optimized code for stack eval comparisons with zero

codeGeneration/src/prog8/codegen/target/cpu6502/codegen/ExpressionsAsmGen.kt

@@ -540,6 +540,13 @@ internal class ExpressionsAsmGen(private val program: Program, private val asmge
                     }
                 }
             }
+            in ComparisonOperators -> {
+                if(leftDt in NumericDatatypes && rightDt in NumericDatatypes) {
+                    val rightVal = expr.right.constValue(program)?.number?.toInt()
+                    if(rightVal==0)
+                        return translateComparisonWithZero(expr.left, leftDt, expr.operator)
+                }
+            }
         }
 
         if((leftDt in ByteDatatypes && rightDt !in ByteDatatypes)
@@ -562,6 +569,69 @@ internal class ExpressionsAsmGen(private val program: Program, private val asmge
         }
     }
 
+    private fun translateComparisonWithZero(expr: Expression, dt: DataType, operator: String) {
+        translateExpressionInternal(expr)
+        when(operator) {
+            "==" -> {
+                when(dt) {
+                    DataType.UBYTE, DataType.BYTE -> asmgen.out("  jsr  prog8_lib.equalzero_b")
+                    DataType.UWORD, DataType.WORD -> asmgen.out("  jsr  prog8_lib.equalzero_w")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.equal_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            "!=" -> {
+                when(dt) {
+                    DataType.UBYTE, DataType.BYTE -> asmgen.out("  jsr  prog8_lib.notequalzero_b")
+                    DataType.UWORD, DataType.WORD -> asmgen.out("  jsr  prog8_lib.notequalzero_w")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.notequal_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            "<" -> {
+                if(dt==DataType.UBYTE || dt==DataType.UWORD)
+                    return translateExpressionInternal(NumericLiteralValue.fromBoolean(false, expr.position))
+                when(dt) {
+                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.lesszero_b")
+                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.lesszero_w")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.less_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            ">" -> {
+                when(dt) {
+                    DataType.UBYTE -> asmgen.out("  jsr  prog8_lib.greaterzero_ub")
+                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.greaterzero_sb")
+                    DataType.UWORD -> asmgen.out("  jsr  prog8_lib.greaterzero_uw")
+                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.greaterzero_sw")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.greater_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            "<=" -> {
+                when(dt) {
+                    DataType.UBYTE -> asmgen.out("  jsr  prog8_lib.equalzero_b")
+                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.lessequalzeros_b")
+                    DataType.UWORD -> asmgen.out("  jsr  prog8_lib.equalzero_w")
+                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.lessequalzero_sw")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.lessequal_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            ">=" -> {
+                if(dt==DataType.UBYTE || dt==DataType.UWORD)
+                    return translateExpressionInternal(NumericLiteralValue.fromBoolean(true, expr.position))
+                when(dt) {
+                    DataType.BYTE -> asmgen.out("  jsr  prog8_lib.greaterequalzero_sb")
+                    DataType.WORD -> asmgen.out("  jsr  prog8_lib.greaterequalzero_sw")
+                    DataType.FLOAT -> asmgen.out("  jsr  floats.greaterequal_zero")
+                    else -> throw AssemblyError("wrong dt")
+                }
+            }
+            else -> throw AssemblyError("invalid comparison operator")
+        }
+    }
+
     private fun translateSquared(variable: IdentifierReference, dt: DataType) {
         val asmVar = asmgen.asmVariableName(variable)
         when(dt) {

compiler/res/prog8lib/c64/floats.asm

@@ -678,3 +678,54 @@ set_array_float		.proc
 		.pend
 
 
+equal_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		beq  _true
+		bne  _false
+_true		lda  #1
+		sta  P8ESTACK_LO,x
+		dex
+		rts
+_false		lda  #0
+		sta  P8ESTACK_LO,x
+		dex
+		rts
+		.pend
+
+notequal_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		bne  equal_zero._true
+		beq  equal_zero._false
+		.pend
+
+greater_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		beq  equal_zero._false
+		bpl  equal_zero._true
+		jmp  equal_zero._false
+		.pend
+
+less_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		bmi  equal_zero._true
+		jmp  equal_zero._false
+		.pend
+
+greaterequal_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		bpl  equal_zero._true
+		jmp  equal_zero._false
+		.pend
+
+lessequal_zero	.proc
+		jsr  floats.pop_float_fac1
+		jsr  floats.SIGN
+		beq  equal_zero._true
+		bmi  equal_zero._true
+		jmp  equal_zero._false
+		.pend

compiler/res/prog8lib/prog8_lib.asm

@@ -4,6 +4,8 @@
 ; Written by Irmen de Jong (irmen@razorvine.net) - license: GNU GPL 3.0
 
 
+orig_stackpointer	.byte  0	; stores the Stack pointer register at program start
+
 read_byte_from_address_on_stack	.proc
 	; -- read the byte from the memory address on the top of the stack, return in A (stack remains unchanged)
 		lda  P8ESTACK_LO+1,x
@@ -683,8 +685,106 @@ shiftright_b	.proc
 		.pend
 
 
-orig_stackpointer	.byte  0	; stores the Stack pointer register at program start
+equalzero_b	.proc
+		lda  P8ESTACK_LO+1,x
+		beq  _true
+		bne  _false
+_true		lda  #1
+		sta  P8ESTACK_LO+1,x
+		rts
+_false		lda  #0
+		sta  P8ESTACK_LO+1,x
+		rts
+		.pend
 
+equalzero_w	.proc
+		lda  P8ESTACK_LO+1,x
+		ora  P8ESTACK_HI+1,x
+		beq  equalzero_b._true
+		bne  equalzero_b._false
+		.pend
+
+notequalzero_b	.proc
+		lda  P8ESTACK_LO+1,x
+		beq  equalzero_b._false
+		bne  equalzero_b._true
+		.pend
+
+notequalzero_w	.proc
+		lda  P8ESTACK_LO+1,x
+		ora  P8ESTACK_HI+1,x
+		beq  equalzero_b._false
+		bne  equalzero_b._true
+		.pend
+
+lesszero_b	.proc
+		lda  P8ESTACK_LO+1,x
+		bmi  equalzero_b._true
+		jmp  equalzero_b._false
+		.pend
+
+lesszero_w	.proc
+		lda  P8ESTACK_HI+1,x
+		bmi  equalzero_b._true
+		jmp  equalzero_b._false
+		.pend
+
+greaterzero_ub	.proc
+		lda  P8ESTACK_LO+1,x
+		bne  equalzero_b._true
+		beq  equalzero_b._false
+		.pend
+
+greaterzero_sb	.proc
+		lda  P8ESTACK_LO+1,x
+		beq  equalzero_b._false
+		bpl  equalzero_b._true
+		bmi  equalzero_b._false
+		.pend
+
+greaterzero_uw	.proc
+		lda  P8ESTACK_LO+1,x
+		ora  P8ESTACK_HI+1,x
+		bne  equalzero_b._true
+		beq  equalzero_b._false
+		.pend
+
+greaterzero_sw	.proc
+		lda  P8ESTACK_HI+1,x
+		bmi  equalzero_b._false
+		ora  P8ESTACK_LO+1,x
+		beq  equalzero_b._false
+		bne  equalzero_b._true
+		.pend
+
+lessequalzero_sb	.proc
+		lda  P8ESTACK_LO+1,x
+		bmi  equalzero_b._true
+		beq  equalzero_b._true
+		bne  equalzero_b._false
+		.pend
+
+lessequalzero_sw	.proc
+		lda  P8ESTACK_HI+1,x
+		bmi  equalzero_b._true
+		ora  P8ESTACK_LO+1,x
+		beq  equalzero_b._true
+		bne  equalzero_b._false
+		.pend
+
+greaterequalzero_sb	.proc
+		lda  P8ESTACK_LO+1,x
+	    	bpl  equalzero_b._true
+	    	bmi  equalzero_b._false
+		.pend
+
+greaterequalzero_sw	.proc
+		lda  P8ESTACK_HI+1,x
+		bmi  equalzero_b._false
+		ora  P8ESTACK_LO+1,x
+		beq  equalzero_b._true
+		bne  equalzero_b._false
+		.pend
 
 
 memcopy16_up	.proc

docs/source/todo.rst

@@ -3,10 +3,10 @@ TODO
 
 For next compiler release (7.7)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-- why is wormfood 40 bytes larger now since 7.6???
+- optimize (u)word comparison against 0 to use MSB only if possible
-- optimize codegen of pipe operator to avoid needless assigns to temp var
 - copying floats around: do it with a subroutine rather than 5 lda/sta pairs .
   is slower but floats are very slow already anyway and this should take a lot less program size.
+- optimize codegen of pipe operator to avoid needless assigns to temp var
 
 
 Need help with

examples/test.p8

@@ -5,11 +5,37 @@
 
 main {
     sub start() {
-        float @shared f1
+        byte bb = 0
+        word ww = 0
+        float fl = 0
 
-        f1 =   1.234 |> addfloat1 |> addfloat2 |> addfloat3         ; TODO fix that the value is actually returned
+        if fl< 0 {
-        floats.print_f(f1)
+            txt.print("wrong fl\n")
-        txt.nl()
+        }
+        fl=-1.111
+        if 0>fl or fl==2 {
+            txt.print("good fl\n")
+        }
+
+        if ww< 0 {
+            txt.print("wrong ww\n")
+        }
+        if bb<0 {
+            txt.print("wrong bb\n")
+        }
+        bb = -1
+        ww = -1111
+        if 0>ww or ww==2 {
+            txt.print("good ww\n")
+        }
+        if 0>bb or bb==2 {
+            txt.print("good bb\n")
+        }
+;        float @shared f1
+;
+;        f1 =   1.234 |> addfloat1 |> addfloat2 |> addfloat3         ; TODO fix that the value is actually returned
+;        floats.print_f(f1)
+;        txt.nl()
 ;        1.234 |> addfloat1
 ;            |> addfloat2 |> addfloat3 |> floats.print_f
 ;        txt.nl()
@@ -21,15 +47,15 @@ main {
 ;            |> txt.print_uw
 ;        txt.nl()
 
-        test_stack.test()
+;        test_stack.test()
-         ; TODO fix that the value is actually returned (398) and that X register is preserved:
+;         ; TODO fix that the value is actually returned (398) and that X register is preserved:
-        uword @shared uw=  9+3 |> assemblything
+;        uword @shared uw=  9+3 |> assemblything
-                             |> sin8u
+;                             |> sin8u
-                             |> add_one
+;                             |> add_one
-                             |> times_two
+;                             |> times_two
-        txt.print_uw(uw)
+;        txt.print_uw(uw)
-        txt.nl()
+;        txt.nl()
-        test_stack.test()
+;        test_stack.test()
     }
 
     sub addfloat1(float fl) -> float {