Decent speedup in XE mode

Noticed that the separate addition for the low 1 bit case was
doing some dupe memory loads. Ended up making separate code
paths for even and odd values so the even saves two cycles
(from 52 down to 50 cycles) and the odd saves 14 cycles (from
69 down to 55 cycles). nice!

This gets the XE runtime on default view down from 3m38s to 3m33s,
a 5 second runtime improvement

mandel-core.s

@@ -593,32 +593,57 @@ bank_switch_table:
 .macro imul8 dest, arg1, arg2, xe
     .if xe
         ; using 64KB lookup table
-        ; 52-69 cycles
+        ; 50-55 cycles
+        ; min: 11 + 9 + 9 + 21 = 50
+        ; max: 11 + 9 + 9 + 26 = 55
         ; clobbers x, y, dest, ptr
         .scope
             output = dest
 
             ; top 2 bits are the table bank selector
+            ; note we keep arg2 in X after this lookup as we need it later!
             ldx arg2                ; 3 cyc
             lda bank_switch_table,x ; 4 cyc
             sta PORTB               ; 4 cyc
 
             ; bottom 14 bits except the LSB are the per-bank table index
             ; add $4000 for the bank pointer
+            ; note the low byte of ptr is assumed to remain at 0! do not clobber it!
             txa          ; 2 cyc
             and #$3f     ; 2 cyc
             ora #$40     ; 2 cyc
             sta ptr + 1  ; 3 cyc
 
-            ; copy the entry into output
+            ; mask off the low bit for the index, as the table contains
+            ; entries for every other arg1 value...
             lda arg1     ; 3 cyc
-            and #$fe     ; 2 cyc
             tay          ; 2 cyc
-            lda (ptr),y  ; 5 cyc
+            and #1       ; 2 cyc
-            sta output   ; 3 cyc
+            bne odd      ; 2 cyc
-            iny          ; 2 cyc
+
-            lda (ptr),y  ; 5 cyc
+        even:
-            sta output+1 ; 3 cyc
+            ; copy the entry into output
+            lda (ptr),y    ; 5 cyc
+            sta output     ; 3 cyc
+            iny            ; 2 cyc
+            lda (ptr),y    ; 5 cyc
+            sta output + 1 ; 3 cyc
+            jmp done       ; 3 cyc
+
+        odd:
+
+            ; skip the LSB for the index
+            dey            ; 2 cyc
+
+            ; add arg2 one last time for the skipped bit
+            txa            ; 2 cyc
+            clc            ; 2 cyc
+            adc (ptr), y   ; 5 cyc
+            sta output     ; 3 cyc
+            lda #0         ; 2 cyc
+            iny            ; 2 cyc
+            adc (ptr),y    ; 5 cyc
+            sta output + 1 ; 3 cyc
 
             ; note: we are not restoring memory to save 6 cycles!
             ; this means those 16kb have to be switched back to base RAM
@@ -627,18 +652,6 @@ bank_switch_table:
             ;;lda #$81     ; 2 cyc - disabled
             ;;sta PORTB    ; 4 cyc - disabled
 
-            ; check that 1 bit we skipped to fit into space
-            lda arg1     ; 3 cyc
-            and #1       ; 2 cyc
-            beq done     ; 2 cyc
-
-            ; add arg2 one last time for the skipped bit
-            clc          ; 2 cyc
-            txa          ; 2 cyc
-            adc output   ; 3 cyc
-            sta output   ; 3 cyc
-            add_carry output + 1 ; 2-7 cyc
-
         done:
         .endscope
     .else
@@ -774,8 +787,8 @@ inner_loop:
 
 ; min base mem: 81 * 4 + 22 * 2 + 10 + 6 = 384 cyc
 ; max base mem: 92 * 4 + 27 * 2 + 50 + 6 = 478 cyc
-; min ext-mem: 52 * 4 + 22 * 2 + 10 + 6 = 268 cyc
+; min ext-mem: 50 * 4 + 22 * 2 + 10 + 6 = 260 cyc
-; max ext-mem: 69 * 4 + 27 * 2 + 50 + 6 = 386 cyc
+; max ext-mem: 55 * 4 + 27 * 2 + 50 + 6 = 330 cyc
 .macro imul16_impl xe
     .local arg1
     .local arg2
@@ -794,18 +807,18 @@ inner_loop:
     ; h1*h2*256*256 + h1*l2*256 + h2*l1*256 + l1*l2
 
     ; l1 * l2
-    imul8 result, arg1, arg2, xe        ; 81-92 or 52-69
+    imul8 result, arg1, arg2, xe        ; 81-92 or 50-55
 
     ; 256 * 256 * h1 * h2
-    imul8 result + 2, arg1 + 1, arg2 + 1, xe ; 81-92 or 52-69
+    imul8 result + 2, arg1 + 1, arg2 + 1, xe ; 81-92 or 50-55
 
     ; 256 * h1 * l1
-    imul8 inter, arg1 + 1, arg2, xe     ; 81-92 or 52-69
+    imul8 inter, arg1 + 1, arg2, xe     ; 81-92 or 50-55
     add16 result + 1, result + 1, inter ; 20 cyc
     add_carry result + 3                ; 2-7 cyc
 
     ; 256 * l1 * h2
-    imul8 inter, arg1, arg2 + 1, xe     ; 81-92 or 52-69
+    imul8 inter, arg1, arg2 + 1, xe     ; 81-92 or 50-55
     add16 result + 1, result + 1, inter ; 20 cyc
     add_carry result + 3                ; 2-7 cyc
 
@@ -826,8 +839,8 @@ arg2_pos:
 
 ; min base ram: 5 + 19 * 2 + 81 + 22 * 2 + 6 = 174
 ; max base ram: 23 + 19 * 2 + 92 + 27 * 2 + 6 = 213
-; min ext ram: 5 + 19 * 2 + 52 + 22 * 2 + 6 = 145
+; min ext ram: 5 + 19 * 2 + 50 + 22 * 2 + 6 = 143
-; max ext ram: 23 + 19 * 2 + 69 + 27 * 2 + 6 = 190
+; max ext ram: 23 + 19 * 2 + 55 + 27 * 2 + 6 = 176
 .macro sqr16_impl xe
     .scope
         arg = FR0    ; 16-bit arg (clobbered)
@@ -853,7 +866,7 @@ arg2_pos:
         sqr8 result + 2, arg + 1  ; 19 cyc
 
         ; 2 * h * l * 256
-        imul8 inter, arg + 1, arg, xe        ; 81-92 / 52-69 cyc
+        imul8 inter, arg + 1, arg, xe        ; 81-92 / 50-55 cyc
         add16 result + 1, result + 1, inter  ; 20 cyc
         add_carry result + 3                 ; 2-7 cyc
         add16 result + 1, result + 1, inter  ; 20 cyc