From 1ddcfcaad365734f6723db62b061a3800e28b6f5 Mon Sep 17 00:00:00 2001
From: cnlohr <lohr85@gmail.com>
Date: Thu, 27 Apr 2023 12:19:51 -0400
Subject: [PATCH] Use fast multiply.
---
examples/ws2812bdemo/color_utilities.h | 65 ++++++++++++++++++++++++--
1 file changed, 62 insertions(+), 3 deletions(-)
diff --git a/examples/ws2812bdemo/color_utilities.h b/examples/ws2812bdemo/color_utilities.h
index 3b6f1b7..15e615c 100644
--- a/examples/ws2812bdemo/color_utilities.h
+++ b/examples/ws2812bdemo/color_utilities.h
@@ -136,6 +136,65 @@ static const unsigned char sintable[] = {
0x26, 0x28, 0x2a, 0x2d, 0x2f, 0x31, 0x34, 0x36, 0x39, 0x3c, 0x3e, 0x41, 0x44, 0x47, 0x49, 0x4c,
0x4f, 0x52, 0x55, 0x58, 0x5b, 0x5e, 0x61, 0x64, 0x67, 0x6a, 0x6d, 0x70, 0x73, 0x76, 0x79, 0x7d, };
+static inline uint32_t FastMultiply( uint32_t big_num, uint32_t small_num ) __attribute__((section(".data")));
+static inline uint32_t FastMultiply( uint32_t big_num, uint32_t small_num )
+{
+ // The CH32V003 is an EC core, so no hardware multiply. GCC's way multiply
+ // is slow, so I wrote this.
+ //
+ // This basically does this:
+ // return small_num * big_num;
+ //
+ // Note: This does NOT check for zero to begin with, though this still
+ // produces the correct results, it is a little weird that even if
+ // small_num is zero it executes once.
+ //
+ // Additionally note, instead of the if( m&1 ) you can do the following:
+ // ret += multiplciant & neg(multiplicand & 1).
+ //
+ // BUT! Shockingly! That is slower than an extra branch! The CH32V003
+ // can branch unbelievably fast.
+ //
+ // This is functionally equivelent and much faster.
+ //
+ // Perf numbers, with small_num set to 180V.
+ // No multiply: 21.3% CPU Usage
+ // Assembly below: 42.4% CPU Usage (1608 bytes for whole program)
+ // C version: 41.4% CPU Usage (1600 bytes for whole program)
+ // Using GCC (__mulsi3) 65.4% CPU Usage (1652 bytes for whole program)
+ //
+ // The multiply can be done manually:
+ uint32_t ret = 0;
+ uint32_t multiplicand = small_num;
+ uint32_t mutliplicant = big_num;
+ do
+ {
+ if( multiplicand & 1 )
+ ret += mutliplicant;
+ mutliplicant<<=1;
+ multiplicand>>=1;
+ } while( multiplicand );
+ return ret;
+
+ // Which is equivelent to the following assembly (If you were curious)
+/*
+ uint32_t ret = 0;
+ asm volatile( "\n\
+ .option rvc;\n\
+ 1: andi t0, %[small], 1\n\
+ beqz t0, 2f\n\
+ add %[ret], %[ret], %[big]\n\
+ 2: srli %[small], %[small], 1\n\
+ slli %[big], %[big], 1\n\
+ bnez %[small], 1b\n\
+ " :
+ [ret]"=&r"(ret), [big]"+&r"(big_num), [small]"+&r"(small_num) : :
+ "t0" );
+ return ret;
+*/
+}
+
+static uint32_t TweenHexColors( uint32_t hexa, uint32_t hexb, int tween ) __attribute__((section(".data")));
static uint32_t TweenHexColors( uint32_t hexa, uint32_t hexb, int tween )
{
if( tween <= 0 ) return hexa;
@@ -148,9 +207,9 @@ static uint32_t TweenHexColors( uint32_t hexa, uint32_t hexb, int tween )
int32_t hbb = hexb & 0xff;
int32_t hbr = (hexb>>8) & 0xff;
int32_t hbg = (hexb>>16) & 0xff;
- int32_t b = (hab * aamt + hbb * bamt + 128) >> 8;
- int32_t r = (har * aamt + hbr * bamt + 128) >> 8;
- int32_t g = (hag * aamt + hbg * bamt + 128) >> 8;
+ int32_t b = (FastMultiply( hab, aamt ) + FastMultiply( hbb, bamt ) + 128) >> 8;
+ int32_t r = (FastMultiply( har, aamt ) + FastMultiply( hbr, bamt ) + 128) >> 8;
+ int32_t g = (FastMultiply( hag, aamt ) + FastMultiply( hbg, bamt ) + 128) >> 8;
return b | (r<<8) | (g<<16);
}
--
GitLab