IR: optimize load+add+load into single loadm with offset (also store)

update gradle wrapper to 9.2.0

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

@@ -55,6 +55,8 @@ class IRPeepholeOptimizer(private val irprog: IRProgram) {
                                 || cleanupPushPop(chunk1, indexedInstructions)
                                 || simplifyConstantReturns(chunk1, indexedInstructions)
                                 || removeNeedlessLoads(chunk1, indexedInstructions)
+                                || useArrayIndexingInsteadOfAdds(chunk1, indexedInstructions)
+                                || removeNops(chunk1, indexedInstructions)   // last time, in case one of the optimizers replaced something with a nop
                     } while (changed)
                 }
             }
@@ -553,4 +555,40 @@ jump p8_label_gen_2
         }
         return changed
     }
+
+    private fun useArrayIndexingInsteadOfAdds(chunk: IRCodeChunk, indexedInstructions: List<IndexedValue<IRInstruction>>): Boolean {
+        var changed = false
+        indexedInstructions.reversed().forEach { (idx, ins) ->
+            if (ins.opcode == Opcode.ADD && ins.immediate!=null && idx>0) {
+                val load = indexedInstructions[idx-1].value
+                if((load.opcode==Opcode.LOAD) && load.labelSymbol!=null) {
+                    val lastInstruction = indexedInstructions[idx+1].value
+                    if(lastInstruction.opcode==Opcode.LOADI) {
+                        val targetRegister = lastInstruction.reg1!!
+                        if(ins.reg1==lastInstruction.reg2!! && load.reg1==lastInstruction.reg2!!) {
+                            val loadm = IRInstruction(Opcode.LOADM, lastInstruction.type, reg1=targetRegister, labelSymbol = load.labelSymbol, symbolOffset = ins.immediate)
+                            chunk.instructions[idx-1] = loadm
+                            chunk.instructions.removeAt(idx+1)
+                            chunk.instructions.removeAt(idx)
+                            changed = true
+                        }
+                    } else if(lastInstruction.opcode==Opcode.STOREI) {
+                        val targetRegister = lastInstruction.reg1!!
+                        if(ins.reg1==lastInstruction.reg2!! && load.reg1==lastInstruction.reg2!!) {
+                            val valueLoad = indexedInstructions[idx-2].value
+                            if(valueLoad.opcode==Opcode.LOAD && valueLoad.reg1==targetRegister) {
+                                val storem = IRInstruction(Opcode.STOREM, lastInstruction.type, reg1=valueLoad.reg1, labelSymbol = load.labelSymbol, symbolOffset = ins.immediate)
+                                chunk.instructions[idx-1] = storem
+                                chunk.instructions.removeAt(idx+1)
+                                chunk.instructions.removeAt(idx)
+                                changed = true
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        return changed
+    }
 }

docs/source/_static/symboldumps/skeletons-c128.txt

@@ -1,6 +1,5 @@
 
 Prog8 compiler v12.0.1 by Irmen de Jong (irmen@razorvine.net)
-Prerelease version from git commit cfa4355a in branch master
 This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/licenses/gpl.html
 
 Compiling program import-all-c128.p8

docs/source/_static/symboldumps/skeletons-c64.txt

@@ -1,6 +1,5 @@
 
 Prog8 compiler v12.0.1 by Irmen de Jong (irmen@razorvine.net)
-Prerelease version from git commit cfa4355a in branch master
 This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/licenses/gpl.html
 
 Compiling program import-all-c64.p8

docs/source/_static/symboldumps/skeletons-cx16.txt

@@ -1,6 +1,5 @@
 
 Prog8 compiler v12.0.1 by Irmen de Jong (irmen@razorvine.net)
-Prerelease version from git commit cfa4355a in branch master
 This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/licenses/gpl.html
 
 Compiling program import-all-cx16.p8

docs/source/_static/symboldumps/skeletons-pet32.txt

@@ -1,6 +1,5 @@
 
 Prog8 compiler v12.0.1 by Irmen de Jong (irmen@razorvine.net)
-Prerelease version from git commit cfa4355a in branch master
 This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/licenses/gpl.html
 
 Compiling program import-all-pet32.p8

docs/source/_static/symboldumps/skeletons-virtual.txt

@@ -1,6 +1,5 @@
 
 Prog8 compiler v12.0.1 by Irmen de Jong (irmen@razorvine.net)
-Prerelease version from git commit cfa4355a in branch master
 This software is licensed under the GNU GPL 3.0, see https://www.gnu.org/licenses/gpl.html
 
 Compiling program import-all-virtual.p8

docs/source/todo.rst

@@ -61,9 +61,8 @@ Future Things and Ideas
 
 IR/VM
 -----
-- optimize away the use of ADD to offset from a label. It's possible to do that with a label-offset in the instruction itself: loadm.b r7,main.start.array2+255   (it works for 255 but after that it starts using ADD!)
-- get rid of LOADX/STOREX/STOREZX, LOADINDEXED/STOREINDEXED just use add + loadi / storei?
 - extend the index range from 0-255 to 0-32767 in the LOADX, STOREX, STOREZX, LOADINDEXED, STOREINDEXED etc instructions (not compatible with 8 bit 6502, but the 68000 can use that)
+- get rid of LOADX/STOREX/STOREZX, LOADINDEXED/STOREINDEXED just use add + loadi / storei?
 - if float<0 / if word<0  uses sgn or load, but still use a bgt etc instruction after that with a #0 operand even though the sgn and load instructions sets the status bits already, so just use bstneg etc
 - add and sub instructions should modify the status flags so an explicit compare to zero can be avoided for example: if cx16.r0sL + cx16.r1sL <= 0  now compiles into:  addr.b r10,r11 /  bgts.b r10,#0,label
 - getting it in shape for code generation: the IR file should be able to encode every detail about a prog8 program (the VM doesn't have to actually be able to run all of it though!)

examples/test.p8

@@ -26,8 +26,6 @@ main {
             %ir {{
 loadm.b r1007,main.start.array2+255
 storem.b r1007,$ff02
-loadm.b r1007,main.start.array2+256
-storem.b r1007,$ff04
 load.w r1009,main.start.array2
 add.w r1009,#$0100
 loadi.b r1008,r1009

gradle/wrapper/gradle-wrapper.properties

@@ -1,6 +1,6 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-9.1.0-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-9.2.0-bin.zip
 networkTimeout=10000
 validateDistributionUrl=true
 zipStoreBase=GRADLE_USER_HOME

intermediate/src/prog8/intermediate/IRProgram.kt

@@ -306,7 +306,7 @@ class IRProgram(val name: String,
         fun convert(asmChunk: IRInlineAsmChunk): IRCodeChunks {
             val chunks = mutableListOf<IRCodeChunkBase>()
             var chunk = IRCodeChunk(asmChunk.label, null)
-            asmChunk.assembly.lineSequence().forEach {
+            asmChunk.assembly.lineSequence().filter{it.isNotBlank()}.forEach {
                 val parsed = parseIRCodeLine(it.trim())
                 parsed.fold(
                     ifLeft = { instruction -> chunk += instruction },