Project: Theora

Revision: 15153

Author: tterribe

Date: 04 Aug 2008 14:37:55

• /trunk/theora/lib/enc/x86_32/recon_mmx.c

  ......	@@ -22,8 +22,8 @@
  22	22	static const __attribute__ ((aligned(8),used)) ogg_int64_t V128 = 0x8080808080808080LL;
  23	23
  24	24	static void copy8x8__mmx (unsigned char *src,
  25		unsigned char *dest,
  26		unsigned int stride)
  	25	unsigned char *dest,
  	26	unsigned int stride)
  27	27	{
  28	28	__asm__ __volatile__ (
  29	29	" .p2align 4 \n\t"
  ......	@@ -35,14 +35,14 @@
  35	35	" movq (%1, %2, 2), %%mm2 \n\t"
  36	36	" movq (%1, %%edi), %%mm3 \n\t"
  37	37
  38		" lea (%1, %2, 4), %1 \n\t"
  	38	" lea (%1, %2, 4), %1 \n\t"
  39	39
  40	40	" movq %%mm0, (%0) \n\t"
  41	41	" movq %%mm1, (%0, %2) \n\t"
  42	42	" movq %%mm2, (%0, %2, 2) \n\t"
  43	43	" movq %%mm3, (%0, %%edi) \n\t"
  44	44
  45		" lea (%0, %2, 4), %0 \n\t"
  	45	" lea (%0, %2, 4), %0 \n\t"
  46	46
  47	47	" movq (%1), %%mm0 \n\t"
  48	48	" movq (%1, %2), %%mm1 \n\t"
  ......	@@ -61,7 +61,7 @@
  61	61	}
  62	62
  63	63	static void recon_intra8x8__mmx (unsigned char *ReconPtr, ogg_int16_t *ChangePtr,
  64		ogg_uint32_t LineStep)
  	64	ogg_uint32_t LineStep)
  65	65	{
  66	66	__asm__ __volatile__ (
  67	67	" .p2align 4 \n\t"
  ......	@@ -69,11 +69,11 @@
  69	69	" movq %[V128], %%mm0 \n\t" /* Set mm0 to 0x8080808080808080 */
  70	70
  71	71	" lea 128(%1), %%edi \n\t" /* Endpoint in input buffer */
  72		"1: \n\t"
  	72	"1: \n\t"
  73	73	" movq (%1), %%mm2 \n\t" /* First four input values */
  74	74
  75	75	" packsswb 8(%1), %%mm2 \n\t" /* pack with next(high) four values */
  76		" por %%mm0, %%mm0 \n\t"
  	76	" por %%mm0, %%mm0 \n\t"
  77	77	" pxor %%mm0, %%mm2 \n\t" /* Convert result to unsigned (same as add 128) */
  78	78	" lea 16(%1), %1 \n\t" /* Step source buffer */
  79	79	" cmp %%edi, %1 \n\t" /* are we done */
  ......	@@ -91,7 +91,7 @@
  91	91	}
  92	92
  93	93	static void recon_inter8x8__mmx (unsigned char *ReconPtr, unsigned char *RefPtr,
  94		ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
  	94	ogg_int16_t *ChangePtr, ogg_uint32_t LineStep)
  95	95	{
  96	96	__asm__ __volatile__ (
  97	97	" .p2align 4 \n\t"
  ......	@@ -127,8 +127,8 @@
  127	127	}
  128	128
  129	129	static void recon_inter8x8_half__mmx (unsigned char *ReconPtr, unsigned char *RefPtr1,
  130		unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
  131		ogg_uint32_t LineStep)
  	130	unsigned char *RefPtr2, ogg_int16_t *ChangePtr,
  	131	ogg_uint32_t LineStep)
  132	132	{
  133	133	__asm__ __volatile__ (
  134	134	" .p2align 4 \n\t"
  ......	@@ -173,7 +173,6 @@
  173	173
  174	174	void dsp_mmx_recon_init(DspFunctions *funcs)
  175	175	{
  176		TH_DEBUG("enabling accelerated x86_32 mmx recon functions.\n");
  177	176	funcs->copy8x8 = copy8x8__mmx;
  178	177	funcs->recon_intra8x8 = recon_intra8x8__mmx;
  179	178	funcs->recon_inter8x8 = recon_inter8x8__mmx;

• /trunk/theora/lib/cpu.c

  ......	@@ -220,24 +220,6 @@
  220	220	/*Implement me.*/
  221	221	flags=0;
  222	222	}
  223		# if defined(DEBUG)
  224		if(flags){
  225		TH_DEBUG("vectorized instruction sets supported:");
  226		if(flags&OC_CPU_X86_MMX)TH_DEBUG(" mmx");
  227		if(flags&OC_CPU_X86_MMXEXT)TH_DEBUG(" mmxext");
  228		if(flags&OC_CPU_X86_SSE)TH_DEBUG(" sse");
  229		if(flags&OC_CPU_X86_SSE2)TH_DEBUG(" sse2");
  230		if(flags&OC_CPU_X86_3DNOW)TH_DEBUG(" 3dnow");
  231		if(flags&OC_CPU_X86_3DNOWEXT)TH_DEBUG(" 3dnowext");
  232		if(flags&OC_CPU_X86_PNI)TH_DEBUG(" pni");
  233		if(flags&OC_CPU_X86_SSSE3)TH_DEBUG(" ssse3");
  234		if(flags&OC_CPU_X86_SSE4_1)TH_DEBUG(" sse4_1");
  235		if(flags&OC_CPU_X86_SSE4_2)TH_DEBUG(" sse4_2");
  236		if(flags&OC_CPU_X86_SSE4A)TH_DEBUG(" sse4a");
  237		if(flags&OC_CPU_X86_SSE5)TH_DEBUG(" sse5");
  238		TH_DEBUG("\n");
  239		}
  240		# endif
  241	223	return flags;
  242	224	}
  243	225	#endif

• /trunk/theora/lib/enc/x86_32/dct_decode_mmx.c

  ......	@@ -27,7 +27,7 @@
  27	27	0x0004000400040004LL;
  28	28
  29	29	static void loop_filter_v(unsigned char *_pix,int _ystride,
  30		const ogg_int16_t *_ll){
  	30	const ogg_int16_t *_ll){
  31	31	long esi;
  32	32	_pix-=_ystride*2;
  33	33	__asm__ __volatile__(
  ......	@@ -210,7 +210,7 @@
  210	210	four p0's to one register we must transpose the values in four mmx regs.
  211	211	When half is done we repeat this for the rest.*/
  212	212	static void loop_filter_h4(unsigned char *_pix,long _ystride,
  213		const ogg_int16_t *_ll){
  	213	const ogg_int16_t *_ll){
  214	214	long esi;
  215	215	long edi;
  216	216	__asm__ __volatile__(
  ......	@@ -343,12 +343,12 @@
  343	343	}
  344	344
  345	345	static void loop_filter_h(unsigned char *_pix,int _ystride,
  346		const ogg_int16_t *_ll){
  	346	const ogg_int16_t *_ll){
  347	347	_pix-=2;
  348	348	loop_filter_h4(_pix,_ystride,_ll);
  349	349	loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
  350	350	}
  351
  	351
  352	352	static void loop_filter_mmx(PB_INSTANCE *pbi, int FLimit){
  353	353	int j;
  354	354	ogg_int16_t __attribute__((aligned(8))) ll[4];
  ......	@@ -359,7 +359,7 @@
  359	359	ll[0]=ll[1]=ll[2]=ll[3]=FLimit;
  360	360
  361	361	for ( j = 0; j < 3 ; j++){
  362		ogg_uint32_t *bp_begin = bp;
  	362	ogg_uint32_t *bp_begin = bp;
  363	363	ogg_uint32_t *bp_end;
  364	364	int stride;
  365	365	int h;
  ......	@@ -376,23 +376,23 @@
  376	376	stride = pbi->UVStride;
  377	377	break;
  378	378	}
  379
  	379
  380	380	while(bp<bp_end){
  381	381	ogg_uint32_t *bp_left = bp;
  382	382	ogg_uint32_t *bp_right = bp + h;
  383	383	while(bp<bp_right){
  384		if(cp[0]){
  385		if(bp>bp_left)
  386		loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
  387		if(bp_left>bp_begin)
  388		loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
  389		if(bp+1<bp_right && !cp[1])
  390		loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
  391		if(bp+h<bp_end && !cp[h])
  392		loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
  393		}
  394		bp++;
  395		cp++;
  	384	if(cp[0]){
  	385	if(bp>bp_left)
  	386	loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
  	387	if(bp_left>bp_begin)
  	388	loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
  	389	if(bp+1<bp_right && !cp[1])
  	390	loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
  	391	if(bp+h<bp_end && !cp[h])
  	392	loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
  	393	}
  	394	bp++;
  	395	cp++;
  396	396	}
  397	397	}
  398	398	}

• /trunk/theora/lib/dec/dequant.c

  ......	@@ -127,65 +127,6 @@
  127	127	}
  128	128	while(qri-->0);
  129	129	}
  130
  131		#ifdef _TH_DEBUG_
  132		/* dump the tables */
  133		{
  134		int i, j, k, l, m;
  135		TH_DEBUG("loop filter limits = {");
  136		for(i=0;i<64;){
  137		TH_DEBUG("\n ");
  138		for(j=0;j<16;i++,j++)
  139		TH_DEBUG("%3d ",_qinfo->loop_filter_limits[i]);
  140		}
  141		TH_DEBUG("\n}\n\n");
  142
  143		TH_DEBUG("ac scale = {");
  144		for(i=0;i<64;){
  145		TH_DEBUG("\n ");
  146		for(j=0;j<16;i++,j++)
  147		TH_DEBUG("%3d ",_qinfo->ac_scale[i]);
  148		}
  149		TH_DEBUG("\n}\n\n");
  150
  151		TH_DEBUG("dc scale = {");
  152		for(i=0;i<64;){
  153		TH_DEBUG("\n ");
  154		for(j=0;j<16;i++,j++)
  155		TH_DEBUG("%3d ",_qinfo->dc_scale[i]);
  156		}
  157		TH_DEBUG("\n}\n\n");
  158
  159		for(k=0;k<2;k++)
  160		for(l=0;l<3;l++){
  161		char *name[2][3]={
  162		{"intra Y bases","intra U bases", "intra V bases"},
  163		{"inter Y bases","inter U bases", "inter V bases"}
  164		};
  165
  166		th_quant_ranges *r = &_qinfo->qi_ranges[k][l];
  167		TH_DEBUG("%s = {\n",name[k][l]);
  168		TH_DEBUG(" ranges = %d\n",r->nranges);
  169		TH_DEBUG(" intervals = { ");
  170		for(i=0;i<r->nranges;i++)
  171		TH_DEBUG("%3d ",r->sizes[i]);
  172		TH_DEBUG("}\n");
  173		TH_DEBUG("\n matricies = { ");
  174		for(m=0;m<r->nranges+1;m++){
  175		TH_DEBUG("\n { ");
  176		for(i=0;i<64;){
  177		TH_DEBUG("\n ");
  178		for(j=0;j<8;i++,j++)
  179		TH_DEBUG("%3d ",r->base_matrices[m][i]);
  180		}
  181		TH_DEBUG("\n }");
  182		}
  183		TH_DEBUG("\n }\n");
  184		}
  185		}
  186
  187		#endif
  188
  189	130	_ogg_free(base_mats);
  190	131	return 0;
  191	132	}
  ......	@@ -227,4 +168,3 @@
  227	168	_ogg_free((void *)_qinfo->qi_ranges[qti][pli].base_matrices);
  228	169	}
  229	170	}
  230

• /trunk/theora/lib/enc/encoder_toplevel.c

  ......	@@ -26,18 +26,13 @@
  26	26	#include "dsp.h"
  27	27	#include "codec_internal.h"
  28	28
  29		#ifdef _TH_DEBUG_
  30		FILE *debugout=NULL;
  31		long dframe=0;
  32		#endif
  33
  34	29	#define A_TABLE_SIZE 29
  35	30	#define DF_CANDIDATE_WINDOW 5
  36	31
  37	32	/*
  38		* th_quant_info for VP3
  	33	* th_quant_info for VP3
  39	34	*/
  40
  	35
  41	36	/*The default quantization parameters used by VP3.1.*/
  42	37	static const int OC_VP31_RANGE_SIZES[1]={63};
  43	38	static const th_quant_base OC_VP31_BASES_INTRA_Y[2]={
  ......	@@ -897,10 +892,6 @@
  897	892
  898	893	CP_INSTANCE *cpi;
  899	894
  900		#ifdef _TH_DEBUG_
  901		debugout=fopen("theoraenc-debugout.txt","w");
  902		#endif
  903
  904	895	memset(th, 0, sizeof(*th));
  905	896	/*Currently only the 4:2:0 format is supported.*/
  906	897	if(c->pixelformat!=OC_PF_420)return OC_IMPL;
  ......	@@ -1043,7 +1034,7 @@
  1043	1034	current clip. */
  1044	1035	cpi->ThisIsFirstFrame = 1;
  1045	1036	cpi->readyflag = 1;
  1046
  	1037
  1047	1038	cpi->pb.HeadersWritten = 0;
  1048	1039	/*We overload this flag to track header output.*/
  1049	1040	cpi->doneflag=-3;
  ......	@@ -1111,7 +1102,7 @@
  1111	1102	if(cpi->LastKeyFrame >= (ogg_uint32_t)
  1112	1103	cpi->pb.info.keyframe_frequency_force)
  1113	1104	cpi->ThisIsKeyFrame = 1;
  1114
  	1105
  1115	1106	if ( cpi->ThisIsKeyFrame ) {
  1116	1107	CompressKeyFrame(cpi);
  1117	1108	cpi->ThisIsKeyFrame = 0;
  ......	@@ -1131,10 +1122,6 @@
  1131	1122	((cpi->CurrentFrame - cpi->LastKeyFrame)<<cpi->pb.keyframe_granule_shift)+
  1132	1123	cpi->LastKeyFrame - 1;
  1133	1124
  1134		#ifdef _TH_DEBUG_
  1135		dframe++;
  1136		#endif
  1137
  1138	1125	return 0;
  1139	1126	}
  1140	1127
  ......	@@ -1170,7 +1157,7 @@
  1170	1157
  1171	1158	static void _tp_writelsbint(oggpack_buffer *opb, long value)
  1172	1159	{
  1173		oggpackB_write(opb, value&0xFF, 8);
  	1160	oggpackB_write(opb, value&0xFF, 8);
  1174	1161	oggpackB_write(opb, value>>8&0xFF, 8);
  1175	1162	oggpackB_write(opb, value>>16&0xFF, 8);
  1176	1163	oggpackB_write(opb, value>>24&0xFF, 8);
  ......	@@ -1197,7 +1184,7 @@
  1197	1184	/* Applications use offset_y to mean offset from the top of the image; the
  1198	1185	* meaning in the bitstream is the opposite (from the bottom). Transform.
  1199	1186	*/
  1200		offset_y = cpi->pb.info.height - cpi->pb.info.frame_height -
  	1187	offset_y = cpi->pb.info.height - cpi->pb.info.frame_height -
  1201	1188	cpi->pb.info.offset_y;
  1202	1189	oggpackB_write(cpi->oggbuffer,offset_y,8);
  1203	1190
  ......	@@ -1321,11 +1308,6 @@
  1321	1308	_ogg_free(cpi);
  1322	1309	}
  1323	1310
  1324		#ifdef _TH_DEBUG_
  1325		fclose(debugout);
  1326		debugout=NULL;
  1327		#endif
  1328
  1329	1311	memset(th,0,sizeof(*th));
  1330	1312	}
  1331	1313
  ......	@@ -1377,59 +1359,59 @@
  1377	1359	CP_INSTANCE *cpi;
  1378	1360	PB_INSTANCE *pbi;
  1379	1361	int value;
  1380
  	1362
  1381	1363	if(th == NULL)
  1382	1364	return TH_EFAULT;
  1383	1365
  1384	1366	cpi = th->internal_encode;
  1385	1367	pbi = &cpi->pb;
  1386
  	1368
  1387	1369	switch(req) {
  1388	1370	case TH_ENCCTL_SET_QUANT_PARAMS:
  1389	1371	if( ( buf==NULL&&buf_sz!=0 )
  1390		|| ( buf!=NULL&&buf_sz!=sizeof(th_quant_info) )
  1391		|| cpi->pb.HeadersWritten ){
  	1372	|| ( buf!=NULL&&buf_sz!=sizeof(th_quant_info) )
  	1373	|| cpi->pb.HeadersWritten ){
  1392	1374	return TH_EINVAL;
  1393	1375	}
  1394
  	1376
  1395	1377	memcpy(&pbi->quant_info, buf, sizeof(th_quant_info));
  1396	1378	InitQTables(pbi);
  1397
  	1379
  1398	1380	return 0;
  1399	1381	case TH_ENCCTL_SET_VP3_COMPATIBLE:
  1400	1382	if(cpi->pb.HeadersWritten)
  1401	1383	return TH_EINVAL;
  1402
  	1384
  1403	1385	memcpy(&pbi->quant_info, &TH_VP31_QUANT_INFO, sizeof(th_quant_info));
  1404	1386	InitQTables(pbi);
  1405
  	1387
  1406	1388	return 0;
  1407	1389	case TH_ENCCTL_SET_SPLEVEL:
  1408	1390	if(buf == NULL || buf_sz != sizeof(int))
  1409	1391	return TH_EINVAL;
  1410
  	1392
  1411	1393	memcpy(&value, buf, sizeof(int));
  1412
  	1394
  1413	1395	switch(value) {
  1414	1396	case 0:
  1415	1397	cpi->MotionCompensation = 1;
  1416	1398	pbi->info.quick_p = 0;
  1417	1399	break;
  1418
  	1400
  1419	1401	case 1:
  1420	1402	cpi->MotionCompensation = 1;
  1421	1403	pbi->info.quick_p = 1;
  1422	1404	break;
  1423
  	1405
  1424	1406	case 2:
  1425	1407	cpi->MotionCompensation = 0;
  1426	1408	pbi->info.quick_p = 1;
  1427	1409	break;
  1428
  	1410
  1429	1411	default:
  1430		return TH_EINVAL;
  	1412	return TH_EINVAL;
  1431	1413	}
  1432
  	1414
  1433	1415	return 0;
  1434	1416	case TH_ENCCTL_GET_SPLEVEL_MAX:
  1435	1417	value = 2;

• /trunk/theora/lib/enc/x86_64/dct_decode_mmx.c

  ......	@@ -27,7 +27,7 @@
  27	27	0x0004000400040004LL;
  28	28
  29	29	static void loop_filter_v(unsigned char *_pix,int _ystride,
  30		const ogg_int16_t *_ll){
  	30	const ogg_int16_t *_ll){
  31	31	long esi;
  32	32	_pix-=_ystride*2;
  33	33	__asm__ __volatile__(
  ......	@@ -210,7 +210,7 @@
  210	210	four p0's to one register we must transpose the values in four mmx regs.
  211	211	When half is done we repeat this for the rest.*/
  212	212	static void loop_filter_h4(unsigned char *_pix,long _ystride,
  213		const ogg_int16_t *_ll){
  	213	const ogg_int16_t *_ll){
  214	214	long esi;
  215	215	long edi;
  216	216	__asm__ __volatile__(
  ......	@@ -343,12 +343,12 @@
  343	343	}
  344	344
  345	345	static void loop_filter_h(unsigned char *_pix,int _ystride,
  346		const ogg_int16_t *_ll){
  	346	const ogg_int16_t *_ll){
  347	347	_pix-=2;
  348	348	loop_filter_h4(_pix,_ystride,_ll);
  349	349	loop_filter_h4(_pix+(_ystride<<2),_ystride,_ll);
  350	350	}
  351
  	351
  352	352	static void loop_filter_mmx(PB_INSTANCE *pbi, int FLimit){
  353	353	int j;
  354	354	ogg_int16_t __attribute__((aligned(8))) ll[4];
  ......	@@ -359,7 +359,7 @@
  359	359	ll[0]=ll[1]=ll[2]=ll[3]=FLimit;
  360	360
  361	361	for ( j = 0; j < 3 ; j++){
  362		ogg_uint32_t *bp_begin = bp;
  	362	ogg_uint32_t *bp_begin = bp;
  363	363	ogg_uint32_t *bp_end;
  364	364	int stride;
  365	365	int h;
  ......	@@ -376,23 +376,23 @@
  376	376	stride = pbi->UVStride;
  377	377	break;
  378	378	}
  379
  	379
  380	380	while(bp<bp_end){
  381	381	ogg_uint32_t *bp_left = bp;
  382	382	ogg_uint32_t *bp_right = bp + h;
  383	383	while(bp<bp_right){
  384		if(cp[0]){
  385		if(bp>bp_left)
  386		loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
  387		if(bp_left>bp_begin)
  388		loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
  389		if(bp+1<bp_right && !cp[1])
  390		loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
  391		if(bp+h<bp_end && !cp[h])
  392		loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
  393		}
  394		bp++;
  395		cp++;
  	384	if(cp[0]){
  	385	if(bp>bp_left)
  	386	loop_filter_h(&pbi->LastFrameRecon[bp[0]],stride,ll);
  	387	if(bp_left>bp_begin)
  	388	loop_filter_v(&pbi->LastFrameRecon[bp[0]],stride,ll);
  	389	if(bp+1<bp_right && !cp[1])
  	390	loop_filter_h(&pbi->LastFrameRecon[bp[0]]+8,stride,ll);
  	391	if(bp+h<bp_end && !cp[h])
  	392	loop_filter_v(&pbi->LastFrameRecon[bp[h]],stride,ll);
  	393	}
  	394	bp++;
  	395	cp++;
  396	396	}
  397	397	}
  398	398	}

• /trunk/theora/lib/enc/encode.c

  ......	@@ -28,10 +28,10 @@
  28	28	#define HIGHBITDUPPED(X) (((ogg_int16_t) X) >> 15)
  29	29
  30	30	static ogg_uint32_t QuadCodeComponent ( CP_INSTANCE *cpi,
  31		ogg_uint32_t FirstSB,
  32		ogg_uint32_t SBRows,
  33		ogg_uint32_t SBCols,
  34		ogg_uint32_t PixelsPerLine){
  	31	ogg_uint32_t FirstSB,
  	32	ogg_uint32_t SBRows,
  	33	ogg_uint32_t SBCols,
  	34	ogg_uint32_t PixelsPerLine){
  35	35
  36	36	ogg_int32_t FragIndex; /* Fragment number */
  37	37	ogg_uint32_t MB, B; /* Macro-Block, Block indices */
  ......	@@ -49,7 +49,7 @@
  49	49	for ( SBcol=0; SBcol<SBCols; SBcol++ ) {
  50	50	/* Check its four Macro-Blocks */
  51	51	/* 'Macro-Block' is a misnomer in the chroma planes; this is
  52		really just a Hilbert curve iterator */
  	52	really just a Hilbert curve iterator */
  53	53	for ( MB=0; MB<4; MB++ ) {
  54	54
  55	55	if ( QuadMapToMBTopLeft(cpi->pb.BlockMap,SB,MB) >= 0 ) {
  ......	@@ -359,31 +359,15 @@
  359	359	/* Add the appropriate mode entropy token. */
  360	360	ModeIndex = SchemeList[cpi->ModeList[i]];
  361	361	oggpackB_write( opb, ModeBitPatterns[ModeIndex],
  362		(ogg_uint32_t)ModeBitLengths[ModeIndex] );
  	362	(ogg_uint32_t)ModeBitLengths[ModeIndex] );
  363	363	}
  364	364	}else{
  365	365	/* Fall back to MODE_BITS per entry */
  366	366	for ( i = 0; i < cpi->ModeListCount; i++)
  367	367	/* Add the appropriate mode entropy token. */
  368		oggpackB_write( opb, cpi->ModeList[i], MODE_BITS );
  	368	oggpackB_write( opb, cpi->ModeList[i], MODE_BITS );
  369	369	}
  370
  371		#ifdef _TH_DEBUG_
  372		TH_DEBUG("mode encode scheme = %d\n",(int)BestScheme);
  373		if ( BestScheme == 0 ) {
  374		TH_DEBUG("mode scheme list = { ");
  375		for ( j = 0; j < MAX_MODES; j++ )
  376		TH_DEBUG("%d ",(int)BestModeSchemes[j]);
  377		TH_DEBUG("}\n");
  378		}
  379		TH_DEBUG("mode list = { ");
  380		for ( i = 0; i < cpi->ModeListCount; i++) {
  381		if((i&0x1f)==0)
  382		TH_DEBUG("\n ");
  383		TH_DEBUG("%d ",cpi->ModeList[i]);
  384		}
  385		TH_DEBUG("\n}\n");
  386		#endif
  	370
  387	371	}
  388	372
  389	373	static void PackMotionVectors (CP_INSTANCE *cpi) {
  ......	@@ -422,15 +406,6 @@
  422	406	(ogg_uint32_t)MvBitsPtr[cpi->MVList[i].y] );
  423	407	}
  424	408
  425		#ifdef _TH_DEBUG_
  426		TH_DEBUG("motion vectors = {");
  427		for ( i = 0; i < (ogg_int32_t)cpi->MvListCount; i++ ) {
  428		if((i&0x7)==0)
  429		TH_DEBUG("\n ");
  430		TH_DEBUG("%+03d,%+03d ",cpi->MVList[i].x,cpi->MVList[i].y);
  431		}
  432		TH_DEBUG("\n}\n");
  433		#endif
  434	409	}
  435	410
  436	411	static void PackEOBRun( CP_INSTANCE *cpi) {
  ......	@@ -905,17 +880,6 @@
  905	880	}
  906	881	}
  907	882
  908		#ifdef _TH_DEBUG_
  909		{
  910		int j;
  911		for ( i = 0; i < cpi->pb.CodedBlockIndex; i++ ) {
  912		FragIndex = cpi->pb.CodedBlockList[i];
  913		for(j=0;j<64;j++)
  914		cpi->pb.QFragQUAN[FragIndex][j] = cpi->pb.QFragData[FragIndex][j];
  915		}
  916		}
  917		#endif
  918
  919	883	/* Pack DC tokens and adjust the ones we couldn't predict 2d */
  920	884	for ( i = 0; i < cpi->pb.CodedBlockIndex; i++ ) {
  921	885	/* Get the linear index for the current coded fragment. */
  ......	@@ -1013,7 +977,7 @@
  1013	977	cpi->pb.FragCodingMethod[cpi->pb.YPlaneFragments +
  1014	978	cpi->pb.UVPlaneFragments + UVFragOffset] =
  1015	979	cpi->MBCodingMode;
  1016		}
  	980	}
  1017	981	}
  1018	982
  1019	983	/* Next Super-Block */
  ......	@@ -1391,7 +1355,7 @@
  1391	1355
  1392	1356	cpi->MBCodingMode = CODE_INTER_PLUS_MV;
  1393	1357	SetMBMotionVectorsAndMode(cpi,YFragIndex,UFragIndex,
  1394		VFragIndex,&InterMVect);
  	1358	VFragIndex,&InterMVect);
  1395	1359
  1396	1360	/* Update Prior last mv with last mv */
  1397	1361	PriorLastInterMVect.x = LastInterMVect.x;
  ......	@@ -1407,7 +1371,7 @@
  1407	1371
  1408	1372	cpi->MBCodingMode = CODE_GOLDEN_MV;
  1409	1373	SetMBMotionVectorsAndMode(cpi,YFragIndex,UFragIndex,
  1410		VFragIndex,&GFMVect);
  	1374	VFragIndex,&GFMVect);
  1411	1375
  1412	1376	/* Note last inter GF MV for future use */
  1413	1377	LastGFMVect.x = GFMVect.x;
  ......	@@ -1463,7 +1427,7 @@
  1463	1427
  1464	1428	cpi->MBCodingMode = CODE_INTRA;
  1465	1429	SetMBMotionVectorsAndMode(cpi,YFragIndex,UFragIndex,
  1466		VFragIndex,&ZeroVect);
  	1430	VFragIndex,&ZeroVect);
  1467	1431	}
  1468	1432
  1469	1433
  ......	@@ -1487,17 +1451,11 @@
  1487	1451	void WriteFrameHeader( CP_INSTANCE *cpi) {
  1488	1452	ogg_uint32_t i;
  1489	1453	oggpack_buffer *opb=cpi->oggbuffer;
  1490
  1491		TH_DEBUG("\n>>>> beginning frame %ld\n\n",dframe);
  1492
  1493	1454	/* Output the frame type (base/key frame or inter frame) */
  1494	1455	oggpackB_write( opb, cpi->pb.FrameType, 1 );
  1495		TH_DEBUG("frame type = video, %s\n",cpi->pb.FrameType?"predicted":"key");
  1496
  1497	1456	/* Write out details of the current value of Q... variable resolution. */
  1498	1457	for ( i = 0; i < Q_TABLE_SIZE; i++ ) {
  1499	1458	if ( cpi->pb.ThisFrameQualityValue == cpi->pb.QThreshTable[i] ) {
  1500		TH_DEBUG("frame quality = { %d }\n",i);
  1501	1459	oggpackB_write( opb, i, 6 );
  1502	1460	break;
  1503	1461	}

• /trunk/theora/lib/enc/x86_32/idct_mmx.c

  ......	@@ -401,7 +401,7 @@
  401	401	" paddsw "r0","r0"\n" \
  402	402	" movq "r1","I(1)"\n" /* save R1 */ \
  403	403	" paddsw "r7","r0"\n" /* r0 = R0 = G. + C. */ \
  404		"#end RowIDCT" \
  	404	"#end RowIDCT" \
  405	405	);
  406	406	// end RowIDCT macro (8 + 38 = 46 cycles)
  407	407
  ......	@@ -465,7 +465,7 @@
  465	465	" movq "r5","J(5)"\n" /* store NR5 at J5 */ \
  466	466	" movq "r7","J(7)"\n" /* store NR7 at J7 */ \
  467	467	" movq "r0","I(0)"\n" /* store NR0 at I0 */ \
  468		"#end ColumnIDCT\n" \
  	468	"#end ColumnIDCT\n" \
  469	469	);
  470	470	// end ColumnIDCT macro (38 + 19 = 57 cycles)
  471	471
  ......	@@ -559,7 +559,7 @@
  559	559	" movq "r4","I(3)"\n" \
  560	560	\
  561	561	" movq "r2","I(2)"\n" \
  562		"#end Transpose\n" \
  	562	"#end Transpose\n" \
  563	563	);
  564	564	// end Transpose macro (19 cycles).
  565	565
  ......	@@ -1013,7 +1013,7 @@
  1013	1013	" paddsw "r0","r0"\n" \
  1014	1014	" movq "r1","I(1)"\n" /* save R1 */ \
  1015	1015	" paddsw "r7","r0"\n" /* r0 = R0 = G. + C. */ \
  1016		"#end RowIDCT_10\n" \
  	1016	"#end RowIDCT_10\n" \
  1017	1017	);
  1018	1018	// end RowIDCT macro (8 + 38 = 46 cycles)
  1019	1019
  ......	@@ -1060,7 +1060,7 @@
  1060	1060	" movq "r7","J(7)"\n" /* store NR7 at J7 */ \
  1061	1061	\
  1062	1062	" movq "r0","I(0)"\n" /* store NR0 at I0 */ \
  1063		"#end ColumnIDCT_10\n" \
  	1063	"#end ColumnIDCT_10\n" \
  1064	1064	);
  1065	1065	// end ColumnIDCT macro (38 + 19 = 57 cycles)
  1066	1066	/* --------------------------------------------------------------- */
  ......	@@ -1389,7 +1389,7 @@
  1389	1389	);
  1390	1390
  1391	1391	ASM(
  1392		"movq (%eax), "r0"\n"
  	1392	"movq (%eax), "r0"\n"
  1393	1393	"pmullw (%esi), "r0"\n" /* r0 = 03 02 01 00 */
  1394	1394	"movq "M(0)", "r2"\n" /* r2 = __ __ __ FF */
  1395	1395	"movq "r0", "r3"\n" /* r3 = 03 02 01 00 */
  ......	@@ -1444,7 +1444,6 @@
  1444	1444	/* install our implementation in the function table */
  1445	1445	void dsp_mmx_idct_init(DspFunctions *funcs)
  1446	1446	{
  1447		TH_DEBUG("enabling accelerated x86_32 mmx idct functions.\n");
  1448	1447	funcs->IDctSlow = IDctSlow__mmx;
  1449	1448	funcs->IDct10 = IDct10__mmx;
  1450	1449	funcs->IDct3 = IDct3__mmx;

• /trunk/theora/lib/enc/x86_32/dsp_mmx.c

  ......	@@ -50,12 +50,12 @@
  50	50
  51	51	static void sub8x8__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr,
  52	52	ogg_int16_t *DctInputPtr, ogg_uint32_t PixelsPerLine,
  53		ogg_uint32_t ReconPixelsPerLine)
  	53	ogg_uint32_t ReconPixelsPerLine)
  54	54	{
  55	55	__asm__ __volatile__ (
  56	56	" .p2align 4 \n\t"
  57	57
  58		" pxor %%mm7, %%mm7 \n\t"
  	58	" pxor %%mm7, %%mm7 \n\t"
  59	59	SUB_LOOP
  60	60	SUB_LOOP
  61	61	SUB_LOOP
  ......	@@ -68,7 +68,7 @@
  68	68	"+r" (ReconPtr),
  69	69	"+r" (DctInputPtr)
  70	70	: "m" (PixelsPerLine),
  71		"m" (ReconPixelsPerLine)
  	71	"m" (ReconPixelsPerLine)
  72	72	: "memory"
  73	73	);
  74	74	}
  ......	@@ -86,16 +86,16 @@
  86	86	" movq %%mm2, 8(%1) \n\t" /* write answer out */ \
  87	87	/* Increment pointers */ \
  88	88	" add $16, %1 \n\t" \
  89		" add %2, %0 \n\t"
  	89	" add %2, %0 \n\t"
  90	90
  91	91
  92	92	static void sub8x8_128__mmx (unsigned char *FiltPtr, ogg_int16_t *DctInputPtr,
  93		ogg_uint32_t PixelsPerLine)
  	93	ogg_uint32_t PixelsPerLine)
  94	94	{
  95	95	__asm__ __volatile__ (
  96	96	" .p2align 4 \n\t"
  97	97
  98		" pxor %%mm7, %%mm7 \n\t"
  	98	" pxor %%mm7, %%mm7 \n\t"
  99	99	" movq %[V128], %%mm1 \n\t"
  100	100	SUB_128_LOOP
  101	101	SUB_128_LOOP
  ......	@@ -140,18 +140,18 @@
  140	140	" add $16, %3 \n\t" \
  141	141	" add %4, %0 \n\t" \
  142	142	" add %5, %1 \n\t" \
  143		" add %5, %2 \n\t"
  	143	" add %5, %2 \n\t"
  144	144
  145	145
  146	146	static void sub8x8avg2__mmx (unsigned char *FiltPtr, unsigned char *ReconPtr1,
  147	147	unsigned char *ReconPtr2, ogg_int16_t *DctInputPtr,
  148	148	ogg_uint32_t PixelsPerLine,
  149		ogg_uint32_t ReconPixelsPerLine)
  	149	ogg_uint32_t ReconPixelsPerLine)
  150	150	{