/
mojoshader.c
3886 lines (3240 loc) · 119 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 <stdint.h>
24
#include <stdarg.h>
25
#include <assert.h>
26
27
#include "mojoshader.h"
28
29
30
31
32
33
34
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
35
36
37
// 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.
38
39
40
41
42
43
44
45
46
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
47
48
49
50
51
52
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
typedef uint8_t uint8;
typedef uint32_t uint32;
53
typedef int32_t int32;
54
55
56
57
58
59
60
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
61
62
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
63
64
65
66
67
68
69
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
70
71
72
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
// 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
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
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;
118
119
// predeclare.
120
typedef struct Context Context;
121
122
// one emit function for each opcode in each profile.
123
typedef void (*emit_function)(Context *ctx);
124
125
// one emit function for comments in each profile.
126
typedef void (*emit_comment)(Context *ctx, const char *str);
127
128
// one emit function for starting output in each profile.
129
typedef void (*emit_start)(Context *ctx);
130
131
// one emit function for ending output in each profile.
132
typedef void (*emit_end)(Context *ctx);
133
134
135
136
137
138
139
140
141
142
// 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);
143
144
145
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
146
// one state function for each opcode where we have state machine updates.
147
typedef void (*state_function)(Context *ctx);
148
149
150
151
152
153
154
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
emit_comment comment_emitter;
155
156
157
emit_global global_emitter;
emit_uniform uniform_emitter;
emit_finalize finalize_emitter;
158
} Profile;
159
160
161
162
163
164
165
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
166
} RastOutType;
167
168
169
170
171
172
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
173
} MiscTypeType;
174
175
176
177
178
179
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
180
} TextureType;
181
182
183
184
185
// 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.
186
typedef struct OutputListNode
187
188
{
char *str;
189
190
191
192
193
194
195
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
196
197
} OutputList;
198
199
200
201
202
203
204
205
206
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
struct RegisterList *next;
} RegisterList;
207
208
209
210
211
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
// 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
230
} SourceMod;
231
232
233
234
typedef struct
{
235
const uint32 *token; // this is the unmolested token in the stream.
236
237
int regnum;
int relative;
238
239
240
241
242
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
243
244
int result_mod;
int result_shift;
245
RegisterType regtype;
246
247
248
249
} DestArgInfo;
typedef struct
{
250
const uint32 *token; // this is the unmolested token in the stream.
251
252
int regnum;
int relative;
253
int swizzle; // xyzw (all four, not split out).
254
255
256
257
258
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
int src_mod;
259
RegisterType regtype;
260
261
262
} SourceArgInfo;
263
264
typedef enum
{
265
// Specific to GLSL profile...
266
267
268
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
CTX_FLAGS_GLSL_DST_OPCODE = (1 << 1),
CTX_FLAGS_GLSL_LRP_OPCODE = (1 << 2),
269
270
271
272
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
273
274
#define SCRATCH_BUFFER_SIZE 256
#define SCRATCH_BUFFERS 10
275
276
// !!! FIXME: the scratch buffers make Context pretty big.
277
278
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
279
280
// Context...this is state that changes as we parse through a shader...
281
struct Context
282
{
283
284
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
285
void *malloc_data;
286
287
const uint32 *tokens;
uint32 tokencount;
288
289
OutputList *output;
OutputList globals;
290
OutputList helpers;
291
292
OutputList subroutines;
OutputList mainline;
293
294
295
296
OutputList ignore;
OutputList *output_stack[2];
int indent_stack[2];
int output_stack_len;
297
int output_len; // total strlen; prevents walking the lists just to malloc.
298
int indent;
299
300
301
const char *endline;
int endline_len;
const char *failstr;
302
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
303
304
int scratchidx; // current scratch buffer.
int profileid;
305
const Profile *profile;
306
MOJOSHADER_shaderType shader_type;
307
308
uint8 major_ver;
uint8 minor_ver;
309
DestArgInfo dest_args[1];
310
SourceArgInfo source_args[5];
311
uint32 dwords[4];
312
int instruction_count;
313
uint32 instruction_controls;
314
315
316
uint32 previous_opcode;
ContextFlags flags;
int loops;
317
318
RegisterList used_registers;
RegisterList defined_registers;
319
320
int uniform_count;
RegisterList uniforms;
321
322
323
};
324
325
326
327
328
329
330
331
332
333
// Convenience functions for allocators...
static inline void *Malloc(Context *ctx, int len)
{
return ctx->malloc(len, ctx->malloc_data);
} // Malloc
static inline void Free(Context *ctx, void *ptr)
{
334
335
336
// check for NULL in case of dumb free() impl.
if (ptr != NULL)
ctx->free(ptr, ctx->malloc_data);
337
338
339
340
} // Free
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
// 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...
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
static int shader_version_supported(uint8 maj, uint8 min)
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
static int shader_version_atleast(const Context *ctx, uint8 maj, uint8 min)
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
381
382
// Bit arrays (for storing large arrays of booleans...
383
static inline char *get_scratch_buffer(Context *ctx)
384
{
385
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
386
return ctx->scratch[ctx->scratchidx];
387
388
} // get_scratch_buffer
389
390
391
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
392
393
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
394
395
static inline int isfail(const Context *ctx)
396
397
398
399
400
{
return (ctx->failstr != NULL);
} // isfail
401
402
403
404
405
static MOJOSHADER_parseData out_of_mem_data = {
"Out of memory", 0, 0, 0, MOJOSHADER_TYPE_UNKNOWN, 0, 0, 0, 0
};
static const char *out_of_mem_str = "Out of memory";
406
407
408
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
409
ctx->failstr = out_of_mem_str; // fail() would call malloc().
410
411
412
413
414
415
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
416
{
417
418
if (ctx->failstr == NULL) // don't change existing error.
{
419
char *scratch = get_scratch_buffer(ctx);
420
421
va_list ap;
va_start(ap, fmt);
422
const int len = vsnprintf(scratch,SCRATCH_BUFFER_SIZE,fmt,ap);
423
424
va_end(ap);
425
char *failstr = (char *) Malloc(ctx, len + 1);
426
427
428
429
430
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
431
if (len < SCRATCH_BUFFER_SIZE)
432
433
434
435
436
437
438
439
440
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;
} // else
441
442
443
444
445
446
} // if
return FAIL;
} // failf
447
static inline int fail(Context *ctx, const char *reason)
448
449
450
451
452
{
return failf(ctx, "%s", reason);
} // fail
453
454
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
455
{
456
457
OutputListNode *item = NULL;
458
if (isfail(ctx))
459
460
return FAIL; // we failed previously, don't go on...
461
char *scratch = get_scratch_buffer(ctx);
462
463
const int indent = ctx->indent;
464
465
if (indent > 0)
memset(scratch, '\t', indent);
466
467
468
va_list ap;
va_start(ap, fmt);
469
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
470
471
va_end(ap);
472
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
473
474
475
if (item == NULL)
return out_of_memory(ctx);
476
item->str = (char *) Malloc(ctx, len + 1);
477
if (item->str == NULL)
478
{
479
Free(ctx, item);
480
return out_of_memory(ctx);
481
482
} // if
483
484
485
// 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().
486
if (len < SCRATCH_BUFFER_SIZE)
487
488
489
strcpy(item->str, scratch); // copy it over.
else
{
490
491
if (indent > 0)
memset(item->str, '\t', indent);
492
va_start(ap, fmt);
493
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
494
495
va_end(ap);
} // else
496
497
item->next = NULL;
498
499
500
ctx->output->tail->next = item;
ctx->output->tail = item;
501
ctx->output_len += len + ctx->endline_len;
502
503
504
505
506
return 0;
} // output_line
507
508
509
510
511
512
513
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
514
// !!! FIXME: this is sort of nasty.
515
516
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
517
518
{
const size_t len = snprintf(buf, bufsize, "%f", f);
519
if ((len+2) >= bufsize)
520
521
522
523
524
525
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
526
527
528
{
if (leavedecimal)
strcat(buf, ".0");
529
return; // done.
530
} // if
531
532
533
534
535
536
537
538
539
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
540
541
if ((leavedecimal) && (end == ptr))
end += 2;
542
543
544
545
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
546
547
548
// Deal with register lists... !!! FIXME: I sort of hate this.
549
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
550
551
552
553
{
while (item != NULL)
{
RegisterList *next = item->next;
554
f(item, d);
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
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
static void reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype, const int regnum)
{
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)
return; // already set, so we're done.
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).
585
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
586
587
588
589
590
591
592
593
594
595
596
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
item->next = prev->next;
prev->next = item;
} // else
} // reglist_insert
597
598
599
static const RegisterList *reglist_exists(const RegisterList *prev,
const RegisterType regtype,
const int regnum)
600
601
{
const uint32 newval = reg_to_ui32(regtype, regnum);
602
const RegisterList *item = prev->next;
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
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.
} // 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);
631
} // get_used_register
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
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
647
// This is used in more than just the d3d profile.
648
static const char *get_D3D_register_string(Context *ctx,
649
RegisterType regtype,
650
651
int regnum, char *regnum_str,
size_t regnum_size)
652
{
653
const char *retval = NULL;
654
int has_number = 1;
655
656
switch (regtype)
657
{
658
case REG_TYPE_TEMP:
659
retval = "r";
660
661
break;
662
case REG_TYPE_INPUT:
663
retval = "v";
664
665
break;
666
case REG_TYPE_CONST:
667
668
669
retval = "c";
break;
670
case REG_TYPE_CONST2:
671
672
673
674
retval = "c";
regnum += 2048;
break;
675
case REG_TYPE_CONST3:
676
677
678
679
retval = "c";
regnum += 4096;
break;
680
case REG_TYPE_CONST4:
681
retval = "c";
682
regnum += 6144;
683
684
break;
685
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
686
retval = (ctx->shader_type == MOJOSHADER_TYPE_VERTEX) ? "a" : "t";
687
688
break;
689
case REG_TYPE_RASTOUT:
690
switch ((RastOutType) regnum)
691
{
692
693
694
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
695
} // switch
696
has_number = 0;
697
698
break;
699
case REG_TYPE_ATTROUT:
700
retval = "oD";
701
702
break;
703
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
704
if ((ctx->shader_type==MOJOSHADER_TYPE_VERTEX) && (ctx->major_ver>=3))
705
retval = "o";
706
else
707
retval = "oT";
708
709
break;
710
case REG_TYPE_CONSTINT:
711
retval = "i";
712
713
break;
714
case REG_TYPE_COLOROUT:
715
retval = "oC";
716
717
break;
718
case REG_TYPE_DEPTHOUT:
719
retval = "oDepth";
720
has_number = 0;
721
722
break;
723
case REG_TYPE_SAMPLER:
724
retval = "s";
725
726
break;
727
case REG_TYPE_CONSTBOOL:
728
retval = "b";
729
730
break;
731
case REG_TYPE_LOOP:
732
retval = "aL";
733
has_number = 0;
734
735
736
break;
// !!! FIXME: don't know what the asm string is for this..
737
case REG_TYPE_TEMPFLOAT16:
738
fail(ctx, "don't know the ASM for tempfloat16 register");
739
740
741
retval = "???";
has_number = 0;
break;
742
743
case REG_TYPE_MISCTYPE:
744
switch ((MiscTypeType) regnum)
745
{
746
747
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
748
} // switch
749
has_number = 0;
750
751
break;
752
case REG_TYPE_LABEL:
753
retval = "l";
754
755
break;
756
case REG_TYPE_PREDICATE:
757
retval = "p";
758
759
760
break;
} // switch
761
762
763
764
765
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
766
767
768
769
return retval;
} // get_D3D_register_string
770
771
772
773
774
775
776
777
778
779
#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,
780
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
781
782
783
784
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
785
return "";
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
} // if
const DestArgInfo *arg = &ctx->dest_args[idx];
const char *result_shift_str = "";
switch (arg->result_shift)
{
case 0x1: result_shift_str = "_x2"; break;
case 0x2: result_shift_str = "_x4"; break;
case 0x3: result_shift_str = "_x8"; break;
case 0xD: result_shift_str = "_d8"; break;
case 0xE: result_shift_str = "_d4"; break;
case 0xF: result_shift_str = "_d2"; break;
} // switch
const char *sat_str = (arg->result_mod & MOD_SATURATE) ? "_sat" : "";
const char *pp_str = (arg->result_mod & MOD_PP) ? "_pp" : "";
const char *cent_str = (arg->result_mod & MOD_CENTROID) ? "_centroid" : "";
char regnum_str[16];
806
807
808
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
809
810
811
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
812
return "";
813
814
} // if
815
816
817
818
819
820
821
822
823
824
825
826
char writemask_str[6];
int i = 0;
if (arg->writemask != 0xF) // 0xF == 1111. No explicit mask.
{
writemask_str[i++] = '.';
if (arg->writemask0) writemask_str[i++] = 'x';
if (arg->writemask1) writemask_str[i++] = 'y';
if (arg->writemask2) writemask_str[i++] = 'z';
if (arg->writemask3) writemask_str[i++] = 'w';
} // if
writemask_str[i] = '\0';
assert(i < sizeof (writemask_str));
827
828
// may turn out something like "_x2_sat_pp_centroid r0.xyzw" ...
829
char *retval = get_scratch_buffer(ctx);
830
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s %s%s%s",
831
result_shift_str, sat_str, pp_str, cent_str,
832
regtype_str, regnum_str, writemask_str);
833
// !!! FIXME: make sure the scratch buffer was large enough.
834
return retval;
835
} // make_D3D_destarg_string
836
837
838
static const char *make_D3D_sourcearg_string(Context *ctx, const int idx)
839
{
840
841
842
if (idx >= STATICARRAYLEN(ctx->source_args))
{
fail(ctx, "Too many source args");
843
return "";
844
} // if
845
846
847
848
849
const SourceArgInfo *arg = &ctx->source_args[idx];
const char *premod_str = "";
const char *postmod_str = "";
850
switch ((SourceMod) arg->src_mod)
851
852
853
854
855
856
857
858
859
860
861
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
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;
899
900
901
902
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
903
904
905
906
} // switch
char regnum_str[16];
907
908
909
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
910
911
912
913
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
914
return "";
915
916
917
918
} // if
char swizzle_str[6];
int i = 0;
919
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
920
921
922
{
static const char channel[] = { 'x', 'y', 'z', 'w' };
swizzle_str[i++] = '.';
923
924
925
926
927
928
929
930
swizzle_str[i++] = channel[arg->swizzle_x];
swizzle_str[i++] = channel[arg->swizzle_y];
swizzle_str[i++] = channel[arg->swizzle_z];
swizzle_str[i++] = channel[arg->swizzle_w];
// .xyzz is the same as .xyz, .z is the same as .zzzz, etc.
while (swizzle_str[i-1] == swizzle_str[i-2])
i--;
931
932
933
934
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
935
char *retval = get_scratch_buffer(ctx);
936
937
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str, swizzle_str);
938
// !!! FIXME: make sure the scratch buffer was large enough.
939
return retval;
940
} // make_D3D_sourcearg_string
941
942
943
static void emit_D3D_start(Context *ctx)
944
945
946
{
const uint major = (uint) ctx->major_ver;
const uint minor = (uint) ctx->minor_ver;
947
948
949
950
951
952
953
954
955
const char *shadertype_str = NULL;
char minor_str[16];
if (minor == 0xFF)
strcpy(minor_str, "sw");
else if (minor == 0x1) // apparently this is "vs_2_x". Weird.
strcpy(minor_str, "x");
else
snprintf(minor_str, sizeof (minor_str), "%u", (uint) minor);
956
957
if (ctx->shader_type == MOJOSHADER_TYPE_PIXEL)
958
shadertype_str = "ps";
959
else if (ctx->shader_type == MOJOSHADER_TYPE_VERTEX)
960
shadertype_str = "vs";
961
else
962
963
964
{
failf(ctx, "Shader type %u unsupported in this profile.",
(uint) ctx->shader_type);
965
return;
966
} // else
967
968
output_line(ctx, "%s_%u_%s", shadertype_str, major, minor_str);
969
970
} // emit_D3D_start
971
972
static void emit_D3D_end(Context *ctx)
973
{
974
output_line(ctx, "end");
975
} // emit_D3D_end
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
static void emit_D3D_finalize(Context *ctx)
{
// no-op.
} // emit_D3D_finalize
static void emit_D3D_global(Context *ctx, RegisterType regtype, int regnum)
{
// no-op.
} // emit_D3D_global
static void emit_D3D_uniform(Context *ctx, RegisterType regtype, int regnum)
{
// no-op.
} // emit_D3D_uniform
996
static void emit_D3D_comment(Context *ctx, const char *str)
997
{
998
999
output_line(ctx, "; %s", str);
} // emit_D3D_comment
1000