/
mojoshader.c
4699 lines (3953 loc) · 147 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
11
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
12
13
14
15
// !!! FIXME: do DEF* opcodes have to come before instructions?
// !!! FIXME: my reading of the msdn spec suggests no, but it sounds like
// !!! FIXME: something they'd require. DCL_* _does_ have to be first.
16
17
18
19
// Shader bytecode format is described at MSDN:
// http://msdn2.microsoft.com/en-us/library/ms800307.aspx
20
#include <stdio.h>
21
#include <string.h>
22
#include <stdlib.h>
23
#include <stdint.h>
24
#include <stdarg.h>
25
#include <assert.h>
26
27
#include "mojoshader.h"
28
29
30
31
32
33
34
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
35
36
37
// If SUPPORT_PROFILE_* isn't defined, we assume an implicit desire to support.
// You get all the profiles unless you go out of your way to disable them.
38
39
40
41
42
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
43
44
45
46
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
47
48
49
50
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
51
52
53
54
55
56
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
typedef uint8_t uint8;
typedef uint32_t uint32;
57
typedef int32_t int32;
58
59
60
61
62
63
64
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
65
66
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
67
68
69
70
71
72
73
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
74
75
76
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
// Byteswap magic...
#if ((defined __GNUC__) && (defined __POWERPC__))
static inline uint32 SWAP32(uint32 x)
{
__asm__ __volatile__("lwbrx %0,0,%1" : "=r" (x) : "r" (&x));
return x;
} // SWAP32
#elif defined(__POWERPC__)
static inline uint32 SWAP32(uint32 x)
{
return ( (((x) >> 24) & 0x000000FF) | (((x) >> 8) & 0x0000FF00) |
(((x) << 8) & 0x00FF0000) | (((x) << 24) & 0xFF000000) );
} // SWAP32
#else
# define SWAP32(x) (x)
#endif
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
typedef enum
{
REG_TYPE_TEMP = 0,
REG_TYPE_INPUT = 1,
REG_TYPE_CONST = 2,
REG_TYPE_ADDRESS = 3,
REG_TYPE_TEXTURE = 3, // ALSO 3!
REG_TYPE_RASTOUT = 4,
REG_TYPE_ATTROUT = 5,
REG_TYPE_TEXCRDOUT = 6,
REG_TYPE_OUTPUT = 6, // ALSO 6!
REG_TYPE_CONSTINT = 7,
REG_TYPE_COLOROUT = 8,
REG_TYPE_DEPTHOUT = 9,
REG_TYPE_SAMPLER = 10,
REG_TYPE_CONST2 = 11,
REG_TYPE_CONST3 = 12,
REG_TYPE_CONST4 = 13,
REG_TYPE_CONSTBOOL = 14,
REG_TYPE_LOOP = 15,
REG_TYPE_TEMPFLOAT16 = 16,
REG_TYPE_MISCTYPE = 17,
REG_TYPE_LABEL = 18,
REG_TYPE_PREDICATE = 19,
REG_TYPE_MAX = 19
} RegisterType;
122
123
124
125
126
127
128
129
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
130
// predeclare.
131
typedef struct Context Context;
132
133
// one emit function for each opcode in each profile.
134
typedef void (*emit_function)(Context *ctx);
135
136
// one emit function for starting output in each profile.
137
typedef void (*emit_start)(Context *ctx);
138
139
// one emit function for ending output in each profile.
140
typedef void (*emit_end)(Context *ctx);
141
142
143
144
145
146
147
148
149
150
// 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);
// one emit function for uniforms in each profile.
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum);
151
152
153
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
154
155
156
157
// one emit function for attributes in each profile.
typedef void (*emit_attribute)(Context *ctx, RegisterType regtype, int regnum,
MOJOSHADER_usage usage, int index, int wmask);
158
159
160
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
161
// one state function for each opcode where we have state machine updates.
162
typedef void (*state_function)(Context *ctx);
163
164
165
166
167
168
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
169
170
emit_global global_emitter;
emit_uniform uniform_emitter;
171
emit_sampler sampler_emitter;
172
emit_attribute attribute_emitter;
173
emit_finalize finalize_emitter;
174
} Profile;
175
176
177
178
179
180
181
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
182
} RastOutType;
183
184
185
186
187
188
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
189
} MiscTypeType;
190
191
192
193
194
// 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.
195
typedef struct OutputListNode
196
197
{
char *str;
198
199
200
201
202
203
204
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
205
206
} OutputList;
207
208
209
210
211
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
212
213
214
MOJOSHADER_usage usage;
int index;
int writemask;
215
216
217
218
struct RegisterList *next;
} RegisterList;
219
220
221
222
223
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
// source modifiers.
typedef enum
{
SRCMOD_NONE,
SRCMOD_NEGATE,
SRCMOD_BIAS,
SRCMOD_BIASNEGATE,
SRCMOD_SIGN,
SRCMOD_SIGNNEGATE,
SRCMOD_COMPLEMENT,
SRCMOD_X2,
SRCMOD_X2NEGATE,
SRCMOD_DZ,
SRCMOD_DW,
SRCMOD_ABS,
SRCMOD_ABSNEGATE,
SRCMOD_NOT,
SRCMOD_TOTAL
242
} SourceMod;
243
244
245
246
typedef struct
{
247
const uint32 *token; // this is the unmolested token in the stream.
248
249
int regnum;
int relative;
250
251
252
253
254
int writemask; // xyzw or rgba (all four, not split out).
int writemask0; // x or red
int writemask1; // y or green
int writemask2; // z or blue
int writemask3; // w or alpha
255
256
int result_mod;
int result_shift;
257
RegisterType regtype;
258
259
260
261
} DestArgInfo;
typedef struct
{
262
const uint32 *token; // this is the unmolested token in the stream.
263
int regnum;
264
int swizzle; // xyzw (all four, not split out).
265
266
267
268
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
269
SourceMod src_mod;
270
RegisterType regtype;
271
272
273
274
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
275
276
277
} SourceArgInfo;
278
279
typedef enum
{
280
// Specific to GLSL profile...
281
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
282
283
284
285
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
286
287
#define SCRATCH_BUFFER_SIZE 128
#define SCRATCH_BUFFERS 16
288
289
// !!! FIXME: the scratch buffers make Context pretty big.
290
291
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
292
293
// Context...this is state that changes as we parse through a shader...
294
struct Context
295
{
296
297
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
298
void *malloc_data;
299
300
const uint32 *tokens;
uint32 tokencount;
301
302
OutputList *output;
OutputList globals;
303
OutputList helpers;
304
305
OutputList subroutines;
OutputList mainline;
306
307
OutputList ignore;
OutputList *output_stack[2];
308
uint8 *output_bytes; // can be used instead of the OutputLists.
309
310
int indent_stack[2];
int output_stack_len;
311
int output_len; // total strlen; prevents walking the lists just to malloc.
312
int indent;
313
314
315
const char *endline;
int endline_len;
const char *failstr;
316
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
317
318
int scratchidx; // current scratch buffer.
int profileid;
319
const Profile *profile;
320
MOJOSHADER_shaderType shader_type;
321
322
uint8 major_ver;
uint8 minor_ver;
323
DestArgInfo dest_args[1];
324
SourceArgInfo source_args[5];
325
SourceArgInfo predicate_arg; // for predicated instructions.
326
uint32 dwords[4];
327
int instruction_count;
328
uint32 instruction_controls;
329
330
uint32 previous_opcode;
ContextFlags flags;
331
int predicated;
332
int loops;
333
int reps;
334
int cmps;
335
336
RegisterList used_registers;
RegisterList defined_registers;
337
338
int uniform_count;
RegisterList uniforms;
339
340
int attribute_count;
RegisterList attributes;
341
342
int sampler_count;
RegisterList samplers;
343
344
345
};
346
347
// Convenience functions for allocators...
348
static inline void *Malloc(const Context *ctx, const int len)
349
350
351
352
353
{
return ctx->malloc(len, ctx->malloc_data);
} // Malloc
354
static inline void Free(const Context *ctx, void *ptr)
355
{
356
if (ptr != NULL) // check for NULL in case of dumb free() impl.
357
ctx->free(ptr, ctx->malloc_data);
358
359
360
} // Free
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
// 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...
385
386
387
388
389
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
390
static inline int shader_version_supported(const uint8 maj, const uint8 min)
391
392
393
394
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
395
396
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
397
398
399
400
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
401
402
403
404
405
406
407
408
409
410
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
411
412
413
// Bit arrays (for storing large arrays of booleans...
414
static inline char *get_scratch_buffer(Context *ctx)
415
{
416
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
417
return ctx->scratch[ctx->scratchidx];
418
419
} // get_scratch_buffer
420
421
422
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
423
424
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
425
426
static inline int isfail(const Context *ctx)
427
428
429
430
431
{
return (ctx->failstr != NULL);
} // isfail
432
433
434
435
436
static MOJOSHADER_parseData out_of_mem_data = {
"Out of memory", 0, 0, 0, MOJOSHADER_TYPE_UNKNOWN, 0, 0, 0, 0
};
static const char *out_of_mem_str = "Out of memory";
437
438
439
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
440
ctx->failstr = out_of_mem_str; // fail() would call malloc().
441
442
443
444
445
446
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
447
{
448
449
if (ctx->failstr == NULL) // don't change existing error.
{
450
char *scratch = get_scratch_buffer(ctx);
451
452
va_list ap;
va_start(ap, fmt);
453
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
454
455
va_end(ap);
456
char *failstr = (char *) Malloc(ctx, len + 1);
457
458
459
460
461
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
462
if (len < SCRATCH_BUFFER_SIZE)
463
464
465
466
467
468
469
470
471
strcpy(failstr, scratch); // copy it over.
else
{
va_start(ap, fmt);
vsnprintf(failstr, len + 1, fmt, ap); // rebuild it.
va_end(ap);
} // else
ctx->failstr = failstr;
} // else
472
473
474
475
476
477
} // if
return FAIL;
} // failf
478
static inline int fail(Context *ctx, const char *reason)
479
480
481
482
483
{
return failf(ctx, "%s", reason);
} // fail
484
485
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
486
{
487
488
OutputListNode *item = NULL;
489
if (isfail(ctx))
490
491
return FAIL; // we failed previously, don't go on...
492
char *scratch = get_scratch_buffer(ctx);
493
494
const int indent = ctx->indent;
495
496
if (indent > 0)
memset(scratch, '\t', indent);
497
498
499
va_list ap;
va_start(ap, fmt);
500
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
501
502
va_end(ap);
503
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
504
505
506
if (item == NULL)
return out_of_memory(ctx);
507
item->str = (char *) Malloc(ctx, len + 1);
508
if (item->str == NULL)
509
{
510
Free(ctx, item);
511
return out_of_memory(ctx);
512
513
} // if
514
515
516
// 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().
517
if (len < SCRATCH_BUFFER_SIZE)
518
519
520
strcpy(item->str, scratch); // copy it over.
else
{
521
522
if (indent > 0)
memset(item->str, '\t', indent);
523
va_start(ap, fmt);
524
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
525
526
va_end(ap);
} // else
527
528
item->next = NULL;
529
530
531
ctx->output->tail->next = item;
ctx->output->tail = item;
532
ctx->output_len += len + ctx->endline_len;
533
534
535
536
537
return 0;
} // output_line
538
539
540
541
542
543
544
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
545
// !!! FIXME: this is sort of nasty.
546
547
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
548
549
{
const size_t len = snprintf(buf, bufsize, "%f", f);
550
if ((len+2) >= bufsize)
551
552
553
554
555
556
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
557
558
559
{
if (leavedecimal)
strcat(buf, ".0");
560
return; // done.
561
} // if
562
563
564
565
566
567
568
569
570
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
571
572
if ((leavedecimal) && (end == ptr))
end += 2;
573
574
575
576
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
577
578
579
// Deal with register lists... !!! FIXME: I sort of hate this.
580
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
581
582
583
584
{
while (item != NULL)
{
RegisterList *next = item->next;
585
f(item, d);
586
587
588
589
590
591
592
593
594
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
595
596
597
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
598
599
600
601
602
603
604
{
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)
605
return item; // already set, so we're done.
606
607
608
609
610
611
612
613
614
615
616
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).
617
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
618
619
620
621
622
623
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
624
625
626
item->usage = 0;
item->index = 0;
item->writemask = 0;
627
628
629
item->next = prev->next;
prev->next = item;
} // else
630
631
return item;
632
633
} // reglist_insert
634
635
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
636
{
637
638
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
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.
655
656
657
658
659
660
661
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
662
663
664
665
666
667
668
669
670
671
672
673
} // 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);
674
} // get_used_register
675
676
677
678
679
680
681
682
683
684
685
686
687
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
688
689
690
691
692
693
694
695
696
697
static void add_attribute_register(Context *ctx, const RegisterType rtype,
const int regnum, const MOJOSHADER_usage usage,
const int index, const int writemask)
{
RegisterList *item = reglist_insert(ctx, &ctx->attributes, rtype, regnum);
item->usage = usage;
item->index = index;
item->writemask = writemask;
} // add_attribute_register
698
699
700
701
702
703
704
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
705
706
707
708
709
710
711
712
713
714
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
715
716
// D3D stuff that's used in more than just the d3d profile...
717
718
719
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
720
721
722
723
724
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
725
726
static const char *get_D3D_register_string(Context *ctx,
727
RegisterType regtype,
728
729
int regnum, char *regnum_str,
size_t regnum_size)
730
{
731
const char *retval = NULL;
732
int has_number = 1;
733
734
switch (regtype)
735
{
736
case REG_TYPE_TEMP:
737
retval = "r";
738
739
break;
740
case REG_TYPE_INPUT:
741
retval = "v";
742
743
break;
744
case REG_TYPE_CONST:
745
746
747
retval = "c";
break;
748
case REG_TYPE_CONST2:
749
750
751
752
retval = "c";
regnum += 2048;
break;
753
case REG_TYPE_CONST3:
754
755
756
757
retval = "c";
regnum += 4096;
break;
758
case REG_TYPE_CONST4:
759
retval = "c";
760
regnum += 6144;
761
762
break;
763
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
764
retval = shader_is_vertex(ctx) ? "a" : "t";
765
766
break;
767
case REG_TYPE_RASTOUT:
768
switch ((RastOutType) regnum)
769
{
770
771
772
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
773
} // switch
774
has_number = 0;
775
776
break;
777
case REG_TYPE_ATTROUT:
778
retval = "oD";
779
780
break;
781
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
782
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
783
retval = "o";
784
else
785
retval = "oT";
786
787
break;
788
case REG_TYPE_CONSTINT:
789
retval = "i";
790
791
break;
792
case REG_TYPE_COLOROUT:
793
retval = "oC";
794
795
break;
796
case REG_TYPE_DEPTHOUT:
797
retval = "oDepth";
798
has_number = 0;
799
800
break;
801
case REG_TYPE_SAMPLER:
802
retval = "s";
803
804
break;
805
case REG_TYPE_CONSTBOOL:
806
retval = "b";
807
808
break;
809
case REG_TYPE_LOOP:
810
retval = "aL";
811
has_number = 0;
812
813
814
break;
// !!! FIXME: don't know what the asm string is for this..
815
case REG_TYPE_TEMPFLOAT16:
816
fail(ctx, "don't know the ASM for tempfloat16 register");
817
818
819
retval = "???";
has_number = 0;
break;
820
821
case REG_TYPE_MISCTYPE:
822
switch ((MiscTypeType) regnum)
823
{
824
825
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
826
} // switch
827
has_number = 0;
828
829
break;
830
case REG_TYPE_LABEL:
831
retval = "l";
832
833
break;
834
case REG_TYPE_PREDICATE:
835
retval = "p";
836
837
838
break;
} // switch
839
840
841
842
843
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
844
845
846
847
return retval;
} // get_D3D_register_string
848
849
850
851
852
853
854
855
856
857
#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,
858
859
860
static const char *make_D3D_sourcearg_string_in_buf(Context *ctx,
const SourceArgInfo *arg,
char *buf, size_t buflen)
861
862
863
{
const char *premod_str = "";
const char *postmod_str = "";
864
switch (arg->src_mod)
865
866
867
868
869
870
871
872
873
874
875
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
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
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;
913
914
915
916
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
917
918
919
920
} // switch
char regnum_str[16];
921
922
923
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
924
925
926
927
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
928
return "";
929
930
} // if
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
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
955
956
char swizzle_str[6];
int i = 0;
957
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
958
959
{
swizzle_str[i++] = '.';
960
961
962
963
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];
964
965
966
967
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
968
969
970
971
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
972
973
974
975
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);
976
// !!! FIXME: make sure the scratch buffer was large enough.
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
return buf;
} // make_D3D_sourcearg_string_in_buf
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
return "";
} // if
const DestArgInfo *arg = &ctx->dest_args[idx];
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