diff --git a/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/AsmsubHelpers.kt b/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/AsmsubHelpers.kt
index 3480282fb..a5a1eff8f 100644
--- a/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/AsmsubHelpers.kt
+++ b/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/AsmsubHelpers.kt
@@ -11,22 +11,25 @@ internal fun asmsub6502ArgsEvalOrder(sub: Subroutine): List<Int> {
     // order is:
     //  1) cx16 virtual word registers,
     //  2) paired CPU registers,
-    //  3) single CPU registers (X last),
+    //  3) single CPU registers (X last), except A,
     //  4) CPU Carry status flag
+    //  5) the A register itself last   (so everything before it can use the accumulator without having to save its value)
     val args = sub.parameters.zip(sub.asmParameterRegisters).withIndex()
     val (cx16regs, args2) = args.partition { it.value.second.registerOrPair in Cx16VirtualRegisters }
     val pairedRegisters = arrayOf(RegisterOrPair.AX, RegisterOrPair.AY, RegisterOrPair.XY)
     val (pairedRegs , args3) = args2.partition { it.value.second.registerOrPair in pairedRegisters }
-    val (regs, rest) = args3.partition { it.value.second.registerOrPair != null }
+    val (regsWithoutA, args4) = args3.partition { it.value.second.registerOrPair != RegisterOrPair.A }
+    val (regA, rest) = args4.partition { it.value.second.registerOrPair != null }
 
     cx16regs.forEach { order += it.index }
     pairedRegs.forEach { order += it.index }
-    regs.forEach {
+    regsWithoutA.forEach {
         if(it.value.second.registerOrPair != RegisterOrPair.X)
             order += it.index
     }
-    regs.firstOrNull { it.value.second.registerOrPair==RegisterOrPair.X } ?.let { order += it.index}
+    regsWithoutA.firstOrNull { it.value.second.registerOrPair==RegisterOrPair.X } ?.let { order += it.index}
     rest.forEach { order += it.index }
+    regA.forEach { order += it.index }
     require(order.size==sub.parameters.size)
     return order
 }
diff --git a/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/BuiltinFunctionsAsmGen.kt b/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/BuiltinFunctionsAsmGen.kt
index e47a7b9e7..7baf66a0e 100644
--- a/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/BuiltinFunctionsAsmGen.kt
+++ b/codeGeneration/src/prog8/compiler/target/cpu6502/codegen/BuiltinFunctionsAsmGen.kt
@@ -77,6 +77,7 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
     }
 
     private fun funcPop(fcall: IFunctionCall, func: FSignature) {
+        // note: because A is pushed first so popped last, saving A is often not required here.
         require(fcall.args[0] is IdentifierReference) {
             "attempt to pop a value into a differently typed variable, or in something else that isn't supported ${(fcall as Node).position}"
         }
@@ -87,15 +88,19 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
             require(sub.isAsmSubroutine) {
                 "push/pop arg passing only supported on asmsubs ${(fcall as Node).position}"
             }
+            val shouldKeepA = sub.asmParameterRegisters.any { it.registerOrPair==RegisterOrPair.AX || it.registerOrPair==RegisterOrPair.AY }
             val reg = sub.asmParameterRegisters[sub.parameters.indexOf(parameter)]
             if(reg.statusflag!=null) {
+                if(shouldKeepA)
+                    asmgen.out("  sta  P8ZP_SCRATCH_REG")
                 asmgen.out("""
-                    sta  P8ZP_SCRATCH_REG
                     clc
                     pla
                     beq  +
                     sec
-+                   lda  P8ZP_SCRATCH_REG""")
++""")
+                if(shouldKeepA)
+                    asmgen.out("  lda  P8ZP_SCRATCH_REG")
             }
             else {
                 if (func.name == "pop") {
@@ -109,15 +114,25 @@ internal class BuiltinFunctionsAsmGen(private val program: Program, private val
                         }
                     } else {
                         when (reg.registerOrPair) {
-                            // TODO make sure that A is pushed first so popped last, so that those store/load to save A are no longer needed (c64)
                             RegisterOrPair.A -> asmgen.out("  pla")
-                            RegisterOrPair.X -> asmgen.out("  sta  P8ZP_SCRATCH_REG |  pla |  tax |  lda  P8ZP_SCRATCH_REG")
-                            RegisterOrPair.Y -> asmgen.out("  sta  P8ZP_SCRATCH_REG |  pla |  tay |  lda  P8ZP_SCRATCH_REG")
+                            RegisterOrPair.X -> {
+                                if(shouldKeepA)
+                                    asmgen.out("  sta  P8ZP_SCRATCH_REG |  pla |  tax |  lda  P8ZP_SCRATCH_REG")
+                                else
+                                    asmgen.out("  pla |  tax")
+                            }
+                            RegisterOrPair.Y -> {
+                                if(shouldKeepA)
+                                    asmgen.out("  sta  P8ZP_SCRATCH_REG |  pla |  tay |  lda  P8ZP_SCRATCH_REG")
+                                else
+                                    asmgen.out("  pla |  tay")
+                            }
                             in Cx16VirtualRegisters -> asmgen.out("  pla |  sta  cx16.${reg.registerOrPair!!.name.lowercase()}")
                             else -> throw AssemblyError("invalid target register ${reg.registerOrPair}")
                         }
                     }
                 } else {
+                    // word pop
                     if (asmgen.isTargetCpu(CpuType.CPU65c02))
                         when (reg.registerOrPair) {
                             RegisterOrPair.AX -> asmgen.out("  plx |  pla")
diff --git a/examples/test.p8 b/examples/test.p8
index af0675272..e807fae03 100644
--- a/examples/test.p8
+++ b/examples/test.p8
@@ -58,19 +58,17 @@ main {
         txt.nl()
     }
 
-    asmsub routine2(uword num @AY, ubyte a1 @R1, ubyte a2 @R2, ubyte switch @Pc, ubyte a3 @X) {
+    asmsub routine2(uword num @AY, ubyte a1 @R1, ubyte a2 @R0, ubyte switch @Pc, ubyte a3 @X) {
         %asm {{
-            pha
+            sta  routine.num
+            sty  routine.num+1
             lda  #0
             adc  #0
             sta  routine.switch
-            pla
-            sta  routine.num
-            sty  routine.num+1
+            lda  cx16.r0
+            sta  routine.a2
             lda  cx16.r1
             sta  routine.a1
-            lda  cx16.r2
-            sta  routine.a2
             stx  routine.a3
             jmp  routine
         }}