From 8769cc16576589d2802c794ca61fa7fd3294dfa8 Mon Sep 17 00:00:00 2001
From: FeralChild64 <you@example.com>
Date: Wed, 29 Jul 2020 17:17:41 +0200
Subject: [PATCH] added alternative, smaller 8x8 multiplication routine

---
 CREDITS.md                         |  2 +-
 src/basic/math/mul_FAC2_FAC1_8x8.s | 35 +++++++++++++++++++++++++++++-
 2 files changed, 35 insertions(+), 2 deletions(-)

diff --git a/CREDITS.md b/CREDITS.md
index c1cfb957..c1a3082a 100644
--- a/CREDITS.md
+++ b/CREDITS.md
@@ -5,7 +5,7 @@
 
 * Retrofan - for his Mega65 banner artwork and PXLfont
 * TWW/CTR - for his keyboard handling code
-* Graham - for his PAL/NTSC detection routine
+* Graham - for his PAL/NTSC detection code and compact 8x8 multiplication routine
 * enthusi/Onslaught - for his example Turbo Tape 64 compatible loader
 * djmips - for fast 8x8 multiplication routine
 * Verz - for the integer division routine
diff --git a/src/basic/math/mul_FAC2_FAC1_8x8.s b/src/basic/math/mul_FAC2_FAC1_8x8.s
index bfc0730c..ff06c0c2 100644
--- a/src/basic/math/mul_FAC2_FAC1_8x8.s
+++ b/src/basic/math/mul_FAC2_FAC1_8x8.s
@@ -3,8 +3,10 @@
 // #LAYOUT# *   *       #IGNORE
 
 //
-// Helper routine, based on code by djmips from:
+// Helper routine, based on code by djmips (fast version) from:
 // - https://codebase64.org/doku.php?id=base:8bit_multiplication_16bit_product_fast_no_tables
+// or by code by Graham (small version) from:
+// - https://codebase64.org/doku.php?id=base:short_8bit_multiplication_16bit_product
 //
 // input: .A (and Zero flag set according to .A), INDEX+3
 // output: .X (high byte), .A (low byte), carry always clear
@@ -22,6 +24,10 @@ mul_FAC2_FAC1_8x8:
 
 	// Multiply INDEX+2 * INDEX+3
 
+#if !HAS_SMALL_BASIC
+
+	// Fast multiplication routine
+
 	lda #$00
 
 	dec INDEX+3	// decrement because we will be adding with carry set for speed (an extra one)
@@ -71,6 +77,33 @@ mul_FAC2_FAC1_8x8:
 	tax
 	lda INDEX+2
 
+#else
+
+	// Small multiplication routine
+
+	lda #$00
+	ldx #$08
+	clc
+!:
+	bcc !+
+	clc
+	adc INDEX+3
+!:
+	ror
+	ror INDEX+2
+	dex
+	bpl !--
+ 	ldx INDEX+2
+
+ 	// Reverse return byte order, to match fast version
+ 	// (such order is easier top handle for the callers)
+
+ 	stx INDEX+2
+ 	tax
+ 	lda INDEX+2 
+
+#endif
+
 	clc
 	rts