/
mojoshader.c
7635 lines (6448 loc) · 245 KB
1
/**
2
3
* MojoShader; generate shader programs from bytecode of compiled
* Direct3D shaders.
4
5
6
7
8
9
*
* Please see the file LICENSE.txt in the source's root directory.
*
* This file written by Ryan C. Gordon.
*/
10
// !!! FIXME: this file really needs to be split up.
11
12
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
13
14
#define __MOJOSHADER_INTERNAL__ 1
#include "mojoshader_internal.h"
15
16
17
// predeclare.
18
typedef struct Context Context;
19
struct ConstantsList;
20
21
// one emit function for each opcode in each profile.
22
typedef void (*emit_function)(Context *ctx);
23
24
// one emit function for starting output in each profile.
25
typedef void (*emit_start)(Context *ctx, const char *profilestr);
26
27
// one emit function for ending output in each profile.
28
typedef void (*emit_end)(Context *ctx);
29
30
31
32
// one emit function for phase opcode output in each profile.
typedef void (*emit_phase)(Context *ctx);
33
34
35
36
37
38
// one emit function for finalizing output in each profile.
typedef void (*emit_finalize)(Context *ctx);
// one emit function for global definitions in each profile.
typedef void (*emit_global)(Context *ctx, RegisterType regtype, int regnum);
39
// one emit function for relative uniform arrays in each profile.
40
typedef void (*emit_array)(Context *ctx, int base, int size);
41
42
43
44
45
46
// one emit function for relative constants arrays in each profile.
typedef void (*emit_const_array)(Context *ctx,
const struct ConstantsList *constslist,
int base, int size);
47
// one emit function for uniforms in each profile.
48
49
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum,
int arraybase, int arraysize);
50
51
52
53
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
54
55
// one emit function for attributes in each profile.
typedef void (*emit_attribute)(Context *ctx, RegisterType regtype, int regnum,
56
57
MOJOSHADER_usage usage, int index, int wmask,
int flags);
58
59
60
61
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
62
// one state function for each opcode where we have state machine updates.
63
typedef void (*state_function)(Context *ctx);
64
65
66
67
68
// one function for varnames in each profile.
typedef const char *(*varname_function)(Context *c, RegisterType t, int num);
// one function for const var array in each profile.
69
typedef const char *(*const_array_varname_function)(Context *c, int base, int size);
70
71
72
73
74
75
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
76
emit_phase phase_emitter;
77
emit_global global_emitter;
78
emit_array array_emitter;
79
emit_const_array const_array_emitter;
80
emit_uniform uniform_emitter;
81
emit_sampler sampler_emitter;
82
emit_attribute attribute_emitter;
83
emit_finalize finalize_emitter;
84
85
varname_function get_varname;
const_array_varname_function get_const_array_varname;
86
} Profile;
87
88
89
90
91
// A simple linked list of strings, so we can build the final output without
// realloc()ing for each new line, and easily insert lines into the middle
// of the output without much trouble.
92
typedef struct OutputListNode
93
94
{
char *str;
95
96
97
98
99
100
101
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
102
103
} OutputList;
104
105
106
107
108
109
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
110
111
112
113
114
typedef struct VariableList
{
MOJOSHADER_uniformType type;
int index;
int count;
115
ConstantsList *constant;
116
int used;
117
118
struct VariableList *next;
} VariableList;
119
120
121
122
123
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
124
125
126
MOJOSHADER_usage usage;
int index;
int writemask;
127
int misc;
128
const VariableList *array;
129
130
131
struct RegisterList *next;
} RegisterList;
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
typedef struct
{
const uint32 *token; // this is the unmolested token in the stream.
int regnum;
int swizzle; // xyzw (all four, not split out).
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
SourceMod src_mod;
RegisterType regtype;
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
const VariableList *relative_array;
} SourceArgInfo;
149
150
#define SCRATCH_BUFFER_SIZE 128
151
#define SCRATCH_BUFFERS 32
152
153
// !!! FIXME: the scratch buffers make Context pretty big.
154
155
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
156
157
// Context...this is state that changes as we parse through a shader...
158
struct Context
159
{
160
161
int isfail;
int out_of_memory;
162
163
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
164
void *malloc_data;
165
const uint32 *orig_tokens;
166
167
const uint32 *tokens;
uint32 tokencount;
168
MOJOSHADER_parsePhase parse_phase;
169
170
const MOJOSHADER_swizzle *swizzles;
unsigned int swizzles_count;
171
172
OutputList *output;
OutputList globals;
173
OutputList helpers;
174
OutputList subroutines;
175
OutputList mainline_intro;
176
OutputList mainline;
177
178
OutputList ignore;
OutputList *output_stack[2];
179
uint8 *output_bytes; // can be used instead of the OutputLists.
180
181
int indent_stack[2];
int output_stack_len;
182
int output_len; // total strlen; prevents walking the lists just to malloc.
183
int indent;
184
const char *shader_type_str;
185
186
const char *endline;
int endline_len;
187
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
188
189
int scratchidx; // current scratch buffer.
int profileid;
190
const Profile *profile;
191
MOJOSHADER_shaderType shader_type;
192
193
uint8 major_ver;
uint8 minor_ver;
194
DestArgInfo dest_arg;
195
SourceArgInfo source_args[5];
196
SourceArgInfo predicate_arg; // for predicated instructions.
197
uint32 dwords[4];
198
uint32 version_token;
199
int instruction_count;
200
uint32 instruction_controls;
201
202
uint32 previous_opcode;
int loops;
203
int reps;
204
int max_reps;
205
int cmps;
206
207
int scratch_registers;
int max_scratch_registers;
208
209
210
int branch_labels_stack_index;
int branch_labels_stack[32];
int assigned_branch_labels;
211
int assigned_vertex_attributes;
212
int last_address_reg_component;
213
214
RegisterList used_registers;
RegisterList defined_registers;
215
216
int error_count;
ErrorList *errors;
217
218
int constant_count;
ConstantsList *constants;
219
220
int uniform_count;
RegisterList uniforms;
221
222
int attribute_count;
RegisterList attributes;
223
224
int sampler_count;
RegisterList samplers;
225
VariableList *variables; // variables to register mapping.
226
227
int centroid_allowed;
int have_ctab;
228
int have_relative_input_registers;
229
230
231
232
233
234
235
int determined_constants_arrays;
int predicated;
int support_nv2;
int support_nv3;
int support_nv4;
int support_glsl120;
int glsl_generated_lit_opcode;
236
237
238
};
239
240
// Convenience functions for allocators...
241
static MOJOSHADER_error out_of_mem_error = { "Out of memory", NULL, -1 };
242
MOJOSHADER_parseData out_of_mem_data = {
243
1, &out_of_mem_error, 0, 0, 0, 0, MOJOSHADER_TYPE_UNKNOWN, 0, 0, 0, 0
244
245
};
246
static inline void out_of_memory(Context *ctx)
247
{
248
ctx->isfail = ctx->out_of_memory = 1;
249
250
} // out_of_memory
251
static inline void *Malloc(Context *ctx, const size_t len)
252
{
253
void *retval = ctx->malloc((int) len, ctx->malloc_data);
254
255
256
if (retval == NULL)
out_of_memory(ctx);
return retval;
257
258
} // Malloc
259
260
261
262
263
264
265
266
267
static inline char *StrDup(Context *ctx, const char *str)
{
char *retval = (char *) Malloc(ctx, strlen(str) + 1);
if (retval == NULL)
out_of_memory(ctx);
else
strcpy(retval, str);
return retval;
} // StrDup
268
269
static inline void Free(Context *ctx, void *ptr)
270
{
271
if (ptr != NULL) // check for NULL in case of dumb free() impl.
272
ctx->free(ptr, ctx->malloc_data);
273
274
275
} // Free
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
// jump between output sections in the context...
static inline void push_output(Context *ctx, OutputList *section)
{
assert(ctx->output_stack_len < STATICARRAYLEN(ctx->output_stack));
ctx->output_stack[ctx->output_stack_len] = ctx->output;
ctx->indent_stack[ctx->output_stack_len] = ctx->indent;
ctx->output_stack_len++;
ctx->output = section;
ctx->indent = 0;
} // push_output
static inline void pop_output(Context *ctx)
{
assert(ctx->output_stack_len > 0);
ctx->output_stack_len--;
ctx->output = ctx->output_stack[ctx->output_stack_len];
ctx->indent = ctx->indent_stack[ctx->output_stack_len];
} // pop_output
// Shader model version magic...
300
301
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
302
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 1 : (minor)) );
303
304
} // version_ui32
305
static inline int shader_version_supported(const uint8 maj, const uint8 min)
306
307
308
309
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
310
311
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
312
313
314
315
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
316
317
318
319
320
321
static inline int shader_version_exactly(const Context *ctx, const uint8 maj,
const uint8 min)
{
return ((ctx->major_ver == maj) && (ctx->minor_ver == min));
} // shader_version_exactly
322
323
324
325
326
327
328
329
330
331
static inline int shader_is_pixel(const Context *ctx)
{
return (ctx->shader_type == MOJOSHADER_TYPE_PIXEL);
} // shader_is_pixel
static inline int shader_is_vertex(const Context *ctx)
{
return (ctx->shader_type == MOJOSHADER_TYPE_VERTEX);
} // shader_is_vertex
332
333
334
static inline int isfail(const Context *ctx)
{
335
return ctx->isfail;
336
337
} // isfail
338
339
static inline char *get_scratch_buffer(Context *ctx)
340
{
341
assert(ctx->scratchidx < SCRATCH_BUFFERS);
342
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
343
return ctx->scratch[ctx->scratchidx];
344
345
} // get_scratch_buffer
346
347
348
static void failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static void failf(Context *ctx, const char *fmt, ...)
349
{
350
351
ctx->isfail = 1;
if (ctx->out_of_memory)
352
return;
353
354
355
int error_position = 0;
switch (ctx->parse_phase)
356
{
357
358
359
360
361
362
363
364
365
366
367
case MOJOSHADER_PARSEPHASE_NOTSTARTED:
error_position = -2;
break;
case MOJOSHADER_PARSEPHASE_WORKING:
error_position = (ctx->tokens - ctx->orig_tokens) * sizeof (uint32);
break;
case MOJOSHADER_PARSEPHASE_DONE:
error_position = -1;
break;
default:
assert(0 && "Unexpected value");
368
return;
369
370
371
372
} // switch
ErrorList *error = (ErrorList *) Malloc(ctx, sizeof (ErrorList));
if (error == NULL)
373
return;
374
375
376
377
378
379
char *scratch = get_scratch_buffer(ctx);
va_list ap;
va_start(ap, fmt);
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
va_end(ap);
380
381
382
383
384
385
386
387
388
389
char *failstr = (char *) Malloc(ctx, len + 1);
if (failstr == NULL)
Free(ctx, error);
else
{
// see comments about scratch buffer overflow in output_line().
if (len < SCRATCH_BUFFER_SIZE)
strcpy(failstr, scratch); // copy it over.
else
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
va_start(ap, fmt);
vsnprintf(failstr, len + 1, fmt, ap); // rebuild it.
va_end(ap);
} // else
error->error.error = failstr;
error->error.filename = NULL; // no filename at this level.
error->error.error_position = error_position;
ErrorList *prev = NULL;
error->next = ctx->errors;
while (error->next != NULL)
{
prev = error->next;
error->next = error->next->next;
} // while
if (prev != NULL)
prev->next = error;
else
ctx->errors = error;
ctx->error_count++;
} // else
415
416
417
} // failf
418
static inline void fail(Context *ctx, const char *reason)
419
{
420
failf(ctx, "%s", reason);
421
422
423
} // fail
424
425
static void output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static void output_line(Context *ctx, const char *fmt, ...)
426
{
427
428
OutputListNode *item = NULL;
429
430
if (isfail(ctx) || ctx->out_of_memory)
return; // we failed previously, don't go on...
431
432
char *scratch = get_scratch_buffer(ctx);
433
434
const int indent = ctx->indent;
435
436
if (indent > 0)
memset(scratch, '\t', indent);
437
438
439
va_list ap;
va_start(ap, fmt);
440
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
441
442
va_end(ap);
443
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
444
if (item == NULL)
445
return;
446
447
item->str = (char *) Malloc(ctx, len + 1);
448
if (item->str == NULL)
449
{
450
Free(ctx, item);
451
return;
452
453
} // if
454
455
456
// If we overflowed our scratch buffer, that's okay. We were going to
// allocate anyhow...the scratch buffer just lets us avoid a second
// run of vsnprintf().
457
if (len < SCRATCH_BUFFER_SIZE)
458
459
460
strcpy(item->str, scratch); // copy it over.
else
{
461
462
if (indent > 0)
memset(item->str, '\t', indent);
463
va_start(ap, fmt);
464
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
465
466
va_end(ap);
} // else
467
468
item->next = NULL;
469
470
471
ctx->output->tail->next = item;
ctx->output->tail = item;
472
ctx->output_len += len + ctx->endline_len;
473
474
475
} // output_line
476
// this is just to stop gcc whining.
477
static inline void output_blank_line(Context *ctx)
478
{
479
output_line(ctx, "%s", "");
480
481
482
} // output_blank_line
483
// !!! FIXME: this is sort of nasty.
484
485
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
486
487
{
const size_t len = snprintf(buf, bufsize, "%f", f);
488
if ((len+2) >= bufsize)
489
490
491
492
493
494
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
495
496
497
{
if (leavedecimal)
strcat(buf, ".0");
498
return; // done.
499
} // if
500
501
502
503
504
505
506
507
508
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
509
510
if ((leavedecimal) && (end == ptr))
end += 2;
511
512
513
514
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
515
516
517
// Deal with register lists... !!! FIXME: I sort of hate this.
518
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
519
520
521
522
{
while (item != NULL)
{
RegisterList *next = item->next;
523
f(item, d);
524
525
526
527
528
529
530
531
532
item = next;
} // while
} // free_reglist
static inline uint32 reg_to_ui32(const RegisterType regtype, const int regnum)
{
return ( ((uint32) regtype) | (((uint32) regnum) << 16) );
} // reg_to_uint32
533
534
535
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
536
537
538
539
540
541
542
{
const uint32 newval = reg_to_ui32(regtype, regnum);
RegisterList *item = prev->next;
while (item != NULL)
{
const uint32 val = reg_to_ui32(item->regtype, item->regnum);
if (newval == val)
543
return item; // already set, so we're done.
544
545
546
547
548
549
550
551
552
553
554
else if (newval < val) // insert it here.
break;
else // if (newval > val)
{
// keep going, we're not to the insertion point yet.
prev = item;
item = item->next;
} // else
} // while
// we need to insert an entry after (prev).
555
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
556
if (item != NULL)
557
558
559
{
item->regtype = regtype;
item->regnum = regnum;
560
item->usage = MOJOSHADER_USAGE_UNKNOWN;
561
562
item->index = 0;
item->writemask = 0;
563
item->misc = 0;
564
item->array = NULL;
565
566
item->next = prev->next;
prev->next = item;
567
} // if
568
569
return item;
570
571
} // reglist_insert
572
573
static RegisterList *reglist_find(const RegisterList *prev,
const RegisterType rtype, const int regnum)
574
{
575
576
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
while (item != NULL)
{
const uint32 val = reg_to_ui32(item->regtype, item->regnum);
if (newval == val)
return item; // here it is.
else if (newval < val) // should have been here if it existed.
return NULL;
else // if (newval > val)
{
// keep going, we're not to the insertion point yet.
prev = item;
item = item->next;
} // else
} // while
return NULL; // wasn't in the list.
593
594
595
596
597
598
599
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
600
601
602
603
604
605
606
607
608
609
610
611
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
} // set_used_register
static inline int get_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
return (reglist_exists(&ctx->used_registers, regtype, regnum) != NULL);
612
} // get_used_register
613
614
615
616
617
618
619
620
621
622
623
624
625
static inline void set_defined_register(Context *ctx, const RegisterType rtype,
const int regnum)
{
reglist_insert(ctx, &ctx->defined_registers, rtype, regnum);
} // set_defined_register
static inline int get_defined_register(Context *ctx, const RegisterType rtype,
const int regnum)
{
return (reglist_exists(&ctx->defined_registers, rtype, regnum) != NULL);
} // get_defined_register
626
627
static void add_attribute_register(Context *ctx, const RegisterType rtype,
const int regnum, const MOJOSHADER_usage usage,
628
const int index, const int writemask, int flags)
629
630
631
632
633
{
RegisterList *item = reglist_insert(ctx, &ctx->attributes, rtype, regnum);
item->usage = usage;
item->index = index;
item->writemask = writemask;
634
item->misc = flags;
635
636
} // add_attribute_register
637
638
639
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
640
// !!! FIXME: make sure it doesn't exist?
641
642
643
644
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
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
static inline int writemask_xyzw(const int writemask)
{
return (writemask == 0xF); // 0xF == 1111. No explicit mask (full!).
} // writemask_xyzw
static inline int writemask_xyz(const int writemask)
{
return (writemask == 0x7); // 0x7 == 0111. (that is: xyz)
} // writemask_xyz
static inline int writemask_xy(const int writemask)
{
return (writemask == 0x3); // 0x3 == 0011. (that is: xy)
} // writemask_xy
static inline int writemask_x(const int writemask)
{
return (writemask == 0x1); // 0x1 == 0001. (that is: x)
} // writemask_x
static inline int writemask_y(const int writemask)
{
return (writemask == 0x2); // 0x1 == 0010. (that is: y)
} // writemask_y
676
677
678
679
680
681
682
683
static inline int replicate_swizzle(const int swizzle)
{
return ( (((swizzle >> 0) & 0x3) == ((swizzle >> 2) & 0x3)) &&
(((swizzle >> 2) & 0x3) == ((swizzle >> 4) & 0x3)) &&
(((swizzle >> 4) & 0x3) == ((swizzle >> 6) & 0x3)) );
} // replicate_swizzle
684
685
static inline int no_swizzle(const int swizzle)
{
686
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
687
688
689
} // no_swizzle
690
691
692
693
694
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
695
696
697
698
699
700
701
702
static int allocate_scratch_register(Context *ctx)
{
const int retval = ctx->scratch_registers++;
if (retval >= ctx->max_scratch_registers)
ctx->max_scratch_registers = retval + 1;
return retval;
} // allocate_scratch_register
703
static int allocate_branch_label(Context *ctx)
704
{
705
706
return ctx->assigned_branch_labels++;
} // allocate_branch_label
707
708
709
710
// D3D stuff that's used in more than just the d3d profile...
711
712
713
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
714
715
716
717
718
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
719
720
static const char *get_D3D_register_string(Context *ctx,
721
RegisterType regtype,
722
723
int regnum, char *regnum_str,
size_t regnum_size)
724
{
725
const char *retval = NULL;
726
int has_number = 1;
727
728
switch (regtype)
729
{
730
case REG_TYPE_TEMP:
731
retval = "r";
732
733
break;
734
case REG_TYPE_INPUT:
735
retval = "v";
736
737
break;
738
case REG_TYPE_CONST:
739
740
741
retval = "c";
break;
742
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
743
retval = shader_is_vertex(ctx) ? "a" : "t";
744
745
break;
746
case REG_TYPE_RASTOUT:
747
switch ((RastOutType) regnum)
748
{
749
750
751
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
752
} // switch
753
has_number = 0;
754
755
break;
756
case REG_TYPE_ATTROUT:
757
retval = "oD";
758
759
break;
760
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
761
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
762
retval = "o";
763
else
764
retval = "oT";
765
766
break;
767
case REG_TYPE_CONSTINT:
768
retval = "i";
769
770
break;
771
case REG_TYPE_COLOROUT:
772
retval = "oC";
773
774
break;
775
case REG_TYPE_DEPTHOUT:
776
retval = "oDepth";
777
has_number = 0;
778
779
break;
780
case REG_TYPE_SAMPLER:
781
retval = "s";
782
783
break;
784
case REG_TYPE_CONSTBOOL:
785
retval = "b";
786
787
break;
788
case REG_TYPE_LOOP:
789
retval = "aL";
790
has_number = 0;
791
792
break;
793
case REG_TYPE_MISCTYPE:
794
switch ((const MiscTypeType) regnum)
795
{
796
797
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
798
} // switch
799
has_number = 0;
800
801
break;
802
case REG_TYPE_LABEL:
803
retval = "l";
804
805
break;
806
case REG_TYPE_PREDICATE:
807
retval = "p";
808
break;
809
810
811
812
813
814
815
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
816
817
} // switch
818
819
820
821
822
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
823
824
825
826
return retval;
} // get_D3D_register_string
827
828
829
830
831
832
833
834
835
#define AT_LEAST_ONE_PROFILE 0
#if !SUPPORT_PROFILE_D3D
#define PROFILE_EMITTER_D3D(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_D3D(op) emit_D3D_##op,
836
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
837
838
const SourceArgInfo *arg,
char *buf, size_t buflen)
839
840
841
{
const char *premod_str = "";
const char *postmod_str = "";
842
switch (arg->src_mod)
843
844
845
846
847
848
849
850
851
852
853
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
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
case SRCMOD_SIGNNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_SIGN:
postmod_str = "_bx2";
break;
case SRCMOD_COMPLEMENT:
premod_str = "1-";
break;
case SRCMOD_X2NEGATE:
premod_str = "-";
// fall through.
case SRCMOD_X2:
postmod_str = "_x2";
break;
case SRCMOD_DZ:
postmod_str = "_dz";
break;
case SRCMOD_DW:
postmod_str = "_dw";
break;
case SRCMOD_ABSNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_ABS:
postmod_str = "_abs";
break;
case SRCMOD_NOT:
premod_str = "!";
break;
891
892
893
894
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
895
896
897
898
} // switch
char regnum_str[16];
899
900
901
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
902
903
904
905
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
906
return "";
907
908
} // if
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
const char *rel_lbracket = "";
const char *rel_rbracket = "";
char rel_swizzle[4] = { '\0' };
char rel_regnum_str[16] = { '\0' };
const char *rel_regtype_str = "";
if (arg->relative)
{
rel_swizzle[0] = '.';
rel_swizzle[1] = swizzle_channels[arg->relative_component];
rel_swizzle[2] = '\0';
rel_lbracket = "[";
rel_rbracket = "]";
rel_regtype_str = get_D3D_register_string(ctx, arg->relative_regtype,
arg->relative_regnum,
rel_regnum_str,
sizeof (rel_regnum_str));
if (regtype_str == NULL)
{
fail(ctx, "Unknown relative source register type.");
return "";
} // if
} // if
933
934
char swizzle_str[6];
int i = 0;
935
const int scalar = scalar_register(ctx->shader_type, arg->regtype, arg->regnum);
936
if (!scalar && !no_swizzle(arg->swizzle))
937
938
{
swizzle_str[i++] = '.';
939
940
941
942
swizzle_str[i++] = swizzle_channels[arg->swizzle_x];
swizzle_str[i++] = swizzle_channels[arg->swizzle_y];
swizzle_str[i++] = swizzle_channels[arg->swizzle_z];
swizzle_str[i++] = swizzle_channels[arg->swizzle_w];
943
944
945
946
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
947
948
949
950
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
951
// !!! FIXME: c12[a0.x] actually needs to be c[a0.x + 12]
952
953
954
955
snprintf(buf, buflen, "%s%s%s%s%s%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str,
rel_lbracket, rel_regtype_str, rel_regnum_str, rel_swizzle,
rel_rbracket, swizzle_str);
956
// !!! FIXME: make sure the scratch buffer was large enough.
957
return buf;
958
} // make_D3D_srcarg_string_in_buf
959
960
961
static const char *make_D3D_destarg_string(Context *ctx)
962
{
963
const DestArgInfo *arg = &ctx->dest_arg;
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
const char *result_shift_str = "";
switch (arg->result_shift)
{
case 0x1: result_shift_str = "_x2"; break;
case 0x2: result_shift_str = "_x4"; break;
case 0x3: result_shift_str = "_x8"; break;
case 0xD: result_shift_str = "_d8"; break;
case 0xE: result_shift_str = "_d4"; break;
case 0xF: result_shift_str = "_d2"; break;
} // switch
const char *sat_str = (arg->result_mod & MOD_SATURATE) ? "_sat" : "";
const char *pp_str = (arg->result_mod & MOD_PP) ? "_pp" : "";
const char *cent_str = (arg->result_mod & MOD_CENTROID) ? "_centroid" : "";
char regnum_str[16];
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
return "";
} // if
char writemask_str[6];
int i = 0;
992
const int scalar = scalar_register(ctx->shader_type, arg->regtype, arg->regnum);
993
if (!scalar && !writemask_xyzw(arg->writemask))
994
995
996
997
998
999
1000
{
writemask_str[i++] = '.';
if (arg->writemask0) writemask_str[i++] = 'x';
if (arg->writemask1) writemask_str[i++] = 'y';
if (arg->writemask2) writemask_str[i++] = 'z';
if (arg->writemask3) writemask_str[i++] = 'w';
} // if