Skip to content

F
ffmpeg.wasm-core
Commit 92cccb7b authored by plepere's avatar plepere Committed by Michael Niedermayer
Browse files
Options

avcodec/hevc: new idct + asm 

Signed-off-by: 's avatarMichael Niedermayer <michaelni@gmx.at>
parent fa0d0fb4
Expand all Show whitespace changes
Inline Side-by-side
Showing with 398 additions and 158 deletions
libavcodec/hevc_cabac.c
View file @ 92cccb7b
......@@ -1388,8 +1388,21 @@ void ff_hevc_hls_residual_coding(HEVCContext *s, int x0, int y0,
s->hevcdsp.transform_skip(dst, coeffs, stride);
else if (lc->cu.pred_mode == MODE_INTRA && c_idx == 0 && log2_trafo_size == 2)
s->hevcdsp.transform_4x4_luma_add(dst, coeffs, stride);
else
s->hevcdsp.transform_add[log2_trafo_size-2](dst, coeffs, stride);
else {
int max_xy = FFMAX(last_significant_coeff_x, last_significant_coeff_y);
if (max_xy == 0)
s->hevcdsp.transform_dc_add[log2_trafo_size-2](dst, coeffs, stride);
else {
int col_limit = last_significant_coeff_x + last_significant_coeff_y + 4;
if (max_xy < 4)
col_limit = FFMIN(4, col_limit);
else if (max_xy < 8)
col_limit = FFMIN(8, col_limit);
else if (max_xy < 12)
col_limit = FFMIN(24, col_limit);
s->hevcdsp.transform_add[log2_trafo_size-2](dst, coeffs, stride, col_limit);
}
}
}
}
......
libavcodec/hevcdsp.c
View file @ 92cccb7b
......@@ -202,6 +202,11 @@ void ff_hevc_dsp_init(HEVCDSPContext *hevcdsp, int bit_depth)
hevcdsp->transform_add[2] = FUNC(transform_16x16_add, depth); \
hevcdsp->transform_add[3] = FUNC(transform_32x32_add, depth); \
\
hevcdsp->transform_dc_add[0] = FUNC(transform_4x4_dc_add, depth); \
hevcdsp->transform_dc_add[1] = FUNC(transform_8x8_dc_add, depth); \
hevcdsp->transform_dc_add[2] = FUNC(transform_16x16_dc_add, depth); \
hevcdsp->transform_dc_add[3] = FUNC(transform_32x32_dc_add, depth); \
\
hevcdsp->sao_band_filter[0] = FUNC(sao_band_filter_0, depth); \
hevcdsp->sao_band_filter[1] = FUNC(sao_band_filter_1, depth); \
hevcdsp->sao_band_filter[2] = FUNC(sao_band_filter_2, depth); \
......
libavcodec/hevcdsp.h
View file @ 92cccb7b
......@@ -50,7 +50,9 @@ typedef struct HEVCDSPContext {
void (*transform_skip)(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void (*transform_4x4_luma_add)(uint8_t *dst, int16_t *coeffs,
ptrdiff_t stride);
void (*transform_add[4])(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void (*transform_add[4])(uint8_t *dst, int16_t *coeffs, ptrdiff_t _stride, int col_limit);
void (*transform_dc_add[4])(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride);
void (*sao_band_filter[4])(uint8_t *dst, uint8_t *src, ptrdiff_t stride,
struct SAOParams *sao, int *borders,
......
libavcodec/hevcdsp_template.c
View file @ 92cccb7b
This diff is collapsed.
libavcodec/x86/Makefile
View file @ 92cccb7b
......@@ -92,7 +92,8 @@ YASM-OBJS-$(CONFIG_H264QPEL) += x86/h264_qpel_8bit.o \
x86/fpel.o \
x86/qpel.o
YASM-OBJS-$(CONFIG_HEVC_DECODER) += x86/hevc_mc.o \
x86/hevc_deblock.o
x86/hevc_deblock.o \
x86/hevc_idct.o
YASM-OBJS-$(CONFIG_HPELDSP) += x86/fpel.o \
x86/hpeldsp.o
YASM-OBJS-$(CONFIG_HUFFYUVDSP) += x86/huffyuvdsp.o
......
libavcodec/x86/hevc_idct.asm 0 → 100644
View file @ 92cccb7b
; /*
; * Provide SSE & MMX idct functions for HEVC decoding
; * Copyright (c) 2014 Pierre-Edouard LEPERE
; *
; * 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/x86/x86util.asm"
SECTION_RODATA
max_pixels_10: times 8 dw ((1 << 10)-1)
dc_add_10: times 4 dd ((1 << 14-10) + 1)
SECTION .text
;the idct_dc_add macros and functions were largely inspired by x264 project's code in the h264_idct.asm file
%macro DC_ADD_INIT 2
add %1w, ((1 << 14-8) + 1)
sar %1w, (15-8)
movd m0, %1
lea %1, [%2*3]
SPLATW m0, m0, 0
pxor m1, m1
psubw m1, m0
packuswb m0, m0
packuswb m1, m1
%endmacro
%macro DC_ADD_OP 4
%1 m2, [%2 ]
%1 m3, [%2+%3 ]
%1 m4, [%2+%3*2]
%1 m5, [%2+%4 ]
paddusb m2, m0
paddusb m3, m0
paddusb m4, m0
paddusb m5, m0
psubusb m2, m1
psubusb m3, m1
psubusb m4, m1
psubusb m5, m1
%1 [%2 ], m2
%1 [%2+%3 ], m3
%1 [%2+%3*2], m4
%1 [%2+%4 ], m5
%endmacro
INIT_MMX mmxext
; void ff_hevc_idct_dc_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
%if ARCH_X86_64
cglobal hevc_idct4_dc_add_8, 3, 4, 0
movsx r3, word [r1]
DC_ADD_INIT r3, r2
DC_ADD_OP movh, r0, r2, r3
RET
; void ff_hevc_idct8_dc_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_idct8_dc_add_8, 3, 4, 0
movsx r3, word [r1]
DC_ADD_INIT r3, r2
DC_ADD_OP mova, r0, r2, r3
lea r0, [r0+r2*4]
DC_ADD_OP mova, r0, r2, r3
RET
%else
; void ff_hevc_idct_dc_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_idct4_dc_add_8, 2, 3, 0
movsx r2, word [r1]
mov r1, r2m
DC_ADD_INIT r2, r1
DC_ADD_OP movh, r0, r1, r2
RET
; void ff_hevc_idct8_dc_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_idct8_dc_add_8, 2, 3, 0
movsx r2, word [r1]
mov r1, r2m
DC_ADD_INIT r2, r1
DC_ADD_OP mova, r0, r1, r2
lea r0, [r0+r1*4]
DC_ADD_OP mova, r0, r1, r2
RET
%endif
INIT_XMM sse2
; void ff_hevc_idct16_dc_add_8_mmxext(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
cglobal hevc_idct16_dc_add_8, 3, 4, 0
movsx r3, word [r1]
DC_ADD_INIT r3, r2
DC_ADD_OP mova, r0, r2, r3
lea r0, [r0+r2*4]
DC_ADD_OP mova, r0, r2, r3
lea r0, [r0+r2*4]
DC_ADD_OP mova, r0, r2, r3
lea r0, [r0+r2*4]
DC_ADD_OP mova, r0, r2, r3
RET
;-----------------------------------------------------------------------------
; void ff_hevc_idct_dc_add_10(pixel *dst, int16_t *block, int stride)
;-----------------------------------------------------------------------------
%macro IDCT_DC_ADD_OP_10 3
pxor m5, m5
%if avx_enabled
paddw m1, m0, [%1+0 ]
paddw m2, m0, [%1+%2 ]
paddw m3, m0, [%1+%2*2]
paddw m4, m0, [%1+%3 ]
%else
mova m1, [%1+0 ]
mova m2, [%1+%2 ]
mova m3, [%1+%2*2]
mova m4, [%1+%3 ]
paddw m1, m0
paddw m2, m0
paddw m3, m0
paddw m4, m0
%endif
CLIPW m1, m5, m6
CLIPW m2, m5, m6
CLIPW m3, m5, m6
CLIPW m4, m5, m6
mova [%1+0 ], m1
mova [%1+%2 ], m2
mova [%1+%2*2], m3
mova [%1+%3 ], m4
%endmacro
INIT_MMX mmxext
cglobal hevc_idct4_dc_add_10,3,3
mov r1w, [r1]
add r1w, ((1 << 4) + 1)
sar r1w, 5
movd m0, r1d
lea r1, [r2*3]
SPLATW m0, m0, 0
mova m6, [max_pixels_10]
IDCT_DC_ADD_OP_10 r0, r2, r1
RET
;-----------------------------------------------------------------------------
; void ff_hevc_idct8_dc_add_10(pixel *dst, int16_t *block, int stride)
;-----------------------------------------------------------------------------
%macro IDCT8_DC_ADD 0
cglobal hevc_idct8_dc_add_10,3,4,7
mov r1w, [r1]
add r1w, ((1 << 4) + 1)
sar r1w, 5
movd m0, r1d
lea r1, [r2*3]
SPLATW m0, m0, 0
mova m6, [max_pixels_10]
IDCT_DC_ADD_OP_10 r0, r2, r1
lea r0, [r0+r2*4]
IDCT_DC_ADD_OP_10 r0, r2, r1
RET
%endmacro
INIT_XMM sse2
IDCT8_DC_ADD
%if HAVE_AVX_EXTERNAL
INIT_XMM avx
IDCT8_DC_ADD
%endif
libavcodec/x86/hevcdsp.h
View file @ 92cccb7b
......@@ -28,6 +28,10 @@
#include <stddef.h>
#include <stdint.h>
#define idct_dc_proto(size, bitd, opt) \
void ff_hevc_idct##size##_dc_add_##bitd##_##opt(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
#define PEL_LINK(dst, idx1, idx2, idx3, name, D, opt) \
dst[idx1][idx2][idx3] = ff_hevc_put_hevc_ ## name ## _ ## D ## _##opt; \
dst ## _bi[idx1][idx2][idx3] = ff_hevc_put_hevc_bi_ ## name ## _ ## D ## _##opt; \
......@@ -119,5 +123,26 @@ QPEL_PROTOTYPES(qpel_hv, 10, sse4);
WEIGHTING_PROTOTYPES(8, sse4);
WEIGHTING_PROTOTYPES(10, sse4);
///////////////////////////////////////////////////////////////////////////////
// IDCT
///////////////////////////////////////////////////////////////////////////////
idct_dc_proto(4, 8,mmxext);
idct_dc_proto(8, 8,mmxext);
idct_dc_proto(16,8, sse2);
idct_dc_proto(32,8, sse2);
idct_dc_proto(4, 10,mmxext);
idct_dc_proto(8, 10, sse2);
idct_dc_proto(16,10, sse2);
idct_dc_proto(32,10, sse2);
idct_dc_proto(8, 10, avx);
idct_dc_proto(16,10, avx);
idct_dc_proto(32,10, avx);
#endif // AVCODEC_X86_HEVCDSP_H
libavcodec/x86/hevcdsp_init.c
View file @ 92cccb7b
......@@ -49,6 +49,48 @@ LFC_FUNCS(uint8_t, 10)
LFL_FUNCS(uint8_t, 8)
LFL_FUNCS(uint8_t, 10)
#if HAVE_SSE2_EXTERNAL
void ff_hevc_idct32_dc_add_8_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
{
ff_hevc_idct16_dc_add_8_sse2(dst, coeffs, stride);
ff_hevc_idct16_dc_add_8_sse2(dst+16, coeffs, stride);
ff_hevc_idct16_dc_add_8_sse2(dst+16*stride, coeffs, stride);
ff_hevc_idct16_dc_add_8_sse2(dst+16*stride+16, coeffs, stride);
}
void ff_hevc_idct16_dc_add_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
{
ff_hevc_idct8_dc_add_10_sse2(dst, coeffs, stride);
ff_hevc_idct8_dc_add_10_sse2(dst+16, coeffs, stride);
ff_hevc_idct8_dc_add_10_sse2(dst+8*stride, coeffs, stride);
ff_hevc_idct8_dc_add_10_sse2(dst+8*stride+16, coeffs, stride);
}
void ff_hevc_idct32_dc_add_10_sse2(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
{
ff_hevc_idct16_dc_add_10_sse2(dst, coeffs, stride);
ff_hevc_idct16_dc_add_10_sse2(dst+32, coeffs, stride);
ff_hevc_idct16_dc_add_10_sse2(dst+16*stride, coeffs, stride);
ff_hevc_idct16_dc_add_10_sse2(dst+16*stride+32, coeffs, stride);
}
#endif //HAVE_SSE2_EXTERNAL
#if HAVE_AVX_EXTERNAL
void ff_hevc_idct16_dc_add_10_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
{
ff_hevc_idct8_dc_add_10_avx(dst, coeffs, stride);
ff_hevc_idct8_dc_add_10_avx(dst+16, coeffs, stride);
ff_hevc_idct8_dc_add_10_avx(dst+8*stride, coeffs, stride);
ff_hevc_idct8_dc_add_10_avx(dst+8*stride+16, coeffs, stride);
}
void ff_hevc_idct32_dc_add_10_avx(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
{
ff_hevc_idct16_dc_add_10_avx(dst, coeffs, stride);
ff_hevc_idct16_dc_add_10_avx(dst+32, coeffs, stride);
ff_hevc_idct16_dc_add_10_avx(dst+16*stride, coeffs, stride);
ff_hevc_idct16_dc_add_10_avx(dst+16*stride+32, coeffs, stride);
}
#endif //HAVE_AVX_EXTERNAL
#define mc_rep_func(name, bitd, step, W, opt) \
void ff_hevc_put_hevc_##name##W##_##bitd##_##opt(int16_t *_dst, ptrdiff_t dststride, \
......@@ -368,9 +410,17 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
int mm_flags = av_get_cpu_flags();
if (bit_depth == 8) {
if (EXTERNAL_MMXEXT(mm_flags)) {
c->transform_dc_add[0] = ff_hevc_idct4_dc_add_8_mmxext;
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_8_mmxext;
}
if (EXTERNAL_SSE2(mm_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_8_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_8_sse2;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_8_sse2;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_8_sse2;
}
if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_8_ssse3;
......@@ -387,12 +437,20 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
QPEL_LINKS(c->put_hevc_qpel, 0, 1, qpel_h, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 8, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 8, sse4);
}
} else if (bit_depth == 10) {
if (EXTERNAL_MMXEXT(mm_flags)) {
c->transform_dc_add[0] = ff_hevc_idct4_dc_add_10_mmxext;
}
if (EXTERNAL_SSE2(mm_flags)) {
c->hevc_v_loop_filter_chroma = ff_hevc_v_loop_filter_chroma_10_sse2;
c->hevc_h_loop_filter_chroma = ff_hevc_h_loop_filter_chroma_10_sse2;
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_10_sse2;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_10_sse2;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_10_sse2;
}
if (EXTERNAL_SSSE3(mm_flags) && ARCH_X86_64) {
c->hevc_v_loop_filter_luma = ff_hevc_v_loop_filter_luma_10_ssse3;
......@@ -410,5 +468,11 @@ void ff_hevcdsp_init_x86(HEVCDSPContext *c, const int bit_depth)
QPEL_LINKS(c->put_hevc_qpel, 1, 0, qpel_v, 10, sse4);
QPEL_LINKS(c->put_hevc_qpel, 1, 1, qpel_hv, 10, sse4);
}
if (EXTERNAL_AVX(mm_flags)) {
c->transform_dc_add[1] = ff_hevc_idct8_dc_add_10_avx;
c->transform_dc_add[2] = ff_hevc_idct16_dc_add_10_avx;
c->transform_dc_add[3] = ff_hevc_idct32_dc_add_10_avx;
}
}
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment