/
mojoshader.c
6865 lines (5794 loc) · 221 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
// Shader bytecode format is described at MSDN:
// http://msdn2.microsoft.com/en-us/library/ms800307.aspx
17
#include <stdio.h>
18
#include <string.h>
19
#include <stdlib.h>
20
#include <stdarg.h>
21
#include <assert.h>
22
23
#include "mojoshader.h"
24
25
26
27
28
29
30
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
31
32
33
// 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.
34
35
36
37
38
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
39
40
41
42
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
43
44
45
46
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
47
48
49
50
#ifndef SUPPORT_PROFILE_ARB1
#define SUPPORT_PROFILE_ARB1 1
#endif
51
52
53
54
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
55
56
57
58
59
60
#ifdef _MSC_VER
#define snprintf _snprintf
typedef unsigned __int8 uint8;
typedef unsigned __int32 uint32;
typedef unsigned __int32 int32;
61
62
63
// Warning Level 4 considered harmful. :)
#pragma warning(disable: 4100) // "unreferenced formal parameter"
#pragma warning(disable: 4389) // "signed/unsigned mismatch"
64
65
#else
#include <stdint.h>
66
67
typedef uint8_t uint8;
typedef uint32_t uint32;
68
typedef int32_t int32;
69
#endif
70
71
72
73
74
75
76
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
77
78
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
79
80
81
82
83
84
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
85
86
87
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
88
89
90
91
92
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
// 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
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
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;
138
139
140
141
142
143
144
145
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
146
// predeclare.
147
typedef struct Context Context;
148
149
// one emit function for each opcode in each profile.
150
typedef void (*emit_function)(Context *ctx);
151
152
// one emit function for starting output in each profile.
153
typedef void (*emit_start)(Context *ctx, const char *profilestr);
154
155
// one emit function for ending output in each profile.
156
typedef void (*emit_end)(Context *ctx);
157
158
159
160
161
162
163
// 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);
164
165
166
// one emit function for relative uniform arrays in each profile.
typedef void (*emit_relative)(Context *ctx, int size);
167
168
169
// one emit function for uniforms in each profile.
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum);
170
171
172
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
173
174
175
176
// 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);
177
178
179
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
180
// one state function for each opcode where we have state machine updates.
181
typedef void (*state_function)(Context *ctx);
182
183
184
185
186
187
188
// 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);
189
190
191
192
193
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
194
emit_global global_emitter;
195
emit_relative relative_emitter;
196
emit_uniform uniform_emitter;
197
emit_sampler sampler_emitter;
198
emit_attribute attribute_emitter;
199
emit_finalize finalize_emitter;
200
201
varname_function get_varname;
const_array_varname_function get_const_array_varname;
202
} Profile;
203
204
205
206
207
208
209
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
210
} RastOutType;
211
212
213
214
215
216
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
217
} MiscTypeType;
218
219
220
221
222
// 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.
223
typedef struct OutputListNode
224
225
{
char *str;
226
227
228
229
230
231
232
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
233
234
} OutputList;
235
236
237
238
239
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
240
241
242
MOJOSHADER_usage usage;
int index;
int writemask;
243
int misc;
244
245
246
struct RegisterList *next;
} RegisterList;
247
248
249
250
251
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
252
253
254
255
256
257
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
// 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
276
} SourceMod;
277
278
279
280
typedef struct
{
281
const uint32 *token; // this is the unmolested token in the stream.
282
283
int regnum;
int relative;
284
285
286
287
288
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
289
int orig_writemask; // writemask before mojoshader tweaks it.
290
291
int result_mod;
int result_shift;
292
RegisterType regtype;
293
294
295
296
} DestArgInfo;
typedef struct
{
297
const uint32 *token; // this is the unmolested token in the stream.
298
int regnum;
299
int swizzle; // xyzw (all four, not split out).
300
301
302
303
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
304
SourceMod src_mod;
305
RegisterType regtype;
306
307
308
309
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
310
311
312
} SourceArgInfo;
313
314
#define SCRATCH_BUFFER_SIZE 128
315
#define SCRATCH_BUFFERS 32
316
317
// !!! FIXME: the scratch buffers make Context pretty big.
318
319
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
320
321
// Context...this is state that changes as we parse through a shader...
322
struct Context
323
{
324
325
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
326
void *malloc_data;
327
328
const uint32 *tokens;
uint32 tokencount;
329
330
OutputList *output;
OutputList globals;
331
OutputList helpers;
332
OutputList subroutines;
333
OutputList mainline_intro;
334
OutputList mainline;
335
336
OutputList ignore;
OutputList *output_stack[2];
337
uint8 *output_bytes; // can be used instead of the OutputLists.
338
339
int indent_stack[2];
int output_stack_len;
340
int output_len; // total strlen; prevents walking the lists just to malloc.
341
int indent;
342
const char *shader_type_str;
343
344
345
const char *endline;
int endline_len;
const char *failstr;
346
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
347
348
int scratchidx; // current scratch buffer.
int profileid;
349
const Profile *profile;
350
MOJOSHADER_shaderType shader_type;
351
352
uint8 major_ver;
uint8 minor_ver;
353
DestArgInfo dest_arg;
354
SourceArgInfo source_args[5];
355
SourceArgInfo predicate_arg; // for predicated instructions.
356
uint32 dwords[4];
357
int instruction_count;
358
uint32 instruction_controls;
359
uint32 previous_opcode;
360
int max_constreg;
361
int loops;
362
int reps;
363
int cmps;
364
365
int scratch_registers;
int max_scratch_registers;
366
367
368
int if_labels_stack_index;
int if_labels_stack[32];
int assigned_if_labels;
369
int assigned_vertex_attributes;
370
int last_address_reg_component;
371
372
RegisterList used_registers;
RegisterList defined_registers;
373
374
int constant_count;
ConstantsList *constants;
375
376
int uniform_count;
RegisterList uniforms;
377
378
int attribute_count;
RegisterList attributes;
379
380
int sampler_count;
RegisterList samplers;
381
382
383
384
int predicated:1;
int uniform_array:1;
int support_nv2:1;
int glsl_generated_lit_opcode:1;
385
386
387
};
388
389
// Convenience functions for allocators...
390
391
392
393
394
395
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)
396
{
397
398
399
400
401
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
402
static inline void *Malloc(Context *ctx, const size_t len)
403
{
404
void *retval = ctx->malloc((int) len, ctx->malloc_data);
405
406
407
if (retval == NULL)
out_of_memory(ctx);
return retval;
408
409
410
} // Malloc
411
static inline void Free(Context *ctx, void *ptr)
412
{
413
if (ptr != NULL) // check for NULL in case of dumb free() impl.
414
ctx->free(ptr, ctx->malloc_data);
415
416
417
} // Free
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
// 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...
442
443
444
445
446
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
447
static inline int shader_version_supported(const uint8 maj, const uint8 min)
448
449
450
451
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
452
453
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
454
455
456
457
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
458
459
460
461
462
463
464
465
466
467
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
468
469
470
471
472
473
static inline int isfail(const Context *ctx)
{
return (ctx->failstr != NULL);
} // isfail
474
475
static inline char *get_scratch_buffer(Context *ctx)
476
{
477
478
479
480
481
482
483
484
485
486
487
488
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
489
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
490
return ctx->scratch[ctx->scratchidx];
491
492
} // get_scratch_buffer
493
494
495
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
496
{
497
498
if (ctx->failstr == NULL) // don't change existing error.
{
499
char *scratch = get_scratch_buffer(ctx);
500
501
va_list ap;
va_start(ap, fmt);
502
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
503
504
va_end(ap);
505
char *failstr = (char *) Malloc(ctx, len + 1);
506
if (failstr != NULL)
507
508
{
// see comments about scratch buffer overflow in output_line().
509
if (len < SCRATCH_BUFFER_SIZE)
510
511
512
513
514
515
516
517
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;
518
} // if
519
520
521
522
523
524
} // if
return FAIL;
} // failf
525
static inline int fail(Context *ctx, const char *reason)
526
527
528
529
530
{
return failf(ctx, "%s", reason);
} // fail
531
532
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
533
{
534
535
OutputListNode *item = NULL;
536
if (isfail(ctx))
537
538
return FAIL; // we failed previously, don't go on...
539
char *scratch = get_scratch_buffer(ctx);
540
541
const int indent = ctx->indent;
542
543
if (indent > 0)
memset(scratch, '\t', indent);
544
545
546
va_list ap;
va_start(ap, fmt);
547
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
548
549
va_end(ap);
550
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
551
if (item == NULL)
552
return FAIL;
553
554
item->str = (char *) Malloc(ctx, len + 1);
555
if (item->str == NULL)
556
{
557
Free(ctx, item);
558
return FAIL;
559
560
} // if
561
562
563
// 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().
564
if (len < SCRATCH_BUFFER_SIZE)
565
566
567
strcpy(item->str, scratch); // copy it over.
else
{
568
569
if (indent > 0)
memset(item->str, '\t', indent);
570
va_start(ap, fmt);
571
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
572
573
va_end(ap);
} // else
574
575
item->next = NULL;
576
577
578
ctx->output->tail->next = item;
ctx->output->tail = item;
579
ctx->output_len += len + ctx->endline_len;
580
581
582
583
584
return 0;
} // output_line
585
586
587
588
589
590
591
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
592
// !!! FIXME: this is sort of nasty.
593
594
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
595
596
{
const size_t len = snprintf(buf, bufsize, "%f", f);
597
if ((len+2) >= bufsize)
598
599
600
601
602
603
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
604
605
606
{
if (leavedecimal)
strcat(buf, ".0");
607
return; // done.
608
} // if
609
610
611
612
613
614
615
616
617
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
618
619
if ((leavedecimal) && (end == ptr))
end += 2;
620
621
622
623
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
624
625
626
// Deal with register lists... !!! FIXME: I sort of hate this.
627
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
628
629
630
631
{
while (item != NULL)
{
RegisterList *next = item->next;
632
f(item, d);
633
634
635
636
637
638
639
640
641
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
642
643
644
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
645
646
647
648
649
650
651
{
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)
652
return item; // already set, so we're done.
653
654
655
656
657
658
659
660
661
662
663
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).
664
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
665
if (item != NULL)
666
667
668
{
item->regtype = regtype;
item->regnum = regnum;
669
item->usage = MOJOSHADER_USAGE_UNKNOWN;
670
671
item->index = 0;
item->writemask = 0;
672
item->misc = 0;
673
674
item->next = prev->next;
prev->next = item;
675
} // if
676
677
return item;
678
679
} // reglist_insert
680
681
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
682
{
683
684
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
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.
701
702
703
704
705
706
707
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
708
709
710
711
712
713
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
714
715
716
if ((regtype == REG_TYPE_CONST) && (ctx->max_constreg < regnum))
ctx->max_constreg = regnum;
717
718
719
720
721
722
} // 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);
723
} // get_used_register
724
725
726
727
728
729
730
731
732
733
734
735
736
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
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
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
752
753
754
755
756
757
758
759
760
761
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
762
763
764
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
765
// !!! FIXME: make sure it doesn't exist?
766
767
768
769
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
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
801
802
803
804
805
806
807
808
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
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
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
831
832
static inline int no_swizzle(const int swizzle)
{
833
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
834
835
836
} // no_swizzle
837
838
839
840
841
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
842
843
844
845
846
847
848
849
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
850
851
852
853
854
static int allocate_if_label(Context *ctx)
{
return ctx->assigned_if_labels++;
} // allocate_if_label
855
856
857
// D3D stuff that's used in more than just the d3d profile...
858
859
860
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
861
862
863
864
865
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
866
867
static const char *get_D3D_register_string(Context *ctx,
868
RegisterType regtype,
869
870
int regnum, char *regnum_str,
size_t regnum_size)
871
{
872
const char *retval = NULL;
873
int has_number = 1;
874
875
switch (regtype)
876
{
877
case REG_TYPE_TEMP:
878
retval = "r";
879
880
break;
881
case REG_TYPE_INPUT:
882
retval = "v";
883
884
break;
885
case REG_TYPE_CONST:
886
887
888
retval = "c";
break;
889
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
890
retval = shader_is_vertex(ctx) ? "a" : "t";
891
892
break;
893
case REG_TYPE_RASTOUT:
894
switch ((RastOutType) regnum)
895
{
896
897
898
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
899
} // switch
900
has_number = 0;
901
902
break;
903
case REG_TYPE_ATTROUT:
904
retval = "oD";
905
906
break;
907
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
908
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
909
retval = "o";
910
else
911
retval = "oT";
912
913
break;
914
case REG_TYPE_CONSTINT:
915
retval = "i";
916
917
break;
918
case REG_TYPE_COLOROUT:
919
retval = "oC";
920
921
break;
922
case REG_TYPE_DEPTHOUT:
923
retval = "oDepth";
924
has_number = 0;
925
926
break;
927
case REG_TYPE_SAMPLER:
928
retval = "s";
929
930
break;
931
case REG_TYPE_CONSTBOOL:
932
retval = "b";
933
934
break;
935
case REG_TYPE_LOOP:
936
retval = "aL";
937
has_number = 0;
938
939
break;
940
case REG_TYPE_MISCTYPE:
941
switch ((const MiscTypeType) regnum)
942
{
943
944
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
945
} // switch
946
has_number = 0;
947
948
break;
949
case REG_TYPE_LABEL:
950
retval = "l";
951
952
break;
953
case REG_TYPE_PREDICATE:
954
retval = "p";
955
break;
956
957
958
959
960
961
962
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
963
964
} // switch
965
966
967
968
969
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
970
971
972
973
return retval;
} // get_D3D_register_string
974
975
976
977
978
979
980
981
982
#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,
983
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
984
985
const SourceArgInfo *arg,
char *buf, size_t buflen)
986
987
988
{
const char *premod_str = "";
const char *postmod_str = "";
989
switch (arg->src_mod)
990
991
992
993
994
995
996
997
998
999
1000
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;