git_clone/thorvg
1
0
Fork 0
mirror of https://github.com/thorvg/thorvg.git synced 2025-06-13 11:36:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

sw_engine : arm neonRasterGrayscale8 function to support aarch64

Browse source 
Improved the speed through neon processing.

Improvements Rate:
Lottie: (0.025986/0.026201) = +4.7%
Performance: (0.014163/0.014785) = +4.3%

issue: https://github.com/thorvg/thorvg/issues/30
This commit is contained in:
rinechran 2024-05-03 01:01:17 +09:00 • committed by Hermet Park
parent c8551d4856
commit c1394668ef
1 changed files with 17 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

20
src/renderer/sw_engine/tvgSwRasterNeon.h
View file

@ -24,6 +24,15 @@
#include <arm_neon.h> #include <arm_neon.h>
//TODO : need to support windows ARM
#if defined(__ARM_64BIT_STATE) || defined(_M_ARM64)
#define TVG_AARCH64 1
#else
#define TVG_AARCH64 0
#endif
static inline uint8x8_t ALPHA_BLEND(uint8x8_t c, uint8x8_t a) static inline uint8x8_t ALPHA_BLEND(uint8x8_t c, uint8x8_t a)
{ {
uint16x8_t t = vmull_u8(c, a); uint16x8_t t = vmull_u8(c, a);
@ -36,12 +45,17 @@ static void neonRasterGrayscale8(uint8_t* dst, uint8_t val, uint32_t offset, int
dst += offset; dst += offset;
int32_t i = 0; int32_t i = 0;
uint8x16_t valVec = vdupq_n_u8(val); const uint8x16_t valVec = vdupq_n_u8(val);
#if TVG_AARCH64
uint8x16x4_t valQuad = {valVec, valVec, valVec, valVec};
for (; i <= len - 16 * 4; i += 16 * 4) {
vst1q_u8_x4(dst + i, valQuad);
}
#else
for (; i <= len - 16; i += 16) { for (; i <= len - 16; i += 16) {
vst1q_u8(dst + i, valVec); vst1q_u8(dst + i, valVec);
} }
#endif
for (; i < len; i++) { for (; i < len; i++) {
dst[i] = val; dst[i] = val;
} }