/
mojoshader.c
4139 lines (3472 loc) · 129 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
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
43
44
45
46
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
47
48
49
50
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
51
52
53
54
55
56
// 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;
57
typedef int32_t int32;
58
59
60
61
62
63
64
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
65
66
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
67
68
69
70
71
72
73
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
74
75
76
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
// 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
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
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;
122
123
// predeclare.
124
typedef struct Context Context;
125
126
// one emit function for each opcode in each profile.
127
typedef void (*emit_function)(Context *ctx);
128
129
// one emit function for comments in each profile.
130
typedef void (*emit_comment)(Context *ctx, const char *str);
131
132
// one emit function for starting output in each profile.
133
typedef void (*emit_start)(Context *ctx);
134
135
// one emit function for ending output in each profile.
136
typedef void (*emit_end)(Context *ctx);
137
138
139
140
141
142
143
144
145
146
// 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);
147
148
149
150
// 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);
151
152
153
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
154
// one state function for each opcode where we have state machine updates.
155
typedef void (*state_function)(Context *ctx);
156
157
158
159
160
161
162
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
emit_comment comment_emitter;
163
164
emit_global global_emitter;
emit_uniform uniform_emitter;
165
emit_attribute attribute_emitter;
166
emit_finalize finalize_emitter;
167
} Profile;
168
169
170
171
172
173
174
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
175
} RastOutType;
176
177
178
179
180
181
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
182
} MiscTypeType;
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
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
212
213
214
MOJOSHADER_usage usage;
int index;
int writemask;
215
216
217
218
struct RegisterList *next;
} RegisterList;
219
220
221
222
223
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
// 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
242
} SourceMod;
243
244
245
246
typedef struct
{
247
const uint32 *token; // this is the unmolested token in the stream.
248
249
int regnum;
int relative;
250
251
252
253
254
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
255
256
int result_mod;
int result_shift;
257
RegisterType regtype;
258
259
260
261
} DestArgInfo;
typedef struct
{
262
const uint32 *token; // this is the unmolested token in the stream.
263
264
int regnum;
int relative;
265
int swizzle; // xyzw (all four, not split out).
266
267
268
269
270
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
int src_mod;
271
RegisterType regtype;
272
273
274
} SourceArgInfo;
275
276
typedef enum
{
277
// Specific to GLSL profile...
278
279
280
CTX_FLAGS_GLSL_LIT_OPCODE = (1 << 0),
CTX_FLAGS_GLSL_DST_OPCODE = (1 << 1),
CTX_FLAGS_GLSL_LRP_OPCODE = (1 << 2),
281
282
283
284
CTX_FLAGS_MASK = 0xFFFFFFFF
} ContextFlags;
285
286
#define SCRATCH_BUFFER_SIZE 256
#define SCRATCH_BUFFERS 10
287
288
// !!! FIXME: the scratch buffers make Context pretty big.
289
290
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
291
292
// Context...this is state that changes as we parse through a shader...
293
struct Context
294
{
295
296
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
297
void *malloc_data;
298
299
const uint32 *tokens;
uint32 tokencount;
300
301
OutputList *output;
OutputList globals;
302
OutputList helpers;
303
304
OutputList subroutines;
OutputList mainline;
305
306
OutputList ignore;
OutputList *output_stack[2];
307
uint8 *output_bytes; // can be used instead of the OutputLists.
308
309
int indent_stack[2];
int output_stack_len;
310
int output_len; // total strlen; prevents walking the lists just to malloc.
311
int indent;
312
313
314
const char *endline;
int endline_len;
const char *failstr;
315
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
316
317
int scratchidx; // current scratch buffer.
int profileid;
318
const Profile *profile;
319
MOJOSHADER_shaderType shader_type;
320
321
uint8 major_ver;
uint8 minor_ver;
322
DestArgInfo dest_args[1];
323
SourceArgInfo source_args[5];
324
uint32 dwords[4];
325
int instruction_count;
326
uint32 instruction_controls;
327
328
329
uint32 previous_opcode;
ContextFlags flags;
int loops;
330
331
RegisterList used_registers;
RegisterList defined_registers;
332
333
int uniform_count;
RegisterList uniforms;
334
335
int attribute_count;
RegisterList attributes;
336
337
338
};
339
340
341
342
343
344
345
346
347
348
// 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)
{
349
350
351
// check for NULL in case of dumb free() impl.
if (ptr != NULL)
ctx->free(ptr, ctx->malloc_data);
352
353
354
355
} // Free
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
// 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...
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
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
396
397
// Bit arrays (for storing large arrays of booleans...
398
static inline char *get_scratch_buffer(Context *ctx)
399
{
400
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
401
return ctx->scratch[ctx->scratchidx];
402
403
} // get_scratch_buffer
404
405
406
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
407
408
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
409
410
static inline int isfail(const Context *ctx)
411
412
413
414
415
{
return (ctx->failstr != NULL);
} // isfail
416
417
418
419
420
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";
421
422
423
static inline int out_of_memory(Context *ctx)
{
if (ctx->failstr == NULL)
424
ctx->failstr = out_of_mem_str; // fail() would call malloc().
425
426
427
428
429
430
return FAIL;
} // out_of_memory
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
431
{
432
433
if (ctx->failstr == NULL) // don't change existing error.
{
434
char *scratch = get_scratch_buffer(ctx);
435
436
va_list ap;
va_start(ap, fmt);
437
const int len = vsnprintf(scratch,SCRATCH_BUFFER_SIZE,fmt,ap);
438
439
va_end(ap);
440
char *failstr = (char *) Malloc(ctx, len + 1);
441
442
443
444
445
if (failstr == NULL)
out_of_memory(ctx);
else
{
// see comments about scratch buffer overflow in output_line().
446
if (len < SCRATCH_BUFFER_SIZE)
447
448
449
450
451
452
453
454
455
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
456
457
458
459
460
461
} // if
return FAIL;
} // failf
462
static inline int fail(Context *ctx, const char *reason)
463
464
465
466
467
{
return failf(ctx, "%s", reason);
} // fail
468
469
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
470
{
471
472
OutputListNode *item = NULL;
473
if (isfail(ctx))
474
475
return FAIL; // we failed previously, don't go on...
476
char *scratch = get_scratch_buffer(ctx);
477
478
const int indent = ctx->indent;
479
480
if (indent > 0)
memset(scratch, '\t', indent);
481
482
483
va_list ap;
va_start(ap, fmt);
484
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
485
486
va_end(ap);
487
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
488
489
490
if (item == NULL)
return out_of_memory(ctx);
491
item->str = (char *) Malloc(ctx, len + 1);
492
if (item->str == NULL)
493
{
494
Free(ctx, item);
495
return out_of_memory(ctx);
496
497
} // if
498
499
500
// 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().
501
if (len < SCRATCH_BUFFER_SIZE)
502
503
504
strcpy(item->str, scratch); // copy it over.
else
{
505
506
if (indent > 0)
memset(item->str, '\t', indent);
507
va_start(ap, fmt);
508
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
509
510
va_end(ap);
} // else
511
512
item->next = NULL;
513
514
515
ctx->output->tail->next = item;
ctx->output->tail = item;
516
ctx->output_len += len + ctx->endline_len;
517
518
519
520
521
return 0;
} // output_line
522
523
524
525
526
527
528
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
529
// !!! FIXME: this is sort of nasty.
530
531
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
532
533
{
const size_t len = snprintf(buf, bufsize, "%f", f);
534
if ((len+2) >= bufsize)
535
536
537
538
539
540
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
541
542
543
{
if (leavedecimal)
strcat(buf, ".0");
544
return; // done.
545
} // if
546
547
548
549
550
551
552
553
554
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
555
556
if ((leavedecimal) && (end == ptr))
end += 2;
557
558
559
560
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
561
562
563
// Deal with register lists... !!! FIXME: I sort of hate this.
564
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
565
566
567
568
{
while (item != NULL)
{
RegisterList *next = item->next;
569
f(item, d);
570
571
572
573
574
575
576
577
578
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
579
580
581
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
582
583
584
585
586
587
588
{
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)
589
return item; // already set, so we're done.
590
591
592
593
594
595
596
597
598
599
600
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).
601
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
602
603
604
605
606
607
if (item == NULL)
out_of_memory(ctx);
else
{
item->regtype = regtype;
item->regnum = regnum;
608
609
610
item->usage = 0;
item->index = 0;
item->writemask = 0;
611
612
613
item->next = prev->next;
prev->next = item;
} // else
614
615
return item;
616
617
} // reglist_insert
618
619
620
static const RegisterList *reglist_exists(const RegisterList *prev,
const RegisterType regtype,
const int regnum)
621
622
{
const uint32 newval = reg_to_ui32(regtype, regnum);
623
const RegisterList *item = prev->next;
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
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);
652
} // get_used_register
653
654
655
656
657
658
659
660
661
662
663
664
665
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
666
667
668
669
670
671
672
673
674
675
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
676
677
678
679
680
681
682
683
// D3D stuff that's used in more than just the d3d profile...
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
684
685
static const char *get_D3D_register_string(Context *ctx,
686
RegisterType regtype,
687
688
int regnum, char *regnum_str,
size_t regnum_size)
689
{
690
const char *retval = NULL;
691
int has_number = 1;
692
693
switch (regtype)
694
{
695
case REG_TYPE_TEMP:
696
retval = "r";
697
698
break;
699
case REG_TYPE_INPUT:
700
retval = "v";
701
702
break;
703
case REG_TYPE_CONST:
704
705
706
retval = "c";
break;
707
case REG_TYPE_CONST2:
708
709
710
711
retval = "c";
regnum += 2048;
break;
712
case REG_TYPE_CONST3:
713
714
715
716
retval = "c";
regnum += 4096;
break;
717
case REG_TYPE_CONST4:
718
retval = "c";
719
regnum += 6144;
720
721
break;
722
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
723
retval = (ctx->shader_type == MOJOSHADER_TYPE_VERTEX) ? "a" : "t";
724
725
break;
726
case REG_TYPE_RASTOUT:
727
switch ((RastOutType) regnum)
728
{
729
730
731
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
732
} // switch
733
has_number = 0;
734
735
break;
736
case REG_TYPE_ATTROUT:
737
retval = "oD";
738
739
break;
740
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
741
if ((ctx->shader_type==MOJOSHADER_TYPE_VERTEX) && (ctx->major_ver>=3))
742
retval = "o";
743
else
744
retval = "oT";
745
746
break;
747
case REG_TYPE_CONSTINT:
748
retval = "i";
749
750
break;
751
case REG_TYPE_COLOROUT:
752
retval = "oC";
753
754
break;
755
case REG_TYPE_DEPTHOUT:
756
retval = "oDepth";
757
has_number = 0;
758
759
break;
760
case REG_TYPE_SAMPLER:
761
retval = "s";
762
763
break;
764
case REG_TYPE_CONSTBOOL:
765
retval = "b";
766
767
break;
768
case REG_TYPE_LOOP:
769
retval = "aL";
770
has_number = 0;
771
772
773
break;
// !!! FIXME: don't know what the asm string is for this..
774
case REG_TYPE_TEMPFLOAT16:
775
fail(ctx, "don't know the ASM for tempfloat16 register");
776
777
778
retval = "???";
has_number = 0;
break;
779
780
case REG_TYPE_MISCTYPE:
781
switch ((MiscTypeType) regnum)
782
{
783
784
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
785
} // switch
786
has_number = 0;
787
788
break;
789
case REG_TYPE_LABEL:
790
retval = "l";
791
792
break;
793
case REG_TYPE_PREDICATE:
794
retval = "p";
795
796
797
break;
} // switch
798
799
800
801
802
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
803
804
805
806
return retval;
} // get_D3D_register_string
807
808
809
810
811
812
813
814
815
816
#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,
817
static const char *make_D3D_destarg_string(Context *ctx, const int idx)
818
819
820
821
{
if (idx >= STATICARRAYLEN(ctx->dest_args))
{
fail(ctx, "Too many destination args");
822
return "";
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
} // 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];
843
844
845
const char *regtype_str = get_D3D_register_string(ctx, arg->regtype,
arg->regnum, regnum_str,
sizeof (regnum_str));
846
847
848
if (regtype_str == NULL)
{
fail(ctx, "Unknown destination register type.");
849
return "";
850
851
} // if
852
853
854
855
856
857
858
859
860
861
862
863
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));
864
865
// may turn out something like "_x2_sat_pp_centroid r0.xyzw" ...
866
char *retval = get_scratch_buffer(ctx);
867
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s %s%s%s",
868
result_shift_str, sat_str, pp_str, cent_str,
869
regtype_str, regnum_str, writemask_str);
870
// !!! FIXME: make sure the scratch buffer was large enough.
871
return retval;
872
} // make_D3D_destarg_string
873
874
875
static const char *make_D3D_sourcearg_string(Context *ctx, const int idx)
876
{
877
878
879
if (idx >= STATICARRAYLEN(ctx->source_args))
{
fail(ctx, "Too many source args");
880
return "";
881
} // if
882
883
884
885
886
const SourceArgInfo *arg = &ctx->source_args[idx];
const char *premod_str = "";
const char *postmod_str = "";
887
switch ((SourceMod) 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
954
955
} // if
char swizzle_str[6];
int i = 0;
956
if (arg->swizzle != 0xE4) // 0xE4 == 11100100 ... 3 2 1 0. No swizzle.
957
958
959
{
static const char channel[] = { 'x', 'y', 'z', 'w' };
swizzle_str[i++] = '.';
960
961
962
963
964
965
966
967
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--;
968
969
970
971
} // if
swizzle_str[i] = '\0';
assert(i < sizeof (swizzle_str));
972
char *retval = get_scratch_buffer(ctx);
973
974
snprintf(retval, SCRATCH_BUFFER_SIZE, "%s%s%s%s%s",
premod_str, regtype_str, regnum_str, postmod_str, swizzle_str);
975
// !!! FIXME: make sure the scratch buffer was large enough.
976
return retval;
977
} // make_D3D_sourcearg_string
978
979
980
static void emit_D3D_start(Context *ctx)
981
982
983
{
const uint major = (uint) ctx->major_ver;
const uint minor = (uint) ctx->minor_ver;
984
985
986
987
988
989
990
991
992
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);
993
994
if (ctx->shader_type == MOJOSHADER_TYPE_PIXEL)
995
shadertype_str = "ps";
996
else if (ctx->shader_type == MOJOSHADER_TYPE_VERTEX)
997
shadertype_str = "vs";
998
else
999
1000
{
failf(ctx, "Shader type %u unsupported in this profile.",