcode.LOAD
-        instr[0].value shouldBe 0
-        instr[1].value shouldBe 0
-        instr[2].value shouldBe 255
-        instr[3].value shouldBe 65535
+        instr[0].immediate shouldBe 0
+        instr[1].immediate shouldBe 0
+        instr[2].immediate shouldBe 255
+        instr[3].immediate shouldBe 65535
     }
 })

codeGenIntermediate/test/TestVmCodeGen.kt

@@ -1,5 +1,6 @@
 import io.kotest.core.spec.style.FunSpec
 import io.kotest.matchers.ints.shouldBeGreaterThan
+import io.kotest.matchers.shouldBe
 import prog8.code.SymbolTableMaker
 import prog8.code.ast.*
 import prog8.code.core.*
@@ -7,6 +8,7 @@ import prog8.code.target.VMTarget
 import prog8.codegen.vm.VmAssemblyProgram
 import prog8.codegen.vm.VmCodeGen
 import prog8.intermediate.IRSubroutine
+import prog8.intermediate.Opcode
 
 class TestVmCodeGen: FunSpec({
 
@@ -39,7 +41,7 @@ class TestVmCodeGen: FunSpec({
 //}
         val codegen = VmCodeGen()
         val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
-        val block = PtBlock("main", null, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
         val sub = PtSub("start", emptyList(), null, Position.DUMMY)
         sub.add(PtVariable("pi", DataType.UBYTE, ZeropageWish.DONTCARE, PtNumber(DataType.UBYTE, 0.0, Position.DUMMY), null, Position.DUMMY))
         sub.add(PtVariable("particleX", DataType.ARRAY_UB, ZeropageWish.DONTCARE, null, 3u, Position.DUMMY))
@@ -89,10 +91,73 @@ class TestVmCodeGen: FunSpec({
         program.add(block)
 
         // define the "cx16.r0" virtual register
-        val cx16block = PtBlock("cx16", null, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val cx16block = PtBlock("cx16", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
         cx16block.add(PtMemMapped("r0", DataType.UWORD, 100u, null, Position.DUMMY))
         program.add(cx16block)
 
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBe 1
+    }
+
+    test("float comparison expressions against zero") {
+//main {
+//    sub start() {
+//        float @shared f1
+//
+//        if f1==0
+//            nop
+//        if f1!=0
+//            nop
+//        if f1>0
+//            nop
+//        if f1<0
+//            nop
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("f1", DataType.FLOAT, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.UBYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.FLOAT, 0.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression("!=", DataType.UBYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.FLOAT, 0.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if2)
+        val if3 = PtIfElse(Position.DUMMY)
+        val cmp3 = PtBinaryExpression("<", DataType.UBYTE, Position.DUMMY)
+        cmp3.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp3.add(PtNumber(DataType.FLOAT, 0.0, Position.DUMMY))
+        if3.add(cmp3)
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if3)
+        val if4 = PtIfElse(Position.DUMMY)
+        val cmp4 = PtBinaryExpression(">", DataType.UBYTE, Position.DUMMY)
+        cmp4.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp4.add(PtNumber(DataType.FLOAT, 0.0, Position.DUMMY))
+        if4.add(cmp4)
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if4)
+        block.add(sub)
+        program.add(block)
+
         val options = getTestOptions()
         val st = SymbolTableMaker(program, options).make()
         val errors = ErrorReporterForTests()
@@ -100,4 +165,313 @@ class TestVmCodeGen: FunSpec({
         val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
         irChunks.size shouldBeGreaterThan 4
     }
+
+    test("float comparison expressions against nonzero") {
+//main {
+//    sub start() {
+//        float @shared f1
+//
+//        if f1==42
+//            nop
+//        if f1!=42
+//            nop
+//        if f1>42
+//            nop
+//        if f1<42
+//            nop
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("f1", DataType.FLOAT, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.UBYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression("!=", DataType.UBYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if2)
+        val if3 = PtIfElse(Position.DUMMY)
+        val cmp3 = PtBinaryExpression("<", DataType.UBYTE, Position.DUMMY)
+        cmp3.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp3.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if3.add(cmp3)
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if3)
+        val if4 = PtIfElse(Position.DUMMY)
+        val cmp4 = PtBinaryExpression(">", DataType.UBYTE, Position.DUMMY)
+        cmp4.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp4.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if4.add(cmp4)
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if4)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBeGreaterThan 4
+    }
+
+    test("float conditional jump") {
+//main {
+//    sub start() {
+//        float @shared f1
+//
+//        if f1==42
+//            goto $c000
+//        if f1>42
+//            goto $c000
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("f1", DataType.FLOAT, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.UBYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtJump(null, 0xc000u, null, Position.DUMMY)) })
+        if1.add(PtNodeGroup())
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression(">", DataType.UBYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.f1", DataType.FLOAT, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.FLOAT, 42.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtJump(null, 0xc000u, null, Position.DUMMY)) })
+        if2.add(PtNodeGroup())
+        sub.add(if2)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBe 1
+    }
+
+    test("integer comparison expressions against zero") {
+//main {
+//    sub start() {
+//        byte @shared sb1
+//
+//        if sb1==0
+//            nop
+//        if sb1!=0
+//            nop
+//        if sb1>0
+//            nop
+//        if sb1<0
+//            nop
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("sb1", DataType.BYTE, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.BYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.BYTE, 0.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression("!=", DataType.BYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.BYTE, 0.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if2)
+        val if3 = PtIfElse(Position.DUMMY)
+        val cmp3 = PtBinaryExpression("<", DataType.BYTE, Position.DUMMY)
+        cmp3.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp3.add(PtNumber(DataType.BYTE, 0.0, Position.DUMMY))
+        if3.add(cmp3)
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if3)
+        val if4 = PtIfElse(Position.DUMMY)
+        val cmp4 = PtBinaryExpression(">", DataType.BYTE, Position.DUMMY)
+        cmp4.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp4.add(PtNumber(DataType.BYTE, 0.0, Position.DUMMY))
+        if4.add(cmp4)
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if4)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBeGreaterThan 4
+    }
+
+    test("integer comparison expressions against nonzero") {
+//main {
+//    sub start() {
+//        byte @shared sb1
+//
+//        if sb1==42
+//            nop
+//        if sb1!=42
+//            nop
+//        if sb1>42
+//            nop
+//        if sb1<42
+//            nop
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("sb1", DataType.BYTE, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.BYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.BYTE, 42.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if1.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression("!=", DataType.BYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.BYTE, 42.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if2.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if2)
+        val if3 = PtIfElse(Position.DUMMY)
+        val cmp3 = PtBinaryExpression("<", DataType.BYTE, Position.DUMMY)
+        cmp3.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp3.add(PtNumber(DataType.BYTE, 42.0, Position.DUMMY))
+        if3.add(cmp3)
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if3.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if3)
+        val if4 = PtIfElse(Position.DUMMY)
+        val cmp4 = PtBinaryExpression(">", DataType.BYTE, Position.DUMMY)
+        cmp4.add(PtIdentifier("main.start.sb1", DataType.BYTE, Position.DUMMY))
+        cmp4.add(PtNumber(DataType.BYTE, 42.0, Position.DUMMY))
+        if4.add(cmp4)
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        if4.add(PtNodeGroup().also { it.add(PtNop(Position.DUMMY)) })
+        sub.add(if4)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBeGreaterThan 4
+    }
+
+    test("integer conditional jump") {
+//main {
+//    sub start() {
+//        ubyte @shared ub1
+//
+//        if ub1==42
+//            goto $c000
+//        if ub1>42
+//            goto $c000
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        sub.add(PtVariable("ub1", DataType.UBYTE, ZeropageWish.DONTCARE, null, null, Position.DUMMY))
+        val if1 = PtIfElse(Position.DUMMY)
+        val cmp1 = PtBinaryExpression("==", DataType.UBYTE, Position.DUMMY)
+        cmp1.add(PtIdentifier("main.start.ub1", DataType.UBYTE, Position.DUMMY))
+        cmp1.add(PtNumber(DataType.UBYTE, 42.0, Position.DUMMY))
+        if1.add(cmp1)
+        if1.add(PtNodeGroup().also { it.add(PtJump(null, 0xc000u, null, Position.DUMMY)) })
+        if1.add(PtNodeGroup())
+        sub.add(if1)
+        val if2 = PtIfElse(Position.DUMMY)
+        val cmp2 = PtBinaryExpression(">", DataType.UBYTE, Position.DUMMY)
+        cmp2.add(PtIdentifier("main.start.ub1", DataType.UBYTE, Position.DUMMY))
+        cmp2.add(PtNumber(DataType.UBYTE, 42.0, Position.DUMMY))
+        if2.add(cmp2)
+        if2.add(PtNodeGroup().also { it.add(PtJump(null, 0xc000u, null, Position.DUMMY)) })
+        if2.add(PtNodeGroup())
+        sub.add(if2)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBe 1
+    }
+
+    test("romsub allowed in ir-codegen") {
+//main {
+//    romsub $5000 = routine()
+//
+//    sub start()  {
+//        routine()
+//    }
+//}
+        val codegen = VmCodeGen()
+        val program = PtProgram("test", DummyMemsizer, DummyStringEncoder)
+        val block = PtBlock("main", null, false, false, false, false, PtBlock.BlockAlignment.NONE, SourceCode.Generated("test"), Position.DUMMY)
+        val romsub = PtAsmSub("routine", 0x5000u, setOf(CpuRegister.Y), emptyList(), emptyList(), false, Position.DUMMY)
+        block.add(romsub)
+        val sub = PtSub("start", emptyList(), null, Position.DUMMY)
+        val call = PtFunctionCall("main.routine", true, DataType.UNDEFINED, Position.DUMMY)
+        sub.add(call)
+        block.add(sub)
+        program.add(block)
+
+        val options = getTestOptions()
+        val st = SymbolTableMaker(program, options).make()
+        val errors = ErrorReporterForTests()
+        val result = codegen.generate(program, st, options, errors) as VmAssemblyProgram
+        val irChunks = (result.irProgram.blocks.first().children.single() as IRSubroutine).chunks
+        irChunks.size shouldBe 1
+        irChunks[0].instructions.size shouldBe 2
+        val preparecallInstr = irChunks[0].instructions[0]
+        preparecallInstr.opcode shouldBe Opcode.PREPARECALL
+        preparecallInstr.immediate shouldBe 0
+        val callInstr = irChunks[0].instructions[1]
+        callInstr.opcode shouldBe Opcode.CALL
+        callInstr.address shouldBe 0x5000
+    }
 })

codeOptimizers/build.gradle

@@ -33,10 +33,10 @@ dependencies {
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
 }

codeOptimizers/src/prog8/optimizer/BinExprSplitter.kt

@@ -1,71 +0,0 @@
-package prog8.optimizer
-
-import prog8.ast.IStatementContainer
-import prog8.ast.Node
-import prog8.ast.Program
-import prog8.ast.expressions.BinaryExpression
-import prog8.ast.statements.AssignTarget
-import prog8.ast.statements.Assignment
-import prog8.ast.statements.AssignmentOrigin
-import prog8.ast.walk.AstWalker
-import prog8.ast.walk.IAstModification
-import prog8.code.core.AugmentAssignmentOperators
-import prog8.code.core.CompilationOptions
-import prog8.code.core.DataType
-import prog8.code.target.VMTarget
-
-
-class BinExprSplitter(private val program: Program, private val options: CompilationOptions) : AstWalker() {
-
-    override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
-
-        if(options.compTarget.name == VMTarget.NAME)
-            return noModifications  // don't split expressions when targeting the vm codegen, it handles nested expressions well
-
-        if(assignment.value.inferType(program) istype DataType.FLOAT && !options.optimizeFloatExpressions)
-            return noModifications
-
-        val binExpr = assignment.value as? BinaryExpression
-        if (binExpr != null) {
-
-            if(binExpr.operator in AugmentAssignmentOperators && isSimpleTarget(assignment.target)) {
-                if(assignment.target isSameAs binExpr.right)
-                    return noModifications
-                if(assignment.target isSameAs binExpr.left) {
-                    if(binExpr.right.isSimple)
-                        return noModifications
-                    val leftBx = binExpr.left as? BinaryExpression
-                    if(leftBx!=null && (!leftBx.left.isSimple || !leftBx.right.isSimple))
-                        return noModifications
-                    val rightBx = binExpr.right as? BinaryExpression
-                    if(rightBx!=null && (!rightBx.left.isSimple || !rightBx.right.isSimple))
-                        return noModifications
-                }
-
-                if(binExpr.right.isSimple) {
-                    val firstAssign = Assignment(assignment.target.copy(), binExpr.left, AssignmentOrigin.OPTIMIZER, binExpr.left.position)
-                    val targetExpr = assignment.target.toExpression()
-                    val augExpr = BinaryExpression(targetExpr, binExpr.operator, binExpr.right, binExpr.right.position)
-                    return listOf(
-                        IAstModification.ReplaceNode(binExpr, augExpr, assignment),
-                        IAstModification.InsertBefore(assignment, firstAssign, assignment.parent as IStatementContainer)
-                    )
-                }
-            }
-
-            // Further unraveling of binary expressions is really complicated here and
-            // often results in much bigger code, thereby defeating the purpose a bit.
-            // All in all this should probably be fixed in a better code generation backend
-            // that doesn't require this at all.
-        }
-
-        return noModifications
-    }
-
-    private fun isSimpleTarget(target: AssignTarget) =
-            if (target.identifier!=null || target.memoryAddress!=null)
-                !target.isIOAddress(options.compTarget.machine)
-            else
-                false
-
-}

codeOptimizers/src/prog8/optimizer/ConstantFoldingOptimizer.kt

@@ -3,27 +3,29 @@ package prog8.optimizer
 import prog8.ast.Node
 import prog8.ast.Program
 import prog8.ast.base.FatalAstException
-import prog8.ast.base.UndefinedSymbolError
 import prog8.ast.expressions.*
 import prog8.ast.maySwapOperandOrder
 import prog8.ast.statements.ForLoop
+import prog8.ast.statements.RepeatLoop
 import prog8.ast.statements.VarDecl
 import prog8.ast.statements.VarDeclType
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
 import prog8.code.core.AssociativeOperators
 import prog8.code.core.DataType
+import kotlin.math.floor
 
 
 class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
 
     override fun before(memread: DirectMemoryRead, parent: Node): Iterable<IAstModification> {
-        // @( &thing )  -->  thing
+        // @( &thing )  -->  thing  (but only if thing is a byte type!)
         val addrOf = memread.addressExpression as? AddressOf
-        return if(addrOf!=null)
-            listOf(IAstModification.ReplaceNode(memread, addrOf.identifier, parent))
-        else
-            noModifications
+        if(addrOf!=null) {
+            if(addrOf.identifier.inferType(program).isBytes)
+                return listOf(IAstModification.ReplaceNode(memread, addrOf.identifier, parent))
+        }
+        return noModifications
     }
 
     override fun after(containment: ContainmentCheck, parent: Node): Iterable<IAstModification> {
@@ -234,6 +236,47 @@ class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
             }
         }
 
+        if(rightconst!=null && (expr.operator=="<<" || expr.operator==">>")) {
+            val dt = expr.left.inferType(program)
+            if(dt.isBytes && rightconst.number>=8) {
+                if(dt.istype(DataType.UBYTE)) {
+                    val zeroUB = NumericLiteral(DataType.UBYTE, 0.0, expr.position)
+                    modifications.add(IAstModification.ReplaceNode(expr, zeroUB, parent))
+                } else {
+                    if(leftconst!=null) {
+                        val zeroB = NumericLiteral(DataType.BYTE, 0.0, expr.position)
+                        val minusoneB = NumericLiteral(DataType.BYTE, -1.0, expr.position)
+                        if(leftconst.number<0.0) {
+                            if(expr.operator=="<<")
+                                modifications.add(IAstModification.ReplaceNode(expr, zeroB, parent))
+                            else
+                                modifications.add(IAstModification.ReplaceNode(expr, minusoneB, parent))
+                        } else {
+                            modifications.add(IAstModification.ReplaceNode(expr, zeroB, parent))
+                        }
+                    }
+                }
+            }
+            else if(dt.isWords && rightconst.number>=16) {
+                if(dt.istype(DataType.UWORD)) {
+                    val zeroUW = NumericLiteral(DataType.UWORD, 0.0, expr.position)
+                    modifications.add(IAstModification.ReplaceNode(expr, zeroUW, parent))
+                } else {
+                    if(leftconst!=null) {
+                        val zeroW = NumericLiteral(DataType.WORD, 0.0, expr.position)
+                        val minusoneW = NumericLiteral(DataType.WORD, -1.0, expr.position)
+                        if(leftconst.number<0.0) {
+                            if(expr.operator=="<<")
+                                modifications.add(IAstModification.ReplaceNode(expr, zeroW, parent))
+                            else
+                                modifications.add(IAstModification.ReplaceNode(expr, minusoneW, parent))
+                        } else {
+                            modifications.add(IAstModification.ReplaceNode(expr, zeroW, parent))
+                        }
+                    }
+                }
+            }
+        }
 
         return modifications
     }
@@ -307,7 +350,7 @@ class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
         val rangeTo = iterableRange.to as? NumericLiteral
         if(rangeFrom==null || rangeTo==null) return noModifications
 
-        val loopvar = forLoop.loopVar.targetVarDecl(program) ?: throw UndefinedSymbolError(forLoop.loopVar)
+        val loopvar = forLoop.loopVar.targetVarDecl(program) ?: return noModifications
 
         val stepLiteral = iterableRange.step as? NumericLiteral
         when(loopvar.datatype) {
@@ -364,6 +407,16 @@ class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
         return noModifications
     }
 
+    override fun after(repeatLoop: RepeatLoop, parent: Node): Iterable<IAstModification> {
+        val count = (repeatLoop.iterations as? NumericLiteral)?.number
+        if(count!=null && floor(count)!=count) {
+            val integer = NumericLiteral.optimalInteger(count.toInt(), repeatLoop.position)
+            repeatLoop.iterations = integer
+            integer.linkParents(repeatLoop)
+        }
+        return noModifications
+    }
+
     private class ShuffleOperands(val expr: BinaryExpression,
                                   val exprOperator: String?,
                                   val subExpr: BinaryExpression,
@@ -576,5 +629,4 @@ class ConstantFoldingOptimizer(private val program: Program) : AstWalker() {
             return null
         }
     }
-
 }

codeOptimizers/src/prog8/optimizer/ConstantIdentifierReplacer.kt

@@ -53,7 +53,8 @@ class VarConstantValueTypeAdjuster(private val program: Program, private val err
         }
 
         if(to==null) {
-            if(!range.to.inferType(program).isInteger)
+            val toType = range.to.inferType(program)
+            if(toType.isKnown && !range.to.inferType(program).isInteger)
                 errors.err("range expression to value must be integer", range.to.position)
         } else if(to-to.toInt()>0) {
             errors.err("range expression to value must be integer", range.to.position)
@@ -68,6 +69,124 @@ class VarConstantValueTypeAdjuster(private val program: Program, private val err
 
         return noModifications
     }
+
+    override fun after(functionCallExpr: FunctionCallExpression, parent: Node): Iterable<IAstModification> {
+        // choose specific builtin function for the given types
+        val func = functionCallExpr.target.nameInSource
+        if(func==listOf("clamp")) {
+            val t1 = functionCallExpr.args[0].inferType(program)
+            if(t1.isKnown) {
+                val replaceFunc: String
+                if(t1.isBytes) {
+                    replaceFunc = if(t1.istype(DataType.BYTE)) "clamp__byte" else "clamp__ubyte"
+                } else if(t1.isInteger) {
+                    replaceFunc = if(t1.istype(DataType.WORD)) "clamp__word" else "clamp__uword"
+                } else {
+                    errors.err("clamp builtin not supported for floats, use floats.clamp", functionCallExpr.position)
+                    return noModifications
+                }
+                return listOf(IAstModification.SetExpression({functionCallExpr.target = it as IdentifierReference},
+                    IdentifierReference(listOf(replaceFunc), functionCallExpr.target.position),
+                    functionCallExpr))
+            }
+        }
+        else if(func==listOf("min") || func==listOf("max")) {
+            val t1 = functionCallExpr.args[0].inferType(program)
+            val t2 = functionCallExpr.args[1].inferType(program)
+            if(t1.isKnown && t2.isKnown) {
+                val funcName = func[0]
+                val replaceFunc: String
+                if(t1.isBytes && t2.isBytes) {
+                    replaceFunc = if(t1.istype(DataType.BYTE) || t2.istype(DataType.BYTE))
+                        "${funcName}__byte"
+                    else
+                        "${funcName}__ubyte"
+                } else if(t1.isInteger && t2.isInteger) {
+                    replaceFunc = if(t1.istype(DataType.WORD) || t2.istype(DataType.WORD))
+                        "${funcName}__word"
+                    else
+                        "${funcName}__uword"
+                } else if(t1.isNumeric && t2.isNumeric) {
+                    errors.err("min/max not supported for floats", functionCallExpr.position)
+                    return noModifications
+                } else {
+                    errors.err("expected numeric arguments", functionCallExpr.args[0].position)
+                    return noModifications
+                }
+                return listOf(IAstModification.SetExpression({functionCallExpr.target = it as IdentifierReference},
+                    IdentifierReference(listOf(replaceFunc), functionCallExpr.target.position),
+                    functionCallExpr))
+            }
+        }
+        else if(func==listOf("abs")) {
+            val t1 = functionCallExpr.args[0].inferType(program)
+            if(t1.isKnown) {
+                val dt = t1.getOrElse { throw InternalCompilerException("invalid dt") }
+                val replaceFunc = when(dt) {
+                    DataType.BYTE -> "abs__byte"
+                    DataType.WORD -> "abs__word"
+                    DataType.FLOAT -> "abs__float"
+                    DataType.UBYTE, DataType.UWORD -> {
+                        return listOf(IAstModification.ReplaceNode(functionCallExpr, functionCallExpr.args[0], parent))
+                    }
+                    else -> {
+                        errors.err("expected numeric argument", functionCallExpr.args[0].position)
+                        return noModifications
+                    }
+                }
+                return listOf(IAstModification.SetExpression({functionCallExpr.target = it as IdentifierReference},
+                    IdentifierReference(listOf(replaceFunc), functionCallExpr.target.position),
+                    functionCallExpr))
+            }
+        }
+        else if(func==listOf("sqrt")) {
+            val t1 = functionCallExpr.args[0].inferType(program)
+            if(t1.isKnown) {
+                val dt = t1.getOrElse { throw InternalCompilerException("invalid dt") }
+                val replaceFunc = when(dt) {
+                    DataType.UBYTE -> "sqrt__ubyte"
+                    DataType.UWORD -> "sqrt__uword"
+                    DataType.FLOAT -> "sqrt__float"
+                    else -> {
+                        errors.err("expected unsigned or float numeric argument", functionCallExpr.args[0].position)
+                        return noModifications
+                    }
+                }
+                return listOf(IAstModification.SetExpression({functionCallExpr.target = it as IdentifierReference},
+                    IdentifierReference(listOf(replaceFunc), functionCallExpr.target.position),
+                    functionCallExpr))
+            }
+        }
+        return noModifications
+    }
+
+    override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
+        // choose specific builtin function for the given types
+        val func = functionCallStatement.target.nameInSource
+        if(func==listOf("divmod")) {
+            val argTypes = functionCallStatement.args.map {it.inferType(program)}.toSet()
+            if(argTypes.size!=1) {
+                errors.err("expected all ubyte or all uword arguments", functionCallStatement.args[0].position)
+                return noModifications
+            }
+            val t1 = argTypes.single()
+            if(t1.isKnown) {
+                val dt = t1.getOrElse { throw InternalCompilerException("invalid dt") }
+                val replaceFunc = when(dt) {
+                    DataType.UBYTE -> "divmod__ubyte"
+                    DataType.UWORD -> "divmod__uword"
+                    else -> {
+                        errors.err("expected all ubyte or all uword arguments", functionCallStatement.args[0].position)
+                        return noModifications
+                    }
+                }
+                return listOf(IAstModification.SetExpression({functionCallStatement.target = it as IdentifierReference},
+                    IdentifierReference(listOf(replaceFunc), functionCallStatement.target.position),
+                    functionCallStatement))
+            }
+        }
+        return noModifications
+    }
 }
 
 
@@ -87,8 +206,27 @@ internal class ConstantIdentifierReplacer(private val program: Program, private
         if(forloop!=null && identifier===forloop.loopVar)
             return noModifications
 
+        val dt = identifier.inferType(program)
+        if(!dt.isKnown || !dt.isNumeric)
+            return noModifications
+
         try {
             val cval = identifier.constValue(program) ?: return noModifications
+            val arrayIdx = identifier.parent as? ArrayIndexedExpression
+            if(arrayIdx!=null && cval.type in NumericDatatypes) {
+                // special case when the identifier is used as a pointer var
+                // var = constpointer[x] --> var = @(constvalue+x) [directmemoryread]
+                // constpointer[x] = var -> @(constvalue+x) [directmemorywrite] = var
+                val add = BinaryExpression(NumericLiteral(cval.type, cval.number, identifier.position), "+", arrayIdx.indexer.indexExpr, identifier.position)
+                return if(arrayIdx.parent is AssignTarget) {
+                    val memwrite = DirectMemoryWrite(add, identifier.position)
+                    val assignTarget = AssignTarget(null, null, memwrite, identifier.position)
+                    listOf(IAstModification.ReplaceNode(arrayIdx.parent, assignTarget, arrayIdx.parent.parent))
+                } else {
+                    val memread = DirectMemoryRead(add, identifier.position)
+                    listOf(IAstModification.ReplaceNode(arrayIdx, memread, arrayIdx.parent))
+                }
+            }
             return when (cval.type) {
                 in NumericDatatypes -> listOf(
                     IAstModification.ReplaceNode(
@@ -138,90 +276,10 @@ internal class ConstantIdentifierReplacer(private val program: Program, private
                         return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
                     }
                 }
-                DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_W -> {
-                    val rangeExpr = decl.value as? RangeExpression
-                    if(rangeExpr!=null) {
-                        // convert the initializer range expression to an actual array
-                        val declArraySize = decl.arraysize?.constIndex()
-                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
-                            errors.err("range expression size (${rangeExpr.size()}) doesn't match declared array size ($declArraySize)", decl.value?.position!!)
-                        val constRange = rangeExpr.toConstantIntegerRange()
-                        if(constRange!=null) {
-                            val eltType = rangeExpr.inferType(program).getOr(DataType.UBYTE)
-                            val newValue = if(eltType in ByteDatatypes) {
-                                ArrayLiteral(InferredTypes.InferredType.known(decl.datatype),
-                                        constRange.map { NumericLiteral(eltType, it.toDouble(), decl.value!!.position) }.toTypedArray(),
-                                        position = decl.value!!.position)
-                            } else {
-                                ArrayLiteral(InferredTypes.InferredType.known(decl.datatype),
-                                        constRange.map { NumericLiteral(eltType, it.toDouble(), decl.value!!.position) }.toTypedArray(),
-                                        position = decl.value!!.position)
-                            }
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
-                        }
-                    }
-                    val numericLv = decl.value as? NumericLiteral
-                    if(numericLv!=null && numericLv.type== DataType.FLOAT)
-                        errors.err("arraysize requires only integers here", numericLv.position)
-                    val size = decl.arraysize?.constIndex() ?: return noModifications
-                    if (rangeExpr==null && numericLv!=null) {
-                        // arraysize initializer is empty or a single int, and we know the size; create the arraysize.
-                        val fillvalue = numericLv.number.toInt()
-                        when(decl.datatype){
-                            DataType.ARRAY_UB -> {
-                                if(fillvalue !in 0..255)
-                                    errors.err("ubyte value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_B -> {
-                                if(fillvalue !in -128..127)
-                                    errors.err("byte value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_UW -> {
-                                if(fillvalue !in 0..65535)
-                                    errors.err("uword value overflow", numericLv.position)
-                            }
-                            DataType.ARRAY_W -> {
-                                if(fillvalue !in -32768..32767)
-                                    errors.err("word value overflow", numericLv.position)
-                            }
-                            else -> {}
-                        }
-                        // create the array itself, filled with the fillvalue.
-                        val array = Array(size) {fillvalue}.map { NumericLiteral(ArrayToElementTypes.getValue(decl.datatype), it.toDouble(), numericLv.position) }.toTypedArray<Expression>()
-                        val refValue = ArrayLiteral(InferredTypes.InferredType.known(decl.datatype), array, position = numericLv.position)
-                        return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
-                    }
-                }
-                DataType.ARRAY_F -> {
-                    val rangeExpr = decl.value as? RangeExpression
-                    if(rangeExpr!=null) {
-                        // convert the initializer range expression to an actual array of floats
-                        val declArraySize = decl.arraysize?.constIndex()
-                        if(declArraySize!=null && declArraySize!=rangeExpr.size())
-                            errors.err("range expression size (${rangeExpr.size()}) doesn't match declared array size ($declArraySize)", decl.value?.position!!)
-                        val constRange = rangeExpr.toConstantIntegerRange()
-                        if(constRange!=null) {
-                            val newValue = ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_F),
-                                    constRange.map { NumericLiteral(DataType.FLOAT, it.toDouble(), decl.value!!.position) }.toTypedArray(),
-                                    position = decl.value!!.position)
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, newValue, decl))
-                        }
-                    }
-
-                    val numericLv = decl.value as? NumericLiteral
-                    val size = decl.arraysize?.constIndex() ?: return noModifications
-                    if(rangeExpr==null && numericLv!=null) {
-                        // arraysize initializer is a single int, and we know the size.
-                        val fillvalue = numericLv.number
-                        if (fillvalue < compTarget.machine.FLOAT_MAX_NEGATIVE || fillvalue > compTarget.machine.FLOAT_MAX_POSITIVE)
-                            errors.err("float value overflow", numericLv.position)
-                        else {
-                            // create the array itself, filled with the fillvalue.
-                            val array = Array(size) {fillvalue}.map { NumericLiteral(DataType.FLOAT, it, numericLv.position) }.toTypedArray<Expression>()
-                            val refValue = ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_F), array, position = numericLv.position)
-                            return listOf(IAstModification.ReplaceNode(decl.value!!, refValue, decl))
-                        }
-                    }
+                in ArrayDatatypes -> {
+                    val replacedArrayInitializer = createConstArrayInitializerValue(decl)
+                    if(replacedArrayInitializer!=null)
+                        return listOf(IAstModification.ReplaceNode(decl.value!!, replacedArrayInitializer, decl))
                 }
                 else -> {
                     // nothing to do for this type
@@ -231,4 +289,109 @@ internal class ConstantIdentifierReplacer(private val program: Program, private
 
         return noModifications
     }
+
+    private fun createConstArrayInitializerValue(decl: VarDecl): ArrayLiteral? {
+        // convert the initializer range expression from a range or int, to an actual array.
+
+        when(decl.datatype) {
+            DataType.ARRAY_UB, DataType.ARRAY_B, DataType.ARRAY_UW, DataType.ARRAY_W, DataType.ARRAY_W_SPLIT, DataType.ARRAY_UW_SPLIT -> {
+                val rangeExpr = decl.value as? RangeExpression
+                if(rangeExpr!=null) {
+                    val constRange = rangeExpr.toConstantIntegerRange()
+                    if(constRange?.isEmpty()==true) {
+                        if(constRange.first>constRange.last && constRange.step>=0)
+                            errors.err("descending range with positive step", decl.value?.position!!)
+                        else if(constRange.first<constRange.last && constRange.step<=0)
+                            errors.err("ascending range with negative step", decl.value?.position!!)
+                    }
+                    val declArraySize = decl.arraysize?.constIndex()
+                    if(declArraySize!=null && declArraySize!=rangeExpr.size())
+                        errors.err("range expression size (${rangeExpr.size()}) doesn't match declared array size ($declArraySize)", decl.value?.position!!)
+                    if(constRange!=null) {
+                        val eltType = rangeExpr.inferType(program).getOr(DataType.UBYTE)
+                        return if(eltType in ByteDatatypes) {
+                            ArrayLiteral(InferredTypes.InferredType.known(decl.datatype),
+                                constRange.map { NumericLiteral(eltType, it.toDouble(), decl.value!!.position) }.toTypedArray(),
+                                position = decl.value!!.position)
+                        } else {
+                            ArrayLiteral(InferredTypes.InferredType.known(decl.datatype),
+                                constRange.map { NumericLiteral(eltType, it.toDouble(), decl.value!!.position) }.toTypedArray(),
+                                position = decl.value!!.position)
+                        }
+                    }
+                }
+                val numericLv = decl.value as? NumericLiteral
+                if(numericLv!=null && numericLv.type== DataType.FLOAT)
+                    errors.err("arraysize requires only integers here", numericLv.position)
+                val size = decl.arraysize?.constIndex() ?: return null
+                if (rangeExpr==null && numericLv!=null) {
+                    // arraysize initializer is empty or a single int, and we know the size; create the arraysize.
+                    val fillvalue = numericLv.number.toInt()
+                    when(decl.datatype){
+                        DataType.ARRAY_UB -> {
+                            if(fillvalue !in 0..255)
+                                errors.err("ubyte value overflow", numericLv.position)
+                        }
+                        DataType.ARRAY_B -> {
+                            if(fillvalue !in -128..127)
+                                errors.err("byte value overflow", numericLv.position)
+                        }
+                        DataType.ARRAY_UW -> {
+                            if(fillvalue !in 0..65535)
+                                errors.err("uword value overflow", numericLv.position)
+                        }
+                        DataType.ARRAY_W -> {
+                            if(fillvalue !in -32768..32767)
+                                errors.err("word value overflow", numericLv.position)
+                        }
+                        else -> {}
+                    }
+                    // create the array itself, filled with the fillvalue.
+                    val array = Array(size) {fillvalue}.map { NumericLiteral(ArrayToElementTypes.getValue(decl.datatype), it.toDouble(), numericLv.position) }.toTypedArray<Expression>()
+                    return ArrayLiteral(InferredTypes.InferredType.known(decl.datatype), array, position = numericLv.position)
+                }
+            }
+            DataType.ARRAY_F -> {
+                val rangeExpr = decl.value as? RangeExpression
+                if(rangeExpr!=null) {
+                    // convert the initializer range expression to an actual array of floats
+                    val declArraySize = decl.arraysize?.constIndex()
+                    if(declArraySize!=null && declArraySize!=rangeExpr.size())
+                        errors.err("range expression size (${rangeExpr.size()}) doesn't match declared array size ($declArraySize)", decl.value?.position!!)
+                    val constRange = rangeExpr.toConstantIntegerRange()
+                    if(constRange!=null) {
+                        return ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_F),
+                            constRange.map { NumericLiteral(DataType.FLOAT, it.toDouble(), decl.value!!.position) }.toTypedArray(),
+                            position = decl.value!!.position)
+                    }
+                }
+
+                val numericLv = decl.value as? NumericLiteral
+                val size = decl.arraysize?.constIndex() ?: return null
+                if(rangeExpr==null && numericLv!=null) {
+                    // arraysize initializer is a single int, and we know the size.
+                    val fillvalue = numericLv.number
+                    if (fillvalue < compTarget.machine.FLOAT_MAX_NEGATIVE || fillvalue > compTarget.machine.FLOAT_MAX_POSITIVE)
+                        errors.err("float value overflow", numericLv.position)
+                    else {
+                        // create the array itself, filled with the fillvalue.
+                        val array = Array(size) {fillvalue}.map { NumericLiteral(DataType.FLOAT, it, numericLv.position) }.toTypedArray<Expression>()
+                        return ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_F), array, position = numericLv.position)
+                    }
+                }
+            }
+            DataType.ARRAY_BOOL -> {
+                val numericLv = decl.value as? NumericLiteral
+                val size = decl.arraysize?.constIndex() ?: return null
+                if(numericLv!=null) {
+                    // arraysize initializer is a single int, and we know the size.
+                    val fillvalue = if(numericLv.number==0.0) 0.0 else 1.0
+                    val array = Array(size) {fillvalue}.map { NumericLiteral(DataType.UBYTE, fillvalue, numericLv.position) }.toTypedArray<Expression>()
+                    return ArrayLiteral(InferredTypes.InferredType.known(DataType.ARRAY_BOOL), array, position = numericLv.position)
+                }
+            }
+            else -> return null
+        }
+        return null
+    }
 }

codeOptimizers/src/prog8/optimizer/ExpressionSimplifier.kt

@@ -18,7 +18,7 @@ import kotlin.math.abs
 import kotlin.math.log2
 import kotlin.math.pow
 
-// TODO add more peephole expression optimizations? Investigate what optimizations binaryen has
+// TODO add more peephole expression optimizations? Investigate what optimizations binaryen has?
 
 class ExpressionSimplifier(private val program: Program,
                            private val errors: IErrorReporter,

codeOptimizers/src/prog8/optimizer/Extensions.kt

@@ -43,9 +43,9 @@ fun Program.constantFold(errors: IErrorReporter, compTarget: ICompilationTarget)
 
 fun Program.optimizeStatements(errors: IErrorReporter,
                                functions: IBuiltinFunctions,
-                               compTarget: ICompilationTarget
+                               options: CompilationOptions
 ): Int {
-    val optimizer = StatementOptimizer(this, errors, functions, compTarget)
+    val optimizer = StatementOptimizer(this, errors, functions, options)
     optimizer.visit(this)
     val optimizationCount = optimizer.applyModifications()
 
@@ -54,8 +54,8 @@ fun Program.optimizeStatements(errors: IErrorReporter,
     return optimizationCount
 }
 
-fun Program.inlineSubroutines(): Int {
-    val inliner = Inliner(this)
+fun Program.inlineSubroutines(options: CompilationOptions): Int {
+    val inliner = Inliner(this, options)
     inliner.visit(this)
     return inliner.applyModifications()
 }
@@ -65,9 +65,3 @@ fun Program.simplifyExpressions(errors: IErrorReporter, target: ICompilationTarg
     opti.visit(this)
     return opti.applyModifications()
 }
-
-fun Program.splitBinaryExpressions(options: CompilationOptions) : Int {
-    val opti = BinExprSplitter(this, options)
-    opti.visit(this)
-    return opti.applyModifications()
-}

codeOptimizers/src/prog8/optimizer/Inliner.kt

@@ -8,7 +8,9 @@ import prog8.ast.statements.*
 import prog8.ast.walk.AstWalker
 import prog8.ast.walk.IAstModification
 import prog8.ast.walk.IAstVisitor
+import prog8.code.core.CompilationOptions
 import prog8.code.core.InternalCompilerException
+import prog8.code.target.VMTarget
 
 
 private  fun isEmptyReturn(stmt: Statement): Boolean = stmt is Return && stmt.value==null
@@ -16,7 +18,7 @@ private  fun isEmptyReturn(stmt: Statement): Boolean = stmt is Return && stmt.va
 
 // inliner potentially enables *ONE LINED* subroutines, wihtout to be inlined.
 
-class Inliner(val program: Program): AstWalker() {
+class Inliner(private val program: Program, private val options: CompilationOptions): AstWalker() {
 
     class DetermineInlineSubs(val program: Program): IAstVisitor {
         private val modifications = mutableListOf<IAstModification>()
@@ -184,6 +186,7 @@ class Inliner(val program: Program): AstWalker() {
             }
             return if(sub.isAsmSubroutine) {
                 // simply insert the asm for the argument-less routine
+                sub.hasBeenInlined=true
                 listOf(IAstModification.ReplaceNode(origNode, sub.statements.single().copy(), parent))
             } else {
                 // note that we don't have to process any args, because we online inline parameterless subroutines.
@@ -192,12 +195,16 @@ class Inliner(val program: Program): AstWalker() {
                         val fcall = toInline.value as? FunctionCallExpression
                         if(fcall!=null) {
                             // insert the function call expression as a void function call directly
+                            sub.hasBeenInlined=true
                             val call = FunctionCallStatement(fcall.target.copy(), fcall.args.map { it.copy() }.toMutableList(), true, fcall.position)
                             listOf(IAstModification.ReplaceNode(origNode, call, parent))
                         } else
                             noModifications
                     }
-                    else -> listOf(IAstModification.ReplaceNode(origNode, toInline.copy(), parent))
+                    else -> {
+                        sub.hasBeenInlined=true
+                        listOf(IAstModification.ReplaceNode(origNode, toInline.copy(), parent))
+                    }
                 }
             }
         }
@@ -206,7 +213,7 @@ class Inliner(val program: Program): AstWalker() {
 
     override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification>  {
         val sub = functionCallStatement.target.targetStatement(program) as? Subroutine
-        return if(sub==null)
+        return if(sub==null || !canInline(sub, functionCallStatement))
             noModifications
         else
             possibleInlineFcallStmt(sub, functionCallStatement, parent)
@@ -214,7 +221,7 @@ class Inliner(val program: Program): AstWalker() {
 
     override fun before(functionCallExpr: FunctionCallExpression, parent: Node): Iterable<IAstModification> {
         val sub = functionCallExpr.target.targetStatement(program) as? Subroutine
-        if(sub!=null && sub.inline && sub.parameters.isEmpty()) {
+        if(sub!=null && sub.inline && sub.parameters.isEmpty() && canInline(sub, functionCallExpr)) {
             require(sub.statements.size == 1 || (sub.statements.size == 2 && isEmptyReturn(sub.statements[1]))) {
                 "invalid inline sub at ${sub.position}"
             }
@@ -226,8 +233,10 @@ class Inliner(val program: Program): AstWalker() {
                     is Return -> {
                         // is an expression, so we have to have a Return here in the inlined sub
                         // note that we don't have to process any args, because we online inline parameterless subroutines.
-                        if(toInline.value!=null)
+                        if(toInline.value!=null) {
+                            sub.hasBeenInlined=true
                             listOf(IAstModification.ReplaceNode(functionCallExpr, toInline.value!!.copy(), parent))
+                        }
                         else
                             noModifications
                     }
@@ -239,5 +248,17 @@ class Inliner(val program: Program): AstWalker() {
         return noModifications
     }
 
+    private fun canInline(sub: Subroutine, fcall: IFunctionCall): Boolean {
+        if(!sub.inline)
+            return false
+        if(options.compTarget.name!=VMTarget.NAME) {
+            val stmt = sub.statements.single()
+            if (stmt is IFunctionCall) {
+                val existing = (fcall as Node).definingScope.lookup(stmt.target.nameInSource.take(1))
+                return existing !is VarDecl
+            }
+        }
+        return true
+    }
 }
 

codeOptimizers/src/prog8/optimizer/StatementOptimizer.kt

@@ -13,7 +13,7 @@ import kotlin.math.floor
 class StatementOptimizer(private val program: Program,
                          private val errors: IErrorReporter,
                          private val functions: IBuiltinFunctions,
-                         private val compTarget: ICompilationTarget
+                         private val options: CompilationOptions
 ) : AstWalker() {
 
     override fun after(functionCallStatement: FunctionCallStatement, parent: Node): Iterable<IAstModification> {
@@ -30,7 +30,7 @@ class StatementOptimizer(private val program: Program,
         if(functionCallStatement.target.nameInSource==listOf("txt", "print")) {
             val arg = functionCallStatement.args.single()
             val stringVar: IdentifierReference? = if(arg is AddressOf) {
-                arg.identifier
+                if(arg.arrayIndex==null) arg.identifier else null
             } else {
                 arg as? IdentifierReference
             }
@@ -39,7 +39,7 @@ class StatementOptimizer(private val program: Program,
                 if(string!=null) {
                     val pos = functionCallStatement.position
                     if (string.value.length == 1) {
-                        val firstCharEncoded = compTarget.encodeString(string.value, string.encoding)[0]
+                        val firstCharEncoded = options.compTarget.encodeString(string.value, string.encoding)[0]
                         val chrout = FunctionCallStatement(
                             IdentifierReference(listOf("txt", "chrout"), pos),
                             mutableListOf(NumericLiteral(DataType.UBYTE, firstCharEncoded.toDouble(), pos)),
@@ -51,7 +51,7 @@ class StatementOptimizer(private val program: Program,
                             IAstModification.Remove(stringDecl, stringDecl.parent as IStatementContainer)
                         )
                     } else if (string.value.length == 2) {
-                        val firstTwoCharsEncoded = compTarget.encodeString(string.value.take(2), string.encoding)
+                        val firstTwoCharsEncoded = options.compTarget.encodeString(string.value.take(2), string.encoding)
                         val chrout1 = FunctionCallStatement(
                             IdentifierReference(listOf("txt", "chrout"), pos),
                             mutableListOf(NumericLiteral(DataType.UBYTE, firstTwoCharsEncoded[0].toDouble(), pos)),
@@ -108,6 +108,14 @@ class StatementOptimizer(private val program: Program,
             }
         }
 
+        // remove obvious dangling elses (else after a return)
+        if(ifElse.elsepart.isNotEmpty() && ifElse.truepart.statements.singleOrNull() is Return) {
+            val elsePart = AnonymousScope(ifElse.elsepart.statements, ifElse.elsepart.position)
+            return listOf(
+                IAstModification.ReplaceNode(ifElse.elsepart, AnonymousScope(mutableListOf(), ifElse.elsepart.position), ifElse),
+                IAstModification.InsertAfter(ifElse, elsePart, parent as IStatementContainer)
+            )
+        }
         return noModifications
     }
 
@@ -141,7 +149,7 @@ class StatementOptimizer(private val program: Program,
                 val size = sv.value.length
                 if(size==1) {
                     // loop over string of length 1 -> just assign the single character
-                    val character = compTarget.encodeString(sv.value, sv.encoding)[0]
+                    val character = options.compTarget.encodeString(sv.value, sv.encoding)[0]
                     val byte = NumericLiteral(DataType.UBYTE, character.toDouble(), iterable.position)
                     val scope = AnonymousScope(mutableListOf(), forLoop.position)
                     scope.statements.add(Assignment(AssignTarget(forLoop.loopVar, null, null, forLoop.position), byte, AssignmentOrigin.OPTIMIZER, forLoop.position))
@@ -316,13 +324,13 @@ class StatementOptimizer(private val program: Program,
                         if (rightCv == 0.0) {
                             return listOf(IAstModification.Remove(assignment, parent as IStatementContainer))
                         } else if (targetDt in IntegerDatatypes && floor(rightCv) == rightCv) {
-                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..4.0 && compTarget.name!=VMTarget.NAME) {
+                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..3.0 && options.compTarget.name!=VMTarget.NAME) {
                                 // replace by several INCs if it's not a memory address (inc on a memory mapped register doesn't work very well)
                                 val incs = AnonymousScope(mutableListOf(), assignment.position)
                                 repeat(rightCv.toInt()) {
                                     incs.statements.add(PostIncrDecr(assignment.target.copy(), "++", assignment.position))
                                 }
-                                listOf(IAstModification.ReplaceNode(assignment, if(incs.statements.size==1) incs.statements[0] else incs, parent))
+                                return listOf(IAstModification.ReplaceNode(assignment, if(incs.statements.size==1) incs.statements[0] else incs, parent))
                             }
                         }
                     }
@@ -330,7 +338,7 @@ class StatementOptimizer(private val program: Program,
                         if (rightCv == 0.0) {
                             return listOf(IAstModification.Remove(assignment, parent as IStatementContainer))
                         } else if (targetDt in IntegerDatatypes && floor(rightCv) == rightCv) {
-                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..4.0 && compTarget.name!=VMTarget.NAME) {
+                            if (vardeclDt != VarDeclType.MEMORY && rightCv in 1.0..3.0 && options.compTarget.name!=VMTarget.NAME) {
                                 // replace by several DECs if it's not a memory address (dec on a memory mapped register doesn't work very well)
                                 val decs = AnonymousScope(mutableListOf(), assignment.position)
                                 repeat(rightCv.toInt()) {
@@ -374,4 +382,55 @@ class StatementOptimizer(private val program: Program,
 
         return noModifications
     }
+
+    override fun before(unrollLoop: UnrollLoop, parent: Node): Iterable<IAstModification> {
+        val iterations = unrollLoop.iterations.constValue(program)?.number?.toInt()
+        return if(iterations!=null && iterations<1)
+            listOf(IAstModification.Remove(unrollLoop, parent as IStatementContainer))
+        else
+            noModifications
+    }
+
+    override fun after(whenStmt: When, parent: Node): Iterable<IAstModification> {
+
+        fun replaceWithIf(condition: Expression, trueBlock: AnonymousScope, elseBlock: AnonymousScope?): List<IAstModification> {
+            val ifStmt = IfElse(condition, trueBlock, elseBlock ?: AnonymousScope(mutableListOf(), whenStmt.position), whenStmt.position)
+            errors.warn("for boolean condition a normal if statement is preferred", whenStmt.position)
+            return listOf(IAstModification.ReplaceNode(whenStmt, ifStmt, parent))
+        }
+
+        if(whenStmt.condition.inferType(program).isBool) {
+            if(whenStmt.choices.all { it.values?.size==1 }) {
+                if (whenStmt.choices.all { it.values!!.single().constValue(program)!!.number in arrayOf(0.0, 1.0) }) {
+                    // it's a when statement on booleans that can just be replaced by an if or if..else.
+                    if (whenStmt.choices.size == 1) {
+                        return if(whenStmt.choices[0].values!![0].constValue(program)!!.number==1.0) {
+                            replaceWithIf(whenStmt.condition, whenStmt.choices[0].statements, null)
+                        } else {
+                            val notCondition = BinaryExpression(whenStmt.condition, "==", NumericLiteral(DataType.UBYTE, 0.0, whenStmt.condition.position), whenStmt.condition.position)
+                            replaceWithIf(notCondition, whenStmt.choices[0].statements, null)
+                        }
+                    } else if (whenStmt.choices.size == 2) {
+                        var trueBlock: AnonymousScope? = null
+                        var elseBlock: AnonymousScope? = null
+                        if(whenStmt.choices[0].values!![0].constValue(program)!!.number==1.0) {
+                            trueBlock = whenStmt.choices[0].statements
+                        } else {
+                            elseBlock = whenStmt.choices[0].statements
+                        }
+                        if(whenStmt.choices[1].values!![0].constValue(program)!!.number==1.0) {
+                            trueBlock = whenStmt.choices[1].statements
+                        } else {
+                            elseBlock = whenStmt.choices[1].statements
+                        }
+                        if(trueBlock!=null && elseBlock!=null) {
+                            return replaceWithIf(whenStmt.condition, trueBlock, elseBlock)
+                        }
+                    }
+                }
+            }
+        }
+        return noModifications
+    }
+
 }

codeOptimizers/src/prog8/optimizer/UnusedCodeRemover.kt

@@ -64,14 +64,20 @@ class UnusedCodeRemover(private val program: Program,
     override fun after(block: Block, parent: Node): Iterable<IAstModification> {
         if("force_output" !in block.options()) {
             if (block.containsNoCodeNorVars) {
-                if(block.name != internedStringsModuleName)
-                    errors.warn("removing unused block '${block.name}'", block.position)
+                if(block.name != internedStringsModuleName) {
+                    if(!block.statements.any { it is Subroutine && it.hasBeenInlined })
+                        errors.warn("removing unused block '${block.name}'", block.position)
+                }
                 return listOf(IAstModification.Remove(block, parent as IStatementContainer))
             }
             if(callgraph.unused(block)) {
-                if(block.statements.any{ it !is VarDecl || it.type== VarDeclType.VAR})
-                    errors.warn("removing unused block '${block.name}'", block.position)
-                program.removeInternedStringsFromRemovedBlock(block)
+                if(block.statements.any{ it !is VarDecl || it.type== VarDeclType.VAR}) {
+                    if(!block.statements.any { it is Subroutine && it.hasBeenInlined })
+                        errors.warn("removing unused block '${block.name}'", block.position)
+                }
+                if(!block.statements.any { it is Subroutine && it.hasBeenInlined }) {
+                    program.removeInternedStringsFromRemovedBlock(block)
+                }
                 return listOf(IAstModification.Remove(block, parent as IStatementContainer))
             }
         }
@@ -81,7 +87,7 @@ class UnusedCodeRemover(private val program: Program,
 
     override fun after(subroutine: Subroutine, parent: Node): Iterable<IAstModification> {
         val forceOutput = "force_output" in subroutine.definingBlock.options()
-        if (subroutine !== program.entrypoint && !forceOutput && !subroutine.inline && !subroutine.isAsmSubroutine) {
+        if (subroutine !== program.entrypoint && !forceOutput && !subroutine.isAsmSubroutine) {
             if(callgraph.unused(subroutine)) {
                 if(subroutine.containsNoCodeNorVars) {
                     if(!subroutine.definingModule.isLibrary)
@@ -93,9 +99,12 @@ class UnusedCodeRemover(private val program: Program,
                     }
                     return removals
                 }
-                if(!subroutine.definingModule.isLibrary)
-                    errors.warn("removing unused subroutine '${subroutine.name}'", subroutine.position)
-                program.removeInternedStringsFromRemovedSubroutine(subroutine)
+                if(!subroutine.definingModule.isLibrary && !subroutine.hasBeenInlined) {
+                    errors.warn("unused subroutine '${subroutine.name}'", subroutine.position)
+                }
+                if(!subroutine.inline) {
+                    program.removeInternedStringsFromRemovedSubroutine(subroutine)
+                }
                 return listOf(IAstModification.Remove(subroutine, parent as IStatementContainer))
             }
         }
@@ -110,7 +119,7 @@ class UnusedCodeRemover(private val program: Program,
             if (!forceOutput && decl.origin==VarDeclOrigin.USERCODE && !decl.sharedWithAsm) {
                 val usages = callgraph.usages(decl)
                 if (usages.isEmpty()) {
-                    // if(!decl.definingModule.isLibrary)
+                    if(!decl.definingModule.isLibrary)
                         errors.warn("removing unused variable '${decl.name}'", decl.position)
                     return listOf(IAstModification.Remove(decl, parent as IStatementContainer))
                 }
@@ -118,8 +127,8 @@ class UnusedCodeRemover(private val program: Program,
                     if(usages.size==1) {
                         val singleUse = usages[0].parent
                         if(singleUse is AssignTarget) {
-                            val assignment = singleUse.parent as Assignment
-                            if(assignment.origin==AssignmentOrigin.VARINIT) {
+                            val assignment = singleUse.parent as? Assignment
+                            if(assignment!=null && assignment.origin==AssignmentOrigin.VARINIT) {
                                 if(assignment.value.isSimple) {
                                     // remove the vardecl
                                     if(!decl.definingModule.isLibrary)
@@ -150,6 +159,12 @@ class UnusedCodeRemover(private val program: Program,
         return noModifications
     }
 
+    override fun after(assignment: Assignment, parent: Node): Iterable<IAstModification> {
+        if(assignment.target isSameAs assignment.value)
+            return listOf(IAstModification.Remove(assignment, parent as IStatementContainer))
+        return noModifications
+    }
+
     private fun deduplicateAssignments(statements: List<Statement>, scope: IStatementContainer): List<IAstModification> {
         // removes 'duplicate' assignments that assign the same target directly after another, unless it is a function call
         val linesToRemove = mutableListOf<Assignment>()

compiler/build.gradle

@@ -1,9 +1,10 @@
 plugins {
     id 'java'
     id 'application'
-    id "org.jetbrains.kotlin.jvm"
+    id 'org.jetbrains.kotlin.jvm'
     id 'com.github.johnrengelman.shadow' version '7.1.2'
-    id "io.kotest" version "0.3.9"
+    id 'io.kotest' version '0.3.9'
+    id 'com.peterabeles.gversion' version '1.10.2'
 }
 
 java {
@@ -24,8 +25,6 @@ compileTestKotlin {
     }
 }
 
-def prog8version = rootProject.file('compiler/res/version.txt').text.trim()
-
 dependencies {
     implementation project(':codeCore')
     implementation project(':codeOptimizers')
@@ -34,14 +33,14 @@ dependencies {
     implementation project(':codeGenIntermediate')
     implementation project(':codeGenExperimental')
     implementation project(':virtualmachine')
-    implementation 'org.antlr:antlr4-runtime:4.11.1'
+    implementation "org.antlr:antlr4-runtime:4.13.1"
     implementation "org.jetbrains.kotlin:kotlin-stdlib-jdk8"
     // implementation "org.jetbrains.kotlin:kotlin-reflect"
-    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.4'
-    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.16"
+    implementation 'org.jetbrains.kotlinx:kotlinx-cli:0.3.6'
+    implementation "com.michael-bull.kotlin-result:kotlin-result-jvm:1.1.18"
 
     testImplementation project(':intermediate')
-    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.3.2'
+    testImplementation 'io.kotest:kotest-runner-junit5-jvm:5.6.2'
 }
 
 configurations.all {
@@ -60,10 +59,10 @@ configurations {
 sourceSets {
     main {
         java {
-            srcDirs = ["${project.projectDir}/src"]
+            srcDir "${project.projectDir}/src"
         }
         resources {
-            srcDirs = ["${project.projectDir}/res"]
+            srcDir "${project.projectDir}/res"
         }
     }
     test {
@@ -82,7 +81,7 @@ application {
 
 shadowJar {
     archiveBaseName = 'prog8compiler'
-    archiveVersion = prog8version
+    archiveVersion = version
     // minimize()
 }
 
@@ -100,4 +99,16 @@ test {
     }
 }
 
+gversion {
+    srcDir       = "src/"           // path is relative to the sub-project by default
+    classPackage = "prog8.buildversion"
+    className    = "BuildVersion"
+    language     = "kotlin"
+}
+
+
 build.finalizedBy installDist, installShadowDist
+
+compileKotlin.dependsOn createVersionFile    // , failDirtyNotSnapshot
+compileJava.dependsOn createVersionFile
+

compiler/res/prog8lib/atari/syslib.p8

@@ -2,44 +2,41 @@
 ; Including memory registers, I/O registers, Basic and Kernal subroutines.
 
 atari {
+    %option no_symbol_prefixing
 
         &uword  NMI_VEC         = $FFFA     ; 6502 nmi vector, determined by the kernal if banked in
         &uword  RESET_VEC       = $FFFC     ; 6502 reset vector, determined by the kernal if banked in
         &uword  IRQ_VEC         = $FFFE     ; 6502 interrupt vector, determined by the kernal if banked in
 
-; ---- kernal routines ----
-; TODO
-
-
-asmsub  init_system()  {
-    ; Initializes the machine to a sane starting state.
-    ; Called automatically by the loader program logic.
-    ; TODO
-    %asm {{
-        sei
-        cld
-        clc
-        ; TODO reset screen mode etc etc
-        clv
-        cli
-        rts
-    }}
 }
 
-asmsub  init_system_phase2()  {
-    %asm {{
-        rts     ; no phase 2 steps on the Atari
-    }}
-}
-
-}
-
-
 sys {
     ; ------- lowlevel system routines --------
 
+    %option no_symbol_prefixing
+
     const ubyte target = 8         ;  compilation target specifier.  64 = C64, 128 = C128,  16 = CommanderX16, 8 = atari800XL
 
+    asmsub  init_system()  {
+        ; Initializes the machine to a sane starting state.
+        ; Called automatically by the loader program logic.
+        ; TODO
+        %asm {{
+            sei
+            cld
+            clc
+            ; TODO reset screen mode etc etc
+            clv
+            cli
+            rts
+        }}
+    }
+
+    asmsub  init_system_phase2()  {
+        %asm {{
+            rts     ; no phase 2 steps on the Atari
+        }}
+    }
 
     asmsub  reset_system()  {
         ; Soft-reset the system back to initial power-on Basic prompt.
@@ -188,6 +185,19 @@ _longcopy
         }}
     }
 
+    inline asmsub irqsafe_set_irqd() {
+        %asm {{
+        php
+        sei
+        }}
+    }
+
+    inline asmsub irqsafe_clear_irqd() {
+        %asm {{
+        plp
+        }}
+    }
+
     inline asmsub exit(ubyte returnvalue @A) {
         ; -- immediately exit the program with a return code in the A register
         ;    TODO
@@ -208,6 +218,7 @@ _longcopy
 }
 
 cx16 {
+    %option no_symbol_prefixing
 
     ; the sixteen virtual 16-bit registers that the CX16 has defined in the zeropage
     ; they are simulated on the Atari as well but their location in memory is different
@@ -313,4 +324,31 @@ cx16 {
     &byte r13sH = $1b1b
     &byte r14sH = $1b1d
     &byte r15sH = $1b1f
+
+    asmsub save_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  cx16.r0,y
+            sta  _cx16_vreg_storage,y
+            dey
+            bpl  -
+            rts
+
+    _cx16_vreg_storage
+            .word 0,0,0,0,0,0,0,0
+            .word 0,0,0,0,0,0,0,0
+        }}
+    }
+
+    asmsub restore_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  save_virtual_registers._cx16_vreg_storage,y
+            sta  cx16.r0,y
+            dey
+            bpl  -
+            rts
+        }}
+    }
+
 }

compiler/res/prog8lib/atari/textio.p8

@@ -6,6 +6,8 @@
 
 txt {
 
+    %option no_symbol_prefixing
+
 const ubyte DEFAULT_WIDTH = 40
 const ubyte DEFAULT_HEIGHT = 24
 
@@ -30,7 +32,7 @@ asmsub column(ubyte col @A) clobbers(A, X, Y) {
     }}
 }
 
-asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
+asmsub  fill_screen (ubyte character @ A, ubyte color @ Y) clobbers(A)  {
     ; ---- fill the character screen with the given fill character and character color.
     ;      (assumes screen and color matrix are at their default addresses)
     ;      TODO
@@ -40,7 +42,7 @@ asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
     }}
 }
 
-asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
+asmsub  clear_screenchars (ubyte character @ A) clobbers(Y)  {
 	; ---- clear the character screen with the given fill character (leaves colors)
 	;      (assumes screen matrix is at the default address)
 	; TODO
@@ -70,7 +72,7 @@ sub uppercase() {
     ; TODO
 }
 
-asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
+asmsub  scroll_left  (bool alsocolors @ Pc) clobbers(A, Y)  {
 	; ---- scroll the whole screen 1 character to the left
 	;      contents of the rightmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
@@ -81,7 +83,7 @@ asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
         }}
 }
 
-asmsub  scroll_right  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_right  (bool alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character to the right
 	;      contents of the leftmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
@@ -91,7 +93,7 @@ asmsub  scroll_right  (ubyte alsocolors @ Pc) clobbers(A)  {
         }}
 }
 
-asmsub  scroll_up  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_up  (bool alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character up
 	;      contents of the bottom row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
@@ -101,7 +103,7 @@ asmsub  scroll_up  (ubyte alsocolors @ Pc) clobbers(A)  {
         }}
 }
 
-asmsub  scroll_down  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_down  (bool alsocolors @ Pc) clobbers(A)  {
 	; ---- scroll the whole screen 1 character down
 	;      contents of the top row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
@@ -112,13 +114,12 @@ asmsub  scroll_down  (ubyte alsocolors @ Pc) clobbers(A)  {
 }
 
 
-romsub $F2B0 = outchar(ubyte char @ A)
+romsub $F2B0 = outchar(ubyte character @ A)
 romsub $F2Fd = waitkey()
 
-asmsub chrout(ubyte char @ A) {
+asmsub chrout(ubyte character @ A) {
 	%asm {{
 		sta  _tmp_outchar+1
-		pha
 		txa
 		pha
 		tya
@@ -130,7 +131,6 @@ _tmp_outchar
 		tay
 		pla
 		tax
-		pla
 		rts
 	}}
 }
@@ -153,10 +153,9 @@ asmsub  print (str text @ AY) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub0  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 		pha
 		tya
@@ -164,16 +163,13 @@ asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
 		pla
 		jsr  chrout
 		txa
-		jsr  chrout
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  chrout
 	}}
 }
 
-asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 _print_byte_digits
 		pha
@@ -189,16 +185,13 @@ _print_byte_digits
         beq  _ones
         jsr  chrout
 _ones   txa
-		jsr  chrout
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  chrout
 	}}
 }
 
-asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
+asmsub  print_b  (byte value @ A) clobbers(A,X,Y)  {
 	; ---- print the byte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		pha
 		cmp  #0
 		bpl  +
@@ -210,10 +203,9 @@ asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubhex  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  +
 		pha
 		lda  #'$'
@@ -222,16 +214,13 @@ asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 +		jsr  conv.ubyte2hex
 		jsr  chrout
 		tya
-		jsr  chrout
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  chrout
 	}}
 }
 
-asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubbin  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'%'
@@ -244,12 +233,11 @@ asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
 +		jsr  chrout
 		dey
 		bne  -
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwbin  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
 		pha
@@ -261,7 +249,7 @@ asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwhex  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in hexadecimal form (4 digits)
 	;      (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
@@ -274,10 +262,9 @@ asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw0  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
@@ -285,17 +272,14 @@ asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
 		jsr  chrout
 		iny
 		bne  -
-+		ldx  P8ZP_SCRATCH_REG
-		rts
++		rts
 	}}
 }
 
-asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, without left padding 0s
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
-		ldx  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
 		beq  _allzero
@@ -316,7 +300,7 @@ _allzero
 	}}
 }
 
-asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
+asmsub  print_w  (word value @ AY) clobbers(A,X,Y)  {
 	; ---- print the (signed) word in A/Y in decimal form, without left padding 0's
 	%asm {{
 		cpy  #0
@@ -380,7 +364,7 @@ asmsub  getclr  (ubyte col @A, ubyte row @Y) clobbers(Y) -> ubyte @ A {
         }}
 }
 
-sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
+sub  setcc  (ubyte col, ubyte row, ubyte char, ubyte charcolor)  {
 	; ---- set char+color at the given position on the screen
 	; TODO
 	%asm {{
@@ -388,7 +372,7 @@ sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
         }}
 }
 
-asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
+asmsub  plot  (ubyte col @ Y, ubyte row @ A) {
 	; ---- set cursor at specific position
 	; TODO
 	%asm  {{

compiler/res/prog8lib/c128/floats.asm

@@ -1,664 +0,0 @@
-; --- low level floating point assembly routines for the C128
-;     these are almost all identical to the C64 except for a few details
-;     so we have to have a separate library file for the C128 unfortunately.
-
-
-FL_ONE_const	.byte  129     			; 1.0
-FL_ZERO_const	.byte  0,0,0,0,0		; 0.0
-FL_LOG2_const	.byte  $80, $31, $72, $17, $f8	; log(2)
-
-
-floats_store_reg	.byte  0		; temp storage
-
-
-ub2float	.proc
-		; -- convert ubyte in SCRATCH_ZPB1 to float at address A/Y
-		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG
-		sta  P8ZP_SCRATCH_W2
-		sty  P8ZP_SCRATCH_W2+1
-		ldy  P8ZP_SCRATCH_B1
-		lda  #0
-		jsr  GIVAYF
-_fac_to_mem	ldx  P8ZP_SCRATCH_W2
-		ldy  P8ZP_SCRATCH_W2+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-b2float		.proc
-		; -- convert byte in SCRATCH_ZPB1 to float at address A/Y
-		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG
-		sta  P8ZP_SCRATCH_W2
-		sty  P8ZP_SCRATCH_W2+1
-		lda  P8ZP_SCRATCH_B1
-		jsr  FREADSA
-		jmp  ub2float._fac_to_mem
-		.pend
-
-uw2float	.proc
-		; -- convert uword in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG
-		sta  P8ZP_SCRATCH_W2
-		sty  P8ZP_SCRATCH_W2+1
-		lda  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  GIVUAYFAY
-		jmp  ub2float._fac_to_mem
-		.pend
-
-w2float		.proc
-		; -- convert word in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG
-		sta  P8ZP_SCRATCH_W2
-		sty  P8ZP_SCRATCH_W2+1
-		ldy  P8ZP_SCRATCH_W1
-		lda  P8ZP_SCRATCH_W1+1
-		jsr  GIVAYF
-		jmp  ub2float._fac_to_mem
-		.pend
-
-
-cast_from_uw	.proc
-		; -- uword in A/Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVUAYFAY
-		jmp  ub2float._fac_to_mem
-		.pend
-
-
-cast_from_w	.proc
-		; -- word in A/Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVAYFAY
-		jmp  ub2float._fac_to_mem
-		.pend
-
-
-cast_from_ub	.proc
-		; -- ubyte in Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
-		jsr  FREADUY
-		jmp  ub2float._fac_to_mem
-		.pend
-
-
-cast_from_b	.proc
-		; -- byte in A into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
-		jsr  FREADSA
-		jmp  ub2float._fac_to_mem
-		.pend
-
-cast_as_uw_into_ya	.proc               ; also used for float 2 ub
-		; -- cast float at A/Y to uword into Y/A
-		jsr  MOVFM
-		jmp  cast_FAC1_as_uw_into_ya
-		.pend
-
-cast_as_w_into_ay	.proc               ; also used for float 2 b
-		; -- cast float at A/Y to word into A/Y
-		jsr  MOVFM
-		jmp  cast_FAC1_as_w_into_ay
-		.pend
-
-cast_FAC1_as_uw_into_ya	.proc               ; also used for float 2 ub
-		; -- cast fac1 to uword into Y/A
-		stx  P8ZP_SCRATCH_REG
-		jsr  GETADR     ; into Y/A
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-cast_FAC1_as_w_into_ay	.proc               ; also used for float 2 b
-		; -- cast fac1 to word into A/Y
-		stx  P8ZP_SCRATCH_REG
-		jsr  AYINT
-		ldy  floats.AYINT_facmo
-		lda  floats.AYINT_facmo+1
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-
-stack_b2float	.proc
-		; -- b2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  FREADSA
-		jmp  push_fac1._internal
-		.pend
-
-stack_w2float	.proc
-		; -- w2float operating on the stack
-		inx
-		ldy  P8ESTACK_LO,x
-		lda  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVAYF
-		jmp  push_fac1._internal
-		.pend
-
-stack_ub2float	.proc
-		; -- ub2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG
-		tay
-		lda  #0
-		jsr  GIVAYF
-		jmp  push_fac1._internal
-		.pend
-
-stack_uw2float	.proc
-		; -- uw2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		ldy  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVUAYFAY
-		jmp  push_fac1._internal
-		.pend
-
-stack_float2w	.proc               ; also used for float2b
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG
-		jsr  AYINT
-		ldx  P8ZP_SCRATCH_REG
-		lda  floats.AYINT_facmo
-		sta  P8ESTACK_HI,x
-		lda  floats.AYINT_facmo+1
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-stack_float2uw	.proc               ; also used for float2ub
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG
-		jsr  GETADR
-		ldx  P8ZP_SCRATCH_REG
-		sta  P8ESTACK_HI,x
-		tya
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-push_float	.proc
-		; ---- push mflpt5 in A/Y onto stack
-		; (taking 3 stack positions = 6 bytes of which 1 is padding)
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		ldy  #0
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_HI,x
-		dex
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_HI,x
-		dex
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-pop_float	.proc
-		; ---- pops mflpt5 from stack to memory A/Y
-		; (frees 3 stack positions = 6 bytes of which 1 is padding)
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		ldy  #4
-		inx
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		inx
-		lda  P8ESTACK_HI,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		inx
-		lda  P8ESTACK_HI,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		rts
-		.pend
-
-pop_float_fac1	.proc
-		; -- pops float from stack into FAC1
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jmp  MOVFM
-		.pend
-
-copy_float	.proc
-		; -- copies the 5 bytes of the mflt value pointed to by SCRATCH_ZPWORD1,
-		;    into the 5 bytes pointed to by A/Y.  Clobbers A,Y.
-		sta  _target+1
-		sty  _target+2
-		ldy  #4
-_loop		lda  (P8ZP_SCRATCH_W1),y
-_target		sta  $ffff,y			; modified
-		dey
-		bpl  _loop
-		rts
-		.pend
-
-inc_var_f	.proc
-		; -- add 1 to float pointed to by A/Y
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG
-		jsr  MOVFM
-		lda  #<FL_ONE_const
-		ldy  #>FL_ONE_const
-		jsr  FADD
-		ldx  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-dec_var_f	.proc
-		; -- subtract 1 from float pointed to by A/Y
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<FL_ONE_const
-		ldy  #>FL_ONE_const
-		jsr  MOVFM
-		lda  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  FSUB
-		ldx  P8ZP_SCRATCH_W1
-		ldy  P8ZP_SCRATCH_W1+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-
-pop_2_floats_f2_in_fac1	.proc
-		; -- pop 2 floats from stack, load the second one in FAC1 as well
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jmp  MOVFM
-		.pend
-
-
-fmath_float1	.byte 0,0,0,0,0	; storage for a mflpt5 value
-fmath_float2	.byte 0,0,0,0,0	; storage for a mflpt5 value
-
-
-push_fac1	.proc
-		; -- push the float in FAC1 onto the stack
-		stx  P8ZP_SCRATCH_REG
-_internal	ldx  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  MOVMF
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		ldx  P8ZP_SCRATCH_REG
-		jmp  push_float
-		.pend
-
-div_f		.proc
-		; -- push f1/f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FDIV
-		jmp  push_fac1._internal
-		.pend
-
-add_f		.proc
-		; -- push f1+f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FADD
-		jmp  push_fac1._internal
-		.pend
-
-sub_f		.proc
-		; -- push f1-f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FSUB
-		jmp  push_fac1._internal
-		.pend
-
-mul_f		.proc
-		; -- push f1*f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FMULT
-		jmp  push_fac1._internal
-		.pend
-
-neg_f		.proc
-		; -- toggle the sign bit on the stack
-		lda  P8ESTACK_HI+3,x
-		eor  #$80
-		sta  P8ESTACK_HI+3,x
-		rts
-		.pend
-
-var_fac1_less_f	.proc
-		; -- is the float in FAC1 < the variable AY?
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #255
-		beq  +
-		lda  #0
-		rts
-+		lda  #1
-		rts
-		.pend
-
-var_fac1_lesseq_f	.proc
-		; -- is the float in FAC1 <= the variable AY?
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #0
-		beq  +
-		cmp  #255
-		beq  +
-		lda  #0
-		rts
-+		lda  #1
-		rts
-		.pend
-
-var_fac1_greater_f	.proc
-		; -- is the float in FAC1 > the variable AY?
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #1
-		beq  +
-		lda  #0
-		rts
-+		lda  #1
-		rts
-		.pend
-
-var_fac1_greatereq_f	.proc
-		; -- is the float in FAC1 >= the variable AY?
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #0
-		beq  +
-		cmp  #1
-		beq  +
-		lda  #0
-		rts
-+		lda  #1
-		rts
-		.pend
-
-var_fac1_notequal_f	.proc
-		; -- are the floats numbers in FAC1 and the variable AY *not* identical?
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		and  #1
-		rts
-		.pend
-
-vars_equal_f	.proc
-		; -- are the mflpt5 numbers in P8ZP_SCRATCH_W1 and AY identical?
-		sta  P8ZP_SCRATCH_W2
-		sty  P8ZP_SCRATCH_W2+1
-		ldy  #0
-		lda  (P8ZP_SCRATCH_W1),y
-		cmp  (P8ZP_SCRATCH_W2),y
-		bne  _false
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		cmp  (P8ZP_SCRATCH_W2),y
-		bne  _false
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		cmp  (P8ZP_SCRATCH_W2),y
-		bne  _false
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		cmp  (P8ZP_SCRATCH_W2),y
-		bne  _false
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		cmp  (P8ZP_SCRATCH_W2),y
-		bne  _false
-		lda  #1
-		rts
-_false		lda  #0
-		rts
-		.pend
-
-equal_f		.proc
-		; -- are the two mflpt5 numbers on the stack identical?
-		inx
-		inx
-		inx
-		inx
-		lda  P8ESTACK_LO-3,x
-		cmp  P8ESTACK_LO,x
-		bne  _equals_false
-		lda  P8ESTACK_LO-2,x
-		cmp  P8ESTACK_LO+1,x
-		bne  _equals_false
-		lda  P8ESTACK_LO-1,x
-		cmp  P8ESTACK_LO+2,x
-		bne  _equals_false
-		lda  P8ESTACK_HI-2,x
-		cmp  P8ESTACK_HI+1,x
-		bne  _equals_false
-		lda  P8ESTACK_HI-1,x
-		cmp  P8ESTACK_HI+2,x
-		bne  _equals_false
-_equals_true	lda  #1
-_equals_store	inx
-		sta  P8ESTACK_LO+1,x
-		rts
-_equals_false	lda  #0
-		beq  _equals_store
-		.pend
-
-notequal_f	.proc
-		; -- are the two mflpt5 numbers on the stack different?
-		jsr  equal_f
-		eor  #1		; invert the result
-		sta  P8ESTACK_LO+1,x
-		rts
-		.pend
-
-vars_less_f	.proc
-		; -- is float in AY < float in P8ZP_SCRATCH_W2 ?
-		jsr  MOVFM
-		lda  P8ZP_SCRATCH_W2
-		ldy  P8ZP_SCRATCH_W2+1
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #255
-		bne  +
-		lda  #1
-		rts
-+		lda  #0
-		rts
-		.pend
-
-vars_lesseq_f	.proc
-		; -- is float in AY <= float in P8ZP_SCRATCH_W2 ?
-		jsr  MOVFM
-		lda  P8ZP_SCRATCH_W2
-		ldy  P8ZP_SCRATCH_W2+1
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
-		cmp  #255
-		bne  +
--		lda  #1
-		rts
-+		cmp  #0
-		beq  -
-		lda  #0
-		rts
-		.pend
-
-less_f		.proc
-		; -- is f1 < f2?
-		jsr  compare_floats
-		cmp  #255
-		beq  compare_floats._return_true
-		bne  compare_floats._return_false
-		.pend
-
-
-lesseq_f	.proc
-		; -- is f1 <= f2?
-		jsr  compare_floats
-		cmp  #255
-		beq  compare_floats._return_true
-		cmp  #0
-		beq  compare_floats._return_true
-		bne  compare_floats._return_false
-		.pend
-
-greater_f	.proc
-		; -- is f1 > f2?
-		jsr  compare_floats
-		cmp  #1
-		beq  compare_floats._return_true
-		bne  compare_floats._return_false
-		.pend
-
-greatereq_f	.proc
-		; -- is f1 >= f2?
-		jsr  compare_floats
-		cmp  #1
-		beq  compare_floats._return_true
-		cmp  #0
-		beq  compare_floats._return_true
-		bne  compare_floats._return_false
-		.pend
-
-compare_floats	.proc
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  MOVFM		; fac1 = flt1
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP		; A = flt1 compared with flt2 (0=equal, 1=flt1>flt2, 255=flt1<flt2)
-		ldx  P8ZP_SCRATCH_REG
-		rts
-_return_false	lda  #0
-_return_result  sta  P8ESTACK_LO,x
-		dex
-		rts
-_return_true	lda  #1
-		bne  _return_result
-		.pend
-
-set_array_float_from_fac1	.proc
-		; -- set the float in FAC1 in the array (index in A, array in P8ZP_SCRATCH_W1)
-		sta  P8ZP_SCRATCH_B1
-		asl  a
-		asl  a
-		clc
-		adc  P8ZP_SCRATCH_B1
-		ldy  P8ZP_SCRATCH_W1+1
-		clc
-		adc  P8ZP_SCRATCH_W1
-		bcc  +
-		iny
-+		stx  floats_store_reg
-		tax
-		jsr  MOVMF
-		ldx  floats_store_reg
-		rts
-		.pend
-
-
-set_0_array_float	.proc
-		; -- set a float in an array to zero (index in A, array in P8ZP_SCRATCH_W1)
-		sta  P8ZP_SCRATCH_B1
-		asl  a
-		asl  a
-		clc
-		adc  P8ZP_SCRATCH_B1
-		tay
-		lda  #0
-		sta  (P8ZP_SCRATCH_W1),y
-		iny
-		sta  (P8ZP_SCRATCH_W1),y
-		iny
-		sta  (P8ZP_SCRATCH_W1),y
-		iny
-		sta  (P8ZP_SCRATCH_W1),y
-		iny
-		sta  (P8ZP_SCRATCH_W1),y
-		rts
-		.pend
-
-
-set_array_float		.proc
-		; -- set a float in an array to a value (index in A, float in P8ZP_SCRATCH_W1, array in P8ZP_SCRATCH_W2)
-		sta  P8ZP_SCRATCH_B1
-		asl  a
-		asl  a
-		clc
-		adc  P8ZP_SCRATCH_B1
-		adc  P8ZP_SCRATCH_W2
-		ldy  P8ZP_SCRATCH_W2+1
-		bcc  +
-		iny
-+		jmp  copy_float
-			; -- copies the 5 bytes of the mflt value pointed to by SCRATCH_ZPWORD1,
-			;    into the 5 bytes pointed to by A/Y.  Clobbers A,Y.
-		.pend
-
-

compiler/res/prog8lib/c128/floats.p8

@@ -1,158 +0,0 @@
-; Prog8 definitions for floating point handling on the Commodore 128
-
-%option enable_floats
-%import floats_functions
-
-floats {
-	; ---- this block contains C-128 compatible floating point related functions ----
-
-        const float  PI     = 3.141592653589793
-        const float  TWOPI  = 6.283185307179586
-
-
-; ---- ROM float functions ----
-
-		; note: the fac1 and fac2 are working registers and take 6 bytes each,
-		; floats in memory  (and rom) are stored in 5-byte MFLPT packed format.
-
-; note: fac1/2 might get clobbered even if not mentioned in the function's name.
-; note: for subtraction and division, the left operand is in fac2, the right operand in fac1.
-
-romsub $af00 = AYINT() clobbers(A,X,Y)          ; fac1-> signed word in 102-103 ($66-$67) MSB FIRST. (might throw ILLEGAL QUANTITY)
-
-; GIVAYF: signed word in Y/A (note different lsb/msb order) -> float in fac1
-; there is also floats.GIVUAYFAY - unsigned word in A/Y (lo/hi) to fac1
-; (tip: use GIVAYFAY to use A/Y input; lo/hi switched to normal order)
-romsub $af03 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y)
-
-romsub $af06 = FOUT() clobbers(X) -> uword @ AY             ; fac1 -> string, address returned in AY
-; romsub $af09 = VAL_1() clobbers(A,X,Y)                      ; convert ASCII string to floating point [not yet implemented!!!]
-
-; fac1 -> unsigned word in Y/A (might throw ILLEGAL QUANTITY) (result also in $14/15)
-; (tip: use GETADRAY to get A/Y output; lo/hi switched to normal little endian order)
-romsub $af0c = GETADR() clobbers(X) -> ubyte @ Y, ubyte @ A
-romsub $af0f = FLOATC() clobbers(A,X,Y)                     ; convert address to floating point
-
-romsub $af12 = FSUB(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt from A/Y - fac1
-romsub $af15 = FSUBT() clobbers(A,X,Y)                      ; fac1 = fac2-fac1   mind the order of the operands         NOTE: use FSUBT2() instead!
-romsub $af18 = FADD(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 += mflpt value from A/Y
-romsub $af1b = FADDT() clobbers(A,X,Y)                      ; fac1 += fac2      NOTE: use FADDT2() instead!
-romsub $af1e = FMULT(uword mflpt @ AY) clobbers(A,X,Y)      ; fac1 *= mflpt value from A/Y
-romsub $af21 = FMULTT() clobbers(A,X,Y)                     ; fac1 *= fac2      NOTE: use FMULTT2() instead!
-romsub $af24 = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1  (remainder in fac2)
-romsub $af27 = FDIVT() clobbers(A,X,Y)                      ; fac1 = fac2/fac1  (remainder in fac2)  mind the order of the operands  NOTE: use FDIVT2() instead!
-romsub $af2a = LOG() clobbers(A,X,Y)                        ; fac1 = LN(fac1)  (natural log)
-romsub $af2d = INT() clobbers(A,X,Y)                        ; INT() truncates, use FADDH first to round instead of trunc
-romsub $af30 = SQR() clobbers(A,X,Y)                        ; fac1 = SQRT(fac1)
-romsub $af33 = NEGOP() clobbers(A)                          ; switch the sign of fac1 (fac1 = -fac1)
-romsub $af36 = FPWR(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = fac2 ** float in A/Y
-romsub $af39 = FPWRT() clobbers(A,X,Y)                      ; fac1 = fac2 ** fac1           NOTE: use FPWRT2() instead!
-romsub $af3c = EXP() clobbers(A,X,Y)                        ; fac1 = EXP(fac1)  (e ** fac1)
-romsub $af3f = COS() clobbers(A,X,Y)                        ; fac1 = COS(fac1)
-romsub $af42 = SIN() clobbers(A,X,Y)                        ; fac1 = SIN(fac1)
-romsub $af45 = TAN() clobbers(A,X,Y)                        ; fac1 = TAN(fac1)
-romsub $af48 = ATN() clobbers(A,X,Y)                        ; fac1 = ATN(fac1)
-romsub $af4b = ROUND() clobbers(A,X,Y)                      ; round fac1
-romsub $af4e = ABS() clobbers(A,X,Y)                        ; fac1 = ABS(fac1)
-romsub $af51 = SIGN() clobbers(X,Y) -> ubyte @ A            ; SIGN(fac1) to A, $ff, $0, $1 for negative, zero, positive
-romsub $af54 = FCOMP(uword mflpt @ AY) clobbers(X,Y) -> ubyte @ A   ; A = compare fac1 to mflpt in A/Y, 0=equal 1=fac1 is greater, 255=fac1 is less than
-romsub $af57 = RND_0() clobbers(A,X,Y)                      ; fac1 = RND(fac1) float random number generator
-romsub $af57 = RND() clobbers(A,X,Y)                        ; alias for RND_0
-romsub $af5a = CONUPK(uword mflpt @ AY) clobbers(A,X,Y)     ; load mflpt value from memory  in A/Y into fac2
-romsub $af5d = ROMUPK(uword mflpt @ AY) clobbers(A,X,Y)     ; load mflpt value from memory in current bank in A/Y into fac2
-romsub $af60 = MOVFRM(uword mflpt @ AY) clobbers(A,X,Y)     ; load mflpt value from memory in A/Y into fac1  (use MOVFM instead)
-romsub $af63 = MOVFM(uword mflpt @ AY) clobbers(A,X,Y)      ; load mflpt value from memory  in A/Y into fac1
-romsub $af66 = MOVMF(uword mflpt @ XY) clobbers(A,X,Y)      ; store fac1 to memory  X/Y as 5-byte mflpt
-romsub $af69 = MOVFA() clobbers(A,X)                        ; copy fac2 to fac1
-romsub $af6c = MOVAF() clobbers(A,X)                        ; copy fac1 to fac2  (rounded)
-
-; X16 additions
-romsub $af6f = FADDH() clobbers(A,X,Y)                      ; fac1 += 0.5, for rounding- call this before INT
-romsub $af72 = FADDT2() clobbers(A,X,Y)                     ; fac1 += fac2
-romsub $af75 = ZEROFC() clobbers(A,X,Y)                     ; fac1 = 0
-romsub $af78 = NORMAL() clobbers(A,X,Y)                     ; normalize fac1 (?)
-romsub $af7b = NEGFAC() clobbers(A)                         ; switch the sign of fac1 (fac1 = -fac1) (juse use NEGOP() instead!)
-romsub $af7e = FMULTT2() clobbers(A,X,Y)                    ; fac1 *= fac2
-romsub $af81 = MUL10() clobbers(A,X,Y)                      ; fac1 *= 10
-romsub $af84 = DIV10() clobbers(A,X,Y)                      ; fac1 /= 10 , CAUTION: result is always positive!
-romsub $af87 = FDIVT2() clobbers(A,X,Y)                     ; fac1 = fac2/fac1  (remainder in fac2)  mind the order of the operands
-romsub $af8a = MOVEF() clobbers(A,X)                        ; copy fac1 to fac2
-romsub $af8d = SGN() clobbers(A,X,Y)                        ; fac1 = SGN(fac1), result of SIGN (-1, 0 or 1)
-romsub $af90 = FLOAT() clobbers(A,X,Y)                      ; FAC = (u8).A
-romsub $af93 = FLOATS() clobbers(A,X,Y)                     ; FAC = (s16)facho+1:facho
-romsub $af9C = QINT() clobbers(A,X,Y)                       ; facho:facho+1:facho+2:facho+3 = u32(FAC)
-romsub $af9f = FINLOG(byte value @A) clobbers (A, X, Y)     ; fac1 += signed byte in A
-romsub $afa5 = FPWRT2() clobbers(A,X,Y)                     ; fac1 = fac2 ** fac1
-
-asmsub  FREADSA  (byte value @A) clobbers(A,X,Y) {
-    ; ---- 8 bit signed A -> float in fac1
-    %asm {{
-        tay
-        bpl  +
-        lda  #$ff
-        jmp  GIVAYF
-+       lda  #0
-        jmp  GIVAYF
-    }}
-}
-
-asmsub  GIVUAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
-	; ---- unsigned 16 bit word in A/Y (lo/hi) to fac1
-	%asm {{
-	phx
-	sty  $64        ; facmo
-	sta  $65        ; facmo+1
-	ldx  #$90
-	sec
-	jsr  FLOATC
-	plx
-	rts
-	}}
-}
-
-asmsub  GIVAYFAY  (uword value @ AY) clobbers(A,X,Y)  {
-	; ---- signed 16 bit word in A/Y (lo/hi) to float in fac1
-	%asm {{
-		sta  P8ZP_SCRATCH_B1
-		tya
-		ldy  P8ZP_SCRATCH_B1
-		jmp  GIVAYF		; this uses the inverse order, Y/A
-	}}
-}
-
-asmsub  GETADRAY  () clobbers(X) -> uword @ AY  {
-	; ---- fac1 to unsigned word in A/Y
-	%asm {{
-		jsr  GETADR		; this uses the inverse order, Y/A
-		sta  P8ZP_SCRATCH_B1
-		tya
-		ldy  P8ZP_SCRATCH_B1
-		rts
-	}}
-}
-
-asmsub  FREADUY (ubyte value @Y) {
-    ; -- 8 bit unsigned Y -> float in fac1
-    %asm {{
-        tya
-        jmp  FLOAT
-    }}
-}
-
-&uword AYINT_facmo = $66      ; $66/$67 contain result of AYINT
-
-sub rndf() -> float {
-    %asm {{
-        stx  P8ZP_SCRATCH_REG
-        lda  #1
-        jsr  RND_0
-        ldx  P8ZP_SCRATCH_REG
-        rts
-    }}
-}
-
-
-%asminclude "library:c128/floats.asm"
-%asminclude "library:c64/floats_funcs.asm"
-
-}

compiler/res/prog8lib/c128/graphics.p8

@@ -1,60 +0,0 @@
-%import syslib
-%import textio
-
-; Bitmap pixel graphics module for the Commodore 128
-
-; TODO c128 actually implement the graphics routines. Ideally a way to 'borrow' the code form the C64 version without just copy-pasting that here?
-
-graphics {
-    const uword WIDTH = 320
-    const ubyte HEIGHT = 200
-
-    sub enable_bitmap_mode() {
-        ; enable bitmap screen, erase it and set colors to black/white.
-        ; TODO
-    }
-
-    sub disable_bitmap_mode() {
-        ; enables text mode, erase the text screen, color white
-        ; TODO
-    }
-
-
-    sub clear_screen(ubyte pixelcolor, ubyte bgcolor) {
-        ; TODO
-    }
-
-    sub line(uword @zp x1, ubyte @zp y1, uword @zp x2, ubyte @zp y2) {
-        ; TODO
-    }
-
-    sub fillrect(uword x, uword y, uword width, uword height) {
-        ; TODO
-    }
-
-    sub rect(uword x, uword y, uword width, uword height) {
-        ; TODO
-    }
-
-    sub horizontal_line(uword x, uword y, uword length) {
-        ; TODO
-    }
-
-    sub vertical_line(uword x, uword y, uword height) {
-        ; TODO
-    }
-
-    sub circle(uword xcenter, ubyte ycenter, ubyte radius) {
-        ; TODO
-    }
-
-    sub disc(uword xcenter, ubyte ycenter, ubyte radius) {
-        ; TODO
-    }
-
-    inline asmsub  plot(uword plotx @R0, uword ploty @R1) clobbers(A, X, Y) {
-        %asm {{
-            nop     ; TODO
-        }}
-    }
-}

compiler/res/prog8lib/c128/syslib.p8

@@ -1,17 +1,18 @@
 ; Prog8 definitions for the Commodore-128
 ; Including memory registers, I/O registers, Basic and Kernal subroutines.
 
-c64 {
+cbm {
+    ; Commodore (CBM) common variables, vectors and kernal routines
+    %option no_symbol_prefixing
 
         &ubyte  TIME_HI         = $a0       ; software jiffy clock, hi byte
         &ubyte  TIME_MID        = $a1       ;  .. mid byte
         &ubyte  TIME_LO         = $a2       ;    .. lo byte. Updated by IRQ every 1/60 sec
         &ubyte  STATUS          = $90       ; kernal status variable for I/O
         &ubyte  STKEY           = $91       ; various keyboard statuses (updated by IRQ)
-        ;;&ubyte  SFDX            = $cb       ; current key pressed (matrix value) (updated by IRQ)     // TODO c128 ??
-
-        &ubyte  COLOR           = $00f1     ; cursor color
-        ;;&ubyte  HIBASE          = $0288     ; screen base address / 256 (hi-byte of screen memory address)        // TODO c128 ??
+        &ubyte  SHFLAG          = $d3       ; various modifier key status (updated by IRQ)
+        &ubyte  SFDX            = $d4       ; current key pressed (matrix value) (updated by IRQ)
+        &ubyte  COLOR           = $f1       ; cursor color
 
         &uword  IERROR          = $0300
         &uword  IMAIN           = $0302
@@ -48,6 +49,92 @@ c64 {
         const  uword  Screen    = $0400     ; to have this as an array[40*25] the compiler would have to support array size > 255
         const  uword  Colors    = $d800     ; to have this as an array[40*25] the compiler would have to support array size > 255
 
+; ---- kernal routines, these are the same as on the Commodore-64 (hence the same block name) ----
+
+; STROUT --> use txt.print
+; CLEARSCR -> use txt.clear_screen
+; HOMECRSR -> use txt.home or txt.plot
+
+romsub $FA65 = IRQDFRT() clobbers(A,X,Y)                        ; default IRQ routine
+romsub $FF33 = IRQDFEND() clobbers(A,X,Y)                       ; default IRQ end/cleanup
+
+; TODO c128 a bunch of kernal routines are missing here that are specific to the c128
+
+romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
+romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
+romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
+romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
+romsub $FF8D = VECTOR(uword userptr @ XY, bool dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
+romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
+romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
+romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
+romsub $FF99 = MEMTOP(uword address @ XY, bool dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
+romsub $FF9C = MEMBOT(uword address @ XY, bool dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
+romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
+romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
+romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
+romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
+romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
+romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
+romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
+romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
+romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
+romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ Y)   ; set logical file parameters
+romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
+romsub $FFC0 = OPEN() clobbers(X,Y) -> bool @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
+romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
+romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> bool @Pc    ; (via 798 ($31E)) define an input channel
+romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
+romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
+romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
+romsub $FFD2 = CHROUT(ubyte character @ A)                           ; (via 806 ($326)) output a character
+romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> bool @Pc, ubyte @ A, uword @ XY     ; (via 816 ($330)) load from device
+romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> bool @ Pc, ubyte @ A          ; (via 818 ($332)) save to a device
+romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
+romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
+romsub $FFE1 = STOP() clobbers(X) -> bool @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
+romsub $FFE4 = GETIN() clobbers(X,Y) -> bool @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
+romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
+romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
+romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
+romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, bool dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
+romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
+
+; ---- end of C64 compatible ROM kernal routines ----
+
+; ---- utilities -----
+
+asmsub STOP2() clobbers(X) -> ubyte @A  {
+    ; -- check if STOP key was pressed, returns true if so.  More convenient to use than STOP() because that only sets the carry status flag.
+    %asm {{
+        jsr  cbm.STOP
+        beq  +
+        lda  #0
+        rts
++       lda  #1
+        rts
+    }}
+}
+
+asmsub RDTIM16() clobbers(X) -> uword @AY {
+    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
+    %asm {{
+        jsr  cbm.RDTIM
+        pha
+        txa
+        tay
+        pla
+        rts
+    }}
+}
+
+
+}
+
+c64 {
+        ; C64 I/O registers (VIC, SID, CIA)
+    %option no_symbol_prefixing
+
         ; the default locations of the 8 sprite pointers (store address of sprite / 64)
         &ubyte  SPRPTR0         = 2040
         &ubyte  SPRPTR1         = 2041
@@ -197,276 +284,11 @@ c64 {
 
     ; ---- end of SID registers ----
 
-; ---- kernal routines, these are the same as on the Commodore-64 (hence the same block name) ----
-
-; STROUT --> use txt.print
-; CLEARSCR -> use txt.clear_screen
-; HOMECRSR -> use txt.home or txt.plot
-
-romsub $FA65 = IRQDFRT() clobbers(A,X,Y)                        ; default IRQ routine
-romsub $FF33 = IRQDFEND() clobbers(A,X,Y)                       ; default IRQ end/cleanup
-
-; TODO c128 a bunch of kernal routines are missing here that are specific to the c128
-
-romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
-romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
-romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
-romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
-romsub $FF8D = VECTOR(uword userptr @ XY, ubyte dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
-romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
-romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
-romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
-romsub $FF99 = MEMTOP(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
-romsub $FF9C = MEMBOT(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
-romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
-romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
-romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
-romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
-romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
-romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
-romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
-romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
-romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
-romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ Y)   ; set logical file parameters
-romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
-romsub $FFC0 = OPEN() clobbers(X,Y) -> ubyte @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
-romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
-romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> ubyte @Pc    ; (via 798 ($31E)) define an input channel
-romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
-romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
-romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
-romsub $FFD2 = CHROUT(ubyte char @ A)                           ; (via 806 ($326)) output a character
-romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, uword @ XY     ; (via 816 ($330)) load from device
-romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> ubyte @ Pc, ubyte @ A          ; (via 818 ($332)) save to a device
-romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
-romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
-romsub $FFE1 = STOP() clobbers(X) -> ubyte @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
-romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
-romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
-romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
-romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
-romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
-romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
-
-; ---- end of C64 compatible ROM kernal routines ----
-
-; ---- utilities -----
-
-asmsub STOP2() -> ubyte @A  {
-    ; -- check if STOP key was pressed, returns true if so.  More convenient to use than STOP() because that only sets the carry status flag.
-    %asm {{
-        txa
-        pha
-        jsr  c64.STOP
-        beq  +
-        pla
-        tax
-        lda  #0
-        rts
-+       pla
-        tax
-        lda  #1
-        rts
-    }}
-}
-
-asmsub RDTIM16() -> uword @AY {
-    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
-    %asm {{
-        stx  P8ZP_SCRATCH_REG
-        jsr  c64.RDTIM
-        pha
-        txa
-        tay
-        pla
-        ldx  P8ZP_SCRATCH_REG
-        rts
-    }}
-}
-
-
-; ---- system utility routines that are essentially the same as on the C64: -----
-asmsub  disable_runstop_and_charsetswitch() clobbers(A) {
-    %asm {{
-        lda  #$80
-        sta  247    ; disable charset switching
-        lda  #112
-        sta  808    ; disable run/stop key
-        rts
-    }}
-}
-
-asmsub  enable_runstop_and_charsetswitch() clobbers(A) {
-    %asm {{
-        lda  #0
-        sta  247    ; enable charset switching
-        lda  #110
-        sta  808    ; enable run/stop key
-        rts
-    }}
-}
-
-asmsub  set_irq(uword handler @AY, ubyte useKernal @Pc) clobbers(A)  {
-	%asm {{
-	        sta  _modified+1
-	        sty  _modified+2
-	        lda  #0
-	        adc  #0
-	        sta  _use_kernal
-		sei
-		lda  #<_irq_handler
-		sta  c64.CINV
-		lda  #>_irq_handler
-		sta  c64.CINV+1
-		cli
-		rts
-_irq_handler    jsr  _irq_handler_init
-_modified	jsr  $ffff                      ; modified
-		jsr  _irq_handler_end
-		lda  _use_kernal
-		bne  +
-		lda  #$ff
-		sta  c64.VICIRQ			; acknowledge raster irq
-		lda  c64.CIA1ICR		; acknowledge CIA1 interrupt
-		; end irq processing - don't use kernal's irq handling
-		pla
-		tay
-		pla
-		tax
-		pla
-		rti
-+		jmp  c64.IRQDFRT		; continue with normal kernal irq routine
-
-_use_kernal     .byte  0
-
-_irq_handler_init
-		; save all zp scratch registers and the X register as these might be clobbered by the irq routine
-		stx  IRQ_X_REG
-		lda  P8ZP_SCRATCH_B1
-		sta  IRQ_SCRATCH_ZPB1
-		lda  P8ZP_SCRATCH_REG
-		sta  IRQ_SCRATCH_ZPREG
-		lda  P8ZP_SCRATCH_W1
-		sta  IRQ_SCRATCH_ZPWORD1
-		lda  P8ZP_SCRATCH_W1+1
-		sta  IRQ_SCRATCH_ZPWORD1+1
-		lda  P8ZP_SCRATCH_W2
-		sta  IRQ_SCRATCH_ZPWORD2
-		lda  P8ZP_SCRATCH_W2+1
-		sta  IRQ_SCRATCH_ZPWORD2+1
-		; stack protector; make sure we don't clobber the top of the evaluation stack
-		dex
-		dex
-		dex
-		dex
-		dex
-		dex
-		cld
-		rts
-
-_irq_handler_end
-		; restore all zp scratch registers and the X register
-		lda  IRQ_SCRATCH_ZPB1
-		sta  P8ZP_SCRATCH_B1
-		lda  IRQ_SCRATCH_ZPREG
-		sta  P8ZP_SCRATCH_REG
-		lda  IRQ_SCRATCH_ZPWORD1
-		sta  P8ZP_SCRATCH_W1
-		lda  IRQ_SCRATCH_ZPWORD1+1
-		sta  P8ZP_SCRATCH_W1+1
-		lda  IRQ_SCRATCH_ZPWORD2
-		sta  P8ZP_SCRATCH_W2
-		lda  IRQ_SCRATCH_ZPWORD2+1
-		sta  P8ZP_SCRATCH_W2+1
-		ldx  IRQ_X_REG
-		rts
-
-IRQ_X_REG		.byte  0
-IRQ_SCRATCH_ZPB1	.byte  0
-IRQ_SCRATCH_ZPREG	.byte  0
-IRQ_SCRATCH_ZPWORD1	.word  0
-IRQ_SCRATCH_ZPWORD2	.word  0
-
-		}}
-}
-
-asmsub  restore_irq() clobbers(A) {
-	%asm {{
-		sei
-		lda  #<c64.IRQDFRT
-		sta  c64.CINV
-		lda  #>c64.IRQDFRT
-		sta  c64.CINV+1
-		lda  #0
-		sta  c64.IREQMASK	; disable raster irq
-		lda  #%10000001
-		sta  c64.CIA1ICR	; restore CIA1 irq
-		cli
-		rts
-	}}
-}
-
-asmsub  set_rasterirq(uword handler @AY, uword rasterpos @R0, ubyte useKernal @Pc) clobbers(A) {
-	%asm {{
-	        sta  _modified+1
-	        sty  _modified+2
-	        lda  #0
-	        adc  #0
-	        sta  set_irq._use_kernal
-		lda  cx16.r0
-		ldy  cx16.r0+1
-		sei
-		jsr  _setup_raster_irq
-		lda  #<_raster_irq_handler
-		sta  c64.CINV
-		lda  #>_raster_irq_handler
-		sta  c64.CINV+1
-		cli
-		rts
-
-_raster_irq_handler
-		jsr  set_irq._irq_handler_init
-_modified	jsr  $ffff              ; modified
-		jsr  set_irq._irq_handler_end
-                lda  #$ff
-                sta  c64.VICIRQ			; acknowledge raster irq
-		lda  set_irq._use_kernal
-		bne  +
-		; end irq processing - don't use kernal's irq handling
-		pla
-		tay
-		pla
-		tax
-		pla
-		rti
-+		jmp  c64.IRQDFRT                ; continue with kernal irq routine
-
-_setup_raster_irq
-		pha
-		lda  #%01111111
-		sta  c64.CIA1ICR    ; "switch off" interrupts signals from cia-1
-		sta  c64.CIA2ICR    ; "switch off" interrupts signals from cia-2
-		and  c64.SCROLY
-		sta  c64.SCROLY     ; clear most significant bit of raster position
-		lda  c64.CIA1ICR    ; ack previous irq
-		lda  c64.CIA2ICR    ; ack previous irq
-		pla
-		sta  c64.RASTER     ; set the raster line number where interrupt should occur
-		cpy  #0
-		beq  +
-		lda  c64.SCROLY
-		ora  #%10000000
-		sta  c64.SCROLY     ; set most significant bit of raster position
-+		lda  #%00000001
-		sta  c64.IREQMASK   ;enable raster interrupt signals from vic
-		rts
-	}}
-}
-
 }
 
 c128 {
 ; ---- C128 specific registers ----
+    %option no_symbol_prefixing
 
     &ubyte  VM1     = $0A2C         ; shadow for VUC $d018 in text mode
     &ubyte  VM2     = $0A2D         ; shadow for VIC $d018 in bitmap screen mode
@@ -476,50 +298,6 @@ c128 {
 
 ; ---- C128 specific system utility routines: ----
 
-asmsub  init_system()  {
-    ; Initializes the machine to a sane starting state.
-    ; Called automatically by the loader program logic.
-    ; This means that the BASIC, KERNAL and CHARGEN ROMs are banked in,
-    ; the VIC, SID and CIA chips are reset, screen is cleared, and the default IRQ is set.
-    ; Also a different color scheme is chosen to identify ourselves a little.
-    ; Uppercase charset is activated, and all three registers set to 0, status flags cleared.
-    %asm {{
-        sei
-        cld
-        ;;lda  #%00101111                     ; TODO c128 ram and rom bank selection how?
-        ;;sta  $00
-        ;;lda  #%00100111
-        ;;sta  $01
-        jsr  c64.IOINIT
-        jsr  c64.RESTOR
-        jsr  c64.CINT
-        lda  #6
-        sta  c64.EXTCOL
-        lda  #7
-        sta  c64.COLOR
-        lda  #0
-        sta  c64.BGCOL0
-        jsr  c64.disable_runstop_and_charsetswitch
-        clc
-        clv
-        cli
-        rts
-    }}
-}
-
-asmsub  init_system_phase2()  {
-    %asm {{
-        rts     ; no phase 2 steps on the C128
-    }}
-}
-
-asmsub  cleanup_at_exit() {
-    ; executed when the main subroutine does rts
-    %asm {{
-        jmp  c64.enable_runstop_and_charsetswitch
-    }}
-}
-
 asmsub  disable_basic() clobbers(A) {
     %asm {{
         lda $0a04   ; disable BASIC shadow registers
@@ -547,17 +325,226 @@ asmsub  disable_basic() clobbers(A) {
 
 sys {
     ; ------- lowlevel system routines --------
+    %option no_symbol_prefixing
 
     const ubyte target = 128         ;  compilation target specifier.  64 = C64, 128 = C128,  16 = CommanderX16.
 
+asmsub  init_system()  {
+    ; Initializes the machine to a sane starting state.
+    ; Called automatically by the loader program logic.
+    ; This means that the BASIC, KERNAL and CHARGEN ROMs are banked in,
+    ; the VIC, SID and CIA chips are reset, screen is cleared, and the default IRQ is set.
+    ; Also a different color scheme is chosen to identify ourselves a little.
+    ; Uppercase charset is activated.
+    %asm {{
+        sei
+        cld
+        lda  #0
+        sta  $ff00      ; select default bank 15
+        jsr  cbm.IOINIT
+        jsr  cbm.RESTOR
+        jsr  cbm.CINT
+        lda  #6
+        sta  c64.EXTCOL
+        lda  #7
+        sta  cbm.COLOR
+        lda  #0
+        sta  c64.BGCOL0
+        jsr  disable_runstop_and_charsetswitch
+        clc
+        clv
+        cli
+        rts
+    }}
+}
+
+asmsub  init_system_phase2()  {
+    %asm {{
+        rts     ; no phase 2 steps on the C128
+    }}
+}
+
+asmsub  cleanup_at_exit() {
+    ; executed when the main subroutine does rts
+    %asm {{
+        jmp  sys.enable_runstop_and_charsetswitch
+    }}
+}
+
+asmsub  disable_runstop_and_charsetswitch() clobbers(A) {
+    %asm {{
+        lda  #$80
+        sta  247    ; disable charset switching
+        lda  #112
+        sta  808    ; disable run/stop key
+        rts
+    }}
+}
+
+asmsub  enable_runstop_and_charsetswitch() clobbers(A) {
+    %asm {{
+        lda  #0
+        sta  247    ; enable charset switching
+        lda  #110
+        sta  808    ; enable run/stop key
+        rts
+    }}
+}
+
+asmsub  set_irq(uword handler @AY, bool useKernal @Pc) clobbers(A)  {
+	%asm {{
+	        sta  _modified+1
+	        sty  _modified+2
+	        lda  #0
+	        rol  a
+	        sta  _use_kernal
+		sei
+		lda  #<_irq_handler
+		sta  cbm.CINV
+		lda  #>_irq_handler
+		sta  cbm.CINV+1
+		cli
+		rts
+_irq_handler    jsr  _irq_handler_init
+_modified	jsr  $ffff                      ; modified
+		jsr  _irq_handler_end
+		lda  _use_kernal
+		bne  +
+		lda  #$ff
+		sta  c64.VICIRQ			; acknowledge raster irq
+		lda  c64.CIA1ICR		; acknowledge CIA1 interrupt
+		; end irq processing - don't use kernal's irq handling
+		pla
+		tay
+		pla
+		tax
+		pla
+		rti
++		jmp  cbm.IRQDFRT		; continue with normal kernal irq routine
+
+_use_kernal     .byte  0
+
+_irq_handler_init
+		; save all zp scratch registers as these might be clobbered by the irq routine
+		lda  P8ZP_SCRATCH_B1
+		sta  IRQ_SCRATCH_ZPB1
+		lda  P8ZP_SCRATCH_REG
+		sta  IRQ_SCRATCH_ZPREG
+		lda  P8ZP_SCRATCH_W1
+		sta  IRQ_SCRATCH_ZPWORD1
+		lda  P8ZP_SCRATCH_W1+1
+		sta  IRQ_SCRATCH_ZPWORD1+1
+		lda  P8ZP_SCRATCH_W2
+		sta  IRQ_SCRATCH_ZPWORD2
+		lda  P8ZP_SCRATCH_W2+1
+		sta  IRQ_SCRATCH_ZPWORD2+1
+		cld
+		rts
+
+_irq_handler_end
+		; restore all zp scratch registers
+		lda  IRQ_SCRATCH_ZPB1
+		sta  P8ZP_SCRATCH_B1
+		lda  IRQ_SCRATCH_ZPREG
+		sta  P8ZP_SCRATCH_REG
+		lda  IRQ_SCRATCH_ZPWORD1
+		sta  P8ZP_SCRATCH_W1
+		lda  IRQ_SCRATCH_ZPWORD1+1
+		sta  P8ZP_SCRATCH_W1+1
+		lda  IRQ_SCRATCH_ZPWORD2
+		sta  P8ZP_SCRATCH_W2
+		lda  IRQ_SCRATCH_ZPWORD2+1
+		sta  P8ZP_SCRATCH_W2+1
+		rts
+
+IRQ_SCRATCH_ZPB1	.byte  0
+IRQ_SCRATCH_ZPREG	.byte  0
+IRQ_SCRATCH_ZPWORD1	.word  0
+IRQ_SCRATCH_ZPWORD2	.word  0
+
+		}}
+}
+
+asmsub  restore_irq() clobbers(A) {
+	%asm {{
+		sei
+		lda  #<cbm.IRQDFRT
+		sta  cbm.CINV
+		lda  #>cbm.IRQDFRT
+		sta  cbm.CINV+1
+		lda  #0
+		sta  c64.IREQMASK	; disable raster irq
+		lda  #%10000001
+		sta  c64.CIA1ICR	; restore CIA1 irq
+		cli
+		rts
+	}}
+}
+
+asmsub  set_rasterirq(uword handler @AY, uword rasterpos @R0, bool useKernal @Pc) clobbers(A) {
+	%asm {{
+	        sta  _modified+1
+	        sty  _modified+2
+	        lda  #0
+	        rol  a
+	        sta  set_irq._use_kernal
+		lda  cx16.r0
+		ldy  cx16.r0+1
+		sei
+		jsr  _setup_raster_irq
+		lda  #<_raster_irq_handler
+		sta  cbm.CINV
+		lda  #>_raster_irq_handler
+		sta  cbm.CINV+1
+		cli
+		rts
+
+_raster_irq_handler
+		jsr  set_irq._irq_handler_init
+_modified	jsr  $ffff              ; modified
+		jsr  set_irq._irq_handler_end
+        lda  #$ff
+        sta  c64.VICIRQ			; acknowledge raster irq
+		lda  set_irq._use_kernal
+		bne  +
+		; end irq processing - don't use kernal's irq handling
+		pla
+		tay
+		pla
+		tax
+		pla
+		rti
++		jmp  cbm.IRQDFRT                ; continue with kernal irq routine
+
+_setup_raster_irq
+		pha
+		lda  #%01111111
+		sta  c64.CIA1ICR    ; "switch off" interrupts signals from cia-1
+		sta  c64.CIA2ICR    ; "switch off" interrupts signals from cia-2
+		and  c64.SCROLY
+		sta  c64.SCROLY     ; clear most significant bit of raster position
+		lda  c64.CIA1ICR    ; ack previous irq
+		lda  c64.CIA2ICR    ; ack previous irq
+		pla
+		sta  c64.RASTER     ; set the raster line number where interrupt should occur
+		cpy  #0
+		beq  +
+		lda  c64.SCROLY
+		ora  #%10000000
+		sta  c64.SCROLY     ; set most significant bit of raster position
++		lda  #%00000001
+		sta  c64.IREQMASK   ;enable raster interrupt signals from vic
+		rts
+	}}
+}
 
     asmsub  reset_system()  {
         ; Soft-reset the system back to initial power-on Basic prompt.
         %asm {{
             sei
-            ;lda  #14
-            ;sta  $01        ; bank the kernal in       TODO c128 how to do this?
-            jmp  (c64.RESET_VEC)
+            lda  #0
+            sta  $ff00      ; default bank 15
+            jmp  (cbm.RESET_VEC)
         }}
     }
 
@@ -575,9 +562,9 @@ _loop       lda  P8ZP_SCRATCH_W1
             ldx  P8ZP_SCRATCH_B1
             rts
 
-+           lda  c64.TIME_LO
++           lda  cbm.TIME_LO
             sta  P8ZP_SCRATCH_B1
--           lda  c64.TIME_LO
+-           lda  cbm.TIME_LO
             cmp  P8ZP_SCRATCH_B1
             beq  -
 
@@ -734,13 +721,26 @@ _longcopy
         }}
     }
 
+    inline asmsub irqsafe_set_irqd() {
+        %asm {{
+        php
+        sei
+        }}
+    }
+
+    inline asmsub irqsafe_clear_irqd() {
+        %asm {{
+        plp
+        }}
+    }
+
     inline asmsub exit(ubyte returnvalue @A) {
         ; -- immediately exit the program with a return code in the A register
         %asm {{
-            ;lda  #14
-            ;sta  $01        ; bank the kernal in       TODO c128 how to do this?
-            jsr  c64.CLRCHN		; reset i/o channels
-            jsr  c64.enable_runstop_and_charsetswitch
+            lda  #0
+            sta  $ff00          ; default bank 15
+            jsr  cbm.CLRCHN		; reset i/o channels
+            jsr  sys.enable_runstop_and_charsetswitch
             ldx  prog8_lib.orig_stackpointer
             txs
             rts		; return to original caller
@@ -757,113 +757,137 @@ _longcopy
 }
 
 cx16 {
+    %option no_symbol_prefixing
 
     ; the sixteen virtual 16-bit registers that the CX16 has defined in the zeropage
     ; they are simulated on the C128 as well but their location in memory is different
     ; (because there's no room for them in the zeropage)
-    ; $1300-$1bff is unused RAM on C128. We'll use $1a00-$1bff as the lo/hi evalstack.
-    ; the virtual registers are allocated at the bottom of the eval-stack (should be ample space unless
-    ; you're doing insane nesting of expressions...)
-    ; NOTE: the memory location of these registers can change based on the "-esa" compiler option
-    &uword r0  = $1b00
-    &uword r1  = $1b02
-    &uword r2  = $1b04
-    &uword r3  = $1b06
-    &uword r4  = $1b08
-    &uword r5  = $1b0a
-    &uword r6  = $1b0c
-    &uword r7  = $1b0e
-    &uword r8  = $1b10
-    &uword r9  = $1b12
-    &uword r10 = $1b14
-    &uword r11 = $1b16
-    &uword r12 = $1b18
-    &uword r13 = $1b1a
-    &uword r14 = $1b1c
-    &uword r15 = $1b1e
+    ; $1300-$1bff is unused RAM on C128.
+    &uword r0  = $1be0
+    &uword r1  = $1be2
+    &uword r2  = $1be4
+    &uword r3  = $1be6
+    &uword r4  = $1be8
+    &uword r5  = $1bea
+    &uword r6  = $1bec
+    &uword r7  = $1bee
+    &uword r8  = $1bf0
+    &uword r9  = $1bf2
+    &uword r10 = $1bf4
+    &uword r11 = $1bf6
+    &uword r12 = $1bf8
+    &uword r13 = $1bfa
+    &uword r14 = $1bfc
+    &uword r15 = $1bfe
 
-    &word r0s  = $1b00
-    &word r1s  = $1b02
-    &word r2s  = $1b04
-    &word r3s  = $1b06
-    &word r4s  = $1b08
-    &word r5s  = $1b0a
-    &word r6s  = $1b0c
-    &word r7s  = $1b0e
-    &word r8s  = $1b10
-    &word r9s  = $1b12
-    &word r10s = $1b14
-    &word r11s = $1b16
-    &word r12s = $1b18
-    &word r13s = $1b1a
-    &word r14s = $1b1c
-    &word r15s = $1b1e
+    &word r0s  = $1be0
+    &word r1s  = $1be2
+    &word r2s  = $1be4
+    &word r3s  = $1be6
+    &word r4s  = $1be8
+    &word r5s  = $1bea
+    &word r6s  = $1bec
+    &word r7s  = $1bee
+    &word r8s  = $1bf0
+    &word r9s  = $1bf2
+    &word r10s = $1bf4
+    &word r11s = $1bf6
+    &word r12s = $1bf8
+    &word r13s = $1bfa
+    &word r14s = $1bfc
+    &word r15s = $1bfe
 
-    &ubyte r0L  = $1b00
-    &ubyte r1L  = $1b02
-    &ubyte r2L  = $1b04
-    &ubyte r3L  = $1b06
-    &ubyte r4L  = $1b08
-    &ubyte r5L  = $1b0a
-    &ubyte r6L  = $1b0c
-    &ubyte r7L  = $1b0e
-    &ubyte r8L  = $1b10
-    &ubyte r9L  = $1b12
-    &ubyte r10L = $1b14
-    &ubyte r11L = $1b16
-    &ubyte r12L = $1b18
-    &ubyte r13L = $1b1a
-    &ubyte r14L = $1b1c
-    &ubyte r15L = $1b1e
+    &ubyte r0L  = $1be0
+    &ubyte r1L  = $1be2
+    &ubyte r2L  = $1be4
+    &ubyte r3L  = $1be6
+    &ubyte r4L  = $1be8
+    &ubyte r5L  = $1bea
+    &ubyte r6L  = $1bec
+    &ubyte r7L  = $1bee
+    &ubyte r8L  = $1bf0
+    &ubyte r9L  = $1bf2
+    &ubyte r10L = $1bf4
+    &ubyte r11L = $1bf6
+    &ubyte r12L = $1bf8
+    &ubyte r13L = $1bfa
+    &ubyte r14L = $1bfc
+    &ubyte r15L = $1bfe
 
-    &ubyte r0H  = $1b01
-    &ubyte r1H  = $1b03
-    &ubyte r2H  = $1b05
-    &ubyte r3H  = $1b07
-    &ubyte r4H  = $1b09
-    &ubyte r5H  = $1b0b
-    &ubyte r6H  = $1b0d
-    &ubyte r7H  = $1b0f
-    &ubyte r8H  = $1b11
-    &ubyte r9H  = $1b13
-    &ubyte r10H = $1b15
-    &ubyte r11H = $1b17
-    &ubyte r12H = $1b19
-    &ubyte r13H = $1b1b
-    &ubyte r14H = $1b1d
-    &ubyte r15H = $1b1f
+    &ubyte r0H  = $1be1
+    &ubyte r1H  = $1be3
+    &ubyte r2H  = $1be5
+    &ubyte r3H  = $1be7
+    &ubyte r4H  = $1be9
+    &ubyte r5H  = $1beb
+    &ubyte r6H  = $1bed
+    &ubyte r7H  = $1bef
+    &ubyte r8H  = $1bf1
+    &ubyte r9H  = $1bf3
+    &ubyte r10H = $1bf5
+    &ubyte r11H = $1bf7
+    &ubyte r12H = $1bf9
+    &ubyte r13H = $1bfb
+    &ubyte r14H = $1bfd
+    &ubyte r15H = $1bff
 
-    &byte r0sL  = $1b00
-    &byte r1sL  = $1b02
-    &byte r2sL  = $1b04
-    &byte r3sL  = $1b06
-    &byte r4sL  = $1b08
-    &byte r5sL  = $1b0a
-    &byte r6sL  = $1b0c
-    &byte r7sL  = $1b0e
-    &byte r8sL  = $1b10
-    &byte r9sL  = $1b12
-    &byte r10sL = $1b14
-    &byte r11sL = $1b16
-    &byte r12sL = $1b18
-    &byte r13sL = $1b1a
-    &byte r14sL = $1b1c
-    &byte r15sL = $1b1e
+    &byte r0sL  = $1be0
+    &byte r1sL  = $1be2
+    &byte r2sL  = $1be4
+    &byte r3sL  = $1be6
+    &byte r4sL  = $1be8
+    &byte r5sL  = $1bea
+    &byte r6sL  = $1bec
+    &byte r7sL  = $1bee
+    &byte r8sL  = $1bf0
+    &byte r9sL  = $1bf2
+    &byte r10sL = $1bf4
+    &byte r11sL = $1bf6
+    &byte r12sL = $1bf8
+    &byte r13sL = $1bfa
+    &byte r14sL = $1bfc
+    &byte r15sL = $1bfe
 
-    &byte r0sH  = $1b01
-    &byte r1sH  = $1b03
-    &byte r2sH  = $1b05
-    &byte r3sH  = $1b07
-    &byte r4sH  = $1b09
-    &byte r5sH  = $1b0b
-    &byte r6sH  = $1b0d
-    &byte r7sH  = $1b0f
-    &byte r8sH  = $1b11
-    &byte r9sH  = $1b13
-    &byte r10sH = $1b15
-    &byte r11sH = $1b17
-    &byte r12sH = $1b19
-    &byte r13sH = $1b1b
-    &byte r14sH = $1b1d
-    &byte r15sH = $1b1f
+    &byte r0sH  = $1be1
+    &byte r1sH  = $1be3
+    &byte r2sH  = $1be5
+    &byte r3sH  = $1be7
+    &byte r4sH  = $1be9
+    &byte r5sH  = $1beb
+    &byte r6sH  = $1bed
+    &byte r7sH  = $1bef
+    &byte r8sH  = $1bf1
+    &byte r9sH  = $1bf3
+    &byte r10sH = $1bf5
+    &byte r11sH = $1bf7
+    &byte r12sH = $1bf9
+    &byte r13sH = $1bfb
+    &byte r14sH = $1bfd
+    &byte r15sH = $1bff
+
+    asmsub save_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  cx16.r0,y
+            sta  _cx16_vreg_storage,y
+            dey
+            bpl  -
+            rts
+
+    _cx16_vreg_storage
+            .word 0,0,0,0,0,0,0,0
+            .word 0,0,0,0,0,0,0,0
+        }}
+    }
+
+    asmsub restore_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  save_virtual_registers._cx16_vreg_storage,y
+            sta  cx16.r0,y
+            dey
+            bpl  -
+            rts
+        }}
+    }
 }

compiler/res/prog8lib/c128/textio.p8

@@ -1,10 +1,11 @@
-; Prog8 definitions for the Text I/O and Screen routines for the Commodore-64
+; Prog8 definitions for the Text I/O and Screen routines for the Commodore-128
 
 %import syslib
 %import conv
 
 
 txt {
+    %option no_symbol_prefixing
 
 const ubyte DEFAULT_WIDTH = 40
 const ubyte DEFAULT_HEIGHT = 25
@@ -30,10 +31,10 @@ asmsub column(ubyte col @A) clobbers(A, X, Y) {
     ; ---- set the cursor on the given column (starting with 0) on the current line
     %asm {{
         sec
-        jsr  c64.PLOT
+        jsr  cbm.PLOT
         tay
         clc
-        jmp  c64.PLOT
+        jmp  cbm.PLOT
     }}
 }
 
@@ -57,10 +58,10 @@ asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
 	;      (assumes screen matrix is at the default address)
 	%asm {{
 		ldy  #250
--		sta  c64.Screen+250*0-1,y
-		sta  c64.Screen+250*1-1,y
-		sta  c64.Screen+250*2-1,y
-		sta  c64.Screen+250*3-1,y
+-		sta  cbm.Screen+250*0-1,y
+		sta  cbm.Screen+250*1-1,y
+		sta  cbm.Screen+250*2-1,y
+		sta  cbm.Screen+250*3-1,y
 		dey
 		bne  -
 		rts
@@ -72,10 +73,10 @@ asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
 	;      (assumes color matrix is at the default address)
 	%asm {{
 		ldy  #250
--		sta  c64.Colors+250*0-1,y
-		sta  c64.Colors+250*1-1,y
-		sta  c64.Colors+250*2-1,y
-		sta  c64.Colors+250*3-1,y
+-		sta  cbm.Colors+250*0-1,y
+		sta  cbm.Colors+250*1-1,y
+		sta  cbm.Colors+250*2-1,y
+		sta  cbm.Colors+250*3-1,y
 		dey
 		bne  -
 		rts
@@ -83,7 +84,7 @@ asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
 }
 
 sub color (ubyte txtcol) {
-    c64.COLOR = txtcol
+    cbm.COLOR = txtcol
 }
 
 sub lowercase() {
@@ -96,13 +97,12 @@ sub uppercase() {
     c128.VM1 &= ~2
 }
 
-asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
+asmsub  scroll_left  (bool alsocolors @ Pc) clobbers(A, X, Y)  {
 	; ---- scroll the whole screen 1 character to the left
 	;      contents of the rightmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc _scroll_screen
 
 +               ; scroll the screen and the color memory
@@ -110,10 +110,10 @@ asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
 		ldy  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 1,x
-            sta  c64.Screen + 40*row + 0,x
-            lda  c64.Colors + 40*row + 1,x
-            sta  c64.Colors + 40*row + 0,x
+            lda  cbm.Screen + 40*row + 1,x
+            sta  cbm.Screen + 40*row + 0,x
+            lda  cbm.Colors + 40*row + 1,x
+            sta  cbm.Colors + 40*row + 0,x
         .next
 		inx
 		dey
@@ -125,34 +125,32 @@ _scroll_screen  ; scroll only the screen memory
 		ldy  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 1,x
-            sta  c64.Screen + 40*row + 0,x
+            lda  cbm.Screen + 40*row + 1,x
+            sta  cbm.Screen + 40*row + 0,x
         .next
 		inx
 		dey
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_right  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_right  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character to the right
 	;      contents of the leftmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 0,x
-            sta  c64.Screen + 40*row + 1,x
-            lda  c64.Colors + 40*row + 0,x
-            sta  c64.Colors + 40*row + 1,x
+            lda  cbm.Screen + 40*row + 0,x
+            sta  cbm.Screen + 40*row + 1,x
+            lda  cbm.Colors + 40*row + 0,x
+            sta  cbm.Colors + 40*row + 1,x
         .next
 		dex
 		bpl  -
@@ -162,33 +160,31 @@ _scroll_screen  ; scroll only the screen memory
 		ldx  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 0,x
-            sta  c64.Screen + 40*row + 1,x
+            lda  cbm.Screen + 40*row + 0,x
+            sta  cbm.Screen + 40*row + 1,x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_up  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_up  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character up
 	;      contents of the bottom row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx #39
 -
         .for row=1, row<=24, row+=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row-1),x
-            lda  c64.Colors + 40*row,x
-            sta  c64.Colors + 40*(row-1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row-1),x
+            lda  cbm.Colors + 40*row,x
+            sta  cbm.Colors + 40*(row-1),x
         .next
 		dex
 		bpl  -
@@ -198,33 +194,31 @@ _scroll_screen  ; scroll only the screen memory
 		ldx #39
 -
         .for row=1, row<=24, row+=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row-1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row-1),x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_down  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_down  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character down
 	;      contents of the top row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx #39
 -
         .for row=23, row>=0, row-=1
-            lda  c64.Colors + 40*row,x
-            sta  c64.Colors + 40*(row+1),x
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row+1),x
+            lda  cbm.Colors + 40*row,x
+            sta  cbm.Colors + 40*(row+1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row+1),x
         .next
 		dex
 		bpl  -
@@ -234,134 +228,121 @@ _scroll_screen  ; scroll only the screen memory
 		ldx #39
 -
         .for row=23, row>=0, row-=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row+1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row+1),x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-romsub $FFD2 = chrout(ubyte char @ A)    ; for consistency. You can also use c64.CHROUT directly ofcourse.
+romsub $FFD2 = chrout(ubyte char @ A)    ; for consistency. You can also use cbm.CHROUT directly ofcourse.
 
 asmsub  print (str text @ AY) clobbers(A,Y)  {
 	; ---- print null terminated string from A/Y
 	; note: the compiler contains an optimization that will replace
 	;       a call to this subroutine with a string argument of just one char,
-	;       by just one call to c64.CHROUT of that single char.
+	;       by just one call to cbm.CHROUT of that single char.
 	%asm {{
 		sta  P8ZP_SCRATCH_B1
 		sty  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  (P8ZP_SCRATCH_B1),y
 		beq  +
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		bne  -
 +		rts
 	}}
 }
 
-asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub0  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 		pha
 		tya
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 _print_byte_digits
 		pha
 		cpy  #'0'
 		beq  +
 		tya
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		jmp  _ones
 +       pla
         cmp  #'0'
         beq  _ones
-        jsr  c64.CHROUT
+        jsr  cbm.CHROUT
 _ones   txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
+asmsub  print_b  (byte value @ A) clobbers(A,X,Y)  {
 	; ---- print the byte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		pha
 		cmp  #0
 		bpl  +
 		lda  #'-'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 +		pla
 		jsr  conv.byte2decimal
 		jmp  print_ub._print_byte_digits
 	}}
 }
 
-asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubhex  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  +
 		pha
 		lda  #'$'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
 +		jsr  conv.ubyte2hex
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		tya
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubbin  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'%'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 +		ldy  #8
 -		lda  #'0'
 		asl  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'1'
-+		jsr  c64.CHROUT
++		jsr  cbm.CHROUT
 		dey
 		bne  -
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwbin  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
 		pha
@@ -373,7 +354,7 @@ asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwhex  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in hexadecimal form (4 digits)
 	;      (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
@@ -386,28 +367,24 @@ asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw0  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
         beq  +
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		bne  -
-+		ldx  P8ZP_SCRATCH_REG
-		rts
++		rts
 	}}
 }
 
-asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, without left padding 0s
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
-		ldx  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
 		beq  _allzero
@@ -417,25 +394,25 @@ asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
 		bne  -
 
 _gotdigit
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		lda  conv.uword2decimal.decTenThousands,y
 		bne  _gotdigit
 		rts
 _allzero
         lda  #'0'
-        jmp  c64.CHROUT
+        jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
+asmsub  print_w  (word value @ AY) clobbers(A,X,Y)  {
 	; ---- print the (signed) word in A/Y in decimal form, without left padding 0's
 	%asm {{
 		cpy  #0
 		bpl  +
 		pha
 		lda  #'-'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		tya
 		eor  #255
 		tay
@@ -457,7 +434,7 @@ asmsub  input_chars  (uword buffer @ AY) clobbers(A) -> ubyte @ Y  {
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
 		ldy  #0				; char counter = 0
--		jsr  c64.CHRIN
+-		jsr  cbm.CHRIN
 		cmp  #$0d			; return (ascii 13) pressed?
 		beq  +				; yes, end.
 		sta  (P8ZP_SCRATCH_W1),y	; else store char in buffer
@@ -489,7 +466,7 @@ asmsub  setchr  (ubyte col @X, ubyte row @Y, ubyte character @A) clobbers(A, Y)
 _mod		sta  $ffff		; modified
 		rts
 
-_screenrows	.word  $0400 + range(0, 1000, 40)
+_screenrows	.word  cbm.Screen + range(0, 1000, 40)
 	}}
 }
 
@@ -556,7 +533,7 @@ _mod		lda  $ffff		; modified
 	}}
 }
 
-sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
+sub  setcc  (ubyte col, ubyte row, ubyte character, ubyte charcolor)  {
 	; ---- set char+color at the given position on the screen
 	%asm {{
 		lda  row
@@ -568,13 +545,13 @@ sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
 		sta  _colormod+2
 		lda  setchr._screenrows,y
 		clc
-		adc  column
+		adc  col
 		sta  _charmod+1
 		sta  _colormod+1
 		bcc  +
 		inc  _charmod+2
 		inc  _colormod+2
-+		lda  char
++		lda  character
 _charmod	sta  $ffff		; modified
 		lda  charcolor
 _colormod	sta  $ffff		; modified
@@ -582,22 +559,17 @@ _colormod	sta  $ffff		; modified
 	}}
 }
 
-asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
-	; ---- safe wrapper around PLOT kernal routine, to save the X register.
+asmsub  plot  (ubyte col @ Y, ubyte row @ X) {
 	%asm  {{
-		stx  P8ZP_SCRATCH_REG
-		tax
 		clc
-		jsr  c64.PLOT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.PLOT
 	}}
 }
 
 asmsub width() clobbers(X,Y) -> ubyte @A {
     ; -- returns the text screen width (number of columns)
     %asm {{
-        jsr  c64.SCREEN
+        jsr  cbm.SCREEN
         txa
         rts
     }}
@@ -606,7 +578,7 @@ asmsub width() clobbers(X,Y) -> ubyte @A {
 asmsub height() clobbers(X, Y) -> ubyte @A {
     ; -- returns the text screen height (number of rows)
     %asm {{
-        jsr  c64.SCREEN
+        jsr  cbm.SCREEN
         tya
         rts
     }}

compiler/res/prog8lib/c64/floats.asm

@@ -5,13 +5,11 @@ FL_ZERO_const	.byte  0,0,0,0,0		; 0.0
 FL_LOG2_const	.byte  $80, $31, $72, $17, $f8	; log(2)
 
 
-floats_store_reg	.byte  0		; temp storage
-
+floats_temp_var         .byte  0,0,0,0,0        ; temporary storage for a float
 
 ub2float	.proc
 		; -- convert ubyte in SCRATCH_ZPB1 to float at address A/Y
-		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers A, X, Y
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		ldy  P8ZP_SCRATCH_B1
@@ -19,15 +17,12 @@ ub2float	.proc
 		jsr  GIVAYF
 _fac_to_mem	ldx  P8ZP_SCRATCH_W2
 		ldy  P8ZP_SCRATCH_W2+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  MOVMF
 		.pend
 
 b2float		.proc
 		; -- convert byte in SCRATCH_ZPB1 to float at address A/Y
-		;    clobbers A, Y
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers A, X, Y
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		lda  P8ZP_SCRATCH_B1
@@ -37,7 +32,7 @@ b2float		.proc
 
 uw2float	.proc
 		; -- convert uword in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		lda  P8ZP_SCRATCH_W1
@@ -48,7 +43,7 @@ uw2float	.proc
 
 w2float		.proc
 		; -- convert word in SCRATCH_ZPWORD1 to float at address A/Y
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		ldy  P8ZP_SCRATCH_W1
@@ -60,7 +55,7 @@ w2float		.proc
 
 cast_from_uw	.proc
 		; -- uword in A/Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		jsr  GIVUAYFAY
 		jmp  ub2float._fac_to_mem
 		.pend
@@ -68,7 +63,7 @@ cast_from_uw	.proc
 
 cast_from_w	.proc
 		; -- word in A/Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		jsr  GIVAYFAY
 		jmp  ub2float._fac_to_mem
 		.pend
@@ -76,7 +71,7 @@ cast_from_w	.proc
 
 cast_from_ub	.proc
 		; -- ubyte in Y into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		jsr  FREADUY
 		jmp  ub2float._fac_to_mem
 		.pend
@@ -84,169 +79,41 @@ cast_from_ub	.proc
 
 cast_from_b	.proc
 		; -- byte in A into float var at (P8ZP_SCRATCH_W2)
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		jsr  FREADSA
 		jmp  ub2float._fac_to_mem
 		.pend
 
 cast_as_uw_into_ya	.proc               ; also used for float 2 ub
 		; -- cast float at A/Y to uword into Y/A
+		;    clobbers X
 		jsr  MOVFM
 		jmp  cast_FAC1_as_uw_into_ya
 		.pend
 
 cast_as_w_into_ay	.proc               ; also used for float 2 b
 		; -- cast float at A/Y to word into A/Y
+		;    clobbers X
 		jsr  MOVFM
 		jmp  cast_FAC1_as_w_into_ay
 		.pend
 
 cast_FAC1_as_uw_into_ya	.proc               ; also used for float 2 ub
 		; -- cast fac1 to uword into Y/A
-		stx  P8ZP_SCRATCH_REG
-		jsr  GETADR     ; into Y/A
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		;    clobbers X
+		jmp  GETADR     ; into Y/A
 		.pend
 
 cast_FAC1_as_w_into_ay	.proc               ; also used for float 2 b
 		; -- cast fac1 to word into A/Y
-		stx  P8ZP_SCRATCH_REG
+		;    clobbers X
 		jsr  AYINT
 		ldy  floats.AYINT_facmo
 		lda  floats.AYINT_facmo+1
-		ldx  P8ZP_SCRATCH_REG
 		rts
 		.pend
 
 
-stack_b2float	.proc
-		; -- b2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  FREADSA
-		jmp  push_fac1._internal
-		.pend
-
-stack_w2float	.proc
-		; -- w2float operating on the stack
-		inx
-		ldy  P8ESTACK_LO,x
-		lda  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVAYF
-		jmp  push_fac1._internal
-		.pend
-
-stack_ub2float	.proc
-		; -- ub2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		stx  P8ZP_SCRATCH_REG
-		tay
-		lda  #0
-		jsr  GIVAYF
-		jmp  push_fac1._internal
-		.pend
-
-stack_uw2float	.proc
-		; -- uw2float operating on the stack
-		inx
-		lda  P8ESTACK_LO,x
-		ldy  P8ESTACK_HI,x
-		stx  P8ZP_SCRATCH_REG
-		jsr  GIVUAYFAY
-		jmp  push_fac1._internal
-		.pend
-
-stack_float2w	.proc               ; also used for float2b
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG
-		jsr  AYINT
-		ldx  P8ZP_SCRATCH_REG
-		lda  floats.AYINT_facmo
-		sta  P8ESTACK_HI,x
-		lda  floats.AYINT_facmo+1
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-stack_float2uw	.proc               ; also used for float2ub
-		jsr  pop_float_fac1
-		stx  P8ZP_SCRATCH_REG
-		jsr  GETADR
-		ldx  P8ZP_SCRATCH_REG
-		sta  P8ESTACK_HI,x
-		tya
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-push_float	.proc
-		; ---- push mflpt5 in A/Y onto stack
-		; (taking 3 stack positions = 6 bytes of which 1 is padding)
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		ldy  #0
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_HI,x
-		dex
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_HI,x
-		dex
-		iny
-		lda  (P8ZP_SCRATCH_W1),y
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
-pop_float	.proc
-		; ---- pops mflpt5 from stack to memory A/Y
-		; (frees 3 stack positions = 6 bytes of which 1 is padding)
-		sta  P8ZP_SCRATCH_W1
-		sty  P8ZP_SCRATCH_W1+1
-		ldy  #4
-		inx
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		inx
-		lda  P8ESTACK_HI,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		inx
-		lda  P8ESTACK_HI,x
-		sta  (P8ZP_SCRATCH_W1),y
-		dey
-		lda  P8ESTACK_LO,x
-		sta  (P8ZP_SCRATCH_W1),y
-		rts
-		.pend
-
-pop_float_fac1	.proc
-		; -- pops float from stack into FAC1
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jmp  MOVFM
-		.pend
-
 copy_float	.proc
 		; -- copies the 5 bytes of the mflt value pointed to by P8ZP_SCRATCH_W1,
 		;    into the 5 bytes pointed to by A/Y.  Clobbers A,Y.
@@ -272,25 +139,23 @@ copy_float	.proc
 
 inc_var_f	.proc
 		; -- add 1 to float pointed to by A/Y
+		;    clobbers X
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG
 		jsr  MOVFM
 		lda  #<FL_ONE_const
 		ldy  #>FL_ONE_const
 		jsr  FADD
 		ldx  P8ZP_SCRATCH_W1
 		ldy  P8ZP_SCRATCH_W1+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  MOVMF
 		.pend
 
 dec_var_f	.proc
 		; -- subtract 1 from float pointed to by A/Y
+		;    clobbers X
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
-		stx  P8ZP_SCRATCH_REG
 		lda  #<FL_ONE_const
 		ldy  #>FL_ONE_const
 		jsr  MOVFM
@@ -299,23 +164,7 @@ dec_var_f	.proc
 		jsr  FSUB
 		ldx  P8ZP_SCRATCH_W1
 		ldy  P8ZP_SCRATCH_W1+1
-		jsr  MOVMF
-		ldx  P8ZP_SCRATCH_REG
-		rts
-		.pend
-
-
-pop_2_floats_f2_in_fac1	.proc
-		; -- pop 2 floats from stack, load the second one in FAC1 as well
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jmp  MOVFM
+		jmp  MOVMF
 		.pend
 
 
@@ -323,71 +172,9 @@ fmath_float1	.byte 0,0,0,0,0	; storage for a mflpt5 value
 fmath_float2	.byte 0,0,0,0,0	; storage for a mflpt5 value
 
 
-push_fac1	.proc
-		; -- push the float in FAC1 onto the stack
-		stx  P8ZP_SCRATCH_REG
-_internal	ldx  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  MOVMF
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		ldx  P8ZP_SCRATCH_REG
-		jmp  push_float
-		.pend
-
-div_f		.proc
-		; -- push f1/f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FDIV
-		jmp  push_fac1._internal
-		.pend
-
-add_f		.proc
-		; -- push f1+f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FADD
-		jmp  push_fac1._internal
-		.pend
-
-sub_f		.proc
-		; -- push f1-f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FSUB
-		jmp  push_fac1._internal
-		.pend
-
-mul_f		.proc
-		; -- push f1*f2 on stack
-		jsr  pop_2_floats_f2_in_fac1
-		stx  P8ZP_SCRATCH_REG
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  FMULT
-		jmp  push_fac1._internal
-		.pend
-
-neg_f		.proc
-		; -- toggle the sign bit on the stack
-		lda  P8ESTACK_HI+3,x
-		eor  #$80
-		sta  P8ESTACK_HI+3,x
-		rts
-		.pend
-
 var_fac1_less_f	.proc
-		; -- is the float in FAC1 < the variable AY?
-		stx  P8ZP_SCRATCH_REG
+		; -- is the float in FAC1 < the variable AY? Result in A. Clobbers X.
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #255
 		beq  +
 		lda  #0
@@ -397,10 +184,8 @@ var_fac1_less_f	.proc
 		.pend
 
 var_fac1_lesseq_f	.proc
-		; -- is the float in FAC1 <= the variable AY?
-		stx  P8ZP_SCRATCH_REG
+		; -- is the float in FAC1 <= the variable AY?  Result in A. Clobbers X.
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #0
 		beq  +
 		cmp  #255
@@ -412,10 +197,8 @@ var_fac1_lesseq_f	.proc
 		.pend
 
 var_fac1_greater_f	.proc
-		; -- is the float in FAC1 > the variable AY?
-		stx  P8ZP_SCRATCH_REG
+		; -- is the float in FAC1 > the variable AY?  Result in A. Clobbers X.
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #1
 		beq  +
 		lda  #0
@@ -425,10 +208,8 @@ var_fac1_greater_f	.proc
 		.pend
 
 var_fac1_greatereq_f	.proc
-		; -- is the float in FAC1 >= the variable AY?
-		stx  P8ZP_SCRATCH_REG
+		; -- is the float in FAC1 >= the variable AY?  Result in A. Clobbers X.
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #0
 		beq  +
 		cmp  #1
@@ -439,17 +220,23 @@ var_fac1_greatereq_f	.proc
 		rts
 		.pend
 
+var_fac1_equal_f	.proc
+		; -- are the floats numbers in FAC1 and the variable AY *not* identical?   Result in A. Clobbers X.
+		jsr  FCOMP
+		and  #1
+		eor  #1
+		rts
+		.pend
+
 var_fac1_notequal_f	.proc
-		; -- are the floats numbers in FAC1 and the variable AY *not* identical?
-		stx  P8ZP_SCRATCH_REG
+		; -- are the floats numbers in FAC1 and the variable AY *not* identical?   Result in A. Clobbers X.
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		and  #1
 		rts
 		.pend
 
 vars_equal_f	.proc
-		; -- are the mflpt5 numbers in P8ZP_SCRATCH_W1 and AY identical?
+		; -- are the mflpt5 numbers in P8ZP_SCRATCH_W1 and AY identical?  Result in A
 		sta  P8ZP_SCRATCH_W2
 		sty  P8ZP_SCRATCH_W2+1
 		ldy  #0
@@ -478,51 +265,13 @@ _false		lda  #0
 		rts
 		.pend
 
-equal_f		.proc
-		; -- are the two mflpt5 numbers on the stack identical?
-		inx
-		inx
-		inx
-		inx
-		lda  P8ESTACK_LO-3,x
-		cmp  P8ESTACK_LO,x
-		bne  _equals_false
-		lda  P8ESTACK_LO-2,x
-		cmp  P8ESTACK_LO+1,x
-		bne  _equals_false
-		lda  P8ESTACK_LO-1,x
-		cmp  P8ESTACK_LO+2,x
-		bne  _equals_false
-		lda  P8ESTACK_HI-2,x
-		cmp  P8ESTACK_HI+1,x
-		bne  _equals_false
-		lda  P8ESTACK_HI-1,x
-		cmp  P8ESTACK_HI+2,x
-		bne  _equals_false
-_equals_true	lda  #1
-_equals_store	inx
-		sta  P8ESTACK_LO+1,x
-		rts
-_equals_false	lda  #0
-		beq  _equals_store
-		.pend
-
-notequal_f	.proc
-		; -- are the two mflpt5 numbers on the stack different?
-		jsr  equal_f
-		eor  #1		; invert the result
-		sta  P8ESTACK_LO+1,x
-		rts
-		.pend
 
 vars_less_f	.proc
-		; -- is float in AY < float in P8ZP_SCRATCH_W2 ?
+		; -- is float in AY < float in P8ZP_SCRATCH_W2 ?   Result in A. Clobbers X.
 		jsr  MOVFM
 		lda  P8ZP_SCRATCH_W2
 		ldy  P8ZP_SCRATCH_W2+1
-		stx  P8ZP_SCRATCH_REG
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #255
 		bne  +
 		lda  #1
@@ -532,13 +281,11 @@ vars_less_f	.proc
 		.pend
 
 vars_lesseq_f	.proc
-		; -- is float in AY <= float in P8ZP_SCRATCH_W2 ?
+		; -- is float in AY <= float in P8ZP_SCRATCH_W2 ?  Result in A. Clobbers X.
 		jsr  MOVFM
 		lda  P8ZP_SCRATCH_W2
 		ldy  P8ZP_SCRATCH_W2+1
-		stx  P8ZP_SCRATCH_REG
 		jsr  FCOMP
-		ldx  P8ZP_SCRATCH_REG
 		cmp  #255
 		bne  +
 -		lda  #1
@@ -550,7 +297,7 @@ vars_lesseq_f	.proc
 		.pend
 
 less_f		.proc
-		; -- is f1 < f2?
+		; -- is f1 < f2?    Result in A. Clobbers X.
 		jsr  compare_floats
 		cmp  #255
 		beq  compare_floats._return_true
@@ -559,7 +306,7 @@ less_f		.proc
 
 
 lesseq_f	.proc
-		; -- is f1 <= f2?
+		; -- is f1 <= f2?  Result in A. Clobbers X.
 		jsr  compare_floats
 		cmp  #255
 		beq  compare_floats._return_true
@@ -569,7 +316,7 @@ lesseq_f	.proc
 		.pend
 
 greater_f	.proc
-		; -- is f1 > f2?
+		; -- is f1 > f2?  Result in A. Clobbers X.
 		jsr  compare_floats
 		cmp  #1
 		beq  compare_floats._return_true
@@ -577,7 +324,7 @@ greater_f	.proc
 		.pend
 
 greatereq_f	.proc
-		; -- is f1 >= f2?
+		; -- is f1 >= f2?  Result in A. Clobbers X.
 		jsr  compare_floats
 		cmp  #1
 		beq  compare_floats._return_true
@@ -586,32 +333,9 @@ greatereq_f	.proc
 		bne  compare_floats._return_false
 		.pend
 
-compare_floats	.proc
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  pop_float
-		lda  #<fmath_float1
-		ldy  #>fmath_float1
-		jsr  MOVFM		; fac1 = flt1
-		lda  #<fmath_float2
-		ldy  #>fmath_float2
-		stx  P8ZP_SCRATCH_REG
-		jsr  FCOMP		; A = flt1 compared with flt2 (0=equal, 1=flt1>flt2, 255=flt1<flt2)
-		ldx  P8ZP_SCRATCH_REG
-		rts
-_return_false	lda  #0
-_return_result  sta  P8ESTACK_LO,x
-		dex
-		rts
-_return_true	lda  #1
-		bne  _return_result
-		.pend
-
 set_array_float_from_fac1	.proc
 		; -- set the float in FAC1 in the array (index in A, array in P8ZP_SCRATCH_W1)
+		;    clobbers X
 		sta  P8ZP_SCRATCH_B1
 		asl  a
 		asl  a
@@ -622,11 +346,8 @@ set_array_float_from_fac1	.proc
 		adc  P8ZP_SCRATCH_W1
 		bcc  +
 		iny
-+		stx  floats_store_reg
-		tax
-		jsr  MOVMF
-		ldx  floats_store_reg
-		rts
++		tax
+		jmp  MOVMF
 		.pend
 
 
@@ -669,54 +390,63 @@ set_array_float		.proc
 		.pend
 
 
-equal_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		beq  _true
-		bne  _false
-_true		lda  #1
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-_false		lda  #0
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
+pushFAC1    .proc
+	;-- push floating point in FAC onto the cpu stack
+	; save return address
+	pla
+	sta  P8ZP_SCRATCH_W2
+	pla
+	sta  P8ZP_SCRATCH_W2+1
+	ldx  #<floats.floats_temp_var
+	ldy  #>floats.floats_temp_var
+	jsr  floats.MOVMF
+	lda  floats.floats_temp_var
+	pha
+	lda  floats.floats_temp_var+1
+	pha
+	lda  floats.floats_temp_var+2
+	pha
+	lda  floats.floats_temp_var+3
+	pha
+	lda  floats.floats_temp_var+4
+	pha
+	; re-push return address
+	lda  P8ZP_SCRATCH_W2+1
+	pha
+	lda  P8ZP_SCRATCH_W2
+	pha
+	rts
+	.pend
 
-notequal_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		bne  equal_zero._true
-		beq  equal_zero._false
-		.pend
+popFAC .proc
+	; -- pop floating point value from cpu stack into FAC1 or FAC2 (
+	;    carry flag clear=FAC1, carry set=FAC2
+	; save return address
+	pla
+	sta  P8ZP_SCRATCH_W2
+	pla
+	sta  P8ZP_SCRATCH_W2+1
+	pla
+	sta  floats.floats_temp_var+4
+	pla
+	sta  floats.floats_temp_var+3
+	pla
+	sta  floats.floats_temp_var+2
+	pla
+	sta  floats.floats_temp_var+1
+	pla
+	sta  floats.floats_temp_var
+	lda  #<floats.floats_temp_var
+	ldy  #>floats.floats_temp_var
+	bcs  +
+	jsr  floats.MOVFM
+	jmp  ++
++       jsr  floats.CONUPK
++	; re-push return address
+	lda  P8ZP_SCRATCH_W2+1
+	pha
+	lda  P8ZP_SCRATCH_W2
+	pha
+	rts
+	.pend
 
-greater_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		beq  equal_zero._false
-		bpl  equal_zero._true
-		jmp  equal_zero._false
-		.pend
-
-less_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		bmi  equal_zero._true
-		jmp  equal_zero._false
-		.pend
-
-greaterequal_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		bpl  equal_zero._true
-		jmp  equal_zero._false
-		.pend
-
-lessequal_zero	.proc
-		jsr  floats.pop_float_fac1
-		jsr  floats.SIGN
-		beq  equal_zero._true
-		bmi  equal_zero._true
-		jmp  equal_zero._false
-		.pend

compiler/res/prog8lib/c64/floats.p8

@@ -5,6 +5,7 @@
 
 floats {
 	; ---- this block contains C-64 floating point related functions ----
+    %option no_symbol_prefixing
 
         const float  PI     = 3.141592653589793
         const float  TWOPI  = 6.283185307179586
@@ -45,7 +46,7 @@ romsub $b391 = GIVAYF(ubyte lo @ Y, ubyte hi @ A) clobbers(A,X,Y)
 
 romsub $b3a2 = FREADUY(ubyte value @ Y) clobbers(A,X,Y)     ; 8 bit unsigned Y -> float in fac1
 romsub $bc3c = FREADSA(byte value @ A) clobbers(A,X,Y)      ; 8 bit signed A -> float in fac1
-romsub $b7b5 = FREADSTR(ubyte length @ A) clobbers(A,X,Y)   ; str -> fac1, $22/23 must point to string, A=string length
+romsub $b7b5 = FREADSTR(ubyte length @ A) clobbers(A,X,Y)   ; str -> fac1, $22/23 must point to string, A=string length.  Also see parse_f()
 romsub $aabc = FPRINTLN() clobbers(A,X,Y)                   ; print string of fac1, on one line (= with newline) destroys fac1.  (consider FOUT + STROUT as well)
 romsub $bddd = FOUT() clobbers(X) -> uword @ AY             ; fac1 -> string, address returned in AY ($0100)
 
@@ -60,8 +61,8 @@ romsub $b853 = FSUBT() clobbers(A,X,Y)                      ; fac1 = fac2-fac1
 romsub $b850 = FSUB(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt from A/Y - fac1
 romsub $ba2b = FMULTT() clobbers(A,X,Y)                     ; fac1 *= fac2
 romsub $ba28 = FMULT(uword mflpt @ AY) clobbers(A,X,Y)      ; fac1 *= mflpt value from A/Y
-romsub $bb12 = FDIVT() clobbers(A,X,Y)                      ; fac1 = fac2/fac1  (remainder in fac2)  mind the order of the operands
-romsub $bb0f = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1  (remainder in fac2)
+romsub $bb12 = FDIVT() clobbers(A,X,Y)                      ; fac1 = fac2/fac1  mind the order of the operands
+romsub $bb0f = FDIV(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = mflpt in A/Y / fac1
 romsub $bf7b = FPWRT() clobbers(A,X,Y)                      ; fac1 = fac2 ** fac1
 romsub $bf78 = FPWR(uword mflpt @ AY) clobbers(A,X,Y)       ; fac1 = fac2 ** mflpt from A/Y
 romsub $bd7e = FINLOG(byte value @A) clobbers (A, X, Y)     ; fac1 += signed byte in A
@@ -158,12 +159,19 @@ asmsub  GETADRAY  () clobbers(X) -> uword @ AY  {
 
 sub rndf() -> float {
     %asm {{
-        stx  P8ZP_SCRATCH_REG
         lda  #1
         jsr  FREADSA
-        jsr  RND		; rng into fac1
-        ldx  P8ZP_SCRATCH_REG
-        rts
+        jmp  RND		; rng into fac1
+    }}
+}
+
+asmsub parse_f(str value @AY) -> float @FAC1 {
+    %asm {{
+        sta  $22
+        sty  $23
+        jsr  prog8_lib.strlen
+        tya
+        jmp  FREADSTR
     }}
 }
 

compiler/res/prog8lib/c64/floats_funcs.asm

@@ -1,13 +1,6 @@
 ; --- floating point builtin functions
 
 
-func_sign_f_stack	.proc
-		jsr  func_sign_f_into_A
-		sta  P8ESTACK_LO,x
-		dex
-		rts
-		.pend
-
 func_sign_f_into_A	.proc
 		jsr  MOVFM
 		jmp  SIGN
@@ -144,3 +137,13 @@ func_all_f_stack	.proc
 		jsr  a_times_5
 		jmp  prog8_lib.func_all_b_stack
 		.pend
+
+func_abs_f_into_FAC1    .proc
+        jsr  MOVFM
+        jmp  ABS
+	.pend
+
+func_sqrt_into_FAC1     .proc
+        jsr  MOVFM
+        jmp  SQR
+	.pend

compiler/res/prog8lib/c64/graphics.p8

@@ -1,33 +1,43 @@
-%import textio
-
 ; bitmap pixel graphics module for the C64
 ; only black/white monochrome 320x200 for now
-; assumes bitmap screen memory is $2000-$3fff
+;
+; Sets character matrix and bitmap screen memory at a higher memory location $5c00-$7fff
+; so that the program itself can be larger without starting to overwrite the graphics memory.
 
 graphics {
-    const uword BITMAP_ADDRESS = $2000
+    %option no_symbol_prefixing
+
     const uword WIDTH = 320
     const ubyte HEIGHT = 200
 
+    const uword BITMAP_ADDRESS = $6000      ; MUST BE IN REGULAR RAM if you are not messing with ROM/RAM banking. (and $2000-aligned)
+                                            ; note: this constant is also used in a asm multiplication table below!
+    const uword CHARS_ADDRESS = $5c00       ; must be in same vic memory bank as the bitmap.
+
     sub enable_bitmap_mode() {
         ; enable bitmap screen, erase it and set colors to black/white.
-        c64.SCROLY = %00111011
-        c64.SCROLX = %00001000
-        c64.VMCSB = (c64.VMCSB & %11110000) | %00001000   ; $2000-$3fff
         clear_screen(1, 0)
+        c64.SCROLY = %00111011      ; enable bitmap graphics mode
+        c64.SCROLX = %00001000      ; 40 column mode, no scrolling, multicolor bitmap off
+        c64.VMCSB = (lsb(CHARS_ADDRESS >> 6) & $F0) | (lsb((BITMAP_ADDRESS & $3fff) / $0800) << 1)     ; set bitmap address
+        c64.CIA2DDRA |= %11
+        c64.CIA2PRA = lsb(BITMAP_ADDRESS >> 14) ^ 3     ; set VIC bank.
     }
 
     sub disable_bitmap_mode() {
-        ; enables text mode, erase the text screen, color white
-        c64.SCROLY = %00011011
-        c64.SCROLX = %00001000
-        c64.VMCSB = (c64.VMCSB & %11110000) | %00000100   ; $1000-$2fff
-        txt.fill_screen(' ', 1)
+        ; enables erase the text screen, text mode
+        sys.memset(CHARS_ADDRESS, 40*25, sc:' ')
+        c64.SCROLY = %00011011      ; disable bitmap graphics mode
+        c64.SCROLX = %00001000      ; 40 column mode, no scrolling
+        c64.VMCSB  = %00010100      ; screen addresses back to defaults
+        c64.CIA2DDRA |= %11
+        c64.CIA2PRA = %11   ; back to VIC bank 0.
     }
 
     sub clear_screen(ubyte pixelcolor, ubyte bgcolor) {
         sys.memset(BITMAP_ADDRESS, 320*200/8, 0)
-        txt.fill_screen(pixelcolor << 4 | bgcolor, 0)
+        sys.memset(CHARS_ADDRESS, 40*25, pixelcolor << 4 | bgcolor)
+        sys.memset(cbm.Colors, 40*25, 255)
     }
 
     sub line(uword @zp x1, ubyte @zp y1, uword @zp x2, ubyte @zp y2) {
@@ -125,46 +135,46 @@ graphics {
         }
     }
 
-    sub rect(uword x, ubyte y, uword width, ubyte height) {
+    sub rect(uword xx, ubyte yy, uword width, ubyte height) {
         if width==0 or height==0
             return
-        horizontal_line(x, y, width)
+        horizontal_line(xx, yy, width)
         if height==1
             return
-        horizontal_line(x, y+height-1, width)
-        vertical_line(x, y+1, height-2)
+        horizontal_line(xx, yy+height-1, width)
+        vertical_line(xx, yy+1, height-2)
         if width==1
             return
-        vertical_line(x+width-1, y+1, height-2)
+        vertical_line(xx+width-1, yy+1, height-2)
     }
 
-    sub fillrect(uword x, ubyte y, uword width, ubyte height) {
+    sub fillrect(uword xx, ubyte yy, uword width, ubyte height) {
         if width==0
             return
         repeat height {
-            horizontal_line(x, y, width)
-            y++
+            horizontal_line(xx, yy, width)
+            yy++
         }
     }
 
-    sub horizontal_line(uword x, ubyte y, uword length) {
+    sub horizontal_line(uword xx, ubyte yy, uword length) {
         if length<8 {
-            internal_plotx=x
+            internal_plotx=xx
             repeat lsb(length) {
-                internal_plot(y)
+                internal_plot(yy)
                 internal_plotx++
             }
             return
         }
 
-        ubyte separate_pixels = lsb(x) & 7
-        uword addr = get_y_lookup(y) + (x&$fff8)
+        ubyte separate_pixels = lsb(xx) & 7
+        uword pixaddr = get_y_lookup(yy) + (xx&$fff8)
 
         if separate_pixels {
             %asm {{
-                lda  addr
+                lda  pixaddr
                 sta  P8ZP_SCRATCH_W1
-                lda  addr+1
+                lda  pixaddr+1
                 sta  P8ZP_SCRATCH_W1+1
                 ldy  separate_pixels
                 lda  hline_filled_right,y
@@ -173,7 +183,7 @@ graphics {
                 ora  (P8ZP_SCRATCH_W1),y
                 sta  (P8ZP_SCRATCH_W1),y
             }}
-            addr += 8
+            pixaddr += 8
             length += separate_pixels
             length -= 8
         }
@@ -183,16 +193,15 @@ graphics {
                 lda  length
                 and  #7
                 sta  separate_pixels
-                stx  P8ZP_SCRATCH_REG
                 lsr  length+1
                 ror  length
                 lsr  length+1
                 ror  length
                 lsr  length+1
                 ror  length
-                lda  addr
+                lda  pixaddr
                 sta  _modified+1
-                lda  addr+1
+                lda  pixaddr+1
                 sta  _modified+2
                 lda  length
                 ora  length+1
@@ -208,8 +217,7 @@ _modified       stx  $ffff      ; modified
                 inc  _modified+2
 +               dey
                 bne  _modified
-_zero           ldx  P8ZP_SCRATCH_REG
-
+_zero
                 ldy  separate_pixels
                 beq  hline_zero2
                 lda  _modified+1
@@ -227,11 +235,11 @@ hline_zero2
         }
     }
 
-    sub vertical_line(uword x, ubyte y, ubyte height) {
-        internal_plotx = x
+    sub vertical_line(uword xx, ubyte yy, ubyte height) {
+        internal_plotx = xx
         repeat height {
-            internal_plot(y)
-            y++
+            internal_plot(yy)
+            yy++
         }
     }
 
@@ -300,8 +308,8 @@ hline_zero2
 ; here is the non-asm code for the plot routine below:
 ;    sub plot_nonasm(uword px, ubyte py) {
 ;        ubyte[] ormask = [128, 64, 32, 16, 8, 4, 2, 1]
-;        uword addr = BITMAP_ADDRESS + 320*(py>>3) + (py & 7) + (px & %0000000111111000)
-;        @(addr) |= ormask[lsb(px) & 7]
+;        uword pixaddr = BITMAP_ADDRESS + 320*(py>>3) + (py & 7) + (px & %0000000111111000)
+;        @(pixaddr) |= ormask[lsb(px) & 7]
 ;    }
 
     inline asmsub  plot(uword plotx @XY, ubyte ploty @A) clobbers (A, X, Y) {
@@ -352,7 +360,7 @@ _ormask     .byte 128, 64, 32, 16, 8, 4, 2, 1
 ; the y lookup tables encodes this formula:  BITMAP_ADDRESS + 320*(py>>3) + (py & 7)    (y from 0..199)
 ; We use the 64tass syntax for range expressions to calculate this table on assembly time.
 
-_plot_y_values := $2000 + 320*(range(200)>>3) + (range(200) & 7)
+_plot_y_values := BITMAP_ADDRESS + 320*(range(200)>>3) + (range(200) & 7)
 
 _y_lookup_lo    .byte  <_plot_y_values
 _y_lookup_hi    .byte  >_plot_y_values
@@ -360,7 +368,7 @@ _y_lookup_hi    .byte  >_plot_y_values
         }}
     }
 
-    asmsub get_y_lookup(ubyte y @Y) -> uword @AY {
+    asmsub get_y_lookup(ubyte yy @Y) -> uword @AY {
         %asm {{
             lda  internal_plot._y_lookup_lo,y
             pha

compiler/res/prog8lib/c64/syslib.p8

@@ -1,13 +1,20 @@
 ; Prog8 definitions for the Commodore-64
 ; Including memory registers, I/O registers, Basic and Kernal subroutines.
 
-c64 {
+
+cbm {
+
+    ; Commodore (CBM) common variables, vectors and kernal routines
+
+    %option no_symbol_prefixing
+
         &ubyte  TIME_HI         = $a0       ; software jiffy clock, hi byte
         &ubyte  TIME_MID        = $a1       ;  .. mid byte
         &ubyte  TIME_LO         = $a2       ;    .. lo byte. Updated by IRQ every 1/60 sec
         &ubyte  STATUS          = $90       ; kernal status variable for I/O
         &ubyte  STKEY           = $91       ; various keyboard statuses (updated by IRQ)
         &ubyte  SFDX            = $cb       ; current key pressed (matrix value) (updated by IRQ)
+        &ubyte  SHFLAG          = $028d     ; various modifier key status (updated by IRQ)
 
         &ubyte  COLOR           = $0286     ; cursor color
         &ubyte  HIBASE          = $0288     ; screen base address / 256 (hi-byte of screen memory address)
@@ -49,7 +56,88 @@ c64 {
         const  uword  Screen    = $0400     ; to have this as an array[40*25] the compiler would have to support array size > 255
         const  uword  Colors    = $d800     ; to have this as an array[40*25] the compiler would have to support array size > 255
 
+
+; ---- CBM ROM kernal routines (C64 addresses) ----
+
+romsub $AB1E = STROUT(uword strptr @ AY) clobbers(A, X, Y)      ; print null-terminated string (use txt.print instead)
+romsub $E544 = CLEARSCR() clobbers(A,X,Y)                       ; clear the screen
+romsub $E566 = HOMECRSR() clobbers(A,X,Y)                       ; cursor to top left of screen
+romsub $EA31 = IRQDFRT() clobbers(A,X,Y)                        ; default IRQ routine
+romsub $EA81 = IRQDFEND() clobbers(A,X,Y)                       ; default IRQ end/cleanup
+romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
+romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
+romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
+romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
+romsub $FF8D = VECTOR(uword userptr @ XY, bool dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
+romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
+romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
+romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
+romsub $FF99 = MEMTOP(uword address @ XY, bool dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
+romsub $FF9C = MEMBOT(uword address @ XY, bool dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
+romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
+romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
+romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
+romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
+romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
+romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
+romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
+romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
+romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
+romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ Y)   ; set logical file parameters
+romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
+romsub $FFC0 = OPEN() clobbers(X,Y) -> bool @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
+romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
+romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> bool @Pc    ; (via 798 ($31E)) define an input channel
+romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
+romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
+romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
+romsub $FFD2 = CHROUT(ubyte character @ A)                      ; (via 806 ($326)) output a character
+romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> bool @Pc, ubyte @ A, uword @ XY     ; (via 816 ($330)) load from device
+romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> bool @ Pc, ubyte @ A          ; (via 818 ($332)) save to a device
+romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
+romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
+romsub $FFE1 = STOP() clobbers(X) -> bool @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
+romsub $FFE4 = GETIN() clobbers(X,Y) -> bool @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
+romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
+romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
+romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
+romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, bool dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
+romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
+
+asmsub STOP2() clobbers(X) -> ubyte @A  {
+    ; -- check if STOP key was pressed, returns true if so.  More convenient to use than STOP() because that only sets the carry status flag.
+    %asm {{
+        jsr  cbm.STOP
+        beq  +
+        lda  #0
+        rts
++       lda  #1
+        rts
+    }}
+}
+
+asmsub RDTIM16() clobbers(X) -> uword @AY {
+    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
+    %asm {{
+        jsr  cbm.RDTIM
+        pha
+        txa
+        tay
+        pla
+        rts
+    }}
+}
+
+}
+
+c64 {
+        ; C64 I/O registers (VIC, SID, CIA)
+
+    %option no_symbol_prefixing
+
         ; the default locations of the 8 sprite pointers (store address of sprite / 64)
+        ; (depending on the VIC bank and screen ram address selection these can be shifted around though,
+        ; see the two routines after this for a dynamic way of determining the correct memory location)
         &ubyte  SPRPTR0         = 2040
         &ubyte  SPRPTR1         = 2041
         &ubyte  SPRPTR2         = 2042
@@ -199,92 +287,47 @@ c64 {
 ; ---- end of SID registers ----
 
 
-; ---- C64 ROM kernal routines ----
+    sub get_vic_memory_base() -> uword {
+        ; one of the 4 possible banks. $0000/$4000/$8000/$c000.
+        c64.CIA2DDRA |= %11
+        return ((c64.CIA2PRA & 3) ^ 3) as uword << 14
+    }
 
-romsub $AB1E = STROUT(uword strptr @ AY) clobbers(A, X, Y)      ; print null-terminated string (use txt.print instead)
-romsub $E544 = CLEARSCR() clobbers(A,X,Y)                       ; clear the screen
-romsub $E566 = HOMECRSR() clobbers(A,X,Y)                       ; cursor to top left of screen
-romsub $EA31 = IRQDFRT() clobbers(A,X,Y)                        ; default IRQ routine
-romsub $EA81 = IRQDFEND() clobbers(A,X,Y)                       ; default IRQ end/cleanup
-romsub $FF81 = CINT() clobbers(A,X,Y)                           ; (alias: SCINIT) initialize screen editor and video chip
-romsub $FF84 = IOINIT() clobbers(A, X)                          ; initialize I/O devices (CIA, SID, IRQ)
-romsub $FF87 = RAMTAS() clobbers(A,X,Y)                         ; initialize RAM, tape buffer, screen
-romsub $FF8A = RESTOR() clobbers(A,X,Y)                         ; restore default I/O vectors
-romsub $FF8D = VECTOR(uword userptr @ XY, ubyte dir @ Pc) clobbers(A,Y)     ; read/set I/O vector table
-romsub $FF90 = SETMSG(ubyte value @ A)                          ; set Kernal message control flag
-romsub $FF93 = SECOND(ubyte address @ A) clobbers(A)            ; (alias: LSTNSA) send secondary address after LISTEN
-romsub $FF96 = TKSA(ubyte address @ A) clobbers(A)              ; (alias: TALKSA) send secondary address after TALK
-romsub $FF99 = MEMTOP(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set top of memory  pointer
-romsub $FF9C = MEMBOT(uword address @ XY, ubyte dir @ Pc) -> uword @ XY     ; read/set bottom of memory  pointer
-romsub $FF9F = SCNKEY() clobbers(A,X,Y)                         ; scan the keyboard
-romsub $FFA2 = SETTMO(ubyte timeout @ A)                        ; set time-out flag for IEEE bus
-romsub $FFA5 = ACPTR() -> ubyte @ A                             ; (alias: IECIN) input byte from serial bus
-romsub $FFA8 = CIOUT(ubyte databyte @ A)                        ; (alias: IECOUT) output byte to serial bus
-romsub $FFAB = UNTLK() clobbers(A)                              ; command serial bus device to UNTALK
-romsub $FFAE = UNLSN() clobbers(A)                              ; command serial bus device to UNLISTEN
-romsub $FFB1 = LISTEN(ubyte device @ A) clobbers(A)             ; command serial bus device to LISTEN
-romsub $FFB4 = TALK(ubyte device @ A) clobbers(A)               ; command serial bus device to TALK
-romsub $FFB7 = READST() -> ubyte @ A                            ; read I/O status word
-romsub $FFBA = SETLFS(ubyte logical @ A, ubyte device @ X, ubyte secondary @ Y)   ; set logical file parameters
-romsub $FFBD = SETNAM(ubyte namelen @ A, str filename @ XY)     ; set filename parameters
-romsub $FFC0 = OPEN() clobbers(X,Y) -> ubyte @Pc, ubyte @A      ; (via 794 ($31A)) open a logical file
-romsub $FFC3 = CLOSE(ubyte logical @ A) clobbers(A,X,Y)         ; (via 796 ($31C)) close a logical file
-romsub $FFC6 = CHKIN(ubyte logical @ X) clobbers(A,X) -> ubyte @Pc    ; (via 798 ($31E)) define an input channel
-romsub $FFC9 = CHKOUT(ubyte logical @ X) clobbers(A,X)          ; (via 800 ($320)) define an output channel
-romsub $FFCC = CLRCHN() clobbers(A,X)                           ; (via 802 ($322)) restore default devices
-romsub $FFCF = CHRIN() clobbers(X, Y) -> ubyte @ A   ; (via 804 ($324)) input a character (for keyboard, read a whole line from the screen) A=byte read.
-romsub $FFD2 = CHROUT(ubyte char @ A)                           ; (via 806 ($326)) output a character
-romsub $FFD5 = LOAD(ubyte verify @ A, uword address @ XY) -> ubyte @Pc, ubyte @ A, uword @ XY     ; (via 816 ($330)) load from device
-romsub $FFD8 = SAVE(ubyte zp_startaddr @ A, uword endaddr @ XY) -> ubyte @ Pc, ubyte @ A          ; (via 818 ($332)) save to a device
-romsub $FFDB = SETTIM(ubyte low @ A, ubyte middle @ X, ubyte high @ Y)      ; set the software clock
-romsub $FFDE = RDTIM() -> ubyte @ A, ubyte @ X, ubyte @ Y       ; read the software clock (A=lo,X=mid,Y=high)
-romsub $FFE1 = STOP() clobbers(X) -> ubyte @ Pz, ubyte @ A      ; (via 808 ($328)) check the STOP key (and some others in A)
-romsub $FFE4 = GETIN() clobbers(X,Y) -> ubyte @Pc, ubyte @ A    ; (via 810 ($32A)) get a character
-romsub $FFE7 = CLALL() clobbers(A,X)                            ; (via 812 ($32C)) close all files
-romsub $FFEA = UDTIM() clobbers(A,X)                            ; update the software clock
-romsub $FFED = SCREEN() -> ubyte @ X, ubyte @ Y                 ; read number of screen rows and columns
-romsub $FFF0 = PLOT(ubyte col @ Y, ubyte row @ X, ubyte dir @ Pc) -> ubyte @ X, ubyte @ Y       ; read/set position of cursor on screen.  Use txt.plot for a 'safe' wrapper that preserves X.
-romsub $FFF3 = IOBASE() -> uword @ XY                           ; read base address of I/O devices
+    sub get_char_matrix_ptr() -> uword {
+        ; Usually the character screen matrix is at 1024-2039 (see above)
+        ; However the vic memory configuration can be altered and this moves these registers with it.
+        ; So this routine determines it dynamically from the VIC memory setup.
+        uword chars_matrix_offset = (c64.VMCSB & $f0) as uword << 6
+        return get_vic_memory_base() + chars_matrix_offset
+    }
 
-; ---- end of C64 ROM kernal routines ----
+    sub get_bitmap_ptr() -> uword {
+        return get_vic_memory_base() + ((c64.VMCSB & %00001000) as uword << 10)
+    }
 
-; ---- utilities -----
+    sub get_sprite_addr_ptrs() -> uword {
+        ; Usually the sprite address pointers are at addresses 2040-2047 (see above)
+        ; However the vic memory configuration can be altered and this moves these registers with it.
+        ; So this routine determines it dynamically from the VIC memory setup.
+        return get_char_matrix_ptr() + 1016
+    }
 
-asmsub STOP2() -> ubyte @A  {
-    ; -- check if STOP key was pressed, returns true if so.  More convenient to use than STOP() because that only sets the carry status flag.
-    %asm {{
-        txa
-        pha
-        jsr  c64.STOP
-        beq  +
-        pla
-        tax
-        lda  #0
-        rts
-+       pla
-        tax
-        lda  #1
-        rts
-    }}
+    sub set_sprite_ptr(ubyte sprite_num, uword sprite_data_address) {
+        ; Sets the sprite data pointer to the given address.
+        ; Because it takes some time to calculate things based on the vic memory setup,
+        ; its only suitable if you're not continuously changing the data address.
+        ; Otherwise store the correct sprite data pointer location somewhere yourself and reuse it.
+        @(get_sprite_addr_ptrs() + sprite_num) = lsb(sprite_data_address / 64)
+    }
 }
 
-asmsub RDTIM16() -> uword @AY {
-    ; --  like RDTIM() but only returning the lower 16 bits in AY for convenience
-    %asm {{
-        stx  P8ZP_SCRATCH_REG
-        jsr  c64.RDTIM
-        pha
-        txa
-        tay
-        pla
-        ldx  P8ZP_SCRATCH_REG
-        rts
-    }}
-}
+sys {
+    ; ------- lowlevel system routines --------
 
+    %option no_symbol_prefixing
 
+    const ubyte target = 64         ;  compilation target specifier.  64 = C64, 128 = C128, 16 = CommanderX16.
 
-; ---- C64 specific system utility routines: ----
 
 asmsub  init_system()  {
     ; Initializes the machine to a sane starting state.
@@ -292,7 +335,7 @@ asmsub  init_system()  {
     ; This means that the BASIC, KERNAL and CHARGEN ROMs are banked in,
     ; the VIC, SID and CIA chips are reset, screen is cleared, and the default IRQ is set.
     ; Also a different color scheme is chosen to identify ourselves a little.
-    ; Uppercase charset is activated, and all three registers set to 0, status flags cleared.
+    ; Uppercase charset is activated.
     %asm {{
         sei
         cld
@@ -300,13 +343,13 @@ asmsub  init_system()  {
         sta  $00
         lda  #%00100111
         sta  $01
-        jsr  c64.IOINIT
-        jsr  c64.RESTOR
-        jsr  c64.CINT
+        jsr  cbm.IOINIT
+        jsr  cbm.RESTOR
+        jsr  cbm.CINT
         lda  #6
         sta  c64.EXTCOL
         lda  #7
-        sta  c64.COLOR
+        sta  cbm.COLOR
         lda  #0
         sta  c64.BGCOL0
         jsr  disable_runstop_and_charsetswitch
@@ -326,7 +369,7 @@ asmsub  init_system_phase2()  {
 asmsub  cleanup_at_exit() {
     ; executed when the main subroutine does rts
     %asm {{
-        jmp  c64.enable_runstop_and_charsetswitch
+        jmp  sys.enable_runstop_and_charsetswitch
     }}
 }
 
@@ -350,18 +393,18 @@ asmsub  enable_runstop_and_charsetswitch() clobbers(A) {
     }}
 }
 
-asmsub  set_irq(uword handler @AY, ubyte useKernal @Pc) clobbers(A)  {
+asmsub  set_irq(uword handler @AY, bool useKernal @Pc) clobbers(A)  {
 	%asm {{
 	        sta  _modified+1
 	        sty  _modified+2
 	        lda  #0
-	        adc  #0
+	        rol  a
 	        sta  _use_kernal
 		sei
 		lda  #<_irq_handler
-		sta  c64.CINV
+		sta  cbm.CINV
 		lda  #>_irq_handler
-		sta  c64.CINV+1
+		sta  cbm.CINV+1
 		cli
 		rts
 _irq_handler    jsr  _irq_handler_init
@@ -379,13 +422,12 @@ _modified	jsr  $ffff                      ; modified
 		tax
 		pla
 		rti
-+		jmp  c64.IRQDFRT		; continue with normal kernal irq routine
++		jmp  cbm.IRQDFRT		; continue with normal kernal irq routine
 
 _use_kernal     .byte  0
 
 _irq_handler_init
-		; save all zp scratch registers and the X register as these might be clobbered by the irq routine
-		stx  IRQ_X_REG
+		; save all zp scratch registers as these might be clobbered by the irq routine
 		lda  P8ZP_SCRATCH_B1
 		sta  IRQ_SCRATCH_ZPB1
 		lda  P8ZP_SCRATCH_REG
@@ -398,18 +440,11 @@ _irq_handler_init
 		sta  IRQ_SCRATCH_ZPWORD2
 		lda  P8ZP_SCRATCH_W2+1
 		sta  IRQ_SCRATCH_ZPWORD2+1
-		; stack protector; make sure we don't clobber the top of the evaluation stack
-		dex
-		dex
-		dex
-		dex
-		dex
-		dex
 		cld
 		rts
 
 _irq_handler_end
-		; restore all zp scratch registers and the X register
+		; restore all zp scratch registers
 		lda  IRQ_SCRATCH_ZPB1
 		sta  P8ZP_SCRATCH_B1
 		lda  IRQ_SCRATCH_ZPREG
@@ -422,10 +457,8 @@ _irq_handler_end
 		sta  P8ZP_SCRATCH_W2
 		lda  IRQ_SCRATCH_ZPWORD2+1
 		sta  P8ZP_SCRATCH_W2+1
-		ldx  IRQ_X_REG
 		rts
 
-IRQ_X_REG		.byte  0
 IRQ_SCRATCH_ZPB1	.byte  0
 IRQ_SCRATCH_ZPREG	.byte  0
 IRQ_SCRATCH_ZPWORD1	.word  0
@@ -437,10 +470,10 @@ IRQ_SCRATCH_ZPWORD2	.word  0
 asmsub  restore_irq() clobbers(A) {
 	%asm {{
 		sei
-		lda  #<c64.IRQDFRT
-		sta  c64.CINV
-		lda  #>c64.IRQDFRT
-		sta  c64.CINV+1
+		lda  #<cbm.IRQDFRT
+		sta  cbm.CINV
+		lda  #>cbm.IRQDFRT
+		sta  cbm.CINV+1
 		lda  #0
 		sta  c64.IREQMASK	; disable raster irq
 		lda  #%10000001
@@ -450,21 +483,21 @@ asmsub  restore_irq() clobbers(A) {
 	}}
 }
 
-asmsub  set_rasterirq(uword handler @AY, uword rasterpos @R0, ubyte useKernal @Pc) clobbers(A) {
+asmsub  set_rasterirq(uword handler @AY, uword rasterpos @R0, bool useKernal @Pc) clobbers(A) {
 	%asm {{
 	        sta  _modified+1
 	        sty  _modified+2
 	        lda  #0
-	        adc  #0
+	        rol  a
 	        sta  set_irq._use_kernal
 		lda  cx16.r0
 		ldy  cx16.r0+1
 		sei
 		jsr  _setup_raster_irq
 		lda  #<_raster_irq_handler
-		sta  c64.CINV
+		sta  cbm.CINV
 		lda  #>_raster_irq_handler
-		sta  c64.CINV+1
+		sta  cbm.CINV+1
 		cli
 		rts
 
@@ -472,8 +505,8 @@ _raster_irq_handler
 		jsr  set_irq._irq_handler_init
 _modified	jsr  $ffff              ; modified
 		jsr  set_irq._irq_handler_end
-                lda  #$ff
-                sta  c64.VICIRQ			; acknowledge raster irq
+        lda  #$ff
+        sta  c64.VICIRQ			; acknowledge raster irq
 		lda  set_irq._use_kernal
 		bne  +
 		; end irq processing - don't use kernal's irq handling
@@ -483,7 +516,7 @@ _modified	jsr  $ffff              ; modified
 		tax
 		pla
 		rti
-+		jmp  c64.IRQDFRT                ; continue with kernal irq routine
++		jmp  cbm.IRQDFRT                ; continue with kernal irq routine
 
 _setup_raster_irq
 		pha
@@ -507,23 +540,14 @@ _setup_raster_irq
 	}}
 }
 
-; ---- end of C64 specific system utility routines ----
 
-}
-
-sys {
-    ; ------- lowlevel system routines --------
-
-    const ubyte target = 64         ;  compilation target specifier.  64 = C64, 128 = C128, 16 = CommanderX16.
-
-
-    asmsub  reset_system()  {
+    asmsub reset_system()  {
         ; Soft-reset the system back to initial power-on Basic prompt.
         %asm {{
             sei
             lda  #14
             sta  $01        ; bank the kernal in
-            jmp  (c64.RESET_VEC)
+            jmp  (cbm.RESET_VEC)
         }}
     }
 
@@ -541,9 +565,9 @@ _loop       lda  P8ZP_SCRATCH_W1
             ldx  P8ZP_SCRATCH_B1
             rts
 
-+           lda  c64.TIME_LO
++           lda  cbm.TIME_LO
             sta  P8ZP_SCRATCH_B1
--           lda  c64.TIME_LO
+-           lda  cbm.TIME_LO
             cmp  P8ZP_SCRATCH_B1
             beq  -
 
@@ -700,13 +724,26 @@ _longcopy
         }}
     }
 
+    inline asmsub irqsafe_set_irqd() {
+        %asm {{
+        php
+        sei
+        }}
+    }
+
+    inline asmsub irqsafe_clear_irqd() {
+        %asm {{
+        plp
+        }}
+    }
+
     inline asmsub exit(ubyte returnvalue @A) {
         ; -- immediately exit the program with a return code in the A register
         %asm {{
             lda  #31
             sta  $01        ; bank the kernal in
-            jsr  c64.CLRCHN		; reset i/o channels
-            jsr  c64.enable_runstop_and_charsetswitch
+            jsr  cbm.CLRCHN		; reset i/o channels
+            jsr  sys.enable_runstop_and_charsetswitch
             ldx  prog8_lib.orig_stackpointer
             txs
             rts		; return to original caller
@@ -724,111 +761,139 @@ _longcopy
 
 cx16 {
 
+    %option no_symbol_prefixing
+
     ; the sixteen virtual 16-bit registers that the CX16 has defined in the zeropage
     ; they are simulated on the C64 as well but their location in memory is different
-    ; (because there's no room for them in the zeropage)
-    ; they are allocated at the bottom of the eval-stack (should be ample space unless
-    ; you're doing insane nesting of expressions...)
-    ; NOTE: the memory location of these registers can change based on the "-esa" compiler option
-    &uword r0  = $cf00
-    &uword r1  = $cf02
-    &uword r2  = $cf04
-    &uword r3  = $cf06
-    &uword r4  = $cf08
-    &uword r5  = $cf0a
-    &uword r6  = $cf0c
-    &uword r7  = $cf0e
-    &uword r8  = $cf10
-    &uword r9  = $cf12
-    &uword r10 = $cf14
-    &uword r11 = $cf16
-    &uword r12 = $cf18
-    &uword r13 = $cf1a
-    &uword r14 = $cf1c
-    &uword r15 = $cf1e
+    ; (because there's no room for them in the zeropage in the default configuration)
+    ; Note that when using ZP options that free up more of the zeropage (such as %zeropage kernalsafe)
+    ; there might be enough space to put them there after all, and the compiler will change these addresses!
+    &uword r0  = $cfe0
+    &uword r1  = $cfe2
+    &uword r2  = $cfe4
+    &uword r3  = $cfe6
+    &uword r4  = $cfe8
+    &uword r5  = $cfea
+    &uword r6  = $cfec
+    &uword r7  = $cfee
+    &uword r8  = $cff0
+    &uword r9  = $cff2
+    &uword r10 = $cff4
+    &uword r11 = $cff6
+    &uword r12 = $cff8
+    &uword r13 = $cffa
+    &uword r14 = $cffc
+    &uword r15 = $cffe
 
-    &word r0s  = $cf00
-    &word r1s  = $cf02
-    &word r2s  = $cf04
-    &word r3s  = $cf06
-    &word r4s  = $cf08
-    &word r5s  = $cf0a
-    &word r6s  = $cf0c
-    &word r7s  = $cf0e
-    &word r8s  = $cf10
-    &word r9s  = $cf12
-    &word r10s = $cf14
-    &word r11s = $cf16
-    &word r12s = $cf18
-    &word r13s = $cf1a
-    &word r14s = $cf1c
-    &word r15s = $cf1e
+    &word r0s  = $cfe0
+    &word r1s  = $cfe2
+    &word r2s  = $cfe4
+    &word r3s  = $cfe6
+    &word r4s  = $cfe8
+    &word r5s  = $cfea
+    &word r6s  = $cfec
+    &word r7s  = $cfee
+    &word r8s  = $cff0
+    &word r9s  = $cff2
+    &word r10s = $cff4
+    &word r11s = $cff6
+    &word r12s = $cff8
+    &word r13s = $cffa
+    &word r14s = $cffc
+    &word r15s = $cffe
 
-    &ubyte r0L  = $cf00
-    &ubyte r1L  = $cf02
-    &ubyte r2L  = $cf04
-    &ubyte r3L  = $cf06
-    &ubyte r4L  = $cf08
-    &ubyte r5L  = $cf0a
-    &ubyte r6L  = $cf0c
-    &ubyte r7L  = $cf0e
-    &ubyte r8L  = $cf10
-    &ubyte r9L  = $cf12
-    &ubyte r10L = $cf14
-    &ubyte r11L = $cf16
-    &ubyte r12L = $cf18
-    &ubyte r13L = $cf1a
-    &ubyte r14L = $cf1c
-    &ubyte r15L = $cf1e
+    &ubyte r0L  = $cfe0
+    &ubyte r1L  = $cfe2
+    &ubyte r2L  = $cfe4
+    &ubyte r3L  = $cfe6
+    &ubyte r4L  = $cfe8
+    &ubyte r5L  = $cfea
+    &ubyte r6L  = $cfec
+    &ubyte r7L  = $cfee
+    &ubyte r8L  = $cff0
+    &ubyte r9L  = $cff2
+    &ubyte r10L = $cff4
+    &ubyte r11L = $cff6
+    &ubyte r12L = $cff8
+    &ubyte r13L = $cffa
+    &ubyte r14L = $cffc
+    &ubyte r15L = $cffe
 
-    &ubyte r0H  = $cf01
-    &ubyte r1H  = $cf03
-    &ubyte r2H  = $cf05
-    &ubyte r3H  = $cf07
-    &ubyte r4H  = $cf09
-    &ubyte r5H  = $cf0b
-    &ubyte r6H  = $cf0d
-    &ubyte r7H  = $cf0f
-    &ubyte r8H  = $cf11
-    &ubyte r9H  = $cf13
-    &ubyte r10H = $cf15
-    &ubyte r11H = $cf17
-    &ubyte r12H = $cf19
-    &ubyte r13H = $cf1b
-    &ubyte r14H = $cf1d
-    &ubyte r15H = $cf1f
+    &ubyte r0H  = $cfe1
+    &ubyte r1H  = $cfe3
+    &ubyte r2H  = $cfe5
+    &ubyte r3H  = $cfe7
+    &ubyte r4H  = $cfe9
+    &ubyte r5H  = $cfeb
+    &ubyte r6H  = $cfed
+    &ubyte r7H  = $cfef
+    &ubyte r8H  = $cff1
+    &ubyte r9H  = $cff3
+    &ubyte r10H = $cff5
+    &ubyte r11H = $cff7
+    &ubyte r12H = $cff9
+    &ubyte r13H = $cffb
+    &ubyte r14H = $cffd
+    &ubyte r15H = $cfff
 
-    &byte r0sL  = $cf00
-    &byte r1sL  = $cf02
-    &byte r2sL  = $cf04
-    &byte r3sL  = $cf06
-    &byte r4sL  = $cf08
-    &byte r5sL  = $cf0a
-    &byte r6sL  = $cf0c
-    &byte r7sL  = $cf0e
-    &byte r8sL  = $cf10
-    &byte r9sL  = $cf12
-    &byte r10sL = $cf14
-    &byte r11sL = $cf16
-    &byte r12sL = $cf18
-    &byte r13sL = $cf1a
-    &byte r14sL = $cf1c
-    &byte r15sL = $cf1e
+    &byte r0sL  = $cfe0
+    &byte r1sL  = $cfe2
+    &byte r2sL  = $cfe4
+    &byte r3sL  = $cfe6
+    &byte r4sL  = $cfe8
+    &byte r5sL  = $cfea
+    &byte r6sL  = $cfec
+    &byte r7sL  = $cfee
+    &byte r8sL  = $cff0
+    &byte r9sL  = $cff2
+    &byte r10sL = $cff4
+    &byte r11sL = $cff6
+    &byte r12sL = $cff8
+    &byte r13sL = $cffa
+    &byte r14sL = $cffc
+    &byte r15sL = $cffe
+
+    &byte r0sH  = $cfe1
+    &byte r1sH  = $cfe3
+    &byte r2sH  = $cfe5
+    &byte r3sH  = $cfe7
+    &byte r4sH  = $cfe9
+    &byte r5sH  = $cfeb
+    &byte r6sH  = $cfed
+    &byte r7sH  = $cfef
+    &byte r8sH  = $cff1
+    &byte r9sH  = $cff3
+    &byte r10sH = $cff5
+    &byte r11sH = $cff7
+    &byte r12sH = $cff9
+    &byte r13sH = $cffb
+    &byte r14sH = $cffd
+    &byte r15sH = $cfff
+
+    asmsub save_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  cx16.r0,y
+            sta  _cx16_vreg_storage,y
+            dey
+            bpl  -
+            rts
+
+    _cx16_vreg_storage
+            .word 0,0,0,0,0,0,0,0
+            .word 0,0,0,0,0,0,0,0
+        }}
+    }
+
+    asmsub restore_virtual_registers() clobbers(A,Y) {
+        %asm {{
+            ldy  #31
+    -       lda  save_virtual_registers._cx16_vreg_storage,y
+            sta  cx16.r0,y
+            dey
+            bpl  -
+            rts
+        }}
+    }
 
-    &byte r0sH  = $cf01
-    &byte r1sH  = $cf03
-    &byte r2sH  = $cf05
-    &byte r3sH  = $cf07
-    &byte r4sH  = $cf09
-    &byte r5sH  = $cf0b
-    &byte r6sH  = $cf0d
-    &byte r7sH  = $cf0f
-    &byte r8sH  = $cf11
-    &byte r9sH  = $cf13
-    &byte r10sH = $cf15
-    &byte r11sH = $cf17
-    &byte r12sH = $cf19
-    &byte r13sH = $cf1b
-    &byte r14sH = $cf1d
-    &byte r15sH = $cf1f
 }

compiler/res/prog8lib/c64/textio.p8

@@ -6,6 +6,8 @@
 
 txt {
 
+    %option no_symbol_prefixing
+
 const ubyte DEFAULT_WIDTH = 40
 const ubyte DEFAULT_HEIGHT = 25
 
@@ -30,14 +32,14 @@ asmsub column(ubyte col @A) clobbers(A, X, Y) {
     ; ---- set the cursor on the given column (starting with 0) on the current line
     %asm {{
         sec
-        jsr  c64.PLOT
+        jsr  cbm.PLOT
         tay
         clc
-        jmp  c64.PLOT
+        jmp  cbm.PLOT
     }}
 }
 
-asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
+asmsub  fill_screen (ubyte character @ A, ubyte color @ Y) clobbers(A)  {
 	; ---- fill the character screen with the given fill character and character color.
 	;      (assumes screen and color matrix are at their default addresses)
 
@@ -52,15 +54,15 @@ asmsub  fill_screen (ubyte char @ A, ubyte color @ Y) clobbers(A)  {
 
 }
 
-asmsub  clear_screenchars (ubyte char @ A) clobbers(Y)  {
+asmsub  clear_screenchars (ubyte character @ A) clobbers(Y)  {
 	; ---- clear the character screen with the given fill character (leaves colors)
 	;      (assumes screen matrix is at the default address)
 	%asm {{
 		ldy  #250
--		sta  c64.Screen+250*0-1,y
-		sta  c64.Screen+250*1-1,y
-		sta  c64.Screen+250*2-1,y
-		sta  c64.Screen+250*3-1,y
+-		sta  cbm.Screen+250*0-1,y
+		sta  cbm.Screen+250*1-1,y
+		sta  cbm.Screen+250*2-1,y
+		sta  cbm.Screen+250*3-1,y
 		dey
 		bne  -
 		rts
@@ -72,10 +74,10 @@ asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
 	;      (assumes color matrix is at the default address)
 	%asm {{
 		ldy  #250
--		sta  c64.Colors+250*0-1,y
-		sta  c64.Colors+250*1-1,y
-		sta  c64.Colors+250*2-1,y
-		sta  c64.Colors+250*3-1,y
+-		sta  cbm.Colors+250*0-1,y
+		sta  cbm.Colors+250*1-1,y
+		sta  cbm.Colors+250*2-1,y
+		sta  cbm.Colors+250*3-1,y
 		dey
 		bne  -
 		rts
@@ -83,7 +85,7 @@ asmsub  clear_screencolors (ubyte color @ A) clobbers(Y)  {
 }
 
 sub color (ubyte txtcol) {
-    c64.COLOR = txtcol
+    cbm.COLOR = txtcol
 }
 
 sub lowercase() {
@@ -94,13 +96,12 @@ sub uppercase() {
     c64.VMCSB &= ~2
 }
 
-asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
+asmsub  scroll_left  (bool alsocolors @ Pc) clobbers(A, X, Y)  {
 	; ---- scroll the whole screen 1 character to the left
 	;      contents of the rightmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc _scroll_screen
 
 +               ; scroll the screen and the color memory
@@ -108,10 +109,10 @@ asmsub  scroll_left  (ubyte alsocolors @ Pc) clobbers(A, Y)  {
 		ldy  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 1,x
-            sta  c64.Screen + 40*row + 0,x
-            lda  c64.Colors + 40*row + 1,x
-            sta  c64.Colors + 40*row + 0,x
+            lda  cbm.Screen + 40*row + 1,x
+            sta  cbm.Screen + 40*row + 0,x
+            lda  cbm.Colors + 40*row + 1,x
+            sta  cbm.Colors + 40*row + 0,x
         .next
 		inx
 		dey
@@ -123,34 +124,32 @@ _scroll_screen  ; scroll only the screen memory
 		ldy  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 1,x
-            sta  c64.Screen + 40*row + 0,x
+            lda  cbm.Screen + 40*row + 1,x
+            sta  cbm.Screen + 40*row + 0,x
         .next
 		inx
 		dey
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_right  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_right  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character to the right
 	;      contents of the leftmost column are unchanged, you should clear/refill this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 0,x
-            sta  c64.Screen + 40*row + 1,x
-            lda  c64.Colors + 40*row + 0,x
-            sta  c64.Colors + 40*row + 1,x
+            lda  cbm.Screen + 40*row + 0,x
+            sta  cbm.Screen + 40*row + 1,x
+            lda  cbm.Colors + 40*row + 0,x
+            sta  cbm.Colors + 40*row + 1,x
         .next
 		dex
 		bpl  -
@@ -160,33 +159,31 @@ _scroll_screen  ; scroll only the screen memory
 		ldx  #38
 -
         .for row=0, row<=24, row+=1
-            lda  c64.Screen + 40*row + 0,x
-            sta  c64.Screen + 40*row + 1,x
+            lda  cbm.Screen + 40*row + 0,x
+            sta  cbm.Screen + 40*row + 1,x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_up  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_up  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character up
 	;      contents of the bottom row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx #39
 -
         .for row=1, row<=24, row+=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row-1),x
-            lda  c64.Colors + 40*row,x
-            sta  c64.Colors + 40*(row-1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row-1),x
+            lda  cbm.Colors + 40*row,x
+            sta  cbm.Colors + 40*(row-1),x
         .next
 		dex
 		bpl  -
@@ -196,33 +193,31 @@ _scroll_screen  ; scroll only the screen memory
 		ldx #39
 -
         .for row=1, row<=24, row+=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row-1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row-1),x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  scroll_down  (ubyte alsocolors @ Pc) clobbers(A)  {
+asmsub  scroll_down  (bool alsocolors @ Pc) clobbers(A,X)  {
 	; ---- scroll the whole screen 1 character down
 	;      contents of the top row are unchanged, you should refill/clear this yourself
 	;      Carry flag determines if screen color data must be scrolled too
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  _scroll_screen
 
 +               ; scroll the screen and the color memory
 		ldx #39
 -
         .for row=23, row>=0, row-=1
-            lda  c64.Colors + 40*row,x
-            sta  c64.Colors + 40*(row+1),x
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row+1),x
+            lda  cbm.Colors + 40*row,x
+            sta  cbm.Colors + 40*(row+1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row+1),x
         .next
 		dex
 		bpl  -
@@ -232,134 +227,121 @@ _scroll_screen  ; scroll only the screen memory
 		ldx #39
 -
         .for row=23, row>=0, row-=1
-            lda  c64.Screen + 40*row,x
-            sta  c64.Screen + 40*(row+1),x
+            lda  cbm.Screen + 40*row,x
+            sta  cbm.Screen + 40*(row+1),x
         .next
 		dex
 		bpl  -
 
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-romsub $FFD2 = chrout(ubyte char @ A)    ; for consistency. You can also use c64.CHROUT directly ofcourse.
+romsub $FFD2 = chrout(ubyte character @ A)    ; for consistency. You can also use cbm.CHROUT directly ofcourse.
 
 asmsub  print (str text @ AY) clobbers(A,Y)  {
 	; ---- print null terminated string from A/Y
 	; note: the compiler contains an optimization that will replace
 	;       a call to this subroutine with a string argument of just one char,
-	;       by just one call to c64.CHROUT of that single char.
+	;       by just one call to cbm.CHROUT of that single char.
 	%asm {{
 		sta  P8ZP_SCRATCH_B1
 		sty  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  (P8ZP_SCRATCH_B1),y
 		beq  +
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		bne  -
 +		rts
 	}}
 }
 
-asmsub  print_ub0  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub0  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, with left padding 0s (3 positions total)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 		pha
 		tya
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_ub  (ubyte value @ A) clobbers(A,Y)  {
+asmsub  print_ub  (ubyte value @ A) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		jsr  conv.ubyte2decimal
 _print_byte_digits
 		pha
 		cpy  #'0'
 		beq  +
 		tya
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		jmp  _ones
 +       pla
         cmp  #'0'
         beq  _ones
-        jsr  c64.CHROUT
+        jsr  cbm.CHROUT
 _ones   txa
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_b  (byte value @ A) clobbers(A,Y)  {
+asmsub  print_b  (byte value @ A) clobbers(A,X,Y)  {
 	; ---- print the byte in A in decimal form, without left padding 0s
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		pha
 		cmp  #0
 		bpl  +
 		lda  #'-'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 +		pla
 		jsr  conv.byte2decimal
 		jmp  print_ub._print_byte_digits
 	}}
 }
 
-asmsub  print_ubhex  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubhex  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in hex form (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		bcc  +
 		pha
 		lda  #'$'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		pla
 +		jsr  conv.ubyte2hex
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		tya
-		jsr  c64.CHROUT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_ubbin  (ubyte value @ A, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_ubbin  (ubyte value @ A, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the ubyte in A in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
-		stx  P8ZP_SCRATCH_REG
 		sta  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'%'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 +		ldy  #8
 -		lda  #'0'
 		asl  P8ZP_SCRATCH_B1
 		bcc  +
 		lda  #'1'
-+		jsr  c64.CHROUT
++		jsr  cbm.CHROUT
 		dey
 		bne  -
-		ldx  P8ZP_SCRATCH_REG
 		rts
 	}}
 }
 
-asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwbin  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in binary form (if Carry is set, a radix prefix '%' is printed as well)
 	%asm {{
 		pha
@@ -371,7 +353,7 @@ asmsub  print_uwbin  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
+asmsub  print_uwhex  (uword value @ AY, bool prefix @ Pc) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in hexadecimal form (4 digits)
 	;      (if Carry is set, a radix prefix '$' is printed as well)
 	%asm {{
@@ -384,28 +366,24 @@ asmsub  print_uwhex  (uword value @ AY, ubyte prefix @ Pc) clobbers(A,Y)  {
 	}}
 }
 
-asmsub  print_uw0  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw0  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, with left padding 0s (5 positions total)
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
         beq  +
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		bne  -
-+		ldx  P8ZP_SCRATCH_REG
-		rts
++		rts
 	}}
 }
 
-asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
+asmsub  print_uw  (uword value @ AY) clobbers(A,X,Y)  {
 	; ---- print the uword in A/Y in decimal form, without left padding 0s
 	%asm {{
-	    stx  P8ZP_SCRATCH_REG
 		jsr  conv.uword2decimal
-		ldx  P8ZP_SCRATCH_REG
 		ldy  #0
 -		lda  conv.uword2decimal.decTenThousands,y
 		beq  _allzero
@@ -415,25 +393,25 @@ asmsub  print_uw  (uword value @ AY) clobbers(A,Y)  {
 		bne  -
 
 _gotdigit
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		iny
 		lda  conv.uword2decimal.decTenThousands,y
 		bne  _gotdigit
 		rts
 _allzero
         lda  #'0'
-        jmp  c64.CHROUT
+        jmp  cbm.CHROUT
 	}}
 }
 
-asmsub  print_w  (word value @ AY) clobbers(A,Y)  {
+asmsub  print_w  (word value @ AY) clobbers(A,X,Y)  {
 	; ---- print the (signed) word in A/Y in decimal form, without left padding 0's
 	%asm {{
 		cpy  #0
 		bpl  +
 		pha
 		lda  #'-'
-		jsr  c64.CHROUT
+		jsr  cbm.CHROUT
 		tya
 		eor  #255
 		tay
@@ -455,7 +433,7 @@ asmsub  input_chars  (uword buffer @ AY) clobbers(A) -> ubyte @ Y  {
 		sta  P8ZP_SCRATCH_W1
 		sty  P8ZP_SCRATCH_W1+1
 		ldy  #0				; char counter = 0
--		jsr  c64.CHRIN
+-		jsr  cbm.CHRIN
 		cmp  #$0d			; return (ascii 13) pressed?
 		beq  +				; yes, end.
 		sta  (P8ZP_SCRATCH_W1),y	; else store char in buffer
@@ -487,7 +465,7 @@ asmsub  setchr  (ubyte col @X, ubyte row @Y, ubyte character @A) clobbers(A, Y)
 _mod		sta  $ffff		; modified
 		rts
 
-_screenrows	.word  $0400 + range(0, 1000, 40)
+_screenrows	.word  cbm.Screen + range(0, 1000, 40)
 	}}
 }
 
@@ -554,7 +532,7 @@ _mod		lda  $ffff		; modified
 	}}
 }
 
-sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
+sub  setcc  (ubyte col, ubyte row, ubyte character, ubyte charcolor)  {
 	; ---- set char+color at the given position on the screen
 	%asm {{
 		lda  row
@@ -566,13 +544,13 @@ sub  setcc  (ubyte column, ubyte row, ubyte char, ubyte charcolor)  {
 		sta  _colormod+2
 		lda  setchr._screenrows,y
 		clc
-		adc  column
+		adc  col
 		sta  _charmod+1
 		sta  _colormod+1
 		bcc  +
 		inc  _charmod+2
 		inc  _colormod+2
-+		lda  char
++		lda  character
 _charmod	sta  $ffff		; modified
 		lda  charcolor
 _colormod	sta  $ffff		; modified
@@ -580,22 +558,17 @@ _colormod	sta  $ffff		; modified
 	}}
 }
 
-asmsub  plot  (ubyte col @ Y, ubyte row @ A) clobbers(A) {
-	; ---- safe wrapper around PLOT kernal routine, to save the X register.
+asmsub  plot  (ubyte col @ Y, ubyte row @ X) {
 	%asm  {{
-		stx  P8ZP_SCRATCH_REG
-		tax
 		clc
-		jsr  c64.PLOT
-		ldx  P8ZP_SCRATCH_REG
-		rts
+		jmp  cbm.PLOT
 	}}
 }
 
 asmsub width() clobbers(X,Y) -> ubyte @A {
     ; -- returns the text screen width (number of columns)
     %asm {{
-        jsr  c64.SCREEN
+        jsr  cbm.SCREEN
         txa
         rts
     }}
@@ -604,7 +577,7 @@ asmsub width() clobbers(X,Y) -> ubyte @A {
 asmsub height() clobbers(X, Y) -> ubyte @A {
     ; -- returns the text screen height (number of rows)
     %asm {{
-        jsr  c64.SCREEN
+        jsr  cbm.SCREEN
         tya
         rts
     }}