/
mojoshader.c
6647 lines (5620 loc) · 213 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: this file really needs to be split up.
12
13
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
14
15
16
17
// !!! 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.
18
19
20
21
// Shader bytecode format is described at MSDN:
// http://msdn2.microsoft.com/en-us/library/ms800307.aspx
22
#include <stdio.h>
23
#include <string.h>
24
#include <stdlib.h>
25
#include <stdarg.h>
26
#include <assert.h>
27
28
#include "mojoshader.h"
29
30
31
32
33
34
35
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
36
37
38
// 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.
39
40
41
42
43
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
44
45
46
47
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
48
49
50
51
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
52
53
54
55
#ifndef SUPPORT_PROFILE_ARB1
#define SUPPORT_PROFILE_ARB1 1
#endif
56
57
58
59
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
60
61
62
63
64
65
#ifdef _MSC_VER
#define snprintf _snprintf
typedef unsigned __int8 uint8;
typedef unsigned __int32 uint32;
typedef unsigned __int32 int32;
66
67
68
// Warning Level 4 considered harmful. :)
#pragma warning(disable: 4100) // "unreferenced formal parameter"
#pragma warning(disable: 4389) // "signed/unsigned mismatch"
69
70
#else
#include <stdint.h>
71
72
typedef uint8_t uint8;
typedef uint32_t uint32;
73
typedef int32_t int32;
74
#endif
75
76
77
78
79
80
81
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
82
83
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
84
85
86
87
88
89
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
90
91
92
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
93
94
95
96
97
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
// 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
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
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;
143
144
145
146
147
148
149
150
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
151
// predeclare.
152
typedef struct Context Context;
153
154
// one emit function for each opcode in each profile.
155
typedef void (*emit_function)(Context *ctx);
156
157
// one emit function for starting output in each profile.
158
typedef void (*emit_start)(Context *ctx);
159
160
// one emit function for ending output in each profile.
161
typedef void (*emit_end)(Context *ctx);
162
163
164
165
166
167
168
// 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);
169
170
171
// one emit function for relative uniform arrays in each profile.
typedef void (*emit_relative)(Context *ctx, int size);
172
173
174
// one emit function for uniforms in each profile.
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum);
175
176
177
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
178
179
180
181
// 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);
182
183
184
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
185
// one state function for each opcode where we have state machine updates.
186
typedef void (*state_function)(Context *ctx);
187
188
189
190
191
192
193
// one function for varnames in each profile.
typedef const char *(*varname_function)(Context *c, RegisterType t, int num);
// one function for const var array in each profile.
typedef const char *(*const_array_varname_function)(Context *c);
194
195
196
197
198
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
199
emit_global global_emitter;
200
emit_relative relative_emitter;
201
emit_uniform uniform_emitter;
202
emit_sampler sampler_emitter;
203
emit_attribute attribute_emitter;
204
emit_finalize finalize_emitter;
205
206
varname_function get_varname;
const_array_varname_function get_const_array_varname;
207
} Profile;
208
209
210
211
212
213
214
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
215
} RastOutType;
216
217
218
219
220
221
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
222
} MiscTypeType;
223
224
225
226
227
// 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.
228
typedef struct OutputListNode
229
230
{
char *str;
231
232
233
234
235
236
237
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
238
239
} OutputList;
240
241
242
243
244
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
245
246
247
MOJOSHADER_usage usage;
int index;
int writemask;
248
int misc;
249
250
251
struct RegisterList *next;
} RegisterList;
252
253
254
255
256
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
257
258
259
260
261
262
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
// 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
281
} SourceMod;
282
283
284
285
typedef struct
{
286
const uint32 *token; // this is the unmolested token in the stream.
287
288
int regnum;
int relative;
289
290
291
292
293
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
294
int orig_writemask; // writemask before mojoshader tweaks it.
295
296
int result_mod;
int result_shift;
297
RegisterType regtype;
298
299
300
301
} DestArgInfo;
typedef struct
{
302
const uint32 *token; // this is the unmolested token in the stream.
303
int regnum;
304
int swizzle; // xyzw (all four, not split out).
305
306
307
308
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
309
SourceMod src_mod;
310
RegisterType regtype;
311
312
313
314
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
315
316
317
} SourceArgInfo;
318
// !!! FIXME: get rid of this. use a bitfield instead.
319
320
typedef enum
{
321
// Specific to GLSL profile...
322
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
323
324
325
326
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
327
#define SCRATCH_BUFFER_SIZE 128
328
#define SCRATCH_BUFFERS 32
329
330
// !!! FIXME: the scratch buffers make Context pretty big.
331
332
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
333
334
// Context...this is state that changes as we parse through a shader...
335
struct Context
336
{
337
338
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
339
void *malloc_data;
340
341
const uint32 *tokens;
uint32 tokencount;
342
343
OutputList *output;
OutputList globals;
344
OutputList helpers;
345
OutputList subroutines;
346
OutputList mainline_intro;
347
OutputList mainline;
348
349
OutputList ignore;
OutputList *output_stack[2];
350
uint8 *output_bytes; // can be used instead of the OutputLists.
351
352
int indent_stack[2];
int output_stack_len;
353
int output_len; // total strlen; prevents walking the lists just to malloc.
354
int indent;
355
const char *shader_type_str;
356
357
358
const char *endline;
int endline_len;
const char *failstr;
359
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
360
361
int scratchidx; // current scratch buffer.
int profileid;
362
const Profile *profile;
363
MOJOSHADER_shaderType shader_type;
364
365
uint8 major_ver;
uint8 minor_ver;
366
DestArgInfo dest_arg;
367
SourceArgInfo source_args[5];
368
SourceArgInfo predicate_arg; // for predicated instructions.
369
uint32 dwords[4];
370
int instruction_count;
371
uint32 instruction_controls;
372
373
uint32 previous_opcode;
ContextFlags flags;
374
int predicated;
375
376
int max_constreg;
int uniform_array;
377
int loops;
378
int reps;
379
int cmps;
380
381
int scratch_registers;
int max_scratch_registers;
382
int assigned_vertex_attributes;
383
int last_address_reg_component;
384
385
RegisterList used_registers;
RegisterList defined_registers;
386
387
int constant_count;
ConstantsList *constants;
388
389
int uniform_count;
RegisterList uniforms;
390
391
int attribute_count;
RegisterList attributes;
392
393
int sampler_count;
RegisterList samplers;
394
395
396
};
397
398
// Convenience functions for allocators...
399
400
401
402
403
404
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)
405
{
406
407
408
409
410
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
411
static inline void *Malloc(Context *ctx, const size_t len)
412
{
413
void *retval = ctx->malloc((int) len, ctx->malloc_data);
414
415
416
if (retval == NULL)
out_of_memory(ctx);
return retval;
417
418
419
} // Malloc
420
static inline void Free(Context *ctx, void *ptr)
421
{
422
if (ptr != NULL) // check for NULL in case of dumb free() impl.
423
ctx->free(ptr, ctx->malloc_data);
424
425
426
} // Free
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
// 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...
451
452
453
454
455
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
456
static inline int shader_version_supported(const uint8 maj, const uint8 min)
457
458
459
460
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
461
462
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
463
464
465
466
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
467
468
469
470
471
472
473
474
475
476
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
477
478
479
480
481
482
static inline int isfail(const Context *ctx)
{
return (ctx->failstr != NULL);
} // isfail
483
484
static inline char *get_scratch_buffer(Context *ctx)
485
{
486
487
488
489
490
491
492
493
494
495
496
497
if ((ctx->scratchidx >= SCRATCH_BUFFERS) && !isfail(ctx))
{
// can't call fail() here, since it calls back into here.
const char *errstr = "BUG: overflowed scratch buffers";
char *failstr = (char *) Malloc(ctx, strlen(errstr) + 1);
if (failstr != NULL)
{
strcpy(failstr, errstr);
ctx->failstr = failstr;
} // if
} // if
498
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
499
return ctx->scratch[ctx->scratchidx];
500
501
} // get_scratch_buffer
502
503
504
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
505
{
506
507
if (ctx->failstr == NULL) // don't change existing error.
{
508
char *scratch = get_scratch_buffer(ctx);
509
510
va_list ap;
va_start(ap, fmt);
511
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
512
513
va_end(ap);
514
char *failstr = (char *) Malloc(ctx, len + 1);
515
if (failstr != NULL)
516
517
{
// see comments about scratch buffer overflow in output_line().
518
if (len < SCRATCH_BUFFER_SIZE)
519
520
521
522
523
524
525
526
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;
527
} // if
528
529
530
531
532
533
} // if
return FAIL;
} // failf
534
static inline int fail(Context *ctx, const char *reason)
535
536
537
538
539
{
return failf(ctx, "%s", reason);
} // fail
540
541
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
542
{
543
544
OutputListNode *item = NULL;
545
if (isfail(ctx))
546
547
return FAIL; // we failed previously, don't go on...
548
char *scratch = get_scratch_buffer(ctx);
549
550
const int indent = ctx->indent;
551
552
if (indent > 0)
memset(scratch, '\t', indent);
553
554
555
va_list ap;
va_start(ap, fmt);
556
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
557
558
va_end(ap);
559
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
560
if (item == NULL)
561
return FAIL;
562
563
item->str = (char *) Malloc(ctx, len + 1);
564
if (item->str == NULL)
565
{
566
Free(ctx, item);
567
return FAIL;
568
569
} // if
570
571
572
// 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().
573
if (len < SCRATCH_BUFFER_SIZE)
574
575
576
strcpy(item->str, scratch); // copy it over.
else
{
577
578
if (indent > 0)
memset(item->str, '\t', indent);
579
va_start(ap, fmt);
580
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
581
582
va_end(ap);
} // else
583
584
item->next = NULL;
585
586
587
ctx->output->tail->next = item;
ctx->output->tail = item;
588
ctx->output_len += len + ctx->endline_len;
589
590
591
592
593
return 0;
} // output_line
594
595
596
597
598
599
600
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
601
// !!! FIXME: this is sort of nasty.
602
603
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
604
605
{
const size_t len = snprintf(buf, bufsize, "%f", f);
606
if ((len+2) >= bufsize)
607
608
609
610
611
612
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
613
614
615
{
if (leavedecimal)
strcat(buf, ".0");
616
return; // done.
617
} // if
618
619
620
621
622
623
624
625
626
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
627
628
if ((leavedecimal) && (end == ptr))
end += 2;
629
630
631
632
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
633
634
635
// Deal with register lists... !!! FIXME: I sort of hate this.
636
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
637
638
639
640
{
while (item != NULL)
{
RegisterList *next = item->next;
641
f(item, d);
642
643
644
645
646
647
648
649
650
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
651
652
653
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
654
655
656
657
658
659
660
{
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)
661
return item; // already set, so we're done.
662
663
664
665
666
667
668
669
670
671
672
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).
673
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
674
if (item != NULL)
675
676
677
{
item->regtype = regtype;
item->regnum = regnum;
678
item->usage = MOJOSHADER_USAGE_UNKNOWN;
679
680
item->index = 0;
item->writemask = 0;
681
item->misc = 0;
682
683
item->next = prev->next;
prev->next = item;
684
} // if
685
686
return item;
687
688
} // reglist_insert
689
690
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
691
{
692
693
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
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.
710
711
712
713
714
715
716
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
717
718
719
720
721
722
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
723
724
725
if ((regtype == REG_TYPE_CONST) && (ctx->max_constreg < regnum))
ctx->max_constreg = regnum;
726
727
728
729
730
731
} // 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);
732
} // get_used_register
733
734
735
736
737
738
739
740
741
742
743
744
745
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
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
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
761
762
763
764
765
766
767
768
769
770
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
771
772
773
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
774
// !!! FIXME: make sure it doesn't exist?
775
776
777
778
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
static inline int writemask_xyzw(const int writemask)
{
return (writemask == 0xF); // 0xF == 1111. No explicit mask (full!).
} // writemask_xyzw
static inline int writemask_xyz(const int writemask)
{
return (writemask == 0x7); // 0x7 == 0111. (that is: xyz)
} // writemask_xyz
static inline int writemask_xy(const int writemask)
{
return (writemask == 0x3); // 0x3 == 0011. (that is: xy)
} // writemask_xy
static inline int writemask_x(const int writemask)
{
return (writemask == 0x1); // 0x1 == 0001. (that is: x)
} // writemask_x
static inline int writemask_y(const int writemask)
{
return (writemask == 0x2); // 0x1 == 0010. (that is: y)
} // writemask_y
810
811
812
813
814
815
816
817
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
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
static inline int scalar_register(const RegisterType regtype, const int regnum)
{
switch (regtype)
{
case REG_TYPE_DEPTHOUT:
case REG_TYPE_CONSTBOOL:
case REG_TYPE_PREDICATE:
case REG_TYPE_LOOP:
return 1;
case REG_TYPE_MISCTYPE:
if ( ((const MiscTypeType) regnum) == MISCTYPE_TYPE_FACE )
return 1;
return 0;
default: break;
} // switch
return 0;
} // scalar_register
840
841
static inline int no_swizzle(const int swizzle)
{
842
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
843
844
845
} // no_swizzle
846
847
848
849
850
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
851
852
853
854
855
856
857
858
static int allocate_scratch_register(Context *ctx)
{
const int retval = ctx->scratch_registers++;
if (retval >= ctx->max_scratch_registers)
ctx->max_scratch_registers = retval + 1;
return retval;
} // allocate_scratch_register
859
860
861
// D3D stuff that's used in more than just the d3d profile...
862
863
864
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
865
866
867
868
869
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
870
871
static const char *get_D3D_register_string(Context *ctx,
872
RegisterType regtype,
873
874
int regnum, char *regnum_str,
size_t regnum_size)
875
{
876
const char *retval = NULL;
877
int has_number = 1;
878
879
switch (regtype)
880
{
881
case REG_TYPE_TEMP:
882
retval = "r";
883
884
break;
885
case REG_TYPE_INPUT:
886
retval = "v";
887
888
break;
889
case REG_TYPE_CONST:
890
891
892
retval = "c";
break;
893
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
894
retval = shader_is_vertex(ctx) ? "a" : "t";
895
896
break;
897
case REG_TYPE_RASTOUT:
898
switch ((RastOutType) regnum)
899
{
900
901
902
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
903
} // switch
904
has_number = 0;
905
906
break;
907
case REG_TYPE_ATTROUT:
908
retval = "oD";
909
910
break;
911
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
912
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
913
retval = "o";
914
else
915
retval = "oT";
916
917
break;
918
case REG_TYPE_CONSTINT:
919
retval = "i";
920
921
break;
922
case REG_TYPE_COLOROUT:
923
retval = "oC";
924
925
break;
926
case REG_TYPE_DEPTHOUT:
927
retval = "oDepth";
928
has_number = 0;
929
930
break;
931
case REG_TYPE_SAMPLER:
932
retval = "s";
933
934
break;
935
case REG_TYPE_CONSTBOOL:
936
retval = "b";
937
938
break;
939
case REG_TYPE_LOOP:
940
retval = "aL";
941
has_number = 0;
942
943
break;
944
case REG_TYPE_MISCTYPE:
945
switch ((const MiscTypeType) regnum)
946
{
947
948
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
949
} // switch
950
has_number = 0;
951
952
break;
953
case REG_TYPE_LABEL:
954
retval = "l";
955
956
break;
957
case REG_TYPE_PREDICATE:
958
retval = "p";
959
break;
960
961
962
963
964
965
966
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
967
968
} // switch
969
970
971
972
973
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
974
975
976
977
return retval;
} // get_D3D_register_string
978
979
980
981
982
983
984
985
986
#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,
987
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
988
989
const SourceArgInfo *arg,
char *buf, size_t buflen)
990
991
992
{
const char *premod_str = "";
const char *postmod_str = "";
993
switch (arg->src_mod)
994
995
996
997
998
999
1000
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";