/
mojoshader.c
6093 lines (5134 loc) · 192 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
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
193
emit_global global_emitter;
194
emit_relative relative_emitter;
195
emit_uniform uniform_emitter;
196
emit_sampler sampler_emitter;
197
emit_attribute attribute_emitter;
198
emit_finalize finalize_emitter;
199
} Profile;
200
201
202
203
204
205
206
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
207
} RastOutType;
208
209
210
211
212
213
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
214
} MiscTypeType;
215
216
217
218
219
// 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.
220
typedef struct OutputListNode
221
222
{
char *str;
223
224
225
226
227
228
229
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
230
231
} OutputList;
232
233
234
235
236
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
237
238
239
MOJOSHADER_usage usage;
int index;
int writemask;
240
int misc;
241
242
243
struct RegisterList *next;
} RegisterList;
244
245
246
247
248
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
249
250
251
252
253
254
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
// 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
273
} SourceMod;
274
275
276
277
typedef struct
{
278
const uint32 *token; // this is the unmolested token in the stream.
279
280
int regnum;
int relative;
281
282
283
284
285
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
286
int orig_writemask; // writemask before mojoshader tweaks it.
287
288
int result_mod;
int result_shift;
289
RegisterType regtype;
290
291
292
293
} DestArgInfo;
typedef struct
{
294
const uint32 *token; // this is the unmolested token in the stream.
295
int regnum;
296
int swizzle; // xyzw (all four, not split out).
297
298
299
300
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
301
SourceMod src_mod;
302
RegisterType regtype;
303
304
305
306
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
307
308
309
} SourceArgInfo;
310
// !!! FIXME: get rid of this. use a bitfield instead.
311
312
typedef enum
{
313
314
315
316
// Specific to ARB1 profile...
CTX_FLAGS_ARB1_USES_SCRATCH1 = (1 << 0),
CTX_FLAGS_ARB1_USES_SCRATCH2 = (1 << 1),
317
// Specific to GLSL profile...
318
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
319
320
321
322
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
323
324
#define SCRATCH_BUFFER_SIZE 128
#define SCRATCH_BUFFERS 16
325
326
// !!! FIXME: the scratch buffers make Context pretty big.
327
328
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
329
330
// Context...this is state that changes as we parse through a shader...
331
struct Context
332
{
333
334
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
335
void *malloc_data;
336
337
const uint32 *tokens;
uint32 tokencount;
338
339
OutputList *output;
OutputList globals;
340
OutputList helpers;
341
OutputList subroutines;
342
OutputList mainline_intro;
343
OutputList mainline;
344
345
OutputList ignore;
OutputList *output_stack[2];
346
uint8 *output_bytes; // can be used instead of the OutputLists.
347
348
int indent_stack[2];
int output_stack_len;
349
int output_len; // total strlen; prevents walking the lists just to malloc.
350
int indent;
351
352
353
const char *endline;
int endline_len;
const char *failstr;
354
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
355
356
int scratchidx; // current scratch buffer.
int profileid;
357
const Profile *profile;
358
MOJOSHADER_shaderType shader_type;
359
360
uint8 major_ver;
uint8 minor_ver;
361
DestArgInfo dest_arg;
362
SourceArgInfo source_args[5];
363
SourceArgInfo predicate_arg; // for predicated instructions.
364
uint32 dwords[4];
365
int instruction_count;
366
uint32 instruction_controls;
367
368
uint32 previous_opcode;
ContextFlags flags;
369
int predicated;
370
371
int max_constreg;
int uniform_array;
372
int loops;
373
int reps;
374
int cmps;
375
376
RegisterList used_registers;
RegisterList defined_registers;
377
378
int constant_count;
ConstantsList *constants;
379
380
int uniform_count;
RegisterList uniforms;
381
382
int attribute_count;
RegisterList attributes;
383
384
int sampler_count;
RegisterList samplers;
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
// Bit arrays (for storing large arrays of booleans...
471
static inline char *get_scratch_buffer(Context *ctx)
472
{
473
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
474
return ctx->scratch[ctx->scratchidx];
475
476
} // get_scratch_buffer
477
478
static inline int isfail(const Context *ctx)
479
480
481
482
483
{
return (ctx->failstr != NULL);
} // isfail
484
485
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
486
{
487
488
if (ctx->failstr == NULL) // don't change existing error.
{
489
char *scratch = get_scratch_buffer(ctx);
490
491
va_list ap;
va_start(ap, fmt);
492
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
493
494
va_end(ap);
495
char *failstr = (char *) Malloc(ctx, len + 1);
496
if (failstr != NULL)
497
498
{
// see comments about scratch buffer overflow in output_line().
499
if (len < SCRATCH_BUFFER_SIZE)
500
501
502
503
504
505
506
507
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;
508
} // if
509
510
511
512
513
514
} // if
return FAIL;
} // failf
515
static inline int fail(Context *ctx, const char *reason)
516
517
518
519
520
{
return failf(ctx, "%s", reason);
} // fail
521
522
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
523
{
524
525
OutputListNode *item = NULL;
526
if (isfail(ctx))
527
528
return FAIL; // we failed previously, don't go on...
529
char *scratch = get_scratch_buffer(ctx);
530
531
const int indent = ctx->indent;
532
533
if (indent > 0)
memset(scratch, '\t', indent);
534
535
536
va_list ap;
va_start(ap, fmt);
537
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
538
539
va_end(ap);
540
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
541
if (item == NULL)
542
return FAIL;
543
544
item->str = (char *) Malloc(ctx, len + 1);
545
if (item->str == NULL)
546
{
547
Free(ctx, item);
548
return FAIL;
549
550
} // if
551
552
553
// 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().
554
if (len < SCRATCH_BUFFER_SIZE)
555
556
557
strcpy(item->str, scratch); // copy it over.
else
{
558
559
if (indent > 0)
memset(item->str, '\t', indent);
560
va_start(ap, fmt);
561
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
562
563
va_end(ap);
} // else
564
565
item->next = NULL;
566
567
568
ctx->output->tail->next = item;
ctx->output->tail = item;
569
ctx->output_len += len + ctx->endline_len;
570
571
572
573
574
return 0;
} // output_line
575
576
577
578
579
580
581
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
582
// !!! FIXME: this is sort of nasty.
583
584
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
585
586
{
const size_t len = snprintf(buf, bufsize, "%f", f);
587
if ((len+2) >= bufsize)
588
589
590
591
592
593
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
594
595
596
{
if (leavedecimal)
strcat(buf, ".0");
597
return; // done.
598
} // if
599
600
601
602
603
604
605
606
607
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
608
609
if ((leavedecimal) && (end == ptr))
end += 2;
610
611
612
613
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
614
615
616
// Deal with register lists... !!! FIXME: I sort of hate this.
617
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
618
619
620
621
{
while (item != NULL)
{
RegisterList *next = item->next;
622
f(item, d);
623
624
625
626
627
628
629
630
631
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
632
633
634
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
635
636
637
638
639
640
641
{
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)
642
return item; // already set, so we're done.
643
644
645
646
647
648
649
650
651
652
653
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).
654
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
655
if (item != NULL)
656
657
658
{
item->regtype = regtype;
item->regnum = regnum;
659
item->usage = MOJOSHADER_USAGE_UNKNOWN;
660
661
item->index = 0;
item->writemask = 0;
662
item->misc = 0;
663
664
item->next = prev->next;
prev->next = item;
665
} // if
666
667
return item;
668
669
} // reglist_insert
670
671
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
672
{
673
674
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
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.
691
692
693
694
695
696
697
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
698
699
700
701
702
703
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
704
705
706
if ((regtype == REG_TYPE_CONST) && (ctx->max_constreg < regnum))
ctx->max_constreg = regnum;
707
708
709
710
711
712
} // 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);
713
} // get_used_register
714
715
716
717
718
719
720
721
722
723
724
725
726
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
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
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
742
743
744
745
746
747
748
749
750
751
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
752
753
754
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
755
// !!! FIXME: make sure it doesn't exist?
756
757
758
759
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
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
791
792
793
794
795
796
797
798
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
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
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
821
822
static inline int no_swizzle(const int swizzle)
{
823
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
824
825
826
} // no_swizzle
827
828
829
830
831
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
832
833
834
// D3D stuff that's used in more than just the d3d profile...
835
836
837
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
838
839
840
841
842
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
843
844
static const char *get_D3D_register_string(Context *ctx,
845
RegisterType regtype,
846
847
int regnum, char *regnum_str,
size_t regnum_size)
848
{
849
const char *retval = NULL;
850
int has_number = 1;
851
852
switch (regtype)
853
{
854
case REG_TYPE_TEMP:
855
retval = "r";
856
857
break;
858
case REG_TYPE_INPUT:
859
retval = "v";
860
861
break;
862
case REG_TYPE_CONST:
863
864
865
retval = "c";
break;
866
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
867
retval = shader_is_vertex(ctx) ? "a" : "t";
868
869
break;
870
case REG_TYPE_RASTOUT:
871
switch ((RastOutType) regnum)
872
{
873
874
875
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
876
} // switch
877
has_number = 0;
878
879
break;
880
case REG_TYPE_ATTROUT:
881
retval = "oD";
882
883
break;
884
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
885
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
886
retval = "o";
887
else
888
retval = "oT";
889
890
break;
891
case REG_TYPE_CONSTINT:
892
retval = "i";
893
894
break;
895
case REG_TYPE_COLOROUT:
896
retval = "oC";
897
898
break;
899
case REG_TYPE_DEPTHOUT:
900
retval = "oDepth";
901
has_number = 0;
902
903
break;
904
case REG_TYPE_SAMPLER:
905
retval = "s";
906
907
break;
908
case REG_TYPE_CONSTBOOL:
909
retval = "b";
910
911
break;
912
case REG_TYPE_LOOP:
913
retval = "aL";
914
has_number = 0;
915
916
break;
917
case REG_TYPE_MISCTYPE:
918
switch ((const MiscTypeType) regnum)
919
{
920
921
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
922
} // switch
923
has_number = 0;
924
925
break;
926
case REG_TYPE_LABEL:
927
retval = "l";
928
929
break;
930
case REG_TYPE_PREDICATE:
931
retval = "p";
932
break;
933
934
935
936
937
938
939
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
940
941
} // switch
942
943
944
945
946
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
947
948
949
950
return retval;
} // get_D3D_register_string
951
952
953
954
955
956
957
958
959
960
static inline const char *get_shader_type_string(Context *ctx)
{
if (shader_is_pixel(ctx))
return "ps";
else if (shader_is_vertex(ctx))
return "vs";
fail(ctx, "Unknown shader type.");
return "";
} // get_shader_type_string
961
962
963
964
965
966
967
968
969
970
#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,
971
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
972
973
const SourceArgInfo *arg,
char *buf, size_t buflen)
974
975
976
{
const char *premod_str = "";
const char *postmod_str = "";
977
switch (arg->src_mod)
978
979
980
981
982
983
984
985
986
987
988
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
989
990
991
992
993
994
995
996
997
998
999
1000
case SRCMOD_SIGNNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_SIGN:
postmod_str = "_bx2";
break;
case SRCMOD_COMPLEMENT:
premod_str = "1-";
break;