dany/libwebp
1
0
Fork 0
mirror of https://github.com/webmproject/libwebp.git synced 2024-11-20 12:28:26 +01:00
Code Issues Actions Packages Projects Releases Wiki Activity

replace some mult-long (vmull_u8) with mult-long-accumulate (vmlal_u8)

Browse Source 
saves few instructions

Change-Id: If8f464bb2894a209bba94825a4db9267df126d47
This commit is contained in:
skal 2014-04-14 14:45:44 +02:00
parent 8ff96a027a
commit 7ca2e74bb4
1 changed files with 6 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
src/dsp/enc_neon.c
View File

@ -788,12 +788,9 @@ static WEBP_INLINE void AccumulateSSE16(const uint8_t* const a,
const uint8x16_t a0 = vld1q_u8(a); const uint8x16_t a0 = vld1q_u8(a);
const uint8x16_t b0 = vld1q_u8(b); const uint8x16_t b0 = vld1q_u8(b);
const uint8x16_t abs_diff = vabdq_u8(a0, b0); const uint8x16_t abs_diff = vabdq_u8(a0, b0);
const uint16x8_t prod_l = vmull_u8(vget_low_u8(abs_diff), uint16x8_t prod = vmull_u8(vget_low_u8(abs_diff), vget_low_u8(abs_diff));
vget_low_u8(abs_diff)); prod = vmlal_u8(prod, vget_high_u8(abs_diff), vget_high_u8(abs_diff));
const uint16x8_t prod_h = vmull_u8(vget_high_u8(abs_diff), *sum = vpadalq_u16(*sum, prod); // pair-wise add and accumulate
vget_high_u8(abs_diff));
const uint16x8_t prod = vaddq_u16(prod_h, prod_l);
*sum = vpadalq_u16(*sum, prod); // pair-wise multiply and accumulate
} }
// Horizontal sum of all four uint32_t values in 'sum'. // Horizontal sum of all four uint32_t values in 'sum'.
@ -838,12 +835,9 @@ static int SSE4x4(const uint8_t* a, const uint8_t* b) {
const uint8x16_t a0 = Load4x4(a); const uint8x16_t a0 = Load4x4(a);
const uint8x16_t b0 = Load4x4(b); const uint8x16_t b0 = Load4x4(b);
const uint8x16_t abs_diff = vabdq_u8(a0, b0); const uint8x16_t abs_diff = vabdq_u8(a0, b0);
const uint16x8_t prod_l = vmull_u8(vget_low_u8(abs_diff), uint16x8_t prod = vmull_u8(vget_low_u8(abs_diff), vget_low_u8(abs_diff));
vget_low_u8(abs_diff)); prod = vmlal_u8(prod, vget_high_u8(abs_diff), vget_high_u8(abs_diff));
const uint16x8_t prod_h = vmull_u8(vget_high_u8(abs_diff), return SumToInt(vpaddlq_u16(prod));
vget_high_u8(abs_diff));
const uint32x4_t sum = vpaddlq_u16(vaddq_u16(prod_h, prod_l));
return SumToInt(sum);
} }
#undef LOAD_LANE_32b #undef LOAD_LANE_32b