/
mojoshader.c
4912 lines (4127 loc) · 154 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
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
73
74
75
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
76
77
78
79
80
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
// 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
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
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;
126
127
128
129
130
131
132
133
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
134
// predeclare.
135
typedef struct Context Context;
136
137
// one emit function for each opcode in each profile.
138
typedef void (*emit_function)(Context *ctx);
139
140
// one emit function for starting output in each profile.
141
typedef void (*emit_start)(Context *ctx);
142
143
// one emit function for ending output in each profile.
144
typedef void (*emit_end)(Context *ctx);
145
146
147
148
149
150
151
152
153
154
// 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);
155
156
157
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
158
159
160
161
// 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);
162
163
164
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
165
// one state function for each opcode where we have state machine updates.
166
typedef void (*state_function)(Context *ctx);
167
168
169
170
171
172
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
173
174
emit_global global_emitter;
emit_uniform uniform_emitter;
175
emit_sampler sampler_emitter;
176
emit_attribute attribute_emitter;
177
emit_finalize finalize_emitter;
178
} Profile;
179
180
181
182
183
184
185
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
186
} RastOutType;
187
188
189
190
191
192
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
193
} MiscTypeType;
194
195
196
197
198
// 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.
199
typedef struct OutputListNode
200
201
{
char *str;
202
203
204
205
206
207
208
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
209
210
} OutputList;
211
212
213
214
215
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
216
217
218
MOJOSHADER_usage usage;
int index;
int writemask;
219
220
221
222
struct RegisterList *next;
} RegisterList;
223
224
225
226
227
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
// 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
246
} SourceMod;
247
248
249
250
typedef struct
{
251
const uint32 *token; // this is the unmolested token in the stream.
252
253
int regnum;
int relative;
254
255
256
257
258
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
259
260
int result_mod;
int result_shift;
261
RegisterType regtype;
262
263
264
265
} DestArgInfo;
typedef struct
{
266
const uint32 *token; // this is the unmolested token in the stream.
267
int regnum;
268
int swizzle; // xyzw (all four, not split out).
269
270
271
272
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
273
SourceMod src_mod;
274
RegisterType regtype;
275
276
277
278
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
279
280
281
} SourceArgInfo;
282
283
typedef enum
{
284
// Specific to GLSL profile...
285
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
286
287
288
289
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
290
291
#define SCRATCH_BUFFER_SIZE 128
#define SCRATCH_BUFFERS 16
292
293
// !!! FIXME: the scratch buffers make Context pretty big.
294
295
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
296
297
// Context...this is state that changes as we parse through a shader...
298
struct Context
299
{
300
301
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
302
void *malloc_data;
303
304
const uint32 *tokens;
uint32 tokencount;
305
306
OutputList *output;
OutputList globals;
307
OutputList helpers;
308
OutputList subroutines;
309
OutputList mainline_intro;
310
OutputList mainline;
311
312
OutputList ignore;
OutputList *output_stack[2];
313
uint8 *output_bytes; // can be used instead of the OutputLists.
314
315
int indent_stack[2];
int output_stack_len;
316
int output_len; // total strlen; prevents walking the lists just to malloc.
317
int indent;
318
319
320
const char *endline;
int endline_len;
const char *failstr;
321
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
322
323
int scratchidx; // current scratch buffer.
int profileid;
324
const Profile *profile;
325
MOJOSHADER_shaderType shader_type;
326
327
uint8 major_ver;
uint8 minor_ver;
328
DestArgInfo dest_arg;
329
SourceArgInfo source_args[5];
330
SourceArgInfo predicate_arg; // for predicated instructions.
331
uint32 dwords[4];
332
int instruction_count;
333
uint32 instruction_controls;
334
335
uint32 previous_opcode;
ContextFlags flags;
336
int predicated;
337
int loops;
338
int reps;
339
int cmps;
340
341
RegisterList used_registers;
RegisterList defined_registers;
342
343
int uniform_count;
RegisterList uniforms;
344
345
int attribute_count;
RegisterList attributes;
346
347
int sampler_count;
RegisterList samplers;
348
349
350
};
351
352
// Convenience functions for allocators...
353
354
355
356
357
358
static MOJOSHADER_parseData out_of_mem_data = {
"Out of memory", 0, 0, 0, 0, MOJOSHADER_TYPE_UNKNOWN, 0, 0, 0, 0
};
static const char *out_of_mem_str = "Out of memory";
static inline int out_of_memory(Context *ctx)
359
{
360
361
362
363
364
365
366
367
368
369
370
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
static inline void *Malloc(Context *ctx, const int len)
{
void *retval = ctx->malloc(len, ctx->malloc_data);
if (retval == NULL)
out_of_memory(ctx);
return retval;
371
372
373
} // Malloc
374
static inline void Free(Context *ctx, void *ptr)
375
{
376
if (ptr != NULL) // check for NULL in case of dumb free() impl.
377
ctx->free(ptr, ctx->malloc_data);
378
379
380
} // Free
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
// 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...
405
406
407
408
409
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
410
static inline int shader_version_supported(const uint8 maj, const uint8 min)
411
412
413
414
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
415
416
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
417
418
419
420
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
421
422
423
424
425
426
427
428
429
430
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
431
432
433
// Bit arrays (for storing large arrays of booleans...
434
static inline char *get_scratch_buffer(Context *ctx)
435
{
436
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
437
return ctx->scratch[ctx->scratchidx];
438
439
} // get_scratch_buffer
440
441
static inline int isfail(const Context *ctx)
442
443
444
445
446
{
return (ctx->failstr != NULL);
} // isfail
447
448
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
449
{
450
451
if (ctx->failstr == NULL) // don't change existing error.
{
452
char *scratch = get_scratch_buffer(ctx);
453
454
va_list ap;
va_start(ap, fmt);
455
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
456
457
va_end(ap);
458
char *failstr = (char *) Malloc(ctx, len + 1);
459
if (failstr != NULL)
460
461
{
// see comments about scratch buffer overflow in output_line().
462
if (len < SCRATCH_BUFFER_SIZE)
463
464
465
466
467
468
469
470
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;
471
} // if
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
if (item == NULL)
505
return FAIL;
506
507
item->str = (char *) Malloc(ctx, len + 1);
508
if (item->str == NULL)
509
{
510
Free(ctx, item);
511
return FAIL;
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
if (item != NULL)
619
620
621
{
item->regtype = regtype;
item->regnum = regnum;
622
623
624
item->usage = 0;
item->index = 0;
item->writemask = 0;
625
626
item->next = prev->next;
prev->next = item;
627
} // if
628
629
return item;
630
631
} // reglist_insert
632
633
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
634
{
635
636
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
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.
653
654
655
656
657
658
659
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
660
661
662
663
664
665
666
667
668
669
670
671
} // 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);
672
} // get_used_register
673
674
675
676
677
678
679
680
681
682
683
684
685
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
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
static const RegisterList *declared_attribute(Context *ctx,
const MOJOSHADER_usage usage,
const int index)
{
const RegisterList *item = ctx->attributes.next;
while (item != NULL)
{
if ((item->usage == usage) && (item->index == index))
return item;
item = item->next;
} // while
return NULL;
} // declared_attribute
701
702
703
704
705
706
707
708
709
710
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
711
712
713
714
715
716
717
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
718
719
720
721
722
723
724
725
726
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
727
728
729
730
731
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
732
733
734
// D3D stuff that's used in more than just the d3d profile...
735
736
737
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
738
739
740
741
742
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
743
744
static const char *get_D3D_register_string(Context *ctx,
745
RegisterType regtype,
746
747
int regnum, char *regnum_str,
size_t regnum_size)
748
{
749
const char *retval = NULL;
750
int has_number = 1;
751
752
switch (regtype)
753
{
754
case REG_TYPE_TEMP:
755
retval = "r";
756
757
break;
758
case REG_TYPE_INPUT:
759
retval = "v";
760
761
break;
762
case REG_TYPE_CONST:
763
764
765
retval = "c";
break;
766
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
767
retval = shader_is_vertex(ctx) ? "a" : "t";
768
769
break;
770
case REG_TYPE_RASTOUT:
771
switch ((RastOutType) regnum)
772
{
773
774
775
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
776
} // switch
777
has_number = 0;
778
779
break;
780
case REG_TYPE_ATTROUT:
781
retval = "oD";
782
783
break;
784
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
785
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
786
retval = "o";
787
else
788
retval = "oT";
789
790
break;
791
case REG_TYPE_CONSTINT:
792
retval = "i";
793
794
break;
795
case REG_TYPE_COLOROUT:
796
retval = "oC";
797
798
break;
799
case REG_TYPE_DEPTHOUT:
800
retval = "oDepth";
801
has_number = 0;
802
803
break;
804
case REG_TYPE_SAMPLER:
805
retval = "s";
806
807
break;
808
case REG_TYPE_CONSTBOOL:
809
retval = "b";
810
811
break;
812
case REG_TYPE_LOOP:
813
retval = "aL";
814
has_number = 0;
815
816
break;
817
case REG_TYPE_MISCTYPE:
818
switch ((MiscTypeType) regnum)
819
{
820
821
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
822
} // switch
823
has_number = 0;
824
825
break;
826
case REG_TYPE_LABEL:
827
retval = "l";
828
829
break;
830
case REG_TYPE_PREDICATE:
831
retval = "p";
832
break;
833
834
835
836
837
838
839
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
840
841
} // switch
842
843
844
845
846
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
847
848
849
850
return retval;
} // get_D3D_register_string
851
852
853
854
855
856
857
858
859
860
static inline const char *get_shader_type_string(Context *ctx)
{
if (shader_is_pixel(ctx))
return "ps";
else if (shader_is_vertex(ctx))
return "vs";
fail(ctx, "Unknown shader type.");
return "";
} // get_shader_type_string
861
862
863
864
865
866
867
868
869
870
#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,
871
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
872
873
const SourceArgInfo *arg,
char *buf, size_t buflen)
874
875
876
{
const char *premod_str = "";
const char *postmod_str = "";
877
switch (arg->src_mod)
878
879
880
881
882
883
884
885
886
887
888
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
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
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;
926
927
928
929
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
930
931
932
933
} // switch
char regnum_str[16];
934
935
936
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
937
938
939
940
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
941
return "";
942
943
} // if
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
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
968
969
char swizzle_str[6];
int i = 0;
970
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
971
972
{
swizzle_str[i++] = '.';
973
974
975
976
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];
977
978
979
980
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
981
982
983
984
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
985
986
987
988
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);
989
// !!! FIXME: make sure the scratch buffer was large enough.
990
return buf;
991
} // make_D3D_srcarg_string_in_buf
992
993
994
static const char *make_D3D_destarg_string(Context *ctx)
995
{
996
const DestArgInfo *arg = &ctx->dest_arg;
997
998
999
1000
const char *result_shift_str = "";
switch (arg->result_shift)
{