1 /*
2 * Copyright (c) 2013 Rob Clark <robdclark@gmail.com>
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 #ifndef INSTR_A3XX_H_
25 #define INSTR_A3XX_H_
26
27 #define PACKED __attribute__((__packed__))
28
29 #include <stdint.h>
30 #include <assert.h>
31
32 /* size of largest OPC field of all the instruction categories: */
33 #define NOPC_BITS 6
34
35 #define _OPC(cat, opc) (((cat) << NOPC_BITS) | opc)
36
37 typedef enum {
38 /* category 0: */
39 OPC_NOP = _OPC(0, 0),
40 OPC_BR = _OPC(0, 1),
41 OPC_JUMP = _OPC(0, 2),
42 OPC_CALL = _OPC(0, 3),
43 OPC_RET = _OPC(0, 4),
44 OPC_KILL = _OPC(0, 5),
45 OPC_END = _OPC(0, 6),
46 OPC_EMIT = _OPC(0, 7),
47 OPC_CUT = _OPC(0, 8),
48 OPC_CHMASK = _OPC(0, 9),
49 OPC_CHSH = _OPC(0, 10),
50 OPC_FLOW_REV = _OPC(0, 11),
51
52 /* category 1: */
53 OPC_MOV = _OPC(1, 0),
54
55 /* category 2: */
56 OPC_ADD_F = _OPC(2, 0),
57 OPC_MIN_F = _OPC(2, 1),
58 OPC_MAX_F = _OPC(2, 2),
59 OPC_MUL_F = _OPC(2, 3),
60 OPC_SIGN_F = _OPC(2, 4),
61 OPC_CMPS_F = _OPC(2, 5),
62 OPC_ABSNEG_F = _OPC(2, 6),
63 OPC_CMPV_F = _OPC(2, 7),
64 /* 8 - invalid */
65 OPC_FLOOR_F = _OPC(2, 9),
66 OPC_CEIL_F = _OPC(2, 10),
67 OPC_RNDNE_F = _OPC(2, 11),
68 OPC_RNDAZ_F = _OPC(2, 12),
69 OPC_TRUNC_F = _OPC(2, 13),
70 /* 14-15 - invalid */
71 OPC_ADD_U = _OPC(2, 16),
72 OPC_ADD_S = _OPC(2, 17),
73 OPC_SUB_U = _OPC(2, 18),
74 OPC_SUB_S = _OPC(2, 19),
75 OPC_CMPS_U = _OPC(2, 20),
76 OPC_CMPS_S = _OPC(2, 21),
77 OPC_MIN_U = _OPC(2, 22),
78 OPC_MIN_S = _OPC(2, 23),
79 OPC_MAX_U = _OPC(2, 24),
80 OPC_MAX_S = _OPC(2, 25),
81 OPC_ABSNEG_S = _OPC(2, 26),
82 /* 27 - invalid */
83 OPC_AND_B = _OPC(2, 28),
84 OPC_OR_B = _OPC(2, 29),
85 OPC_NOT_B = _OPC(2, 30),
86 OPC_XOR_B = _OPC(2, 31),
87 /* 32 - invalid */
88 OPC_CMPV_U = _OPC(2, 33),
89 OPC_CMPV_S = _OPC(2, 34),
90 /* 35-47 - invalid */
91 OPC_MUL_U = _OPC(2, 48),
92 OPC_MUL_S = _OPC(2, 49),
93 OPC_MULL_U = _OPC(2, 50),
94 OPC_BFREV_B = _OPC(2, 51),
95 OPC_CLZ_S = _OPC(2, 52),
96 OPC_CLZ_B = _OPC(2, 53),
97 OPC_SHL_B = _OPC(2, 54),
98 OPC_SHR_B = _OPC(2, 55),
99 OPC_ASHR_B = _OPC(2, 56),
100 OPC_BARY_F = _OPC(2, 57),
101 OPC_MGEN_B = _OPC(2, 58),
102 OPC_GETBIT_B = _OPC(2, 59),
103 OPC_SETRM = _OPC(2, 60),
104 OPC_CBITS_B = _OPC(2, 61),
105 OPC_SHB = _OPC(2, 62),
106 OPC_MSAD = _OPC(2, 63),
107
108 /* category 3: */
109 OPC_MAD_U16 = _OPC(3, 0),
110 OPC_MADSH_U16 = _OPC(3, 1),
111 OPC_MAD_S16 = _OPC(3, 2),
112 OPC_MADSH_M16 = _OPC(3, 3), /* should this be .s16? */
113 OPC_MAD_U24 = _OPC(3, 4),
114 OPC_MAD_S24 = _OPC(3, 5),
115 OPC_MAD_F16 = _OPC(3, 6),
116 OPC_MAD_F32 = _OPC(3, 7),
117 OPC_SEL_B16 = _OPC(3, 8),
118 OPC_SEL_B32 = _OPC(3, 9),
119 OPC_SEL_S16 = _OPC(3, 10),
120 OPC_SEL_S32 = _OPC(3, 11),
121 OPC_SEL_F16 = _OPC(3, 12),
122 OPC_SEL_F32 = _OPC(3, 13),
123 OPC_SAD_S16 = _OPC(3, 14),
124 OPC_SAD_S32 = _OPC(3, 15),
125
126 /* category 4: */
127 OPC_RCP = _OPC(4, 0),
128 OPC_RSQ = _OPC(4, 1),
129 OPC_LOG2 = _OPC(4, 2),
130 OPC_EXP2 = _OPC(4, 3),
131 OPC_SIN = _OPC(4, 4),
132 OPC_COS = _OPC(4, 5),
133 OPC_SQRT = _OPC(4, 6),
134 // 7-63 - invalid
135
136 /* category 5: */
137 OPC_ISAM = _OPC(5, 0),
138 OPC_ISAML = _OPC(5, 1),
139 OPC_ISAMM = _OPC(5, 2),
140 OPC_SAM = _OPC(5, 3),
141 OPC_SAMB = _OPC(5, 4),
142 OPC_SAML = _OPC(5, 5),
143 OPC_SAMGQ = _OPC(5, 6),
144 OPC_GETLOD = _OPC(5, 7),
145 OPC_CONV = _OPC(5, 8),
146 OPC_CONVM = _OPC(5, 9),
147 OPC_GETSIZE = _OPC(5, 10),
148 OPC_GETBUF = _OPC(5, 11),
149 OPC_GETPOS = _OPC(5, 12),
150 OPC_GETINFO = _OPC(5, 13),
151 OPC_DSX = _OPC(5, 14),
152 OPC_DSY = _OPC(5, 15),
153 OPC_GATHER4R = _OPC(5, 16),
154 OPC_GATHER4G = _OPC(5, 17),
155 OPC_GATHER4B = _OPC(5, 18),
156 OPC_GATHER4A = _OPC(5, 19),
157 OPC_SAMGP0 = _OPC(5, 20),
158 OPC_SAMGP1 = _OPC(5, 21),
159 OPC_SAMGP2 = _OPC(5, 22),
160 OPC_SAMGP3 = _OPC(5, 23),
161 OPC_DSXPP_1 = _OPC(5, 24),
162 OPC_DSYPP_1 = _OPC(5, 25),
163 OPC_RGETPOS = _OPC(5, 26),
164 OPC_RGETINFO = _OPC(5, 27),
165
166 /* category 6: */
167 OPC_LDG = _OPC(6, 0), /* load-global */
168 OPC_LDL = _OPC(6, 1),
169 OPC_LDP = _OPC(6, 2),
170 OPC_STG = _OPC(6, 3), /* store-global */
171 OPC_STL = _OPC(6, 4),
172 OPC_STP = _OPC(6, 5),
173 OPC_STI = _OPC(6, 6),
174 OPC_G2L = _OPC(6, 7),
175 OPC_L2G = _OPC(6, 8),
176 OPC_PREFETCH = _OPC(6, 9),
177 OPC_LDLW = _OPC(6, 10),
178 OPC_STLW = _OPC(6, 11),
179 OPC_RESFMT = _OPC(6, 14),
180 OPC_RESINFO = _OPC(6, 15),
181 OPC_ATOMIC_ADD = _OPC(6, 16),
182 OPC_ATOMIC_SUB = _OPC(6, 17),
183 OPC_ATOMIC_XCHG = _OPC(6, 18),
184 OPC_ATOMIC_INC = _OPC(6, 19),
185 OPC_ATOMIC_DEC = _OPC(6, 20),
186 OPC_ATOMIC_CMPXCHG = _OPC(6, 21),
187 OPC_ATOMIC_MIN = _OPC(6, 22),
188 OPC_ATOMIC_MAX = _OPC(6, 23),
189 OPC_ATOMIC_AND = _OPC(6, 24),
190 OPC_ATOMIC_OR = _OPC(6, 25),
191 OPC_ATOMIC_XOR = _OPC(6, 26),
192 OPC_LDGB = _OPC(6, 27),
193 OPC_STGB = _OPC(6, 28),
194 OPC_STIB = _OPC(6, 29),
195 OPC_LDC = _OPC(6, 30),
196 OPC_LDLV = _OPC(6, 31),
197
198 /* category 7: */
199 OPC_BAR = _OPC(7, 0),
200 OPC_FENCE = _OPC(7, 1),
201
202 /* meta instructions (category -1): */
203 /* placeholder instr to mark shader inputs: */
204 OPC_META_INPUT = _OPC(-1, 0),
205 OPC_META_PHI = _OPC(-1, 1),
206 /* The "fan-in" and "fan-out" instructions are used for keeping
207 * track of instructions that write to multiple dst registers
208 * (fan-out) like texture sample instructions, or read multiple
209 * consecutive scalar registers (fan-in) (bary.f, texture samp)
210 */
211 OPC_META_FO = _OPC(-1, 2),
212 OPC_META_FI = _OPC(-1, 3),
213
214 } opc_t;
215
216 #define opc_cat(opc) ((int)((opc) >> NOPC_BITS))
217 #define opc_op(opc) ((unsigned)((opc) & ((1 << NOPC_BITS) - 1)))
218
219 typedef enum {
220 TYPE_F16 = 0,
221 TYPE_F32 = 1,
222 TYPE_U16 = 2,
223 TYPE_U32 = 3,
224 TYPE_S16 = 4,
225 TYPE_S32 = 5,
226 TYPE_U8 = 6,
227 TYPE_S8 = 7, // XXX I assume?
228 } type_t;
229
type_size(type_t type)230 static inline uint32_t type_size(type_t type)
231 {
232 switch (type) {
233 case TYPE_F32:
234 case TYPE_U32:
235 case TYPE_S32:
236 return 32;
237 case TYPE_F16:
238 case TYPE_U16:
239 case TYPE_S16:
240 return 16;
241 case TYPE_U8:
242 case TYPE_S8:
243 return 8;
244 default:
245 assert(0); /* invalid type */
246 return 0;
247 }
248 }
249
type_float(type_t type)250 static inline int type_float(type_t type)
251 {
252 return (type == TYPE_F32) || (type == TYPE_F16);
253 }
254
type_uint(type_t type)255 static inline int type_uint(type_t type)
256 {
257 return (type == TYPE_U32) || (type == TYPE_U16) || (type == TYPE_U8);
258 }
259
type_sint(type_t type)260 static inline int type_sint(type_t type)
261 {
262 return (type == TYPE_S32) || (type == TYPE_S16) || (type == TYPE_S8);
263 }
264
265 typedef union PACKED {
266 /* normal gpr or const src register: */
267 struct PACKED {
268 uint32_t comp : 2;
269 uint32_t num : 10;
270 };
271 /* for immediate val: */
272 int32_t iim_val : 11;
273 /* to make compiler happy: */
274 uint32_t dummy32;
275 uint32_t dummy10 : 10;
276 int32_t idummy10 : 10;
277 uint32_t dummy11 : 11;
278 uint32_t dummy12 : 12;
279 uint32_t dummy13 : 13;
280 uint32_t dummy8 : 8;
281 } reg_t;
282
283 /* special registers: */
284 #define REG_A0 61 /* address register */
285 #define REG_P0 62 /* predicate register */
286
reg_special(reg_t reg)287 static inline int reg_special(reg_t reg)
288 {
289 return (reg.num == REG_A0) || (reg.num == REG_P0);
290 }
291
292 typedef struct PACKED {
293 /* dword0: */
294 union PACKED {
295 struct PACKED {
296 int16_t immed : 16;
297 uint32_t dummy1 : 16;
298 } a3xx;
299 struct PACKED {
300 int32_t immed : 20;
301 uint32_t dummy1 : 12;
302 } a4xx;
303 struct PACKED {
304 uint32_t immed : 32;
305 } a5xx;
306 };
307
308 /* dword1: */
309 uint32_t dummy2 : 8;
310 uint32_t repeat : 3;
311 uint32_t dummy3 : 1;
312 uint32_t ss : 1;
313 uint32_t dummy4 : 7;
314 uint32_t inv : 1;
315 uint32_t comp : 2;
316 uint32_t opc : 4;
317 uint32_t jmp_tgt : 1;
318 uint32_t sync : 1;
319 uint32_t opc_cat : 3;
320 } instr_cat0_t;
321
322 typedef struct PACKED {
323 /* dword0: */
324 union PACKED {
325 /* for normal src register: */
326 struct PACKED {
327 uint32_t src : 11;
328 /* at least low bit of pad must be zero or it will
329 * look like a address relative src
330 */
331 uint32_t pad : 21;
332 };
333 /* for address relative: */
334 struct PACKED {
335 int32_t off : 10;
336 uint32_t src_rel_c : 1;
337 uint32_t src_rel : 1;
338 uint32_t unknown : 20;
339 };
340 /* for immediate: */
341 int32_t iim_val;
342 uint32_t uim_val;
343 float fim_val;
344 };
345
346 /* dword1: */
347 uint32_t dst : 8;
348 uint32_t repeat : 3;
349 uint32_t src_r : 1;
350 uint32_t ss : 1;
351 uint32_t ul : 1;
352 uint32_t dst_type : 3;
353 uint32_t dst_rel : 1;
354 uint32_t src_type : 3;
355 uint32_t src_c : 1;
356 uint32_t src_im : 1;
357 uint32_t even : 1;
358 uint32_t pos_inf : 1;
359 uint32_t must_be_0 : 2;
360 uint32_t jmp_tgt : 1;
361 uint32_t sync : 1;
362 uint32_t opc_cat : 3;
363 } instr_cat1_t;
364
365 typedef struct PACKED {
366 /* dword0: */
367 union PACKED {
368 struct PACKED {
369 uint32_t src1 : 11;
370 uint32_t must_be_zero1: 2;
371 uint32_t src1_im : 1; /* immediate */
372 uint32_t src1_neg : 1; /* negate */
373 uint32_t src1_abs : 1; /* absolute value */
374 };
375 struct PACKED {
376 uint32_t src1 : 10;
377 uint32_t src1_c : 1; /* relative-const */
378 uint32_t src1_rel : 1; /* relative address */
379 uint32_t must_be_zero : 1;
380 uint32_t dummy : 3;
381 } rel1;
382 struct PACKED {
383 uint32_t src1 : 12;
384 uint32_t src1_c : 1; /* const */
385 uint32_t dummy : 3;
386 } c1;
387 };
388
389 union PACKED {
390 struct PACKED {
391 uint32_t src2 : 11;
392 uint32_t must_be_zero2: 2;
393 uint32_t src2_im : 1; /* immediate */
394 uint32_t src2_neg : 1; /* negate */
395 uint32_t src2_abs : 1; /* absolute value */
396 };
397 struct PACKED {
398 uint32_t src2 : 10;
399 uint32_t src2_c : 1; /* relative-const */
400 uint32_t src2_rel : 1; /* relative address */
401 uint32_t must_be_zero : 1;
402 uint32_t dummy : 3;
403 } rel2;
404 struct PACKED {
405 uint32_t src2 : 12;
406 uint32_t src2_c : 1; /* const */
407 uint32_t dummy : 3;
408 } c2;
409 };
410
411 /* dword1: */
412 uint32_t dst : 8;
413 uint32_t repeat : 3;
414 uint32_t src1_r : 1;
415 uint32_t ss : 1;
416 uint32_t ul : 1; /* dunno */
417 uint32_t dst_half : 1; /* or widen/narrow.. ie. dst hrN <-> rN */
418 uint32_t ei : 1;
419 uint32_t cond : 3;
420 uint32_t src2_r : 1;
421 uint32_t full : 1; /* not half */
422 uint32_t opc : 6;
423 uint32_t jmp_tgt : 1;
424 uint32_t sync : 1;
425 uint32_t opc_cat : 3;
426 } instr_cat2_t;
427
428 typedef struct PACKED {
429 /* dword0: */
430 union PACKED {
431 struct PACKED {
432 uint32_t src1 : 11;
433 uint32_t must_be_zero1: 2;
434 uint32_t src2_c : 1;
435 uint32_t src1_neg : 1;
436 uint32_t src2_r : 1;
437 };
438 struct PACKED {
439 uint32_t src1 : 10;
440 uint32_t src1_c : 1;
441 uint32_t src1_rel : 1;
442 uint32_t must_be_zero : 1;
443 uint32_t dummy : 3;
444 } rel1;
445 struct PACKED {
446 uint32_t src1 : 12;
447 uint32_t src1_c : 1;
448 uint32_t dummy : 3;
449 } c1;
450 };
451
452 union PACKED {
453 struct PACKED {
454 uint32_t src3 : 11;
455 uint32_t must_be_zero2: 2;
456 uint32_t src3_r : 1;
457 uint32_t src2_neg : 1;
458 uint32_t src3_neg : 1;
459 };
460 struct PACKED {
461 uint32_t src3 : 10;
462 uint32_t src3_c : 1;
463 uint32_t src3_rel : 1;
464 uint32_t must_be_zero : 1;
465 uint32_t dummy : 3;
466 } rel2;
467 struct PACKED {
468 uint32_t src3 : 12;
469 uint32_t src3_c : 1;
470 uint32_t dummy : 3;
471 } c2;
472 };
473
474 /* dword1: */
475 uint32_t dst : 8;
476 uint32_t repeat : 3;
477 uint32_t src1_r : 1;
478 uint32_t ss : 1;
479 uint32_t ul : 1;
480 uint32_t dst_half : 1; /* or widen/narrow.. ie. dst hrN <-> rN */
481 uint32_t src2 : 8;
482 uint32_t opc : 4;
483 uint32_t jmp_tgt : 1;
484 uint32_t sync : 1;
485 uint32_t opc_cat : 3;
486 } instr_cat3_t;
487
instr_cat3_full(instr_cat3_t * cat3)488 static inline bool instr_cat3_full(instr_cat3_t *cat3)
489 {
490 switch (_OPC(3, cat3->opc)) {
491 case OPC_MAD_F16:
492 case OPC_MAD_U16:
493 case OPC_MAD_S16:
494 case OPC_SEL_B16:
495 case OPC_SEL_S16:
496 case OPC_SEL_F16:
497 case OPC_SAD_S16:
498 case OPC_SAD_S32: // really??
499 return false;
500 default:
501 return true;
502 }
503 }
504
505 typedef struct PACKED {
506 /* dword0: */
507 union PACKED {
508 struct PACKED {
509 uint32_t src : 11;
510 uint32_t must_be_zero1: 2;
511 uint32_t src_im : 1; /* immediate */
512 uint32_t src_neg : 1; /* negate */
513 uint32_t src_abs : 1; /* absolute value */
514 };
515 struct PACKED {
516 uint32_t src : 10;
517 uint32_t src_c : 1; /* relative-const */
518 uint32_t src_rel : 1; /* relative address */
519 uint32_t must_be_zero : 1;
520 uint32_t dummy : 3;
521 } rel;
522 struct PACKED {
523 uint32_t src : 12;
524 uint32_t src_c : 1; /* const */
525 uint32_t dummy : 3;
526 } c;
527 };
528 uint32_t dummy1 : 16; /* seem to be ignored */
529
530 /* dword1: */
531 uint32_t dst : 8;
532 uint32_t repeat : 3;
533 uint32_t src_r : 1;
534 uint32_t ss : 1;
535 uint32_t ul : 1;
536 uint32_t dst_half : 1; /* or widen/narrow.. ie. dst hrN <-> rN */
537 uint32_t dummy2 : 5; /* seem to be ignored */
538 uint32_t full : 1; /* not half */
539 uint32_t opc : 6;
540 uint32_t jmp_tgt : 1;
541 uint32_t sync : 1;
542 uint32_t opc_cat : 3;
543 } instr_cat4_t;
544
545 typedef struct PACKED {
546 /* dword0: */
547 union PACKED {
548 /* normal case: */
549 struct PACKED {
550 uint32_t full : 1; /* not half */
551 uint32_t src1 : 8;
552 uint32_t src2 : 8;
553 uint32_t dummy1 : 4; /* seem to be ignored */
554 uint32_t samp : 4;
555 uint32_t tex : 7;
556 } norm;
557 /* s2en case: */
558 struct PACKED {
559 uint32_t full : 1; /* not half */
560 uint32_t src1 : 8;
561 uint32_t src2 : 11;
562 uint32_t dummy1 : 1;
563 uint32_t src3 : 8;
564 uint32_t dummy2 : 3;
565 } s2en;
566 /* same in either case: */
567 // XXX I think, confirm this
568 struct PACKED {
569 uint32_t full : 1; /* not half */
570 uint32_t src1 : 8;
571 uint32_t pad : 23;
572 };
573 };
574
575 /* dword1: */
576 uint32_t dst : 8;
577 uint32_t wrmask : 4; /* write-mask */
578 uint32_t type : 3;
579 uint32_t dummy2 : 1; /* seems to be ignored */
580 uint32_t is_3d : 1;
581
582 uint32_t is_a : 1;
583 uint32_t is_s : 1;
584 uint32_t is_s2en : 1;
585 uint32_t is_o : 1;
586 uint32_t is_p : 1;
587
588 uint32_t opc : 5;
589 uint32_t jmp_tgt : 1;
590 uint32_t sync : 1;
591 uint32_t opc_cat : 3;
592 } instr_cat5_t;
593
594 /* dword0 encoding for src_off: [src1 + off], src2: */
595 typedef struct PACKED {
596 /* dword0: */
597 uint32_t mustbe1 : 1;
598 int32_t off : 13;
599 uint32_t src1 : 8;
600 uint32_t src1_im : 1;
601 uint32_t src2_im : 1;
602 uint32_t src2 : 8;
603
604 /* dword1: */
605 uint32_t dword1;
606 } instr_cat6a_t;
607
608 /* dword0 encoding for !src_off: [src1], src2 */
609 typedef struct PACKED {
610 /* dword0: */
611 uint32_t mustbe0 : 1;
612 uint32_t src1 : 13;
613 uint32_t ignore0 : 8;
614 uint32_t src1_im : 1;
615 uint32_t src2_im : 1;
616 uint32_t src2 : 8;
617
618 /* dword1: */
619 uint32_t dword1;
620 } instr_cat6b_t;
621
622 /* dword1 encoding for dst_off: */
623 typedef struct PACKED {
624 /* dword0: */
625 uint32_t dword0;
626
627 /* note: there is some weird stuff going on where sometimes
628 * cat6->a.off is involved.. but that seems like a bug in
629 * the blob, since it is used even if !cat6->src_off
630 * It would make sense for there to be some more bits to
631 * bring us to 11 bits worth of offset, but not sure..
632 */
633 int32_t off : 8;
634 uint32_t mustbe1 : 1;
635 uint32_t dst : 8;
636 uint32_t pad1 : 15;
637 } instr_cat6c_t;
638
639 /* dword1 encoding for !dst_off: */
640 typedef struct PACKED {
641 /* dword0: */
642 uint32_t dword0;
643
644 uint32_t dst : 8;
645 uint32_t mustbe0 : 1;
646 uint32_t idx : 8;
647 uint32_t pad0 : 15;
648 } instr_cat6d_t;
649
650 /* ldgb and atomics..
651 *
652 * ldgb: pad0=0, pad3=1
653 * atomic .g: pad0=1, pad3=1
654 * .l: pad0=1, pad3=0
655 */
656 typedef struct PACKED {
657 /* dword0: */
658 uint32_t pad0 : 1;
659 uint32_t src3 : 8;
660 uint32_t d : 2;
661 uint32_t typed : 1;
662 uint32_t type_size : 2;
663 uint32_t src1 : 8;
664 uint32_t src1_im : 1;
665 uint32_t src2_im : 1;
666 uint32_t src2 : 8;
667
668 /* dword1: */
669 uint32_t dst : 8;
670 uint32_t mustbe0 : 1;
671 uint32_t src_ssbo : 8;
672 uint32_t pad2 : 3; // type
673 uint32_t g : 1;
674 uint32_t pad3 : 1;
675 uint32_t pad4 : 10; // opc/jmp_tgt/sync/opc_cat
676 } instr_cat6ldgb_t;
677
678 /* stgb, pad0=0, pad3=2
679 */
680 typedef struct PACKED {
681 /* dword0: */
682 uint32_t mustbe1 : 1; // ???
683 uint32_t src1 : 8;
684 uint32_t d : 2;
685 uint32_t typed : 1;
686 uint32_t type_size : 2;
687 uint32_t pad0 : 9;
688 uint32_t src2_im : 1;
689 uint32_t src2 : 8;
690
691 /* dword1: */
692 uint32_t src3 : 8;
693 uint32_t src3_im : 1;
694 uint32_t dst_ssbo : 8;
695 uint32_t pad2 : 3; // type
696 uint32_t pad3 : 2;
697 uint32_t pad4 : 10; // opc/jmp_tgt/sync/opc_cat
698 } instr_cat6stgb_t;
699
700 typedef union PACKED {
701 instr_cat6a_t a;
702 instr_cat6b_t b;
703 instr_cat6c_t c;
704 instr_cat6d_t d;
705 instr_cat6ldgb_t ldgb;
706 instr_cat6stgb_t stgb;
707 struct PACKED {
708 /* dword0: */
709 uint32_t src_off : 1;
710 uint32_t pad1 : 31;
711
712 /* dword1: */
713 uint32_t pad2 : 8;
714 uint32_t dst_off : 1;
715 uint32_t pad3 : 8;
716 uint32_t type : 3;
717 uint32_t g : 1; /* or in some cases it means dst immed */
718 uint32_t pad4 : 1;
719 uint32_t opc : 5;
720 uint32_t jmp_tgt : 1;
721 uint32_t sync : 1;
722 uint32_t opc_cat : 3;
723 };
724 } instr_cat6_t;
725
726 typedef struct PACKED {
727 /* dword0: */
728 uint32_t pad1 : 32;
729
730 /* dword1: */
731 uint32_t pad2 : 12;
732 uint32_t ss : 1; /* maybe in the encoding, but blob only uses (sy) */
733 uint32_t pad3 : 6;
734 uint32_t w : 1; /* write */
735 uint32_t r : 1; /* read */
736 uint32_t l : 1; /* local */
737 uint32_t g : 1; /* global */
738 uint32_t opc : 4; /* presumed, but only a couple known OPCs */
739 uint32_t jmp_tgt : 1; /* (jp) */
740 uint32_t sync : 1; /* (sy) */
741 uint32_t opc_cat : 3;
742 } instr_cat7_t;
743
744 typedef union PACKED {
745 instr_cat0_t cat0;
746 instr_cat1_t cat1;
747 instr_cat2_t cat2;
748 instr_cat3_t cat3;
749 instr_cat4_t cat4;
750 instr_cat5_t cat5;
751 instr_cat6_t cat6;
752 instr_cat7_t cat7;
753 struct PACKED {
754 /* dword0: */
755 uint64_t pad1 : 40;
756 uint32_t repeat : 3; /* cat0-cat4 */
757 uint32_t pad2 : 1;
758 uint32_t ss : 1; /* cat1-cat4 (cat0??) and cat7 (?) */
759 uint32_t ul : 1; /* cat2-cat4 (and cat1 in blob.. which may be bug??) */
760 uint32_t pad3 : 13;
761 uint32_t jmp_tgt : 1;
762 uint32_t sync : 1;
763 uint32_t opc_cat : 3;
764
765 };
766 } instr_t;
767
instr_opc(instr_t * instr)768 static inline uint32_t instr_opc(instr_t *instr)
769 {
770 switch (instr->opc_cat) {
771 case 0: return instr->cat0.opc;
772 case 1: return 0;
773 case 2: return instr->cat2.opc;
774 case 3: return instr->cat3.opc;
775 case 4: return instr->cat4.opc;
776 case 5: return instr->cat5.opc;
777 case 6: return instr->cat6.opc;
778 case 7: return instr->cat7.opc;
779 default: return 0;
780 }
781 }
782
is_mad(opc_t opc)783 static inline bool is_mad(opc_t opc)
784 {
785 switch (opc) {
786 case OPC_MAD_U16:
787 case OPC_MAD_S16:
788 case OPC_MAD_U24:
789 case OPC_MAD_S24:
790 case OPC_MAD_F16:
791 case OPC_MAD_F32:
792 return true;
793 default:
794 return false;
795 }
796 }
797
is_madsh(opc_t opc)798 static inline bool is_madsh(opc_t opc)
799 {
800 switch (opc) {
801 case OPC_MADSH_U16:
802 case OPC_MADSH_M16:
803 return true;
804 default:
805 return false;
806 }
807 }
808
is_atomic(opc_t opc)809 static inline bool is_atomic(opc_t opc)
810 {
811 switch (opc) {
812 case OPC_ATOMIC_ADD:
813 case OPC_ATOMIC_SUB:
814 case OPC_ATOMIC_XCHG:
815 case OPC_ATOMIC_INC:
816 case OPC_ATOMIC_DEC:
817 case OPC_ATOMIC_CMPXCHG:
818 case OPC_ATOMIC_MIN:
819 case OPC_ATOMIC_MAX:
820 case OPC_ATOMIC_AND:
821 case OPC_ATOMIC_OR:
822 case OPC_ATOMIC_XOR:
823 return true;
824 default:
825 return false;
826 }
827 }
828
is_ssbo(opc_t opc)829 static inline bool is_ssbo(opc_t opc)
830 {
831 switch (opc) {
832 case OPC_RESFMT:
833 case OPC_RESINFO:
834 case OPC_LDGB:
835 case OPC_STGB:
836 case OPC_STIB:
837 return true;
838 default:
839 return false;
840 }
841 }
842
843 #endif /* INSTR_A3XX_H_ */
844