xref: /aosp_15_r20/external/libaom/aom_dsp/x86/intrapred_x86.h (revision 77c1e3ccc04c968bd2bc212e87364f250e820521)
1*77c1e3ccSAndroid Build Coastguard Worker /*
2*77c1e3ccSAndroid Build Coastguard Worker  * Copyright (c) 2020, Alliance for Open Media. All rights reserved.
3*77c1e3ccSAndroid Build Coastguard Worker  *
4*77c1e3ccSAndroid Build Coastguard Worker  * This source code is subject to the terms of the BSD 2 Clause License and
5*77c1e3ccSAndroid Build Coastguard Worker  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6*77c1e3ccSAndroid Build Coastguard Worker  * was not distributed with this source code in the LICENSE file, you can
7*77c1e3ccSAndroid Build Coastguard Worker  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8*77c1e3ccSAndroid Build Coastguard Worker  * Media Patent License 1.0 was not distributed with this source code in the
9*77c1e3ccSAndroid Build Coastguard Worker  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10*77c1e3ccSAndroid Build Coastguard Worker  */
11*77c1e3ccSAndroid Build Coastguard Worker 
12*77c1e3ccSAndroid Build Coastguard Worker #ifndef AOM_AOM_DSP_X86_INTRAPRED_X86_H_
13*77c1e3ccSAndroid Build Coastguard Worker #define AOM_AOM_DSP_X86_INTRAPRED_X86_H_
14*77c1e3ccSAndroid Build Coastguard Worker 
15*77c1e3ccSAndroid Build Coastguard Worker #include <emmintrin.h>  // SSE2
16*77c1e3ccSAndroid Build Coastguard Worker #include "aom/aom_integer.h"
17*77c1e3ccSAndroid Build Coastguard Worker #include "config/aom_config.h"
18*77c1e3ccSAndroid Build Coastguard Worker 
dc_sum_16_sse2(const uint8_t * ref)19*77c1e3ccSAndroid Build Coastguard Worker static inline __m128i dc_sum_16_sse2(const uint8_t *ref) {
20*77c1e3ccSAndroid Build Coastguard Worker   __m128i x = _mm_load_si128((__m128i const *)ref);
21*77c1e3ccSAndroid Build Coastguard Worker   const __m128i zero = _mm_setzero_si128();
22*77c1e3ccSAndroid Build Coastguard Worker   x = _mm_sad_epu8(x, zero);
23*77c1e3ccSAndroid Build Coastguard Worker   const __m128i high = _mm_unpackhi_epi64(x, x);
24*77c1e3ccSAndroid Build Coastguard Worker   return _mm_add_epi16(x, high);
25*77c1e3ccSAndroid Build Coastguard Worker }
26*77c1e3ccSAndroid Build Coastguard Worker 
dc_sum_32_sse2(const uint8_t * ref)27*77c1e3ccSAndroid Build Coastguard Worker static inline __m128i dc_sum_32_sse2(const uint8_t *ref) {
28*77c1e3ccSAndroid Build Coastguard Worker   __m128i x0 = _mm_load_si128((__m128i const *)ref);
29*77c1e3ccSAndroid Build Coastguard Worker   __m128i x1 = _mm_load_si128((__m128i const *)(ref + 16));
30*77c1e3ccSAndroid Build Coastguard Worker   const __m128i zero = _mm_setzero_si128();
31*77c1e3ccSAndroid Build Coastguard Worker   x0 = _mm_sad_epu8(x0, zero);
32*77c1e3ccSAndroid Build Coastguard Worker   x1 = _mm_sad_epu8(x1, zero);
33*77c1e3ccSAndroid Build Coastguard Worker   x0 = _mm_add_epi16(x0, x1);
34*77c1e3ccSAndroid Build Coastguard Worker   const __m128i high = _mm_unpackhi_epi64(x0, x0);
35*77c1e3ccSAndroid Build Coastguard Worker   return _mm_add_epi16(x0, high);
36*77c1e3ccSAndroid Build Coastguard Worker }
37*77c1e3ccSAndroid Build Coastguard Worker 
38*77c1e3ccSAndroid Build Coastguard Worker #endif  // AOM_AOM_DSP_X86_INTRAPRED_X86_H_
39