Line data    Source code

       1             : /*
       2             :  *  Copyright (c) 2016 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_FDCT_H_
      11             : #define VPX_DSP_X86_FDCT_H_
      12             : 
      13             : #include <xmmintrin.h>
      14             : 
      15             : #include "./vpx_config.h"
      16             : #include "vpx/vpx_integer.h"
      17             : #include "vpx_dsp/vpx_dsp_common.h"
      18             : 
      19             : // Load 8 16 bit values. If the source is 32 bits then cast down.
      20             : // This does not saturate values. It only truncates.
      21           0 : static INLINE __m128i load_tran_low(const tran_low_t *a) {
      22             : #if CONFIG_VP9_HIGHBITDEPTH
      23             :   return _mm_setr_epi16((int16_t)a[0], (int16_t)a[1], (int16_t)a[2],
      24             :                         (int16_t)a[3], (int16_t)a[4], (int16_t)a[5],
      25             :                         (int16_t)a[6], (int16_t)a[7]);
      26             : #else
      27           0 :   return _mm_load_si128((const __m128i *)a);
      28             : #endif
      29             : }
      30             : 
      31             : // Store 8 16 bit values. If the destination is 32 bits then sign extend the
      32             : // values by multiplying by 1.
      33           0 : static INLINE void store_tran_low(__m128i a, tran_low_t *b) {
      34             : #if CONFIG_VP9_HIGHBITDEPTH
      35             :   const __m128i one = _mm_set1_epi16(1);
      36             :   const __m128i a_hi = _mm_mulhi_epi16(a, one);
      37             :   const __m128i a_lo = _mm_mullo_epi16(a, one);
      38             :   const __m128i a_1 = _mm_unpacklo_epi16(a_lo, a_hi);
      39             :   const __m128i a_2 = _mm_unpackhi_epi16(a_lo, a_hi);
      40             :   _mm_store_si128((__m128i *)(b), a_1);
      41             :   _mm_store_si128((__m128i *)(b + 4), a_2);
      42             : #else
      43             :   _mm_store_si128((__m128i *)(b), a);
      44             : #endif
      45           0 : }
      46             : 
      47             : // Zero fill 8 positions in the output buffer.
      48           0 : static INLINE void store_zero_tran_low(tran_low_t *a) {
      49           0 :   const __m128i zero = _mm_setzero_si128();
      50             : #if CONFIG_VP9_HIGHBITDEPTH
      51             :   _mm_store_si128((__m128i *)(a), zero);
      52             :   _mm_store_si128((__m128i *)(a + 4), zero);
      53             : #else
      54             :   _mm_store_si128((__m128i *)(a), zero);
      55             : #endif
      56           0 : }
      57             : #endif  // VPX_DSP_X86_FDCT_H_