videolan
/
ffmpeg
mirror of https://git.videolan.org/git/ffmpeg.git
1
Fork 0
Code Releases Activity

hevc: ppc: Add HEVC 4x4 IDCT for PowerPC 

Signed-off-by: Diego Biurrun <diego@biurrun.de>
This commit is contained in:
Alexandra Hajkova 2016-12-11 12:10:19 +01:00 committed by Diego Biurrun
parent fc368497f2
commit b0e6b3f477
5 changed files with 160 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
libavcodec/hevcdsp.c
View File

@ -245,6 +245,8 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int bit_depth)
break;
}
if (ARCH_PPC)
ff_hevc_dsp_init_ppc(hevcdsp, bit_depth);
if (ARCH_X86)
ff_hevc_dsp_init_x86(hevcdsp, bit_depth);
}

1
libavcodec/hevcdsp.h
View File

@ -115,6 +115,7 @@ typedef struct HEVCDSPContext {
void ff_hevc_dsp_init(HEVCDSPContext *hpc, int bit_depth);
void ff_hevc_dsp_init_ppc(HEVCDSPContext *c, const int bit_depth);
void ff_hevc_dsp_init_x86(HEVCDSPContext *c, const int bit_depth);
extern const int16_t ff_hevc_epel_coeffs[7][16];

1
libavcodec/ppc/Makefile
View File

@ -25,6 +25,7 @@ OBJS-$(CONFIG_VP8DSP) += ppc/vp8dsp_altivec.o
# decoders/encoders
OBJS-$(CONFIG_APE_DECODER) += ppc/apedsp_altivec.o
OBJS-$(CONFIG_HEVC_DECODER) += ppc/hevcdsp.o
OBJS-$(CONFIG_SVQ1_ENCODER) += ppc/svq1enc_altivec.o
OBJS-$(CONFIG_VORBIS_DECODER) += ppc/vorbisdsp_altivec.o
OBJS-$(CONFIG_VP7_DECODER) += ppc/vp8dsp_altivec.o

108
libavcodec/ppc/hevcdsp.c Normal file
View File

@ -0,0 +1,108 @@
/* SIMD-optimized IDCT functions for HEVC decoding
* Copyright (c) Alexandra Hajkova
*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "config.h"
#if HAVE_ALTIVEC_H
#include <altivec.h>
#endif
#include "libavutil/cpu.h"
#include "libavutil/ppc/cpu.h"
#include "libavutil/ppc/types_altivec.h"
#include "libavutil/ppc/util_altivec.h"
#include "libavcodec/hevcdsp.h"
#if HAVE_ALTIVEC
static const vector int16_t trans4[4] = {
{ 64, 64, 64, 64, 64, 64, 64, 64 },
{ 83, 36, 83, 36, 83, 36, 83, 36 },
{ 64, -64, 64, -64, 64, -64, 64, -64 },
{ 36, -83, 36, -83, 36, -83, 36, -83 },
};
static const vec_u8 mask[2] = {
{ 0x00, 0x01, 0x08, 0x09, 0x10, 0x11, 0x18, 0x19, 0x02, 0x03, 0x0A, 0x0B, 0x12, 0x13, 0x1A, 0x1B },
{ 0x04, 0x05, 0x0C, 0x0D, 0x14, 0x15, 0x1C, 0x1D, 0x06, 0x07, 0x0E, 0x0F, 0x16, 0x17, 0x1E, 0x1F },
};
static void transform4x4(vector int16_t src_01, vector int16_t src_23,
vector int32_t res[4], const int shift, int16_t *coeffs)
{
vector int16_t src_02, src_13;
vector int32_t zero = vec_splat_s32(0);
vector int32_t e0, o0, e1, o1;
vector int32_t add;
src_13 = vec_mergel(src_01, src_23);
src_02 = vec_mergeh(src_01, src_23);
e0 = vec_msums(src_02, trans4[0], zero);
o0 = vec_msums(src_13, trans4[1], zero);
e1 = vec_msums(src_02, trans4[2], zero);
o1 = vec_msums(src_13, trans4[3], zero);
add = vec_sl(vec_splat_s32(1), vec_splat_u32(shift - 1));
e0 = vec_add(e0, add);
e1 = vec_add(e1, add);
res[0] = vec_add(e0, o0);
res[1] = vec_add(e1, o1);
res[2] = vec_sub(e1, o1);
res[3] = vec_sub(e0, o0);
}
static void scale(vector int32_t res[4], vector int16_t res_packed[2], int shift)
{
int i;
vector unsigned int v_shift = vec_splat_u32(shift);
for (i = 0; i < 4; i++)
res[i] = vec_sra(res[i], v_shift);
// clip16
res_packed[0] = vec_packs(res[0], res[1]);
res_packed[1] = vec_packs(res[2], res[3]);
}
#define FUNCDECL(a, depth) a ## _ ## depth ## _altivec
#define FUNC(a, b) FUNCDECL(a, b)
#define BIT_DEPTH 8
#include "hevcdsp_template.c"
#undef BIT_DEPTH
#define BIT_DEPTH 10
#include "hevcdsp_template.c"
#undef BIT_DEPTH
#endif /* HAVE_ALTIVEC */
av_cold void ff_hevc_dsp_init_ppc(HEVCDSPContext *c, const int bit_depth)
{
#if HAVE_ALTIVEC
if (!PPC_ALTIVEC(av_get_cpu_flags()))
return;
if (bit_depth == 8)
c->idct[0] = ff_hevc_idct_4x4_8_altivec;
if (bit_depth == 10)
c->idct[0] = ff_hevc_idct_4x4_10_altivec;
#endif /* HAVE_ALTIVEC */
}

48
libavcodec/ppc/hevcdsp_template.c Normal file
View File

@ -0,0 +1,48 @@
/*
* Copyright (c) Alexandra Hajkova
*
* This file is part of Libav.
*
* Libav 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.
*
* Libav 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 Libav; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
static void FUNC(ff_hevc_idct_4x4, BIT_DEPTH)(int16_t *coeffs, int col_limit)
{
const int shift = 7;
const int shift2 = 20 - BIT_DEPTH;
vector int16_t src_01, src_23;
vector int32_t res[4];
vector int16_t res_packed[2];
src_01 = vec_ld(0, coeffs);
src_23 = vec_ld(16, coeffs);
transform4x4(src_01, src_23, res, shift, coeffs);
src_01 = vec_packs(res[0], res[1]);
src_23 = vec_packs(res[2], res[3]);
scale(res, res_packed, shift);
// transpose
src_01 = vec_perm(res_packed[0], res_packed[1], mask[0]);
src_23 = vec_perm(res_packed[0], res_packed[1], mask[1]);
transform4x4(src_01, src_23, res, shift2, coeffs);
scale(res, res_packed, shift2);
// transpose
src_01 = vec_perm(res_packed[0], res_packed[1], mask[0]);
src_23 = vec_perm(res_packed[0], res_packed[1], mask[1]);
vec_st(src_01, 0, coeffs);
vec_st(src_23, 16, coeffs);
}