summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/r300/radeon_program_pair.c
blob: 4aa2319a454ff1eeeba8aca4a0af8068df3f43f8 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
/*
 * Copyright (C) 2008 Nicolai Haehnle.
 *
 * All Rights Reserved.
 *
 * Permission is hereby granted, free of charge, to any person obtaining
 * a copy of this software and associated documentation files (the
 * "Software"), to deal in the Software without restriction, including
 * without limitation the rights to use, copy, modify, merge, publish,
 * distribute, sublicense, and/or sell copies of the Software, and to
 * permit persons to whom the Software is furnished to do so, subject to
 * the following conditions:
 *
 * The above copyright notice and this permission notice (including the
 * next paragraph) shall be included in all copies or substantial
 * portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 */

/**
 * @file
 *
 * Perform temporary register allocation and attempt to pair off instructions
 * in RGB and Alpha pairs. Also attempts to optimize the TEX instruction
 * vs. ALU instruction scheduling.
 */

#include "radeon_program_pair.h"

#include "radeon_context.h"

#include "shader/prog_print.h"

#define error(fmt, args...) do { \
	_mesa_problem(s->Ctx, "%s::%s(): " fmt "\n",	\
		__FILE__, __FUNCTION__, ##args);	\
	s->Error = GL_TRUE;				\
} while(0)

struct pair_state_instruction {
	GLuint IsTex:1; /**< Is a texture instruction */
	GLuint NeedRGB:1; /**< Needs the RGB ALU */
	GLuint NeedAlpha:1; /**< Needs the Alpha ALU */
	GLuint IsTranscendent:1; /**< Is a special transcendent instruction */

	/**
	 * Number of (read and write) dependencies that must be resolved before
	 * this instruction can be scheduled.
	 */
	GLuint NumDependencies:5;

	/**
	 * Next instruction in the linked list of ready instructions.
	 */
	struct pair_state_instruction *NextReady;

	/**
	 * Values that this instruction writes
	 */
	struct reg_value *Values[4];
};


/**
 * Used to keep track of which instructions read a value.
 */
struct reg_value_reader {
	GLuint IP; /**< IP of the instruction that performs this access */
	struct reg_value_reader *Next;
};

/**
 * Used to keep track which values are stored in each component of a
 * PROGRAM_TEMPORARY.
 */
struct reg_value {
	GLuint IP; /**< IP of the instruction that writes this value */
	struct reg_value *Next; /**< Pointer to the next value to be written to the same PROGRAM_TEMPORARY component */

	/**
	 * Unordered linked list of instructions that read from this value.
	 */
	struct reg_value_reader *Readers;

	/**
	 * Number of readers of this value. This is calculated during @ref scan_instructions
	 * and continually decremented during code emission.
	 * When this count reaches zero, the instruction that writes the @ref Next value
	 * can be scheduled.
	 */
	GLuint NumReaders;
};

/**
 * Used to translate a PROGRAM_INPUT or PROGRAM_TEMPORARY Mesa register
 * to the proper hardware temporary.
 */
struct pair_register_translation {
	GLuint Allocated:1;
	GLuint HwIndex:8;
	GLuint RefCount:23; /**< # of times this occurs in an unscheduled instruction SrcReg or DstReg */

	/**
	 * Notes the value that is currently contained in each component
	 * (only used for PROGRAM_TEMPORARY registers).
	 */
	struct reg_value *Value[4];
};

struct pair_state {
	GLcontext *Ctx;
	struct gl_program *Program;
	const struct radeon_pair_handler *Handler;
	GLboolean Error;
	GLboolean Debug;
	GLboolean Verbose;
	void *UserData;

	/**
	 * Translate Mesa registers to hardware registers
	 */
	struct pair_register_translation Inputs[FRAG_ATTRIB_MAX];
	struct pair_register_translation Temps[MAX_PROGRAM_TEMPS];

	/**
	 * Derived information about program instructions.
	 */
	struct pair_state_instruction *Instructions;

	struct {
		GLuint RefCount; /**< # of times this occurs in an unscheduled SrcReg or DstReg */
	} HwTemps[128];

	/**
	 * Linked list of instructions that can be scheduled right now,
	 * based on which ALU/TEX resources they require.
	 */
	struct pair_state_instruction *ReadyFullALU;
	struct pair_state_instruction *ReadyRGB;
	struct pair_state_instruction *ReadyAlpha;
	struct pair_state_instruction *ReadyTEX;

	/**
	 * Pool of @ref reg_value structures for fast allocation.
	 */
	struct reg_value *ValuePool;
	GLuint ValuePoolUsed;
	struct reg_value_reader *ReaderPool;
	GLuint ReaderPoolUsed;
};


static struct pair_register_translation *get_register(struct pair_state *s, GLuint file, GLuint index)
{
	switch(file) {
	case PROGRAM_TEMPORARY: return &s->Temps[index];
	case PROGRAM_INPUT: return &s->Inputs[index];
	default: return 0;
	}
}

static void alloc_hw_reg(struct pair_state *s, GLuint file, GLuint index, GLuint hwindex)
{
	struct pair_register_translation *t = get_register(s, file, index);
	ASSERT(!s->HwTemps[hwindex].RefCount);
	ASSERT(!t->Allocated);
	s->HwTemps[hwindex].RefCount = t->RefCount;
	t->Allocated = 1;
	t->HwIndex = hwindex;
}

static GLuint get_hw_reg(struct pair_state *s, GLuint file, GLuint index)
{
	GLuint hwindex;

	struct pair_register_translation *t = get_register(s, file, index);
	if (!t) {
		_mesa_problem(s->Ctx, "get_hw_reg: %i[%i]\n", file, index);
		return 0;
	}

	if (t->Allocated)
		return t->HwIndex;

	for(hwindex = 0; hwindex < s->Handler->MaxHwTemps; ++hwindex)
		if (!s->HwTemps[hwindex].RefCount)
			break;

	if (hwindex >= s->Handler->MaxHwTemps) {
		error("Ran out of hardware temporaries");
		return 0;
	}

	alloc_hw_reg(s, file, index, hwindex);
	return hwindex;
}


static void deref_hw_reg(struct pair_state *s, GLuint hwindex)
{
	if (!s->HwTemps[hwindex].RefCount) {
		error("Hwindex %i refcount error", hwindex);
		return;
	}

	s->HwTemps[hwindex].RefCount--;
}

static void add_pairinst_to_list(struct pair_state_instruction **list, struct pair_state_instruction *pairinst)
{
	pairinst->NextReady = *list;
	*list = pairinst;
}

/**
 * The instruction at the given IP has become ready. Link it into the ready
 * instructions.
 */
static void instruction_ready(struct pair_state *s, int ip)
{
	struct pair_state_instruction *pairinst = s->Instructions + ip;

	if (s->Verbose)
		_mesa_printf("instruction_ready(%i)\n", ip);

	if (pairinst->IsTex)
		add_pairinst_to_list(&s->ReadyTEX, pairinst);
	else if (!pairinst->NeedAlpha)
		add_pairinst_to_list(&s->ReadyRGB, pairinst);
	else if (!pairinst->NeedRGB)
		add_pairinst_to_list(&s->ReadyAlpha, pairinst);
	else
		add_pairinst_to_list(&s->ReadyFullALU, pairinst);
}


/**
 * Finally rewrite ADD, MOV, MUL as the appropriate native instruction
 * and reverse the order of arguments for CMP.
 */
static void final_rewrite(struct pair_state *s, struct prog_instruction *inst)
{
	struct prog_src_register tmp;

	switch(inst->Opcode) {
	case OPCODE_ADD:
		inst->SrcReg[2] = inst->SrcReg[1];
		inst->SrcReg[1].File = PROGRAM_BUILTIN;
		inst->SrcReg[1].Swizzle = SWIZZLE_1111;
		inst->SrcReg[1].Negate = NEGATE_NONE;
		inst->Opcode = OPCODE_MAD;
		break;
	case OPCODE_CMP:
		tmp = inst->SrcReg[2];
		inst->SrcReg[2] = inst->SrcReg[0];
		inst->SrcReg[0] = tmp;
		break;
	case OPCODE_MOV:
		/* AMD say we should use CMP.
		 * However, when we transform
		 *  KIL -r0;
		 * into
		 *  CMP tmp, -r0, -r0, 0;
		 *  KIL tmp;
		 * we get incorrect behaviour on R500 when r0 == 0.0.
		 * It appears that the R500 KIL hardware treats -0.0 as less
		 * than zero.
		 */
		inst->SrcReg[1].File = PROGRAM_BUILTIN;
		inst->SrcReg[1].Swizzle = SWIZZLE_1111;
		inst->SrcReg[2].File = PROGRAM_BUILTIN;
		inst->SrcReg[2].Swizzle = SWIZZLE_0000;
		inst->Opcode = OPCODE_MAD;
		break;
	case OPCODE_MUL:
		inst->SrcReg[2].File = PROGRAM_BUILTIN;
		inst->SrcReg[2].Swizzle = SWIZZLE_0000;
		inst->Opcode = OPCODE_MAD;
		break;
	default:
		/* nothing to do */
		break;
	}
}


/**
 * Classify an instruction according to which ALUs etc. it needs
 */
static void classify_instruction(struct pair_state *s,
	struct prog_instruction *inst, struct pair_state_instruction *pairinst)
{
	pairinst->NeedRGB = (inst->DstReg.WriteMask & WRITEMASK_XYZ) ? 1 : 0;
	pairinst->NeedAlpha = (inst->DstReg.WriteMask & WRITEMASK_W) ? 1 : 0;

	switch(inst->Opcode) {
	case OPCODE_ADD:
	case OPCODE_CMP:
	case OPCODE_DDX:
	case OPCODE_DDY:
	case OPCODE_FRC:
	case OPCODE_MAD:
	case OPCODE_MAX:
	case OPCODE_MIN:
	case OPCODE_MOV:
	case OPCODE_MUL:
		break;
	case OPCODE_COS:
	case OPCODE_EX2:
	case OPCODE_LG2:
	case OPCODE_RCP:
	case OPCODE_RSQ:
	case OPCODE_SIN:
		pairinst->IsTranscendent = 1;
		pairinst->NeedAlpha = 1;
		break;
	case OPCODE_DP4:
		pairinst->NeedAlpha = 1;
		/* fall through */
	case OPCODE_DP3:
		pairinst->NeedRGB = 1;
		break;
	case OPCODE_KIL:
	case OPCODE_TEX:
	case OPCODE_TXB:
	case OPCODE_TXP:
	case OPCODE_END:
		pairinst->IsTex = 1;
		break;
	default:
		error("Unknown opcode %d\n", inst->Opcode);
		break;
	}
}


/**
 * Count which (input, temporary) register is read and written how often,
 * and scan the instruction stream to find dependencies.
 */
static void scan_instructions(struct pair_state *s)
{
	struct prog_instruction *inst;
	struct pair_state_instruction *pairinst;
	GLuint ip;

	for(inst = s->Program->Instructions, pairinst = s->Instructions, ip = 0;
	    inst->Opcode != OPCODE_END;
	    ++inst, ++pairinst, ++ip) {
		final_rewrite(s, inst);
		classify_instruction(s, inst, pairinst);

		int nsrc = _mesa_num_inst_src_regs(inst->Opcode);
		int j;
		for(j = 0; j < nsrc; j++) {
			struct pair_register_translation *t =
				get_register(s, inst->SrcReg[j].File, inst->SrcReg[j].Index);
			if (!t)
				continue;

			t->RefCount++;

			if (inst->SrcReg[j].File == PROGRAM_TEMPORARY) {
				int i;
				for(i = 0; i < 4; ++i) {
					GLuint swz = GET_SWZ(inst->SrcReg[j].Swizzle, i);
					if (swz >= 4)
						continue; /* constant or NIL swizzle */
					if (!t->Value[swz])
						continue; /* this is an undefined read */

					/* Do not add a dependency if this instruction
					 * also rewrites the value. The code below adds
					 * a dependency for the DstReg, which is a superset
					 * of the SrcReg dependency. */
					if (inst->DstReg.File == PROGRAM_TEMPORARY &&
					    inst->DstReg.Index == inst->SrcReg[j].Index &&
					    GET_BIT(inst->DstReg.WriteMask, swz))
						continue;

					struct reg_value_reader* r = &s->ReaderPool[s->ReaderPoolUsed++];
					pairinst->NumDependencies++;
					t->Value[swz]->NumReaders++;
					r->IP = ip;
					r->Next = t->Value[swz]->Readers;
					t->Value[swz]->Readers = r;
				}
			}
		}

		int ndst = _mesa_num_inst_dst_regs(inst->Opcode);
		if (ndst) {
			struct pair_register_translation *t =
				get_register(s, inst->DstReg.File, inst->DstReg.Index);
			if (t) {
				t->RefCount++;

				if (inst->DstReg.File == PROGRAM_TEMPORARY) {
					int j;
					for(j = 0; j < 4; ++j) {
						if (!GET_BIT(inst->DstReg.WriteMask, j))
							continue;

						struct reg_value* v = &s->ValuePool[s->ValuePoolUsed++];
						v->IP = ip;
						if (t->Value[j]) {
							pairinst->NumDependencies++;
							t->Value[j]->Next = v;
						}
						t->Value[j] = v;
						pairinst->Values[j] = v;
					}
				}
			}
		}

		if (s->Verbose)
			_mesa_printf("scan(%i): NumDeps = %i\n", ip, pairinst->NumDependencies);

		if (!pairinst->NumDependencies)
			instruction_ready(s, ip);
	}

	/* Clear the PROGRAM_TEMPORARY state */
	int i, j;
	for(i = 0; i < MAX_PROGRAM_TEMPS; ++i) {
		for(j = 0; j < 4; ++j)
			s->Temps[i].Value[j] = 0;
	}
}


/**
 * Reserve hardware temporary registers for the program inputs.
 *
 * @note This allocation is performed explicitly, because the order of inputs
 * is determined by the RS hardware.
 */
static void allocate_input_registers(struct pair_state *s)
{
	GLuint InputsRead = s->Program->InputsRead;
	int i;
	GLuint hwindex = 0;

	/* Primary colour */
	if (InputsRead & FRAG_BIT_COL0)
		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_COL0, hwindex++);
	InputsRead &= ~FRAG_BIT_COL0;

	/* Secondary color */
	if (InputsRead & FRAG_BIT_COL1)
		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_COL1, hwindex++);
	InputsRead &= ~FRAG_BIT_COL1;

	/* Texcoords */
	for (i = 0; i < s->Ctx->Const.MaxTextureUnits; i++) {
		if (InputsRead & (FRAG_BIT_TEX0 << i))
			alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_TEX0+i, hwindex++);
	}
	InputsRead &= ~FRAG_BITS_TEX_ANY;

	/* Fogcoords treated as a texcoord */
	if (InputsRead & FRAG_BIT_FOGC)
		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_FOGC, hwindex++);
	InputsRead &= ~FRAG_BIT_FOGC;

	/* fragment position treated as a texcoord */
	if (InputsRead & FRAG_BIT_WPOS)
		alloc_hw_reg(s, PROGRAM_INPUT, FRAG_ATTRIB_WPOS, hwindex++);
	InputsRead &= ~FRAG_BIT_WPOS;

	/* Anything else */
	if (InputsRead)
		error("Don't know how to handle inputs 0x%x\n", InputsRead);
}


static void decrement_dependencies(struct pair_state *s, int ip)
{
	struct pair_state_instruction *pairinst = s->Instructions + ip;
	ASSERT(pairinst->NumDependencies > 0);
	if (!--pairinst->NumDependencies)
		instruction_ready(s, ip);
}

/**
 * Update the dependency tracking state based on what the instruction
 * at the given IP does.
 */
static void commit_instruction(struct pair_state *s, int ip)
{
	struct prog_instruction *inst = s->Program->Instructions + ip;
	struct pair_state_instruction *pairinst = s->Instructions + ip;

	if (s->Verbose)
		_mesa_printf("commit_instruction(%i)\n", ip);

	if (inst->DstReg.File == PROGRAM_TEMPORARY) {
		struct pair_register_translation *t = &s->Temps[inst->DstReg.Index];
		deref_hw_reg(s, t->HwIndex);

		int i;
		for(i = 0; i < 4; ++i) {
			if (!GET_BIT(inst->DstReg.WriteMask, i))
				continue;

			t->Value[i] = pairinst->Values[i];
			if (t->Value[i]->NumReaders) {
				struct reg_value_reader *r;
				for(r = pairinst->Values[i]->Readers; r; r = r->Next)
					decrement_dependencies(s, r->IP);
			} else if (t->Value[i]->Next) {
				/* This happens when the only reader writes
				 * the register at the same time */
				decrement_dependencies(s, t->Value[i]->Next->IP);
			}
		}
	}

	int nsrc = _mesa_num_inst_src_regs(inst->Opcode);
	int i;
	for(i = 0; i < nsrc; i++) {
		struct pair_register_translation *t = get_register(s, inst->SrcReg[i].File, inst->SrcReg[i].Index);
		if (!t)
			continue;

		deref_hw_reg(s, get_hw_reg(s, inst->SrcReg[i].File, inst->SrcReg[i].Index));

		if (inst->SrcReg[i].File != PROGRAM_TEMPORARY)
			continue;

		int j;
		for(j = 0; j < 4; ++j) {
			GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j);
			if (swz >= 4)
				continue;
			if (!t->Value[swz])
				continue;

			/* Do not free a dependency if this instruction
			 * also rewrites the value. See scan_instructions. */
			if (inst->DstReg.File == PROGRAM_TEMPORARY &&
			    inst->DstReg.Index == inst->SrcReg[i].Index &&
			    GET_BIT(inst->DstReg.WriteMask, swz))
				continue;

			if (!--t->Value[swz]->NumReaders) {
				if (t->Value[swz]->Next)
					decrement_dependencies(s, t->Value[swz]->Next->IP);
			}
		}
	}
}


/**
 * Emit all ready texture instructions in a single block.
 *
 * Emit as a single block to (hopefully) sample many textures in parallel,
 * and to avoid hardware indirections on R300.
 *
 * In R500, we don't really know when the result of a texture instruction
 * arrives. So allocate all destinations first, to make sure they do not
 * arrive early and overwrite a texture coordinate we're going to use later
 * in the block.
 */
static void emit_all_tex(struct pair_state *s)
{
	struct pair_state_instruction *readytex;
	struct pair_state_instruction *pairinst;

	ASSERT(s->ReadyTEX);

	// Don't let the ready list change under us!
	readytex = s->ReadyTEX;
	s->ReadyTEX = 0;

	// Allocate destination hardware registers in one block to avoid conflicts.
	for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) {
		int ip = pairinst - s->Instructions;
		struct prog_instruction *inst = s->Program->Instructions + ip;
		if (inst->Opcode != OPCODE_KIL)
			get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
	}

	if (s->Debug)
		_mesa_printf(" BEGIN_TEX\n");

	if (s->Handler->BeginTexBlock)
		s->Error = s->Error || !s->Handler->BeginTexBlock(s->UserData);

	for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) {
		int ip = pairinst - s->Instructions;
		struct prog_instruction *inst = s->Program->Instructions + ip;
		commit_instruction(s, ip);

		if (inst->Opcode != OPCODE_KIL)
			inst->DstReg.Index = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
		inst->SrcReg[0].Index = get_hw_reg(s, inst->SrcReg[0].File, inst->SrcReg[0].Index);

		if (s->Debug) {
			_mesa_printf("   ");
			_mesa_print_instruction(inst);
		}
		s->Error = s->Error || !s->Handler->EmitTex(s->UserData, inst);
	}

	if (s->Debug)
		_mesa_printf(" END_TEX\n");
}


static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instruction *pair,
	struct prog_src_register src, GLboolean rgb, GLboolean alpha)
{
	int candidate = -1;
	int candidate_quality = -1;
	int i;

	if (!rgb && !alpha)
		return 0;

	GLuint constant;
	GLuint index;

	if (src.File == PROGRAM_TEMPORARY || src.File == PROGRAM_INPUT) {
		constant = 0;
		index = get_hw_reg(s, src.File, src.Index);
	} else {
		constant = 1;
		s->Error |= !s->Handler->EmitConst(s->UserData, src.File, src.Index, &index);
	}

	for(i = 0; i < 3; ++i) {
		int q = 0;
		if (rgb) {
			if (pair->RGB.Src[i].Used) {
				if (pair->RGB.Src[i].Constant != constant ||
				    pair->RGB.Src[i].Index != index)
					continue;
				q++;
			}
		}
		if (alpha) {
			if (pair->Alpha.Src[i].Used) {
				if (pair->Alpha.Src[i].Constant != constant ||
				    pair->Alpha.Src[i].Index != index)
					continue;
				q++;
			}
		}
		if (q > candidate_quality) {
			candidate_quality = q;
			candidate = i;
		}
	}

	if (candidate >= 0) {
		if (rgb) {
			pair->RGB.Src[candidate].Used = 1;
			pair->RGB.Src[candidate].Constant = constant;
			pair->RGB.Src[candidate].Index = index;
		}
		if (alpha) {
			pair->Alpha.Src[candidate].Used = 1;
			pair->Alpha.Src[candidate].Constant = constant;
			pair->Alpha.Src[candidate].Index = index;
		}
	}

	return candidate;
}

/**
 * Fill the given ALU instruction's opcodes and source operands into the given pair,
 * if possible.
 */
static GLboolean fill_instruction_into_pair(struct pair_state *s, struct radeon_pair_instruction *pair, int ip)
{
	struct pair_state_instruction *pairinst = s->Instructions + ip;
	struct prog_instruction *inst = s->Program->Instructions + ip;

	ASSERT(!pairinst->NeedRGB || pair->RGB.Opcode == OPCODE_NOP);
	ASSERT(!pairinst->NeedAlpha || pair->Alpha.Opcode == OPCODE_NOP);

	if (pairinst->NeedRGB) {
		if (pairinst->IsTranscendent)
			pair->RGB.Opcode = OPCODE_REPL_ALPHA;
		else
			pair->RGB.Opcode = inst->Opcode;
		if (inst->SaturateMode == SATURATE_ZERO_ONE)
			pair->RGB.Saturate = 1;
	}
	if (pairinst->NeedAlpha) {
		pair->Alpha.Opcode = inst->Opcode;
		if (inst->SaturateMode == SATURATE_ZERO_ONE)
			pair->Alpha.Saturate = 1;
	}

	int nargs = _mesa_num_inst_src_regs(inst->Opcode);
	int i;

	/* Special case for DDX/DDY (MDH/MDV). */
	if (inst->Opcode == OPCODE_DDX || inst->Opcode == OPCODE_DDY) {
		if (pair->RGB.Src[0].Used || pair->Alpha.Src[0].Used)
			return GL_FALSE;
		else
			nargs++;
	}

	for(i = 0; i < nargs; ++i) {
		int source;
		if (pairinst->NeedRGB && !pairinst->IsTranscendent) {
			GLboolean srcrgb = GL_FALSE;
			GLboolean srcalpha = GL_FALSE;
			int j;
			for(j = 0; j < 3; ++j) {
				GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j);
				if (swz < 3)
					srcrgb = GL_TRUE;
				else if (swz < 4)
					srcalpha = GL_TRUE;
			}
			source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha);
			if (source < 0)
				return GL_FALSE;
			pair->RGB.Arg[i].Source = source;
			pair->RGB.Arg[i].Swizzle = inst->SrcReg[i].Swizzle & 0x1ff;
			pair->RGB.Arg[i].Abs = inst->SrcReg[i].Abs;
			pair->RGB.Arg[i].Negate = !!(inst->SrcReg[i].Negate & (NEGATE_X | NEGATE_Y | NEGATE_Z));
		}
		if (pairinst->NeedAlpha) {
			GLboolean srcrgb = GL_FALSE;
			GLboolean srcalpha = GL_FALSE;
			GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, pairinst->IsTranscendent ? 0 : 3);
			if (swz < 3)
				srcrgb = GL_TRUE;
			else if (swz < 4)
				srcalpha = GL_TRUE;
			source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha);
			if (source < 0)
				return GL_FALSE;
			pair->Alpha.Arg[i].Source = source;
			pair->Alpha.Arg[i].Swizzle = swz;
			pair->Alpha.Arg[i].Abs = inst->SrcReg[i].Abs;
			pair->Alpha.Arg[i].Negate = !!(inst->SrcReg[i].Negate & NEGATE_W);
		}
	}

	return GL_TRUE;
}


/**
 * Fill in the destination register information.
 *
 * This is split from filling in source registers because we want
 * to avoid allocating hardware temporaries for destinations until
 * we are absolutely certain that we're going to emit a certain
 * instruction pairing.
 */
static void fill_dest_into_pair(struct pair_state *s, struct radeon_pair_instruction *pair, int ip)
{
	struct pair_state_instruction *pairinst = s->Instructions + ip;
	struct prog_instruction *inst = s->Program->Instructions + ip;

	if (inst->DstReg.File == PROGRAM_OUTPUT) {
		if (inst->DstReg.Index == FRAG_RESULT_COLOR) {
			pair->RGB.OutputWriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ;
			pair->Alpha.OutputWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
		} else if (inst->DstReg.Index == FRAG_RESULT_DEPTH) {
			pair->Alpha.DepthWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
		}
	} else {
		GLuint hwindex = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
		if (pairinst->NeedRGB) {
			pair->RGB.DestIndex = hwindex;
			pair->RGB.WriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ;
		}
		if (pairinst->NeedAlpha) {
			pair->Alpha.DestIndex = hwindex;
			pair->Alpha.WriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
		}
	}
}


/**
 * Find a good ALU instruction or pair of ALU instruction and emit it.
 *
 * Prefer emitting full ALU instructions, so that when we reach a point
 * where no full ALU instruction can be emitted, we have more candidates
 * for RGB/Alpha pairing.
 */
static void emit_alu(struct pair_state *s)
{
	struct radeon_pair_instruction pair;

	if (s->ReadyFullALU || !(s->ReadyRGB && s->ReadyAlpha)) {
		int ip;
		if (s->ReadyFullALU) {
			ip = s->ReadyFullALU - s->Instructions;
			s->ReadyFullALU = s->ReadyFullALU->NextReady;
		} else if (s->ReadyRGB) {
			ip = s->ReadyRGB - s->Instructions;
			s->ReadyRGB = s->ReadyRGB->NextReady;
		} else {
			ip = s->ReadyAlpha - s->Instructions;
			s->ReadyAlpha = s->ReadyAlpha->NextReady;
		}

		_mesa_bzero(&pair, sizeof(pair));
		fill_instruction_into_pair(s, &pair, ip);
		fill_dest_into_pair(s, &pair, ip);
		commit_instruction(s, ip);
	} else {
		struct pair_state_instruction **prgb;
		struct pair_state_instruction **palpha;

		/* Some pairings might fail because they require too
		 * many source slots; try all possible pairings if necessary */
		for(prgb = &s->ReadyRGB; *prgb; prgb = &(*prgb)->NextReady) {
			for(palpha = &s->ReadyAlpha; *palpha; palpha = &(*palpha)->NextReady) {
				int rgbip = *prgb - s->Instructions;
				int alphaip = *palpha - s->Instructions;
				_mesa_bzero(&pair, sizeof(pair));
				fill_instruction_into_pair(s, &pair, rgbip);
				if (!fill_instruction_into_pair(s, &pair, alphaip))
					continue;
				*prgb = (*prgb)->NextReady;
				*palpha = (*palpha)->NextReady;
				fill_dest_into_pair(s, &pair, rgbip);
				fill_dest_into_pair(s, &pair, alphaip);
				commit_instruction(s, rgbip);
				commit_instruction(s, alphaip);
				goto success;
			}
		}

		/* No success in pairing; just take the first RGB instruction */
		int ip = s->ReadyRGB - s->Instructions;
		s->ReadyRGB = s->ReadyRGB->NextReady;
		_mesa_bzero(&pair, sizeof(pair));
		fill_instruction_into_pair(s, &pair, ip);
		fill_dest_into_pair(s, &pair, ip);
		commit_instruction(s, ip);
	success: ;
	}

	if (s->Debug)
		radeonPrintPairInstruction(&pair);

	s->Error = s->Error || !s->Handler->EmitPaired(s->UserData, &pair);
}


GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
	const struct radeon_pair_handler* handler, void *userdata)
{
	struct pair_state s;

	_mesa_bzero(&s, sizeof(s));
	s.Ctx = ctx;
	s.Program = program;
	s.Handler = handler;
	s.UserData = userdata;
	s.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE;
	s.Verbose = GL_FALSE && s.Debug;

	s.Instructions = (struct pair_state_instruction*)_mesa_calloc(
		sizeof(struct pair_state_instruction)*s.Program->NumInstructions);
	s.ValuePool = (struct reg_value*)_mesa_calloc(sizeof(struct reg_value)*s.Program->NumInstructions*4);
	s.ReaderPool = (struct reg_value_reader*)_mesa_calloc(
		sizeof(struct reg_value_reader)*s.Program->NumInstructions*12);

	if (s.Debug)
		_mesa_printf("Emit paired program\n");

	scan_instructions(&s);
	allocate_input_registers(&s);

	while(!s.Error &&
	      (s.ReadyTEX || s.ReadyRGB || s.ReadyAlpha || s.ReadyFullALU)) {
		if (s.ReadyTEX)
			emit_all_tex(&s);

		while(s.ReadyFullALU || s.ReadyRGB || s.ReadyAlpha)
			emit_alu(&s);
	}

	if (s.Debug)
		_mesa_printf(" END\n");

	_mesa_free(s.Instructions);
	_mesa_free(s.ValuePool);
	_mesa_free(s.ReaderPool);

	return !s.Error;
}


static void print_pair_src(int i, struct radeon_pair_instruction_source* src)
{
	_mesa_printf("  Src%i = %s[%i]", i, src->Constant ? "CNST" : "TEMP", src->Index);
}

static const char* opcode_string(GLuint opcode)
{
	if (opcode == OPCODE_REPL_ALPHA)
		return "SOP";
	else
		return _mesa_opcode_string(opcode);
}

static int num_pairinst_args(GLuint opcode)
{
	if (opcode == OPCODE_REPL_ALPHA)
		return 0;
	else
		return _mesa_num_inst_src_regs(opcode);
}

static char swizzle_char(GLuint swz)
{
	switch(swz) {
	case SWIZZLE_X: return 'x';
	case SWIZZLE_Y: return 'y';
	case SWIZZLE_Z: return 'z';
	case SWIZZLE_W: return 'w';
	case SWIZZLE_ZERO: return '0';
	case SWIZZLE_ONE: return '1';
	case SWIZZLE_NIL: return '_';
	default: return '?';
	}
}

void radeonPrintPairInstruction(struct radeon_pair_instruction *inst)
{
	int nargs;
	int i;

	_mesa_printf("       RGB:  ");
	for(i = 0; i < 3; ++i) {
		if (inst->RGB.Src[i].Used)
			print_pair_src(i, inst->RGB.Src + i);
	}
	_mesa_printf("\n");
	_mesa_printf("       Alpha:");
	for(i = 0; i < 3; ++i) {
		if (inst->Alpha.Src[i].Used)
			print_pair_src(i, inst->Alpha.Src + i);
	}
	_mesa_printf("\n");

	_mesa_printf("  %s%s", opcode_string(inst->RGB.Opcode), inst->RGB.Saturate ? "_SAT" : "");
	if (inst->RGB.WriteMask)
		_mesa_printf(" TEMP[%i].%s%s%s", inst->RGB.DestIndex,
			(inst->RGB.WriteMask & 1) ? "x" : "",
			(inst->RGB.WriteMask & 2) ? "y" : "",
			(inst->RGB.WriteMask & 4) ? "z" : "");
	if (inst->RGB.OutputWriteMask)
		_mesa_printf(" COLOR.%s%s%s",
			(inst->RGB.OutputWriteMask & 1) ? "x" : "",
			(inst->RGB.OutputWriteMask & 2) ? "y" : "",
			(inst->RGB.OutputWriteMask & 4) ? "z" : "");
	nargs = num_pairinst_args(inst->RGB.Opcode);
	for(i = 0; i < nargs; ++i) {
		const char* abs = inst->RGB.Arg[i].Abs ? "|" : "";
		const char* neg = inst->RGB.Arg[i].Negate ? "-" : "";
		_mesa_printf(", %s%sSrc%i.%c%c%c%s", neg, abs, inst->RGB.Arg[i].Source,
			swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 0)),
			swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 1)),
			swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 2)),
			abs);
	}
	_mesa_printf("\n");

	_mesa_printf("  %s%s", opcode_string(inst->Alpha.Opcode), inst->Alpha.Saturate ? "_SAT" : "");
	if (inst->Alpha.WriteMask)
		_mesa_printf(" TEMP[%i].w", inst->Alpha.DestIndex);
	if (inst->Alpha.OutputWriteMask)
		_mesa_printf(" COLOR.w");
	if (inst->Alpha.DepthWriteMask)
		_mesa_printf(" DEPTH.w");
	nargs = num_pairinst_args(inst->Alpha.Opcode);
	for(i = 0; i < nargs; ++i) {
		const char* abs = inst->Alpha.Arg[i].Abs ? "|" : "";
		const char* neg = inst->Alpha.Arg[i].Negate ? "-" : "";
		_mesa_printf(", %s%sSrc%i.%c%s", neg, abs, inst->Alpha.Arg[i].Source,
			swizzle_char(inst->Alpha.Arg[i].Swizzle), abs);
	}
	_mesa_printf("\n");
}