]> git.sesse.net Git - ffmpeg/commitdiff
Merge commit '89d9869d2491d4209d707a8e7f29c58227ae5a4e'
authorJames Almer <jamrial@gmail.com>
Fri, 27 Oct 2017 21:22:39 +0000 (18:22 -0300)
committerJames Almer <jamrial@gmail.com>
Fri, 27 Oct 2017 21:22:39 +0000 (18:22 -0300)
* commit '89d9869d2491d4209d707a8e7f29c58227ae5a4e':
  hevc: Add NEON 16x16 IDCT

Merged-by: James Almer <jamrial@gmail.com>
1  2 
libavcodec/arm/hevcdsp_idct_neon.S
libavcodec/arm/hevcdsp_init_neon.c

Simple merge
index 8bc430e0121d0c6b7ce4791932363208300717f2,0000000000000000000000000000000000000000..562ebf96d129e13cbe9f018d433aa2f1eb8be35f
mode 100644,000000..100644
--- /dev/null
@@@ -1,231 -1,0 +1,235 @@@
 +/*
 + * Copyright (c) 2014 Seppo Tomperi <seppo.tomperi@vtt.fi>
 + *
 + * This file is part of FFmpeg.
 + *
 + * FFmpeg is free software; you can redistribute it and/or
 + * modify it under the terms of the GNU Lesser General Public
 + * License as published by the Free Software Foundation; either
 + * version 2.1 of the License, or (at your option) any later version.
 + *
 + * FFmpeg is distributed in the hope that it will be useful,
 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 + * Lesser General Public License for more details.
 + *
 + * You should have received a copy of the GNU Lesser General Public
 + * License along with FFmpeg; if not, write to the Free Software
 + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 + */
 +
 +#include "libavutil/attributes.h"
 +#include "libavutil/arm/cpu.h"
 +#include "libavcodec/hevcdsp.h"
 +#include "hevcdsp_arm.h"
 +
 +void ff_hevc_v_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
 +void ff_hevc_h_loop_filter_luma_neon(uint8_t *_pix, ptrdiff_t _stride, int _beta, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
 +void ff_hevc_v_loop_filter_chroma_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
 +void ff_hevc_h_loop_filter_chroma_neon(uint8_t *_pix, ptrdiff_t _stride, int *_tc, uint8_t *_no_p, uint8_t *_no_q);
 +void ff_hevc_idct_4x4_8_neon(int16_t *coeffs, int col_limit);
 +void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
++void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
 +void ff_hevc_idct_4x4_10_neon(int16_t *coeffs, int col_limit);
 +void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
++void ff_hevc_idct_16x16_10_neon(int16_t *coeffs, int col_limit);
 +void ff_hevc_idct_4x4_dc_neon_8(int16_t *coeffs);
 +void ff_hevc_idct_8x8_dc_neon_8(int16_t *coeffs);
 +void ff_hevc_idct_16x16_dc_neon_8(int16_t *coeffs);
 +void ff_hevc_idct_32x32_dc_neon_8(int16_t *coeffs);
 +void ff_hevc_transform_luma_4x4_neon_8(int16_t *coeffs);
 +void ff_hevc_add_residual_4x4_neon_8(uint8_t *_dst, int16_t *coeffs,
 +                                     ptrdiff_t stride);
 +void ff_hevc_add_residual_8x8_neon_8(uint8_t *_dst, int16_t *coeffs,
 +                                     ptrdiff_t stride);
 +void ff_hevc_add_residual_16x16_neon_8(uint8_t *_dst, int16_t *coeffs,
 +                                       ptrdiff_t stride);
 +void ff_hevc_add_residual_32x32_neon_8(uint8_t *_dst, int16_t *coeffs,
 +                                       ptrdiff_t stride);
 +
 +#define PUT_PIXELS(name) \
 +    void name(int16_t *dst, uint8_t *src, \
 +                                ptrdiff_t srcstride, int height, \
 +                                intptr_t mx, intptr_t my, int width)
 +PUT_PIXELS(ff_hevc_put_pixels_w2_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w4_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w6_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w8_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w12_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w16_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w24_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w32_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w48_neon_8);
 +PUT_PIXELS(ff_hevc_put_pixels_w64_neon_8);
 +#undef PUT_PIXELS
 +
 +static void (*put_hevc_qpel_neon[4][4])(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
 +                                   int height, int width);
 +static void (*put_hevc_qpel_uw_neon[4][4])(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride,
 +                                   int width, int height, int16_t* src2, ptrdiff_t src2stride);
 +void ff_hevc_put_qpel_neon_wrapper(int16_t *dst, uint8_t *src, ptrdiff_t srcstride,
 +                                   int height, intptr_t mx, intptr_t my, int width);
 +void ff_hevc_put_qpel_uni_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
 +                                   int height, intptr_t mx, intptr_t my, int width);
 +void ff_hevc_put_qpel_bi_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
 +                                       int16_t *src2,
 +                                       int height, intptr_t mx, intptr_t my, int width);
 +#define QPEL_FUNC(name) \
 +    void name(int16_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride, \
 +                                   int height, int width)
 +
 +QPEL_FUNC(ff_hevc_put_qpel_v1_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_v2_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_v3_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h1_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h2_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h3_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h1v1_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h1v2_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h1v3_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h2v1_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h2v2_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h2v3_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h3v1_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h3v2_neon_8);
 +QPEL_FUNC(ff_hevc_put_qpel_h3v3_neon_8);
 +#undef QPEL_FUNC
 +
 +#define QPEL_FUNC_UW_PIX(name) \
 +    void name(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride, \
 +                                   int height, intptr_t mx, intptr_t my, int width);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w4_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w8_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w16_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w24_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w32_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w48_neon_8);
 +QPEL_FUNC_UW_PIX(ff_hevc_put_qpel_uw_pixels_w64_neon_8);
 +#undef QPEL_FUNC_UW_PIX
 +
 +#define QPEL_FUNC_UW(name) \
 +    void name(uint8_t *dst, ptrdiff_t dststride, uint8_t *_src, ptrdiff_t _srcstride, \
 +                                   int width, int height, int16_t* src2, ptrdiff_t src2stride);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_pixels_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v1_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v2_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_v3_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v1_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v2_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h1v3_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v1_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v2_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h2v3_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v1_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v2_neon_8);
 +QPEL_FUNC_UW(ff_hevc_put_qpel_uw_h3v3_neon_8);
 +#undef QPEL_FUNC_UW
 +
 +void ff_hevc_put_qpel_neon_wrapper(int16_t *dst, uint8_t *src, ptrdiff_t srcstride,
 +                                   int height, intptr_t mx, intptr_t my, int width) {
 +
 +    put_hevc_qpel_neon[my][mx](dst, MAX_PB_SIZE, src, srcstride, height, width);
 +}
 +
 +void ff_hevc_put_qpel_uni_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
 +                                   int height, intptr_t mx, intptr_t my, int width) {
 +
 +    put_hevc_qpel_uw_neon[my][mx](dst, dststride, src, srcstride, width, height, NULL, 0);
 +}
 +
 +void ff_hevc_put_qpel_bi_neon_wrapper(uint8_t *dst, ptrdiff_t dststride, uint8_t *src, ptrdiff_t srcstride,
 +                                       int16_t *src2,
 +                                       int height, intptr_t mx, intptr_t my, int width) {
 +    put_hevc_qpel_uw_neon[my][mx](dst, dststride, src, srcstride, width, height, src2, MAX_PB_SIZE);
 +}
 +
 +av_cold void ff_hevc_dsp_init_neon(HEVCDSPContext *c, const int bit_depth)
 +{
 +    if (bit_depth == 8) {
 +        int x;
 +        c->hevc_v_loop_filter_luma     = ff_hevc_v_loop_filter_luma_neon;
 +        c->hevc_h_loop_filter_luma     = ff_hevc_h_loop_filter_luma_neon;
 +        c->hevc_v_loop_filter_chroma   = ff_hevc_v_loop_filter_chroma_neon;
 +        c->hevc_h_loop_filter_chroma   = ff_hevc_h_loop_filter_chroma_neon;
 +        c->idct[0]                     = ff_hevc_idct_4x4_8_neon;
 +        c->idct[1]                     = ff_hevc_idct_8x8_8_neon;
++        c->idct[2]                     = ff_hevc_idct_16x16_8_neon;
 +        c->idct_dc[0]                  = ff_hevc_idct_4x4_dc_neon_8;
 +        c->idct_dc[1]                  = ff_hevc_idct_8x8_dc_neon_8;
 +        c->idct_dc[2]                  = ff_hevc_idct_16x16_dc_neon_8;
 +        c->idct_dc[3]                  = ff_hevc_idct_32x32_dc_neon_8;
 +        c->add_residual[0]             = ff_hevc_add_residual_4x4_neon_8;
 +        c->add_residual[1]             = ff_hevc_add_residual_8x8_neon_8;
 +        c->add_residual[2]             = ff_hevc_add_residual_16x16_neon_8;
 +        c->add_residual[3]             = ff_hevc_add_residual_32x32_neon_8;
 +        c->transform_4x4_luma          = ff_hevc_transform_luma_4x4_neon_8;
 +        put_hevc_qpel_neon[1][0]       = ff_hevc_put_qpel_v1_neon_8;
 +        put_hevc_qpel_neon[2][0]       = ff_hevc_put_qpel_v2_neon_8;
 +        put_hevc_qpel_neon[3][0]       = ff_hevc_put_qpel_v3_neon_8;
 +        put_hevc_qpel_neon[0][1]       = ff_hevc_put_qpel_h1_neon_8;
 +        put_hevc_qpel_neon[0][2]       = ff_hevc_put_qpel_h2_neon_8;
 +        put_hevc_qpel_neon[0][3]       = ff_hevc_put_qpel_h3_neon_8;
 +        put_hevc_qpel_neon[1][1]       = ff_hevc_put_qpel_h1v1_neon_8;
 +        put_hevc_qpel_neon[1][2]       = ff_hevc_put_qpel_h2v1_neon_8;
 +        put_hevc_qpel_neon[1][3]       = ff_hevc_put_qpel_h3v1_neon_8;
 +        put_hevc_qpel_neon[2][1]       = ff_hevc_put_qpel_h1v2_neon_8;
 +        put_hevc_qpel_neon[2][2]       = ff_hevc_put_qpel_h2v2_neon_8;
 +        put_hevc_qpel_neon[2][3]       = ff_hevc_put_qpel_h3v2_neon_8;
 +        put_hevc_qpel_neon[3][1]       = ff_hevc_put_qpel_h1v3_neon_8;
 +        put_hevc_qpel_neon[3][2]       = ff_hevc_put_qpel_h2v3_neon_8;
 +        put_hevc_qpel_neon[3][3]       = ff_hevc_put_qpel_h3v3_neon_8;
 +        put_hevc_qpel_uw_neon[1][0]      = ff_hevc_put_qpel_uw_v1_neon_8;
 +        put_hevc_qpel_uw_neon[2][0]      = ff_hevc_put_qpel_uw_v2_neon_8;
 +        put_hevc_qpel_uw_neon[3][0]      = ff_hevc_put_qpel_uw_v3_neon_8;
 +        put_hevc_qpel_uw_neon[0][1]      = ff_hevc_put_qpel_uw_h1_neon_8;
 +        put_hevc_qpel_uw_neon[0][2]      = ff_hevc_put_qpel_uw_h2_neon_8;
 +        put_hevc_qpel_uw_neon[0][3]      = ff_hevc_put_qpel_uw_h3_neon_8;
 +        put_hevc_qpel_uw_neon[1][1]      = ff_hevc_put_qpel_uw_h1v1_neon_8;
 +        put_hevc_qpel_uw_neon[1][2]      = ff_hevc_put_qpel_uw_h2v1_neon_8;
 +        put_hevc_qpel_uw_neon[1][3]      = ff_hevc_put_qpel_uw_h3v1_neon_8;
 +        put_hevc_qpel_uw_neon[2][1]      = ff_hevc_put_qpel_uw_h1v2_neon_8;
 +        put_hevc_qpel_uw_neon[2][2]      = ff_hevc_put_qpel_uw_h2v2_neon_8;
 +        put_hevc_qpel_uw_neon[2][3]      = ff_hevc_put_qpel_uw_h3v2_neon_8;
 +        put_hevc_qpel_uw_neon[3][1]      = ff_hevc_put_qpel_uw_h1v3_neon_8;
 +        put_hevc_qpel_uw_neon[3][2]      = ff_hevc_put_qpel_uw_h2v3_neon_8;
 +        put_hevc_qpel_uw_neon[3][3]      = ff_hevc_put_qpel_uw_h3v3_neon_8;
 +        for (x = 0; x < 10; x++) {
 +            c->put_hevc_qpel[x][1][0]         = ff_hevc_put_qpel_neon_wrapper;
 +            c->put_hevc_qpel[x][0][1]         = ff_hevc_put_qpel_neon_wrapper;
 +            c->put_hevc_qpel[x][1][1]         = ff_hevc_put_qpel_neon_wrapper;
 +            c->put_hevc_qpel_uni[x][1][0]     = ff_hevc_put_qpel_uni_neon_wrapper;
 +            c->put_hevc_qpel_uni[x][0][1]     = ff_hevc_put_qpel_uni_neon_wrapper;
 +            c->put_hevc_qpel_uni[x][1][1]     = ff_hevc_put_qpel_uni_neon_wrapper;
 +            c->put_hevc_qpel_bi[x][1][0]      = ff_hevc_put_qpel_bi_neon_wrapper;
 +            c->put_hevc_qpel_bi[x][0][1]      = ff_hevc_put_qpel_bi_neon_wrapper;
 +            c->put_hevc_qpel_bi[x][1][1]      = ff_hevc_put_qpel_bi_neon_wrapper;
 +        }
 +        c->put_hevc_qpel[0][0][0]  = ff_hevc_put_pixels_w2_neon_8;
 +        c->put_hevc_qpel[1][0][0]  = ff_hevc_put_pixels_w4_neon_8;
 +        c->put_hevc_qpel[2][0][0]  = ff_hevc_put_pixels_w6_neon_8;
 +        c->put_hevc_qpel[3][0][0]  = ff_hevc_put_pixels_w8_neon_8;
 +        c->put_hevc_qpel[4][0][0]  = ff_hevc_put_pixels_w12_neon_8;
 +        c->put_hevc_qpel[5][0][0]  = ff_hevc_put_pixels_w16_neon_8;
 +        c->put_hevc_qpel[6][0][0]  = ff_hevc_put_pixels_w24_neon_8;
 +        c->put_hevc_qpel[7][0][0]  = ff_hevc_put_pixels_w32_neon_8;
 +        c->put_hevc_qpel[8][0][0]  = ff_hevc_put_pixels_w48_neon_8;
 +        c->put_hevc_qpel[9][0][0]  = ff_hevc_put_pixels_w64_neon_8;
 +
 +        c->put_hevc_qpel_uni[1][0][0]  = ff_hevc_put_qpel_uw_pixels_w4_neon_8;
 +        c->put_hevc_qpel_uni[3][0][0]  = ff_hevc_put_qpel_uw_pixels_w8_neon_8;
 +        c->put_hevc_qpel_uni[5][0][0]  = ff_hevc_put_qpel_uw_pixels_w16_neon_8;
 +        c->put_hevc_qpel_uni[6][0][0]  = ff_hevc_put_qpel_uw_pixels_w24_neon_8;
 +        c->put_hevc_qpel_uni[7][0][0]  = ff_hevc_put_qpel_uw_pixels_w32_neon_8;
 +        c->put_hevc_qpel_uni[8][0][0]  = ff_hevc_put_qpel_uw_pixels_w48_neon_8;
 +        c->put_hevc_qpel_uni[9][0][0]  = ff_hevc_put_qpel_uw_pixels_w64_neon_8;
 +    }
 +
 +    if (bit_depth == 10) {
 +        c->idct[0] = ff_hevc_idct_4x4_10_neon;
 +        c->idct[1] = ff_hevc_idct_8x8_10_neon;
++        c->idct[2] = ff_hevc_idct_16x16_10_neon;
 +    }
 +}