From 533090a68e5307baeec77270a59a71d09115974a Mon Sep 17 00:00:00 2001
From: Irmen de Jong <irmen@razorvine.net>
Date: Sun, 26 Nov 2023 22:55:17 +0100
Subject: [PATCH] fix expression result register of square, callfar, string
 compare functions in certain situations

---
 .../codegen/cpu6502/BuiltinFunctionsAsmGen.kt | 26 +++++++++++-----
 docs/source/todo.rst                          |  9 ------
 examples/test.p8                              | 31 ++++++++++++-------
 3 files changed, 38 insertions(+), 28 deletions(-)

diff --git a/codeGenCpu6502/src/prog8/codegen/cpu6502/BuiltinFunctionsAsmGen.kt b/codeGenCpu6502/src/prog8/codegen/cpu6502/BuiltinFunctionsAsmGen.kt
index 97c48232f..0adfed014 100644
--- a/codeGenCpu6502/src/prog8/codegen/cpu6502/BuiltinFunctionsAsmGen.kt
+++ b/codeGenCpu6502/src/prog8/codegen/cpu6502/BuiltinFunctionsAsmGen.kt
@@ -81,26 +81,32 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
             "rsave" -> funcRsave()
             "rrestore" -> funcRrestore()
             "cmp" -> funcCmp(fcall)
-            "callfar" -> funcCallFar(fcall)
-            "prog8_lib_stringcompare" -> funcStringCompare(fcall)
-            "prog8_lib_square_byte" -> funcSquare(fcall, DataType.UBYTE)
-            "prog8_lib_square_word" -> funcSquare(fcall, DataType.UWORD)
+            "callfar" -> funcCallFar(fcall, resultRegister)
+            "prog8_lib_stringcompare" -> funcStringCompare(fcall, resultRegister)
+            "prog8_lib_square_byte" -> funcSquare(fcall, DataType.UBYTE, resultRegister)
+            "prog8_lib_square_word" -> funcSquare(fcall, DataType.UWORD, resultRegister)
             else -> throw AssemblyError("missing asmgen for builtin func ${fcall.name}")
         }
 
         return BuiltinFunctions.getValue(fcall.name).returnType
     }
 
-    private fun funcSquare(fcall: PtBuiltinFunctionCall, resultType: DataType) {
+    private fun funcSquare(fcall: PtBuiltinFunctionCall, resultType: DataType, resultRegister: RegisterOrPair?) {
         // square of word value is faster with dedicated routine, square of byte just use the regular multiplication routine.
         when (resultType) {
             DataType.UBYTE -> {
                 asmgen.assignExpressionToRegister(fcall.args[0], RegisterOrPair.A)
                 asmgen.out("  tay  |  jsr  math.multiply_bytes")
+                if(resultRegister!=null)  {
+                    assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister, false, fcall.position, null, asmgen), CpuRegister.A, false, false)
+                }
             }
             DataType.UWORD -> {
                 asmgen.assignExpressionToRegister(fcall.args[0], RegisterOrPair.AY)
                 asmgen.out("  jsr  math.square")
+                if(resultRegister!=null)  {
+                    assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister, false, fcall.position, null, asmgen), RegisterOrPair.AY)
+                }
             }
             else -> {
                 throw AssemblyError("optimized square only for integer types")
@@ -140,9 +146,12 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
             sty  $remainderVar+1""")
     }
 
-    private fun funcStringCompare(fcall: PtBuiltinFunctionCall) {
+    private fun funcStringCompare(fcall: PtBuiltinFunctionCall, resultRegister: RegisterOrPair?) {
         asmgen.assignWordOperandsToAYAndVar(fcall.args[0], fcall.args[1], "P8ZP_SCRATCH_W2")
         asmgen.out("  jsr  prog8_lib.strcmp_mem")
+        if(resultRegister!=null)  {
+            assignAsmGen.assignRegisterByte(AsmAssignTarget.fromRegisters(resultRegister, false, fcall.position, null, asmgen), CpuRegister.A, false, false)
+        }
     }
 
     private fun funcRsave() {
@@ -182,7 +191,7 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
                 plp""")
     }
 
-    private fun funcCallFar(fcall: PtBuiltinFunctionCall) {
+    private fun funcCallFar(fcall: PtBuiltinFunctionCall, resultRegister: RegisterOrPair?) {
         if(asmgen.options.compTarget.name != "cx16")
             throw AssemblyError("callfar only works on cx16 target at this time")
 
@@ -196,6 +205,9 @@ internal class BuiltinFunctionsAsmGen(private val program: PtProgram,
 +           .word  0
 +           .byte  0""")
         // note that by convention the values in A+Y registers are now the return value of the call.
+        if(resultRegister!=null)  {
+            assignAsmGen.assignRegisterpairWord(AsmAssignTarget.fromRegisters(resultRegister, false, fcall.position, null, asmgen), RegisterOrPair.AY)
+        }
     }
 
     private fun funcCmp(fcall: PtBuiltinFunctionCall) {
diff --git a/docs/source/todo.rst b/docs/source/todo.rst
index ae4eb5225..b79ffce3f 100644
--- a/docs/source/todo.rst
+++ b/docs/source/todo.rst
@@ -2,15 +2,6 @@
 TODO
 ====
 
-- fix wrong result of:
-        ubyte b = 4
-        float c
-        %asm {{
-            nop
-        }}
-        c += b*b
-  which works if c is a uword instead...
-
 - [on branch: shortcircuit] investigate McCarthy evaluation again? this may also reduce code size perhaps for things like if a>4 or a<2 ....
 
 ...
diff --git a/examples/test.p8 b/examples/test.p8
index dbba2d187..5732c61af 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -1,19 +1,26 @@
 %import textio
-%import string
+%import floats
 %zeropage basicsafe
 
 main {
     sub start() {
-        str[] names = [iso:"irmen", iso:"jurrian", iso:"houtzaagmolen 41", iso:"the Quick Brown Fox jumps Over the LAZY dog!"]
-
-        ; txt.iso()
-
-        uword name
-        for name in names {
-            txt.print_ub(string.hash(name))
-            txt.spc()
-            txt.print(name)
-            txt.nl()
-        }
+        ubyte b = 4
+        ubyte b2 = 4
+        uword w = 4
+        uword w2 = 4
+        float c
+        %asm {{
+            nop
+        }}
+        c += b*b
+        floats.print_f(c)
+        txt.nl()
+        c=0
+        %asm {{
+            nop
+        }}
+        c += w*w
+        floats.print_f(c)
+        txt.nl()
     }
 }