/
mojoshader.c
3561 lines (2963 loc) · 109 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
// predeclare.
94
typedef struct Context Context;
95
96
// one emit function for each opcode in each profile.
97
typedef void (*emit_function)(Context *ctx);
98
99
// one emit function for comments in each profile.
100
typedef void (*emit_comment)(Context *ctx, const char *str);
101
102
// one emit function for starting output in each profile.
103
typedef void (*emit_start)(Context *ctx);
104
105
// one emit function for ending output in each profile.
106
typedef void (*emit_end)(Context *ctx);
107
108
109
110
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
111
// one state function for each opcode where we have state machine updates.
112
typedef void (*state_function)(Context *ctx);
113
114
115
116
117
118
119
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
emit_comment comment_emitter;
120
} Profile;
121
122
typedef MOJOSHADER_shaderType ShaderType;
123
124
125
typedef enum
{
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
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
149
} RegisterType;
150
151
152
153
154
155
156
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
157
} RastOutType;
158
159
160
161
162
163
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
164
} MiscTypeType;
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
typedef enum
{
DECLUSAGE_POSITION = 0,
DECLUSAGE_BLENDWEIGHT = 1,
DECLUSAGE_BLENDINDICES = 2,
DECLUSAGE_NORMAL = 3,
DECLUSAGE_PSIZE = 4,
DECLUSAGE_TEXCOORD = 5,
DECLUSAGE_TANGENT = 6,
DECLUSAGE_BINORMAL = 7,
DECLUSAGE_TESSFACTOR = 8,
DECLUSAGE_POSITIONT = 9,
DECLUSAGE_COLOR = 10,
DECLUSAGE_FOG = 11,
DECLUSAGE_DEPTH = 12,
DECLUSAGE_SAMPLE = 13
182
} DeclUsageType;
183
184
185
186
187
188
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
189
} TextureType;
190
191
192
193
194
// 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.
195
typedef struct OutputListNode
196
197
{
char *str;
198
199
200
201
202
203
204
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
205
206
} OutputList;
207
208
209
210
211
212
213
214
215
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
struct RegisterList *next;
} RegisterList;
216
217
218
219
220
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
// 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
239
} SourceMod;
240
241
242
243
typedef struct
{
244
const uint32 *token; // this is the unmolested token in the stream.
245
246
int regnum;
int relative;
247
248
249
250
251
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
252
253
int result_mod;
int result_shift;
254
RegisterType regtype;
255
256
257
258
} DestArgInfo;
typedef struct
{
259
const uint32 *token; // this is the unmolested token in the stream.
260
261
int regnum;
int relative;
262
int swizzle; // xyzw (all four, not split out).
263
264
265
266
267
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
int src_mod;
268
RegisterType regtype;
269
270
271
} SourceArgInfo;
272
273
typedef enum
{
274
// Specific to GLSL profile...
275
276
277
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
CTX_FLAGS_GLSL_DST_OPCODE = (1 << 1),
CTX_FLAGS_GLSL_LRP_OPCODE = (1 << 2),
278
279
280
281
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
282
283
#define SCRATCH_BUFFER_SIZE 256
#define SCRATCH_BUFFERS 10
284
285
// !!! FIXME: the scratch buffers make Context pretty big.
286
287
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
288
289
// Context...this is state that changes as we parse through a shader...
290
struct Context
291
{
292
293
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
294
295
const uint32 *tokens;
uint32 tokencount;
296
297
OutputList *output;
OutputList globals;
298
OutputList helpers;
299
300
OutputList subroutines;
OutputList mainline;
301
302
303
304
OutputList ignore;
OutputList *output_stack[2];
int indent_stack[2];
int output_stack_len;
305
int output_len; // total strlen; prevents walking the lists just to malloc.
306
int indent;
307
308
309
const char *endline;
int endline_len;
const char *failstr;
310
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
311
312
int scratchidx; // current scratch buffer.
int profileid;
313
314
const Profile *profile;
ShaderType shader_type;
315
316
uint8 major_ver;
uint8 minor_ver;
317
DestArgInfo dest_args[1];
318
SourceArgInfo source_args[5];
319
uint32 dwords[4];
320
int instruction_count;
321
uint32 instruction_controls;
322
323
324
uint32 previous_opcode;
ContextFlags flags;
int loops;
325
326
RegisterList used_registers;
RegisterList defined_registers;
327
328
329
};
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
// 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...
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
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
370
371
// Bit arrays (for storing large arrays of booleans...
372
static inline char *get_scratch_buffer(Context *ctx)
373
{
374
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
375
return ctx->scratch[ctx->scratchidx];
376
377
} // get_scratch_buffer
378
379
380
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
381
382
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
383
384
static inline int isfail(const Context *ctx)
385
386
387
388
389
{
return (ctx->failstr != NULL);
} // isfail
390
391
392
393
394
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";
395
396
397
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
398
ctx->failstr = out_of_mem_str; // fail() would call malloc().
399
400
401
402
403
404
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
405
{
406
407
if (ctx->failstr == NULL) // don't change existing error.
{
408
char *scratch = get_scratch_buffer(ctx);
409
410
va_list ap;
va_start(ap, fmt);
411
const int len = vsnprintf(scratch,SCRATCH_BUFFER_SIZE,fmt,ap);
412
413
va_end(ap);
414
415
416
417
418
419
char *failstr = (char *) ctx->malloc(len + 1);
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
420
if (len < SCRATCH_BUFFER_SIZE)
421
422
423
424
425
426
427
428
429
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
430
431
432
433
434
435
} // if
return FAIL;
} // failf
436
static inline int fail(Context *ctx, const char *reason)
437
438
439
440
441
{
return failf(ctx, "%s", reason);
} // fail
442
443
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
444
{
445
446
OutputListNode *item = NULL;
447
if (isfail(ctx))
448
449
return FAIL; // we failed previously, don't go on...
450
char *scratch = get_scratch_buffer(ctx);
451
452
const int indent = ctx->indent;
453
454
if (indent > 0)
memset(scratch, '\t', indent);
455
456
457
va_list ap;
va_start(ap, fmt);
458
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
459
460
va_end(ap);
461
462
463
464
item = (OutputListNode *) ctx->malloc(sizeof (OutputListNode));
if (item == NULL)
return out_of_memory(ctx);
465
466
item->str = (char *) ctx->malloc(len + 1);
if (item->str == NULL)
467
{
468
469
free(item);
return out_of_memory(ctx);
470
471
} // if
472
473
474
// 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().
475
if (len < SCRATCH_BUFFER_SIZE)
476
477
478
strcpy(item->str, scratch); // copy it over.
else
{
479
480
if (indent > 0)
memset(item->str, '\t', indent);
481
va_start(ap, fmt);
482
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
483
484
va_end(ap);
} // else
485
486
item->next = NULL;
487
488
489
ctx->output->tail->next = item;
ctx->output->tail = item;
490
ctx->output_len += len + ctx->endline_len;
491
492
493
494
495
return 0;
} // output_line
496
497
498
499
500
501
502
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
503
// !!! FIXME: this is sort of nasty.
504
505
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
506
507
{
const size_t len = snprintf(buf, bufsize, "%f", f);
508
if ((len+2) >= bufsize)
509
510
511
512
513
514
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
515
516
517
{
if (leavedecimal)
strcat(buf, ".0");
518
return; // done.
519
} // if
520
521
522
523
524
525
526
527
528
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
529
530
if ((leavedecimal) && (end == ptr))
end += 2;
531
532
533
534
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
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
585
// Deal with register lists... !!! FIXME: I sort of hate this.
static void free_reglist(MOJOSHADER_free f, RegisterList *item)
{
while (item != NULL)
{
RegisterList *next = item->next;
f(item);
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).
item = (RegisterList *) ctx->malloc(sizeof (RegisterList));
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
item->next = prev->next;
prev->next = item;
} // else
} // reglist_insert
586
587
588
static const RegisterList *reglist_exists(const RegisterList *prev,
const RegisterType regtype,
const int regnum)
589
590
{
const uint32 newval = reg_to_ui32(regtype, regnum);
591
const RegisterList *item = prev->next;
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
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);
620
} // get_used_register
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
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
636
// This is used in more than just the d3d profile.
637
static const char *get_D3D_register_string(Context *ctx,
638
RegisterType regtype,
639
640
int regnum, char *regnum_str,
size_t regnum_size)
641
{
642
const char *retval = NULL;
643
int has_number = 1;
644
645
switch (regtype)
646
{
647
case REG_TYPE_TEMP:
648
retval = "r";
649
650
break;
651
case REG_TYPE_INPUT:
652
retval = "v";
653
654
break;
655
case REG_TYPE_CONST:
656
657
658
retval = "c";
break;
659
case REG_TYPE_CONST2:
660
661
662
663
retval = "c";
regnum += 2048;
break;
664
case REG_TYPE_CONST3:
665
666
667
668
retval = "c";
regnum += 4096;
break;
669
case REG_TYPE_CONST4:
670
retval = "c";
671
regnum += 6144;
672
673
break;
674
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
675
retval = (ctx->shader_type == MOJOSHADER_TYPE_VERTEX) ? "a" : "t";
676
677
break;
678
case REG_TYPE_RASTOUT:
679
switch ((RastOutType) regnum)
680
{
681
682
683
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
684
} // switch
685
has_number = 0;
686
687
break;
688
case REG_TYPE_ATTROUT:
689
retval = "oD";
690
691
break;
692
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
693
if ((ctx->shader_type==MOJOSHADER_TYPE_VERTEX) && (ctx->major_ver>=3))
694
retval = "o";
695
else
696
retval = "oT";
697
698
break;
699
case REG_TYPE_CONSTINT:
700
retval = "i";
701
702
break;
703
case REG_TYPE_COLOROUT:
704
retval = "oC";
705
706
break;
707
case REG_TYPE_DEPTHOUT:
708
retval = "oDepth";
709
has_number = 0;
710
711
break;
712
case REG_TYPE_SAMPLER:
713
retval = "s";
714
715
break;
716
case REG_TYPE_CONSTBOOL:
717
retval = "b";
718
719
break;
720
case REG_TYPE_LOOP:
721
retval = "aL";
722
has_number = 0;
723
724
725
break;
// !!! FIXME: don't know what the asm string is for this..
726
case REG_TYPE_TEMPFLOAT16:
727
fail(ctx, "don't know the ASM for tempfloat16 register");
728
729
730
retval = "???";
has_number = 0;
break;
731
732
case REG_TYPE_MISCTYPE:
733
switch ((MiscTypeType) regnum)
734
{
735
736
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
737
} // switch
738
has_number = 0;
739
740
break;
741
case REG_TYPE_LABEL:
742
retval = "l";
743
744
break;
745
case REG_TYPE_PREDICATE:
746
retval = "p";
747
748
749
break;
} // switch
750
751
752
753
754
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
755
756
757
758
return retval;
} // get_D3D_register_string
759
760
761
762
763
764
765
766
767
768
#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,
769
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
770
771
772
773
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
774
return "";
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
} // 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];
795
796
797
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
798
799
800
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
801
return "";
802
803
} // if
804
805
806
807
808
809
810
811
812
813
814
815
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));
816
817
// may turn out something like "_x2_sat_pp_centroid r0.xyzw" ...
818
char *retval = get_scratch_buffer(ctx);
819
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s %s%s%s",
820
result_shift_str, sat_str, pp_str, cent_str,
821
regtype_str, regnum_str, writemask_str);
822
// !!! FIXME: make sure the scratch buffer was large enough.
823
return retval;
824
} // make_D3D_destarg_string
825
826
827
static const char *make_D3D_sourcearg_string(Context *ctx, const int idx)
828
{
829
830
831
if (idx >= STATICARRAYLEN(ctx->source_args))
{
fail(ctx, "Too many source args");
832
return "";
833
} // if
834
835
836
837
838
const SourceArgInfo *arg = &ctx->source_args[idx];
const char *premod_str = "";
const char *postmod_str = "";
839
switch ((SourceMod) arg->src_mod)
840
841
842
843
844
845
846
847
848
849
850
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS:
postmod_str = "_bias";
break;
851
852
853
854
855
856
857
858
859
860
861
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
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;
888
889
890
891
case SRCMOD_NONE:
case SRCMOD_TOTAL:
break; // stop compiler whining.
892
893
894
895
} // switch
char regnum_str[16];
896
897
898
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
899
900
901
902
if (regtype_str == NULL)
{
fail(ctx, "Unknown source register type.");
903
return "";
904
905
906
907
} // if
char swizzle_str[6];
int i = 0;
908
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
909
910
911
{
static const char channel[] = { 'x', 'y', 'z', 'w' };
swizzle_str[i++] = '.';
912
913
914
915
916
917
918
919
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--;
920
921
922
923
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
924
char *retval = get_scratch_buffer(ctx);
925
926
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str, swizzle_str);
927
// !!! FIXME: make sure the scratch buffer was large enough.
928
return retval;
929
} // make_D3D_sourcearg_string
930
931
932
static void emit_D3D_start(Context *ctx)
933
934
935
{
const uint major = (uint) ctx->major_ver;
const uint minor = (uint) ctx->minor_ver;
936
937
938
939
940
941
942
943
944
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);
945
946
if (ctx->shader_type == MOJOSHADER_TYPE_PIXEL)
947
shadertype_str = "ps";
948
else if (ctx->shader_type == MOJOSHADER_TYPE_VERTEX)
949
shadertype_str = "vs";
950
else
951
952
953
{
failf(ctx, "Shader type %u unsupported in this profile.",
(uint) ctx->shader_type);
954
return;
955
} // else
956
957
output_line(ctx, "%s_%u_%s", shadertype_str, major, minor_str);
958
959
} // emit_D3D_start
960
961
static void emit_D3D_end(Context *ctx)
962
{
963
output_line(ctx, "end");
964
} // emit_D3D_end
965
966
967
static void emit_D3D_comment(Context *ctx, const char *str)
968
{
969
970
output_line(ctx, "; %s", str);
} // emit_D3D_comment
971
972
973
static void emit_D3D_RESERVED(Context *ctx)
974
{
975
976
977
// do nothing; fails in the state machine.
} // emit_D3D_RESERVED
978
979
980
981
// Generic D3D opcode emitters. A list of macros generate all the entry points
// that call into these...
982
983
984
985
986
987
988
989
990
991
992
993
static char *lowercase(char *dst, const char *src)
{
int i = 0;
do
{
const char ch = src[i];
dst[i] = (((ch >= 'A') && (ch <= 'Z')) ? (ch - ('A' - 'a')) : ch);
} while (src[i++]);
return dst;
} // lowercase
994
static void emit_D3D_opcode_d(Context *ctx, const char *opcode)
995
{
996
const char *dst0 = make_D3D_destarg_string(ctx, 0);
997
opcode = lowercase(get_scratch_buffer(ctx), opcode);
998
output_line(ctx, "%s%s", opcode, dst0);
999
1000
} // emit_D3D_opcode_d