/
mojoshader.c
4008 lines (3345 loc) · 123 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
146
// 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);
147
148
149
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
150
// one state function for each opcode where we have state machine updates.
151
typedef void (*state_function)(Context *ctx);
152
153
154
155
156
157
158
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
emit_comment comment_emitter;
159
160
emit_global global_emitter;
emit_uniform uniform_emitter;
161
emit_attribute attribute_emitter;
162
emit_finalize finalize_emitter;
163
} Profile;
164
165
166
167
168
169
170
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
171
} RastOutType;
172
173
174
175
176
177
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
178
} MiscTypeType;
179
180
181
182
183
184
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
185
} TextureType;
186
187
188
189
190
// 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.
191
typedef struct OutputListNode
192
193
{
char *str;
194
195
196
197
198
199
200
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
201
202
} OutputList;
203
204
205
206
207
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
208
209
210
MOJOSHADER_usage usage;
int index;
int writemask;
211
212
213
214
struct RegisterList *next;
} RegisterList;
215
216
217
218
219
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
// 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
238
} SourceMod;
239
240
241
242
typedef struct
{
243
const uint32 *token; // this is the unmolested token in the stream.
244
245
int regnum;
int relative;
246
247
248
249
250
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
251
252
int result_mod;
int result_shift;
253
RegisterType regtype;
254
255
256
257
} DestArgInfo;
typedef struct
{
258
const uint32 *token; // this is the unmolested token in the stream.
259
260
int regnum;
int relative;
261
int swizzle; // xyzw (all four, not split out).
262
263
264
265
266
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
int src_mod;
267
RegisterType regtype;
268
269
270
} SourceArgInfo;
271
272
typedef enum
{
273
// Specific to GLSL profile...
274
275
276
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
CTX_FLAGS_GLSL_DST_OPCODE = (1 << 1),
CTX_FLAGS_GLSL_LRP_OPCODE = (1 << 2),
277
278
279
280
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
281
282
#define SCRATCH_BUFFER_SIZE 256
#define SCRATCH_BUFFERS 10
283
284
// !!! FIXME: the scratch buffers make Context pretty big.
285
286
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
287
288
// Context...this is state that changes as we parse through a shader...
289
struct Context
290
{
291
292
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
293
void *malloc_data;
294
295
const uint32 *tokens;
uint32 tokencount;
296
297
OutputList *output;
OutputList globals;
298
OutputList helpers;
299
300
OutputList subroutines;
OutputList mainline;
301
302
303
304
OutputList ignore;
OutputList *output_stack[2];
int indent_stack[2];
int output_stack_len;
305
int output_len; // total strlen; prevents walking the lists just to malloc.
306
int indent;
307
308
309
const char *endline;
int endline_len;
const char *failstr;
310
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
311
312
int scratchidx; // current scratch buffer.
int profileid;
313
const Profile *profile;
314
MOJOSHADER_shaderType shader_type;
315
316
uint8 major_ver;
uint8 minor_ver;
317
DestArgInfo dest_args[1];
318
SourceArgInfo source_args[5];
319
uint32 dwords[4];
320
int instruction_count;
321
uint32 instruction_controls;
322
323
324
uint32 previous_opcode;
ContextFlags flags;
int loops;
325
326
RegisterList used_registers;
RegisterList defined_registers;
327
328
int uniform_count;
RegisterList uniforms;
329
330
int attribute_count;
RegisterList attributes;
331
332
333
};
334
335
336
337
338
339
340
341
342
343
// 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)
{
344
345
346
// check for NULL in case of dumb free() impl.
if (ptr != NULL)
ctx->free(ptr, ctx->malloc_data);
347
348
349
350
} // Free
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
// 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...
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
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
391
392
// Bit arrays (for storing large arrays of booleans...
393
static inline char *get_scratch_buffer(Context *ctx)
394
{
395
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
396
return ctx->scratch[ctx->scratchidx];
397
398
} // get_scratch_buffer
399
400
401
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
402
403
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
404
405
static inline int isfail(const Context *ctx)
406
407
408
409
410
{
return (ctx->failstr != NULL);
} // isfail
411
412
413
414
415
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";
416
417
418
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
419
ctx->failstr = out_of_mem_str; // fail() would call malloc().
420
421
422
423
424
425
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
426
{
427
428
if (ctx->failstr == NULL) // don't change existing error.
{
429
char *scratch = get_scratch_buffer(ctx);
430
431
va_list ap;
va_start(ap, fmt);
432
const int len = vsnprintf(scratch,SCRATCH_BUFFER_SIZE,fmt,ap);
433
434
va_end(ap);
435
char *failstr = (char *) Malloc(ctx, len + 1);
436
437
438
439
440
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
441
if (len < SCRATCH_BUFFER_SIZE)
442
443
444
445
446
447
448
449
450
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
451
452
453
454
455
456
} // if
return FAIL;
} // failf
457
static inline int fail(Context *ctx, const char *reason)
458
459
460
461
462
{
return failf(ctx, "%s", reason);
} // fail
463
464
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
465
{
466
467
OutputListNode *item = NULL;
468
if (isfail(ctx))
469
470
return FAIL; // we failed previously, don't go on...
471
char *scratch = get_scratch_buffer(ctx);
472
473
const int indent = ctx->indent;
474
475
if (indent > 0)
memset(scratch, '\t', indent);
476
477
478
va_list ap;
va_start(ap, fmt);
479
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
480
481
va_end(ap);
482
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
483
484
485
if (item == NULL)
return out_of_memory(ctx);
486
item->str = (char *) Malloc(ctx, len + 1);
487
if (item->str == NULL)
488
{
489
Free(ctx, item);
490
return out_of_memory(ctx);
491
492
} // if
493
494
495
// 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().
496
if (len < SCRATCH_BUFFER_SIZE)
497
498
499
strcpy(item->str, scratch); // copy it over.
else
{
500
501
if (indent > 0)
memset(item->str, '\t', indent);
502
va_start(ap, fmt);
503
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
504
505
va_end(ap);
} // else
506
507
item->next = NULL;
508
509
510
ctx->output->tail->next = item;
ctx->output->tail = item;
511
ctx->output_len += len + ctx->endline_len;
512
513
514
515
516
return 0;
} // output_line
517
518
519
520
521
522
523
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
524
// !!! FIXME: this is sort of nasty.
525
526
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
527
528
{
const size_t len = snprintf(buf, bufsize, "%f", f);
529
if ((len+2) >= bufsize)
530
531
532
533
534
535
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
536
537
538
{
if (leavedecimal)
strcat(buf, ".0");
539
return; // done.
540
} // if
541
542
543
544
545
546
547
548
549
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
550
551
if ((leavedecimal) && (end == ptr))
end += 2;
552
553
554
555
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
556
557
558
// Deal with register lists... !!! FIXME: I sort of hate this.
559
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
560
561
562
563
{
while (item != NULL)
{
RegisterList *next = item->next;
564
f(item, d);
565
566
567
568
569
570
571
572
573
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
574
575
576
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
577
578
579
580
581
582
583
{
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)
584
return item; // already set, so we're done.
585
586
587
588
589
590
591
592
593
594
595
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).
596
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
597
598
599
600
601
602
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
603
604
605
item->usage = 0;
item->index = 0;
item->writemask = 0;
606
607
608
item->next = prev->next;
prev->next = item;
} // else
609
610
return item;
611
612
} // reglist_insert
613
614
615
static const RegisterList *reglist_exists(const RegisterList *prev,
const RegisterType regtype,
const int regnum)
616
617
{
const uint32 newval = reg_to_ui32(regtype, regnum);
618
const RegisterList *item = prev->next;
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
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);
647
} // get_used_register
648
649
650
651
652
653
654
655
656
657
658
659
660
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
661
662
663
664
665
666
667
668
669
670
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
671
672
673
674
675
676
677
678
// D3D stuff that's used in more than just the d3d profile...
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
679
680
static const char *get_D3D_register_string(Context *ctx,
681
RegisterType regtype,
682
683
int regnum, char *regnum_str,
size_t regnum_size)
684
{
685
const char *retval = NULL;
686
int has_number = 1;
687
688
switch (regtype)
689
{
690
case REG_TYPE_TEMP:
691
retval = "r";
692
693
break;
694
case REG_TYPE_INPUT:
695
retval = "v";
696
697
break;
698
case REG_TYPE_CONST:
699
700
701
retval = "c";
break;
702
case REG_TYPE_CONST2:
703
704
705
706
retval = "c";
regnum += 2048;
break;
707
case REG_TYPE_CONST3:
708
709
710
711
retval = "c";
regnum += 4096;
break;
712
case REG_TYPE_CONST4:
713
retval = "c";
714
regnum += 6144;
715
716
break;
717
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
718
retval = (ctx->shader_type == MOJOSHADER_TYPE_VERTEX) ? "a" : "t";
719
720
break;
721
case REG_TYPE_RASTOUT:
722
switch ((RastOutType) regnum)
723
{
724
725
726
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
727
} // switch
728
has_number = 0;
729
730
break;
731
case REG_TYPE_ATTROUT:
732
retval = "oD";
733
734
break;
735
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
736
if ((ctx->shader_type==MOJOSHADER_TYPE_VERTEX) && (ctx->major_ver>=3))
737
retval = "o";
738
else
739
retval = "oT";
740
741
break;
742
case REG_TYPE_CONSTINT:
743
retval = "i";
744
745
break;
746
case REG_TYPE_COLOROUT:
747
retval = "oC";
748
749
break;
750
case REG_TYPE_DEPTHOUT:
751
retval = "oDepth";
752
has_number = 0;
753
754
break;
755
case REG_TYPE_SAMPLER:
756
retval = "s";
757
758
break;
759
case REG_TYPE_CONSTBOOL:
760
retval = "b";
761
762
break;
763
case REG_TYPE_LOOP:
764
retval = "aL";
765
has_number = 0;
766
767
768
break;
// !!! FIXME: don't know what the asm string is for this..
769
case REG_TYPE_TEMPFLOAT16:
770
fail(ctx, "don't know the ASM for tempfloat16 register");
771
772
773
retval = "???";
has_number = 0;
break;
774
775
case REG_TYPE_MISCTYPE:
776
switch ((MiscTypeType) regnum)
777
{
778
779
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
780
} // switch
781
has_number = 0;
782
783
break;
784
case REG_TYPE_LABEL:
785
retval = "l";
786
787
break;
788
case REG_TYPE_PREDICATE:
789
retval = "p";
790
791
792
break;
} // switch
793
794
795
796
797
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
798
799
800
801
return retval;
} // get_D3D_register_string
802
803
804
805
806
807
808
809
810
811
#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,
812
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
813
814
815
816
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
817
return "";
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
} // 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];
838
839
840
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
841
842
843
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
844
return "";
845
846
} // if
847
848
849
850
851
852
853
854
855
856
857
858
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));
859
860
// may turn out something like "_x2_sat_pp_centroid r0.xyzw" ...
861
char *retval = get_scratch_buffer(ctx);
862
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s %s%s%s",
863
result_shift_str, sat_str, pp_str, cent_str,
864
regtype_str, regnum_str, writemask_str);
865
// !!! FIXME: make sure the scratch buffer was large enough.
866
return retval;
867
} // make_D3D_destarg_string
868
869
870
static const char *make_D3D_sourcearg_string(Context *ctx, const int idx)
871
{
872
873
874
if (idx >= STATICARRAYLEN(ctx->source_args))
{
fail(ctx, "Too many source args");
875
return "";
876
} // if
877
878
879
880
881
const SourceArgInfo *arg = &ctx->source_args[idx];
const char *premod_str = "";
const char *postmod_str = "";
882
switch ((SourceMod) arg->src_mod)
883
884
885
886
887
888
889
890
891
892
893
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
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;
931
932
933
934
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
935
936
937
938
} // switch
char regnum_str[16];
939
940
941
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
942
943
944
945
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
946
return "";
947
948
949
950
} // if
char swizzle_str[6];
int i = 0;
951
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
952
953
954
{
static const char channel[] = { 'x', 'y', 'z', 'w' };
swizzle_str[i++] = '.';
955
956
957
958
959
960
961
962
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--;
963
964
965
966
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
967
char *retval = get_scratch_buffer(ctx);
968
969
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str, swizzle_str);
970
// !!! FIXME: make sure the scratch buffer was large enough.
971
return retval;
972
} // make_D3D_sourcearg_string
973
974
975
static void emit_D3D_start(Context *ctx)
976
977
978
{
const uint major = (uint) ctx->major_ver;
const uint minor = (uint) ctx->minor_ver;
979
980
981
982
983
984
985
986
987
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);
988
989
if (ctx->shader_type == MOJOSHADER_TYPE_PIXEL)
990
shadertype_str = "ps";
991
else if (ctx->shader_type == MOJOSHADER_TYPE_VERTEX)
992
shadertype_str = "vs";
993
else
994
995
996
{
failf(ctx, "Shader type %u unsupported in this profile.",
(uint) ctx->shader_type);
997
return;
998
} // else
999
1000
output_line(ctx, "%s_%u_%s", shadertype_str, major, minor_str);