From 4b1001bfdc5b9abe08b6ce0c277e150a37f5a8ab Mon Sep 17 00:00:00 2001
From: Brion Vibber <brion@pobox.com>
Date: Fri, 30 Dec 2022 17:33:18 -0800
Subject: [PATCH] annotate cycle counts

---
 mandel.s | 72 ++++++++++++++++++++++++++++++--------------------------
 1 file changed, 38 insertions(+), 34 deletions(-)

diff --git a/mandel.s b/mandel.s
index 465ed96..39571b3 100644
--- a/mandel.s
+++ b/mandel.s
@@ -57,19 +57,22 @@ plus:
     neg 4, arg
 .endmacro
 
+; 2 + 9 * bytes cycles
 .macro add bytes, arg1, arg2
-    clc
+    clc               ; 2 cyc
     .repeat bytes, byte
-        lda arg1+byte
-        adc arg2+byte
-        sta arg1+byte
+        lda arg1+byte ; 3 cyc
+        adc arg2+byte ; 3 cyc
+        sta arg1+byte ; 3 cyc
     .endrepeat
 .endmacro
 
+; 20 cycles
 .macro add16 arg1, arg2
     add 2, arg1, arg2
 .endmacro
 
+; 38 cycles
 .macro add32 arg1, arg2
     add 4, arg1, arg2
 .endmacro
@@ -112,6 +115,7 @@ plus:
     shr 4, arg
 .endmacro
 
+; 24 to 49 cycles
 .macro bitmul16 arg1, arg2, result, bitnum
     .local next
 
@@ -119,79 +123,79 @@ plus:
     ; arg1 must be 0 or positive
     ; arg2 must be 0 or positive
 
-    clc
+    clc ; 2 cyc
 
     ; check if arg1 has 0 or 1 bit in this place
+    ; 5 cycles either way
     .if bitnum < 8
-        lda arg1
-        and #(1 << bitnum)
+        lda arg1                 ; 3 cyc
+        and #(1 << bitnum)       ; 2 cyc
     .else
-        lda arg1 + 1
-        and #(1 << (bitnum - 8))
+        lda arg1 + 1             ; 3 cyc
+        and #(1 << (bitnum - 8)) ; 2 cyc
     .endif
-    beq next
+    beq next ; 2 cyc
 
     ; 16-bit add on the top bits
-    lda result + 2
-    adc arg2
-    sta result + 2
-    lda result + 3
-    adc arg2 + 1
-    sta result + 3
+    add16 result + 2, arg2 ; 20 cyc
 
 next:
     ; Shift the 32-bit result down by one bit,
     ; saving the previous carry.
-    ror result + 3
-    ror result + 2
-    ror result + 1
+    ror result + 3 ; 5 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
+        ror result ; 5 cyc
     .endif
 .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 X register.
     .local positive
-    lda arg + 1
-    bpl positive
-    neg16 arg
-    inx
+    lda arg + 1   ; 3 cyc
+    bpl positive  ; 2 cyc
+    neg16 arg     ; 18 cyc
+    inx           ; 2 cyc
 positive:
 .endmacro
 
+; 579 to 725 cycles
 .proc imul16
     arg1 = FR0   ; 16-bit arg (clobbered)
     arg2 = FR1   ; 16-bit arg (clobbered)
     result = FR2 ; 32-bit result
 
-    ldx #0
+    ldx #0          ; 2 cyc
     ; counts the number of sign bits in X
-    check_sign arg1
-    check_sign arg2
+    check_sign arg1 ; 5 to 25 cyc
+    check_sign arg2 ; 5 to 25 cyc
     
     ; zero out the 32-bit temp's top 16 bits
-    lda #0
-    sta result + 2
-    sta result + 3
+    lda #0          ; 2 cyc
+    sta result + 2  ; 3 cyc
+    sta result + 3  ; 3 cyc
     ; 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
+        ; first half: 24 to 45 cycles
+        ; second half: 29 to 49 cycles
         bitmul16 arg1, arg2, result, bitnum
     .endrepeat
 
     ; In case of mixed input signs, return a negative result.
-    cpx #1
-    bne positive_result
-    neg32 result
+    cpx #1              ; 2 cyc
+    bne positive_result ; 2 cyc
+    neg32 result        ; 34 cyc
 positive_result:
 
-    rts
+    rts ; 6 cyc
 .endproc
 
 .proc iter