/
mojoshader.c
3928 lines (3328 loc) · 131 KB
1
/**
2
3
* MojoShader; generate shader programs from bytecode of compiled
* Direct3D shaders.
4
5
6
7
8
9
*
* Please see the file LICENSE.txt in the source's root directory.
*
* This file written by Ryan C. Gordon.
*/
10
// !!! FIXME: this file really needs to be split up.
11
12
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
13
14
15
16
17
// !!! FIXME: rules from MSDN about temp registers we probably don't check.
// - There are limited temporaries: vs_1_1 has 12 (ps_1_1 has _2_!).
// - SM2 apparently was variable, between 12 and 32. Shader Model 3 has 32.
// - A maximum of three temp registers can be used in a single instruction.
18
#define __MOJOSHADER_INTERNAL__ 1
19
#include "profiles/mojoshader_profile.h"
20
21
22
// Deal with register lists... !!! FIXME: I sort of hate this.
23
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
24
25
26
27
{
while (item != NULL)
{
RegisterList *next = item->next;
28
f(item, d);
29
30
31
32
item = next;
} // while
} // free_reglist
33
34
35
36
37
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
38
39
} // reglist_exists
40
41
42
43
44
45
46
static inline int register_was_written(Context *ctx, const RegisterType rtype,
const int regnum)
{
RegisterList *reg = reglist_find(&ctx->used_registers, rtype, regnum);
return (reg && reg->written);
} // register_was_written
47
48
49
50
51
52
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
53
54
static void add_attribute_register(Context *ctx, const RegisterType rtype,
const int regnum, const MOJOSHADER_usage usage,
55
const int index, const int writemask, int flags)
56
57
58
59
60
{
RegisterList *item = reglist_insert(ctx, &ctx->attributes, rtype, regnum);
item->usage = usage;
item->index = index;
item->writemask = writemask;
61
item->misc = flags;
62
63
64
if ((rtype == REG_TYPE_OUTPUT) && (usage == MOJOSHADER_USAGE_POINTSIZE))
ctx->uses_pointsize = 1; // note that we have to check this later.
65
66
else if ((rtype == REG_TYPE_OUTPUT) && (usage == MOJOSHADER_USAGE_FOG))
ctx->uses_fog = 1; // note that we have to check this later.
67
68
} // add_attribute_register
69
70
71
72
73
74
75
76
77
78
static inline TextureType cvtMojoToD3DSamplerType(const MOJOSHADER_samplerType type)
{
return (TextureType) (((int) type) + 2);
} // cvtMojoToD3DSamplerType
static inline MOJOSHADER_samplerType cvtD3DToMojoSamplerType(const TextureType type)
{
return (MOJOSHADER_samplerType) (((int) type) - 2);
} // cvtD3DToMojoSamplerType
79
80
static inline void add_sampler(Context *ctx, const int regnum,
TextureType ttype, const int texbem)
81
{
82
83
const RegisterType rtype = REG_TYPE_SAMPLER;
84
// !!! FIXME: make sure it doesn't exist?
85
// !!! FIXME: (ps_1_1 assume we can add it multiple times...)
86
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
87
88
89
90
91
92
93
94
95
96
97
98
99
100
if (ctx->samplermap != NULL)
{
unsigned int i;
for (i = 0; i < ctx->samplermap_count; i++)
{
if (ctx->samplermap[i].index == regnum)
{
ttype = cvtMojoToD3DSamplerType(ctx->samplermap[i].type);
break;
} // if
} // for
} // if
101
item->index = (int) ttype;
102
item->misc |= texbem;
103
104
} // add_sampler
105
106
107
108
109
110
111
static inline void adjust_token_position(Context *ctx, const int incr)
{
ctx->tokens += incr;
ctx->tokencount -= incr;
ctx->current_position += incr * sizeof (uint32);
} // adjust_token_position
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
// Generate emitter declarations for each profile with this macro...
#define PREDECLARE_PROFILE(prof) \
void emit_##prof##_start(Context *ctx, const char *profilestr); \
void emit_##prof##_end(Context *ctx); \
void emit_##prof##_phase(Context *ctx); \
void emit_##prof##_finalize(Context *ctx); \
void emit_##prof##_global(Context *ctx, RegisterType regtype, int regnum);\
void emit_##prof##_array(Context *ctx, VariableList *var); \
void emit_##prof##_const_array(Context *ctx, const ConstantsList *clist, \
int base, int size); \
void emit_##prof##_uniform(Context *ctx, RegisterType regtype, int regnum,\
const VariableList *var); \
void emit_##prof##_sampler(Context *ctx, int stage, TextureType ttype, \
int tb); \
void emit_##prof##_attribute(Context *ctx, RegisterType regtype, \
int regnum, MOJOSHADER_usage usage, \
int index, int wmask, int flags); \
void emit_##prof##_NOP(Context *ctx); \
void emit_##prof##_MOV(Context *ctx); \
void emit_##prof##_ADD(Context *ctx); \
void emit_##prof##_SUB(Context *ctx); \
void emit_##prof##_MAD(Context *ctx); \
void emit_##prof##_MUL(Context *ctx); \
void emit_##prof##_RCP(Context *ctx); \
void emit_##prof##_RSQ(Context *ctx); \
void emit_##prof##_DP3(Context *ctx); \
void emit_##prof##_DP4(Context *ctx); \
void emit_##prof##_MIN(Context *ctx); \
void emit_##prof##_MAX(Context *ctx); \
void emit_##prof##_SLT(Context *ctx); \
void emit_##prof##_SGE(Context *ctx); \
void emit_##prof##_EXP(Context *ctx); \
void emit_##prof##_LOG(Context *ctx); \
void emit_##prof##_LIT(Context *ctx); \
void emit_##prof##_DST(Context *ctx); \
void emit_##prof##_LRP(Context *ctx); \
void emit_##prof##_FRC(Context *ctx); \
void emit_##prof##_M4X4(Context *ctx); \
void emit_##prof##_M4X3(Context *ctx); \
void emit_##prof##_M3X4(Context *ctx); \
void emit_##prof##_M3X3(Context *ctx); \
void emit_##prof##_M3X2(Context *ctx); \
void emit_##prof##_CALL(Context *ctx); \
void emit_##prof##_CALLNZ(Context *ctx); \
void emit_##prof##_LOOP(Context *ctx); \
void emit_##prof##_ENDLOOP(Context *ctx); \
void emit_##prof##_LABEL(Context *ctx); \
void emit_##prof##_DCL(Context *ctx); \
void emit_##prof##_POW(Context *ctx); \
void emit_##prof##_CRS(Context *ctx); \
void emit_##prof##_SGN(Context *ctx); \
void emit_##prof##_ABS(Context *ctx); \
void emit_##prof##_NRM(Context *ctx); \
void emit_##prof##_SINCOS(Context *ctx); \
void emit_##prof##_REP(Context *ctx); \
void emit_##prof##_ENDREP(Context *ctx); \
void emit_##prof##_IF(Context *ctx); \
void emit_##prof##_IFC(Context *ctx); \
void emit_##prof##_ELSE(Context *ctx); \
void emit_##prof##_ENDIF(Context *ctx); \
void emit_##prof##_BREAK(Context *ctx); \
void emit_##prof##_BREAKC(Context *ctx); \
void emit_##prof##_MOVA(Context *ctx); \
void emit_##prof##_DEFB(Context *ctx); \
void emit_##prof##_DEFI(Context *ctx); \
void emit_##prof##_TEXCRD(Context *ctx); \
void emit_##prof##_TEXKILL(Context *ctx); \
void emit_##prof##_TEXLD(Context *ctx); \
void emit_##prof##_TEXBEM(Context *ctx); \
void emit_##prof##_TEXBEML(Context *ctx); \
void emit_##prof##_TEXREG2AR(Context *ctx); \
void emit_##prof##_TEXREG2GB(Context *ctx); \
void emit_##prof##_TEXM3X2PAD(Context *ctx); \
void emit_##prof##_TEXM3X2TEX(Context *ctx); \
void emit_##prof##_TEXM3X3PAD(Context *ctx); \
void emit_##prof##_TEXM3X3TEX(Context *ctx); \
void emit_##prof##_TEXM3X3SPEC(Context *ctx); \
void emit_##prof##_TEXM3X3VSPEC(Context *ctx); \
void emit_##prof##_EXPP(Context *ctx); \
void emit_##prof##_LOGP(Context *ctx); \
void emit_##prof##_CND(Context *ctx); \
void emit_##prof##_DEF(Context *ctx); \
void emit_##prof##_TEXREG2RGB(Context *ctx); \
void emit_##prof##_TEXDP3TEX(Context *ctx); \
void emit_##prof##_TEXM3X2DEPTH(Context *ctx); \
void emit_##prof##_TEXDP3(Context *ctx); \
void emit_##prof##_TEXM3X3(Context *ctx); \
void emit_##prof##_TEXDEPTH(Context *ctx); \
void emit_##prof##_CMP(Context *ctx); \
void emit_##prof##_BEM(Context *ctx); \
void emit_##prof##_DP2ADD(Context *ctx); \
void emit_##prof##_DSX(Context *ctx); \
void emit_##prof##_DSY(Context *ctx); \
void emit_##prof##_TEXLDD(Context *ctx); \
void emit_##prof##_SETP(Context *ctx); \
void emit_##prof##_TEXLDL(Context *ctx); \
void emit_##prof##_BREAKP(Context *ctx); \
void emit_##prof##_RESERVED(Context *ctx); \
void emit_##prof##_RET(Context *ctx); \
const char *get_##prof##_varname(Context *ctx, RegisterType rt, \
int regnum); \
const char *get_##prof##_const_array_varname(Context *ctx, \
int base, int size);
// Check for profile support...
218
219
220
#define AT_LEAST_ONE_PROFILE 0
221
222
223
224
225
226
227
228
229
#if !SUPPORT_PROFILE_BYTECODE
#define PROFILE_EMITTER_BYTECODE(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_BYTECODE(op) emit_BYTECODE_##op,
PREDECLARE_PROFILE(BYTECODE)
#endif
230
231
232
233
234
235
#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,
236
237
PREDECLARE_PROFILE(D3D)
#endif
238
239
240
241
242
243
244
245
246
#if !SUPPORT_PROFILE_GLSL
#define PROFILE_EMITTER_GLSL(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_GLSL(op) emit_GLSL_##op,
PREDECLARE_PROFILE(GLSL)
#endif
247
248
249
250
251
252
253
254
255
#if !SUPPORT_PROFILE_METAL
#define PROFILE_EMITTER_METAL(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_METAL(op) emit_METAL_##op,
PREDECLARE_PROFILE(METAL)
#endif
256
257
258
259
260
261
262
263
264
#if !SUPPORT_PROFILE_ARB1
#define PROFILE_EMITTER_ARB1(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_ARB1(op) emit_ARB1_##op,
PREDECLARE_PROFILE(ARB1)
#endif
265
266
267
268
269
270
271
272
273
274
#if !SUPPORT_PROFILE_SPIRV
#define PROFILE_EMITTER_SPIRV(op)
#else
#undef AT_LEAST_ONE_PROFILE
#define AT_LEAST_ONE_PROFILE 1
#define PROFILE_EMITTER_SPIRV(op) emit_SPIRV_##op,
PREDECLARE_PROFILE(SPIRV)
#endif
275
276
277
278
#if !AT_LEAST_ONE_PROFILE
#error No profiles are supported. Fix your build.
#endif
279
280
281
282
#define DEFINE_PROFILE(prof) { \
MOJOSHADER_PROFILE_##prof, \
emit_##prof##_start, \
emit_##prof##_end, \
283
emit_##prof##_phase, \
284
emit_##prof##_global, \
285
emit_##prof##_array, \
286
emit_##prof##_const_array, \
287
emit_##prof##_uniform, \
288
emit_##prof##_sampler, \
289
emit_##prof##_attribute, \
290
emit_##prof##_finalize, \
291
292
get_##prof##_varname, \
get_##prof##_const_array_varname, \
293
294
},
295
static const Profile profiles[] =
296
297
{
#if SUPPORT_PROFILE_D3D
298
DEFINE_PROFILE(D3D)
299
#endif
300
301
#if SUPPORT_PROFILE_BYTECODE
DEFINE_PROFILE(BYTECODE)
302
#endif
303
#if SUPPORT_PROFILE_GLSL
304
DEFINE_PROFILE(GLSL)
305
#endif
306
307
308
#if SUPPORT_PROFILE_ARB1
DEFINE_PROFILE(ARB1)
#endif
309
310
311
#if SUPPORT_PROFILE_METAL
DEFINE_PROFILE(METAL)
#endif
312
313
314
#if SUPPORT_PROFILE_SPIRV
DEFINE_PROFILE(SPIRV)
#endif
315
316
};
317
318
#undef DEFINE_PROFILE
319
320
321
// This is for profiles that extend other profiles...
static const struct { const char *from; const char *to; } profileMap[] =
{
322
{ MOJOSHADER_PROFILE_GLSPIRV, MOJOSHADER_PROFILE_SPIRV },
323
{ MOJOSHADER_PROFILE_GLSLES, MOJOSHADER_PROFILE_GLSL },
324
{ MOJOSHADER_PROFILE_GLSL120, MOJOSHADER_PROFILE_GLSL },
325
{ MOJOSHADER_PROFILE_NV2, MOJOSHADER_PROFILE_ARB1 },
326
{ MOJOSHADER_PROFILE_NV3, MOJOSHADER_PROFILE_ARB1 },
327
{ MOJOSHADER_PROFILE_NV4, MOJOSHADER_PROFILE_ARB1 },
328
329
};
330
331
332
// The PROFILE_EMITTER_* items MUST be in the same order as profiles[]!
#define PROFILE_EMITTERS(op) { \
PROFILE_EMITTER_D3D(op) \
333
PROFILE_EMITTER_BYTECODE(op) \
334
PROFILE_EMITTER_GLSL(op) \
335
PROFILE_EMITTER_ARB1(op) \
336
PROFILE_EMITTER_METAL(op) \
337
PROFILE_EMITTER_SPIRV(op) \
338
}
339
340
341
static int parse_destination_token(Context *ctx, DestArgInfo *info)
{
342
// !!! FIXME: recheck against the spec for ranges (like RASTOUT values, etc).
343
if (ctx->tokencount == 0)
344
345
346
347
{
fail(ctx, "Out of tokens in destination parameter");
return 0;
} // if
348
349
350
351
352
const uint32 token = SWAP32(*(ctx->tokens));
const int reserved1 = (int) ((token >> 14) & 0x3); // bits 14 through 15
const int reserved2 = (int) ((token >> 31) & 0x1); // bit 31
353
info->token = ctx->tokens;
354
355
info->regnum = (int) (token & 0x7ff); // bits 0 through 10
info->relative = (int) ((token >> 13) & 0x1); // bit 13
356
info->orig_writemask = (int) ((token >> 16) & 0xF); // bits 16 through 19
357
info->result_mod = (int) ((token >> 20) & 0xF); // bits 20 through 23
358
359
info->result_shift = (int) ((token >> 24) & 0xF); // bits 24 through 27 abc
info->regtype = (RegisterType) (((token >> 28) & 0x7) | ((token >> 8) & 0x18)); // bits 28-30, 11-12
360
361
int writemask;
362
if (isscalar(ctx, ctx->shader_type, info->regtype, info->regnum))
363
364
365
366
writemask = 0x1; // just x.
else
writemask = info->orig_writemask;
367
set_dstarg_writemask(info, writemask); // bits 16 through 19.
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
// all the REG_TYPE_CONSTx types are the same register type, it's just
// split up so its regnum can be > 2047 in the bytecode. Clean it up.
if (info->regtype == REG_TYPE_CONST2)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 2048;
} // else if
else if (info->regtype == REG_TYPE_CONST3)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 4096;
} // else if
else if (info->regtype == REG_TYPE_CONST4)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 6144;
} // else if
387
388
// swallow token for now, for multiple calls in a row.
adjust_token_position(ctx, 1);
389
390
if (reserved1 != 0x0)
391
fail(ctx, "Reserved bit #1 in destination token must be zero");
392
393
if (reserved2 != 0x1)
394
fail(ctx, "Reserved bit #2 in destination token must be one");
395
396
397
if (info->relative)
{
398
if (!shader_is_vertex(ctx))
399
400
401
fail(ctx, "Relative addressing in non-vertex shader");
if (!shader_version_atleast(ctx, 3, 0))
fail(ctx, "Relative addressing in vertex shader version < 3.0");
402
403
404
if ((!ctx->ctab.have_ctab) && (!ctx->ignores_ctab))
{
// it's hard to do this efficiently without!
405
fail(ctx, "relative addressing unsupported without a CTAB");
406
} // if
407
408
// !!! FIXME: I don't have a shader that has a relative dest currently.
409
410
fail(ctx, "Relative addressing of dest tokens is unsupported");
return 2;
411
412
} // if
413
414
const int s = info->result_shift;
if (s != 0)
415
{
416
if (!shader_is_pixel(ctx))
417
418
419
420
421
fail(ctx, "Result shift scale in non-pixel shader");
if (shader_version_atleast(ctx, 2, 0))
fail(ctx, "Result shift scale in pixel shader version >= 2.0");
if ( ! (((s >= 1) && (s <= 3)) || ((s >= 0xD) && (s <= 0xF))) )
fail(ctx, "Result shift scale isn't 1 to 3, or 13 to 15.");
422
423
} // if
424
if (info->result_mod & MOD_PP) // Partial precision (pixel shaders only)
425
{
426
if (!shader_is_pixel(ctx))
427
fail(ctx, "Partial precision result mod in non-pixel shader");
428
429
} // if
430
if (info->result_mod & MOD_CENTROID) // Centroid (pixel shaders only)
431
{
432
if (!shader_is_pixel(ctx))
433
fail(ctx, "Centroid result mod in non-pixel shader");
434
else if (!ctx->centroid_allowed) // only on DCL opcodes!
435
fail(ctx, "Centroid modifier not allowed here");
436
437
} // if
438
if (/*(info->regtype < 0) ||*/ (info->regtype > REG_TYPE_MAX))
439
fail(ctx, "Register type is out of range");
440
441
if (!isfail(ctx))
442
set_used_register(ctx, info->regtype, info->regnum, 1);
443
444
445
446
447
return 1;
} // parse_destination_token
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
static void determine_constants_arrays(Context *ctx)
{
// Only process this stuff once. This is called after all DEF* opcodes
// could have been parsed.
if (ctx->determined_constants_arrays)
return;
ctx->determined_constants_arrays = 1;
if (ctx->constant_count <= 1)
return; // nothing to sort or group.
// Sort the linked list into an array for easier tapdancing...
ConstantsList **array = (ConstantsList **) alloca(sizeof (ConstantsList *) * (ctx->constant_count + 1));
ConstantsList *item = ctx->constants;
int i;
for (i = 0; i < ctx->constant_count; i++)
{
if (item == NULL)
{
fail(ctx, "BUG: mismatched constant list and count");
return;
} // if
array[i] = item;
item = item->next;
} // for
array[ctx->constant_count] = NULL;
// bubble sort ftw.
int sorted;
do
{
sorted = 1;
for (i = 0; i < ctx->constant_count-1; i++)
{
if (array[i]->constant.index > array[i+1]->constant.index)
{
ConstantsList *tmp = array[i];
array[i] = array[i+1];
array[i+1] = tmp;
sorted = 0;
} // if
} // for
} while (!sorted);
// okay, sorted. While we're here, let's redo the linked list in order...
for (i = 0; i < ctx->constant_count; i++)
array[i]->next = array[i+1];
ctx->constants = array[0];
// now figure out the groupings of constants and add to ctx->variables...
int start = -1;
int prev = -1;
int count = 0;
const int hi = ctx->constant_count;
for (i = 0; i <= hi; i++)
{
if (array[i] && (array[i]->constant.type != MOJOSHADER_UNIFORM_FLOAT))
continue; // we only care about REG_TYPE_CONST for array groups.
if (start == -1)
{
prev = start = i; // first REG_TYPE_CONST we've seen. Mark it!
continue;
} // if
// not a match (or last item in the array)...see if we had a
// contiguous set before this point...
if ( (array[i]) && (array[i]->constant.index == (array[prev]->constant.index + 1)) )
count++;
else
{
if (count > 0) // multiple constants in the set?
{
VariableList *var;
var = (VariableList *) Malloc(ctx, sizeof (VariableList));
if (var == NULL)
break;
var->type = MOJOSHADER_UNIFORM_FLOAT;
var->index = array[start]->constant.index;
var->count = (array[prev]->constant.index - var->index) + 1;
var->constant = array[start];
var->used = 0;
535
var->emit_position = -1;
536
537
var->next = ctx->variables;
ctx->variables = var;
538
} // if
539
540
541
542
543
544
545
546
547
start = i; // set this as new start of sequence.
} // if
prev = i;
} // for
} // determine_constants_arrays
548
549
550
551
552
553
554
555
556
557
558
559
static int adjust_swizzle(const Context *ctx, const RegisterType regtype,
const int regnum, const int swizzle)
{
if (regtype != REG_TYPE_INPUT) // !!! FIXME: maybe lift this later?
return swizzle;
else if (ctx->swizzles_count == 0)
return swizzle;
const RegisterList *reg = reglist_find(&ctx->attributes, regtype, regnum);
if (reg == NULL)
return swizzle;
560
561
size_t i;
for (i = 0; i < ctx->swizzles_count; i++)
562
{
563
const MOJOSHADER_swizzle *swiz = &ctx->swizzles[i];
564
565
566
567
568
569
570
571
572
573
574
575
576
if ((swiz->usage == reg->usage) && (swiz->index == reg->index))
{
return ( (((int)(swiz->swizzles[((swizzle >> 0) & 0x3)])) << 0) |
(((int)(swiz->swizzles[((swizzle >> 2) & 0x3)])) << 2) |
(((int)(swiz->swizzles[((swizzle >> 4) & 0x3)])) << 4) |
(((int)(swiz->swizzles[((swizzle >> 6) & 0x3)])) << 6) );
} // if
} // for
return swizzle;
} // adjust_swizzle
577
578
static int parse_source_token(Context *ctx, SourceArgInfo *info)
{
579
580
int retval = 1;
581
if (ctx->tokencount == 0)
582
583
584
585
{
fail(ctx, "Out of tokens in source parameter");
return 0;
} // if
586
587
588
589
590
const uint32 token = SWAP32(*(ctx->tokens));
const int reserved1 = (int) ((token >> 14) & 0x3); // bits 14 through 15
const int reserved2 = (int) ((token >> 31) & 0x1); // bit 31
591
info->token = ctx->tokens;
592
593
info->regnum = (int) (token & 0x7ff); // bits 0 through 10
info->relative = (int) ((token >> 13) & 0x1); // bit 13
594
const int swizzle = (int) ((token >> 16) & 0xFF); // bits 16 through 23
595
info->src_mod = (SourceMod) ((token >> 24) & 0xF); // bits 24 through 27
596
info->regtype = (RegisterType) (((token >> 28) & 0x7) | ((token >> 8) & 0x18)); // bits 28-30, 11-12
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
// all the REG_TYPE_CONSTx types are the same register type, it's just
// split up so its regnum can be > 2047 in the bytecode. Clean it up.
if (info->regtype == REG_TYPE_CONST2)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 2048;
} // else if
else if (info->regtype == REG_TYPE_CONST3)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 4096;
} // else if
else if (info->regtype == REG_TYPE_CONST4)
{
info->regtype = REG_TYPE_CONST;
info->regnum += 6144;
} // else if
616
617
618
619
620
621
info->swizzle = adjust_swizzle(ctx, info->regtype, info->regnum, swizzle);
info->swizzle_x = ((info->swizzle >> 0) & 0x3);
info->swizzle_y = ((info->swizzle >> 2) & 0x3);
info->swizzle_z = ((info->swizzle >> 4) & 0x3);
info->swizzle_w = ((info->swizzle >> 6) & 0x3);
622
623
// swallow token for now, for multiple calls in a row.
adjust_token_position(ctx, 1);
624
625
if (reserved1 != 0x0)
626
fail(ctx, "Reserved bits #1 in source token must be zero");
627
628
if (reserved2 != 0x1)
629
630
631
632
633
634
635
fail(ctx, "Reserved bit #2 in source token must be one");
if ((info->relative) && (ctx->tokencount == 0))
{
fail(ctx, "Out of tokens in relative source parameter");
info->relative = 0; // don't try to process it.
} // if
636
637
638
if (info->relative)
{
639
if ( (shader_is_pixel(ctx)) && (!shader_version_atleast(ctx, 3, 0)) )
640
fail(ctx, "Relative addressing in pixel shader version < 3.0");
641
642
643
644
645
646
647
648
649
// Shader Model 1 doesn't have an extra token to specify the
// relative register: it's always a0.x.
if (!shader_version_atleast(ctx, 2, 0))
{
info->relative_regnum = 0;
info->relative_regtype = REG_TYPE_ADDRESS;
info->relative_component = 0;
} // if
650
651
652
653
654
655
else // Shader Model 2 and later...
{
const uint32 reltoken = SWAP32(*(ctx->tokens));
// swallow token for now, for multiple calls in a row.
adjust_token_position(ctx, 1);
656
657
658
659
660
661
const int relswiz = (int) ((reltoken >> 16) & 0xFF);
info->relative_regnum = (int) (reltoken & 0x7ff);
info->relative_regtype = (RegisterType)
(((reltoken >> 28) & 0x7) |
((reltoken >> 8) & 0x18));
662
663
664
if (((reltoken >> 31) & 0x1) == 0)
fail(ctx, "bit #31 in relative address must be set");
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
if ((reltoken & 0xF00E000) != 0) // usused bits.
fail(ctx, "relative address reserved bit must be zero");
switch (info->relative_regtype)
{
case REG_TYPE_LOOP:
case REG_TYPE_ADDRESS:
break;
default:
fail(ctx, "invalid register for relative address");
break;
} // switch
if (info->relative_regnum != 0) // true for now.
680
fail(ctx, "invalid register for relative address");
681
682
if ( (info->relative_regtype != REG_TYPE_LOOP) && !replicate_swizzle(relswiz) )
683
fail(ctx, "relative address needs replicate swizzle");
684
685
info->relative_component = (relswiz & 0x3);
686
687
688
retval++;
} // else
689
690
691
692
if (info->regtype == REG_TYPE_INPUT)
{
if ( (shader_is_pixel(ctx)) || (!shader_version_atleast(ctx, 3, 0)) )
693
fail(ctx, "relative addressing of input registers not supported in this shader model");
694
695
696
697
698
ctx->have_relative_input_registers = 1;
} // if
else if (info->regtype == REG_TYPE_CONST)
{
// figure out what array we're in...
699
if (!ctx->ignores_ctab)
700
{
701
702
703
if (!ctx->ctab.have_ctab) // hard to do efficiently without!
fail(ctx, "relative addressing unsupported without a CTAB");
else
704
{
705
706
707
708
709
710
711
712
713
714
715
716
717
determine_constants_arrays(ctx);
VariableList *var;
const int reltarget = info->regnum;
for (var = ctx->variables; var != NULL; var = var->next)
{
const int lo = var->index;
if ( (reltarget >= lo) && (reltarget < (lo + var->count)) )
break; // match!
} // for
if (var == NULL)
fail(ctx, "relative addressing of indeterminate array");
718
719
720
721
else
{
var->used = 1;
info->relative_array = var;
722
set_used_register(ctx, info->relative_regtype, info->relative_regnum, 0);
723
} // else
724
725
} // else
} // if
726
727
} // else if
else
728
{
729
fail(ctx, "relative addressing of invalid register");
730
} // else
731
732
} // if
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
switch (info->src_mod)
{
case SRCMOD_NONE:
case SRCMOD_ABSNEGATE:
case SRCMOD_ABS:
case SRCMOD_NEGATE:
break; // okay in any shader model.
// apparently these are only legal in Shader Model 1.x ...
case SRCMOD_BIASNEGATE:
case SRCMOD_BIAS:
case SRCMOD_SIGNNEGATE:
case SRCMOD_SIGN:
case SRCMOD_COMPLEMENT:
case SRCMOD_X2NEGATE:
case SRCMOD_X2:
case SRCMOD_DZ:
case SRCMOD_DW:
if (shader_version_atleast(ctx, 2, 0))
752
fail(ctx, "illegal source mod for this Shader Model.");
753
754
break;
755
756
757
case SRCMOD_NOT: // !!! FIXME: I _think_ this is right...
if (shader_version_atleast(ctx, 2, 0))
{
758
759
760
if (info->regtype != REG_TYPE_PREDICATE
&& info->regtype != REG_TYPE_CONSTBOOL)
fail(ctx, "NOT only allowed on bool registers.");
761
762
763
} // if
break;
764
default:
765
fail(ctx, "Unknown source modifier");
766
767
768
769
770
771
772
773
774
775
} // switch
// !!! FIXME: docs say this for sm3 ... check these!
// "The negate modifier cannot be used on second source register of these
// instructions: m3x2 - ps, m3x3 - ps, m3x4 - ps, m4x3 - ps, and
// m4x4 - ps."
// "If any version 3 shader reads from one or more constant float
// registers (c#), one of the following must be true.
// All of the constant floating-point registers must use the abs modifier.
// None of the constant floating-point registers can use the abs modifier.
776
777
if (!isfail(ctx))
778
779
780
781
782
783
784
785
786
787
788
{
RegisterList *reg;
reg = set_used_register(ctx, info->regtype, info->regnum, 0);
// !!! FIXME: this test passes if you write to the register
// !!! FIXME: in this same instruction, because we parse the
// !!! FIXME: destination token first.
// !!! FIXME: Microsoft's shader validation explicitly checks temp
// !!! FIXME: registers for this...do they check other writable ones?
if ((info->regtype == REG_TYPE_TEMP) && (reg) && (!reg->written))
failf(ctx, "Temp register r%d used uninitialized", info->regnum);
} // if
789
790
return retval;
791
792
793
} // parse_source_token
794
795
796
static int parse_predicated_token(Context *ctx)
{
SourceArgInfo *arg = &ctx->predicate_arg;
797
798
799
800
801
802
803
804
805
parse_source_token(ctx, arg);
if (arg->regtype != REG_TYPE_PREDICATE)
fail(ctx, "Predicated instruction but not predicate register!");
if ((arg->src_mod != SRCMOD_NONE) && (arg->src_mod != SRCMOD_NOT))
fail(ctx, "Predicated instruction register is not NONE or NOT");
if ( !no_swizzle(arg->swizzle) && !replicate_swizzle(arg->swizzle) )
fail(ctx, "Predicated instruction register has wrong swizzle");
if (arg->relative) // I'm pretty sure this is illegal...?
fail(ctx, "relative addressing in predicated token");
806
807
return 1;
808
809
810
} // parse_predicated_token
811
812
static int parse_args_NULL(Context *ctx)
{
813
return 1;
814
815
816
} // parse_args_NULL
817
818
static int parse_args_DEF(Context *ctx)
{
819
820
821
822
823
parse_destination_token(ctx, &ctx->dest_arg);
if (ctx->dest_arg.regtype != REG_TYPE_CONST)
fail(ctx, "DEF using non-CONST register");
if (ctx->dest_arg.relative) // I'm pretty sure this is illegal...?
fail(ctx, "relative addressing in DEF");
824
825
826
827
828
829
ctx->dwords[0] = SWAP32(ctx->tokens[0]);
ctx->dwords[1] = SWAP32(ctx->tokens[1]);
ctx->dwords[2] = SWAP32(ctx->tokens[2]);
ctx->dwords[3] = SWAP32(ctx->tokens[3]);
830
return 6;
831
832
833
} // parse_args_DEF
834
static int parse_args_DEFI(Context *ctx)
835
{
836
837
838
839
840
parse_destination_token(ctx, &ctx->dest_arg);
if (ctx->dest_arg.regtype != REG_TYPE_CONSTINT)
fail(ctx, "DEFI using non-CONSTING register");
if (ctx->dest_arg.relative) // I'm pretty sure this is illegal...?
fail(ctx, "relative addressing in DEFI");
841
842
843
844
845
ctx->dwords[0] = SWAP32(ctx->tokens[0]);
ctx->dwords[1] = SWAP32(ctx->tokens[1]);
ctx->dwords[2] = SWAP32(ctx->tokens[2]);
ctx->dwords[3] = SWAP32(ctx->tokens[3]);
846
847
848
849
850
851
852
return 6;
} // parse_args_DEFI
static int parse_args_DEFB(Context *ctx)
{
853
854
855
856
857
parse_destination_token(ctx, &ctx->dest_arg);
if (ctx->dest_arg.regtype != REG_TYPE_CONSTBOOL)
fail(ctx, "DEFB using non-CONSTBOOL register");
if (ctx->dest_arg.relative) // I'm pretty sure this is illegal...?
fail(ctx, "relative addressing in DEFB");
858
859
860
ctx->dwords[0] = *(ctx->tokens) ? 1 : 0;
861
return 3;
862
} // parse_args_DEFB
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
static int valid_texture_type(const uint32 ttype)
{
switch ((const TextureType) ttype)
{
case TEXTURE_TYPE_2D:
case TEXTURE_TYPE_CUBE:
case TEXTURE_TYPE_VOLUME:
return 1; // it's okay.
} // switch
return 0;
} // valid_texture_type
879
// !!! FIXME: this function is kind of a mess.
880
881
882
883
884
885
886
887
static int parse_args_DCL(Context *ctx)
{
int unsupported = 0;
const uint32 token = SWAP32(*(ctx->tokens));
const int reserved1 = (int) ((token >> 31) & 0x1); // bit 31
uint32 reserved_mask = 0x00000000;
if (reserved1 != 0x1)
888
fail(ctx, "Bit #31 in DCL token must be one");
889
890
ctx->centroid_allowed = 1;
891
adjust_token_position(ctx, 1);
892
parse_destination_token(ctx, &ctx->dest_arg);
893
ctx->centroid_allowed = 0;
894
895
if (ctx->dest_arg.result_shift != 0) // I'm pretty sure this is illegal...?
896
897
898
fail(ctx, "shift scale in DCL");
if (ctx->dest_arg.relative) // I'm pretty sure this is illegal...?
fail(ctx, "relative addressing in DCL");
899
900
901
const RegisterType regtype = ctx->dest_arg.regtype;
const int regnum = ctx->dest_arg.regnum;
902
if ( (shader_is_pixel(ctx)) && (shader_version_atleast(ctx, 3, 0)) )
903
{
904
if (regtype == REG_TYPE_INPUT)
905
906
907
908
909
910
911
{
const uint32 usage = (token & 0xF);
const uint32 index = ((token >> 16) & 0xF);
reserved_mask = 0x7FF0FFE0;
ctx->dwords[0] = usage;
ctx->dwords[1] = index;
} // if
912
913
else if (regtype == REG_TYPE_MISCTYPE)
914
{
915
const MiscTypeType mt = (MiscTypeType) regnum;
916
917
918
919
920
if (mt == MISCTYPE_TYPE_POSITION)
reserved_mask = 0x7FFFFFFF;
else if (mt == MISCTYPE_TYPE_FACE)
{
reserved_mask = 0x7FFFFFFF;
921
if (!writemask_xyzw(ctx->dest_arg.orig_writemask))
922
923
924
925
926
fail(ctx, "DCL face writemask must be full");
if (ctx->dest_arg.result_mod != 0)
fail(ctx, "DCL face result modifier must be zero");
if (ctx->dest_arg.result_shift != 0)
fail(ctx, "DCL face shift scale must be zero");
927
928
929
930
931
} // else if
else
{
unsupported = 1;
} // else
932
933
934
ctx->dwords[0] = (uint32) MOJOSHADER_USAGE_UNKNOWN;
ctx->dwords[1] = 0;
935
936
} // else if
937
else if (regtype == REG_TYPE_TEXTURE)
938
939
940
{
const uint32 usage = (token & 0xF);
const uint32 index = ((token >> 16) & 0xF);
941
if (usage == MOJOSHADER_USAGE_TEXCOORD)
942
943
{
if (index > 7)
944
fail(ctx, "DCL texcoord usage must have 0-7 index");
945
} // if
946
else if (usage == MOJOSHADER_USAGE_COLOR)
947
948
{
if (index != 0)
949
fail(ctx, "DCL color usage must have 0 index");
950
951
952
} // else if
else
{
953
fail(ctx, "Invalid DCL texture usage");
954
955
956
957
958
959
960
} // else
reserved_mask = 0x7FF0FFE0;
ctx->dwords[0] = usage;
ctx->dwords[1] = index;
} // else if
961
else if (regtype == REG_TYPE_SAMPLER)
962
{
963
964
const uint32 ttype = ((token >> 27) & 0xF);
if (!valid_texture_type(ttype))
965
fail(ctx, "unknown sampler texture type");
966
reserved_mask = 0x7FFFFFF;
967
ctx->dwords[0] = ttype;
968
969
970
971
972
973
974
975
} // else if
else
{
unsupported = 1;
} // else
} // if
976
else if ( (shader_is_pixel(ctx)) && (shader_version_atleast(ctx, 2, 0)) )
977
{
978
if (regtype == REG_TYPE_INPUT)
979
980
981
{
ctx->dwords[0] = (uint32) MOJOSHADER_USAGE_COLOR;
ctx->dwords[1] = regnum;
982
reserved_mask = 0x7FFFFFFF;
983
} // if
984
else if (regtype == REG_TYPE_TEXTURE)
985
986
987
{
ctx->dwords[0] = (uint32) MOJOSHADER_USAGE_TEXCOORD;
ctx->dwords[1] = regnum;
988
reserved_mask = 0x7FFFFFFF;
989
} // else if
990
else if (regtype == REG_TYPE_SAMPLER)
991
{
992
993
const uint32 ttype = ((token >> 27) & 0xF);
if (!valid_texture_type(ttype))
994
fail(ctx, "unknown sampler texture type");
995
reserved_mask = 0x7FFFFFF;
996
ctx->dwords[0] = ttype;
997
998
999
1000
} // else if
else
{
unsupported = 1;