1 /*
2 * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10 #ifndef VPX_DSP_X86_BITDEPTH_CONVERSION_AVX2_H_
11 #define VPX_DSP_X86_BITDEPTH_CONVERSION_AVX2_H_
12
13 #include <immintrin.h>
14
15 #include "./vpx_config.h"
16 #include "vpx/vpx_integer.h"
17 #include "vpx_dsp/vpx_dsp_common.h"
18
19 // Load 16 16 bit values. If the source is 32 bits then pack down with
20 // saturation.
load_tran_low(const tran_low_t * a)21 static INLINE __m256i load_tran_low(const tran_low_t *a) {
22 #if CONFIG_VP9_HIGHBITDEPTH
23 const __m256i a_low = _mm256_loadu_si256((const __m256i *)a);
24 const __m256i a_high = _mm256_loadu_si256((const __m256i *)(a + 8));
25 return _mm256_packs_epi32(a_low, a_high);
26 #else
27 return _mm256_loadu_si256((const __m256i *)a);
28 #endif
29 }
30
store_tran_low(__m256i a,tran_low_t * b)31 static INLINE void store_tran_low(__m256i a, tran_low_t *b) {
32 #if CONFIG_VP9_HIGHBITDEPTH
33 const __m256i one = _mm256_set1_epi16(1);
34 const __m256i a_hi = _mm256_mulhi_epi16(a, one);
35 const __m256i a_lo = _mm256_mullo_epi16(a, one);
36 const __m256i a_1 = _mm256_unpacklo_epi16(a_lo, a_hi);
37 const __m256i a_2 = _mm256_unpackhi_epi16(a_lo, a_hi);
38 _mm256_storeu_si256((__m256i *)b, a_1);
39 _mm256_storeu_si256((__m256i *)(b + 8), a_2);
40 #else
41 _mm256_storeu_si256((__m256i *)b, a);
42 #endif
43 }
44 #endif // VPX_DSP_X86_BITDEPTH_CONVERSION_AVX2_H_
45