summaryrefslogtreecommitdiff
path: root/src/mesa/tnl/t_vertex_sse.c
blob: 109fdfb3b0d16ba0a855096d1c69b3e858a5396e (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
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
/*
 * Copyright 2003 Tungsten Graphics, inc.
 * 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
 * on the rights to use, copy, modify, merge, publish, distribute, sub
 * license, 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 NON-INFRINGEMENT.  IN NO EVENT SHALL
 * TUNGSTEN GRAPHICS AND/OR THEIR 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.
 *
 * Authors:
 *    Keith Whitwell <keithw@tungstengraphics.com>
 */

#include "glheader.h"
#include "context.h"
#include "colormac.h"
#include "t_context.h"
#include "t_vertex.h"
#include "simple_list.h"

#define X    0
#define Y    1
#define Z    2
#define W    3

#define DISASSEM 1

struct x86_reg {
   GLuint file:3;
   GLuint idx:3;
   GLuint mod:2;		/* mod_REG if this is just a register */
   GLint  disp:24;		/* only +/- 23bits of offset - should be enough... */
};

struct x86_program {
   GLcontext *ctx;
   
   GLubyte *store;
   GLubyte *csr;

   GLuint stack_offset;

   GLboolean inputs_safe;
   GLboolean outputs_safe;
   GLboolean have_sse2;
   GLboolean need_emms;
   
   struct x86_reg identity;
   struct x86_reg vp0;
   struct x86_reg vp1;
};


#define X86_TWOB 0x0f

/* There are more but these are all we'll use:
 */
enum x86_reg_file {
   file_REG32,
   file_MMX,
   file_XMM
};

/* Values for mod field of modr/m byte
 */
enum x86_reg_mod {
   mod_INDIRECT,
   mod_DISP8,
   mod_DISP32,
   mod_REG
};

enum x86_reg_name {
   reg_AX,
   reg_CX,
   reg_DX,
   reg_BX,
   reg_SP,
   reg_BP,
   reg_SI,
   reg_DI
};


enum x86_cc {
   cc_O,			/* overflow */
   cc_NO,			/* not overflow */
   cc_NAE,			/* not above or equal / carry */
   cc_AE,			/* above or equal / not carry */
   cc_E,			/* equal / zero */
   cc_NE			/* not equal / not zero */
};

#define cc_Z  cc_E
#define cc_NZ cc_NE


/* Create and manipulate registers and regmem values:
 */
static struct x86_reg make_reg( GLuint file,
				GLuint idx )
{
   struct x86_reg reg;

   reg.file = file;
   reg.idx = idx;
   reg.mod = mod_REG;
   reg.disp = 0;

   return reg;
}

static struct x86_reg make_disp( struct x86_reg reg,
				 GLint disp )
{
   assert(reg.file == file_REG32);

   if (reg.mod == mod_REG)
      reg.disp = disp;
   else
      reg.disp += disp;

   if (reg.disp == 0)
      reg.mod = mod_INDIRECT;
   else if (reg.disp <= 127 && reg.disp >= -128)
      reg.mod = mod_DISP8;
   else
      reg.mod = mod_DISP32;

   return reg;
}

static struct x86_reg deref( struct x86_reg reg )
{
   return make_disp(reg, 0);
}

static struct x86_reg get_base_reg( struct x86_reg reg )
{
   return make_reg( reg.file, reg.idx );
}


/* Retreive a reference to one of the function arguments, taking into
 * account any push/pop activity:
 */
static struct x86_reg make_fn_arg( struct x86_program *p,
				   GLuint arg )
{
   return make_disp(make_reg(file_REG32, reg_SP), 
		    p->stack_offset + arg * 4);	/* ??? */
}

static struct x86_reg get_identity( struct x86_program *p )
{
   return p->identity;
}


/* Emit bytes to the instruction stream:
 */
static void emit_1b( struct x86_program *p, GLbyte b0 )
{
   *(GLbyte *)(p->csr++) = b0;
}

static void emit_1i( struct x86_program *p, GLint i0 )
{
   *(GLint *)(p->csr) = i0;
   p->csr += 4;
}

static void disassem( struct x86_program *p, const char *fn )
{
#if DISASSEM
   static const char *last_fn;
   if (fn && fn != last_fn) {
      _mesa_printf("0x%x: %s\n", p->csr, fn);
      last_fn = fn;
   }
#endif
}

static void emit_1ub_fn( struct x86_program *p, GLubyte b0, const char *fn )
{
   disassem(p, fn);
   *(p->csr++) = b0;
}

static void emit_2ub_fn( struct x86_program *p, GLubyte b0, GLubyte b1, const char *fn )
{
   disassem(p, fn);
   *(p->csr++) = b0;
   *(p->csr++) = b1;
}

static void emit_3ub_fn( struct x86_program *p, GLubyte b0, GLubyte b1, GLubyte b2, const char *fn )
{
   disassem(p, fn);
   *(p->csr++) = b0;
   *(p->csr++) = b1;
   *(p->csr++) = b2;
}

#define emit_1ub(p, b0)         emit_1ub_fn(p, b0, __FUNCTION__)
#define emit_2ub(p, b0, b1)     emit_2ub_fn(p, b0, b1, __FUNCTION__)
#define emit_3ub(p, b0, b1, b2) emit_3ub_fn(p, b0, b1, b2, __FUNCTION__)


/* Labels, jumps and fixup:
 */
static GLubyte *get_label( struct x86_program *p )
{
   return p->csr;
}

static void x86_jcc( struct x86_program *p,
		      GLuint cc,
		      GLubyte *label )
{
   GLint offset = label - (get_label(p) + 2);
   
   if (offset <= 127 && offset >= -128) {
      emit_1ub(p, 0x70 + cc);
      emit_1b(p, (GLbyte) offset);
   }
   else {
      offset = label - (get_label(p) + 6);
      emit_2ub(p, 0x0f, 0x80 + cc);
      emit_1i(p, offset);
   }
}

/* Always use a 32bit offset for forward jumps:
 */
static GLubyte *x86_jcc_forward( struct x86_program *p,
				 GLuint cc )
{
   emit_2ub(p, 0x0f, 0x80 + cc);
   emit_1i(p, 0);
   return get_label(p);
}

/* Fixup offset from forward jump:
 */
static void do_fixup( struct x86_program *p,
		      GLubyte *fixup )
{
   *(int *)(fixup - 4) = get_label(p) - fixup;
}

static void x86_push( struct x86_program *p,
		       struct x86_reg reg )
{
   assert(reg.mod == mod_REG);
   emit_1ub(p, 0x50 + reg.idx);
   p->stack_offset += 4;
}

static void x86_pop( struct x86_program *p,
		       struct x86_reg reg )
{
   assert(reg.mod == mod_REG);
   emit_1ub(p, 0x58 + reg.idx);
   p->stack_offset -= 4;
}

static void x86_inc( struct x86_program *p,
		       struct x86_reg reg )
{
   assert(reg.mod == mod_REG);
   emit_1ub(p, 0x40 + reg.idx);
}

static void x86_dec( struct x86_program *p,
		       struct x86_reg reg )
{
   assert(reg.mod == mod_REG);
   emit_1ub(p, 0x48 + reg.idx);
}

static void x86_ret( struct x86_program *p )
{
   emit_1ub(p, 0xc3);
}

static void mmx_emms( struct x86_program *p )
{
   assert(p->need_emms);
   emit_2ub(p, 0x0f, 0x0e);
   p->need_emms = 0;
}




/* Build a modRM byte + possible displacement.  No treatment of SIB
 * indexing.  BZZT - no way to encode an absolute address.
 */
static void emit_modrm( struct x86_program *p, 
			struct x86_reg reg, 
			struct x86_reg regmem )
{
   GLubyte val = 0;
   
   assert(reg.mod == mod_REG);
   
   val |= regmem.mod << 6;     	/* mod field */
   val |= reg.idx << 3;		/* reg field */
   val |= regmem.idx;		/* r/m field */
   
   emit_1ub_fn(p, val, 0);

   /* Oh-oh we've stumbled into the SIB thing.
    */
   if (regmem.idx == reg_SP) {
      emit_1ub_fn(p, 0x24, 0);		/* simplistic! */
   }

   switch (regmem.mod) {
   case mod_REG:
   case mod_INDIRECT:
      break;
   case mod_DISP8:
      emit_1b(p, regmem.disp);
      break;
   case mod_DISP32:
      emit_1i(p, regmem.disp);
      break;
   }
}

/* Many x86 instructions have two opcodes to cope with the situations
 * where the destination is a register or memory reference
 * respectively.  This function selects the correct opcode based on
 * the arguments presented.
 */
static void emit_op_modrm( struct x86_program *p,
			   GLubyte op_dst_is_reg, 
			   GLubyte op_dst_is_mem,
			   struct x86_reg dst,
			   struct x86_reg src )
{
   switch (dst.mod) {
   case mod_REG:
      emit_1ub_fn(p, op_dst_is_reg, 0);
      emit_modrm(p, dst, src);
      break;
   case mod_INDIRECT:
   case mod_DISP32:
   case mod_DISP8:
      assert(src.mod == mod_REG);
      emit_1ub_fn(p, op_dst_is_mem, 0);
      emit_modrm(p, src, dst);
      break;
   }
}

static void x86_mov( struct x86_program *p,
		      struct x86_reg dst,
		      struct x86_reg src )
{
   emit_op_modrm( p, 0x8b, 0x89, dst, src );
}

static void x86_xor( struct x86_program *p,
		      struct x86_reg dst,
		      struct x86_reg src )
{
   emit_op_modrm( p, 0x33, 0x31, dst, src );
}

static void x86_cmp( struct x86_program *p,
		      struct x86_reg dst,
		      struct x86_reg src )
{
   emit_op_modrm( p, 0x3b, 0x39, dst, src );
}

static void sse2_movd( struct x86_program *p,
		       struct x86_reg dst,
		       struct x86_reg src )
{
   assert(p->have_sse2);
   emit_2ub(p, 0x66, X86_TWOB);
   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
}

static void mmx_movd( struct x86_program *p,
		       struct x86_reg dst,
		       struct x86_reg src )
{
   p->need_emms = 1;
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
}

static void mmx_movq( struct x86_program *p,
		       struct x86_reg dst,
		       struct x86_reg src )
{
   p->need_emms = 1;
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x6f, 0x7f, dst, src );
}


static void sse_movss( struct x86_program *p,
		       struct x86_reg dst,
		       struct x86_reg src )
{
   emit_2ub(p, 0xF3, X86_TWOB);
   emit_op_modrm( p, 0x10, 0x11, dst, src );
}

static void sse_movaps( struct x86_program *p,
			 struct x86_reg dst,
			 struct x86_reg src )
{
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x28, 0x29, dst, src );
}

static void sse_movups( struct x86_program *p,
			 struct x86_reg dst,
			 struct x86_reg src )
{
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x10, 0x11, dst, src );
}

static void sse_movhps( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   assert(dst.mod != mod_REG || src.mod != mod_REG);
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x16, 0x17, dst, src ); /* cf movlhps */
}

static void sse_movlps( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   assert(dst.mod != mod_REG || src.mod != mod_REG);
   emit_1ub(p, X86_TWOB);
   emit_op_modrm( p, 0x12, 0x13, dst, src ); /* cf movhlps */
}

/* SSE operations often only have one format, with dest constrained to
 * be a register:
 */
static void sse_mulps( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   emit_2ub(p, X86_TWOB, 0x59);
   emit_modrm( p, dst, src );
}

static void sse_addps( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   emit_2ub(p, X86_TWOB, 0x58);
   emit_modrm( p, dst, src );
}

static void sse_movhlps( struct x86_program *p,
			 struct x86_reg dst,
			 struct x86_reg src )
{
   assert(dst.mod == mod_REG && src.mod == mod_REG);
   emit_2ub(p, X86_TWOB, 0x12);
   emit_modrm( p, dst, src );
}

static void sse_movlhps( struct x86_program *p,
			 struct x86_reg dst,
			 struct x86_reg src )
{
   assert(dst.mod == mod_REG && src.mod == mod_REG);
   emit_2ub(p, X86_TWOB, 0x16);
   emit_modrm( p, dst, src );
}

static void sse2_cvtps2dq( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   assert(p->have_sse2);
   emit_3ub(p, 0x66, X86_TWOB, 0x5B);
   emit_modrm( p, dst, src );
}

static void sse2_packssdw( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   assert(p->have_sse2);
   emit_3ub(p, 0x66, X86_TWOB, 0x6B);
   emit_modrm( p, dst, src );
}

static void sse2_packsswb( struct x86_program *p,
			struct x86_reg dst,
			struct x86_reg src )
{
   assert(p->have_sse2);
   emit_3ub(p, 0x66, X86_TWOB, 0x63);
   emit_modrm( p, dst, src );
}

static void sse2_packuswb( struct x86_program *p,
			   struct x86_reg dst,
			   struct x86_reg src )
{
   assert(p->have_sse2);
   emit_3ub(p, 0x66, X86_TWOB, 0x67);
   emit_modrm( p, dst, src );
}

static void sse_cvtps2pi( struct x86_program *p,
			  struct x86_reg dst,
			  struct x86_reg src )
{
   assert(dst.file == file_MMX && 
	  (src.file == file_XMM || src.mod != mod_REG));

   p->need_emms = 1;

   emit_2ub(p, X86_TWOB, 0x2d);
   emit_modrm( p, dst, src );
}

static void mmx_packssdw( struct x86_program *p,
			  struct x86_reg dst,
			  struct x86_reg src )
{
   assert(dst.file == file_MMX && 
	  (src.file == file_MMX || src.mod != mod_REG));

   p->need_emms = 1;

   emit_2ub(p, X86_TWOB, 0x6b);
   emit_modrm( p, dst, src );
}

static void mmx_packuswb( struct x86_program *p,
			  struct x86_reg dst,
			  struct x86_reg src )
{
   assert(dst.file == file_MMX && 
	  (src.file == file_MMX || src.mod != mod_REG));

   p->need_emms = 1;

   emit_2ub(p, X86_TWOB, 0x67);
   emit_modrm( p, dst, src );
}


/* Load effective address:
 */
static void x86_lea( struct x86_program *p,
		     struct x86_reg dst,
		     struct x86_reg src )
{
   emit_1ub(p, 0x8d);
   emit_modrm( p, dst, src );
}

static void x86_test( struct x86_program *p,
		      struct x86_reg dst,
		      struct x86_reg src )
{
   emit_1ub(p, 0x85);
   emit_modrm( p, dst, src );
}




/**
 * Perform a reduced swizzle:
 */
static void sse2_pshufd( struct x86_program *p,
			 struct x86_reg dest,
			 struct x86_reg arg0,
			 GLubyte x,
			 GLubyte y,
			 GLubyte z,
			 GLubyte w) 
{
   assert(p->have_sse2);
   emit_3ub(p, 0x66, X86_TWOB, 0x70);
   emit_modrm(p, dest, arg0);
   emit_1ub(p, (x|(y<<2)|(z<<4)|w<<6)); 
}


/* Shufps can also be used to implement a reduced swizzle when dest ==
 * arg0.
 */
static void sse_shufps( struct x86_program *p,
			 struct x86_reg dest,
			 struct x86_reg arg0,
			 GLubyte x,
			 GLubyte y,
			 GLubyte z,
			 GLubyte w) 
{
   emit_2ub(p, X86_TWOB, 0xC6);
   emit_modrm(p, dest, arg0);
   emit_1ub(p, (x|(y<<2)|(z<<4)|w<<6)); 
}


static void emit_load4f_4( struct x86_program *p, 			   
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   sse_movups(p, dest, arg0);
}

static void emit_load4f_3( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   /* Have to jump through some hoops:
    *
    * 0 0 0 1 -- skip if reg[3] preserved over loop iterations
    * c 0 0 1
    * 0 0 c 1
    * a b c 1 
    */
   sse_movaps(p, dest, get_identity(p));
   sse_movss(p, dest, make_disp(arg0, 8));
   sse_shufps(p, dest, dest, Y,Z,X,W );
   sse_movlps(p, dest, arg0);
}

static void emit_load4f_2( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   /* Pull in 2 dwords, then copy the top 2 dwords with 0,1 from id.
    */
   sse_movlps(p, dest, arg0);
   sse_movhps(p, dest, get_identity(p));
}

static void emit_load4f_1( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   /* Initialized with [0,0,0,1] from id, then pull in the single low
    * word.
    */
   sse_movaps(p, dest, get_identity(p));
   sse_movss(p, dest, arg0);
}



static void emit_load3f_3( struct x86_program *p, 			   
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   /* Over-reads by 1 dword - potential SEGV if input is a vertex
    * array.
    */
   if (p->inputs_safe) {
      sse_movups(p, dest, arg0);
   } 
   else {
      /* c . . .
       * c c c c
       * a b c c 
       */
      sse_movss(p, dest, make_disp(arg0, 8));
      sse_shufps(p, dest, dest, X,X,X,X);
      sse_movlps(p, dest, arg0);
   }
}

static void emit_load3f_2( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   emit_load4f_2(p, dest, arg0);
}

static void emit_load3f_1( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   emit_load4f_1(p, dest, arg0);
}

static void emit_load2f_2( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   sse_movlps(p, dest, arg0);
}

static void emit_load2f_1( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   emit_load4f_1(p, dest, arg0);
}

static void emit_load1f_1( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   sse_movss(p, dest, arg0);
}

static void (*load[4][4])( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 ) = {
   { emit_load1f_1, 
     emit_load1f_1, 
     emit_load1f_1, 
     emit_load1f_1 },

   { emit_load2f_1, 
     emit_load2f_2, 
     emit_load2f_2, 
     emit_load2f_2 },

   { emit_load3f_1, 
     emit_load3f_2, 
     emit_load3f_3, 
     emit_load3f_3 },

   { emit_load4f_1, 
     emit_load4f_2, 
     emit_load4f_3, 
     emit_load4f_4 } 
};

static void emit_load( struct x86_program *p,
		       struct x86_reg dest,
		       GLuint sz,
		       struct x86_reg src,
		       GLuint src_sz)
{
   if (DISASSEM)
      _mesa_printf("load %d/%d\n", sz, src_sz);
   load[sz-1][src_sz-1](p, dest, src);
}


static void emit_store4f( struct x86_program *p, 			   
			  struct x86_reg dest,
			  struct x86_reg arg0 )
{
   sse_movups(p, dest, arg0);
}

static void emit_store3f( struct x86_program *p, 
			  struct x86_reg dest,
			  struct x86_reg arg0 )
{
   if (p->outputs_safe) {
      /* Emit the extra dword anyway.  This may hurt writecombining,
       * may cause other problems.
       */
      sse_movups(p, dest, arg0);
   }
   else {
      /* Alternate strategy - emit two, shuffle, emit one.
       */
      sse_movlps(p, dest, arg0);
      sse_shufps(p, arg0, arg0, Z, Z, Z, Z ); /* NOTE! destructive */
      sse_movss(p, make_disp(dest,8), arg0);
   }
}

static void emit_store2f( struct x86_program *p, 
			   struct x86_reg dest,
			   struct x86_reg arg0 )
{
   sse_movlps(p, dest, arg0);
}

static void emit_store1f( struct x86_program *p, 
			  struct x86_reg dest,
			  struct x86_reg arg0 )
{
   sse_movss(p, dest, arg0);
}


static void (*store[4])( struct x86_program *p, 
			 struct x86_reg dest,
			 struct x86_reg arg0 ) = 
{
   emit_store1f, 
   emit_store2f, 
   emit_store3f, 
   emit_store4f 
};

static void emit_store( struct x86_program *p,
			struct x86_reg dest,
			GLuint sz,
			struct x86_reg temp )

{
   if (DISASSEM)
      _mesa_printf("store %d\n", sz);
   store[sz-1](p, dest, temp);
}

static void emit_pack_store_4ub( struct x86_program *p,
				 struct x86_reg dest,
				 struct x86_reg temp )
{
   if (p->have_sse2) {
      sse2_cvtps2dq(p, temp, temp);
      sse2_packssdw(p, temp, temp);
      sse2_packuswb(p, temp, temp);
      sse_movss(p, dest, temp);
   }
   else {
      struct x86_reg mmx0 = make_reg(file_MMX, 0);
      struct x86_reg mmx1 = make_reg(file_MMX, 1);
      sse_cvtps2pi(p, mmx0, temp);
      sse_movhlps(p, temp, temp);
      sse_cvtps2pi(p, mmx1, temp);
      mmx_packssdw(p, mmx0, mmx1);
      mmx_packuswb(p, mmx0, mmx0);
      mmx_movd(p, dest, mmx0);
   }
}

static GLint get_offset( const void *a, const void *b )
{
   return (const char *)b - (const char *)a;
}



/* Lots of hardcoding
 *
 * EAX -- pointer to current output vertex
 * ECX -- pointer to current attribute 
 * 
 */
static GLboolean build_vertex_emit( struct x86_program *p )
{
   GLcontext *ctx = p->ctx;
   TNLcontext *tnl = TNL_CONTEXT(ctx);
   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
   struct tnl_clipspace_attr *a = vtx->attr;
   GLuint j;

   struct x86_reg vertexEAX = make_reg(file_REG32, reg_AX);
   struct x86_reg srcEDI = make_reg(file_REG32, reg_CX);
   struct x86_reg countEBP = make_reg(file_REG32, reg_BP);
   struct x86_reg vtxESI = make_reg(file_REG32, reg_SI);
   struct x86_reg temp = make_reg(file_XMM, 0);
   struct x86_reg vp0 = make_reg(file_XMM, 1);
   struct x86_reg vp1 = make_reg(file_XMM, 2);
   struct x86_reg chan0 = make_reg(file_XMM, 3);
   GLubyte *fixup, *label;

   p->csr = p->store;
   
   /* Push a few regs?
    */
   x86_push(p, srcEDI);
   x86_push(p, countEBP);
   x86_push(p, vtxESI);


   /* Get vertex count, compare to zero
    */
   x86_xor(p, srcEDI, srcEDI);
   x86_mov(p, countEBP, make_fn_arg(p, 2));
   x86_cmp(p, countEBP, srcEDI);
   fixup = x86_jcc_forward(p, cc_E);

   /* Initialize destination register. 
    */
   x86_mov(p, vertexEAX, make_fn_arg(p, 3));

   /* Dereference ctx to get tnl, then vtx:
    */
   x86_mov(p, vtxESI, make_fn_arg(p, 1));
   x86_mov(p, vtxESI, make_disp(vtxESI, get_offset(ctx, &ctx->swtnl_context)));
   vtxESI = make_disp(vtxESI, get_offset(tnl, &tnl->clipspace));

   
   /* Possibly load vp0, vp1 for viewport calcs:
    */
   if (vtx->need_viewport) {
      sse_movups(p, vp0, make_disp(vtxESI, get_offset(vtx, &vtx->vp_scale[0])));
      sse_movups(p, vp1, make_disp(vtxESI, get_offset(vtx, &vtx->vp_xlate[0])));
   }

   /* always load, needed or not:
    */
   sse_movups(p, chan0, make_disp(vtxESI, get_offset(vtx, &vtx->chan_scale[0])));
   sse_movups(p, p->identity, make_disp(vtxESI, get_offset(vtx, &vtx->identity[0])));

   /* Note address for loop jump */
   label = get_label(p);

   /* Emit code for each of the attributes.  Currently routes
    * everything through SSE registers, even when it might be more
    * efficient to stick with regular old x86.  No optimization or
    * other tricks - enough new ground to cover here just getting
    * things working.
    */
   for (j = 0; j < vtx->attr_count; j++) {
      struct x86_reg dest = make_disp(vertexEAX, vtx->attr[j].vertoffset);
      struct x86_reg ptr_to_src = make_disp(vtxESI, get_offset(vtx, &vtx->attr[j].inputptr));

      /* Load current a[j].inputptr
       */
      x86_mov(p, srcEDI, ptr_to_src);

      /* Now, load an XMM reg from src, perhaps transform, then save.
       * Could be shortcircuited in specific cases:
       */
      switch (a[j].format) {
      case EMIT_1F:
	 emit_load(p, temp, 1, deref(srcEDI), vtx->attr[j].inputsize);
	 emit_store(p, dest, 1, temp);
	 break;
      case EMIT_2F:
	 emit_load(p, temp, 2, deref(srcEDI), vtx->attr[j].inputsize);
	 emit_store(p, dest, 2, temp);
	 break;
      case EMIT_3F:
	 /* Potentially the worst case - hardcode 2+1 copying:
	  */
	 emit_load(p, temp, 3, deref(srcEDI), vtx->attr[j].inputsize);
	 emit_store(p, dest, 3, temp);
	 break;
      case EMIT_4F:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 emit_store(p, dest, 4, temp);
	 break;
      case EMIT_2F_VIEWPORT: 
	 emit_load(p, temp, 2, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_mulps(p, temp, vp0);
	 sse_addps(p, temp, vp1);
	 emit_store(p, dest, 2, temp);
	 break;
      case EMIT_3F_VIEWPORT: 
	 emit_load(p, temp, 3, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_mulps(p, temp, vp0);
	 sse_addps(p, temp, vp1);
	 emit_store(p, dest, 3, temp);
	 break;
      case EMIT_4F_VIEWPORT: 
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_mulps(p, temp, vp0);
	 sse_addps(p, temp, vp1);
	 emit_store(p, dest, 4, temp);
	 break;
      case EMIT_3F_XYW:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_shufps(p, temp, temp, X, Y, W, Z);
	 emit_store(p, dest, 3, temp);
	 break;

	 /* Try and bond 3ub + 1ub pairs into a single 4ub operation?
	  */	 
      case EMIT_1UB_1F:	 
      case EMIT_3UB_3F_RGB:
      case EMIT_3UB_3F_BGR:
	 _mesa_printf("non-implemneted format %d\n", a[j].format);
	 return GL_FALSE;	/* add this later */

      case EMIT_4UB_4F_RGBA:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_mulps(p, temp, chan0);
	 emit_pack_store_4ub(p, dest, temp);
	 break;
      case EMIT_4UB_4F_BGRA:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_shufps(p, temp, temp, Z, Y, X, W);
	 sse_mulps(p, temp, chan0);
	 emit_pack_store_4ub(p, dest, temp);
	 break;
      case EMIT_4UB_4F_ARGB:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_shufps(p, temp, temp, W, X, Y, Z);
	 sse_mulps(p, temp, chan0);
	 emit_pack_store_4ub(p, dest, temp);
	 break;
      case EMIT_4UB_4F_ABGR:
	 emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	 sse_shufps(p, temp, temp, W, Z, Y, X);
	 sse_mulps(p, temp, chan0);
	 emit_pack_store_4ub(p, dest, temp);
	 break;
      case EMIT_4CHAN_4F_RGBA:
	 switch (CHAN_TYPE) {
	 case GL_UNSIGNED_BYTE:
	    emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	    sse_mulps(p, temp, chan0);
	    emit_pack_store_4ub(p, dest, temp);
	    break;
	 case GL_FLOAT:
	    emit_load(p, temp, 4, deref(srcEDI), vtx->attr[j].inputsize);
	    emit_store(p, dest, 4, temp);
	    break;
	 case GL_UNSIGNED_SHORT:
	 default:
	    _mesa_printf("unknown CHAN_TYPE %s\n", _mesa_lookup_enum_by_nr(CHAN_TYPE));
	    return GL_FALSE;
	 }
	 break;
      default:
	 _mesa_printf("unknown a[%d].format %d\n", j, a[j].format);
	 return GL_FALSE;	/* catch any new opcodes */
      }
      
      /* add a[j].inputstride (hardcoded value - could just as easily
       * pull the stride value from memory each time).
       */
      x86_lea(p, srcEDI, make_disp(srcEDI, a[j].inputstride));
      
      /* save new value of a[j].inputptr 
       */
      x86_mov(p, ptr_to_src, srcEDI);

   }

   /* Next vertex:
    */
   x86_lea(p, vertexEAX, make_disp(vertexEAX, vtx->vertex_size));

   /* decr count, loop if not zero
    */
   x86_dec(p, countEBP);
   x86_test(p, countEBP, countEBP); 
   x86_jcc(p, cc_NZ, label);

   /* Exit mmx state?
    */
   if (p->need_emms)
      mmx_emms(p);

   /* Land forward jump here:
    */
   do_fixup(p, fixup);

   /* Pop regs and return
    */
   x86_pop(p, get_base_reg(vtxESI));
   x86_pop(p, countEBP);
   x86_pop(p, srcEDI);
   x86_ret(p);

   vtx->emit = (tnl_emit_func)p->store;
   return GL_TRUE;
}

void _tnl_generate_sse_emit( GLcontext *ctx )
{
   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
   struct x86_program p;

   memset(&p, 0, sizeof(p));
   p.ctx = ctx;
   p.store = MALLOC(1024);

   p.inputs_safe = 1;		/* for now */
   p.outputs_safe = 1;		/* for now */
   p.have_sse2 = 0;		/* testing */
   p.identity = make_reg(file_XMM, 6);
   
   if (build_vertex_emit(&p)) {
      _tnl_register_fastpath( vtx, GL_TRUE );
      if (DISASSEM)
	 _mesa_printf("disassemble 0x%x 0x%x\n", p.store, p.csr);
   }
   else {
      /* Note the failure:
       */
      _tnl_register_fastpath( vtx, GL_FALSE );
      FREE(p.store);
   }

   (void)sse2_movd;
   (void)x86_inc;
   (void)x86_xor;
}