From fa0de6dc776a875ed97f74c4261d91629ee58fb7 Mon Sep 17 00:00:00 2001
From: Brooke Vibber <bvibber@pobox.com>
Date: Tue, 16 Sep 2025 21:29:40 -0700
Subject: [PATCH 1/2] 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 | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)

diff --git a/mandel.s b/mandel.s
index b0f9c28..ec9f17f 100644
--- a/mandel.s
+++ b/mandel.s
@@ -461,7 +461,7 @@ input_max:
     sub 4, dest, arg1, arg2
 .endmacro
 
-; 3 + 5 * bytes cycles
+; 3 + 5 * (bytes - 1) cycles
 .macro shl bytes, arg
     asl arg              ; 3 cyc
     .repeat bytes-1, i
@@ -469,17 +469,17 @@ input_max:
     .endrepeat
 .endmacro
 
-; 13 cycles
+; 8 cycles
 .macro shl16 arg
     shl 2, arg
 .endmacro
 
-; 18 cycles
+; 13 cycles
 .macro shl24 arg
     shl 3, arg
 .endmacro
 
-; 23 cycles
+; 18 cycles
 .macro shl32 arg
     shl 4, arg
 .endmacro
@@ -529,11 +529,11 @@ input_max:
     neg 4, arg
 .endmacro
 
-; 11-27 + 23 * shift cycles
-; 103-119 cycles for shift=4
+; 11-27 + 18 * shift cycles
+; 65-81 cycles for shift=3
 .macro shift_round_16 arg, shift
     .repeat shift
-        shl32 arg ; 23 cycles
+        shl32 arg ; 18 cycles
     .endrepeat
     round16 arg ; 11-27 cycles
 .endmacro
@@ -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

From 6479cf530c1c584f33b96f2b19885d02415863bb Mon Sep 17 00:00:00 2001
From: Brooke Vibber <bvibber@pobox.com>
Date: Tue, 16 Sep 2025 21:29:40 -0700
Subject: [PATCH 2/2] update some timings

---
 mandel.s | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/mandel.s b/mandel.s
index b0f9c28..b52f24a 100644
--- a/mandel.s
+++ b/mandel.s
@@ -461,7 +461,7 @@ input_max:
     sub 4, dest, arg1, arg2
 .endmacro
 
-; 3 + 5 * bytes cycles
+; 3 + 5 * (bytes - 1) cycles
 .macro shl bytes, arg
     asl arg              ; 3 cyc
     .repeat bytes-1, i
@@ -469,17 +469,17 @@ input_max:
     .endrepeat
 .endmacro
 
-; 13 cycles
+; 8 cycles
 .macro shl16 arg
     shl 2, arg
 .endmacro
 
-; 18 cycles
+; 13 cycles
 .macro shl24 arg
     shl 3, arg
 .endmacro
 
-; 23 cycles
+; 18 cycles
 .macro shl32 arg
     shl 4, arg
 .endmacro
@@ -529,11 +529,11 @@ input_max:
     neg 4, arg
 .endmacro
 
-; 11-27 + 23 * shift cycles
-; 103-119 cycles for shift=4
+; 11-27 + 18 * shift cycles
+; 65-81 cycles for shift=3
 .macro shift_round_16 arg, shift
     .repeat shift
-        shl32 arg ; 23 cycles
+        shl32 arg ; 18 cycles
     .endrepeat
     round16 arg ; 11-27 cycles
 .endmacro