diff options
author | Matt A. Tobin <email@mattatobin.com> | 2020-04-07 23:30:51 -0400 |
---|---|---|
committer | wolfbeast <mcwerewolf@wolfbeast.com> | 2020-04-14 13:26:42 +0200 |
commit | 277f2116b6660e9bbe7f5d67524be57eceb49b8b (patch) | |
tree | 4595f7cc71418f71b9a97dfaeb03a30aa60f336a /media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c | |
parent | d270404436f6e84ffa3b92af537ac721bf10d66e (diff) | |
download | UXP-277f2116b6660e9bbe7f5d67524be57eceb49b8b.tar UXP-277f2116b6660e9bbe7f5d67524be57eceb49b8b.tar.gz UXP-277f2116b6660e9bbe7f5d67524be57eceb49b8b.tar.lz UXP-277f2116b6660e9bbe7f5d67524be57eceb49b8b.tar.xz UXP-277f2116b6660e9bbe7f5d67524be57eceb49b8b.zip |
Move aom source to a sub-directory under media/libaom
There is no damned reason to treat this differently than any other media lib given its license and there never was.
Diffstat (limited to 'media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c')
-rw-r--r-- | media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c | 69 |
1 files changed, 69 insertions, 0 deletions
diff --git a/media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c b/media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c new file mode 100644 index 000000000..2d8f8f71e --- /dev/null +++ b/media/libaom/src/aom_dsp/x86/fwd_txfm_sse2.c @@ -0,0 +1,69 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include <emmintrin.h> // SSE2 + +#include "config/aom_config.h" +#include "config/aom_dsp_rtcd.h" + +#include "aom_dsp/aom_dsp_common.h" +#include "aom_dsp/x86/fwd_txfm_sse2.h" + +void aom_fdct8x8_1_sse2(const int16_t *input, tran_low_t *output, int stride) { + __m128i in0 = _mm_load_si128((const __m128i *)(input + 0 * stride)); + __m128i in1 = _mm_load_si128((const __m128i *)(input + 1 * stride)); + __m128i in2 = _mm_load_si128((const __m128i *)(input + 2 * stride)); + __m128i in3 = _mm_load_si128((const __m128i *)(input + 3 * stride)); + __m128i u0, u1, sum; + + u0 = _mm_add_epi16(in0, in1); + u1 = _mm_add_epi16(in2, in3); + + in0 = _mm_load_si128((const __m128i *)(input + 4 * stride)); + in1 = _mm_load_si128((const __m128i *)(input + 5 * stride)); + in2 = _mm_load_si128((const __m128i *)(input + 6 * stride)); + in3 = _mm_load_si128((const __m128i *)(input + 7 * stride)); + + sum = _mm_add_epi16(u0, u1); + + in0 = _mm_add_epi16(in0, in1); + in2 = _mm_add_epi16(in2, in3); + sum = _mm_add_epi16(sum, in0); + + u0 = _mm_setzero_si128(); + sum = _mm_add_epi16(sum, in2); + + in0 = _mm_unpacklo_epi16(u0, sum); + in1 = _mm_unpackhi_epi16(u0, sum); + in0 = _mm_srai_epi32(in0, 16); + in1 = _mm_srai_epi32(in1, 16); + + sum = _mm_add_epi32(in0, in1); + in0 = _mm_unpacklo_epi32(sum, u0); + in1 = _mm_unpackhi_epi32(sum, u0); + + sum = _mm_add_epi32(in0, in1); + in0 = _mm_srli_si128(sum, 8); + + in1 = _mm_add_epi32(sum, in0); + output[0] = (tran_low_t)_mm_cvtsi128_si32(in1); +} + +#define DCT_HIGH_BIT_DEPTH 0 +#define FDCT8x8_2D aom_fdct8x8_sse2 +#include "aom_dsp/x86/fwd_txfm_impl_sse2.h" +#undef FDCT8x8_2D + +#undef DCT_HIGH_BIT_DEPTH +#define DCT_HIGH_BIT_DEPTH 1 +#define FDCT8x8_2D aom_highbd_fdct8x8_sse2 +#include "aom_dsp/x86/fwd_txfm_impl_sse2.h" // NOLINT +#undef FDCT8x8_2D |