From d93dd2e6a6b6c9adedd9cb718c5227f0bff62558 Mon Sep 17 00:00:00 2001
From: Brooke Vibber <bvibber@pobox.com>
Date: Sun, 23 Aug 2026 15:18:25 -0700
Subject: [PATCH] step 1: sqr16_impl and imul16_impl take argument addresses

first step to removing function call overhead
---
 mandel-core.s | 49 ++++++++++++++++++++++++++-----------------------
 1 file changed, 26 insertions(+), 23 deletions(-)

diff --git a/mandel-core.s b/mandel-core.s
index 64c5540..2e1c8b2 100644
--- a/mandel-core.s
+++ b/mandel-core.s
@@ -799,15 +799,15 @@ inner_loop:
 
 .endproc
 
-; 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: 50 * 4 + 22 * 2 + 10 + 6 = 260 cyc
-; max ext-mem: 55 * 4 + 27 * 2 + 50 + 6 = 330 cyc
-.macro imul16_impl xe
+; min base mem: 81 * 4 + 22 * 2 + 10 = 378 cyc
+; max base mem: 92 * 4 + 27 * 2 + 50 = 472 cyc
+; min ext-mem: 50 * 4 + 22 * 2 + 10 = 254 cyc
+; max ext-mem: 55 * 4 + 27 * 2 + 50 = 324 cyc
+.macro imul16_impl result, arg1, arg2, xe
     .scope
-        arg1 = FR0    ; 16-bit arg
-        arg2 = FR1    ; 16-bit arg
-        result = FR2  ; 32-bit result (output)
+        ;arg1         ; 16-bit arg
+        ;arg2         ; 16-bit arg
+        ;result       ; 32-bit result (output)
         inter = temp2 ; 16-bit temporary (clobbered)
 
         ; h1l1 * h2l2
@@ -843,8 +843,6 @@ inner_loop:
         bpl arg2_pos                       ; 2 cyc
         sub16 result + 2, result + 2, arg1 ; 20 cyc
     arg2_pos:
-
-        rts ; 6 cyc
     .endscope
 .endmacro
 
@@ -877,15 +875,15 @@ inner_loop:
 .endmacro
 
 
-; min base ram: 5 + 163 + 6 = 174
-; max base ram: 7 + 184 + 18 + 6 = 215
-; min ext ram: 5 + 132 + 6 = 143
-; max ext ram: 7 + 147 + 18 + 6 = 178
-.macro sqr16_impl xe
+; min base ram: 5 + 163 + 3 = 171
+; max base ram: 7 + 184 + 18 = 209
+; min ext ram: 5 + 132 + 3 = 140
+; max ext ram: 7 + 147 + 18 = 172
+.macro sqr16_impl result, arg, xe
     .scope
-        arg = FR0    ; 16-bit arg
+        ; arg        ; 16-bit arg
         negated = FR1
-        result = FR2 ; 32-bit result
+        ; result     ; 32-bit result
 
         ; 5-7 cycles
         lda arg + 1                ; 3 cyc
@@ -894,29 +892,34 @@ inner_loop:
 
     arg_pos:
         sqr16_impl_inner result, arg, xe      ; 163-184 / 132-147 cyc
-        rts                                   ; 6 cyc
+        jmp done                              ; 3 cyc
 
     arg_neg:
         copy_neg16 negated, arg               ; 18 cyc
         sqr16_impl_inner result, negated, xe  ; 163-184 / 132-147 cyc
-        rts                                   ; 6 cyc
+
+    done:
     .endscope
 .endmacro
 
 .proc imul16_func
-    imul16_impl 0
+    imul16_impl FR2, FR0, FR1, 0
+    rts
 .endproc
 
 .proc imul16xe_func
-    imul16_impl 1
+    imul16_impl FR2, FR0, FR1, 1
+    rts
 .endproc
 
 .proc sqr16_func
-    sqr16_impl 0
+    sqr16_impl FR2, FR0, 0
+    rts
 .endproc
 
 .proc sqr16xe_func
-    sqr16_impl 1
+    sqr16_impl FR2, FR0, 1
+    rts
 .endproc
 
 ; 11-27 cycles