git_clone/thorvg
1
0
Fork 0
mirror of https://github.com/thorvg/thorvg.git synced 2025-06-14 12:04:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

sw_engine/neon: arm neonRasterPixel32 function to support aarch64

Browse source 
Improved the speed through neon processing.

Improvements Rate:
Lottie: (0.026321/0.026779) = +1.8%
Performance: (0.015411/0.015732) = +2.1%

issue: https://github.com/thorvg/thorvg/issues/30
This commit is contained in:
rinechran 2024-05-07 15:33:58 +09:00 • committed by Hermet Park
parent 8c04b9d65e
commit 477714778c
1 changed files with 14 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
src/renderer/sw_engine/tvgSwRasterNeon.h
View file

@ -64,17 +64,26 @@ static void neonRasterGrayscale8(uint8_t* dst, uint8_t val, uint32_t offset, int
static void neonRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int32_t len) static void neonRasterPixel32(uint32_t *dst, uint32_t val, uint32_t offset, int32_t len)
{ {
dst += offset;
uint32x4_t vectorVal = vdupq_n_u32(val);
#if TVG_AARCH64
uint32_t iterations = len / 16;
uint32_t neonFilled = iterations * 16;
uint32x4x4_t valQuad = {vectorVal, vectorVal, vectorVal, vectorVal};
for (uint32_t i = 0; i < iterations; ++i) {
vst4q_u32(dst, valQuad);
dst += 16;
}
#else
uint32_t iterations = len / 4; uint32_t iterations = len / 4;
uint32_t neonFilled = iterations * 4; uint32_t neonFilled = iterations * 4;
dst += offset;
uint32x4_t vectorVal = {val, val, val, val};
for (uint32_t i = 0; i < iterations; ++i) { for (uint32_t i = 0; i < iterations; ++i) {
vst1q_u32(dst, vectorVal); vst1q_u32(dst, vectorVal);
dst += 4; dst += 4;
} }
#endif
int32_t leftovers = len - neonFilled; int32_t leftovers = len - neonFilled;
while (leftovers--) *dst++ = val; while (leftovers--) *dst++ = val;
} }