WIP savings of half a cycle per imul8_xe

Uses X to cache arg1, which is always used, instead of
arg2, which is only used on odds.

Should save half a cycle per imul8_xe, untested

mandel.s

@@ -588,7 +588,7 @@ bank_switch_table:
 .macro imul8 dest, arg1, arg2, xe
     .if xe
         ; using 64KB lookup table
-        ; 51-70 cycles
+        ; 50-70 cycles
         ; clobbers x, y, dest, ptr
         .scope
             output = dest
@@ -600,13 +600,13 @@ bank_switch_table:
 
             ; bottom 14 bits except the LSB are the per-bank table index
             ; add $4000 for the bank pointer
-            txa          ; 2 cyc
             and #$3f     ; 2 cyc
             ora #$40     ; 2 cyc
             sta ptr + 1  ; 3 cyc
 
             ; copy the entry into output
             lda arg1     ; 3 cyc
+            tax          ; 2 cyc
             and #$fe     ; 2 cyc
             tay          ; 2 cyc
             lda (ptr),y  ; 5 cyc
@@ -623,13 +623,13 @@ bank_switch_table:
             ;;sta PORTB    ; 4 cyc - disabled
 
             ; check that 1 bit we skipped to fit into space
-            lda arg1     ; 3 cyc
+            txa          ; 2 cyc
             and #1       ; 2 cyc
             beq done     ; 2 cyc
 
             ; add arg2 one last time for the skipped bit
             clc          ; 2 cyc
-            txa          ; 2 cyc
+            lda arg1     ; 3 cyc
             adc output   ; 3 cyc
             sta output   ; 3 cyc
             lda #0       ; 2 cyc