From cf0b7cf953a80e78c1778e819b4d5b03216dc165 Mon Sep 17 00:00:00 2001 From: Christophe Massiot Date: Wed, 17 Jan 2001 18:17:31 +0000 Subject: [PATCH] * Borrowed LiViD's MMX and MMX EXT IDCT. This might break things. Expect a performance increase. --- Makefile.in | 7 +- include/modules.h | 1 + plugins/idct/idct.c | 29 +- plugins/idct/idct.h | 13 +- plugins/idct/idct_common.c | 21 +- plugins/idct/idct_mmxext.c | 79 --- plugins/idct/idctclassic.c | 29 +- plugins/idct/idctmmx.c | 374 +++++++++++++- plugins/idct/idctmmx_asm.S | 829 ------------------------------- plugins/idct/idctmmxext.c | 355 ++++++++++++- src/video_decoder/vdec_idct.h | 4 +- src/video_decoder/video_parser.h | 10 +- src/video_decoder/vpar_blocks.h | 7 +- src/video_decoder/vpar_headers.h | 4 +- src/video_parser/video_parser.c | 4 +- src/video_parser/vpar_blocks.c | 39 +- src/video_parser/vpar_headers.c | 53 +- 17 files changed, 802 insertions(+), 1056 deletions(-) delete mode 100644 plugins/idct/idct_mmxext.c delete mode 100644 plugins/idct/idctmmx_asm.S diff --git a/Makefile.in b/Makefile.in index ee645bb392..dc20060c6c 100644 --- a/Makefile.in +++ b/Makefile.in @@ -278,7 +278,7 @@ endif ifneq (,$(findstring 86,$(ARCH))) ifneq (,$(findstring mmx,$(ARCH))) ASM_OBJ = -STD_PLUGIN_ASM = plugins/idct/idctmmx_asm.o +STD_PLUGIN_ASM = endif endif @@ -351,11 +351,10 @@ PLUGIN_IDCTCLASSIC = plugins/idct/idctclassic.o \ plugins/idct/idct_common.o PLUGIN_IDCTMMX = plugins/idct/idctmmx.o \ - plugins/idct/idct_common.o \ - plugins/idct/idctmmx_asm.o + plugins/idct/idct_common.o PLUGIN_IDCTMMXEXT = plugins/idct/idctmmxext.o \ - plugins/idct/idct_mmxext.o + plugins/idct/idct_common.o PLUGIN_ALSA = plugins/alsa/alsa.o \ plugins/alsa/aout_alsa.o diff --git a/include/modules.h b/include/modules.h index e65ebdd736..57cf042ad9 100644 --- a/include/modules.h +++ b/include/modules.h @@ -83,6 +83,7 @@ typedef struct function_list_s void ( * pf_idct ) ( struct vdec_thread_s * p_vdec, dctelem_t * p_block, int i_idontcare ); + void ( * pf_norm_scan ) ( u8 ppi_scan[2][64] ); } idct; struct diff --git a/plugins/idct/idct.c b/plugins/idct/idct.c index b6927cedb2..7b9738df4f 100644 --- a/plugins/idct/idct.c +++ b/plugins/idct/idct.c @@ -2,7 +2,7 @@ * idct.c : IDCT module ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idct.c,v 1.4 2001/01/16 05:04:25 sam Exp $ + * $Id: idct.c,v 1.5 2001/01/17 18:17:30 massiot Exp $ * * Authors: Gaël Hendryckx * @@ -50,13 +50,8 @@ * Local and extern prototypes. *****************************************************************************/ static void idct_getfunctions( function_list_t * p_function_list ); - static int idct_Probe ( probedata_t *p_data ); -static void vdec_InitIDCT ( vdec_thread_t * p_vdec); - void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); -static void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ); +static void vdec_NormScan ( u8 ppi_scan[2][64] ); /***************************************************************************** @@ -136,6 +131,7 @@ static void idct_getfunctions( function_list_t * p_function_list ) p_function_list->functions.idct.pf_init = vdec_InitIDCT; p_function_list->functions.idct.pf_sparse_idct = vdec_SparseIDCT; p_function_list->functions.idct.pf_idct = vdec_IDCT; + p_function_list->functions.idct.pf_norm_scan = vdec_NormScan; } /***************************************************************************** @@ -153,28 +149,17 @@ static int idct_Probe( probedata_t *p_data ) } /***************************************************************************** - * vdec_InitIDCT : initialize datas for vdec_SparseIDCT + * vdec_NormScan : Unused in this IDCT *****************************************************************************/ -static void vdec_InitIDCT (vdec_thread_t * p_vdec) +static void vdec_NormScan( u8 ppi_scan[2][64] ) { - int i; - - dctelem_t * p_pre = p_vdec->p_pre_idct; - memset( p_pre, 0, 64*64*sizeof(dctelem_t) ); - - for( i=0 ; i < 64 ; i++ ) - { - p_pre[i*64+i] = 1 << SPARSE_SCALE_FACTOR; - vdec_IDCT( p_vdec, &p_pre[i*64], 0) ; - } - return; } /***************************************************************************** * vdec_IDCT : IDCT function for normal matrices *****************************************************************************/ -static void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ) +void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, + int i_idontcare ) { s32 tmp0, tmp1, tmp2, tmp3; s32 tmp10, tmp11, tmp12, tmp13; diff --git a/plugins/idct/idct.h b/plugins/idct/idct.h index d1b25be68d..3159ec3151 100644 --- a/plugins/idct/idct.h +++ b/plugins/idct/idct.h @@ -2,7 +2,7 @@ * idct.h : macros for the inverse discrete cosine transform ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idct.h,v 1.1 2001/01/13 12:57:20 sam Exp $ + * $Id: idct.h,v 1.2 2001/01/17 18:17:30 massiot Exp $ * * Authors: Gaël Hendryckx * Christophe Massiot @@ -26,7 +26,7 @@ #define DCTSIZE 8 /* 8*8 DCT */ /***************************************************************************** - * Macros + * Macros *****************************************************************************/ /* We assume that right shift corresponds to signed division by 2 with @@ -140,3 +140,12 @@ #define MULTIPLY(var,const) ((var) * (const)) #endif +/***************************************************************************** + * Protoypes + *****************************************************************************/ +void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, + int i_sparse_pos); +void vdec_InitIDCT ( vdec_thread_t * p_vdec ); +void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, + int i_idontcare ); + diff --git a/plugins/idct/idct_common.c b/plugins/idct/idct_common.c index 90b308207d..601186222b 100644 --- a/plugins/idct/idct_common.c +++ b/plugins/idct/idct_common.c @@ -2,7 +2,7 @@ * idct_common.c : common IDCT functions ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idct_common.c,v 1.2 2001/01/15 06:18:23 sam Exp $ + * $Id: idct_common.c,v 1.3 2001/01/17 18:17:30 massiot Exp $ * * Authors: Gaël Hendryckx * @@ -40,8 +40,23 @@ #include "idct.h" -void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); +/***************************************************************************** + * vdec_InitIDCT : initialize datas for vdec_SparseIDCT + *****************************************************************************/ +void vdec_InitIDCT (vdec_thread_t * p_vdec) +{ + int i; + + dctelem_t * p_pre = p_vdec->p_pre_idct; + memset( p_pre, 0, 64*64*sizeof(dctelem_t) ); + + for( i=0 ; i < 64 ; i++ ) + { + p_pre[i*64+i] = 1 << SPARSE_SCALE_FACTOR; + vdec_IDCT( p_vdec, &p_pre[i*64], 0) ; + } + return; +} /***************************************************************************** * vdec_SparseIDCT : IDCT function for sparse matrices diff --git a/plugins/idct/idct_mmxext.c b/plugins/idct/idct_mmxext.c deleted file mode 100644 index 0d8122842f..0000000000 --- a/plugins/idct/idct_mmxext.c +++ /dev/null @@ -1,79 +0,0 @@ -/***************************************************************************** - * idct_mmxext.c : MMX EXT IDCT functions - ***************************************************************************** - * Copyright (C) 1999, 2000 VideoLAN - * $Id: idct_mmxext.c,v 1.1 2001/01/16 13:26:46 sam Exp $ - * - * Authors: - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or - * (at your option) any later version. - * - * This program 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 General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA. - *****************************************************************************/ - -/***************************************************************************** - * Preamble - *****************************************************************************/ -#include "defs.h" - -#include - -#include "config.h" -#include "common.h" -#include "threads.h" -#include "mtime.h" -#include "tests.h" - -#include "video.h" -#include "video_output.h" - -#include "video_decoder.h" - -#include "idct.h" - -/***************************************************************************** - * Local and extern prototypes. - *****************************************************************************/ -void vdec_InitIDCT ( vdec_thread_t * p_vdec); -void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); -void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ); - - -/***************************************************************************** - * vdec_InitIDCT : initialize datas for vdec_SparseIDCT - *****************************************************************************/ -void vdec_InitIDCT (vdec_thread_t * p_vdec) -{ - return; -} - -/***************************************************************************** - * vdec_IDCT : IDCT function for normal matrices - *****************************************************************************/ -void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ) -{ - return; -} - -/***************************************************************************** - * vdec_SparseIDCT : IDCT function for sparse matrices - *****************************************************************************/ -void vdec_SparseIDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos ) -{ - return; -} - diff --git a/plugins/idct/idctclassic.c b/plugins/idct/idctclassic.c index 1c6928a2c0..d823d37829 100644 --- a/plugins/idct/idctclassic.c +++ b/plugins/idct/idctclassic.c @@ -2,7 +2,7 @@ * idctclassic.c : Classic IDCT module ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idctclassic.c,v 1.4 2001/01/16 05:04:25 sam Exp $ + * $Id: idctclassic.c,v 1.5 2001/01/17 18:17:30 massiot Exp $ * * Authors: Gaël Hendryckx * @@ -50,13 +50,8 @@ * Local and extern prototypes. *****************************************************************************/ static void idct_getfunctions( function_list_t * p_function_list ); - static int idct_Probe ( probedata_t *p_data ); -static void vdec_InitIDCT ( vdec_thread_t * p_vdec); - void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); -static void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ); +static void vdec_NormScan ( u8 ppi_scan[2][64] ); /***************************************************************************** @@ -136,6 +131,7 @@ static void idct_getfunctions( function_list_t * p_function_list ) p_function_list->functions.idct.pf_init = vdec_InitIDCT; p_function_list->functions.idct.pf_sparse_idct = vdec_SparseIDCT; p_function_list->functions.idct.pf_idct = vdec_IDCT; + p_function_list->functions.idct.pf_norm_scan = vdec_NormScan; } /***************************************************************************** @@ -153,28 +149,17 @@ static int idct_Probe( probedata_t *p_data ) } /***************************************************************************** - * vdec_InitIDCT : initialize datas for vdec_SparseIDCT + * vdec_NormScan : Unused in this IDCT *****************************************************************************/ -static void vdec_InitIDCT (vdec_thread_t * p_vdec) +static void vdec_NormScan( u8 ppi_scan[2][64] ) { - int i; - - dctelem_t * p_pre = p_vdec->p_pre_idct; - memset( p_pre, 0, 64*64*sizeof(dctelem_t) ); - - for( i=0 ; i < 64 ; i++ ) - { - p_pre[i*64+i] = 1 << SPARSE_SCALE_FACTOR; - vdec_IDCT( p_vdec, &p_pre[i*64], 0) ; - } - return; } /***************************************************************************** * vdec_IDCT : IDCT function for normal matrices *****************************************************************************/ -static void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ) +void vdec_IDCT( vdec_thread_t * p_vdec, dctelem_t * p_block, + int i_idontcare ) { /* dct classique: pour tester la meilleure entre la classique et la */ /* no classique */ diff --git a/plugins/idct/idctmmx.c b/plugins/idct/idctmmx.c index 4457bb8650..39416be8bd 100644 --- a/plugins/idct/idctmmx.c +++ b/plugins/idct/idctmmx.c @@ -2,9 +2,13 @@ * idctmmx.c : MMX IDCT module ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idctmmx.c,v 1.5 2001/01/16 16:09:52 sam Exp $ + * $Id: idctmmx.c,v 1.6 2001/01/17 18:17:30 massiot Exp $ * - * Authors: Gaël Hendryckx + * Authors: Aaron Holtzman + * Michel Lespinasse + * Peter Gubanov + * (from the LiViD project) + * Christophe Massiot * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -46,17 +50,15 @@ #include "idct.h" +#include "attributes.h" +#include "mmx.h" + /***************************************************************************** * Local prototypes. *****************************************************************************/ static void idct_getfunctions( function_list_t * p_function_list ); - static int idct_Probe ( probedata_t *p_data ); -static void vdec_InitIDCT ( vdec_thread_t * p_vdec); - void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); - void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ); +static void vdec_NormScan ( u8 ppi_scan[2][64] ); /***************************************************************************** @@ -127,8 +129,7 @@ int DeactivateModule( module_t * p_module ) /* Following functions are local */ /***************************************************************************** - * Functions exported as capabilities. They are declared as static so that - * we don't pollute the namespace too much. + * Functions exported as capabilities. *****************************************************************************/ static void idct_getfunctions( function_list_t * p_function_list ) { @@ -136,6 +137,7 @@ static void idct_getfunctions( function_list_t * p_function_list ) p_function_list->functions.idct.pf_init = vdec_InitIDCT; p_function_list->functions.idct.pf_sparse_idct = vdec_SparseIDCT; p_function_list->functions.idct.pf_idct = vdec_IDCT; + p_function_list->functions.idct.pf_norm_scan = vdec_NormScan; } /***************************************************************************** @@ -149,10 +151,10 @@ static int idct_Probe( probedata_t *p_data ) { return( 999 ); } - else + else { return( 100 ); - } + } } else { @@ -161,20 +163,348 @@ static int idct_Probe( probedata_t *p_data ) } /***************************************************************************** - * vdec_InitIDCT : initialize datas for vdec_SparceIDCT + * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table *****************************************************************************/ -static void vdec_InitIDCT (vdec_thread_t * p_vdec) +static void vdec_NormScan( u8 ppi_scan[2][64] ) { - int i; - - dctelem_t * p_pre = p_vdec->p_pre_idct; - memset( p_pre, 0, 64*64*sizeof(dctelem_t) ); + int i, j; - for( i=0 ; i < 64 ; i++ ) + for( i = 0; i < 64; i++ ) { - p_pre[i*64+i] = 1 << SPARSE_SCALE_FACTOR; - vdec_IDCT( p_vdec, &p_pre[i*64], 0) ; + j = ppi_scan[0][i]; + ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + + j = ppi_scan[1][i]; + ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); } - return; +} + +/***************************************************************************** + * vdec_IDCT : + *****************************************************************************/ +#define ROW_SHIFT 11 +#define COL_SHIFT 6 + +#define round(bias) ((int)(((bias)+0.5) * (1<V18 */ - paddsw %mm0, %mm2 /* V17 */ - movq %mm2, %mm0 /* duplicate V17 */ - psraw $1, %mm2 /* t75=t82 */ - psraw $2, %mm0 /* t72 */ - movq %mm3, %mm4 /* duplicate V0 */ - paddsw %mm5, %mm3 /* V19 */ - psubsw %mm5, %mm4 /* V20 ;mm5 free */ -/* moved from the block below */ - movq 8*10(%ecx), %mm7 - psraw $1, %mm3 /* t74=t81 */ - movq %mm3, %mm6 /* duplicate t74=t81 */ - psraw $2, %mm4 /* t77=t79 */ - psubsw %mm0, %mm1 /* V21 ; mm0 free */ - paddsw %mm2, %mm3 /* V22 */ - movq %mm1, %mm5 /* duplicate V21 */ - paddsw %mm4, %mm1 /* V23 */ - movq %mm3, 8*4(%esi) /* V22 */ - psubsw %mm5, %mm4 /* V24; mm5 free */ - movq %mm1, 8*12(%esi) /* V23 */ - psubsw %mm2, %mm6 /* V25; mm2 free */ - movq %mm4, (%esi) /* V24 */ -/* keep mm6 alive all along the next block */ - /* movq %mm6, 8*8(%esi) V25 */ -/* column 0: odd part - * use V2, V6, V10, V14 to produce V31, V39, V40, V41 - */ -/* moved above: movq 8*10(%ecx), %mm7 */ - - pmulhw 8*10(%esi), %mm7 /* V10 */ - movq 8*6(%ecx), %mm0 - pmulhw 8*6(%esi), %mm0 /* V6 */ - movq 8*2(%ecx), %mm5 - movq %mm7, %mm3 /* duplicate V10 */ - pmulhw 8*2(%esi), %mm5 /* V2 */ - movq 8*14(%ecx), %mm4 - psubsw %mm0, %mm7 /* V26 */ - pmulhw 8*14(%esi), %mm4 /* V14 */ - paddsw %mm0, %mm3 /* V29 ; free mm0 */ - movq %mm7, %mm1 /* duplicate V26 */ - psraw $1, %mm3 /* t91=t94 */ - pmulhw x539f539f539f539f,%mm7 /* V33 */ - psraw $1, %mm1 /* t96 */ - movq %mm5, %mm0 /* duplicate V2 */ - psraw $2, %mm4 /* t85=t87 */ - paddsw %mm4,%mm5 /* V27 */ - psubsw %mm4, %mm0 /* V28 ; free mm4 */ - movq %mm0, %mm2 /* duplicate V28 */ - psraw $1, %mm5 /* t90=t93 */ - pmulhw x4546454645464546,%mm0 /* V35 */ - psraw $1, %mm2 /* t97 */ - movq %mm5, %mm4 /* duplicate t90=t93 */ - psubsw %mm2, %mm1 /* V32 ; free mm2 */ - pmulhw x61f861f861f861f8,%mm1 /* V36 */ - psllw $1, %mm7 /* t107 */ - paddsw %mm3, %mm5 /* V31 */ - psubsw %mm3, %mm4 /* V30 ; free mm3 */ - pmulhw x5a825a825a825a82,%mm4 /* V34 */ - nop - psubsw %mm1, %mm0 /* V38 */ - psubsw %mm7, %mm1 /* V37 ; free mm7 */ - psllw $1, %mm1 /* t114 */ -/* move from the next block */ - movq %mm6, %mm3 /* duplicate V25 */ -/* move from the next block */ - movq 8*4(%esi), %mm7 /* V22 */ - psllw $1, %mm0 /* t110 */ - psubsw %mm5, %mm0 /* V39 (mm5 needed for next block) */ - psllw $2, %mm4 /* t112 */ -/* moved from the next block */ - movq 8*12(%esi), %mm2 /* V23 */ - psubsw %mm0, %mm4 /* V40 */ - paddsw %mm4, %mm1 /* V41; free mm0 */ -/* moved from the next block */ - psllw $1, %mm2 /* t117=t125 */ -/* column 0: output butterfly */ -/* moved above: - * movq %mm6, %mm3 duplicate V25 - * movq 8*4(%esi), %mm7 V22 - * movq 8*12(%esi), %mm2 V23 - * psllw $1, %mm2 t117=t125 - */ - psubsw %mm1, %mm6 /* tm6 */ - paddsw %mm1, %mm3 /* tm8; free mm1 */ - movq %mm7, %mm1 /* duplicate V22 */ - paddsw %mm5, %mm7 /* tm0 */ - movq %mm3, 8*8(%esi) /* tm8; free mm3 */ - psubsw %mm5, %mm1 /* tm14; free mm5 */ - movq %mm6, 8*6(%esi) /* tm6; free mm6 */ - movq %mm2, %mm3 /* duplicate t117=t125 */ - movq (%esi), %mm6 /* V24 */ - paddsw %mm0, %mm2 /* tm2 */ - movq %mm7, (%esi) /* tm0; free mm7 */ - psubsw %mm0, %mm3 /* tm12; free mm0 */ - movq %mm1, 8*14(%esi) /* tm14; free mm1 */ - psllw $1, %mm6 /* t119=t123 */ - movq %mm2, 8*2(%esi) /* tm2; free mm2 */ - movq %mm6, %mm0 /* duplicate t119=t123 */ - movq %mm3, 8*12(%esi) /* tm12; free mm3 */ - paddsw %mm4, %mm6 /* tm4 */ -/* moved from next block */ - movq 8*5(%ecx), %mm1 - psubsw %mm4, %mm0 /* tm10; free mm4 */ -/* moved from next block */ - pmulhw 8*5(%esi), %mm1 /* V5 */ - movq %mm6, 8*4(%esi) /* tm4; free mm6 */ - movq %mm0, 8*10(%esi) /* tm10; free mm0 */ -/* column 1: even part - * use V5, V13, V1, V9 to produce V56..V59 - */ -/* moved to prev block: - * movq 8*5(%ecx), %mm1 - * pmulhw 8*5(%esi), %mm1 V5 - */ - movq 8*13(%ecx), %mm7 - psllw $1, %mm1 /* t128=t130 */ - pmulhw 8*13(%esi), %mm7 /* V13 */ - movq %mm1, %mm2 /* duplicate t128=t130 */ - movq 8(%ecx), %mm3 - pmulhw 8(%esi), %mm3 /* V1 */ - movq 8*9(%ecx), %mm5 - psubsw %mm7, %mm1 /* V50 */ - pmulhw 8*9(%esi), %mm5 /* V9 */ - paddsw %mm7, %mm2 /* V51 */ - pmulhw x5a825a825a825a82, %mm1 /* 23170 ->V52 */ - movq %mm2, %mm6 /* duplicate V51 */ - psraw $1, %mm2 /* t138=t144 */ - movq %mm3, %mm4 /* duplicate V1 */ - psraw $2, %mm6 /* t136 */ - paddsw %mm5, %mm3 /* V53 */ - psubsw %mm5, %mm4 /* V54 ;mm5 free */ - movq %mm3, %mm7 /* duplicate V53 */ -/* moved from next block */ - movq 8*11(%ecx), %mm0 - psraw $1, %mm4 /* t140=t142 */ - psubsw %mm6, %mm1 /* V55 ; mm6 free */ - paddsw %mm2, %mm3 /* V56 */ - movq %mm4, %mm5 /* duplicate t140=t142 */ - paddsw %mm1, %mm4 /* V57 */ - movq %mm3, 8*5(%esi) /* V56 */ - psubsw %mm1, %mm5 /* V58; mm1 free */ - movq %mm4, 8*13(%esi) /* V57 */ - psubsw %mm2, %mm7 /* V59; mm2 free */ - movq %mm5, 8*9(%esi) /* V58 */ -/* keep mm7 alive all along the next block - * movq %mm7, 8(%esi) V59 - * moved above - * movq 8*11(%ecx), %mm0 - */ - pmulhw 8*11(%esi), %mm0 /* V11 */ - movq 8*7(%ecx), %mm6 - pmulhw 8*7(%esi), %mm6 /* V7 */ - movq 8*15(%ecx), %mm4 - movq %mm0, %mm3 /* duplicate V11 */ - pmulhw 8*15(%esi), %mm4 /* V15 */ - movq 8*3(%ecx), %mm5 - psllw $1, %mm6 /* t146=t152 */ - pmulhw 8*3(%esi), %mm5 /* V3 */ - paddsw %mm6, %mm0 /* V63 */ -/* note that V15 computation has a correction step: - * this is a 'magic' constant that rebiases the results to be closer to the - * expected result. this magic constant can be refined to reduce the error - * even more by doing the correction step in a later stage when the number - * is actually multiplied by 16 - */ - paddw x0005000200010001, %mm4 - psubsw %mm6, %mm3 /* V60 ; free mm6 */ - psraw $1, %mm0 /* t154=t156 */ - movq %mm3, %mm1 /* duplicate V60 */ - pmulhw x539f539f539f539f, %mm1 /* V67 */ - movq %mm5, %mm6 /* duplicate V3 */ - psraw $2, %mm4 /* t148=t150 */ - paddsw %mm4, %mm5 /* V61 */ - psubsw %mm4, %mm6 /* V62 ; free mm4 */ - movq %mm5, %mm4 /* duplicate V61 */ - psllw $1, %mm1 /* t169 */ - paddsw %mm0, %mm5 /* V65 -> result */ - psubsw %mm0, %mm4 /* V64 ; free mm0 */ - pmulhw x5a825a825a825a82, %mm4 /* V68 */ - psraw $1, %mm3 /* t158 */ - psubsw %mm6, %mm3 /* V66 */ - movq %mm5, %mm2 /* duplicate V65 */ - pmulhw x61f861f861f861f8, %mm3 /* V70 */ - psllw $1, %mm6 /* t165 */ - pmulhw x4546454645464546, %mm6 /* V69 */ - psraw $1, %mm2 /* t172 */ -/* moved from next block */ - movq 8*5(%esi), %mm0 /* V56 */ - psllw $1, %mm4 /* t174 */ -/* moved from next block */ - psraw $1, %mm0 /* t177=t188 */ - nop - psubsw %mm3, %mm6 /* V72 */ - psubsw %mm1, %mm3 /* V71 ; free mm1 */ - psubsw %mm2, %mm6 /* V73 ; free mm2 */ -/* moved from next block */ - psraw $1, %mm5 /* t178=t189 */ - psubsw %mm6, %mm4 /* V74 */ -/* moved from next block */ - movq %mm0, %mm1 /* duplicate t177=t188 */ - paddsw %mm4, %mm3 /* V75 */ -/* moved from next block */ - paddsw %mm5, %mm0 /* tm1 */ -/* location - * 5 - V56 - * 13 - V57 - * 9 - V58 - * X - V59, mm7 - * X - V65, mm5 - * X - V73, mm6 - * X - V74, mm4 - * X - V75, mm3 - * free mm0, mm1 & mm2 - * moved above - * movq 8*5(%esi), %mm0 V56 - * psllw $1, %mm0 t177=t188 ! new !! - * psllw $1, %mm5 t178=t189 ! new !! - * movq %mm0, %mm1 duplicate t177=t188 - * paddsw %mm5, %mm0 tm1 - */ - movq 8*13(%esi), %mm2 /* V57 */ - psubsw %mm5, %mm1 /* tm15; free mm5 */ - movq %mm0, 8(%esi) /* tm1; free mm0 */ - psraw $1, %mm7 /* t182=t184 ! new !! */ -/* save the store as used directly in the transpose - * movq %mm1, 120(%esi) tm15; free mm1 - */ - movq %mm7, %mm5 /* duplicate t182=t184 */ - psubsw %mm3, %mm7 /* tm7 */ - paddsw %mm3, %mm5 /* tm9; free mm3 */ - movq 8*9(%esi), %mm0 /* V58 */ - movq %mm2, %mm3 /* duplicate V57 */ - movq %mm7, 8*7(%esi) /* tm7; free mm7 */ - psubsw %mm6, %mm3 /* tm13 */ - paddsw %mm6, %mm2 /* tm3 ; free mm6 */ -/* moved up from the transpose */ - movq %mm3, %mm7 -/* moved up from the transpose */ - punpcklwd %mm1, %mm3 - movq %mm0, %mm6 /* duplicate V58 */ - movq %mm2, 8*3(%esi) /* tm3; free mm2 */ - paddsw %mm4, %mm0 /* tm5 */ - psubsw %mm4, %mm6 /* tm11; free mm4 */ -/* moved up from the transpose */ - punpckhwd %mm1, %mm7 - movq %mm0, 8*5(%esi) /* tm5; free mm0 */ -/* moved up from the transpose */ - movq %mm5, %mm2 -/* transpose - M4 part - * --------- --------- - * | M1 | M2 | | M1'| M3'| - * --------- --> --------- - * | M3 | M4 | | M2'| M4'| - * --------- --------- - * Two alternatives: use full mmword approach so the following code can be - * scheduled before the transpose is done without stores, or use the faster - * half mmword stores (when possible) - */ - movd %mm3, 8*9+4(%esi) /* MS part of tmt9 */ - punpcklwd %mm6, %mm5 - movd %mm7, 8*13+4(%esi) /* MS part of tmt13 */ - punpckhwd %mm6, %mm2 - movd %mm5, 8*9(%esi) /* LS part of tmt9 */ - punpckhdq %mm3, %mm5 /* free mm3 */ - movd %mm2, 8*13(%esi) /* LS part of tmt13 */ - punpckhdq %mm7, %mm2 /* free mm7 */ -/* moved up from the M3 transpose */ - movq 8*8(%esi), %mm0 -/* moved up from the M3 transpose */ - movq 8*10(%esi), %mm1 -/* moved up from the M3 transpose */ - movq %mm0, %mm3 -/* shuffle the rest of the data, and write it with 2 mmword writes */ - movq %mm5, 8*11(%esi) /* tmt11 */ -/* moved up from the M3 transpose */ - punpcklwd %mm1, %mm0 - movq %mm2, 8*15(%esi) /* tmt15 */ -/* moved up from the M3 transpose */ - punpckhwd %mm1, %mm3 -/* transpose - M3 part - * moved up to previous code section - * movq 8*8(%esi), %mm0 - * movq 8*10(%esi), %mm1 - * movq %mm0, %mm3 - * punpcklwd %mm1, %mm0 - * punpckhwd %mm1, %mm3 - */ - movq 8*12(%esi), %mm6 - movq 8*14(%esi), %mm4 - movq %mm6, %mm2 -/* shuffle the data and write the lower parts of the transposed in 4 dwords */ - punpcklwd %mm4, %mm6 - movq %mm0, %mm1 - punpckhdq %mm6, %mm1 - movq %mm3, %mm7 - punpckhwd %mm4, %mm2 /* free mm4 */ - punpckldq %mm6, %mm0 /* free mm6 */ -/* moved from next block */ - movq 8*13(%esi), %mm4 /* tmt13 */ - punpckldq %mm2, %mm3 - punpckhdq %mm2, %mm7 /* free mm2 */ -/* moved from next block */ - movq %mm3, %mm5 /* duplicate tmt5 */ -/* column 1: even part (after transpose) -* moved above -* movq %mm3, %mm5 duplicate tmt5 -* movq 8*13(%esi), %mm4 tmt13 -*/ - psubsw %mm4, %mm3 /* V134 */ - pmulhw x5a825a825a825a82, %mm3 /* 23170 ->V136 */ - movq 8*9(%esi), %mm6 /* tmt9 */ - paddsw %mm4, %mm5 /* V135 ; mm4 free */ - movq %mm0, %mm4 /* duplicate tmt1 */ - paddsw %mm6, %mm0 /* V137 */ - psubsw %mm6, %mm4 /* V138 ; mm6 free */ - psllw $2, %mm3 /* t290 */ - psubsw %mm5, %mm3 /* V139 */ - movq %mm0, %mm6 /* duplicate V137 */ - paddsw %mm5, %mm0 /* V140 */ - movq %mm4, %mm2 /* duplicate V138 */ - paddsw %mm3, %mm2 /* V141 */ - psubsw %mm3, %mm4 /* V142 ; mm3 free */ - movq %mm0, 8*9(%esi) /* V140 */ - psubsw %mm5, %mm6 /* V143 ; mm5 free */ -/* moved from next block */ - movq 8*11(%esi), %mm0 /* tmt11 */ - movq %mm2, 8*13(%esi) /* V141 */ -/* moved from next block */ - movq %mm0, %mm2 /* duplicate tmt11 */ -/* column 1: odd part (after transpose) */ -/* moved up to the prev block - * movq 8*11(%esi), %mm0 tmt11 - * movq %mm0, %mm2 duplicate tmt11 - */ - movq 8*15(%esi), %mm5 /* tmt15 */ - psubsw %mm7, %mm0 /* V144 */ - movq %mm0, %mm3 /* duplicate V144 */ - paddsw %mm7, %mm2 /* V147 ; free mm7 */ - pmulhw x539f539f539f539f, %mm0 /* 21407-> V151 */ - movq %mm1, %mm7 /* duplicate tmt3 */ - paddsw %mm5, %mm7 /* V145 */ - psubsw %mm5, %mm1 /* V146 ; free mm5 */ - psubsw %mm1, %mm3 /* V150 */ - movq %mm7, %mm5 /* duplicate V145 */ - pmulhw x4546454645464546, %mm1 /* 17734-> V153 */ - psubsw %mm2, %mm5 /* V148 */ - pmulhw x61f861f861f861f8, %mm3 /* 25080-> V154 */ - psllw $2, %mm0 /* t311 */ - pmulhw x5a825a825a825a82, %mm5 /* 23170-> V152 */ - paddsw %mm2, %mm7 /* V149 ; free mm2 */ - psllw $1, %mm1 /* t313 */ - nop /* without the nop - freeze here for one clock */ - movq %mm3, %mm2 /* duplicate V154 */ - psubsw %mm0, %mm3 /* V155 ; free mm0 */ - psubsw %mm2, %mm1 /* V156 ; free mm2 */ -/* moved from the next block */ - movq %mm6, %mm2 /* duplicate V143 */ -/* moved from the next block */ - movq 8*13(%esi), %mm0 /* V141 */ - psllw $1, %mm1 /* t315 */ - psubsw %mm7, %mm1 /* V157 (keep V149) */ - psllw $2, %mm5 /* t317 */ - psubsw %mm1, %mm5 /* V158 */ - psllw $1, %mm3 /* t319 */ - paddsw %mm5, %mm3 /* V159 */ -/* column 1: output butterfly (after transform) - * moved to the prev block - * movq %mm6, %mm2 duplicate V143 - * movq 8*13(%esi), %mm0 V141 - */ - psubsw %mm3, %mm2 /* V163 */ - paddsw %mm3, %mm6 /* V164 ; free mm3 */ - movq %mm4, %mm3 /* duplicate V142 */ - psubsw %mm5, %mm4 /* V165 ; free mm5 */ - movq %mm2, scratch7 /* out7 */ - psraw $4, %mm6 - psraw $4, %mm4 - paddsw %mm5, %mm3 /* V162 */ - movq 8*9(%esi), %mm2 /* V140 */ - movq %mm0, %mm5 /* duplicate V141 */ -/* in order not to perculate this line up, - * we read 72(%esi) very near to this location - */ - movq %mm6, 8*9(%esi) /* out9 */ - paddsw %mm1, %mm0 /* V161 */ - movq %mm3, scratch5 /* out5 */ - psubsw %mm1, %mm5 /* V166 ; free mm1 */ - movq %mm4, 8*11(%esi) /* out11 */ - psraw $4, %mm5 - movq %mm0, scratch3 /* out3 */ - movq %mm2, %mm4 /* duplicate V140 */ - movq %mm5, 8*13(%esi) /* out13 */ - paddsw %mm7, %mm2 /* V160 */ -/* moved from the next block */ - movq 8(%esi), %mm0 - psubsw %mm7, %mm4 /* V167 ; free mm7 */ -/* moved from the next block */ - movq 8*3(%esi), %mm7 - psraw $4, %mm4 - movq %mm2, scratch1 /* out1 */ -/* moved from the next block */ - movq %mm0, %mm1 - movq %mm4, 8*15(%esi) /* out15 */ -/* moved from the next block */ - punpcklwd %mm7, %mm0 -/* transpose - M2 parts - * moved up to the prev block - * movq 8(%esi), %mm0 - * movq 8*3(%esi), %mm7 - * movq %mm0, %mm1 - * punpcklwd %mm7, %mm0 - */ - movq 8*5(%esi), %mm5 - punpckhwd %mm7, %mm1 - movq 8*7(%esi), %mm4 - movq %mm5, %mm3 -/* shuffle the data and write the lower parts of the trasposed in 4 dwords */ - movd %mm0, 8*8(%esi) /* LS part of tmt8 */ - punpcklwd %mm4, %mm5 - movd %mm1, 8*12(%esi) /* LS part of tmt12 */ - punpckhwd %mm4, %mm3 - movd %mm5, 8*8+4(%esi) /* MS part of tmt8 */ - punpckhdq %mm5, %mm0 /* tmt10 */ - movd %mm3, 8*12+4(%esi) /* MS part of tmt12 */ - punpckhdq %mm3, %mm1 /* tmt14 */ -/* transpose - M1 parts */ - movq (%esi), %mm7 - movq 8*2(%esi), %mm2 - movq %mm7, %mm6 - movq 8*4(%esi), %mm5 - punpcklwd %mm2, %mm7 - movq 8*6(%esi), %mm4 - punpckhwd %mm2, %mm6 /* free mm2 */ - movq %mm5, %mm3 - punpcklwd %mm4, %mm5 - punpckhwd %mm4, %mm3 /* free mm4 */ - movq %mm7, %mm2 - movq %mm6, %mm4 - punpckldq %mm5, %mm7 /* tmt0 */ - punpckhdq %mm5, %mm2 /* tmt2 ; free mm5 */ -/* shuffle the rest of the data, and write it with 2 mmword writes */ - punpckldq %mm3, %mm6 /* tmt4 */ -/* moved from next block */ - movq %mm2, %mm5 /* duplicate tmt2 */ - punpckhdq %mm3, %mm4 /* tmt6 ; free mm3 */ -/* moved from next block */ - movq %mm0, %mm3 /* duplicate tmt10 */ -/* column 0: odd part (after transpose) - *moved up to prev block - * movq %mm0, %mm3 duplicate tmt10 - * movq %mm2, %mm5 duplicate tmt2 - */ - psubsw %mm4, %mm0 /* V110 */ - paddsw %mm4, %mm3 /* V113 ; free mm4 */ - movq %mm0, %mm4 /* duplicate V110 */ - paddsw %mm1, %mm2 /* V111 */ - pmulhw x539f539f539f539f, %mm0 /* 21407-> V117 */ - psubsw %mm1, %mm5 /* V112 ; free mm1 */ - psubsw %mm5, %mm4 /* V116 */ - movq %mm2, %mm1 /* duplicate V111 */ - pmulhw x4546454645464546, %mm5 /* 17734-> V119 */ - psubsw %mm3, %mm2 /* V114 */ - pmulhw x61f861f861f861f8, %mm4 /* 25080-> V120 */ - paddsw %mm3, %mm1 /* V115 ; free mm3 */ - pmulhw x5a825a825a825a82, %mm2 /* 23170-> V118 */ - psllw $2, %mm0 /* t266 */ - movq %mm1, (%esi) /* save V115 */ - psllw $1, %mm5 /* t268 */ - psubsw %mm4, %mm5 /* V122 */ - psubsw %mm0, %mm4 /* V121 ; free mm0 */ - psllw $1, %mm5 /* t270 */ - psubsw %mm1, %mm5 /* V123 ; free mm1 */ - psllw $2, %mm2 /* t272 */ - psubsw %mm5, %mm2 /* V124 (keep V123) */ - psllw $1, %mm4 /* t274 */ - movq %mm5, 8*2(%esi) /* save V123 ; free mm5 */ - paddsw %mm2, %mm4 /* V125 (keep V124) */ -/* column 0: even part (after transpose) */ - movq 8*12(%esi), %mm0 /* tmt12 */ - movq %mm6, %mm3 /* duplicate tmt4 */ - psubsw %mm0, %mm6 /* V100 */ - paddsw %mm0, %mm3 /* V101 ; free mm0 */ - pmulhw x5a825a825a825a82, %mm6 /* 23170 ->V102 */ - movq %mm7, %mm5 /* duplicate tmt0 */ - movq 8*8(%esi), %mm1 /* tmt8 */ - paddsw %mm1, %mm7 /* V103 */ - psubsw %mm1, %mm5 /* V104 ; free mm1 */ - movq %mm7, %mm0 /* duplicate V103 */ - psllw $2, %mm6 /* t245 */ - paddsw %mm3, %mm7 /* V106 */ - movq %mm5, %mm1 /* duplicate V104 */ - psubsw %mm3, %mm6 /* V105 */ - psubsw %mm3, %mm0 /* V109; free mm3 */ - paddsw %mm6, %mm5 /* V107 */ - psubsw %mm6, %mm1 /* V108 ; free mm6 */ -/* column 0: output butterfly (after transform) */ - movq %mm1, %mm3 /* duplicate V108 */ - paddsw %mm2, %mm1 /* out4 */ - psraw $4, %mm1 - psubsw %mm2, %mm3 /* out10 ; free mm2 */ - psraw $4, %mm3 - movq %mm0, %mm6 /* duplicate V109 */ - movq %mm1, 8*4(%esi) /* out4 ; free mm1 */ - psubsw %mm4, %mm0 /* out6 */ - movq %mm3, 8*10(%esi) /* out10 ; free mm3 */ - psraw $4, %mm0 - paddsw %mm4, %mm6 /* out8 ; free mm4 */ - movq %mm7, %mm1 /* duplicate V106 */ - movq %mm0, 8*6(%esi) /* out6 ; free mm0 */ - psraw $4, %mm6 - movq (%esi), %mm4 /* V115 */ - movq %mm6, 8*8(%esi) /* out8 ; free mm6 */ - movq %mm5, %mm2 /* duplicate V107 */ - movq 8*2(%esi), %mm3 /* V123 */ - paddsw %mm4, %mm7 /* out0 */ -/* moved up from next block */ - movq scratch3, %mm0 - psraw $4, %mm7 -/* moved up from next block */ - movq scratch5, %mm6 - psubsw %mm4, %mm1 /* out14 ; free mm4 */ - paddsw %mm3, %mm5 /* out2 */ - psraw $4, %mm1 - movq %mm7, (%esi) /* out0 ; free mm7 */ - psraw $4, %mm5 - movq %mm1, 8*14(%esi) /* out14 ; free mm1 */ - psubsw %mm3, %mm2 /* out12 ; free mm3 */ - movq %mm5, 8*2(%esi) /* out2 ; free mm5 */ - psraw $4, %mm2 -/* moved up to the prev block */ - movq scratch7, %mm4 -/* moved up to the prev block */ - psraw $4, %mm0 - movq %mm2, 8*12(%esi) /* out12 ; free mm2 */ -/* moved up to the prev block */ - psraw $4, %mm6 -/* move back the data to its correct place -* moved up to the prev block - * movq scratch3, %mm0 - * movq scratch5, %mm6 - * movq scratch7, %mm4 - * psraw $4, %mm0 - * psraw $4, %mm6 -*/ - movq scratch1, %mm1 - psraw $4, %mm4 - movq %mm0, 8*3(%esi) /* out3 */ - psraw $4, %mm1 - movq %mm6, 8*5(%esi) /* out5 */ - movq %mm4, 8*7(%esi) /* out7 */ - movq %mm1, 8(%esi) /* out1 */ -/* transpose matrix */ - movl $8, %ebx /* ebx is x_size */ - movl %esi, %edi /* pointer to the matrix */ - movl %ebx, %ecx - sal $2, %ecx - movl %ebx, %eax - addl %ebx, %ecx - subl $4, %eax /* eax is inner loop variable */ - addl %ebx, %ecx /* ecx is 6*row size */ - movl %eax, %edx /* edx is the outer loop variable */ -.L1: movq (%esi), %mm0 /* first line */ - movq (%esi,%ebx,4), %mm2 /* third line */ - movq %mm0, %mm6 /* copy first line */ - punpcklwd (%esi,%ebx,2), %mm0 /* interleave fist and second lines */ - movq %mm2, %mm7 /* copy third line */ - punpcklwd (%esi,%ecx), %mm2 /* interleave third and fourth lines */ - movq %mm0, %mm4 /* copy first intermediate result */ - movq (%esi,%ebx,2), %mm1 /* second line */ -/* the next line 'punpcklwd %mm2, %mm0' inverted two pixels. */ -/* punpckldq make printing cleaner */ - punpckldq %mm2, %mm0 /* interleave to produce result 1 */ - movq (%esi,%ecx), %mm3 /* fourth line */ - punpckhdq %mm2, %mm4 /* interleave to produce result 2 */ - movq %mm0, (%esi) /* write result 1 */ - punpckhwd %mm1, %mm6 /* interleave first and second lines */ - movq %mm4, (%esi,%ebx,2) /* write result 2 */ - punpckhwd %mm3, %mm7 /* interleave 3rd and 4th lines */ - movq %mm6, %mm5 /* copy first intermediate result */ - punpckldq %mm7, %mm6 /* interleave to produce result 3 */ - leal (%edi,%ebx,8), %edi /* point to 4x4 set 4 rows down */ - punpckhdq %mm7, %mm5 /* interleave to produce result 4 */ - movq %mm6, (%esi,%ebx,4) /* write result 3 */ - movq %mm5, (%esi,%ecx) /* write result 4 */ - /* check to see if number of rows left is zero */ - cmpl $0, %edx - /* last time through you are done and ready to exit */ - je .L3 -.L2: movq 8(%esi), %mm0 /* first line */ - movq 8(%esi,%ebx,4), %mm2 /* third line */ - movq %mm0, %mm6 /* copy first line */ - punpcklwd 8(%esi,%ebx,2), %mm0 /* interleave first and second lines */ - movq %mm2, %mm7 /* copy third line */ - punpcklwd 8(%esi,%ecx), %mm2 /* interleave 3rd and 4th lines */ - movq %mm0, %mm4 /* copy first intermediate */ - movq (%edi), %mm1 /* first line */ - punpckldq %mm2, %mm0 /* interleave to produce 1st result */ - movq (%edi,%ebx,4), %mm3 /* third line */ - punpckhdq %mm2, %mm4 /* interleave to produce 2nd result */ - punpckhwd 8(%esi,%ebx,2), %mm6 /* interleave 1st and 2nd lines */ - movq %mm1, %mm2 /* copy first line */ - punpckhwd 8(%esi,%ecx), %mm7 /* interleave 3rd and 4th lines */ - movq %mm6, %mm5 /* copy first intermediate */ - movq %mm0, (%edi) /* write result 1 */ - punpckhdq %mm7, %mm5 /* produce third result */ - punpcklwd (%edi,%ebx,2), %mm1 /* interleave 1st and 2nd lines */ - movq %mm3, %mm0 /* copy third line */ - punpckhwd (%edi,%ebx,2), %mm2 /* interleave 1st and 2nd lines */ - movq %mm4, (%edi,%ebx,2) /* write result 2 */ - punpckldq %mm7, %mm6 /* produce fourth result */ - punpcklwd (%edi,%ecx), %mm3 /* interleave 3rd and 4th lines */ - movq %mm1, %mm4 /* copy first intermediate */ - movq %mm6, (%edi,%ebx,4) /* write result 3 */ - punpckldq %mm3, %mm1 - punpckhwd (%edi,%ecx), %mm0 /* interleave 3rd and 4th lines */ - movq %mm2, %mm6 /* copy second intermediate */ - movq %mm5, (%edi,%ecx) /* write result 4 */ - punpckhdq %mm3, %mm4 /* produce second result */ - movq %mm1, 8(%esi) /* write result 5 */ - punpckldq %mm0, %mm2 /* produce third result */ - movq %mm4, 8(%esi,%ebx,2) /* write result 6 */ - punpckhdq %mm0, %mm6 /* produce fourth result */ - movq %mm2, 8(%esi,%ebx,4) /* write result 7 */ - movq %mm6, 8(%esi,%ecx) /* write result 8 */ - /* increment %esi to point to next 4x4 block in same row */ - addl $8, %esi - /* increment %edi to point to nxt 4x4 block below current */ - leal (%edi,%ebx,8), %edi - sub $4, %eax /* decrement inner loop var */ - jnz .L2 - /* %edi points to start of second row in block just finished */ - sal $1, %edx - leal 8(%esi,%ebx,8), %esi - subl %edx, %esi - /* subtract the number of bytes in last row */ - /* now we point to spot where row=col */ - subl $8, %edx /* sub 4 from row number */ - sarl $1, %edx - mov %esi, %edi - mov %edx, %eax - /* reset x_size to outer loop variable to start new row */ - jmp .L1 -.L3: emms - popl %edi - popl %esi - popl %edx - popl %ecx - popl %ebx - movl %ebp,%esp - popl %ebp - ret -.Lfe1: - .size vdec_IDCT,.Lfe1-vdec_IDCT diff --git a/plugins/idct/idctmmxext.c b/plugins/idct/idctmmxext.c index f09005c3b4..017f177eff 100644 --- a/plugins/idct/idctmmxext.c +++ b/plugins/idct/idctmmxext.c @@ -2,9 +2,13 @@ * idctmmxext.c : MMX EXT IDCT module ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: idctmmxext.c,v 1.3 2001/01/16 16:09:52 sam Exp $ + * $Id: idctmmxext.c,v 1.4 2001/01/17 18:17:30 massiot Exp $ * - * Authors: + * Authors: Aaron Holtzman + * Michel Lespinasse + * Peter Gubanov + * (from the LiViD project) + * Christophe Massiot * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by @@ -46,18 +50,15 @@ #include "idct.h" +#include "attributes.h" +#include "mmx.h" + /***************************************************************************** * Local prototypes. *****************************************************************************/ static void idct_getfunctions( function_list_t * p_function_list ); - static int idct_Probe ( probedata_t *p_data ); - - void vdec_InitIDCT ( vdec_thread_t * p_vdec); - void vdec_SparseIDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_sparse_pos); - void vdec_IDCT ( vdec_thread_t * p_vdec, dctelem_t * p_block, - int i_idontcare ); +static void vdec_NormScan ( u8 ppi_scan[2][64] ); /***************************************************************************** @@ -136,6 +137,7 @@ static void idct_getfunctions( function_list_t * p_function_list ) p_function_list->functions.idct.pf_init = vdec_InitIDCT; p_function_list->functions.idct.pf_sparse_idct = vdec_SparseIDCT; p_function_list->functions.idct.pf_idct = vdec_IDCT; + p_function_list->functions.idct.pf_norm_scan = vdec_NormScan; } /***************************************************************************** @@ -143,16 +145,16 @@ static void idct_getfunctions( function_list_t * p_function_list ) *****************************************************************************/ static int idct_Probe( probedata_t *p_data ) { - if( 0/*TestCPU( CPU_CAPABILITY_MMXEXT )*/ ) + if( TestCPU( CPU_CAPABILITY_MMXEXT ) ) { if( TestMethod( IDCT_METHOD_VAR, "idctmmxext" ) ) { return( 999 ); } - else + else { return( 200 ); - } + } } else { @@ -160,3 +162,332 @@ static int idct_Probe( probedata_t *p_data ) } } +/***************************************************************************** + * vdec_NormScan : This IDCT uses reordered coeffs, so we patch the scan table + *****************************************************************************/ +static void vdec_NormScan( u8 ppi_scan[2][64] ) +{ + int i, j; + + for( i = 0; i < 64; i++ ) + { + j = ppi_scan[0][i]; + ppi_scan[0][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + + j = ppi_scan[1][i]; + ppi_scan[1][i] = (j & 0x38) | ((j & 6) >> 1) | ((j & 1) << 2); + } +} + +/***************************************************************************** + * vdec_IDCT : + *****************************************************************************/ +#define ROW_SHIFT 11 +#define COL_SHIFT 6 + +#define round(bias) ((int)(((bias)+0.5) * (1< * Christophe Massiot @@ -26,4 +26,4 @@ struct vdec_thread_s; typedef void ( *idct_init_t ) ( struct vdec_thread_s * ); typedef void ( *f_idct_t ) ( struct vdec_thread_s *, dctelem_t*, int ); - +typedef void ( *norm_scan_t ) ( u8 ppi_scan[2][64] ); diff --git a/src/video_decoder/video_parser.h b/src/video_decoder/video_parser.h index bf96ce4af4..4dd14b261e 100644 --- a/src/video_decoder/video_parser.h +++ b/src/video_decoder/video_parser.h @@ -2,7 +2,7 @@ * video_parser.h : video parser thread ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: video_parser.h,v 1.4 2001/01/15 13:25:09 massiot Exp $ + * $Id: video_parser.h,v 1.5 2001/01/17 18:17:30 massiot Exp $ * * Authors: Christophe Massiot * @@ -125,14 +125,20 @@ typedef struct vpar_thread_s lookup_t * pl_coded_pattern; /* variable length codes for the structure dct_dc_size for intra blocks */ lookup_t * pppl_dct_dc_size[2][2]; - /* Structure to store the tables B14 & B15 (ISO/CEI 13818-2 B.4) */ + /* Structure to store the tables B14 & B15 (ISO/IEC 13818-2 B.4) */ dct_lookup_t ppl_dct_coef[2][16384]; + /* Scan table */ + u8 ppi_scan[2][64]; + /* Default quantization matrices */ + u8 pi_default_intra_quant[64]; + u8 pi_default_nonintra_quant[64]; /* IDCT plugin used and shortcuts to access its capabilities */ struct module_s * p_module; idct_init_t pf_init; f_idct_t pf_sparse_idct; f_idct_t pf_idct; + norm_scan_t pf_norm_scan; #ifdef STATS /* Statistics */ diff --git a/src/video_decoder/vpar_blocks.h b/src/video_decoder/vpar_blocks.h index 99236d4564..907e8eb46a 100644 --- a/src/video_decoder/vpar_blocks.h +++ b/src/video_decoder/vpar_blocks.h @@ -2,7 +2,7 @@ * vpar_blocks.h : video parser blocks management ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: vpar_blocks.h,v 1.1 2000/12/21 17:19:52 massiot Exp $ + * $Id: vpar_blocks.h,v 1.2 2001/01/17 18:17:30 massiot Exp $ * * Authors: Christophe Massiot * Jean-Marc Dressler @@ -157,8 +157,8 @@ typedef struct dct_lookup_s /***************************************************************************** * Constants *****************************************************************************/ -extern int pi_default_intra_quant[]; -extern int pi_default_nonintra_quant[]; +extern u8 pi_default_intra_quant[64]; +extern u8 pi_default_nonintra_quant[64]; extern u8 pi_scan[2][64]; /***************************************************************************** @@ -170,4 +170,5 @@ void vpar_InitPMBType( struct vpar_thread_s * p_vpar ); void vpar_InitBMBType( struct vpar_thread_s * p_vpar ); void vpar_InitCodedPattern( struct vpar_thread_s * p_vpar ); void vpar_InitDCTTables( struct vpar_thread_s * p_vpar ); +void vpar_InitScanTable( struct vpar_thread_s * p_vpar ); void vpar_PictureData( struct vpar_thread_s * p_vpar, int i_mb_base ); diff --git a/src/video_decoder/vpar_headers.h b/src/video_decoder/vpar_headers.h index 2b08a7eb7c..0e32bea3f0 100644 --- a/src/video_decoder/vpar_headers.h +++ b/src/video_decoder/vpar_headers.h @@ -2,7 +2,7 @@ * vpar_headers.h : video parser : headers parsing ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: vpar_headers.h,v 1.3 2001/01/15 18:02:48 massiot Exp $ + * $Id: vpar_headers.h,v 1.4 2001/01/17 18:17:30 massiot Exp $ * * Authors: Christophe Massiot * Stéphane Borel @@ -40,7 +40,7 @@ *****************************************************************************/ typedef struct quant_matrix_s { - int * pi_matrix; + u8 * pi_matrix; boolean_t b_allocated; /* Has the matrix been allocated by vpar_headers ? */ } quant_matrix_t; diff --git a/src/video_parser/video_parser.c b/src/video_parser/video_parser.c index 3c2f214950..7add0ba9fa 100644 --- a/src/video_parser/video_parser.c +++ b/src/video_parser/video_parser.c @@ -2,7 +2,7 @@ * video_parser.c : video parser thread ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: video_parser.c,v 1.67 2001/01/15 13:25:09 massiot Exp $ + * $Id: video_parser.c,v 1.68 2001/01/17 18:17:31 massiot Exp $ * * Authors: Christophe Massiot * Samuel Hocevar @@ -116,6 +116,7 @@ vlc_thread_t vpar_CreateThread( vdec_config_t * p_config ) p_vpar->pf_init = idct_functions.pf_init; p_vpar->pf_sparse_idct = idct_functions.pf_sparse_idct; p_vpar->pf_idct = idct_functions.pf_idct; + p_vpar->pf_norm_scan = idct_functions.pf_norm_scan; #undef idct_functions /* Spawn the video parser thread */ @@ -228,6 +229,7 @@ static int InitThread( vpar_thread_t *p_vpar ) vpar_InitPMBType( p_vpar ); vpar_InitBMBType( p_vpar ); vpar_InitDCTTables( p_vpar ); + vpar_InitScanTable( p_vpar ); /* * Initialize the synchro properties diff --git a/src/video_parser/vpar_blocks.c b/src/video_parser/vpar_blocks.c index d41a4922b8..132f9a5007 100644 --- a/src/video_parser/vpar_blocks.c +++ b/src/video_parser/vpar_blocks.c @@ -2,7 +2,7 @@ * vpar_blocks.c : blocks parsing ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: vpar_blocks.c,v 1.70 2001/01/13 12:57:21 sam Exp $ + * $Id: vpar_blocks.c,v 1.71 2001/01/17 18:17:31 massiot Exp $ * * Authors: Christophe Massiot * Jean-Marc Dressler @@ -609,6 +609,26 @@ void vpar_InitDCTTables( vpar_thread_t * p_vpar ) FillDCTTable( p_vpar->ppl_dct_coef[1], pl_DCT_tab6, 1, 16, 16 ); } +/***************************************************************************** + * vpar_InitScanTable : Initialize scan table + *****************************************************************************/ +void vpar_InitScanTable( vpar_thread_t * p_vpar ) +{ + int i; + + memcpy( p_vpar->ppi_scan, pi_scan, sizeof(pi_scan) ); + p_vpar->pf_norm_scan( p_vpar->ppi_scan ); + + /* If scan table has changed, we must change the quantization matrices. */ + for( i = 0; i < 64; i++ ) + { + p_vpar->pi_default_intra_quant[ p_vpar->ppi_scan[0][i] ] = + pi_default_intra_quant[ pi_scan[0][i] ]; + p_vpar->pi_default_nonintra_quant[ p_vpar->ppi_scan[0][i] ] = + pi_default_nonintra_quant[ pi_scan[0][i] ]; + } +} + /* * Block parsing @@ -686,11 +706,6 @@ static __inline__ void DecodeMPEG1NonIntra( vpar_thread_t * p_vpar, break; case DCT_EOB: -#ifdef HAVE_MMX - /* The MMX IDCT has a precision problem with non-intra - * blocks. */ - p_mb->ppi_blocks[i_b][0] += 4; -#endif if( i_nc <= 1 ) { p_mb->pf_idct[i_b] = p_vpar->pf_sparse_idct; @@ -715,7 +730,7 @@ static __inline__ void DecodeMPEG1NonIntra( vpar_thread_t * p_vpar, break; } - i_pos = pi_scan[p_vpar->picture.b_alternate_scan][i_parse]; + i_pos = p_vpar->ppi_scan[p_vpar->picture.b_alternate_scan][i_parse]; i_level = ( ((i_level << 1) + 1) * p_vpar->mb.i_quantizer_scale * p_vpar->sequence.nonintra_quant.pi_matrix[i_pos] ) >> 4; @@ -870,7 +885,7 @@ static __inline__ void DecodeMPEG1Intra( vpar_thread_t * p_vpar, } /* Determine the position of the block in the frame */ - i_pos = pi_scan[p_vpar->picture.b_alternate_scan][i_parse]; + i_pos = p_vpar->ppi_scan[p_vpar->picture.b_alternate_scan][i_parse]; i_level = ( i_level * p_vpar->mb.i_quantizer_scale * p_vpar->sequence.intra_quant.pi_matrix[i_pos] ) >> 3; @@ -908,7 +923,7 @@ static __inline__ void DecodeMPEG2NonIntra( vpar_thread_t * p_vpar, boolean_t b_dc; boolean_t b_sign; boolean_t b_chroma; - int * pi_quant; + u8 * pi_quant; /* Give the chromatic component (0, 1, 2) */ i_cc = pi_cc_index[i_b]; @@ -988,7 +1003,7 @@ static __inline__ void DecodeMPEG2NonIntra( vpar_thread_t * p_vpar, break; } - i_pos = pi_scan[p_vpar->picture.b_alternate_scan][i_parse]; + i_pos = p_vpar->ppi_scan[p_vpar->picture.b_alternate_scan][i_parse]; i_level = ( ((i_level << 1) + 1) * p_vpar->mb.i_quantizer_scale * pi_quant[i_pos] ) >> 5; p_mb->ppi_blocks[i_b][i_pos] = b_sign ? -i_level : i_level; @@ -1019,7 +1034,7 @@ static __inline__ void DecodeMPEG2Intra( vpar_thread_t * p_vpar, boolean_t b_vlc_intra; boolean_t b_sign; boolean_t b_chroma; - int * pi_quant; + u8 * pi_quant; /* Give the chromatic component (0, 1, 2) */ i_cc = pi_cc_index[i_b]; @@ -1132,7 +1147,7 @@ static __inline__ void DecodeMPEG2Intra( vpar_thread_t * p_vpar, } /* Determine the position of the block in the frame */ - i_pos = pi_scan[p_vpar->picture.b_alternate_scan][i_parse]; + i_pos = p_vpar->ppi_scan[p_vpar->picture.b_alternate_scan][i_parse]; i_level = ( i_level * p_vpar->mb.i_quantizer_scale * pi_quant[i_pos] ) >> 4; diff --git a/src/video_parser/vpar_headers.c b/src/video_parser/vpar_headers.c index 28de7f60bf..47e03ed58e 100644 --- a/src/video_parser/vpar_headers.c +++ b/src/video_parser/vpar_headers.c @@ -2,7 +2,7 @@ * vpar_headers.c : headers parsing ***************************************************************************** * Copyright (C) 1999, 2000 VideoLAN - * $Id: vpar_headers.c,v 1.72 2001/01/16 18:06:01 massiot Exp $ + * $Id: vpar_headers.c,v 1.73 2001/01/17 18:17:31 massiot Exp $ * * Authors: Christophe Massiot * Stéphane Borel @@ -76,8 +76,7 @@ static void CopyrightExtension( vpar_thread_t * p_vpar ); /***************************************************************************** * pi_default_intra_quant : default quantization matrix *****************************************************************************/ -#ifndef VDEC_DFT -int pi_default_intra_quant[] = +u8 pi_default_intra_quant[] = { 8, 16, 19, 22, 26, 27, 29, 34, 16, 16, 22, 24, 27, 29, 34, 37, @@ -88,25 +87,11 @@ int pi_default_intra_quant[] = 26, 27, 29, 34, 38, 46, 56, 69, 27, 29, 35, 38, 46, 56, 69, 83 }; -#else -int pi_default_intra_quant[] = -{ - 2048, 5681, 6355, 6623, 6656, 5431, 4018, 2401, - 5681, 7880, 10207, 10021, 9587, 8091, 6534, 3625, - 6355, 10207, 11363, 10619, 9700, 8935, 6155, 3507, - 6623, 9186, 10226, 9557, 8730, 8041, 6028, 3322, - 5632, 9232, 9031, 8730, 8192, 7040, 5542, 3390, - 5230, 7533, 7621, 7568, 7040, 6321, 5225, 3219, - 3602, 5189, 5250, 5539, 5265, 5007, 4199, 2638, - 1907, 2841, 3230, 3156, 3249, 3108, 2638, 1617 -}; -#endif /***************************************************************************** * pi_default_nonintra_quant : default quantization matrix *****************************************************************************/ -#ifndef VDEC_DFT -int pi_default_nonintra_quant[] = +u8 pi_default_nonintra_quant[] = { 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, @@ -117,19 +102,6 @@ int pi_default_nonintra_quant[] = 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16 }; -#else -int pi_default_nonintra_quanit[] = -{ - 4096, 5680, 5344, 4816, 4096, 3216, 2224, 1136, - 5680, 7888, 7424, 6688, 5680, 4464, 3072, 1568, - 5344, 7424, 6992, 6288, 5344, 4208, 2896, 1472, - 4816, 6688, 6288, 5664, 4816, 3792, 2608, 1328, - 4096, 5680, 5344, 4816, 4096, 3216, 2224, 1136, - 3216, 4464, 4208, 3792, 3216, 2528, 1744, 880, - 2224, 3072, 2896, 2608, 2224, 1744, 1200, 608, - 1136, 1568, 1472, 1328, 1136, 880, 608, 304 -}; -#endif /***************************************************************************** * pi_scan : zig-zag and alternate scan patterns @@ -211,14 +183,15 @@ static void __inline__ ReferenceReplace( vpar_thread_t * p_vpar, /***************************************************************************** * LoadMatrix : Load a quantization matrix *****************************************************************************/ -static __inline__ void LoadMatrix( vpar_thread_t * p_vpar, quant_matrix_t * p_matrix ) +static __inline__ void LoadMatrix( vpar_thread_t * p_vpar, + quant_matrix_t * p_matrix ) { int i_dummy; if( !p_matrix->b_allocated ) { /* Allocate a piece of memory to load the matrix. */ - if( (p_matrix->pi_matrix = (int *)malloc( 64*sizeof(int) )) == NULL ) + if( (p_matrix->pi_matrix = (u8 *)malloc( 64*sizeof(u8) )) == NULL ) { intf_ErrMsg( "vpar error: allocation error in LoadMatrix()" ); p_vpar->p_fifo->b_error = 1; @@ -229,7 +202,7 @@ static __inline__ void LoadMatrix( vpar_thread_t * p_vpar, quant_matrix_t * p_ma for( i_dummy = 0; i_dummy < 64; i_dummy++ ) { - p_matrix->pi_matrix[pi_scan[SCAN_ZIGZAG][i_dummy]] + p_matrix->pi_matrix[p_vpar->ppi_scan[SCAN_ZIGZAG][i_dummy]] = GetBits( &p_vpar->bit_stream, 8 ); } @@ -243,7 +216,7 @@ static __inline__ void LoadMatrix( vpar_thread_t * p_vpar, quant_matrix_t * p_ma /***************************************************************************** * LinkMatrix : Link a quantization matrix to another *****************************************************************************/ -static __inline__ void LinkMatrix( quant_matrix_t * p_matrix, int * pi_array ) +static __inline__ void LinkMatrix( quant_matrix_t * p_matrix, u8 * pi_array ) { if( p_matrix->b_allocated ) { @@ -366,7 +339,8 @@ static void SequenceHeader( vpar_thread_t * p_vpar ) else { /* Use default matrix. */ - LinkMatrix( &p_vpar->sequence.intra_quant, pi_default_intra_quant ); + LinkMatrix( &p_vpar->sequence.intra_quant, + p_vpar->pi_default_intra_quant ); } if( GetBits( &p_vpar->bit_stream, 1 ) ) /* load_non_intra_quantizer_matrix */ @@ -376,7 +350,8 @@ static void SequenceHeader( vpar_thread_t * p_vpar ) else { /* Use default matrix. */ - LinkMatrix( &p_vpar->sequence.nonintra_quant, pi_default_nonintra_quant ); + LinkMatrix( &p_vpar->sequence.nonintra_quant, + p_vpar->pi_default_nonintra_quant ); } /* Unless later overwritten by a matrix extension, we have the same @@ -905,7 +880,7 @@ static void QuantMatrixExtension( vpar_thread_t * p_vpar ) { /* Use the default matrix. */ LinkMatrix( &p_vpar->sequence.intra_quant, - pi_default_intra_quant ); + p_vpar->pi_default_intra_quant ); } if( GetBits( &p_vpar->bit_stream, 1 ) ) { @@ -916,7 +891,7 @@ static void QuantMatrixExtension( vpar_thread_t * p_vpar ) { /* Use the default matrix. */ LinkMatrix( &p_vpar->sequence.nonintra_quant, - pi_default_nonintra_quant ); + p_vpar->pi_default_nonintra_quant ); } if( GetBits( &p_vpar->bit_stream, 1 ) ) { -- 2.39.2