/
mojoshader.c
3758 lines (3122 loc) · 114 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
43
44
45
46
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
47
48
49
50
51
52
// 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;
53
typedef int32_t int32;
54
55
56
57
58
59
60
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
61
62
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
63
64
65
66
67
68
69
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
70
71
72
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
// 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
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
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;
118
119
// predeclare.
120
typedef struct Context Context;
121
122
// one emit function for each opcode in each profile.
123
typedef void (*emit_function)(Context *ctx);
124
125
// one emit function for comments in each profile.
126
typedef void (*emit_comment)(Context *ctx, const char *str);
127
128
// one emit function for starting output in each profile.
129
typedef void (*emit_start)(Context *ctx);
130
131
// one emit function for ending output in each profile.
132
typedef void (*emit_end)(Context *ctx);
133
134
135
136
137
138
139
140
141
142
// 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);
143
144
145
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
146
// one state function for each opcode where we have state machine updates.
147
typedef void (*state_function)(Context *ctx);
148
149
150
151
152
153
154
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
emit_comment comment_emitter;
155
156
157
emit_global global_emitter;
emit_uniform uniform_emitter;
emit_finalize finalize_emitter;
158
} Profile;
159
160
161
162
163
164
165
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
166
} RastOutType;
167
168
169
170
171
172
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
173
} MiscTypeType;
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
typedef enum
{
DECLUSAGE_POSITION = 0,
DECLUSAGE_BLENDWEIGHT = 1,
DECLUSAGE_BLENDINDICES = 2,
DECLUSAGE_NORMAL = 3,
DECLUSAGE_PSIZE = 4,
DECLUSAGE_TEXCOORD = 5,
DECLUSAGE_TANGENT = 6,
DECLUSAGE_BINORMAL = 7,
DECLUSAGE_TESSFACTOR = 8,
DECLUSAGE_POSITIONT = 9,
DECLUSAGE_COLOR = 10,
DECLUSAGE_FOG = 11,
DECLUSAGE_DEPTH = 12,
DECLUSAGE_SAMPLE = 13
191
} DeclUsageType;
192
193
194
195
196
197
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
198
} TextureType;
199
200
201
202
203
// 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.
204
typedef struct OutputListNode
205
206
{
char *str;
207
208
209
210
211
212
213
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
214
215
} OutputList;
216
217
218
219
220
221
222
223
224
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
struct RegisterList *next;
} RegisterList;
225
226
227
228
229
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
// 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
248
} SourceMod;
249
250
251
252
typedef struct
{
253
const uint32 *token; // this is the unmolested token in the stream.
254
255
int regnum;
int relative;
256
257
258
259
260
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
261
262
int result_mod;
int result_shift;
263
RegisterType regtype;
264
265
266
267
} DestArgInfo;
typedef struct
{
268
const uint32 *token; // this is the unmolested token in the stream.
269
270
int regnum;
int relative;
271
int swizzle; // xyzw (all four, not split out).
272
273
274
275
276
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
int src_mod;
277
RegisterType regtype;
278
279
280
} SourceArgInfo;
281
282
typedef enum
{
283
// Specific to GLSL profile...
284
285
286
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
CTX_FLAGS_GLSL_DST_OPCODE = (1 << 1),
CTX_FLAGS_GLSL_LRP_OPCODE = (1 << 2),
287
288
289
290
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
291
292
#define SCRATCH_BUFFER_SIZE 256
#define SCRATCH_BUFFERS 10
293
294
// !!! FIXME: the scratch buffers make Context pretty big.
295
296
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
297
298
// Context...this is state that changes as we parse through a shader...
299
struct Context
300
{
301
302
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
303
void *malloc_data;
304
305
const uint32 *tokens;
uint32 tokencount;
306
307
OutputList *output;
OutputList globals;
308
OutputList helpers;
309
310
OutputList subroutines;
OutputList mainline;
311
312
313
314
OutputList ignore;
OutputList *output_stack[2];
int indent_stack[2];
int output_stack_len;
315
int output_len; // total strlen; prevents walking the lists just to malloc.
316
int indent;
317
318
319
const char *endline;
int endline_len;
const char *failstr;
320
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
321
322
int scratchidx; // current scratch buffer.
int profileid;
323
const Profile *profile;
324
MOJOSHADER_shaderType shader_type;
325
326
uint8 major_ver;
uint8 minor_ver;
327
DestArgInfo dest_args[1];
328
SourceArgInfo source_args[5];
329
uint32 dwords[4];
330
int instruction_count;
331
uint32 instruction_controls;
332
333
334
uint32 previous_opcode;
ContextFlags flags;
int loops;
335
336
RegisterList used_registers;
RegisterList defined_registers;
337
338
int uniform_count;
RegisterList uniforms;
339
340
341
};
342
343
344
345
346
347
348
349
350
351
// Convenience functions for allocators...
static inline void *Malloc(Context *ctx, int len)
{
return ctx->malloc(len, ctx->malloc_data);
} // Malloc
static inline void Free(Context *ctx, void *ptr)
{
352
353
354
// check for NULL in case of dumb free() impl.
if (ptr != NULL)
ctx->free(ptr, ctx->malloc_data);
355
356
357
358
} // Free
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
// 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...
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
static int shader_version_supported(uint8 maj, uint8 min)
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
static int shader_version_atleast(const Context *ctx, uint8 maj, uint8 min)
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
399
400
// Bit arrays (for storing large arrays of booleans...
401
static inline char *get_scratch_buffer(Context *ctx)
402
{
403
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
404
return ctx->scratch[ctx->scratchidx];
405
406
} // get_scratch_buffer
407
408
409
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
410
411
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
412
413
static inline int isfail(const Context *ctx)
414
415
416
417
418
{
return (ctx->failstr != NULL);
} // isfail
419
420
421
422
423
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";
424
425
426
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
427
ctx->failstr = out_of_mem_str; // fail() would call malloc().
428
429
430
431
432
433
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
434
{
435
436
if (ctx->failstr == NULL) // don't change existing error.
{
437
char *scratch = get_scratch_buffer(ctx);
438
439
va_list ap;
va_start(ap, fmt);
440
const int len = vsnprintf(scratch,SCRATCH_BUFFER_SIZE,fmt,ap);
441
442
va_end(ap);
443
char *failstr = (char *) Malloc(ctx, len + 1);
444
445
446
447
448
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
449
if (len < SCRATCH_BUFFER_SIZE)
450
451
452
453
454
455
456
457
458
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
459
460
461
462
463
464
} // if
return FAIL;
} // failf
465
static inline int fail(Context *ctx, const char *reason)
466
467
468
469
470
{
return failf(ctx, "%s", reason);
} // fail
471
472
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
473
{
474
475
OutputListNode *item = NULL;
476
if (isfail(ctx))
477
478
return FAIL; // we failed previously, don't go on...
479
char *scratch = get_scratch_buffer(ctx);
480
481
const int indent = ctx->indent;
482
483
if (indent > 0)
memset(scratch, '\t', indent);
484
485
486
va_list ap;
va_start(ap, fmt);
487
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
488
489
va_end(ap);
490
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
491
492
493
if (item == NULL)
return out_of_memory(ctx);
494
item->str = (char *) Malloc(ctx, len + 1);
495
if (item->str == NULL)
496
{
497
Free(ctx, item);
498
return out_of_memory(ctx);
499
500
} // if
501
502
503
// 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().
504
if (len < SCRATCH_BUFFER_SIZE)
505
506
507
strcpy(item->str, scratch); // copy it over.
else
{
508
509
if (indent > 0)
memset(item->str, '\t', indent);
510
va_start(ap, fmt);
511
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
512
513
va_end(ap);
} // else
514
515
item->next = NULL;
516
517
518
ctx->output->tail->next = item;
ctx->output->tail = item;
519
ctx->output_len += len + ctx->endline_len;
520
521
522
523
524
return 0;
} // output_line
525
526
527
528
529
530
531
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
532
// !!! FIXME: this is sort of nasty.
533
534
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
535
536
{
const size_t len = snprintf(buf, bufsize, "%f", f);
537
if ((len+2) >= bufsize)
538
539
540
541
542
543
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
544
545
546
{
if (leavedecimal)
strcat(buf, ".0");
547
return; // done.
548
} // if
549
550
551
552
553
554
555
556
557
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
558
559
if ((leavedecimal) && (end == ptr))
end += 2;
560
561
562
563
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
564
565
566
// Deal with register lists... !!! FIXME: I sort of hate this.
567
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
568
569
570
571
{
while (item != NULL)
{
RegisterList *next = item->next;
572
f(item, d);
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
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
static void reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype, const int regnum)
{
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)
return; // already set, so we're done.
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).
603
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
604
605
606
607
608
609
610
611
612
613
614
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
item->next = prev->next;
prev->next = item;
} // else
} // reglist_insert
615
616
617
static const RegisterList *reglist_exists(const RegisterList *prev,
const RegisterType regtype,
const int regnum)
618
619
{
const uint32 newval = reg_to_ui32(regtype, regnum);
620
const RegisterList *item = prev->next;
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
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.
} // 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);
649
} // get_used_register
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
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
665
// This is used in more than just the d3d profile.
666
static const char *get_D3D_register_string(Context *ctx,
667
RegisterType regtype,
668
669
int regnum, char *regnum_str,
size_t regnum_size)
670
{
671
const char *retval = NULL;
672
int has_number = 1;
673
674
switch (regtype)
675
{
676
case REG_TYPE_TEMP:
677
retval = "r";
678
679
break;
680
case REG_TYPE_INPUT:
681
retval = "v";
682
683
break;
684
case REG_TYPE_CONST:
685
686
687
retval = "c";
break;
688
case REG_TYPE_CONST2:
689
690
691
692
retval = "c";
regnum += 2048;
break;
693
case REG_TYPE_CONST3:
694
695
696
697
retval = "c";
regnum += 4096;
break;
698
case REG_TYPE_CONST4:
699
retval = "c";
700
regnum += 6144;
701
702
break;
703
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
704
retval = (ctx->shader_type == MOJOSHADER_TYPE_VERTEX) ? "a" : "t";
705
706
break;
707
case REG_TYPE_RASTOUT:
708
switch ((RastOutType) regnum)
709
{
710
711
712
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
713
} // switch
714
has_number = 0;
715
716
break;
717
case REG_TYPE_ATTROUT:
718
retval = "oD";
719
720
break;
721
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
722
if ((ctx->shader_type==MOJOSHADER_TYPE_VERTEX) && (ctx->major_ver>=3))
723
retval = "o";
724
else
725
retval = "oT";
726
727
break;
728
case REG_TYPE_CONSTINT:
729
retval = "i";
730
731
break;
732
case REG_TYPE_COLOROUT:
733
retval = "oC";
734
735
break;
736
case REG_TYPE_DEPTHOUT:
737
retval = "oDepth";
738
has_number = 0;
739
740
break;
741
case REG_TYPE_SAMPLER:
742
retval = "s";
743
744
break;
745
case REG_TYPE_CONSTBOOL:
746
retval = "b";
747
748
break;
749
case REG_TYPE_LOOP:
750
retval = "aL";
751
has_number = 0;
752
753
754
break;
// !!! FIXME: don't know what the asm string is for this..
755
case REG_TYPE_TEMPFLOAT16:
756
fail(ctx, "don't know the ASM for tempfloat16 register");
757
758
759
retval = "???";
has_number = 0;
break;
760
761
case REG_TYPE_MISCTYPE:
762
switch ((MiscTypeType) regnum)
763
{
764
765
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
766
} // switch
767
has_number = 0;
768
769
break;
770
case REG_TYPE_LABEL:
771
retval = "l";
772
773
break;
774
case REG_TYPE_PREDICATE:
775
retval = "p";
776
777
778
break;
} // switch
779
780
781
782
783
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
784
785
786
787
return retval;
} // get_D3D_register_string
788
789
790
791
792
793
794
795
796
797
#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,
798
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
799
800
801
802
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
803
return "";
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
} // 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
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];
824
825
826
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
827
828
829
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
830
return "";
831
832
} // if
833
834
835
836
837
838
839
840
841
842
843
844
char writemask_str[6];
int i = 0;
if (arg->writemask != 0xF) // 0xF == 1111. No explicit mask.
{
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
writemask_str[i] = '\0';
assert(i < sizeof (writemask_str));
845
846
// may turn out something like "_x2_sat_pp_centroid r0.xyzw" ...
847
char *retval = get_scratch_buffer(ctx);
848
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s %s%s%s",
849
result_shift_str, sat_str, pp_str, cent_str,
850
regtype_str, regnum_str, writemask_str);
851
// !!! FIXME: make sure the scratch buffer was large enough.
852
return retval;
853
} // make_D3D_destarg_string
854
855
856
static const char *make_D3D_sourcearg_string(Context *ctx, const int idx)
857
{
858
859
860
if (idx >= STATICARRAYLEN(ctx->source_args))
{
fail(ctx, "Too many source args");
861
return "";
862
} // if
863
864
865
866
867
const SourceArgInfo *arg = &ctx->source_args[idx];
const char *premod_str = "";
const char *postmod_str = "";
868
switch ((SourceMod) arg->src_mod)
869
870
871
872
873
874
875
876
877
878
879
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
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
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;
917
918
919
920
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
921
922
923
924
} // switch
char regnum_str[16];
925
926
927
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
928
929
930
931
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
932
return "";
933
934
935
936
} // if
char swizzle_str[6];
int i = 0;
937
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
938
939
940
{
static const char channel[] = { 'x', 'y', 'z', 'w' };
swizzle_str[i++] = '.';
941
942
943
944
945
946
947
948
swizzle_str[i++] = channel[arg->swizzle_x];
swizzle_str[i++] = channel[arg->swizzle_y];
swizzle_str[i++] = channel[arg->swizzle_z];
swizzle_str[i++] = channel[arg->swizzle_w];
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
949
950
951
952
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
953
char *retval = get_scratch_buffer(ctx);
954
955
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str, swizzle_str);
956
// !!! FIXME: make sure the scratch buffer was large enough.
957
return retval;
958
} // make_D3D_sourcearg_string
959
960
961
static void emit_D3D_start(Context *ctx)
962
963
964
{
const uint major = (uint) ctx->major_ver;
const uint minor = (uint) ctx->minor_ver;
965
966
967
968
969
970
971
972
973
const char *shadertype_str = NULL;
char minor_str[16];
if (minor == 0xFF)
strcpy(minor_str, "sw");
else if (minor == 0x1) // apparently this is "vs_2_x". Weird.
strcpy(minor_str, "x");
else
snprintf(minor_str, sizeof (minor_str), "%u", (uint) minor);
974
975
if (ctx->shader_type == MOJOSHADER_TYPE_PIXEL)
976
shadertype_str = "ps";
977
else if (ctx->shader_type == MOJOSHADER_TYPE_VERTEX)
978
shadertype_str = "vs";
979
else
980
981
982
{
failf(ctx, "Shader type %u unsupported in this profile.",
(uint) ctx->shader_type);
983
return;
984
} // else
985
986
output_line(ctx, "%s_%u_%s", shadertype_str, major, minor_str);
987
988
} // emit_D3D_start
989
990
static void emit_D3D_end(Context *ctx)
991
{
992
output_line(ctx, "end");
993
} // emit_D3D_end
994
995
996
997
998
999
1000
static void emit_D3D_finalize(Context *ctx)
{
// no-op.
} // emit_D3D_finalize