slightly faster handling of signed mul

previously we were flipping the inputs if negative, and then the
output if both inputs were negative

turns out you can just treat the whole thing as an unsigned mul
and then subtract each term from the high word if the other term
is negative.

https://stackoverflow.com/a/28827013

this saves a handful of cycles, reducing our runtime to a merge
14.211 ms/px \o/

mandel.s

@@ -344,18 +344,6 @@ fill_masks:
     neg 4, arg
 .endmacro
 
-; 5 to 25 cycles
-.macro check_sign arg
-    ; Check sign bit and flip argument to postive,
-    ; keeping a count of sign bits in the Y register.
-    .local positive
-    lda arg + 1   ; 3 cyc
-    bpl positive  ; 2 cyc
-    neg16 arg     ; 18 cyc
-    iny           ; 2 cyc
-positive:
-.endmacro
-
 ; 518 - 828 cyc
 .macro imul16 dest, arg1, arg2
     copy16 FR0, arg1  ; 12 cyc
@@ -438,11 +426,6 @@ positive:
     result = FR2 ; 32-bit result
     inter = temp2
 
-    ldy #0          ; 2 cyc
-    ; counts the number of sign bits in Y
-    check_sign arg1 ; 5 to 25 cyc
-    check_sign arg2 ; 5 to 25 cyc
-
     ; h1l1 * h2l2
     ; (h1*256 + l1) * (h2*256 + l2)
     ; h1*256*(h2*256 + l2) + l1*(h2*256 + l2)
@@ -464,11 +447,16 @@ positive:
     imul8 inter, arg1 + 1, arg2 + 1
     add16 result + 2, result + 2, inter
 
-    ; In case of mixed input signs, return a negative result.
-    cpy #1              ; 2 cyc
-    bne positive_result ; 2 cyc
-    neg32 result        ; 34 cyc
-positive_result:
+    ; In case of negative inputs, adjust high word
+    ; https://stackoverflow.com/a/28827013
+    lda arg1 + 1
+    bpl arg1_pos
+    sub16 result + 2, result + 2, arg2
+arg1_pos:
+    lda arg2 + 1
+    bpl arg2_pos
+    sub16 result + 2, result + 2, arg1
+arg2_pos:
 
     rts ; 6 cyc
 .endproc