/
mojoshader.c
7710 lines (6507 loc) · 249 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
// Get basic wankery out of the way here...
typedef unsigned int uint; // this is a printf() helper. don't use for code.
55
56
#ifdef _MSC_VER
57
#include <malloc.h>
58
59
#define snprintf _snprintf
typedef unsigned __int8 uint8;
60
typedef unsigned __int16 uint16;
61
typedef unsigned __int32 uint32;
62
typedef __int32 int32;
63
64
65
// Warning Level 4 considered harmful. :)
#pragma warning(disable: 4100) // "unreferenced formal parameter"
#pragma warning(disable: 4389) // "signed/unsigned mismatch"
66
67
#else
#include <stdint.h>
68
typedef uint8_t uint8;
69
typedef uint16_t uint16;
70
typedef uint32_t uint32;
71
typedef int32_t int32;
72
#endif
73
74
75
76
77
78
79
#ifdef __GNUC__
#define ISPRINTF(x,y) __attribute__((format (printf, x, y)))
#else
#define ISPRINTF(x,y)
#endif
80
81
#define STATICARRAYLEN(x) ( (sizeof ((x))) / (sizeof ((x)[0])) )
82
83
84
85
86
87
#ifdef _WINDOWS // !!! FIXME: bleh
const char *endline_str = "\r\n";
#else
const char *endline_str = "\n";
#endif
88
89
90
// we need to reference this by explicit value occasionally.
#define OPCODE_RET 28
91
92
93
94
95
// Special-case return values from the parsing pipeline...
#define FAIL (-1)
#define NOFAIL (-2)
#define END_OF_STREAM (-3)
96
97
98
99
100
101
102
103
104
// 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
105
106
107
108
109
static inline uint16 SWAP16(uint16 x)
{
__asm__ __volatile__("lhbrx %0,0,%1" : "=r" (x) : "r" (&x));
return x;
} // SWAP16
110
111
112
113
114
115
#elif defined(__POWERPC__)
static inline uint32 SWAP32(uint32 x)
{
return ( (((x) >> 24) & 0x000000FF) | (((x) >> 8) & 0x0000FF00) |
(((x) << 8) & 0x00FF0000) | (((x) << 24) & 0xFF000000) );
} // SWAP32
116
117
118
119
static inline uint16 SWAP16(uint16 x)
{
return ( (((x) >> 8) & 0x00FF) | (((x) << 8) & 0xFF00) );
} // SWAP16
120
#else
121
# define SWAP16(x) (x)
122
123
124
# define SWAP32(x) (x)
#endif
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
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;
151
152
153
154
155
156
157
158
typedef enum
{
TEXTURE_TYPE_2D = 2,
TEXTURE_TYPE_CUBE = 3,
TEXTURE_TYPE_VOLUME = 4,
} TextureType;
159
// predeclare.
160
typedef struct Context Context;
161
struct ConstantsList;
162
163
// one emit function for each opcode in each profile.
164
typedef void (*emit_function)(Context *ctx);
165
166
// one emit function for starting output in each profile.
167
typedef void (*emit_start)(Context *ctx, const char *profilestr);
168
169
// one emit function for ending output in each profile.
170
typedef void (*emit_end)(Context *ctx);
171
172
173
174
// one emit function for phase opcode output in each profile.
typedef void (*emit_phase)(Context *ctx);
175
176
177
178
179
180
// 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);
181
// one emit function for relative uniform arrays in each profile.
182
typedef void (*emit_array)(Context *ctx, int base, int size);
183
184
185
186
187
188
// one emit function for relative constants arrays in each profile.
typedef void (*emit_const_array)(Context *ctx,
const struct ConstantsList *constslist,
int base, int size);
189
// one emit function for uniforms in each profile.
190
191
typedef void (*emit_uniform)(Context *ctx, RegisterType regtype, int regnum,
int arraybase, int arraysize);
192
193
194
195
// one emit function for samplers in each profile.
typedef void (*emit_sampler)(Context *ctx, int stage, TextureType ttype);
196
197
// one emit function for attributes in each profile.
typedef void (*emit_attribute)(Context *ctx, RegisterType regtype, int regnum,
198
199
MOJOSHADER_usage usage, int index, int wmask,
int flags);
200
201
202
203
// one args function for each possible sequence of opcode arguments.
typedef int (*args_function)(Context *ctx);
204
// one state function for each opcode where we have state machine updates.
205
typedef void (*state_function)(Context *ctx);
206
207
208
209
210
// 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.
211
typedef const char *(*const_array_varname_function)(Context *c, int base, int size);
212
213
214
215
216
217
typedef struct
{
const char *name;
emit_start start_emitter;
emit_end end_emitter;
218
emit_phase phase_emitter;
219
emit_global global_emitter;
220
emit_array array_emitter;
221
emit_const_array const_array_emitter;
222
emit_uniform uniform_emitter;
223
emit_sampler sampler_emitter;
224
emit_attribute attribute_emitter;
225
emit_finalize finalize_emitter;
226
227
varname_function get_varname;
const_array_varname_function get_const_array_varname;
228
} Profile;
229
230
231
232
233
234
235
typedef enum
{
RASTOUT_TYPE_POSITION = 0,
RASTOUT_TYPE_FOG = 1,
RASTOUT_TYPE_POINT_SIZE = 2,
RASTOUT_TYPE_MAX = 2
236
} RastOutType;
237
238
239
240
241
242
typedef enum
{
MISCTYPE_TYPE_POSITION = 0,
MISCTYPE_TYPE_FACE = 1,
MISCTYPE_TYPE_MAX = 1
243
} MiscTypeType;
244
245
246
247
248
// 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.
249
typedef struct OutputListNode
250
251
{
char *str;
252
253
254
255
256
257
258
struct OutputListNode *next;
} OutputListNode;
typedef struct OutputList
{
OutputListNode head;
OutputListNode *tail;
259
260
} OutputList;
261
262
263
264
265
266
typedef struct ConstantsList
{
MOJOSHADER_constant constant;
struct ConstantsList *next;
} ConstantsList;
267
268
269
270
271
typedef struct VariableList
{
MOJOSHADER_uniformType type;
int index;
int count;
272
ConstantsList *constant;
273
int used;
274
275
struct VariableList *next;
} VariableList;
276
277
278
279
280
typedef struct RegisterList
{
RegisterType regtype;
int regnum;
281
282
283
MOJOSHADER_usage usage;
int index;
int writemask;
284
int misc;
285
const VariableList *array;
286
287
288
struct RegisterList *next;
} RegisterList;
289
// result modifiers.
290
// !!! FIXME: why isn't this an enum?
291
292
293
294
#define MOD_SATURATE 0x01
#define MOD_PP 0x02
#define MOD_CENTROID 0x04
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
// 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
313
} SourceMod;
314
315
316
317
typedef struct
{
318
const uint32 *token; // this is the unmolested token in the stream.
319
320
int regnum;
int relative;
321
322
323
324
325
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
326
int orig_writemask; // writemask before mojoshader tweaks it.
327
328
int result_mod;
int result_shift;
329
RegisterType regtype;
330
331
332
333
} DestArgInfo;
typedef struct
{
334
const uint32 *token; // this is the unmolested token in the stream.
335
int regnum;
336
int swizzle; // xyzw (all four, not split out).
337
338
339
340
int swizzle_x;
int swizzle_y;
int swizzle_z;
int swizzle_w;
341
SourceMod src_mod;
342
RegisterType regtype;
343
344
345
346
int relative;
RegisterType relative_regtype;
int relative_regnum;
int relative_component;
347
const VariableList *relative_array;
348
349
350
} SourceArgInfo;
351
352
#define SCRATCH_BUFFER_SIZE 128
353
#define SCRATCH_BUFFERS 32
354
355
// !!! FIXME: the scratch buffers make Context pretty big.
356
357
// !!! FIXME: might be worth having one set of static scratch buffers that
// !!! FIXME: are mutex protected?
358
359
// Context...this is state that changes as we parse through a shader...
360
struct Context
361
{
362
363
MOJOSHADER_malloc malloc;
MOJOSHADER_free free;
364
void *malloc_data;
365
366
const uint32 *tokens;
uint32 tokencount;
367
368
const MOJOSHADER_swizzle *swizzles;
unsigned int swizzles_count;
369
370
OutputList *output;
OutputList globals;
371
OutputList helpers;
372
OutputList subroutines;
373
OutputList mainline_intro;
374
OutputList mainline;
375
376
OutputList ignore;
OutputList *output_stack[2];
377
uint8 *output_bytes; // can be used instead of the OutputLists.
378
379
int indent_stack[2];
int output_stack_len;
380
int output_len; // total strlen; prevents walking the lists just to malloc.
381
int indent;
382
const char *shader_type_str;
383
384
385
const char *endline;
int endline_len;
const char *failstr;
386
char scratch[SCRATCH_BUFFERS][SCRATCH_BUFFER_SIZE];
387
388
int scratchidx; // current scratch buffer.
int profileid;
389
const Profile *profile;
390
MOJOSHADER_shaderType shader_type;
391
392
uint8 major_ver;
uint8 minor_ver;
393
DestArgInfo dest_arg;
394
SourceArgInfo source_args[5];
395
SourceArgInfo predicate_arg; // for predicated instructions.
396
uint32 dwords[4];
397
uint32 version_token;
398
int instruction_count;
399
uint32 instruction_controls;
400
401
uint32 previous_opcode;
int loops;
402
int reps;
403
int max_reps;
404
int cmps;
405
406
int scratch_registers;
int max_scratch_registers;
407
408
409
int branch_labels_stack_index;
int branch_labels_stack[32];
int assigned_branch_labels;
410
int assigned_vertex_attributes;
411
int last_address_reg_component;
412
413
RegisterList used_registers;
RegisterList defined_registers;
414
415
int constant_count;
ConstantsList *constants;
416
417
int uniform_count;
RegisterList uniforms;
418
419
int attribute_count;
RegisterList attributes;
420
421
int sampler_count;
RegisterList samplers;
422
VariableList *variables; // variables to register mapping.
423
int centroid_allowed:1;
424
int have_ctab:1;
425
int determined_constants_arrays:1;
426
427
int predicated:1;
int support_nv2:1;
428
int support_nv3:1;
429
int support_nv4:1;
430
int support_glsl120:1;
431
int glsl_generated_lit_opcode:1;
432
433
434
};
435
436
// Convenience functions for allocators...
437
438
439
440
441
442
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)
443
{
444
445
446
447
448
if (ctx->failstr == NULL)
ctx->failstr = out_of_mem_str; // fail() would call malloc().
return FAIL;
} // out_of_memory
449
static inline void *Malloc(Context *ctx, const size_t len)
450
{
451
void *retval = ctx->malloc((int) len, ctx->malloc_data);
452
453
454
if (retval == NULL)
out_of_memory(ctx);
return retval;
455
456
457
} // Malloc
458
static inline void Free(Context *ctx, void *ptr)
459
{
460
if (ptr != NULL) // check for NULL in case of dumb free() impl.
461
ctx->free(ptr, ctx->malloc_data);
462
463
464
} // Free
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
// 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...
489
490
491
492
493
static inline uint32 ver_ui32(const uint8 major, const uint8 minor)
{
return ( (((uint32) major) << 16) | (((minor) == 0xFF) ? 0 : (minor)) );
} // version_ui32
494
static inline int shader_version_supported(const uint8 maj, const uint8 min)
495
496
497
498
{
return (ver_ui32(maj,min) <= ver_ui32(MAX_SHADER_MAJOR, MAX_SHADER_MINOR));
} // shader_version_supported
499
500
static inline int shader_version_atleast(const Context *ctx, const uint8 maj,
const uint8 min)
501
502
503
504
{
return (ver_ui32(ctx->major_ver, ctx->minor_ver) >= ver_ui32(maj, min));
} // shader_version_atleast
505
506
507
508
509
510
static inline int shader_version_exactly(const Context *ctx, const uint8 maj,
const uint8 min)
{
return ((ctx->major_ver == maj) && (ctx->minor_ver == min));
} // shader_version_exactly
511
512
513
514
515
516
517
518
519
520
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
521
522
523
524
525
526
static inline int isfail(const Context *ctx)
{
return (ctx->failstr != NULL);
} // isfail
527
528
static inline char *get_scratch_buffer(Context *ctx)
529
{
530
531
532
533
534
535
536
537
538
539
540
541
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
542
ctx->scratchidx = (ctx->scratchidx + 1) % SCRATCH_BUFFERS;
543
return ctx->scratch[ctx->scratchidx];
544
545
} // get_scratch_buffer
546
547
548
static int failf(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int failf(Context *ctx, const char *fmt, ...)
549
{
550
551
if (ctx->failstr == NULL) // don't change existing error.
{
552
char *scratch = get_scratch_buffer(ctx);
553
554
va_list ap;
va_start(ap, fmt);
555
const int len = vsnprintf(scratch, SCRATCH_BUFFER_SIZE, fmt, ap);
556
557
va_end(ap);
558
char *failstr = (char *) Malloc(ctx, len + 1);
559
if (failstr != NULL)
560
561
{
// see comments about scratch buffer overflow in output_line().
562
if (len < SCRATCH_BUFFER_SIZE)
563
564
565
566
567
568
569
570
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;
571
} // if
572
573
574
575
576
577
} // if
return FAIL;
} // failf
578
static inline int fail(Context *ctx, const char *reason)
579
580
581
582
583
{
return failf(ctx, "%s", reason);
} // fail
584
585
static int output_line(Context *ctx, const char *fmt, ...) ISPRINTF(2,3);
static int output_line(Context *ctx, const char *fmt, ...)
586
{
587
588
OutputListNode *item = NULL;
589
if (isfail(ctx))
590
591
return FAIL; // we failed previously, don't go on...
592
char *scratch = get_scratch_buffer(ctx);
593
594
const int indent = ctx->indent;
595
596
if (indent > 0)
memset(scratch, '\t', indent);
597
598
599
va_list ap;
va_start(ap, fmt);
600
const int len = vsnprintf(scratch+indent, SCRATCH_BUFFER_SIZE-indent, fmt, ap) + indent;
601
602
va_end(ap);
603
item = (OutputListNode *) Malloc(ctx, sizeof (OutputListNode));
604
if (item == NULL)
605
return FAIL;
606
607
item->str = (char *) Malloc(ctx, len + 1);
608
if (item->str == NULL)
609
{
610
Free(ctx, item);
611
return FAIL;
612
613
} // if
614
615
616
// 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().
617
if (len < SCRATCH_BUFFER_SIZE)
618
619
620
strcpy(item->str, scratch); // copy it over.
else
{
621
622
if (indent > 0)
memset(item->str, '\t', indent);
623
va_start(ap, fmt);
624
vsnprintf(item->str+indent, len + 1, fmt, ap); // rebuild it.
625
626
va_end(ap);
} // else
627
628
item->next = NULL;
629
630
631
ctx->output->tail->next = item;
ctx->output->tail = item;
632
ctx->output_len += len + ctx->endline_len;
633
634
635
636
637
return 0;
} // output_line
638
639
640
641
642
643
644
// this is just to stop gcc whining.
static inline int output_blank_line(Context *ctx)
{
return output_line(ctx, "%s", "");
} // output_blank_line
645
// !!! FIXME: this is sort of nasty.
646
647
static void floatstr(Context *ctx, char *buf, size_t bufsize, float f,
int leavedecimal)
648
649
{
const size_t len = snprintf(buf, bufsize, "%f", f);
650
if ((len+2) >= bufsize)
651
652
653
654
655
656
fail(ctx, "BUG: internal buffer is too small");
else
{
char *end = buf + len;
char *ptr = strchr(buf, '.');
if (ptr == NULL)
657
658
659
{
if (leavedecimal)
strcat(buf, ".0");
660
return; // done.
661
} // if
662
663
664
665
666
667
668
669
670
while (--end != ptr)
{
if (*end != '0')
{
end++;
break;
} // if
} // while
671
672
if ((leavedecimal) && (end == ptr))
end += 2;
673
674
675
676
*end = '\0'; // chop extra '0' or all decimal places off.
} // else
} // floatstr
677
678
679
// Deal with register lists... !!! FIXME: I sort of hate this.
680
static void free_reglist(MOJOSHADER_free f, void *d, RegisterList *item)
681
682
683
684
{
while (item != NULL)
{
RegisterList *next = item->next;
685
f(item, d);
686
687
688
689
690
691
692
693
694
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
695
696
697
static RegisterList *reglist_insert(Context *ctx, RegisterList *prev,
const RegisterType regtype,
const int regnum)
698
699
700
701
702
703
704
{
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)
705
return item; // already set, so we're done.
706
707
708
709
710
711
712
713
714
715
716
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).
717
item = (RegisterList *) Malloc(ctx, sizeof (RegisterList));
718
if (item != NULL)
719
720
721
{
item->regtype = regtype;
item->regnum = regnum;
722
item->usage = MOJOSHADER_USAGE_UNKNOWN;
723
724
item->index = 0;
item->writemask = 0;
725
item->misc = 0;
726
item->array = NULL;
727
728
item->next = prev->next;
prev->next = item;
729
} // if
730
731
return item;
732
733
} // reglist_insert
734
735
static RegisterList *reglist_find(const RegisterList *prev,
const RegisterType rtype, const int regnum)
736
{
737
738
const uint32 newval = reg_to_ui32(rtype, regnum);
RegisterList *item = prev->next;
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
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.
755
756
757
758
759
760
761
} // reglist_find
static inline const RegisterList *reglist_exists(RegisterList *prev,
const RegisterType regtype,
const int regnum)
{
return (reglist_find(prev, regtype, regnum));
762
763
764
765
766
767
768
769
770
771
772
773
} // 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);
774
} // get_used_register
775
776
777
778
779
780
781
782
783
784
785
786
787
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
788
789
static void add_attribute_register(Context *ctx, const RegisterType rtype,
const int regnum, const MOJOSHADER_usage usage,
790
const int index, const int writemask, int flags)
791
792
793
794
795
{
RegisterList *item = reglist_insert(ctx, &ctx->attributes, rtype, regnum);
item->usage = usage;
item->index = index;
item->writemask = writemask;
796
item->misc = flags;
797
798
} // add_attribute_register
799
800
801
static inline void add_sampler(Context *ctx, const RegisterType rtype,
const int regnum, const TextureType ttype)
{
802
// !!! FIXME: make sure it doesn't exist?
803
804
805
806
RegisterList *item = reglist_insert(ctx, &ctx->samplers, rtype, regnum);
item->index = (int) ttype;
} // add_sampler
807
808
809
810
811
812
813
814
815
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 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
838
839
840
841
842
843
844
845
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
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
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
868
869
static inline int no_swizzle(const int swizzle)
{
870
return (swizzle == 0xE4); // 0xE4 == 11100100 ... 0 1 2 3. No swizzle.
871
872
873
} // no_swizzle
874
875
876
877
878
static inline int vecsize_from_writemask(const int m)
{
return (m & 1) + ((m >> 1) & 1) + ((m >> 2) & 1) + ((m >> 3) & 1);
} // vecsize_from_writemask
879
880
881
882
883
884
885
886
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
887
static int allocate_branch_label(Context *ctx)
888
{
889
890
return ctx->assigned_branch_labels++;
} // allocate_branch_label
891
892
893
894
// D3D stuff that's used in more than just the d3d profile...
895
896
897
static const char swizzle_channels[] = { 'x', 'y', 'z', 'w' };
898
899
900
901
902
static const char *usagestrs[] = {
"_position", "_blendweight", "_blendindices", "_normal", "_psize",
"_texcoord", "_tangent", "_binormal", "_tessfactor", "_positiont",
"_color", "_fog", "_depth", "_sample"
};
903
904
static const char *get_D3D_register_string(Context *ctx,
905
RegisterType regtype,
906
907
int regnum, char *regnum_str,
size_t regnum_size)
908
{
909
const char *retval = NULL;
910
int has_number = 1;
911
912
switch (regtype)
913
{
914
case REG_TYPE_TEMP:
915
retval = "r";
916
917
break;
918
case REG_TYPE_INPUT:
919
retval = "v";
920
921
break;
922
case REG_TYPE_CONST:
923
924
925
retval = "c";
break;
926
case REG_TYPE_ADDRESS: // (or REG_TYPE_TEXTURE, same value.)
927
retval = shader_is_vertex(ctx) ? "a" : "t";
928
929
break;
930
case REG_TYPE_RASTOUT:
931
switch ((RastOutType) regnum)
932
{
933
934
935
case RASTOUT_TYPE_POSITION: retval = "oPos"; break;
case RASTOUT_TYPE_FOG: retval = "oFog"; break;
case RASTOUT_TYPE_POINT_SIZE: retval = "oPts"; break;
936
} // switch
937
has_number = 0;
938
939
break;
940
case REG_TYPE_ATTROUT:
941
retval = "oD";
942
943
break;
944
case REG_TYPE_OUTPUT: // (or REG_TYPE_TEXCRDOUT, same value.)
945
if (shader_is_vertex(ctx) && shader_version_atleast(ctx, 3, 0))
946
retval = "o";
947
else
948
retval = "oT";
949
950
break;
951
case REG_TYPE_CONSTINT:
952
retval = "i";
953
954
break;
955
case REG_TYPE_COLOROUT:
956
retval = "oC";
957
958
break;
959
case REG_TYPE_DEPTHOUT:
960
retval = "oDepth";
961
has_number = 0;
962
963
break;
964
case REG_TYPE_SAMPLER:
965
retval = "s";
966
967
break;
968
case REG_TYPE_CONSTBOOL:
969
retval = "b";
970
971
break;
972
case REG_TYPE_LOOP:
973
retval = "aL";
974
has_number = 0;
975
976
break;
977
case REG_TYPE_MISCTYPE:
978
switch ((const MiscTypeType) regnum)
979
{
980
981
case MISCTYPE_TYPE_POSITION: retval = "vPos"; break;
case MISCTYPE_TYPE_FACE: retval = "vFace"; break;
982
} // switch
983
has_number = 0;
984
985
break;
986
case REG_TYPE_LABEL:
987
retval = "l";
988
989
break;
990
case REG_TYPE_PREDICATE:
991
retval = "p";
992
break;
993
994
995
996
997
998
999
//case REG_TYPE_TEMPFLOAT16: // !!! FIXME: don't know this asm string
default:
fail(ctx, "unknown register type");
retval = "???";
has_number = 0;
break;
1000
} // switch