/
mojoshader.c
6748 lines (5702 loc) · 217 KB
1
/**
2
3
* MojoShader; generate shader programs from bytecode of compiled
* Direct3D shaders.
4
5
6
7
8
9
*
* Please see the file LICENSE.txt in the source's root directory.
*
* This file written by Ryan C. Gordon.
*/
10
11
// !!! FIXME: this file really needs to be split up.
12
13
// !!! FIXME: I keep changing coding styles for symbols and typedefs.
14
15
16
// Shader bytecode format is described at MSDN:
// http://msdn2.microsoft.com/en-us/library/ms800307.aspx
17
#include <stdio.h>
18
#include <string.h>
19
#include <stdlib.h>
20
#include <stdarg.h>
21
#include <assert.h>
22
23
#include "mojoshader.h"
24
25
26
27
28
29
30
// This is the highest shader version we currently support.
#define MAX_SHADER_MAJOR 3
#define MAX_SHADER_MINOR 0
31
32
33
// 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.
34
35
36
37
38
#ifndef SUPPORT_PROFILE_D3D
#define SUPPORT_PROFILE_D3D 1
#endif
39
40
41
42
#ifndef SUPPORT_PROFILE_PASSTHROUGH
#define SUPPORT_PROFILE_PASSTHROUGH 1
#endif
43
44
45
46
#ifndef SUPPORT_PROFILE_GLSL
#define SUPPORT_PROFILE_GLSL 1
#endif
47
48
49
50
#ifndef SUPPORT_PROFILE_ARB1
#define SUPPORT_PROFILE_ARB1 1
#endif
51
52
53
54
55
56
57
58
59
60
#ifndef SUPPORT_PROFILE_NV2
#define SUPPORT_PROFILE_NV2 1
#endif
#if SUPPORT_PROFILE_NV2 && !SUPPORT_PROFILE_ARB1
#error nv2 profile requires arb1 profile.
#endif
61
62
63
64
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
65
66
67
68
69
70
#ifdef _MSC_VER
#define snprintf _snprintf
typedef unsigned __int8 uint8;
typedef unsigned __int32 uint32;
typedef unsigned __int32 int32;
71
72
73
// Warning Level 4 considered harmful. :)
#pragma warning(disable: 4100) // "unreferenced formal parameter"
#pragma warning(disable: 4389) // "signed/unsigned mismatch"
74
75
#else
#include <stdint.h>
76
77
typedef uint8_t uint8;
typedef uint32_t uint32;
78
typedef int32_t int32;
79
#endif
80
81
82
83
84
85
86
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
87
88
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
89
90
91
92
93
94
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
95
96
97
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
98
99
100
101
102
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
// 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
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
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;
148
149
150
151
152
153
154
155
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
156
// predeclare.
157
typedef struct Context Context;
158
159
// one emit function for each opcode in each profile.
160
typedef void (*emit_function)(Context *ctx);
161
162
// one emit function for starting output in each profile.
163
typedef void (*emit_start)(Context *ctx, const char *profilestr);
164
165
// one emit function for ending output in each profile.
166
typedef void (*emit_end)(Context *ctx);
167
168
169
170
171
172
173
// 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);
174
175
176
// one emit function for relative uniform arrays in each profile.
typedef void (*emit_relative)(Context *ctx, int size);
177
178
179
// one emit function for uniforms in each profile.
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum);
180
181
182
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
183
184
185
186
// 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);
187
188
189
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
190
// one state function for each opcode where we have state machine updates.
191
typedef void (*state_function)(Context *ctx);
192
193
194
195
196
197
198
// one function for varnames in each profile.
typedef const char *(*varname_function)(Context *c, RegisterType t, int num);
// one function for const var array in each profile.
typedef const char *(*const_array_varname_function)(Context *c);
199
200
201
202
203
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
204
emit_global global_emitter;
205
emit_relative relative_emitter;
206
emit_uniform uniform_emitter;
207
emit_sampler sampler_emitter;
208
emit_attribute attribute_emitter;
209
emit_finalize finalize_emitter;
210
211
varname_function get_varname;
const_array_varname_function get_const_array_varname;
212
} Profile;
213
214
215
216
217
218
219
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
220
} RastOutType;
221
222
223
224
225
226
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
227
} MiscTypeType;
228
229
230
231
232
// 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.
233
typedef struct OutputListNode
234
235
{
char *str;
236
237
238
239
240
241
242
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
243
244
} OutputList;
245
246
247
248
249
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
250
251
252
MOJOSHADER_usage usage;
int index;
int writemask;
253
int misc;
254
255
256
struct RegisterList *next;
} RegisterList;
257
258
259
260
261
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
262
263
264
265
266
267
// result modifiers.
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
// 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
286
} SourceMod;
287
288
289
290
typedef struct
{
291
const uint32 *token; // this is the unmolested token in the stream.
292
293
int regnum;
int relative;
294
295
296
297
298
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
299
int orig_writemask; // writemask before mojoshader tweaks it.
300
301
int result_mod;
int result_shift;
302
RegisterType regtype;
303
304
305
306
} DestArgInfo;
typedef struct
{
307
const uint32 *token; // this is the unmolested token in the stream.
308
int regnum;
309
int swizzle; // xyzw (all four, not split out).
310
311
312
313
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
314
SourceMod src_mod;
315
RegisterType regtype;
316
317
318
319
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
320
321
322
} SourceArgInfo;
323
324
#define SCRATCH_BUFFER_SIZE 128
325
#define SCRATCH_BUFFERS 32
326
327
// !!! FIXME: the scratch buffers make Context pretty big.
328
329
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
330
331
// Context...this is state that changes as we parse through a shader...
332
struct Context
333
{
334
335
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
336
void *malloc_data;
337
338
const uint32 *tokens;
uint32 tokencount;
339
340
OutputList *output;
OutputList globals;
341
OutputList helpers;
342
OutputList subroutines;
343
OutputList mainline_intro;
344
OutputList mainline;
345
346
OutputList ignore;
OutputList *output_stack[2];
347
uint8 *output_bytes; // can be used instead of the OutputLists.
348
349
int indent_stack[2];
int output_stack_len;
350
int output_len; // total strlen; prevents walking the lists just to malloc.
351
int indent;
352
const char *shader_type_str;
353
354
355
const char *endline;
int endline_len;
const char *failstr;
356
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
357
358
int scratchidx; // current scratch buffer.
int profileid;
359
const Profile *profile;
360
MOJOSHADER_shaderType shader_type;
361
362
uint8 major_ver;
uint8 minor_ver;
363
DestArgInfo dest_arg;
364
SourceArgInfo source_args[5];
365
SourceArgInfo predicate_arg; // for predicated instructions.
366
uint32 dwords[4];
367
int instruction_count;
368
uint32 instruction_controls;
369
uint32 previous_opcode;
370
int max_constreg;
371
int loops;
372
int reps;
373
int cmps;
374
375
int scratch_registers;
int max_scratch_registers;
376
int assigned_vertex_attributes;
377
int last_address_reg_component;
378
379
RegisterList used_registers;
RegisterList defined_registers;
380
381
int constant_count;
ConstantsList *constants;
382
383
int uniform_count;
RegisterList uniforms;
384
385
int attribute_count;
RegisterList attributes;
386
387
int sampler_count;
RegisterList samplers;
388
389
390
391
int predicated:1;
int uniform_array:1;
int support_nv2:1;
int glsl_generated_lit_opcode:1;
392
393
394
};
395
396
// Convenience functions for allocators...
397
398
399
400
401
402
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)
403
{
404
405
406
407
408
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
409
static inline void *Malloc(Context *ctx, const size_t len)
410
{
411
void *retval = ctx->malloc((int) len, ctx->malloc_data);
412
413
414
if (retval == NULL)
out_of_memory(ctx);
return retval;
415
416
417
} // Malloc
418
static inline void Free(Context *ctx, void *ptr)
419
{
420
if (ptr != NULL) // check for NULL in case of dumb free() impl.
421
ctx->free(ptr, ctx->malloc_data);
422
423
424
} // Free
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
// 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...
449
450
451
452
453
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
454
static inline int shader_version_supported(const uint8 maj, const uint8 min)
455
456
457
458
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
459
460
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
461
462
463
464
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
465
466
467
468
469
470
471
472
473
474
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
475
476
477
478
479
480
static inline int isfail(const Context *ctx)
{
return (ctx->failstr != NULL);
} // isfail
481
482
static inline char *get_scratch_buffer(Context *ctx)
483
{
484
485
486
487
488
489
490
491
492
493
494
495
if ((ctx->scratchidx >= SCRATCH_BUFFERS) && !isfail(ctx))
{
// can't call fail() here, since it calls back into here.
const char *errstr = "BUG: overflowed scratch buffers";
char *failstr = (char *) Malloc(ctx, strlen(errstr) + 1);
if (failstr != NULL)
{
strcpy(failstr, errstr);
ctx->failstr = failstr;
} // if
} // if
496
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
497
return ctx->scratch[ctx->scratchidx];
498
499
} // get_scratch_buffer
500
501
502
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
503
{
504
505
if (ctx->failstr == NULL) // don't change existing error.
{
506
char *scratch = get_scratch_buffer(ctx);
507
508
va_list ap;
va_start(ap, fmt);
509
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
510
511
va_end(ap);
512
char *failstr = (char *) Malloc(ctx, len + 1);
513
if (failstr != NULL)
514
515
{
// see comments about scratch buffer overflow in output_line().
516
if (len < SCRATCH_BUFFER_SIZE)
517
518
519
520
521
522
523
524
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;
525
} // if
526
527
528
529
530
531
} // if
return FAIL;
} // failf
532
static inline int fail(Context *ctx, const char *reason)
533
534
535
536
537
{
return failf(ctx, "%s", reason);
} // fail
538
539
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
540
{
541
542
OutputListNode *item = NULL;
543
if (isfail(ctx))
544
545
return FAIL; // we failed previously, don't go on...
546
char *scratch = get_scratch_buffer(ctx);
547
548
const int indent = ctx->indent;
549
550
if (indent > 0)
memset(scratch, '\t', indent);
551
552
553
va_list ap;
va_start(ap, fmt);
554
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
555
556
va_end(ap);
557
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
558
if (item == NULL)
559
return FAIL;
560
561
item->str = (char *) Malloc(ctx, len + 1);
562
if (item->str == NULL)
563
{
564
Free(ctx, item);
565
return FAIL;
566
567
} // if
568
569
570
// 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().
571
if (len < SCRATCH_BUFFER_SIZE)
572
573
574
strcpy(item->str, scratch); // copy it over.
else
{
575
576
if (indent > 0)
memset(item->str, '\t', indent);
577
va_start(ap, fmt);
578
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
579
580
va_end(ap);
} // else
581
582
item->next = NULL;
583
584
585
ctx->output->tail->next = item;
ctx->output->tail = item;
586
ctx->output_len += len + ctx->endline_len;
587
588
589
590
591
return 0;
} // output_line
592
593
594
595
596
597
598
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
599
// !!! FIXME: this is sort of nasty.
600
601
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
602
603
{
const size_t len = snprintf(buf, bufsize, "%f", f);
604
if ((len+2) >= bufsize)
605
606
607
608
609
610
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
611
612
613
{
if (leavedecimal)
strcat(buf, ".0");
614
return; // done.
615
} // if
616
617
618
619
620
621
622
623
624
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
625
626
if ((leavedecimal) && (end == ptr))
end += 2;
627
628
629
630
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
631
632
633
// Deal with register lists... !!! FIXME: I sort of hate this.
634
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
635
636
637
638
{
while (item != NULL)
{
RegisterList *next = item->next;
639
f(item, d);
640
641
642
643
644
645
646
647
648
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
649
650
651
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
652
653
654
655
656
657
658
{
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)
659
return item; // already set, so we're done.
660
661
662
663
664
665
666
667
668
669
670
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).
671
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
672
if (item != NULL)
673
674
675
{
item->regtype = regtype;
item->regnum = regnum;
676
item->usage = MOJOSHADER_USAGE_UNKNOWN;
677
678
item->index = 0;
item->writemask = 0;
679
item->misc = 0;
680
681
item->next = prev->next;
prev->next = item;
682
} // if
683
684
return item;
685
686
} // reglist_insert
687
688
static RegisterList *reglist_find(RegisterList *prev, const RegisterType rtype,
const int regnum)
689
{
690
691
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
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.
708
709
710
711
712
713
714
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
715
716
717
718
719
720
} // reglist_exists
static inline void set_used_register(Context *ctx, const RegisterType regtype,
const int regnum)
{
reglist_insert(ctx, &ctx->used_registers, regtype, regnum);
721
722
723
if ((regtype == REG_TYPE_CONST) && (ctx->max_constreg < regnum))
ctx->max_constreg = regnum;
724
725
726
727
728
729
} // 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);
730
} // get_used_register
731
732
733
734
735
736
737
738
739
740
741
742
743
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
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
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
759
760
761
762
763
764
765
766
767
768
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
769
770
771
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
772
// !!! FIXME: make sure it doesn't exist?
773
774
775
776
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
static inline int writemask_xyzw(const int writemask)
{
return (writemask == 0xF); // 0xF == 1111. No explicit mask (full!).
} // writemask_xyzw
static inline int writemask_xyz(const int writemask)
{
return (writemask == 0x7); // 0x7 == 0111. (that is: xyz)
} // writemask_xyz
static inline int writemask_xy(const int writemask)
{
return (writemask == 0x3); // 0x3 == 0011. (that is: xy)
} // writemask_xy
static inline int writemask_x(const int writemask)
{
return (writemask == 0x1); // 0x1 == 0001. (that is: x)
} // writemask_x
static inline int writemask_y(const int writemask)
{
return (writemask == 0x2); // 0x1 == 0010. (that is: y)
} // writemask_y
808
809
810
811
812
813
814
815
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
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
static inline int scalar_register(const RegisterType regtype, const int regnum)
{
switch (regtype)
{
case REG_TYPE_DEPTHOUT:
case REG_TYPE_CONSTBOOL:
case REG_TYPE_PREDICATE:
case REG_TYPE_LOOP:
return 1;
case REG_TYPE_MISCTYPE:
if ( ((const MiscTypeType) regnum) == MISCTYPE_TYPE_FACE )
return 1;
return 0;
default: break;
} // switch
return 0;
} // scalar_register
838
839
static inline int no_swizzle(const int swizzle)
{
840
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
841
842
843
} // no_swizzle
844
845
846
847
848
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
849
850
851
852
853
854
855
856
static int allocate_scratch_register(Context *ctx)
{
const int retval = ctx->scratch_registers++;
if (retval >= ctx->max_scratch_registers)
ctx->max_scratch_registers = retval + 1;
return retval;
} // allocate_scratch_register
857
858
859
// D3D stuff that's used in more than just the d3d profile...
860
861
862
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
863
864
865
866
867
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
868
869
static const char *get_D3D_register_string(Context *ctx,
870
RegisterType regtype,
871
872
int regnum, char *regnum_str,
size_t regnum_size)
873
{
874
const char *retval = NULL;
875
int has_number = 1;
876
877
switch (regtype)
878
{
879
case REG_TYPE_TEMP:
880
retval = "r";
881
882
break;
883
case REG_TYPE_INPUT:
884
retval = "v";
885
886
break;
887
case REG_TYPE_CONST:
888
889
890
retval = "c";
break;
891
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
892
retval = shader_is_vertex(ctx) ? "a" : "t";
893
894
break;
895
case REG_TYPE_RASTOUT:
896
switch ((RastOutType) regnum)
897
{
898
899
900
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
901
} // switch
902
has_number = 0;
903
904
break;
905
case REG_TYPE_ATTROUT:
906
retval = "oD";
907
908
break;
909
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
910
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
911
retval = "o";
912
else
913
retval = "oT";
914
915
break;
916
case REG_TYPE_CONSTINT:
917
retval = "i";
918
919
break;
920
case REG_TYPE_COLOROUT:
921
retval = "oC";
922
923
break;
924
case REG_TYPE_DEPTHOUT:
925
retval = "oDepth";
926
has_number = 0;
927
928
break;
929
case REG_TYPE_SAMPLER:
930
retval = "s";
931
932
break;
933
case REG_TYPE_CONSTBOOL:
934
retval = "b";
935
936
break;
937
case REG_TYPE_LOOP:
938
retval = "aL";
939
has_number = 0;
940
941
break;
942
case REG_TYPE_MISCTYPE:
943
switch ((const MiscTypeType) regnum)
944
{
945
946
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
947
} // switch
948
has_number = 0;
949
950
break;
951
case REG_TYPE_LABEL:
952
retval = "l";
953
954
break;
955
case REG_TYPE_PREDICATE:
956
retval = "p";
957
break;
958
959
960
961
962
963
964
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
965
966
} // switch
967
968
969
970
971
if (has_number)
snprintf(regnum_str, regnum_size, "%u", (uint) regnum);
else
regnum_str[0] = '\0';
972
973
974
975
return retval;
} // get_D3D_register_string
976
977
978
979
980
981
982
983
984
#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,
985
static const char *make_D3D_srcarg_string_in_buf(Context *ctx,
986
987
const SourceArgInfo *arg,
char *buf, size_t buflen)
988
989
990
{
const char *premod_str = "";
const char *postmod_str = "";
991
switch (arg->src_mod)
992
993
994
995
996
997
998
999
1000
{
case SRCMOD_NEGATE:
premod_str = "-";
break;
case SRCMOD_BIASNEGATE:
premod_str = "-";
// fall through.
case SRCMOD_BIAS: