added some cpu stack related assembly-level optimizations

codeGeneration/src/prog8/compiler/target/cpu6502/codegen/AsmOptimizer.kt

@@ -7,10 +7,6 @@ import prog8.ast.statements.VarDecl
 import prog8.compilerinterface.IMachineDefinition
 
 
-// TODO optimize subsequent  pha/pla,  phx/plx,  phy/ply  pairs
-// TODO optimize pha/plx -> tax,  pha/ply -> tay,  etc.
-
-
 // note: see https://wiki.nesdev.org/w/index.php/6502_assembly_optimisations
 
 
@@ -357,6 +353,31 @@ private fun optimizeStoreLoadSame(linesByFour: List<List<IndexedValue<String>>>,
                     mods.add(Modification(lines[2].index, true, null))
             }
         }
+        else if(first=="pha" && second=="pla" ||
+            first=="phx" && second=="plx" ||
+            first=="phy" && second=="ply" ||
+            first=="php" && second=="plp") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, true, null))
+        } else if(first=="pha" && second=="plx") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  tax"))
+        } else if(first=="pha" && second=="ply") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  tay"))
+        } else if(first=="phx" && second=="pla") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  txa"))
+        } else if(first=="phx" && second=="ply") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  txy"))
+        } else if(first=="phy" && second=="pla") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  tya"))
+        } else if(first=="phy" && second=="plx") {
+            mods.add(Modification(lines[1].index, true, null))
+            mods.add(Modification(lines[2].index, false, "  tyx"))
+        }
     }
     return mods
 }

codeOptimizers/src/prog8/optimizer/StatementOptimizer.kt

@@ -143,21 +143,6 @@ class StatementOptimizer(private val program: Program,
         return noModifications
     }
 
-    // TODO WHAT TO DO WITH THIS:
-//    override fun before(functionCall: FunctionCall, parent: Node): Iterable<IAstModification> {
-//        // if the first instruction in the called subroutine is a return statement with constant value, replace with the constant value
-//        val subroutine = functionCall.target.targetSubroutine(program)
-//        if(subroutine!=null) {
-//            val first = subroutine.statements.asSequence().filterNot { it is VarDecl || it is Directive }.firstOrNull()
-//            if(first is Return && first.value!=null) {
-//                val constval = first.value?.constValue(program)
-//                if(constval!=null)
-//                    return listOf(IAstModification.ReplaceNode(functionCall, constval, parent))
-//            }
-//        }
-//        return noModifications
-//    }
-
     override fun after(ifStatement: IfStatement, parent: Node): Iterable<IAstModification> {
         // remove empty if statements
         if(ifStatement.truepart.isEmpty() && ifStatement.elsepart.isEmpty())

compiler/src/prog8/compiler/astprocessing/StatementReorderer.kt

@@ -412,9 +412,9 @@ internal class StatementReorderer(val program: Program,
             }
             return listOf(IAstModification.ReplaceNode(call, scope, parent))
         } else if(!options.compTarget.asmsubArgsHaveRegisterClobberRisk(call.args)) {
-            // no register clobber risk, let the asmgen assign values to the registers directly.
+            // No register clobber risk, let the asmgen assign values to the registers directly.
-            // this is more efficient than first evaluating them to the stack
+            // this is more efficient than first evaluating them to the stack.
-            // TODO but what about complex expressions?
+            // As complex expressions will be flagged as a clobber-risk, these will be simplified below.
             return noModifications
         } else {
             // clobber risk; evaluate the arguments on the CPU stack first (in reverse order)...

examples/test.p8

@@ -1,40 +1,19 @@
 %import textio
-%import string
 %import test_stack
 
 main {
 
-    ubyte[23]  savedata
-    ubyte[17] cargohold = 0
-    uword filenameptr = $c000
-
     sub start() {
         test_stack.test()
 
-        save(8, "wob", 0, 10)
+        ubyte x1 = 10
+        ubyte x2 = 20
+        ubyte x3 = 30
 
-        uword @shared uw
+        x1 += x2+x3         ; TODO WHY SLOW EVAL????
-        ubyte @shared ub
+        x1 += x2-x3         ; TODO WHY SLOW EVAL????
-        word @shared ww
-        byte @shared bb
 
-        push(ub+1)
+        txt.print_ub(x1)
-        pop(ub)
-        txt.print_ub(ub)
-        txt.nl()
-        pushw(32767)
-        popw(uw)
-        txt.print_uw(uw)
-        txt.nl()
-
-        single(ub+1)
-
-;        uw=10000
-;        routines(44,uw+123)
-;        routines2(44,uw+123)
-;
-;        routine(uw+123, 22,33, true, 44)
-;        routine2(uw+123, 22,33, true, 44)
 
         test_stack.test()
 
@@ -42,58 +21,4 @@ main {
         }
 
     }
-
-    sub save(ubyte drivenumber, uword filenameptr, uword address, uword size)  {
-        c64.SETNAM(string.length(filenameptr), filenameptr)
-    }
-
-    sub single(ubyte num) {
-        num++
-    }
-
-    sub routine(uword num, ubyte a1, ubyte a2, ubyte switch, byte a3) {
-        txt.print_uw(num)
-        txt.spc()
-        txt.print_ub(a1)
-        txt.spc()
-        txt.print_ub(a2)
-        txt.spc()
-        txt.print_ub(switch)
-        txt.spc()
-        txt.print_b(a3)
-        txt.nl()
-    }
-
-    sub routines(ubyte bb, uword num) {
-        txt.print_ub(bb)
-        txt.spc()
-        txt.print_uw(num)
-        txt.nl()
-    }
-
-    asmsub routine2(uword num @AY, ubyte a1 @R1, ubyte a2 @R0, ubyte switch @Pc, ubyte a3 @X) {
-        %asm {{
-            sta  routine.num
-            sty  routine.num+1
-            lda  #0
-            adc  #0
-            sta  routine.switch
-            lda  cx16.r0
-            sta  routine.a2
-            lda  cx16.r1
-            sta  routine.a1
-            stx  routine.a3
-            jmp  routine
-        }}
-    }
-
-    asmsub routines2(ubyte bb @X, uword num @AY) {
-        %asm {{
-            sta  routines.num
-            sty  routines.num+1
-            stx  routines.bb
-            jmp  routines
-        }}
-    }
-
 }