chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/highbd_inv_txfm_sse2.h

/*
 *  Copyright (c) 2015 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_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE2_H_
#define VPX_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE2_H_

#include <emmintrin.h>  // SSE2

#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#include "vpx_dsp/inv_txfm.h"
#include "vpx_dsp/x86/transpose_sse2.h"
#include "vpx_dsp/x86/txfm_common_sse2.h"

// Note: There is no 64-bit bit-level shifting SIMD instruction. All
// coefficients are left shifted by 2, so that dct_const_round_shift() can be
// done by right shifting 2 bytes.

static INLINE void extend_64bit(const __m128i in,
                                __m128i *const out /*out[2]*/) { … }

static INLINE __m128i wraplow_16bit_shift4(const __m128i in0, const __m128i in1,
                                           const __m128i rounding) { … }

static INLINE __m128i wraplow_16bit_shift5(const __m128i in0, const __m128i in1,
                                           const __m128i rounding) { … }

static INLINE __m128i dct_const_round_shift_64bit(const __m128i in) { … }

static INLINE __m128i pack_4(const __m128i in0, const __m128i in1) { … }

static INLINE void abs_extend_64bit_sse2(const __m128i in,
                                         __m128i *const out /*out[2]*/,
                                         __m128i *const sign /*sign[2]*/) { … }

// Note: cospi must be non negative.
static INLINE __m128i multiply_apply_sign_sse2(const __m128i in,
                                               const __m128i sign,
                                               const __m128i cospi) { … }

// Note: c must be non negative.
static INLINE __m128i multiplication_round_shift_sse2(
    const __m128i *const in /*in[2]*/, const __m128i *const sign /*sign[2]*/,
    const int c) { … }

// Note: c must be non negative.
static INLINE __m128i multiplication_neg_round_shift_sse2(
    const __m128i *const in /*in[2]*/, const __m128i *const sign /*sign[2]*/,
    const int c) { … }

// Note: c0 and c1 must be non negative.
static INLINE void highbd_butterfly_sse2(const __m128i in0, const __m128i in1,
                                         const int c0, const int c1,
                                         __m128i *const out0,
                                         __m128i *const out1) { … }

// Note: c0 and c1 must be non negative.
static INLINE void highbd_partial_butterfly_sse2(const __m128i in, const int c0,
                                                 const int c1,
                                                 __m128i *const out0,
                                                 __m128i *const out1) { … }

// Note: c0 and c1 must be non negative.
static INLINE void highbd_partial_butterfly_neg_sse2(const __m128i in,
                                                     const int c0, const int c1,
                                                     __m128i *const out0,
                                                     __m128i *const out1) { … }

static INLINE void highbd_butterfly_cospi16_sse2(const __m128i in0,
                                                 const __m128i in1,
                                                 __m128i *const out0,
                                                 __m128i *const out1) { … }

// Only do addition and subtraction butterfly, size = 16, 32
static INLINE void highbd_add_sub_butterfly(const __m128i *in, __m128i *out,
                                            int size) { … }

static INLINE void highbd_idct8_stage4(const __m128i *const in,
                                       __m128i *const out) { … }

static INLINE void highbd_idct8x8_final_round(__m128i *const io) { … }

static INLINE void highbd_idct16_4col_stage7(const __m128i *const in,
                                             __m128i *const out) { … }

static INLINE __m128i add_clamp(const __m128i in0, const __m128i in1,
                                const int bd) { … }

static INLINE void highbd_idct_1_add_kernel(const tran_low_t *input,
                                            uint16_t *dest, int stride, int bd,
                                            const int size) { … }

static INLINE void recon_and_store_4(const __m128i in, uint16_t *const dest,
                                     const int bd) { … }

static INLINE void recon_and_store_4x2(const __m128i in, uint16_t *const dest,
                                       const int stride, const int bd) { … }

static INLINE void recon_and_store_4x4(const __m128i *const in, uint16_t *dest,
                                       const int stride, const int bd) { … }

static INLINE void recon_and_store_8(const __m128i in, uint16_t **const dest,
                                     const int stride, const int bd) { … }

static INLINE void recon_and_store_8x8(const __m128i *const in, uint16_t *dest,
                                       const int stride, const int bd) { … }

static INLINE __m128i load_pack_8_32bit(const tran_low_t *const input) { … }

static INLINE void highbd_load_pack_transpose_32bit_8x8(const tran_low_t *input,
                                                        const int stride,
                                                        __m128i *const in) { … }

static INLINE void highbd_load_transpose_32bit_8x4(const tran_low_t *input,
                                                   const int stride,
                                                   __m128i *in) { … }

static INLINE void highbd_load_transpose_32bit_4x4(const tran_low_t *input,
                                                   const int stride,
                                                   __m128i *in) { … }

static INLINE void highbd_write_buffer_8(uint16_t *dest, const __m128i in,
                                         const int bd) { … }

static INLINE void highbd_write_buffer_4(uint16_t *const dest, const __m128i in,
                                         const int bd) { … }

#endif  // VPX_VPX_DSP_X86_HIGHBD_INV_TXFM_SSE2_H_