From 059715a41cf67834c1ce153caa245c394abb82e0 Mon Sep 17 00:00:00 2001 From: Robert Edele Date: Thu, 16 Mar 2006 19:18:18 +0000 Subject: First part of a series of speed-enchancing patches. This one sets up a snow.h and makes snow use the dsputil function pointer framework to access the three functions that will be implemented in asm in the other parts of the patchset. Patch by Robert Edele < yartrebo AH earthlink POIS net> Original thread: Subject: [Ffmpeg-devel] [PATCH] Snow mmx+sse2 asm optimizations Date: Sun, 05 Feb 2006 12:47:14 -0500 Originally committed as revision 5172 to svn://svn.ffmpeg.org/ffmpeg/trunk --- libavcodec/dsputil.c | 5 ++ libavcodec/dsputil.h | 9 +++ libavcodec/i386/mmx.h | 2 + libavcodec/snow.c | 184 ++++++++++++++------------------------------------ libavcodec/snow.h | 123 +++++++++++++++++++++++++++++++++ 5 files changed, 190 insertions(+), 133 deletions(-) create mode 100644 libavcodec/snow.h diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index 5a518f4c8c..6bc59db9fa 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -30,6 +30,7 @@ #include "mpegvideo.h" #include "simple_idct.h" #include "faandct.h" +#include "snow.h" /* snow.c */ void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count); @@ -4047,6 +4048,10 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx) c->try_8x8basis= try_8x8basis_c; c->add_8x8basis= add_8x8basis_c; + c->vertical_compose97i = ff_snow_vertical_compose97i; + c->horizontal_compose97i = ff_snow_horizontal_compose97i; + c->inner_add_yblock = ff_snow_inner_add_yblock; + #ifdef HAVE_MMX dsputil_init_mmx(c, avctx); #endif diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index e165eede2d..8fe9be7798 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -35,6 +35,7 @@ //#define DEBUG /* dct code */ typedef short DCTELEM; +typedef int DWTELEM; void fdct_ifast (DCTELEM *data); void fdct_ifast248 (DCTELEM *data); @@ -133,6 +134,9 @@ static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; +// for snow slices +typedef struct slice_buffer_s slice_buffer; + /** * DSPContext. */ @@ -334,6 +338,11 @@ typedef struct DSPContext { void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); + + /* snow wavelet */ + void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width); + void (*horizontal_compose97i)(DWTELEM *b, int width); + void (*inner_add_yblock)(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); } DSPContext; void dsputil_static_init(void); diff --git a/libavcodec/i386/mmx.h b/libavcodec/i386/mmx.h index df4620e0ab..df1791823b 100644 --- a/libavcodec/i386/mmx.h +++ b/libavcodec/i386/mmx.h @@ -12,6 +12,7 @@ # define REG_d "rdx" # define REG_D "rdi" # define REG_S "rsi" +# define PTR_SIZE "8" #else # define REG_a "eax" # define REG_b "ebx" @@ -19,6 +20,7 @@ # define REG_d "edx" # define REG_D "edi" # define REG_S "esi" +# define PTR_SIZE "4" #endif /* diff --git a/libavcodec/snow.c b/libavcodec/snow.c index b22d8b8a74..8fdc956c2b 100644 --- a/libavcodec/snow.c +++ b/libavcodec/snow.c @@ -19,23 +19,15 @@ #include "avcodec.h" #include "common.h" #include "dsputil.h" +#include "snow.h" #include "rangecoder.h" -#define MID_STATE 128 #include "mpegvideo.h" #undef NDEBUG #include -#define MAX_DECOMPOSITIONS 8 -#define MAX_PLANES 4 -#define DWTELEM int -#define QSHIFT 5 -#define QROOT (1<>shift), inverse); } } +#endif +#ifndef liftS static always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int dst_step, int src_step, int ref_step, int width, int mul, int add, int shift, int highpass, int inverse){ const int mirror_left= !highpass; const int mirror_right= (width&1) ^ highpass; @@ -793,6 +775,7 @@ static always_inline void liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref, int ds dst[w*dst_step] = LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse); } } +#endif static void inplace_lift(DWTELEM *dst, int width, int *coeffs, int n, int shift, int start, int inverse){ @@ -1111,76 +1094,6 @@ STOP_TIMER("vertical_decompose53i*")} } } -#define liftS lift -#define lift5 lift -#if 1 -#define W_AM 3 -#define W_AO 0 -#define W_AS 1 - -#undef liftS -#define W_BM 1 -#define W_BO 8 -#define W_BS 4 - -#define W_CM 1 -#define W_CO 0 -#define W_CS 0 - -#define W_DM 3 -#define W_DO 4 -#define W_DS 3 -#elif 0 -#define W_AM 55 -#define W_AO 16 -#define W_AS 5 - -#define W_BM 3 -#define W_BO 32 -#define W_BS 6 - -#define W_CM 127 -#define W_CO 64 -#define W_CS 7 - -#define W_DM 7 -#define W_DO 8 -#define W_DS 4 -#elif 0 -#define W_AM 97 -#define W_AO 32 -#define W_AS 6 - -#define W_BM 63 -#define W_BO 512 -#define W_BS 10 - -#define W_CM 13 -#define W_CO 8 -#define W_CS 4 - -#define W_DM 15 -#define W_DO 16 -#define W_DS 5 - -#else - -#define W_AM 203 -#define W_AO 64 -#define W_AS 7 - -#define W_BM 217 -#define W_BO 2048 -#define W_BS 12 - -#define W_CM 113 -#define W_CO 64 -#define W_CS 7 - -#define W_DM 227 -#define W_DO 128 -#define W_DS 9 -#endif static void horizontal_decompose97i(DWTELEM *b, int width){ DWTELEM temp[width]; const int w2= (width+1)>>1; @@ -1410,7 +1323,7 @@ static void spatial_compose53i(DWTELEM *buffer, int width, int height, int strid } -static void horizontal_compose97i(DWTELEM *b, int width){ +void ff_snow_horizontal_compose97i(DWTELEM *b, int width){ DWTELEM temp[width]; const int w2= (width+1)>>1; @@ -1463,7 +1376,7 @@ static void vertical_compose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int wid } } -static void vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){ +void ff_snow_vertical_compose97i(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width){ int i; for(i=0; iy = -3; } -static void spatial_compose97i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ +static void spatial_compose97i_dy_buffered(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * sb, int width, int height, int stride_line){ int y = cs->y; DWTELEM *b0= cs->b0; @@ -1516,7 +1429,7 @@ static void spatial_compose97i_dy_buffered(dwt_compose_t *cs, slice_buffer * sb, {START_TIMER if(y>0 && y+4vertical_compose97i(b0, b1, b2, b3, b4, b5, width); }else{ if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width); if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width); @@ -1527,8 +1440,8 @@ if(width>400){ STOP_TIMER("vertical_compose97i")}} {START_TIMER - if(y-1<(unsigned)height) horizontal_compose97i(b0, width); - if(y+0<(unsigned)height) horizontal_compose97i(b1, width); + if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width); + if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width); if(width>400 && y+0<(unsigned)height){ STOP_TIMER("horizontal_compose97i")}} @@ -1557,8 +1470,8 @@ if(width>400){ STOP_TIMER("vertical_compose97i")}} {START_TIMER - if(y-1<(unsigned)height) horizontal_compose97i(b0, width); - if(y+0<(unsigned)height) horizontal_compose97i(b1, width); + if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width); + if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width); if(width>400 && b0 <= b2){ STOP_TIMER("horizontal_compose97i")}} @@ -1619,7 +1532,7 @@ static void ff_spatial_idwt_slice(dwt_compose_t *cs, DWTELEM *buffer, int width, } } -static void ff_spatial_idwt_buffered_slice(dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ +static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, dwt_compose_t *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){ const int support = type==1 ? 3 : 5; int level; if(type==2) return; @@ -1627,7 +1540,7 @@ static void ff_spatial_idwt_buffered_slice(dwt_compose_t *cs, slice_buffer * sli for(level=decomposition_count-1; level>=0; level--){ while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){ switch(type){ - case 0: spatial_compose97i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<>level, height>>level, stride_line<>level, height>>level, stride_line<>1); + uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); + uint8_t *obmc4= obmc3+ (obmc_stride>>1); + dst = slice_buffer_get_line(sb, src_y + y); + for(x=0; x>= 8 - FRAC_BITS; + } + if(add){ + v += dst[x + src_x]; + v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; + if(v&(~255)) v= ~(v>>31); + dst8[x + y*src_stride] = v; + }else{ + dst[x + src_x] -= v; + } + } + } +} + //FIXME name clenup (b_w, block_w, b_width stuff) static always_inline void add_yblock_buffered(SnowContext *s, slice_buffer * sb, DWTELEM *old_dst, uint8_t *dst8, uint8_t *src, uint8_t *obmc, int src_x, int src_y, int b_w, int b_h, int w, int h, int dst_stride, int src_stride, int obmc_stride, int b_x, int b_y, int add, int plane_index){ DWTELEM * dst = NULL; @@ -2669,36 +2616,7 @@ assert(src_stride > 2*MB_SIZE + 5); START_TIMER - for(y=0; y>1); - uint8_t *obmc3= obmc1+ obmc_stride*(obmc_stride>>1); - uint8_t *obmc4= obmc3+ (obmc_stride>>1); - dst = slice_buffer_get_line(sb, src_y + y); - for(x=0; x>= 8 - FRAC_BITS; - } - if(add){ -// v += old_dst[x + y*dst_stride]; - v += dst[x + src_x]; - v = (v + (1<<(FRAC_BITS-1))) >> FRAC_BITS; - if(v&(~255)) v= ~(v>>31); - dst8[x + y*src_stride] = v; - }else{ -// old_dst[x + y*dst_stride] -= v; - dst[x + src_x] -= v; - } - } - } + s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8); STOP_TIMER("Inner add y block") } #endif @@ -4399,7 +4317,7 @@ if(s->avctx->debug&2048){ { START_TIMER for(; ydsb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); + ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd); } STOP_TIMER("idwt slice");} diff --git a/libavcodec/snow.h b/libavcodec/snow.h new file mode 100644 index 0000000000..11fd15a295 --- /dev/null +++ b/libavcodec/snow.h @@ -0,0 +1,123 @@ +/* + * Copyright (C) 2004 Michael Niedermayer + * Copyright (C) 2006 Robert Edele + * + * This library 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 of the License, or (at your option) any later version. + * + * This library 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 this library; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#ifndef _SNOW_H +#define _SNOW_H + +#include "dsputil.h" + +#define MID_STATE 128 + +#define MAX_DECOMPOSITIONS 8 +#define MAX_PLANES 4 +#define QSHIFT 5 +#define QROOT (1<