chromium/third_party/libaom/source/libaom/av1/common/x86/wiener_convolve_avx2.c

/*
 * Copyright (c) 2018, Alliance for Open Media. All rights reserved.
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <immintrin.h>
#include <assert.h>

#include "config/av1_rtcd.h"

#include "av1/common/convolve.h"
#include "aom_dsp/aom_dsp_common.h"
#include "aom_dsp/aom_filter.h"
#include "aom_dsp/x86/convolve_avx2.h"
#include "aom_dsp/x86/synonyms.h"
#include "aom_dsp/x86/synonyms_avx2.h"

// 128-bit xmmwords are written as [ ... ] with the MSB on the left.
// 256-bit ymmwords are written as two xmmwords, [ ... ][ ... ] with the MSB
// on the left.
// A row of, say, 8-bit pixels with values p0, p1, p2, ..., p30, p31 will be
// loaded and stored as [ p31 ... p17 p16 ][ p15 ... p1 p0 ].

// Exploiting the range of wiener filter coefficients,
// horizontal filtering can be done in 16 bit intermediate precision.
// The details are as follows :
// Consider the horizontal wiener filter coefficients of the following form :
//      [C0, C1, C2, 2^(FILTER_BITS) -2 * (C0 + C1 + C2), C2, C1, C0]
// Subtracting  2^(FILTER_BITS) from the centre tap we get the following  :
//      [C0, C1, C2,     -2 * (C0 + C1 + C2),             C2, C1, C0]
// The sum of the product "C0 * p0 + C1 * p1 + C2 * p2 -2 * (C0 + C1 + C2) * p3
// + C2 * p4 + C1 * p5 + C0 * p6" would be in the range of signed 16 bit
// precision. Finally, after rounding the above result by round_0, we multiply
// the centre pixel by 2^(FILTER_BITS - round_0) and add it to get the
// horizontal filter output.

void av1_wiener_convolve_add_src_avx2(const uint8_t *src, ptrdiff_t src_stride,
                                      uint8_t *dst, ptrdiff_t dst_stride,
                                      const int16_t *filter_x, int x_step_q4,
                                      const int16_t *filter_y, int y_step_q4,
                                      int w, int h,
                                      const WienerConvolveParams *conv_params) { … }