1 /*
2 * Copyright (c) 2013 Rob Clark <robclark@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 */
23
24 %code requires {
25 #include "ir3/ir3_assembler.h"
26 #include "ir3/ir3_shader.h"
27
28 struct ir3 * ir3_parse(struct ir3_shader_variant *v,
29 struct ir3_kernel_info *k, FILE *f);
30 }
31
32 %{
33 #define YYDEBUG 0
34
35 #include <stdlib.h>
36 #include <stdio.h>
37 #include <string.h>
38 #include <math.h>
39
40 #include "util/half_float.h"
41 #include "util/u_math.h"
42
43 #include "ir3/ir3.h"
44 #include "ir3/ir3_shader.h"
45 #include "ir3/instr-a3xx.h"
46
47 #include "ir3_parser.h"
48
49 #define swap(a, b) \
50 do { __typeof(a) __tmp = (a); (a) = (b); (b) = __tmp; } while (0)
51
52 /* ir3 treats the abs/neg flags as separate flags for float vs integer,
53 * but in the instruction encoding they are the same thing. Tracking
54 * them separately is only for the benefit of ir3 opt passes, and not
55 * required here, so just use the float versions:
56 */
57 #define IR3_REG_ABS IR3_REG_FABS
58 #define IR3_REG_NEGATE IR3_REG_FNEG
59
60 static struct ir3_kernel_info *info;
61 static struct ir3_shader_variant *variant;
62 /* NOTE the assembler doesn't really use the ir3_block construction
63 * like the compiler does. Everything is treated as one large block.
64 * Which might happen to contain flow control. But since we don't
65 * use any of the ir3 backend passes (sched, RA, etc) this doesn't
66 * really matter.
67 */
68 static struct ir3_block *block; /* current shader block */
69 static struct ir3_instruction *instr; /* current instruction */
70 static unsigned ip; /* current instruction pointer */
71 static struct hash_table *labels;
72
73 void *ir3_parser_dead_ctx;
74
75 static struct {
76 unsigned flags;
77 unsigned repeat;
78 unsigned nop;
79 } iflags;
80
81 static struct {
82 unsigned flags;
83 unsigned wrmask;
84 } rflags;
85
86 static struct {
87 uint32_t reg_address_hi;
88 uint32_t reg_address_lo;
89 uint32_t reg_tmp;
90
91 uint32_t regs_to_dump[128];
92 uint32_t regs_count;
93 } meta_print_data;
94
95 int ir3_yyget_lineno(void);
96
new_label(const char * name)97 static void new_label(const char *name)
98 {
99 ralloc_steal(labels, (void *) name);
100 _mesa_hash_table_insert(labels, name, (void *)(uintptr_t)ip);
101 }
102
new_instr(opc_t opc)103 static struct ir3_instruction * new_instr(opc_t opc)
104 {
105 instr = ir3_instr_create(block, opc, 4, 6);
106 instr->flags = iflags.flags;
107 instr->repeat = iflags.repeat;
108 instr->nop = iflags.nop;
109 instr->line = ir3_yyget_lineno();
110 iflags.flags = iflags.repeat = iflags.nop = 0;
111 ip++;
112 return instr;
113 }
114
new_shader(void)115 static void new_shader(void)
116 {
117 variant->ir = ir3_create(variant->compiler, variant);
118 block = ir3_block_create(variant->ir);
119 list_addtail(&block->node, &variant->ir->block_list);
120 ip = 0;
121 labels = _mesa_hash_table_create(variant, _mesa_hash_string, _mesa_key_string_equal);
122 ir3_parser_dead_ctx = ralloc_context(NULL);
123 }
124
parse_type(const char ** type)125 static type_t parse_type(const char **type)
126 {
127 if (!strncmp("f16", *type, 3)) {
128 *type += 3;
129 return TYPE_F16;
130 } else if (!strncmp("f32", *type, 3)) {
131 *type += 3;
132 return TYPE_F32;
133 } else if (!strncmp("u16", *type, 3)) {
134 *type += 3;
135 return TYPE_U16;
136 } else if (!strncmp("u32", *type, 3)) {
137 *type += 3;
138 return TYPE_U32;
139 } else if (!strncmp("s16", *type, 3)) {
140 *type += 3;
141 return TYPE_S16;
142 } else if (!strncmp("s32", *type, 3)) {
143 *type += 3;
144 return TYPE_S32;
145 } else if (!strncmp("u8", *type, 2)) {
146 *type += 2;
147 return TYPE_U8;
148 } else if (!strncmp("s8", *type, 2)) {
149 *type += 2;
150 return TYPE_S8;
151 } else {
152 assert(0); /* shouldn't get here */
153 return ~0;
154 }
155 }
156
parse_type_type(struct ir3_instruction * instr,const char * type_type)157 static struct ir3_instruction * parse_type_type(struct ir3_instruction *instr,
158 const char *type_type)
159 {
160 instr->cat1.src_type = parse_type(&type_type);
161 instr->cat1.dst_type = parse_type(&type_type);
162 return instr;
163 }
164
new_src(int num,unsigned flags)165 static struct ir3_register * new_src(int num, unsigned flags)
166 {
167 struct ir3_register *reg;
168 flags |= rflags.flags;
169 if (num & 0x1)
170 flags |= IR3_REG_HALF;
171 reg = ir3_src_create(instr, num>>1, flags);
172 reg->wrmask = MAX2(1, rflags.wrmask);
173 rflags.flags = rflags.wrmask = 0;
174 return reg;
175 }
176
new_dst(int num,unsigned flags)177 static struct ir3_register * new_dst(int num, unsigned flags)
178 {
179 struct ir3_register *reg;
180 flags |= rflags.flags;
181 if (num & 0x1)
182 flags |= IR3_REG_HALF;
183 reg = ir3_dst_create(instr, num>>1, flags);
184 reg->wrmask = MAX2(1, rflags.wrmask);
185 rflags.flags = rflags.wrmask = 0;
186 return reg;
187 }
188
dummy_dst(void)189 static struct ir3_register * dummy_dst(void)
190 {
191 return new_dst(0, 0);
192 }
193
fixup_cat5_s2en(void)194 static void fixup_cat5_s2en(void)
195 {
196 assert(opc_cat(instr->opc) == 5);
197 if (!(instr->flags & IR3_INSTR_S2EN))
198 return;
199 /* For various reasons (ie. mainly to make the .s2en src easier to
200 * find, given that various different cat5 tex instructions can have
201 * different # of src registers), in ir3 the samp/tex src register
202 * is first, rather than last. So we have to detect this case and
203 * fix things up.
204 */
205 struct ir3_register *s2en_src = instr->srcs[instr->srcs_count - 1];
206
207 if (instr->flags & IR3_INSTR_B)
208 assert(!(s2en_src->flags & IR3_REG_HALF));
209 else
210 assert(s2en_src->flags & IR3_REG_HALF);
211
212 for (int i = 0; i < instr->srcs_count - 1; i++) {
213 instr->srcs[i+1] = instr->srcs[i];
214 }
215 instr->srcs[0] = s2en_src;
216 }
217
add_const(unsigned reg,unsigned c0,unsigned c1,unsigned c2,unsigned c3)218 static void add_const(unsigned reg, unsigned c0, unsigned c1, unsigned c2, unsigned c3)
219 {
220 struct ir3_const_state *const_state = ir3_const_state(variant);
221 assert((reg & 0x7) == 0);
222 int idx = reg >> (1 + 2); /* low bit is half vs full, next two bits are swiz */
223 if (idx * 4 + 4 > const_state->immediates_size) {
224 const_state->immediates = rerzalloc(const_state,
225 const_state->immediates,
226 __typeof__(const_state->immediates[0]),
227 const_state->immediates_size,
228 idx * 4 + 4);
229 for (unsigned i = const_state->immediates_size; i < idx * 4; i++)
230 const_state->immediates[i] = 0xd0d0d0d0;
231 const_state->immediates_size = const_state->immediates_count = idx * 4 + 4;
232 }
233 const_state->immediates[idx * 4 + 0] = c0;
234 const_state->immediates[idx * 4 + 1] = c1;
235 const_state->immediates[idx * 4 + 2] = c2;
236 const_state->immediates[idx * 4 + 3] = c3;
237 }
238
add_sysval(unsigned reg,unsigned compmask,gl_system_value sysval)239 static void add_sysval(unsigned reg, unsigned compmask, gl_system_value sysval)
240 {
241 unsigned n = variant->inputs_count++;
242 variant->inputs[n].regid = reg;
243 variant->inputs[n].sysval = true;
244 variant->inputs[n].slot = sysval;
245 variant->inputs[n].compmask = compmask;
246 variant->total_in++;
247 }
248
resolve_labels(void)249 static bool resolve_labels(void)
250 {
251 int instr_ip = 0;
252 foreach_instr (instr, &block->instr_list) {
253 if (opc_cat(instr->opc) == 0 && instr->cat0.target_label) {
254 struct hash_entry *entry = _mesa_hash_table_search(labels, instr->cat0.target_label);
255 if (!entry) {
256 fprintf(stderr, "unknown label %s\n", instr->cat0.target_label);
257 return false;
258 }
259 int target_ip = (uintptr_t)entry->data;
260 instr->cat0.immed = target_ip - instr_ip;
261 }
262 instr_ip++;
263 }
264 return true;
265 }
266
267 #ifdef YYDEBUG
268 int yydebug;
269 #endif
270
271 extern int yylex(void);
272 void ir3_yyset_lineno(int _line_number);
273 void ir3_yyset_input(FILE *f);
274
275 int yyparse(void);
276
yyerror(const char * error)277 static void yyerror(const char *error)
278 {
279 fprintf(stderr, "error at line %d: %s\n", ir3_yyget_lineno(), error);
280 }
281
ir3_parse(struct ir3_shader_variant * v,struct ir3_kernel_info * k,FILE * f)282 struct ir3 * ir3_parse(struct ir3_shader_variant *v,
283 struct ir3_kernel_info *k, FILE *f)
284 {
285 ir3_yyset_lineno(1);
286 ir3_yyset_input(f);
287 #ifdef YYDEBUG
288 yydebug = 1;
289 #endif
290 info = k;
291 variant = v;
292 if (yyparse() || !resolve_labels()) {
293 ir3_destroy(variant->ir);
294 variant->ir = NULL;
295 }
296 ralloc_free(labels);
297 ralloc_free(ir3_parser_dead_ctx);
298 return variant->ir;
299 }
300 %}
301
302 %union {
303 int tok;
304 int num;
305 uint32_t unum;
306 uint64_t u64;
307 double flt;
308 const char *str;
309 struct ir3_register *reg;
310 struct {
311 int start;
312 int num;
313 } range;
314 type_t type;
315 }
316
317 %{
318 #if YYDEBUG
print_token(FILE * file,int type,YYSTYPE value)319 static void print_token(FILE *file, int type, YYSTYPE value)
320 {
321 fprintf(file, "\ntype: %d\n", type);
322 }
323
324 #define YYPRINT(file, type, value) print_token(file, type, value)
325 #endif
326 %}
327
328 %token <num> T_INT
329 %token <unum> T_HEX
330 %token <flt> T_FLOAT
331 %token <str> T_IDENTIFIER
332 %token <num> T_REGISTER
333 %token <num> T_CONSTANT
334
335 /* @ headers (@const/@sampler/@uniform/@varying) */
336 %token <tok> T_A_LOCALSIZE
337 %token <tok> T_A_CONST
338 %token <tok> T_A_BUF
339 %token <tok> T_A_INVOCATIONID
340 %token <tok> T_A_WGID
341 %token <tok> T_A_NUMWG
342 %token <tok> T_A_BRANCHSTACK
343 %token <tok> T_A_IN
344 %token <tok> T_A_OUT
345 %token <tok> T_A_TEX
346 %token <tok> T_A_PVTMEM
347 %token <tok> T_A_EARLYPREAMBLE
348 /* todo, re-add @sampler/@uniform/@varying if needed someday */
349
350 /* src register flags */
351 %token <tok> T_ABSNEG
352 %token <tok> T_NEG
353 %token <tok> T_ABS
354 %token <tok> T_R
355 %token <tok> T_LAST
356
357 %token <tok> T_HR
358 %token <tok> T_HC
359
360 /* dst register flags */
361 %token <tok> T_EVEN
362 %token <tok> T_POS_INFINITY
363 %token <tok> T_NEG_INFINITY
364 %token <tok> T_EI
365 %token <num> T_WRMASK
366
367 /* Float LUT values accepted as immed: */
368 %token <num> T_FLUT_0_0
369 %token <num> T_FLUT_0_5
370 %token <num> T_FLUT_1_0
371 %token <num> T_FLUT_2_0
372 %token <num> T_FLUT_E
373 %token <num> T_FLUT_PI
374 %token <num> T_FLUT_INV_PI
375 %token <num> T_FLUT_INV_LOG2_E
376 %token <num> T_FLUT_LOG2_E
377 %token <num> T_FLUT_INV_LOG2_10
378 %token <num> T_FLUT_LOG2_10
379 %token <num> T_FLUT_4_0
380
381 /* instruction flags */
382 %token <tok> T_SY
383 %token <tok> T_SS
384 %token <tok> T_JP
385 %token <tok> T_EQ_FLAG
386 %token <tok> T_SAT
387 %token <num> T_RPT
388 %token <tok> T_UL
389 %token <tok> T_NOP
390
391 /* category 0: */
392 %token <tok> T_OP_NOP
393 %token <tok> T_OP_BR
394 %token <tok> T_OP_BRAO
395 %token <tok> T_OP_BRAA
396 %token <tok> T_OP_BRAC
397 %token <tok> T_OP_BANY
398 %token <tok> T_OP_BALL
399 %token <tok> T_OP_BRAX
400 %token <tok> T_OP_JUMP
401 %token <tok> T_OP_CALL
402 %token <tok> T_OP_RET
403 %token <tok> T_OP_KILL
404 %token <tok> T_OP_END
405 %token <tok> T_OP_EMIT
406 %token <tok> T_OP_CUT
407 %token <tok> T_OP_CHMASK
408 %token <tok> T_OP_CHSH
409 %token <tok> T_OP_FLOW_REV
410 %token <tok> T_OP_BKT
411 %token <tok> T_OP_STKS
412 %token <tok> T_OP_STKR
413 %token <tok> T_OP_XSET
414 %token <tok> T_OP_XCLR
415 %token <tok> T_OP_GETLAST
416 %token <tok> T_OP_GETONE
417 %token <tok> T_OP_DBG
418 %token <tok> T_OP_SHPS
419 %token <tok> T_OP_SHPE
420 %token <tok> T_OP_PREDT
421 %token <tok> T_OP_PREDF
422 %token <tok> T_OP_PREDE
423
424 /* category 1: */
425 %token <tok> T_OP_MOVMSK
426 %token <tok> T_OP_MOVA1
427 %token <tok> T_OP_MOVA
428 %token <tok> T_OP_MOV
429 %token <tok> T_OP_COV
430 %token <tok> T_OP_SWZ
431 %token <tok> T_OP_GAT
432 %token <tok> T_OP_SCT
433
434 /* category 2: */
435 %token <tok> T_OP_ADD_F
436 %token <tok> T_OP_MIN_F
437 %token <tok> T_OP_MAX_F
438 %token <tok> T_OP_MUL_F
439 %token <tok> T_OP_SIGN_F
440 %token <tok> T_OP_CMPS_F
441 %token <tok> T_OP_ABSNEG_F
442 %token <tok> T_OP_CMPV_F
443 %token <tok> T_OP_FLOOR_F
444 %token <tok> T_OP_CEIL_F
445 %token <tok> T_OP_RNDNE_F
446 %token <tok> T_OP_RNDAZ_F
447 %token <tok> T_OP_TRUNC_F
448 %token <tok> T_OP_ADD_U
449 %token <tok> T_OP_ADD_S
450 %token <tok> T_OP_SUB_U
451 %token <tok> T_OP_SUB_S
452 %token <tok> T_OP_CMPS_U
453 %token <tok> T_OP_CMPS_S
454 %token <tok> T_OP_MIN_U
455 %token <tok> T_OP_MIN_S
456 %token <tok> T_OP_MAX_U
457 %token <tok> T_OP_MAX_S
458 %token <tok> T_OP_ABSNEG_S
459 %token <tok> T_OP_AND_B
460 %token <tok> T_OP_OR_B
461 %token <tok> T_OP_NOT_B
462 %token <tok> T_OP_XOR_B
463 %token <tok> T_OP_CMPV_U
464 %token <tok> T_OP_CMPV_S
465 %token <tok> T_OP_MUL_U24
466 %token <tok> T_OP_MUL_S24
467 %token <tok> T_OP_MULL_U
468 %token <tok> T_OP_BFREV_B
469 %token <tok> T_OP_CLZ_S
470 %token <tok> T_OP_CLZ_B
471 %token <tok> T_OP_SHL_B
472 %token <tok> T_OP_SHR_B
473 %token <tok> T_OP_ASHR_B
474 %token <tok> T_OP_BARY_F
475 %token <tok> T_OP_FLAT_B
476 %token <tok> T_OP_MGEN_B
477 %token <tok> T_OP_GETBIT_B
478 %token <tok> T_OP_SETRM
479 %token <tok> T_OP_CBITS_B
480 %token <tok> T_OP_SHB
481 %token <tok> T_OP_MSAD
482
483 /* category 3: */
484 %token <tok> T_OP_MAD_U16
485 %token <tok> T_OP_MADSH_U16
486 %token <tok> T_OP_MAD_S16
487 %token <tok> T_OP_MADSH_M16
488 %token <tok> T_OP_MAD_U24
489 %token <tok> T_OP_MAD_S24
490 %token <tok> T_OP_MAD_F16
491 %token <tok> T_OP_MAD_F32
492 %token <tok> T_OP_SEL_B16
493 %token <tok> T_OP_SEL_B32
494 %token <tok> T_OP_SEL_S16
495 %token <tok> T_OP_SEL_S32
496 %token <tok> T_OP_SEL_F16
497 %token <tok> T_OP_SEL_F32
498 %token <tok> T_OP_SAD_S16
499 %token <tok> T_OP_SAD_S32
500 %token <tok> T_OP_SHRM
501 %token <tok> T_OP_SHLM
502 %token <tok> T_OP_SHRG
503 %token <tok> T_OP_SHLG
504 %token <tok> T_OP_ANDG
505 %token <tok> T_OP_DP2ACC
506 %token <tok> T_OP_DP4ACC
507 %token <tok> T_OP_WMM
508 %token <tok> T_OP_WMM_ACCU
509
510 /* category 4: */
511 %token <tok> T_OP_RCP
512 %token <tok> T_OP_RSQ
513 %token <tok> T_OP_LOG2
514 %token <tok> T_OP_EXP2
515 %token <tok> T_OP_SIN
516 %token <tok> T_OP_COS
517 %token <tok> T_OP_SQRT
518 %token <tok> T_OP_HRSQ
519 %token <tok> T_OP_HLOG2
520 %token <tok> T_OP_HEXP2
521
522 /* category 5: */
523 %token <tok> T_OP_ISAM
524 %token <tok> T_OP_ISAML
525 %token <tok> T_OP_ISAMM
526 %token <tok> T_OP_SAM
527 %token <tok> T_OP_SAMB
528 %token <tok> T_OP_SAML
529 %token <tok> T_OP_SAMGQ
530 %token <tok> T_OP_GETLOD
531 %token <tok> T_OP_CONV
532 %token <tok> T_OP_CONVM
533 %token <tok> T_OP_GETSIZE
534 %token <tok> T_OP_GETBUF
535 %token <tok> T_OP_GETPOS
536 %token <tok> T_OP_GETINFO
537 %token <tok> T_OP_DSX
538 %token <tok> T_OP_DSY
539 %token <tok> T_OP_GATHER4R
540 %token <tok> T_OP_GATHER4G
541 %token <tok> T_OP_GATHER4B
542 %token <tok> T_OP_GATHER4A
543 %token <tok> T_OP_SAMGP0
544 %token <tok> T_OP_SAMGP1
545 %token <tok> T_OP_SAMGP2
546 %token <tok> T_OP_SAMGP3
547 %token <tok> T_OP_DSXPP_1
548 %token <tok> T_OP_DSYPP_1
549 %token <tok> T_OP_RGETPOS
550 %token <tok> T_OP_RGETINFO
551 %token <tok> T_OP_BRCST_A
552 %token <tok> T_OP_QSHUFFLE_BRCST
553 %token <tok> T_OP_QSHUFFLE_H
554 %token <tok> T_OP_QSHUFFLE_V
555 %token <tok> T_OP_QSHUFFLE_DIAG
556 %token <tok> T_OP_TCINV
557
558 /* category 6: */
559 %token <tok> T_OP_LDG
560 %token <tok> T_OP_LDG_A
561 %token <tok> T_OP_LDG_K
562 %token <tok> T_OP_LDL
563 %token <tok> T_OP_LDP
564 %token <tok> T_OP_STG
565 %token <tok> T_OP_STG_A
566 %token <tok> T_OP_STL
567 %token <tok> T_OP_STP
568 %token <tok> T_OP_LDIB
569 %token <tok> T_OP_G2L
570 %token <tok> T_OP_L2G
571 %token <tok> T_OP_PREFETCH
572 %token <tok> T_OP_LDLW
573 %token <tok> T_OP_STLW
574 %token <tok> T_OP_RESFMT
575 %token <tok> T_OP_RESINFO
576 %token <tok> T_OP_ATOMIC_ADD
577 %token <tok> T_OP_ATOMIC_SUB
578 %token <tok> T_OP_ATOMIC_XCHG
579 %token <tok> T_OP_ATOMIC_INC
580 %token <tok> T_OP_ATOMIC_DEC
581 %token <tok> T_OP_ATOMIC_CMPXCHG
582 %token <tok> T_OP_ATOMIC_MIN
583 %token <tok> T_OP_ATOMIC_MAX
584 %token <tok> T_OP_ATOMIC_AND
585 %token <tok> T_OP_ATOMIC_OR
586 %token <tok> T_OP_ATOMIC_XOR
587 %token <tok> T_OP_RESINFO_B
588 %token <tok> T_OP_LDIB_B
589 %token <tok> T_OP_STIB_B
590 %token <tok> T_OP_ATOMIC_B_ADD
591 %token <tok> T_OP_ATOMIC_B_SUB
592 %token <tok> T_OP_ATOMIC_B_XCHG
593 %token <tok> T_OP_ATOMIC_B_INC
594 %token <tok> T_OP_ATOMIC_B_DEC
595 %token <tok> T_OP_ATOMIC_B_CMPXCHG
596 %token <tok> T_OP_ATOMIC_B_MIN
597 %token <tok> T_OP_ATOMIC_B_MAX
598 %token <tok> T_OP_ATOMIC_B_AND
599 %token <tok> T_OP_ATOMIC_B_OR
600 %token <tok> T_OP_ATOMIC_B_XOR
601 %token <tok> T_OP_ATOMIC_S_ADD
602 %token <tok> T_OP_ATOMIC_S_SUB
603 %token <tok> T_OP_ATOMIC_S_XCHG
604 %token <tok> T_OP_ATOMIC_S_INC
605 %token <tok> T_OP_ATOMIC_S_DEC
606 %token <tok> T_OP_ATOMIC_S_CMPXCHG
607 %token <tok> T_OP_ATOMIC_S_MIN
608 %token <tok> T_OP_ATOMIC_S_MAX
609 %token <tok> T_OP_ATOMIC_S_AND
610 %token <tok> T_OP_ATOMIC_S_OR
611 %token <tok> T_OP_ATOMIC_S_XOR
612 %token <tok> T_OP_ATOMIC_G_ADD
613 %token <tok> T_OP_ATOMIC_G_SUB
614 %token <tok> T_OP_ATOMIC_G_XCHG
615 %token <tok> T_OP_ATOMIC_G_INC
616 %token <tok> T_OP_ATOMIC_G_DEC
617 %token <tok> T_OP_ATOMIC_G_CMPXCHG
618 %token <tok> T_OP_ATOMIC_G_MIN
619 %token <tok> T_OP_ATOMIC_G_MAX
620 %token <tok> T_OP_ATOMIC_G_AND
621 %token <tok> T_OP_ATOMIC_G_OR
622 %token <tok> T_OP_ATOMIC_G_XOR
623 %token <tok> T_OP_LDGB
624 %token <tok> T_OP_STGB
625 %token <tok> T_OP_STIB
626 %token <tok> T_OP_LDC
627 %token <tok> T_OP_LDLV
628 %token <tok> T_OP_GETSPID
629 %token <tok> T_OP_GETWID
630 %token <tok> T_OP_GETFIBERID
631 %token <tok> T_OP_STC
632 %token <tok> T_OP_STSC
633
634 /* category 7: */
635 %token <tok> T_OP_BAR
636 %token <tok> T_OP_FENCE
637 %token <tok> T_OP_SLEEP
638 %token <tok> T_OP_ICINV
639 %token <tok> T_OP_DCCLN
640 %token <tok> T_OP_DCINV
641 %token <tok> T_OP_DCFLU
642 %token <tok> T_OP_CCINV
643 %token <tok> T_OP_LOCK
644 %token <tok> T_OP_UNLOCK
645 %token <tok> T_OP_ALIAS
646
647 %token <u64> T_RAW
648
649 %token <tok> T_OP_PRINT
650
651 /* type qualifiers: */
652 %token <tok> T_TYPE_F16
653 %token <tok> T_TYPE_F32
654 %token <tok> T_TYPE_U16
655 %token <tok> T_TYPE_U32
656 %token <tok> T_TYPE_S16
657 %token <tok> T_TYPE_S32
658 %token <tok> T_TYPE_U8
659 %token <tok> T_TYPE_S8
660
661 %token <tok> T_UNTYPED
662 %token <tok> T_TYPED
663
664 %token <tok> T_MIXED
665 %token <tok> T_UNSIGNED
666 %token <tok> T_LOW
667 %token <tok> T_HIGH
668
669 %token <tok> T_1D
670 %token <tok> T_2D
671 %token <tok> T_3D
672 %token <tok> T_4D
673
674 /* condition qualifiers: */
675 %token <tok> T_LT
676 %token <tok> T_LE
677 %token <tok> T_GT
678 %token <tok> T_GE
679 %token <tok> T_EQ
680 %token <tok> T_NE
681
682 %token <tok> T_S2EN
683 %token <tok> T_SAMP
684 %token <tok> T_TEX
685 %token <tok> T_BASE
686 %token <tok> T_OFFSET
687 %token <tok> T_UNIFORM
688 %token <tok> T_NONUNIFORM
689 %token <tok> T_IMM
690
691 %token <tok> T_NAN
692 %token <tok> T_INF
693 %token <num> T_A0
694 %token <num> T_A1
695 %token <num> T_P0
696 %token <num> T_W
697 %token <str> T_CAT1_TYPE_TYPE
698 %token <str> T_INSTR_TYPE
699
700 %token <tok> T_MOD_TEX
701 %token <tok> T_MOD_MEM
702 %token <tok> T_MOD_RT
703
704 %type <num> integer offset
705 %type <num> flut_immed
706 %type <flt> float
707 %type <reg> src dst const
708 %type <tok> cat1_opc
709 %type <tok> cat2_opc_1src cat2_opc_2src_cnd cat2_opc_2src
710 %type <tok> cat3_opc
711 %type <tok> cat4_opc
712 %type <tok> cat5_opc cat5_samp cat5_tex cat5_type
713 %type <type> type
714 %type <unum> const_val
715
716 %error-verbose
717
718 %start shader
719
720 %%
721
722 shader: { new_shader(); } headers instrs
723
724 headers:
725 | header headers
726
727 header: localsize_header
728 | const_header
729 | buf_header
730 | invocationid_header
731 | wgid_header
732 | numwg_header
733 | branchstack_header
734 | in_header
735 | out_header
736 | tex_header
737 | pvtmem_header
738 | earlypreamble_header
739
740 const_val: T_FLOAT { $$ = fui($1); }
741 | T_INT { $$ = $1; }
742 | '-' T_INT { $$ = -$2; }
743 | T_HEX { $$ = $1; }
744
745 localsize_header: T_A_LOCALSIZE const_val ',' const_val ',' const_val {
746 variant->local_size[0] = $2;
747 variant->local_size[1] = $4;
748 variant->local_size[2] = $6;
749 }
750
751 const_header: T_A_CONST '(' T_CONSTANT ')' const_val ',' const_val ',' const_val ',' const_val {
752 add_const($3, $5, $7, $9, $11);
753 }
754
755 buf_header_addr_reg:
756 '(' T_CONSTANT ')' {
757 assert(($2 & 0x1) == 0); /* half-reg not allowed */
758 unsigned reg = $2 >> 1;
759
760 info->buf_addr_regs[info->num_bufs - 1] = reg;
761 /* reserve space in immediates for the actual value to be plugged in later: */
762 add_const($2, 0, 0, 0, 0);
763 }
764 |
765
766 buf_header: T_A_BUF const_val {
767 int idx = info->num_bufs++;
768 assert(idx < MAX_BUFS);
769 info->buf_sizes[idx] = $2;
770 } buf_header_addr_reg
771
772 invocationid_header: T_A_INVOCATIONID '(' T_REGISTER ')' {
773 assert(($3 & 0x1) == 0); /* half-reg not allowed */
774 unsigned reg = $3 >> 1;
775 add_sysval(reg, 0x7, SYSTEM_VALUE_LOCAL_INVOCATION_ID);
776 }
777
778 wgid_header: T_A_WGID '(' T_REGISTER ')' {
779 assert(($3 & 0x1) == 0); /* half-reg not allowed */
780 unsigned reg = $3 >> 1;
781 assert(variant->compiler->gen >= 5);
782 assert(reg >= regid(48, 0)); /* must be a high reg */
783 add_sysval(reg, 0x7, SYSTEM_VALUE_WORKGROUP_ID);
784 }
785 | T_A_WGID '(' T_CONSTANT ')' {
786 assert(($3 & 0x1) == 0); /* half-reg not allowed */
787 unsigned reg = $3 >> 1;
788 assert(variant->compiler->gen < 5);
789 info->wgid = reg;
790 }
791
792 numwg_header: T_A_NUMWG '(' T_CONSTANT ')' {
793 assert(($3 & 0x1) == 0); /* half-reg not allowed */
794 unsigned reg = $3 >> 1;
795 info->numwg = reg;
796 /* reserve space in immediates for the actual value to be plugged in later: */
797 if (variant->compiler->gen >= 5)
798 add_const($3, 0, 0, 0, 0);
799 }
800
801 branchstack_header: T_A_BRANCHSTACK const_val { variant->branchstack = $2; }
802
803 pvtmem_header: T_A_PVTMEM const_val { variant->pvtmem_size = $2; }
804
805 earlypreamble_header: T_A_EARLYPREAMBLE { info->early_preamble = 1; }
806
807 /* Stubs for now */
808 in_header: T_A_IN '(' T_REGISTER ')' T_IDENTIFIER '(' T_IDENTIFIER '=' integer ')' { }
809
810 out_header: T_A_OUT '(' T_REGISTER ')' T_IDENTIFIER '(' T_IDENTIFIER '=' integer ')' { }
811
812 tex_header: T_A_TEX '(' T_REGISTER ')'
813 T_IDENTIFIER '=' integer ',' /* src */
814 T_IDENTIFIER '=' integer ',' /* samp */
815 T_IDENTIFIER '=' integer ',' /* tex */
816 T_IDENTIFIER '=' integer ',' /* wrmask */
817 T_IDENTIFIER '=' integer /* cmd */ { }
818
819 iflag: T_SY { iflags.flags |= IR3_INSTR_SY; }
820 | T_SS { iflags.flags |= IR3_INSTR_SS; }
821 | T_JP { iflags.flags |= IR3_INSTR_JP; }
822 | T_EQ_FLAG { iflags.flags |= IR3_INSTR_EQ; }
823 | T_SAT { iflags.flags |= IR3_INSTR_SAT; }
824 | T_RPT { iflags.repeat = $1; }
825 | T_UL { iflags.flags |= IR3_INSTR_UL; }
826 | T_NOP { iflags.nop = $1; }
827
828 iflags:
829 | iflag iflags
830
831 instrs: instrs instr
832 | instr
833
834 instr: iflags cat0_instr
835 | iflags cat1_instr
836 | iflags cat2_instr
837 | iflags cat3_instr
838 | iflags cat4_instr
839 | iflags cat5_instr { fixup_cat5_s2en(); }
840 | iflags cat6_instr
841 | iflags cat7_instr
842 | raw_instr
843 | meta_print
844 | label
845
846 label: T_IDENTIFIER ':' { new_label($1); }
847
848 cat0_src1: '!' T_P0 { instr->cat0.inv1 = true; instr->cat0.comp1 = $2 >> 1; }
849 | T_P0 { instr->cat0.comp1 = $1 >> 1; }
850
851 cat0_src2: '!' T_P0 { instr->cat0.inv2 = true; instr->cat0.comp2 = $2 >> 1; }
852 | T_P0 { instr->cat0.comp2 = $1 >> 1; }
853
854 cat0_immed: '#' integer { instr->cat0.immed = $2; }
855 | '#' T_IDENTIFIER { ralloc_steal(instr, (void *)$2); instr->cat0.target_label = $2; }
856
857 cat0_instr: T_OP_NOP { new_instr(OPC_NOP); }
858 | T_OP_BR { new_instr(OPC_B)->cat0.brtype = BRANCH_PLAIN; } cat0_src1 ',' cat0_immed
859 | T_OP_BRAO { new_instr(OPC_B)->cat0.brtype = BRANCH_OR; } cat0_src1 ',' cat0_src2 ',' cat0_immed
860 | T_OP_BRAA { new_instr(OPC_B)->cat0.brtype = BRANCH_AND; } cat0_src1 ',' cat0_src2 ',' cat0_immed
861 | T_OP_BRAC '.' integer { new_instr(OPC_B)->cat0.brtype = BRANCH_CONST; instr->cat0.idx = $3; } cat0_immed
862 | T_OP_BANY { new_instr(OPC_B)->cat0.brtype = BRANCH_ANY; } cat0_src1 ',' cat0_immed
863 | T_OP_BALL { new_instr(OPC_B)->cat0.brtype = BRANCH_ALL; } cat0_src1 ',' cat0_immed
864 | T_OP_BRAX { new_instr(OPC_B)->cat0.brtype = BRANCH_X; } cat0_immed
865 | T_OP_JUMP { new_instr(OPC_JUMP); } cat0_immed
866 | T_OP_CALL { new_instr(OPC_CALL); } cat0_immed
867 | T_OP_RET { new_instr(OPC_RET); }
868 | T_OP_KILL { new_instr(OPC_KILL); } cat0_src1
869 | T_OP_END { new_instr(OPC_END); }
870 | T_OP_EMIT { new_instr(OPC_EMIT); }
871 | T_OP_CUT { new_instr(OPC_CUT); }
872 | T_OP_CHMASK { new_instr(OPC_CHMASK); }
873 | T_OP_CHSH { new_instr(OPC_CHSH); }
874 | T_OP_FLOW_REV { new_instr(OPC_FLOW_REV); }
875 | T_OP_BKT { new_instr(OPC_BKT); } cat0_immed
876 | T_OP_STKS { new_instr(OPC_STKS); }
877 | T_OP_STKR { new_instr(OPC_STKR); }
878 | T_OP_XSET { new_instr(OPC_XSET); }
879 | T_OP_XCLR { new_instr(OPC_XCLR); }
880 | T_OP_GETONE { new_instr(OPC_GETONE); } cat0_immed
881 | T_OP_DBG { new_instr(OPC_DBG); }
882 | T_OP_SHPS { new_instr(OPC_SHPS); } cat0_immed
883 | T_OP_SHPE { new_instr(OPC_SHPE); }
884 | T_OP_PREDT { new_instr(OPC_PREDT); } cat0_src1
885 | T_OP_PREDF { new_instr(OPC_PREDF); } cat0_src1
886 | T_OP_PREDE { new_instr(OPC_PREDE); }
887 | T_OP_GETLAST '.' T_W { new_instr(OPC_GETLAST); } cat0_immed
888
889 cat1_opc: T_OP_MOV '.' T_CAT1_TYPE_TYPE {
890 parse_type_type(new_instr(OPC_MOV), $3);
891 }
892 | T_OP_COV '.' T_CAT1_TYPE_TYPE {
893 parse_type_type(new_instr(OPC_MOV), $3);
894 }
895
896 cat1_src: src_reg_or_const_or_rel
897 | immediate_cat1
898
899 cat1_movmsk: T_OP_MOVMSK '.' T_W {
900 new_instr(OPC_MOVMSK);
901 instr->cat1.src_type = TYPE_U32;
902 instr->cat1.dst_type = TYPE_U32;
903 } dst_reg {
904 if (($3 % 32) != 0)
905 yyerror("w# must be multiple of 32");
906 if ($3 < 32)
907 yyerror("w# must be at least 32");
908
909 int num = $3 / 32;
910
911 instr->repeat = num - 1;
912 instr->dsts[0]->wrmask = (1 << num) - 1;
913 }
914
915 cat1_mova1: T_OP_MOVA1 T_A1 ',' {
916 new_instr(OPC_MOV);
917 instr->cat1.src_type = TYPE_U16;
918 instr->cat1.dst_type = TYPE_U16;
919 new_dst((61 << 3) + 2, IR3_REG_HALF);
920 } cat1_src
921
922 cat1_mova: T_OP_MOVA T_A0 ',' {
923 new_instr(OPC_MOV);
924 instr->cat1.src_type = TYPE_S16;
925 instr->cat1.dst_type = TYPE_S16;
926 new_dst((61 << 3), IR3_REG_HALF);
927 } cat1_src
928
929 cat1_swz: T_OP_SWZ '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_SWZ), $3); } dst_reg ',' dst_reg ',' src_reg ',' src_reg
930
931 cat1_gat: T_OP_GAT '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_GAT), $3); } dst_reg ',' src_reg ',' src_reg ',' src_reg ',' src_reg
932
933 cat1_sct: T_OP_SCT '.' T_CAT1_TYPE_TYPE { parse_type_type(new_instr(OPC_SCT), $3); } dst_reg ',' dst_reg ',' dst_reg ',' dst_reg ',' src_reg
934
935 /* NOTE: cat1 can also *write* to relative gpr */
936 cat1_instr: cat1_movmsk
937 | cat1_mova1
938 | cat1_mova
939 | cat1_swz
940 | cat1_gat
941 | cat1_sct
942 | cat1_opc dst_reg ',' cat1_src
943 | cat1_opc relative_gpr_dst ',' cat1_src
944
945 cat2_opc_1src: T_OP_ABSNEG_F { new_instr(OPC_ABSNEG_F); }
946 | T_OP_ABSNEG_S { new_instr(OPC_ABSNEG_S); }
947 | T_OP_CLZ_B { new_instr(OPC_CLZ_B); }
948 | T_OP_CLZ_S { new_instr(OPC_CLZ_S); }
949 | T_OP_SIGN_F { new_instr(OPC_SIGN_F); }
950 | T_OP_FLOOR_F { new_instr(OPC_FLOOR_F); }
951 | T_OP_CEIL_F { new_instr(OPC_CEIL_F); }
952 | T_OP_RNDNE_F { new_instr(OPC_RNDNE_F); }
953 | T_OP_RNDAZ_F { new_instr(OPC_RNDAZ_F); }
954 | T_OP_TRUNC_F { new_instr(OPC_TRUNC_F); }
955 | T_OP_NOT_B { new_instr(OPC_NOT_B); }
956 | T_OP_BFREV_B { new_instr(OPC_BFREV_B); }
957 | T_OP_SETRM { new_instr(OPC_SETRM); }
958 | T_OP_CBITS_B { new_instr(OPC_CBITS_B); }
959
960 cat2_opc_2src_cnd: T_OP_CMPS_F { new_instr(OPC_CMPS_F); }
961 | T_OP_CMPS_U { new_instr(OPC_CMPS_U); }
962 | T_OP_CMPS_S { new_instr(OPC_CMPS_S); }
963 | T_OP_CMPV_F { new_instr(OPC_CMPV_F); }
964 | T_OP_CMPV_U { new_instr(OPC_CMPV_U); }
965 | T_OP_CMPV_S { new_instr(OPC_CMPV_S); }
966
967 cat2_opc_2src: T_OP_ADD_F { new_instr(OPC_ADD_F); }
968 | T_OP_MIN_F { new_instr(OPC_MIN_F); }
969 | T_OP_MAX_F { new_instr(OPC_MAX_F); }
970 | T_OP_MUL_F { new_instr(OPC_MUL_F); }
971 | T_OP_ADD_U { new_instr(OPC_ADD_U); }
972 | T_OP_ADD_S { new_instr(OPC_ADD_S); }
973 | T_OP_SUB_U { new_instr(OPC_SUB_U); }
974 | T_OP_SUB_S { new_instr(OPC_SUB_S); }
975 | T_OP_MIN_U { new_instr(OPC_MIN_U); }
976 | T_OP_MIN_S { new_instr(OPC_MIN_S); }
977 | T_OP_MAX_U { new_instr(OPC_MAX_U); }
978 | T_OP_MAX_S { new_instr(OPC_MAX_S); }
979 | T_OP_AND_B { new_instr(OPC_AND_B); }
980 | T_OP_OR_B { new_instr(OPC_OR_B); }
981 | T_OP_XOR_B { new_instr(OPC_XOR_B); }
982 | T_OP_MUL_U24 { new_instr(OPC_MUL_U24); }
983 | T_OP_MUL_S24 { new_instr(OPC_MUL_S24); }
984 | T_OP_MULL_U { new_instr(OPC_MULL_U); }
985 | T_OP_SHL_B { new_instr(OPC_SHL_B); }
986 | T_OP_SHR_B { new_instr(OPC_SHR_B); }
987 | T_OP_ASHR_B { new_instr(OPC_ASHR_B); }
988 | T_OP_BARY_F { new_instr(OPC_BARY_F); }
989 | T_OP_FLAT_B { new_instr(OPC_FLAT_B); }
990 | T_OP_MGEN_B { new_instr(OPC_MGEN_B); }
991 | T_OP_GETBIT_B { new_instr(OPC_GETBIT_B); }
992 | T_OP_SHB { new_instr(OPC_SHB); }
993 | T_OP_MSAD { new_instr(OPC_MSAD); }
994
995 cond: T_LT { instr->cat2.condition = IR3_COND_LT; }
996 | T_LE { instr->cat2.condition = IR3_COND_LE; }
997 | T_GT { instr->cat2.condition = IR3_COND_GT; }
998 | T_GE { instr->cat2.condition = IR3_COND_GE; }
999 | T_EQ { instr->cat2.condition = IR3_COND_EQ; }
1000 | T_NE { instr->cat2.condition = IR3_COND_NE; }
1001
1002 cat2_instr: cat2_opc_1src dst_reg ',' src_reg_or_const_or_rel_or_imm
1003 | cat2_opc_2src_cnd '.' cond dst_reg ',' src_reg_or_const_or_rel_or_imm ',' src_reg_or_const_or_rel_or_imm
1004 | cat2_opc_2src dst_reg ',' src_reg_or_const_or_rel_or_imm ',' src_reg_or_const_or_rel_or_imm
1005
1006 cat3_dp_signedness:'.' T_MIXED { instr->cat3.signedness = IR3_SRC_MIXED; }
1007 | '.' T_UNSIGNED{ instr->cat3.signedness = IR3_SRC_UNSIGNED; }
1008
1009 cat3_dp_pack: '.' T_LOW { instr->cat3.packed = IR3_SRC_PACKED_LOW; }
1010 | '.' T_HIGH { instr->cat3.packed = IR3_SRC_PACKED_HIGH; }
1011
1012 cat3_opc: T_OP_MAD_U16 { new_instr(OPC_MAD_U16); }
1013 | T_OP_MADSH_U16 { new_instr(OPC_MADSH_U16); }
1014 | T_OP_MAD_S16 { new_instr(OPC_MAD_S16); }
1015 | T_OP_MADSH_M16 { new_instr(OPC_MADSH_M16); }
1016 | T_OP_MAD_U24 { new_instr(OPC_MAD_U24); }
1017 | T_OP_MAD_S24 { new_instr(OPC_MAD_S24); }
1018 | T_OP_MAD_F16 { new_instr(OPC_MAD_F16); }
1019 | T_OP_MAD_F32 { new_instr(OPC_MAD_F32); }
1020 | T_OP_SEL_B16 { new_instr(OPC_SEL_B16); }
1021 | T_OP_SEL_B32 { new_instr(OPC_SEL_B32); }
1022 | T_OP_SEL_S16 { new_instr(OPC_SEL_S16); }
1023 | T_OP_SEL_S32 { new_instr(OPC_SEL_S32); }
1024 | T_OP_SEL_F16 { new_instr(OPC_SEL_F16); }
1025 | T_OP_SEL_F32 { new_instr(OPC_SEL_F32); }
1026 | T_OP_SAD_S16 { new_instr(OPC_SAD_S16); }
1027 | T_OP_SAD_S32 { new_instr(OPC_SAD_S32); }
1028
1029 cat3_imm_reg_opc: T_OP_SHRM { new_instr(OPC_SHRM); }
1030 | T_OP_SHLM { new_instr(OPC_SHLM); }
1031 | T_OP_SHRG { new_instr(OPC_SHRG); }
1032 | T_OP_SHLG { new_instr(OPC_SHLG); }
1033 | T_OP_ANDG { new_instr(OPC_ANDG); }
1034
1035 cat3_wmm: T_OP_WMM { new_instr(OPC_WMM); }
1036 | T_OP_WMM_ACCU { new_instr(OPC_WMM_ACCU); }
1037
1038 cat3_dp: T_OP_DP2ACC { new_instr(OPC_DP2ACC); }
1039 | T_OP_DP4ACC { new_instr(OPC_DP4ACC); }
1040
1041 cat3_instr: cat3_opc dst_reg ',' src_reg_or_const_or_rel ',' src_reg_or_const ',' src_reg_or_const_or_rel
1042 | cat3_imm_reg_opc dst_reg ',' src_reg_or_rel_or_imm ',' src_reg_or_const ',' src_reg_or_rel_or_imm
1043 | cat3_wmm dst_reg ',' src_reg_gpr ',' src_reg ',' immediate
1044 | cat3_dp cat3_dp_signedness cat3_dp_pack dst_reg ',' src_reg_or_rel_or_imm ',' src_reg_or_const ',' src_reg_or_rel_or_imm
1045
1046 cat4_opc: T_OP_RCP { new_instr(OPC_RCP); }
1047 | T_OP_RSQ { new_instr(OPC_RSQ); }
1048 | T_OP_LOG2 { new_instr(OPC_LOG2); }
1049 | T_OP_EXP2 { new_instr(OPC_EXP2); }
1050 | T_OP_SIN { new_instr(OPC_SIN); }
1051 | T_OP_COS { new_instr(OPC_COS); }
1052 | T_OP_SQRT { new_instr(OPC_SQRT); }
1053 | T_OP_HRSQ { new_instr(OPC_HRSQ); }
1054 | T_OP_HLOG2 { new_instr(OPC_HLOG2); }
1055 | T_OP_HEXP2 { new_instr(OPC_HEXP2); }
1056
1057 cat4_instr: cat4_opc dst_reg ',' src_reg_or_const_or_rel_or_imm
1058
1059 cat5_opc_dsxypp: T_OP_DSXPP_1 { new_instr(OPC_DSXPP_1)->cat5.type = TYPE_F32; }
1060 | T_OP_DSYPP_1 { new_instr(OPC_DSYPP_1)->cat5.type = TYPE_F32; }
1061
1062 cat5_opc: T_OP_ISAM { new_instr(OPC_ISAM); }
1063 | T_OP_ISAML { new_instr(OPC_ISAML); }
1064 | T_OP_ISAMM { new_instr(OPC_ISAMM); }
1065 | T_OP_SAM { new_instr(OPC_SAM); }
1066 | T_OP_SAMB { new_instr(OPC_SAMB); }
1067 | T_OP_SAML { new_instr(OPC_SAML); }
1068 | T_OP_SAMGQ { new_instr(OPC_SAMGQ); }
1069 | T_OP_GETLOD { new_instr(OPC_GETLOD); }
1070 | T_OP_CONV { new_instr(OPC_CONV); }
1071 | T_OP_CONVM { new_instr(OPC_CONVM); }
1072 | T_OP_GETSIZE { new_instr(OPC_GETSIZE); }
1073 | T_OP_GETBUF { new_instr(OPC_GETBUF); }
1074 | T_OP_GETPOS { new_instr(OPC_GETPOS); }
1075 | T_OP_GETINFO { new_instr(OPC_GETINFO); }
1076 | T_OP_DSX { new_instr(OPC_DSX); }
1077 | T_OP_DSY { new_instr(OPC_DSY); }
1078 | T_OP_GATHER4R { new_instr(OPC_GATHER4R); }
1079 | T_OP_GATHER4G { new_instr(OPC_GATHER4G); }
1080 | T_OP_GATHER4B { new_instr(OPC_GATHER4B); }
1081 | T_OP_GATHER4A { new_instr(OPC_GATHER4A); }
1082 | T_OP_SAMGP0 { new_instr(OPC_SAMGP0); }
1083 | T_OP_SAMGP1 { new_instr(OPC_SAMGP1); }
1084 | T_OP_SAMGP2 { new_instr(OPC_SAMGP2); }
1085 | T_OP_SAMGP3 { new_instr(OPC_SAMGP3); }
1086 | T_OP_RGETPOS { new_instr(OPC_RGETPOS); }
1087 | T_OP_RGETINFO { new_instr(OPC_RGETINFO); }
1088 | T_OP_BRCST_A { new_instr(OPC_BRCST_ACTIVE); }
1089 | T_OP_QSHUFFLE_BRCST { new_instr(OPC_QUAD_SHUFFLE_BRCST); }
1090 | T_OP_QSHUFFLE_H { new_instr(OPC_QUAD_SHUFFLE_HORIZ); }
1091 | T_OP_QSHUFFLE_V { new_instr(OPC_QUAD_SHUFFLE_VERT); }
1092 | T_OP_QSHUFFLE_DIAG { new_instr(OPC_QUAD_SHUFFLE_DIAG); }
1093
1094 cat5_flag: '.' T_3D { instr->flags |= IR3_INSTR_3D; }
1095 | '.' 'a' { instr->flags |= IR3_INSTR_A; }
1096 | '.' 'o' { instr->flags |= IR3_INSTR_O; }
1097 | '.' 'p' { instr->flags |= IR3_INSTR_P; }
1098 | '.' 's' { instr->flags |= IR3_INSTR_S; }
1099 | '.' T_S2EN { instr->flags |= IR3_INSTR_S2EN; }
1100 | '.' T_UNIFORM { }
1101 | '.' T_NONUNIFORM { instr->flags |= IR3_INSTR_NONUNIF; }
1102 | '.' T_BASE { instr->flags |= IR3_INSTR_B; instr->cat5.tex_base = $2; }
1103 | '.' T_W { instr->cat5.cluster_size = $2; }
1104 cat5_flags:
1105 | cat5_flag cat5_flags
1106
1107 cat5_samp: T_SAMP { instr->cat5.samp = $1; }
1108 cat5_tex: T_TEX { instr->cat5.tex = $1; }
1109 cat5_type: '(' type ')' { instr->cat5.type = $2; }
1110 cat5_a1: src_reg { instr->flags |= IR3_INSTR_A1EN; }
1111
1112 cat5_instr: cat5_opc_dsxypp cat5_flags dst_reg ',' src_reg
1113 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' src_reg
1114 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' cat5_samp ',' cat5_tex
1115 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' cat5_samp ',' cat5_a1
1116 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' cat5_tex ',' cat5_a1
1117 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' cat5_samp
1118 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg ',' cat5_tex
1119 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' src_reg
1120 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' cat5_samp ',' cat5_tex
1121 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' cat5_samp ',' cat5_a1
1122 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' cat5_tex ',' cat5_a1
1123 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' cat5_samp
1124 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg ',' cat5_tex
1125 | cat5_opc cat5_flags cat5_type dst_reg ',' src_reg
1126 | cat5_opc cat5_flags cat5_type dst_reg ',' cat5_samp ',' cat5_tex
1127 | cat5_opc cat5_flags cat5_type dst_reg ',' cat5_samp
1128 | cat5_opc cat5_flags cat5_type dst_reg ',' cat5_tex
1129 | cat5_opc cat5_flags cat5_type dst_reg
1130 | T_OP_TCINV { new_instr(OPC_TCINV); }
1131
1132 cat6_typed: '.' T_UNTYPED { instr->cat6.typed = 0; }
1133 | '.' T_TYPED { instr->cat6.typed = 1; }
1134
1135 cat6_dim: '.' T_1D { instr->cat6.d = 1; }
1136 | '.' T_2D { instr->cat6.d = 2; }
1137 | '.' T_3D { instr->cat6.d = 3; }
1138 | '.' T_4D { instr->cat6.d = 4; }
1139
1140 cat6_type: '.' type { instr->cat6.type = $2; }
1141 cat6_imm_offset: offset { new_src(0, IR3_REG_IMMED)->iim_val = $1; }
1142 cat6_offset: cat6_imm_offset
1143 | '+' src
1144 cat6_dst_offset: offset { instr->cat6.dst_offset = $1; }
1145 | '+' src
1146
1147 cat6_immed: integer { instr->cat6.iim_val = $1; }
1148
1149 cat6_a6xx_global_address_pt3:
1150 '<' '<' integer offset '<' '<' integer {
1151 assert($7 == 2);
1152 new_src(0, IR3_REG_IMMED)->uim_val = $3 - 2;
1153 new_src(0, IR3_REG_IMMED)->uim_val = $4;
1154 }
1155 | '+' cat6_reg_or_immed {
1156 // Dummy src to smooth the difference between a6xx and a7xx
1157 new_src(0, IR3_REG_IMMED)->uim_val = 0;
1158 }
1159
1160 cat6_a6xx_global_address_pt2:
1161 '(' src offset ')' '<' '<' integer {
1162 assert($7 == 2);
1163 new_src(0, IR3_REG_IMMED)->uim_val = 0;
1164 new_src(0, IR3_REG_IMMED)->uim_val = $3;
1165 }
1166
1167 | src cat6_a6xx_global_address_pt3
1168
1169 cat6_a6xx_global_address:
1170 src_reg_or_const '+' cat6_a6xx_global_address_pt2
1171
1172 cat6_load: T_OP_LDG { new_instr(OPC_LDG); } cat6_type dst_reg ',' 'g' '[' src cat6_offset ']' ',' immediate
1173 | T_OP_LDG_A { new_instr(OPC_LDG_A); } cat6_type dst_reg ',' 'g' '[' cat6_a6xx_global_address ']' ',' immediate
1174 | T_OP_LDG_K { new_instr(OPC_LDG_K); } cat6_type 'c' '[' const_dst ']' ',' 'g' '[' src cat6_offset ']' ',' immediate
1175 | T_OP_LDP { new_instr(OPC_LDP); } cat6_type dst_reg ',' 'p' '[' src cat6_offset ']' ',' immediate
1176 | T_OP_LDL { new_instr(OPC_LDL); } cat6_type dst_reg ',' 'l' '[' src cat6_offset ']' ',' immediate
1177 | T_OP_LDLW { new_instr(OPC_LDLW); } cat6_type dst_reg ',' 'l' '[' src cat6_offset ']' ',' immediate
1178 | T_OP_LDLV { new_instr(OPC_LDLV); } cat6_type dst_reg ',' 'l' '[' integer ']' {
1179 new_src(0, IR3_REG_IMMED)->iim_val = $8;
1180 } ',' immediate
1181
1182 cat6_store: T_OP_STG { new_instr(OPC_STG); dummy_dst(); } cat6_type 'g' '[' src cat6_imm_offset ']' ',' src ',' immediate
1183 | T_OP_STG_A { new_instr(OPC_STG_A); dummy_dst(); } cat6_type 'g' '[' cat6_a6xx_global_address ']' ',' src ',' immediate
1184 | T_OP_STP { new_instr(OPC_STP); dummy_dst(); } cat6_type 'p' '[' src cat6_dst_offset ']' ',' src ',' immediate
1185 | T_OP_STL { new_instr(OPC_STL); dummy_dst(); } cat6_type 'l' '[' src cat6_dst_offset ']' ',' src ',' immediate
1186 | T_OP_STLW { new_instr(OPC_STLW); dummy_dst(); } cat6_type 'l' '[' src cat6_dst_offset ']' ',' src ',' immediate
1187
1188 cat6_loadib: T_OP_LDIB { new_instr(OPC_LDIB); } cat6_typed cat6_dim cat6_type '.' cat6_immed dst_reg ',' 'g' '[' immediate ']' ',' src ',' src
1189 cat6_storeib: T_OP_STIB { new_instr(OPC_STIB); dummy_dst(); } cat6_typed cat6_dim cat6_type '.' cat6_immed'g' '[' immediate ']' ',' src ',' src ',' src
1190
1191 cat6_prefetch: T_OP_PREFETCH { new_instr(OPC_PREFETCH); new_dst(0,0); /* dummy dst */ } 'g' '[' src cat6_offset ']' ',' cat6_immed
1192
1193 cat6_atomic_opc: T_OP_ATOMIC_ADD { new_instr(OPC_ATOMIC_ADD); }
1194 | T_OP_ATOMIC_SUB { new_instr(OPC_ATOMIC_SUB); }
1195 | T_OP_ATOMIC_XCHG { new_instr(OPC_ATOMIC_XCHG); }
1196 | T_OP_ATOMIC_INC { new_instr(OPC_ATOMIC_INC); }
1197 | T_OP_ATOMIC_DEC { new_instr(OPC_ATOMIC_DEC); }
1198 | T_OP_ATOMIC_CMPXCHG { new_instr(OPC_ATOMIC_CMPXCHG); }
1199 | T_OP_ATOMIC_MIN { new_instr(OPC_ATOMIC_MIN); }
1200 | T_OP_ATOMIC_MAX { new_instr(OPC_ATOMIC_MAX); }
1201 | T_OP_ATOMIC_AND { new_instr(OPC_ATOMIC_AND); }
1202 | T_OP_ATOMIC_OR { new_instr(OPC_ATOMIC_OR); }
1203 | T_OP_ATOMIC_XOR { new_instr(OPC_ATOMIC_XOR); }
1204
1205 cat6_a3xx_atomic_opc: T_OP_ATOMIC_S_ADD { new_instr(OPC_ATOMIC_S_ADD); }
1206 | T_OP_ATOMIC_S_SUB { new_instr(OPC_ATOMIC_S_SUB); }
1207 | T_OP_ATOMIC_S_XCHG { new_instr(OPC_ATOMIC_S_XCHG); }
1208 | T_OP_ATOMIC_S_INC { new_instr(OPC_ATOMIC_S_INC); }
1209 | T_OP_ATOMIC_S_DEC { new_instr(OPC_ATOMIC_S_DEC); }
1210 | T_OP_ATOMIC_S_CMPXCHG { new_instr(OPC_ATOMIC_S_CMPXCHG); }
1211 | T_OP_ATOMIC_S_MIN { new_instr(OPC_ATOMIC_S_MIN); }
1212 | T_OP_ATOMIC_S_MAX { new_instr(OPC_ATOMIC_S_MAX); }
1213 | T_OP_ATOMIC_S_AND { new_instr(OPC_ATOMIC_S_AND); }
1214 | T_OP_ATOMIC_S_OR { new_instr(OPC_ATOMIC_S_OR); }
1215 | T_OP_ATOMIC_S_XOR { new_instr(OPC_ATOMIC_S_XOR); }
1216
1217 cat6_a6xx_atomic_opc: T_OP_ATOMIC_G_ADD { new_instr(OPC_ATOMIC_G_ADD); }
1218 | T_OP_ATOMIC_G_SUB { new_instr(OPC_ATOMIC_G_SUB); }
1219 | T_OP_ATOMIC_G_XCHG { new_instr(OPC_ATOMIC_G_XCHG); }
1220 | T_OP_ATOMIC_G_INC { new_instr(OPC_ATOMIC_G_INC); }
1221 | T_OP_ATOMIC_G_DEC { new_instr(OPC_ATOMIC_G_DEC); }
1222 | T_OP_ATOMIC_G_CMPXCHG { new_instr(OPC_ATOMIC_G_CMPXCHG); }
1223 | T_OP_ATOMIC_G_MIN { new_instr(OPC_ATOMIC_G_MIN); }
1224 | T_OP_ATOMIC_G_MAX { new_instr(OPC_ATOMIC_G_MAX); }
1225 | T_OP_ATOMIC_G_AND { new_instr(OPC_ATOMIC_G_AND); }
1226 | T_OP_ATOMIC_G_OR { new_instr(OPC_ATOMIC_G_OR); }
1227 | T_OP_ATOMIC_G_XOR { new_instr(OPC_ATOMIC_G_XOR); }
1228
1229 cat6_a3xx_atomic_s: cat6_a3xx_atomic_opc cat6_typed cat6_dim cat6_type '.' cat6_immed '.' 'g' dst_reg ',' 'g' '[' cat6_reg_or_immed ']' ',' src ',' src ',' src
1230
1231 cat6_a6xx_atomic_g: cat6_a6xx_atomic_opc cat6_typed cat6_dim cat6_type '.' cat6_immed '.' 'g' dst_reg ',' src ',' src
1232
1233 cat6_atomic_l: cat6_atomic_opc cat6_typed cat6_dim cat6_type '.' cat6_immed '.' 'l' dst_reg ',' 'l' '[' cat6_reg_or_immed ']' ',' src
1234
1235 cat6_atomic: cat6_atomic_l
1236 | cat6_a3xx_atomic_s
1237 | cat6_a6xx_atomic_g
1238
1239 cat6_ibo_opc_1src: T_OP_RESINFO { new_instr(OPC_RESINFO); }
1240
1241 cat6_ibo_opc_ldgb: T_OP_LDGB { new_instr(OPC_LDGB); }
1242 cat6_ibo_opc_stgb: T_OP_STGB { new_instr(OPC_STGB); }
1243
1244 cat6_ibo: cat6_ibo_opc_1src cat6_type cat6_dim dst_reg ',' 'g' '[' cat6_reg_or_immed ']'
1245 | cat6_ibo_opc_ldgb cat6_typed cat6_dim cat6_type '.' cat6_immed dst_reg ',' 'g' '[' cat6_reg_or_immed ']' ',' src ',' src
1246 | cat6_ibo_opc_stgb cat6_typed cat6_dim cat6_type '.' cat6_immed { dummy_dst(); } 'g' '[' cat6_reg_or_immed ']' ',' src ',' cat6_reg_or_immed ',' src
1247
1248 cat6_id_opc:
1249 T_OP_GETSPID { new_instr(OPC_GETSPID); }
1250 | T_OP_GETWID { new_instr(OPC_GETWID); }
1251 | T_OP_GETFIBERID { new_instr(OPC_GETFIBERID); }
1252
1253 cat6_id: cat6_id_opc cat6_type dst_reg
1254
1255 cat6_bindless_base:
1256 | '.' T_BASE { instr->flags |= IR3_INSTR_B; instr->cat6.base = $2; }
1257
1258 cat6_bindless_mode: T_IMM cat6_bindless_base
1259 | T_UNIFORM cat6_bindless_base
1260 | T_NONUNIFORM cat6_bindless_base { instr->flags |= IR3_INSTR_NONUNIF; }
1261
1262 cat6_reg_or_immed: src
1263 | integer { new_src(0, IR3_REG_IMMED)->iim_val = $1; }
1264
1265 cat6_bindless_ibo_opc_1src: T_OP_RESINFO_B { new_instr(OPC_RESINFO); }
1266
1267 cat6_bindless_ibo_opc_2src: T_OP_ATOMIC_B_ADD { new_instr(OPC_ATOMIC_B_ADD); dummy_dst(); }
1268 | T_OP_ATOMIC_B_SUB { new_instr(OPC_ATOMIC_B_SUB); dummy_dst(); }
1269 | T_OP_ATOMIC_B_XCHG { new_instr(OPC_ATOMIC_B_XCHG); dummy_dst(); }
1270 | T_OP_ATOMIC_B_INC { new_instr(OPC_ATOMIC_B_INC); dummy_dst(); }
1271 | T_OP_ATOMIC_B_DEC { new_instr(OPC_ATOMIC_B_DEC); dummy_dst(); }
1272 | T_OP_ATOMIC_B_CMPXCHG { new_instr(OPC_ATOMIC_B_CMPXCHG); dummy_dst(); }
1273 | T_OP_ATOMIC_B_MIN { new_instr(OPC_ATOMIC_B_MIN); dummy_dst(); }
1274 | T_OP_ATOMIC_B_MAX { new_instr(OPC_ATOMIC_B_MAX); dummy_dst(); }
1275 | T_OP_ATOMIC_B_AND { new_instr(OPC_ATOMIC_B_AND); dummy_dst(); }
1276 | T_OP_ATOMIC_B_OR { new_instr(OPC_ATOMIC_B_OR); dummy_dst(); }
1277 | T_OP_ATOMIC_B_XOR { new_instr(OPC_ATOMIC_B_XOR); dummy_dst(); }
1278 | T_OP_STIB_B { new_instr(OPC_STIB); dummy_dst(); }
1279
1280 cat6_bindless_ibo_opc_2src_dst: T_OP_LDIB_B { new_instr(OPC_LDIB); }
1281
1282 cat6_bindless_ibo: cat6_bindless_ibo_opc_1src cat6_typed cat6_dim cat6_type '.' cat6_immed '.' cat6_bindless_mode dst_reg ',' cat6_reg_or_immed
1283 | cat6_bindless_ibo_opc_2src cat6_typed cat6_dim cat6_type '.' cat6_immed '.' cat6_bindless_mode src_reg ',' cat6_reg_or_immed ',' cat6_reg_or_immed { swap(instr->srcs[0], instr->srcs[2]); }
1284 | cat6_bindless_ibo_opc_2src_dst cat6_typed cat6_dim cat6_type '.' cat6_immed '.' cat6_bindless_mode dst_reg ',' cat6_reg_or_immed ',' cat6_reg_or_immed { swap(instr->srcs[0], instr->srcs[1]); }
1285
1286 cat6_bindless_ldc_opc: T_OP_LDC { new_instr(OPC_LDC); }
1287
1288 /* This is separated from the opcode to avoid lookahead/shift-reduce conflicts */
1289 cat6_bindless_ldc_middle:
1290 T_OFFSET '.' cat6_immed '.' cat6_bindless_mode dst_reg { instr->cat6.d = $1; }
1291 | cat6_immed '.' 'k' '.' cat6_bindless_mode 'c' '[' T_A1 ']' { instr->opc = OPC_LDC_K; }
1292
1293 cat6_bindless_ldc: cat6_bindless_ldc_opc '.' cat6_bindless_ldc_middle ',' cat6_reg_or_immed ',' cat6_reg_or_immed {
1294 instr->cat6.type = TYPE_U32;
1295 /* TODO cleanup ir3 src order: */
1296 swap(instr->srcs[0], instr->srcs[1]);
1297 }
1298
1299 const_dst: integer { new_src(0, IR3_REG_IMMED)->iim_val = $1; }
1300 | T_A1 { new_src(0, IR3_REG_IMMED)->iim_val = 0; instr->flags |= IR3_INSTR_A1EN; }
1301 | T_A1 '+' integer { new_src(0, IR3_REG_IMMED)->iim_val = $3; instr->flags |= IR3_INSTR_A1EN; }
1302
1303 cat6_stc:
1304 T_OP_STC { new_instr(OPC_STC); } cat6_type 'c' '[' const_dst ']' ',' src_reg ',' cat6_immed
1305 | T_OP_STSC { new_instr(OPC_STSC); } cat6_type 'c' '[' const_dst ']' ',' immediate ',' cat6_immed
1306
1307 cat6_todo: T_OP_G2L { new_instr(OPC_G2L); }
1308 | T_OP_L2G { new_instr(OPC_L2G); }
1309 | T_OP_RESFMT { new_instr(OPC_RESFMT); }
1310
1311 cat6_instr: cat6_load
1312 | cat6_loadib
1313 | cat6_store
1314 | cat6_storeib
1315 | cat6_prefetch
1316 | cat6_atomic
1317 | cat6_ibo
1318 | cat6_id
1319 | cat6_bindless_ldc
1320 | cat6_bindless_ibo
1321 | cat6_stc
1322 | cat6_todo
1323
1324 cat7_scope: '.' 'w' { instr->cat7.w = true; }
1325 | '.' 'r' { instr->cat7.r = true; }
1326 | '.' 'l' { instr->cat7.l = true; }
1327 | '.' 'g' { instr->cat7.g = true; }
1328
1329 cat7_scopes:
1330 | cat7_scope cat7_scopes
1331
1332 cat7_barrier: T_OP_BAR { new_instr(OPC_BAR); } cat7_scopes
1333 | T_OP_FENCE { new_instr(OPC_FENCE); } cat7_scopes
1334
1335 cat7_data_cache: T_OP_DCCLN { new_instr(OPC_DCCLN); }
1336 | T_OP_DCINV { new_instr(OPC_DCINV); }
1337 | T_OP_DCFLU { new_instr(OPC_DCFLU); }
1338
1339 cat7_alias_src: src_reg_or_const
1340 | immediate_cat1
1341
1342 cat7_alias_scope: T_MOD_TEX { instr->cat7.alias_scope = ALIAS_TEX; }
1343 | T_MOD_MEM { instr->cat7.alias_scope = ALIAS_MEM; }
1344 | T_MOD_RT { instr->cat7.alias_scope = ALIAS_RT; }
1345
1346 cat7_instr: cat7_barrier
1347 | cat7_data_cache
1348 | T_OP_SLEEP { new_instr(OPC_SLEEP); }
1349 | T_OP_CCINV { new_instr(OPC_CCINV); }
1350 | T_OP_ICINV { new_instr(OPC_ICINV); }
1351 | T_OP_LOCK { new_instr(OPC_LOCK); }
1352 | T_OP_UNLOCK { new_instr(OPC_UNLOCK); }
1353 | T_OP_ALIAS {
1354 /* TODO: handle T_INSTR_TYPE */
1355 new_instr(OPC_ALIAS);
1356 } '.' cat7_alias_scope '.' T_INSTR_TYPE '.' integer dst_reg ',' cat7_alias_src {
1357 new_src(0, IR3_REG_IMMED)->uim_val = $8;
1358 }
1359
1360 raw_instr: T_RAW {new_instr(OPC_META_RAW)->raw.value = $1;}
1361
1362 meta_print_regs: meta_print_reg
1363 | meta_print_reg meta_print_regs
1364
1365 meta_print_reg: ',' T_REGISTER {
1366 meta_print_data.regs_to_dump[meta_print_data.regs_count++] = $2;
1367 }
1368
1369 meta_print_start: T_OP_PRINT T_REGISTER {
1370 meta_print_data.reg_address_lo = $2;
1371 meta_print_data.reg_address_hi = $2 + 2;
1372 meta_print_data.reg_tmp = $2 + 4;
1373 meta_print_data.regs_count = 0;
1374 }
1375
1376 meta_print: meta_print_start meta_print_regs {
1377 /* low */
1378 new_instr(OPC_MOV);
1379 instr->cat1.src_type = TYPE_U32;
1380 instr->cat1.dst_type = TYPE_U32;
1381 new_dst(meta_print_data.reg_address_lo, 0);
1382 new_src(0, IR3_REG_IMMED)->uim_val = info->shader_print_buffer_iova & 0xffffffff;
1383
1384 /* high */
1385 new_instr(OPC_MOV);
1386 instr->cat1.src_type = TYPE_U32;
1387 instr->cat1.dst_type = TYPE_U32;
1388 new_dst(meta_print_data.reg_address_hi, 0);
1389 new_src(0, IR3_REG_IMMED)->uim_val = info->shader_print_buffer_iova >> 32;
1390
1391 /* offset */
1392 new_instr(OPC_MOV);
1393 instr->cat1.src_type = TYPE_U32;
1394 instr->cat1.dst_type = TYPE_U32;
1395 new_dst(meta_print_data.reg_tmp, 0);
1396 new_src(0, IR3_REG_IMMED)->uim_val = 4 * meta_print_data.regs_count;
1397
1398 new_instr(OPC_NOP);
1399 instr->repeat = 5;
1400
1401 /* Increment and get current offset into print buffer */
1402 new_instr(OPC_ATOMIC_G_ADD);
1403 instr->cat6.d = 1;
1404 instr->cat6.typed = 0;
1405 instr->cat6.type = TYPE_U32;
1406 instr->cat6.iim_val = 1;
1407
1408 new_dst(meta_print_data.reg_address_lo, 0);
1409 new_src(meta_print_data.reg_address_lo, 0);
1410 new_src(meta_print_data.reg_tmp, 0);
1411
1412 /* Store all regs */
1413 for (uint32_t i = 0; i < meta_print_data.regs_count; i++) {
1414 new_instr(OPC_STG);
1415 dummy_dst();
1416 instr->cat6.type = TYPE_U32;
1417 instr->flags = IR3_INSTR_SY;
1418 new_src(meta_print_data.reg_address_lo, 0);
1419 new_src(0, IR3_REG_IMMED)->iim_val = 0;
1420 new_src(meta_print_data.regs_to_dump[i], IR3_REG_R);
1421 new_src(0, IR3_REG_IMMED)->iim_val = 1;
1422
1423 new_instr(OPC_ADD_U);
1424 instr->flags = IR3_INSTR_SS;
1425 new_dst(meta_print_data.reg_address_lo, 0);
1426 new_src(meta_print_data.reg_address_lo, 0);
1427 new_src(0, IR3_REG_IMMED)->uim_val = 4;
1428
1429 new_instr(OPC_NOP);
1430 instr->repeat = 5;
1431 }
1432 }
1433
1434 src: T_REGISTER { $$ = new_src($1, 0); }
1435 | T_A0 { $$ = new_src((61 << 3), IR3_REG_HALF); }
1436 | T_A1 { $$ = new_src((61 << 3) + 1, IR3_REG_HALF); }
1437 | T_P0 { $$ = new_src((62 << 3) + $1, 0); }
1438
1439 dst: T_REGISTER { $$ = new_dst($1, 0); }
1440 | T_A0 { $$ = new_dst((61 << 3), IR3_REG_HALF); }
1441 | T_A1 { $$ = new_dst((61 << 3) + 1, IR3_REG_HALF); }
1442 | T_P0 { $$ = new_dst((62 << 3) + $1, 0); }
1443
1444 const: T_CONSTANT { $$ = new_src($1, IR3_REG_CONST); }
1445
1446 dst_reg_flag: T_EVEN { instr->cat1.round = ROUND_EVEN; }
1447 | T_POS_INFINITY { instr->cat1.round = ROUND_POS_INF; }
1448 | T_NEG_INFINITY { instr->cat1.round = ROUND_NEG_INF; }
1449 | T_EI { rflags.flags |= IR3_REG_EI; }
1450 | T_WRMASK { rflags.wrmask = $1; }
1451
1452 dst_reg_flags: dst_reg_flag
1453 | dst_reg_flag dst_reg_flags
1454
1455 /* note: destination registers are always incremented in repeat */
1456 dst_reg: dst { $1->flags |= IR3_REG_R; }
1457 | dst_reg_flags dst { $2->flags |= IR3_REG_R; }
1458
1459 src_reg_flag: T_ABSNEG { rflags.flags |= IR3_REG_ABS|IR3_REG_NEGATE; }
1460 | T_NEG { rflags.flags |= IR3_REG_NEGATE; }
1461 | T_ABS { rflags.flags |= IR3_REG_ABS; }
1462 | T_R { rflags.flags |= IR3_REG_R; }
1463 | T_LAST { rflags.flags |= IR3_REG_LAST_USE; }
1464
1465 src_reg_flags: src_reg_flag
1466 | src_reg_flag src_reg_flags
1467
1468 src_reg: src
1469 | src_reg_flags src
1470
1471 src_reg_gpr: src_reg
1472 | relative_gpr_src
1473
1474 src_const: const
1475 | src_reg_flags const
1476
1477 src_reg_or_const: src_reg
1478 | src_const
1479
1480 src_reg_or_const_or_rel: src_reg_or_const
1481 | relative
1482 | src_reg_flags relative
1483
1484 src_reg_or_const_or_rel_or_imm: src_reg_or_const_or_rel
1485 | src_reg_flags immediate
1486 | immediate
1487
1488 src_reg_or_rel_or_imm: src_reg
1489 | relative
1490 | immediate
1491
1492 offset: { $$ = 0; }
1493 | '+' integer { $$ = $2; }
1494 | '-' integer { $$ = -$2; }
1495
1496 relative_gpr_src: 'r' '<' T_A0 offset '>' { new_src(0, IR3_REG_RELATIV)->array.offset = $4; }
1497 | T_HR '<' T_A0 offset '>' { new_src(0, IR3_REG_RELATIV | IR3_REG_HALF)->array.offset = $4; }
1498
1499 relative_gpr_dst: 'r' '<' T_A0 offset '>' { new_dst(0, IR3_REG_RELATIV)->array.offset = $4; }
1500 | T_HR '<' T_A0 offset '>' { new_dst(0, IR3_REG_RELATIV | IR3_REG_HALF)->array.offset = $4; }
1501
1502 relative_const: 'c' '<' T_A0 offset '>' { new_src(0, IR3_REG_RELATIV | IR3_REG_CONST)->array.offset = $4; }
1503 | T_HC '<' T_A0 offset '>' { new_src(0, IR3_REG_RELATIV | IR3_REG_CONST | IR3_REG_HALF)->array.offset = $4; }
1504
1505 relative: relative_gpr_src
1506 | relative_const
1507
1508 /* cat1 immediates differ slighly in the floating point case from the cat2
1509 * case which can only encode certain predefined values (ie. and index into
1510 * the FLUT table)
1511 */
1512 immediate_cat1: integer { new_src(0, IR3_REG_IMMED)->iim_val = type_size(instr->cat1.src_type) < 32 ? $1 & 0xffff : $1; }
1513 | '(' integer ')' { new_src(0, IR3_REG_IMMED)->fim_val = $2; }
1514 | '(' float ')' { new_src(0, IR3_REG_IMMED)->fim_val = $2; }
1515 | 'h' '(' integer ')' { new_src(0, IR3_REG_IMMED | IR3_REG_HALF)->iim_val = $3 & 0xffff; }
1516 | 'h' '(' float ')' { new_src(0, IR3_REG_IMMED | IR3_REG_HALF)->uim_val = _mesa_float_to_half($3); }
1517 | '(' T_NAN ')' { new_src(0, IR3_REG_IMMED)->fim_val = NAN; }
1518 | '(' T_INF ')' { new_src(0, IR3_REG_IMMED)->fim_val = INFINITY; }
1519
1520 immediate: integer { new_src(0, IR3_REG_IMMED)->iim_val = $1; }
1521 | '(' integer ')' { new_src(0, IR3_REG_IMMED)->fim_val = $2; }
1522 | flut_immed { new_src(0, IR3_REG_IMMED)->uim_val = $1; }
1523 | 'h' '(' integer ')' { new_src(0, IR3_REG_IMMED | IR3_REG_HALF)->iim_val = $3; }
1524 | 'h' flut_immed { new_src(0, IR3_REG_IMMED | IR3_REG_HALF)->uim_val = $2; }
1525
1526 /* Float LUT values accepted as immed: */
1527 flut_immed: T_FLUT_0_0
1528 | T_FLUT_0_5
1529 | T_FLUT_1_0
1530 | T_FLUT_2_0
1531 | T_FLUT_E
1532 | T_FLUT_PI
1533 | T_FLUT_INV_PI
1534 | T_FLUT_INV_LOG2_E
1535 | T_FLUT_LOG2_E
1536 | T_FLUT_INV_LOG2_10
1537 | T_FLUT_LOG2_10
1538 | T_FLUT_4_0
1539
1540 integer: T_INT { $$ = $1; }
1541 | '-' T_INT { $$ = -$2; }
1542 | T_HEX { $$ = $1; }
1543 | '-' T_HEX { $$ = -$2; }
1544
1545 float: T_FLOAT { $$ = $1; }
1546 | '-' T_FLOAT { $$ = -$2; }
1547
1548 type: T_TYPE_F16 { $$ = TYPE_F16; }
1549 | T_TYPE_F32 { $$ = TYPE_F32; }
1550 | T_TYPE_U16 { $$ = TYPE_U16; }
1551 | T_TYPE_U32 { $$ = TYPE_U32; }
1552 | T_TYPE_S16 { $$ = TYPE_S16; }
1553 | T_TYPE_S32 { $$ = TYPE_S32; }
1554 | T_TYPE_U8 { $$ = TYPE_U8; }
1555 | T_TYPE_S8 { $$ = TYPE_S8; }
1556