/
mojoshader.c
4991 lines (4199 loc) · 157 KB
1
/**
2
3
* MojoShader; generate shader programs from bytecode of compiled
* Direct3D shaders.
4
5
6
7
8
9
*
* Please see the file LICENSE.txt in the source's root directory.
*
* This file written by Ryan C. Gordon.
*/
10
11
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
12
13
14
15
// !!! FIXME: do DEF* opcodes have to come before instructions?
// !!! FIXME: my reading of the msdn spec suggests no, but it sounds like
// !!! FIXME: something they'd require. DCL_* _does_ have to be first.
16
17
18
19
// Shader bytecode format is described at MSDN:
// http://msdn2.microsoft.com/en-us/library/ms800307.aspx
20
#include <stdio.h>
21
#include <string.h>
22
#include <stdlib.h>
23
#include <stdarg.h>
24
#include <assert.h>
25
26
#include "mojoshader.h"
27
28
29
30
31
32
33
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
34
35
36
// 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.
37
38
39
40
41
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
42
43
44
45
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
46
47
48
49
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
50
51
52
53
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
54
55
56
57
58
59
60
61
#ifdef _MSC_VER
#define snprintf _snprintf
typedef unsigned __int8 uint8;
typedef unsigned __int32 uint32;
typedef unsigned __int32 int32;
#else
#include <stdint.h>
62
63
typedef uint8_t uint8;
typedef uint32_t uint32;
64
typedef int32_t int32;
65
#endif
66
67
68
69
70
71
72
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
73
74
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
75
76
77
78
79
80
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
81
82
83
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
84
85
86
87
88
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
// 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
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
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;
134
135
136
137
138
139
140
141
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
142
// predeclare.
143
typedef struct Context Context;
144
145
// one emit function for each opcode in each profile.
146
typedef void (*emit_function)(Context *ctx);
147
148
// one emit function for starting output in each profile.
149
typedef void (*emit_start)(Context *ctx);
150
151
// one emit function for ending output in each profile.
152
typedef void (*emit_end)(Context *ctx);
153
154
155
156
157
158
159
160
161
162
// one emit function for finalizing output in each profile.
typedef void (*emit_finalize)(Context *ctx);
// one emit function for global definitions in each profile.
typedef void (*emit_global)(Context *ctx, RegisterType regtype, int regnum);
// one emit function for uniforms in each profile.
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum);
163
164
165
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
166
167
168
169
// 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);
170
171
172
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
173
// one state function for each opcode where we have state machine updates.
174
typedef void (*state_function)(Context *ctx);
175
176
177
178
179
180
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
181
182
emit_global global_emitter;
emit_uniform uniform_emitter;
183
emit_sampler sampler_emitter;
184
emit_attribute attribute_emitter;
185
emit_finalize finalize_emitter;
186
} Profile;
187
188
189
190
191
192
193
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
194
} RastOutType;
195
196
197
198
199
200
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
201
} MiscTypeType;
202
203
204
205
206
// 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.
207
typedef struct OutputListNode
208
209
{
char *str;
210
211
212
213
214
215
216
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
217
218
} OutputList;
219
220
221
222
223
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
224
225
226
MOJOSHADER_usage usage;
int index;
int writemask;
227
int misc;
228
229
230
231
struct RegisterList *next;
} RegisterList;
232
233
234
235
236
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
// 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
255
} SourceMod;
256
257
258
259
typedef struct
{
260
const uint32 *token; // this is the unmolested token in the stream.
261
262
int regnum;
int relative;
263
264
265
266
267
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
268
269
int result_mod;
int result_shift;
270
RegisterType regtype;
271
272
273
274
} DestArgInfo;
typedef struct
{
275
const uint32 *token; // this is the unmolested token in the stream.
276
int regnum;
277
int swizzle; // xyzw (all four, not split out).
278
279
280
281
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
282
SourceMod src_mod;
283
RegisterType regtype;
284
285
286
287
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
288
289
290
} SourceArgInfo;
291
292
typedef enum
{
293
// Specific to GLSL profile...
294
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
295
296
297
298
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
299
300
#define SCRATCH_BUFFER_SIZE 128
#define SCRATCH_BUFFERS 16
301
302
// !!! FIXME: the scratch buffers make Context pretty big.
303
304
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
305
306
// Context...this is state that changes as we parse through a shader...
307
struct Context
308
{
309
310
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
311
void *malloc_data;
312
313
const uint32 *tokens;
uint32 tokencount;
314
315
OutputList *output;
OutputList globals;
316
OutputList helpers;
317
OutputList subroutines;
318
OutputList mainline_intro;
319
OutputList mainline;
320
321
OutputList ignore;
OutputList *output_stack[2];
322
uint8 *output_bytes; // can be used instead of the OutputLists.
323
324
int indent_stack[2];
int output_stack_len;
325
int output_len; // total strlen; prevents walking the lists just to malloc.
326
int indent;
327
328
329
const char *endline;
int endline_len;
const char *failstr;
330
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
331
332
int scratchidx; // current scratch buffer.
int profileid;
333
const Profile *profile;
334
MOJOSHADER_shaderType shader_type;
335
336
uint8 major_ver;
uint8 minor_ver;
337
DestArgInfo dest_arg;
338
SourceArgInfo source_args[5];
339
SourceArgInfo predicate_arg; // for predicated instructions.
340
uint32 dwords[4];
341
int instruction_count;
342
uint32 instruction_controls;
343
344
uint32 previous_opcode;
ContextFlags flags;
345
int predicated;
346
int loops;
347
int reps;
348
int cmps;
349
350
RegisterList used_registers;
RegisterList defined_registers;
351
352
int uniform_count;
RegisterList uniforms;
353
354
int attribute_count;
RegisterList attributes;
355
356
int sampler_count;
RegisterList samplers;
357
358
359
};
360
361
// Convenience functions for allocators...
362
363
364
365
366
367
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)
368
{
369
370
371
372
373
374
375
376
377
378
379
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
static inline void *Malloc(Context *ctx, const int len)
{
void *retval = ctx->malloc(len, ctx->malloc_data);
if (retval == NULL)
out_of_memory(ctx);
return retval;
380
381
382
} // Malloc
383
static inline void Free(Context *ctx, void *ptr)
384
{
385
if (ptr != NULL) // check for NULL in case of dumb free() impl.
386
ctx->free(ptr, ctx->malloc_data);
387
388
389
} // Free
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
// 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...
414
415
416
417
418
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
419
static inline int shader_version_supported(const uint8 maj, const uint8 min)
420
421
422
423
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
424
425
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
426
427
428
429
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
430
431
432
433
434
435
436
437
438
439
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
440
441
442
// Bit arrays (for storing large arrays of booleans...
443
static inline char *get_scratch_buffer(Context *ctx)
444
{
445
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
446
return ctx->scratch[ctx->scratchidx];
447
448
} // get_scratch_buffer
449
450
static inline int isfail(const Context *ctx)
451
452
453
454
455
{
return (ctx->failstr != NULL);
} // isfail
456
457
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
458
{
459
460
if (ctx->failstr == NULL) // don't change existing error.
{
461
char *scratch = get_scratch_buffer(ctx);
462
463
va_list ap;
va_start(ap, fmt);
464
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
465
466
va_end(ap);
467
char *failstr = (char *) Malloc(ctx, len + 1);
468
if (failstr != NULL)
469
470
{
// see comments about scratch buffer overflow in output_line().
471
if (len < SCRATCH_BUFFER_SIZE)
472
473
474
475
476
477
478
479
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;
480
} // if
481
482
483
484
485
486
} // if
return FAIL;
} // failf
487
static inline int fail(Context *ctx, const char *reason)
488
489
490
491
492
{
return failf(ctx, "%s", reason);
} // fail
493
494
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
495
{
496
497
OutputListNode *item = NULL;
498
if (isfail(ctx))
499
500
return FAIL; // we failed previously, don't go on...
501
char *scratch = get_scratch_buffer(ctx);
502
503
const int indent = ctx->indent;
504
505
if (indent > 0)
memset(scratch, '\t', indent);
506
507
508
va_list ap;
va_start(ap, fmt);
509
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
510
511
va_end(ap);
512
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
513
if (item == NULL)
514
return FAIL;
515
516
item->str = (char *) Malloc(ctx, len + 1);
517
if (item->str == NULL)
518
{
519
Free(ctx, item);
520
return FAIL;
521
522
} // if
523
524
525
// 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().
526
if (len < SCRATCH_BUFFER_SIZE)
527
528
529
strcpy(item->str, scratch); // copy it over.
else
{
530
531
if (indent > 0)
memset(item->str, '\t', indent);
532
va_start(ap, fmt);
533
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
534
535
va_end(ap);
} // else
536
537
item->next = NULL;
538
539
540
ctx->output->tail->next = item;
ctx->output->tail = item;
541
ctx->output_len += len + ctx->endline_len;
542
543
544
545
546
return 0;
} // output_line
547
548
549
550
551
552
553
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
554
// !!! FIXME: this is sort of nasty.
555
556
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
557
558
{
const size_t len = snprintf(buf, bufsize, "%f", f);
559
if ((len+2) >= bufsize)
560
561
562
563
564
565
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
566
567
568
{
if (leavedecimal)
strcat(buf, ".0");
569
return; // done.
570
} // if
571
572
573
574
575
576
577
578
579
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
580
581
if ((leavedecimal) && (end == ptr))
end += 2;
582
583
584
585
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
586
587
588
// Deal with register lists... !!! FIXME: I sort of hate this.
589
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
590
591
592
593
{
while (item != NULL)
{
RegisterList *next = item->next;
594
f(item, d);
595
596
597
598
599
600
601
602
603
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
604
605
606
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
607
608
609
610
611
612
613
{
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)
614
return item; // already set, so we're done.
615
616
617
618
619
620
621
622
623
624
625
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).
626
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
627
if (item != NULL)
628
629
630
{
item->regtype = regtype;
item->regnum = regnum;
631
item->usage = MOJOSHADER_USAGE_UNKNOWN;
632
633
item->index = 0;
item->writemask = 0;
634
item->misc = 0;
635
636
item->next = prev->next;
prev->next = item;
637
} // if
638
639
return item;
640
641
} // reglist_insert
642
643
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
644
{
645
646
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
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.
663
664
665
666
667
668
669
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
670
671
672
673
674
675
676
677
678
679
680
681
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
} // set_used_register
static inline int get_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
return (reglist_exists(&ctx->used_registers, regtype, regnum) != NULL);
682
} // get_used_register
683
684
685
686
687
688
689
690
691
692
693
694
695
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
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
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
711
712
713
714
715
716
717
718
719
720
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
721
722
723
724
725
726
727
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
728
729
730
731
732
733
734
735
736
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
737
738
739
740
741
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
742
743
744
// D3D stuff that's used in more than just the d3d profile...
745
746
747
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
748
749
750
751
752
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
753
754
static const char *get_D3D_register_string(Context *ctx,
755
RegisterType regtype,
756
757
int regnum, char *regnum_str,
size_t regnum_size)
758
{
759
const char *retval = NULL;
760
int has_number = 1;
761
762
switch (regtype)
763
{
764
case REG_TYPE_TEMP:
765
retval = "r";
766
767
break;
768
case REG_TYPE_INPUT:
769
retval = "v";
770
771
break;
772
case REG_TYPE_CONST:
773
774
775
retval = "c";
break;
776
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
777
retval = shader_is_vertex(ctx) ? "a" : "t";
778
779
break;
780
case REG_TYPE_RASTOUT:
781
switch ((RastOutType) regnum)
782
{
783
784
785
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
786
} // switch
787
has_number = 0;
788
789
break;
790
case REG_TYPE_ATTROUT:
791
retval = "oD";
792
793
break;
794
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
795
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
796
retval = "o";
797
else
798
retval = "oT";
799
800
break;
801
case REG_TYPE_CONSTINT:
802
retval = "i";
803
804
break;
805
case REG_TYPE_COLOROUT:
806
retval = "oC";
807
808
break;
809
case REG_TYPE_DEPTHOUT:
810
retval = "oDepth";
811
has_number = 0;
812
813
break;
814
case REG_TYPE_SAMPLER:
815
retval = "s";
816
817
break;
818
case REG_TYPE_CONSTBOOL:
819
retval = "b";
820
821
break;
822
case REG_TYPE_LOOP:
823
retval = "aL";
824
has_number = 0;
825
826
break;
827
case REG_TYPE_MISCTYPE:
828
switch ((MiscTypeType) regnum)
829
{
830
831
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
832
} // switch
833
has_number = 0;
834
835
break;
836
case REG_TYPE_LABEL:
837
retval = "l";
838
839
break;
840
case REG_TYPE_PREDICATE:
841
retval = "p";
842
break;
843
844
845
846
847
848
849
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
850
851
} // switch
852
853
854
855
856
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
857
858
859
860
return retval;
} // get_D3D_register_string
861
862
863
864
865
866
867
868
869
870
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
871
872
873
874
875
876
877
878
879
880
#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,
881
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
882
883
const SourceArgInfo *arg,
char *buf, size_t buflen)
884
885
886
{
const char *premod_str = "";
const char *postmod_str = "";
887
switch (arg->src_mod)
888
889
890
891
892
893
894
895
896
897
898
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
case SRCMOD_SIGNNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_SIGN:
postmod_str = "_bx2";
break;
case SRCMOD_COMPLEMENT:
premod_str = "1-";
break;
case SRCMOD_X2NEGATE:
premod_str = "-";
// fall through.
case SRCMOD_X2:
postmod_str = "_x2";
break;
case SRCMOD_DZ:
postmod_str = "_dz";
break;
case SRCMOD_DW:
postmod_str = "_dw";
break;
case SRCMOD_ABSNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_ABS:
postmod_str = "_abs";
break;
case SRCMOD_NOT:
premod_str = "!";
break;
936
937
938
939
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
940
941
942
943
} // switch
char regnum_str[16];
944
945
946
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
947
948
949
950
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
951
return "";
952
953
} // if
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
const char *rel_lbracket = "";
const char *rel_rbracket = "";
char rel_swizzle[4] = { '\0' };
char rel_regnum_str[16] = { '\0' };
const char *rel_regtype_str = "";
if (arg->relative)
{
rel_swizzle[0] = '.';
rel_swizzle[1] = swizzle_channels[arg->relative_component];
rel_swizzle[2] = '\0';
rel_lbracket = "[";
rel_rbracket = "]";
rel_regtype_str = get_D3D_register_string(ctx, arg->relative_regtype,
arg->relative_regnum,
rel_regnum_str,
sizeof (rel_regnum_str));
if (regtype_str == NULL)
{
fail(ctx, "Unknown relative source register type.");
return "";
} // if
} // if
978
979
char swizzle_str[6];
int i = 0;
980
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
981
982
{
swizzle_str[i++] = '.';
983
984
985
986
swizzle_str[i++] = swizzle_channels[arg->swizzle_x];
swizzle_str[i++] = swizzle_channels[arg->swizzle_y];
swizzle_str[i++] = swizzle_channels[arg->swizzle_z];
swizzle_str[i++] = swizzle_channels[arg->swizzle_w];
987
988
989
990
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
991
992
993
994
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
995
996
997
998
snprintf(buf, buflen, "%s%s%s%s%s%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str,
rel_lbracket, rel_regtype_str, rel_regnum_str, rel_swizzle,
rel_rbracket, swizzle_str);
999
// !!! FIXME: make sure the scratch buffer was large enough.
1000
return buf;