looks workable

mandel.s

@@ -43,12 +43,11 @@ LOG10 = $ded1
     .local plus
     .local minus
     lda arg+1
-    bpl plus
+    asl ; sign -> carry
     lda #$ff
-    jmp minus
+    bcc plus
-plus:
     lda #$00
-minus:
+plus:
     sta arg+2
     sta arg+3
 .endmacro
@@ -68,6 +67,26 @@ minus:
     copy 4, arg1, arg2
 .endmacro
 
+; 2 + 8 * byte cycles
+.macro neg bytes, arg
+    sec ; 2 cyc
+    .repeat bytes, byte ; 8 * byte cycles
+        lda #00         ; 2 cyc
+        sbc arg + byte  ; 3 cyc
+        sta arg + byte  ; 3 cyc
+    .endrepeat
+.endmacro
+
+; 18 cycles
+.macro neg16 arg
+    neg 2, arg
+.endmacro
+
+; 34 cycles
+.macro neg32 arg
+    neg 4, arg
+.endmacro
+
 .macro add bytes, arg1, arg2
     clc
     .repeat bytes, byte
@@ -126,6 +145,10 @@ minus:
 .macro bitmul16 arg1, arg2, result, bitnum
     .local next
 
+    ; does 16-bit adds
+    ; arg1 must be 0 or positive
+    ; arg2 must be 0 or positive
+
     clc
 
     ; check if arg1 has 0 or 1 bit in this place
@@ -159,21 +182,45 @@ next:
     .endif
 .endmacro
 
+.macro check_sign arg
+    ; Check sign bit and flip argument to postive,
+    ; keeping a count of sign bits in the X register.
+    .local positive
+    lda arg + 1
+    bpl positive
+    neg16 arg
+    inx
+positive:
+.endmacro
+
 .proc imul16
-    ; 16-bit arg in FR0
+    arg1 = FR0   ; 16-bit arg (clobbered)
-    ; 16-bit arg in FR1
+    arg2 = FR1   ; 16-bit arg (clobbered)
-    ; 32-bit result in FR2
+    result = FR2 ; 32-bit result
+
+    ldx #0
+    ; counts the number of sign bits in X
+    check_sign arg1
+    check_sign arg2
     
     ; zero out the 32-bit temp's top 16 bits
     lda #0
-    sta FR2 + 2
+    sta result + 2
-    sta FR2 + 3
+    sta result + 3
     ; the bottom two bytes will get cleared by the shifts
 
+    ; unrolled loop for maximum speed, at the cost
+    ; of a larger routine
     .repeat 16, bitnum
-        bitmul16 FR0, FR1, FR2, bitnum
+        bitmul16 arg1, arg2, result, bitnum
     .endrepeat
 
+    ; In case of mixed input signs, return a negative result.
+    cpx #1
+    bne positive_result
+    neg32 result
+positive_result:
+
     rts
 .endproc
 
@@ -202,19 +249,21 @@ loop:
 .endproc
 
 .proc start
-    ; FR0 = 3
+
-    ; FR1 = 5
+loop:
-    lda #3
+    ; FR0 = 5
-    sta FR0
+    ; FR1 = -3
     lda #5
-    sta FR1
+    sta FR0
     lda #0
     sta FR0 + 1
+    lda #$fd
+    sta FR1
+    lda #$ff
     sta FR1 + 1
 
     jsr imul16
-    ; should have 32-bit 15 in FR2
+    ; should have 32-bit -15 in FR2
 
-loop:
     jmp loop
 .endproc