diff options
| author | Johann Koenig <johannkoenig@google.com> | 2017-02-16 02:40:59 +0000 |
|---|---|---|
| committer | Gerrit Code Review <noreply-gerritcodereview@google.com> | 2017-02-16 02:41:00 +0000 |
| commit | b63e88e506a28b0e2814259c1af04ef319dac77b (patch) | |
| tree | edee188c85d4462cb9feb5abb7754cc84806660c | |
| parent | 61d05c1e67371c7468b6e4dbc414bf0c16414e6d (diff) | |
| parent | 327a02d77ef45932c7cce39fca53765b8acc158d (diff) | |
| download | libvpx-b63e88e506a28b0e2814259c1af04ef319dac77b.tar libvpx-b63e88e506a28b0e2814259c1af04ef319dac77b.tar.gz libvpx-b63e88e506a28b0e2814259c1af04ef319dac77b.tar.bz2 libvpx-b63e88e506a28b0e2814259c1af04ef319dac77b.zip | |
Merge "Use 'packssdw' for loading tran_low_t values"
| -rw-r--r-- | vpx_dsp/x86/bitdepth_conversion_sse2.h | 9 |
1 files changed, 4 insertions, 5 deletions
diff --git a/vpx_dsp/x86/bitdepth_conversion_sse2.h b/vpx_dsp/x86/bitdepth_conversion_sse2.h index bff62a431..c5a29efbe 100644 --- a/vpx_dsp/x86/bitdepth_conversion_sse2.h +++ b/vpx_dsp/x86/bitdepth_conversion_sse2.h @@ -16,13 +16,12 @@ #include "vpx/vpx_integer.h" #include "vpx_dsp/vpx_dsp_common.h" -// Load 8 16 bit values. If the source is 32 bits then cast down. -// This does not saturate values. It only truncates. +// Load 8 16 bit values. If the source is 32 bits then pack down with +// saturation. static INLINE __m128i load_tran_low(const tran_low_t *a) { #if CONFIG_VP9_HIGHBITDEPTH - return _mm_setr_epi16((int16_t)a[0], (int16_t)a[1], (int16_t)a[2], - (int16_t)a[3], (int16_t)a[4], (int16_t)a[5], - (int16_t)a[6], (int16_t)a[7]); + const __m128i a_low = _mm_load_si128((const __m128i *)a); + return _mm_packs_epi32(a_low, *(const __m128i *)(a + 4)); #else return _mm_load_si128((const __m128i *)a); #endif |