Line data    Source code

       1             : /*
       2             :  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
       3             :  *
       4             :  * This source code is subject to the terms of the BSD 2 Clause License and
       5             :  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
       6             :  * was not distributed with this source code in the LICENSE file, you can
       7             :  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
       8             :  * Media Patent License 1.0 was not distributed with this source code in the
       9             :  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
      10             :  */
      11             : 
      12             : #ifndef AOM_DSP_X86_SYNONYMS_H_
      13             : #define AOM_DSP_X86_SYNONYMS_H_
      14             : 
      15             : #include <immintrin.h>
      16             : 
      17             : #include "./aom_config.h"
      18             : #include "aom/aom_integer.h"
      19             : 
      20             : /**
      21             :  * Various reusable shorthands for x86 SIMD intrinsics.
      22             :  *
      23             :  * Intrinsics prefixed with xx_ operate on or return 128bit XMM registers.
      24             :  * Intrinsics prefixed with yy_ operate on or return 256bit YMM registers.
      25             :  */
      26             : 
      27             : // Loads and stores to do away with the tedium of casting the address
      28             : // to the right type.
      29           0 : static INLINE __m128i xx_loadl_32(const void *a) {
      30           0 :   return _mm_cvtsi32_si128(*(const uint32_t *)a);
      31             : }
      32             : 
      33           0 : static INLINE __m128i xx_loadl_64(const void *a) {
      34           0 :   return _mm_loadl_epi64((const __m128i *)a);
      35             : }
      36             : 
      37           0 : static INLINE __m128i xx_load_128(const void *a) {
      38           0 :   return _mm_load_si128((const __m128i *)a);
      39             : }
      40             : 
      41           0 : static INLINE __m128i xx_loadu_128(const void *a) {
      42           0 :   return _mm_loadu_si128((const __m128i *)a);
      43             : }
      44             : 
      45           0 : static INLINE void xx_storel_32(void *const a, const __m128i v) {
      46           0 :   *(uint32_t *)a = _mm_cvtsi128_si32(v);
      47           0 : }
      48             : 
      49           0 : static INLINE void xx_storel_64(void *const a, const __m128i v) {
      50             :   _mm_storel_epi64((__m128i *)a, v);
      51           0 : }
      52             : 
      53           0 : static INLINE void xx_store_128(void *const a, const __m128i v) {
      54             :   _mm_store_si128((__m128i *)a, v);
      55           0 : }
      56             : 
      57           0 : static INLINE void xx_storeu_128(void *const a, const __m128i v) {
      58             :   _mm_storeu_si128((__m128i *)a, v);
      59           0 : }
      60             : 
      61           0 : static INLINE __m128i xx_round_epu16(__m128i v_val_w) {
      62           0 :   return _mm_avg_epu16(v_val_w, _mm_setzero_si128());
      63             : }
      64             : 
      65           0 : static INLINE __m128i xx_roundn_epu16(__m128i v_val_w, int bits) {
      66           0 :   const __m128i v_s_w = _mm_srli_epi16(v_val_w, bits - 1);
      67           0 :   return _mm_avg_epu16(v_s_w, _mm_setzero_si128());
      68             : }
      69             : 
      70           0 : static INLINE __m128i xx_roundn_epu32(__m128i v_val_d, int bits) {
      71           0 :   const __m128i v_bias_d = _mm_set1_epi32((1 << bits) >> 1);
      72           0 :   const __m128i v_tmp_d = _mm_add_epi32(v_val_d, v_bias_d);
      73           0 :   return _mm_srli_epi32(v_tmp_d, bits);
      74             : }
      75             : 
      76             : // This is equivalent to ROUND_POWER_OF_TWO(v_val_d, bits)
      77           0 : static INLINE __m128i xx_roundn_epi32_unsigned(__m128i v_val_d, int bits) {
      78           0 :   const __m128i v_bias_d = _mm_set1_epi32((1 << bits) >> 1);
      79           0 :   const __m128i v_tmp_d = _mm_add_epi32(v_val_d, v_bias_d);
      80           0 :   return _mm_srai_epi32(v_tmp_d, bits);
      81             : }
      82             : 
      83             : // This is equivalent to ROUND_POWER_OF_TWO_SIGNED(v_val_d, bits)
      84           0 : static INLINE __m128i xx_roundn_epi32(__m128i v_val_d, int bits) {
      85           0 :   const __m128i v_bias_d = _mm_set1_epi32((1 << bits) >> 1);
      86           0 :   const __m128i v_sign_d = _mm_srai_epi32(v_val_d, 31);
      87           0 :   const __m128i v_tmp_d =
      88           0 :       _mm_add_epi32(_mm_add_epi32(v_val_d, v_bias_d), v_sign_d);
      89           0 :   return _mm_srai_epi32(v_tmp_d, bits);
      90             : }
      91             : 
      92             : #endif  // AOM_DSP_X86_SYNONYMS_H_