wuggy/Dactyloidae
wuggy/Dactyloidae
1
0
Fork 0
mirror of https://repo.dactyloidae.xyz/Dactyloidae/UXP.git synced 2026-10-09 16:57:30 +09:00
Code Issues Projects Releases Packages Wiki Activity

Update aom to commit id e87fb2378f01103d5d6e477a4ef6892dc714e614

Browse source
This commit is contained in:
trav90 2018-10-18 21:53:44 -05:00 • committed by Roy Tam
commit 992c6637e3
429 changed files with 76047 additions and 40937 deletions
Download patch file Download diff file Expand all files Collapse all files

26
third_party/aom/av1/common/x86/convolve_2d_sse2.c vendored
View file

@ -31,6 +31,7 @@ void av1_convolve_2d_sse2(const uint8_t *src, int src_stride,
int i, j;
const int fo_vert = filter_params_y->taps / 2 - 1;
const int fo_horiz = filter_params_x->taps / 2 - 1;
const int do_average = conv_params->do_average;
const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
const __m128i zero = _mm_setzero_si128();
@ -181,9 +182,15 @@ void av1_convolve_2d_sse2(const uint8_t *src, int src_stride,
// Accumulate values into the destination buffer
__m128i *const p = (__m128i *)&dst[i * dst_stride + j];
_mm_storeu_si128(p, _mm_add_epi32(_mm_loadu_si128(p), res_lo_round));
_mm_storeu_si128(p + 1,
_mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
if (do_average) {
_mm_storeu_si128(p + 0,
_mm_add_epi32(_mm_loadu_si128(p + 0), res_lo_round));
_mm_storeu_si128(p + 1,
_mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
} else {
_mm_storeu_si128(p + 0, res_lo_round);
_mm_storeu_si128(p + 1, res_hi_round);
}
}
}
}
@ -204,6 +211,7 @@ void av1_convolve_2d_sse2(const uint8_t *src, int src_stride,
int i, j;
const int fo_vert = filter_params_y->taps / 2 - 1;
const int fo_horiz = filter_params_x->taps / 2 - 1;
const int do_average = conv_params->do_average;
const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
const __m128i zero = _mm_setzero_si128();
@ -357,9 +365,15 @@ void av1_convolve_2d_sse2(const uint8_t *src, int src_stride,
// Accumulate values into the destination buffer
__m128i *const p = (__m128i *)&dst[i * dst_stride + j];
_mm_storeu_si128(p, _mm_add_epi32(_mm_loadu_si128(p), res_lo_round));
_mm_storeu_si128(p + 1,
_mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
if (do_average) {
_mm_storeu_si128(p + 0,
_mm_add_epi32(_mm_loadu_si128(p + 0), res_lo_round));
_mm_storeu_si128(p + 1,
_mm_add_epi32(_mm_loadu_si128(p + 1), res_hi_round));
} else {
_mm_storeu_si128(p + 0, res_lo_round);
_mm_storeu_si128(p + 1, res_hi_round);
}
}
}
}