shithub: libvpx

ref: 7af6c6c9ca56ee62dd5de4dd73f0acae1d65ba09
dir: /vpx_dsp/x86/convolve_ssse3.h/

/*
 *  Copyright (c) 2017 The WebM project authors. All Rights Reserved.
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
 *  in the file PATENTS.  All contributing project authors may
 *  be found in the AUTHORS file in the root of the source tree.
 */

#ifndef VPX_DSP_X86_CONVOLVE_SSSE3_H_
#define VPX_DSP_X86_CONVOLVE_SSSE3_H_

#include <tmmintrin.h>  // SSSE3

#include "./vpx_config.h"

static INLINE void shuffle_filter_ssse3(const int16_t *const filter,
                                        __m128i *const f) {
  const __m128i f_values = _mm_load_si128((const __m128i *)filter);
  // pack and duplicate the filter values
  f[0] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
  f[1] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
  f[2] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
  f[3] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
}

static INLINE __m128i convolve8_8_ssse3(const __m128i *const s,
                                        const __m128i *const f) {
  // multiply 2 adjacent elements with the filter and add the result
  const __m128i k_64 = _mm_set1_epi16(1 << 6);
  const __m128i x0 = _mm_maddubs_epi16(s[0], f[0]);
  const __m128i x1 = _mm_maddubs_epi16(s[1], f[1]);
  const __m128i x2 = _mm_maddubs_epi16(s[2], f[2]);
  const __m128i x3 = _mm_maddubs_epi16(s[3], f[3]);
  // add and saturate the results together
  const __m128i min_x2x1 = _mm_min_epi16(x2, x1);
  const __m128i max_x2x1 = _mm_max_epi16(x2, x1);
  __m128i temp = _mm_adds_epi16(x0, x3);
  temp = _mm_adds_epi16(temp, min_x2x1);
  temp = _mm_adds_epi16(temp, max_x2x1);
  // round and shift by 7 bit each 16 bit
  temp = _mm_adds_epi16(temp, k_64);
  temp = _mm_srai_epi16(temp, 7);
  return temp;
}

#endif  // VPX_DSP_X86_CONVOLVE_SSSE3_H_