diff --git a/mandel.s b/mandel.s
index d8b25e1..4444274 100644
--- a/mandel.s
+++ b/mandel.s
@@ -30,8 +30,8 @@ FRX = $ec
 .endmacro
 
 ; inner loop for imul16
-; bitnum < 8: 25 or 30 cycles
-; bitnum >= 8: 39 or 44 cycles
+; bitnum < 8: 25 or 41 cycles
+; bitnum >= 8: 30 or 46 cycles
 .macro bitmul16 arg1, arg2, result, bitnum
     .local zero
     .local one
@@ -54,7 +54,7 @@ FRX = $ec
     .endif
     bne one ; 2 cyc
 
-zero: ; 18 cyc / 23 cyc
+zero: ; 18 cyc, 23 cyc
     lsr result + 3 ; 5 cyc
     ror result + 2 ; 5 cyc
     ror result + 1 ; 5 cyc
@@ -65,7 +65,7 @@ zero: ; 18 cyc / 23 cyc
     .endif
     jmp next       ; 3 cyc
 
-one: ; 32 cyc / 37 cyc
+one: ; 32 cyc, 37 cyc
     ; 16-bit add on the top bits
     clc            ; 2 cyc
     lda result + 2 ; 3 cyc
@@ -73,13 +73,16 @@ one: ; 32 cyc / 37 cyc
     sta result + 2 ; 3 cyc
     lda result + 3 ; 3 cyc
     adc arg2 + 1   ; 3 cyc
-    ror a          ; 2 cyc - get a jump on the shift
+    ror a          ; 2 cyc
     sta result + 3 ; 3 cyc
     ror result + 2 ; 5 cyc
     ror result + 1 ; 5 cyc
     .if bitnum >= 8
+        ; we can save 5 cycles * 8 bits = 40 cycles total by skipping this byte
+        ; when it's all uninitialized data
         ror result ; 5 cyc
     .endif
+
 next:
 
 .endmacro
@@ -118,8 +121,8 @@ positive:
     ; of a larger routine
     ; 424 to 672 cycles
     .repeat 16, bitnum
-        ; first half: 24 to 40 cycles
-        ; second half: 29 to 44 cycles
+        ; first half: 22 to 40 cycles
+        ; second half: 29 to 47 cycles
         bitmul16 arg1, arg2, result, bitnum
     .endrepeat