optimize asm: don't use temp var for some additions

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

@@ -43,6 +43,13 @@ internal fun optimizeAssembly(lines: MutableList<String>, machine: IMachineDefin
         numberOfOptimizations++
     }
 
+    mods = optimizeUnneededTempvarInAdd(linesByFour)
+    if(mods.isNotEmpty()) {
+        apply(mods, lines)
+        linesByFour = getLinesBy(lines, 4)
+        numberOfOptimizations++
+    }
+
     var linesByFourteen = getLinesBy(lines, 14)
     mods = optimizeSameAssignments(linesByFourteen, machine, symbolTable)
     if(mods.isNotEmpty()) {
@@ -509,3 +516,26 @@ private fun optimizeUselessPushPopStack(linesByFour: List<List<IndexedValue<Stri
     }
     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/assignment/AugmentableAssignmentAsmGen.kt

@@ -707,6 +707,7 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
     }
 
     private fun inplacemodificationByteVariableWithValue(name: String, dt: DataType, operator: String, value: PtExpression) {
+        // TODO optimize: don't use scratch var if possible
         val tmpVar = if(name!="P8ZP_SCRATCH_B1") "P8ZP_SCRATCH_B1" else "P8ZP_SCRATCH_REG"
         asmgen.assignExpressionToVariable(value, tmpVar, value.type)
         asmgen.out("  lda  $name")
@@ -1080,6 +1081,7 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
             }
             "-" -> {
                 asmgen.translateDirectMemReadExpressionToRegA(memread)
+                // TODO optimize: don't use scratch var if possible
                 val tmpByte = if(name!="P8ZP_SCRATCH_B1") "P8ZP_SCRATCH_B1" else "P8ZP_SCRATCH_REG"
                 asmgen.out("""
                     sta  $tmpByte
@@ -2025,8 +2027,9 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
                 // the other variable is a BYTE type so optimize for that
                 when (operator) {
                     "+" -> {
-                        asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_B1", valueDt)
+                        if(valueDt==DataType.UBYTE) {
-                        if(valueDt==DataType.UBYTE)
+                            // TODO optimize: don't use scratch var
+                            asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_B1", valueDt)
                             asmgen.out("""
                                 lda  $name
                                 clc
@@ -2035,7 +2038,9 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
                                 bcc  +
                                 inc  $name+1
 +""")
-                        else
+                        } else {
+                            // TODO optimize: don't use scratch var
+                            asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_B1", valueDt)
                             asmgen.out("""
                                 ldy  #0
                                 lda  P8ZP_SCRATCH_B1
@@ -2047,8 +2052,10 @@ internal class AugmentableAssignmentAsmGen(private val program: PtProgram,
                                 tya
                                 adc  $name+1
                                 sta  $name+1""")
+                        }
                     }
                     "-" -> {
+                        // TODO optimize: don't use scratch var
                         asmgen.assignExpressionToVariable(value, "P8ZP_SCRATCH_B1", valueDt)
                         if(valueDt==DataType.UBYTE)
                             asmgen.out("""

docs/source/todo.rst

@@ -1,5 +1,7 @@
 TODO
 ====
+
+- optimize: search for TODO optimize: don't use scratch var
 - prefix prog8 subroutines with p8s_ instead of p8_ to not let them clash with variables in the asm?
 - allow 'chained' array indexing for expressions:  value = ptrarray[0][0]
 - allow 'chained' array indexing for assign targets:  ptrarray[0][0] = 42   this is just evaluating the lhs as a uword pointer expression

examples/test.p8

@@ -4,42 +4,11 @@
 
 main {
     sub start() {
-        uword w1
+        ubyte index
-        uword w2
+        ubyte[] t_index = [1,2,3,4,5]
+        ubyte nibble = 0
 
-        ubyte b1
+        index -= t_index[4]
-        ubyte b2
+        index -= t_index[nibble]
-
-        b1 = 255
-        b2 = 1
-        txt.print_ub(b1&b2!=0)  ; true
-        txt.spc()
-        txt.print_ub(b1&b2==0)  ; false
-        txt.nl()
-
-        b1 = 255
-        b2 = 0
-        txt.print_ub(b1&b2!=0)  ; false
-        txt.spc()
-        txt.print_ub(b1&b2==0)  ; true
-        txt.nl()
-
-        w1 = $ff0f
-        w2 = $0101
-        txt.print_ub(w1&w2!=0)  ; true
-        txt.spc()
-        txt.print_ub(w1&w2==0)  ; false
-        txt.nl()
-
-        w1 = $ff0f
-        w2 = $00f0
-        txt.print_ub(w1&w2!=0)  ; false
-        txt.spc()
-        txt.print_ub(w1&w2==0)  ; true
-        txt.nl()
-
-
-;        if w1 & w2
-;            cx16.r0++
     }
 }