Merge "vpx_idct32x32_34_add_sse2: rm unneeded transposes"
This commit is contained in:
commit
deadda3dea
@ -3066,17 +3066,7 @@ void vpx_idct32x32_34_add_sse2(const tran_low_t *input, uint8_t *dest,
|
||||
in[6] = load_input_data(input + 192);
|
||||
in[7] = load_input_data(input + 224);
|
||||
|
||||
for (i = 8; i < 32; ++i) {
|
||||
in[i] = _mm_setzero_si128();
|
||||
}
|
||||
|
||||
array_transpose_8x8(in, in);
|
||||
// TODO(hkuang): Following transposes are unnecessary. But remove them will
|
||||
// lead to performance drop on some devices.
|
||||
array_transpose_8x8(in + 8, in + 8);
|
||||
array_transpose_8x8(in + 16, in + 16);
|
||||
array_transpose_8x8(in + 24, in + 24);
|
||||
|
||||
IDCT32_34
|
||||
|
||||
// 1_D: Store 32 intermediate results for each 8x32 block.
|
||||
|
Loading…
x
Reference in New Issue
Block a user