diff options
| author | Scott LaVarnway <slavarnway@google.com> | 2018-06-20 15:10:46 -0700 |
|---|---|---|
| committer | James Zern <jzern@google.com> | 2018-06-22 17:55:48 +0000 |
| commit | a3c2774126a00674ea70fe6e2d722ea36b9f3527 (patch) | |
| tree | c3607a9a341cd96c2ee8cdf0e88fe2ed7748d628 /vpx_dsp | |
| parent | 729d7d6a2ff0f9a9d2f5534767ddb7edd6f598a7 (diff) | |
| download | libvpx-a3c2774126a00674ea70fe6e2d722ea36b9f3527.tar libvpx-a3c2774126a00674ea70fe6e2d722ea36b9f3527.tar.gz libvpx-a3c2774126a00674ea70fe6e2d722ea36b9f3527.tar.bz2 libvpx-a3c2774126a00674ea70fe6e2d722ea36b9f3527.zip | |
Add vpx_highbd_avg_8x8, vpx_highbd_avg_4x4
BUG=webm:1537
Change-Id: I5f216f35436189b67d9f350991f41ed31431d4fe
Diffstat (limited to 'vpx_dsp')
| -rw-r--r-- | vpx_dsp/vpx_dsp_rtcd_defs.pl | 4 | ||||
| -rw-r--r-- | vpx_dsp/x86/avg_intrin_sse2.c | 50 |
2 files changed, 54 insertions, 0 deletions
diff --git a/vpx_dsp/vpx_dsp_rtcd_defs.pl b/vpx_dsp/vpx_dsp_rtcd_defs.pl index 9f3e268cc..0f33562cc 100644 --- a/vpx_dsp/vpx_dsp_rtcd_defs.pl +++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl @@ -985,7 +985,11 @@ if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { # Avg # add_proto qw/unsigned int vpx_highbd_avg_8x8/, "const uint8_t *, int p"; + specialize qw/vpx_highbd_avg_8x8 sse2/; + add_proto qw/unsigned int vpx_highbd_avg_4x4/, "const uint8_t *, int p"; + specialize qw/vpx_highbd_avg_4x4 sse2/; + add_proto qw/void vpx_highbd_minmax_8x8/, "const uint8_t *s, int p, const uint8_t *d, int dp, int *min, int *max"; add_proto qw/unsigned int vpx_highbd_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred"; diff --git a/vpx_dsp/x86/avg_intrin_sse2.c b/vpx_dsp/x86/avg_intrin_sse2.c index a235ba41d..0362c63c5 100644 --- a/vpx_dsp/x86/avg_intrin_sse2.c +++ b/vpx_dsp/x86/avg_intrin_sse2.c @@ -138,6 +138,56 @@ unsigned int vpx_avg_4x4_sse2(const uint8_t *s, int p) { return (avg + 8) >> 4; } +#if CONFIG_VP9_HIGHBITDEPTH +unsigned int vpx_highbd_avg_8x8_sse2(const uint8_t *s8, int p) { + __m128i s0, s1; + unsigned int avg; + const uint16_t *s = CONVERT_TO_SHORTPTR(s8); + const __m128i zero = _mm_setzero_si128(); + s0 = _mm_loadu_si128((const __m128i *)(s)); + s1 = _mm_loadu_si128((const __m128i *)(s + p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 2 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 3 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 4 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 5 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 6 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadu_si128((const __m128i *)(s + 7 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_unpackhi_epi16(s0, zero); + s0 = _mm_unpacklo_epi16(s0, zero); + s0 = _mm_add_epi32(s0, s1); + s0 = _mm_add_epi32(s0, _mm_srli_si128(s0, 8)); + s0 = _mm_add_epi32(s0, _mm_srli_si128(s0, 4)); + avg = _mm_cvtsi128_si32(s0); + + return (avg + 32) >> 6; +} + +unsigned int vpx_highbd_avg_4x4_sse2(const uint8_t *s8, int p) { + __m128i s0, s1; + unsigned int avg; + const uint16_t *s = CONVERT_TO_SHORTPTR(s8); + s0 = _mm_loadl_epi64((const __m128i *)(s)); + s1 = _mm_loadl_epi64((const __m128i *)(s + p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadl_epi64((const __m128i *)(s + 2 * p)); + s0 = _mm_adds_epu16(s0, s1); + s1 = _mm_loadl_epi64((const __m128i *)(s + 3 * p)); + s0 = _mm_adds_epu16(s0, s1); + s0 = _mm_add_epi16(s0, _mm_srli_si128(s0, 4)); + s0 = _mm_add_epi16(s0, _mm_srli_si128(s0, 2)); + avg = _mm_extract_epi16(s0, 0); + + return (avg + 8) >> 4; +} +#endif // CONFIG_VP9_HIGHBITDEPTH + static void hadamard_col8_sse2(__m128i *in, int iter) { __m128i a0 = in[0]; __m128i a1 = in[1]; |