pan/bi: Pack BI_BITWISE
[mesa.git] / src / panfrost / bifrost / bifrost.h
1 /*
2 * Copyright (C) 2019 Connor Abbott <cwabbott0@gmail.com>
3 * Copyright (C) 2019 Lyude Paul <thatslyude@gmail.com>
4 * Copyright (C) 2019 Ryan Houdek <Sonicadvance1@gmail.com>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 */
25
26 #ifndef __bifrost_h__
27 #define __bifrost_h__
28
29 #include <stdint.h>
30 #include <stdbool.h>
31
32 enum bifrost_clause_type {
33 BIFROST_CLAUSE_NONE = 0,
34 BIFROST_CLAUSE_LOAD_VARY = 1,
35 BIFROST_CLAUSE_UBO = 2,
36 BIFROST_CLAUSE_TEX = 3,
37 BIFROST_CLAUSE_SSBO_LOAD = 5,
38 BIFROST_CLAUSE_SSBO_STORE = 6,
39 BIFROST_CLAUSE_BLEND = 9,
40 BIFROST_CLAUSE_ATEST = 13,
41 BIFROST_CLAUSE_64BIT = 15
42 };
43
44 struct bifrost_header {
45 unsigned unk0 : 7;
46 // If true, convert any infinite result of any floating-point operation to
47 // the biggest representable number.
48 unsigned suppress_inf: 1;
49 // Convert any NaN results to 0.
50 unsigned suppress_nan : 1;
51 unsigned unk1 : 2;
52 // true if the execution mask of the next clause is the same as the mask of
53 // the current clause.
54 unsigned back_to_back : 1;
55 unsigned no_end_of_shader: 1;
56 unsigned unk2 : 2;
57 // Set to true for fragment shaders, to implement this bit of spec text
58 // from section 7.1.5 of the GLSL ES spec:
59 //
60 // "Stores to image and buffer variables performed by helper invocations
61 // have no effect on the underlying image or buffer memory."
62 //
63 // Helper invocations are threads (invocations) corresponding to pixels in
64 // a quad that aren't actually part of the triangle, but are included to
65 // make derivatives work correctly. They're usually turned on, but they
66 // need to be masked off for GLSL-level stores. This bit seems to be the
67 // only bit that's actually different between fragment shaders and other
68 // shaders, so this is probably what it's doing.
69 unsigned elide_writes : 1;
70 // If backToBack is off:
71 // - true for conditional branches and fallthrough
72 // - false for unconditional branches
73 // The blob seems to always set it to true if back-to-back is on.
74 unsigned branch_cond : 1;
75 // This bit is set when the next clause writes to the data register of some
76 // previous clause.
77 unsigned datareg_writebarrier: 1;
78 unsigned datareg : 6;
79 unsigned scoreboard_deps: 8;
80 unsigned scoreboard_index: 3;
81 enum bifrost_clause_type clause_type: 4;
82 unsigned unk3 : 1; // part of clauseType?
83 enum bifrost_clause_type next_clause_type: 4;
84 unsigned unk4 : 1; // part of nextClauseType?
85 } __attribute__((packed));
86
87 enum bifrost_packed_src {
88 BIFROST_SRC_PORT0 = 0,
89 BIFROST_SRC_PORT1 = 1,
90 BIFROST_SRC_PORT3 = 2,
91 BIFROST_SRC_STAGE = 3,
92 BIFROST_SRC_CONST_LO = 4,
93 BIFROST_SRC_CONST_HI = 5,
94 BIFROST_SRC_PASS_FMA = 6,
95 BIFROST_SRC_PASS_ADD = 7,
96 };
97
98 #define BIFROST_FMA_EXT (0xe0000)
99 #define BIFROST_FMA_OP_MOV BIFROST_FMA_EXT | (0x32d)
100 #define BIFROST_FMA_OP_FREXPE_LOG BIFROST_FMA_EXT | 0x3c5
101 #define BIFROST_FMA_OP_ADD_FREXPM ((BIFROST_FMA_EXT | 0x1e80) >> 3)
102 #define BIFROST_FMA_SEL_16(swiz) (((BIFROST_FMA_EXT | 0x1e00) >> 3) | (swiz))
103
104 struct bifrost_fma_inst {
105 unsigned src0 : 3;
106 unsigned op : 20;
107 } __attribute__((packed));
108
109 struct bifrost_fma_2src {
110 unsigned src0 : 3;
111 unsigned src1 : 3;
112 unsigned op : 17;
113 } __attribute__((packed));
114
115 #define BIFROST_FMA_OP_SEL8 (0x71)
116
117 struct bifrost_fma_sel8 {
118 unsigned src0 : 3;
119 unsigned src1 : 3;
120 unsigned src2 : 3;
121 unsigned src3 : 3;
122 unsigned swizzle : 4;
123 unsigned op : 7;
124 } __attribute__((packed));
125
126 #define BIFROST_FMA_OP_MSCALE (0x50 >> 3)
127
128 struct bifrost_fma_mscale {
129 unsigned src0 : 3;
130 unsigned src1 : 3;
131 unsigned src2 : 3;
132 unsigned src3 : 3;
133
134 /* If mscale_mode is set - an MSCALE specific mode. If it is not set, a
135 * regular outmod */
136 unsigned mode : 2;
137 unsigned mscale_mode : 1;
138
139 unsigned src0_abs : 1;
140 unsigned src1_neg : 1;
141 unsigned src2_neg : 1;
142 unsigned op : 5;
143 } __attribute__((packed));
144
145 #define BIFROST_ADD_OP_BLEND (0x1952c)
146 #define BIFROST_ADD_OP_FRCP_FAST_F32 (0x0cc00)
147 #define BIFROST_ADD_OP_FRCP_FAST_F16_X (0x0ce10)
148 #define BIFROST_ADD_OP_FRCP_FAST_F16_Y (0x0ce30)
149 #define BIFROST_ADD_OP_FRSQ_FAST_F32 (0x0cc20)
150 #define BIFROST_ADD_OP_FRSQ_FAST_F16_X (0x0ce50)
151 #define BIFROST_ADD_OP_FRSQ_FAST_F16_Y (0x0ce70)
152 #define BIFROST_ADD_OP_LOG2_HELP (0x0cc68)
153 #define BIFROST_ADD_OP_FEXP2_FAST (0x0cd58)
154
155 struct bifrost_add_inst {
156 unsigned src0 : 3;
157 unsigned op : 17;
158 } __attribute__((packed));
159
160 #define BIFROST_ADD_OP_LD_UBO_1 (0x0c1a0 >> 3)
161 #define BIFROST_ADD_OP_LD_UBO_2 (0x0c1e0 >> 3)
162 #define BIFROST_ADD_OP_LD_UBO_3 (0x0caa0 >> 3)
163 #define BIFROST_ADD_OP_LD_UBO_4 (0x0c220 >> 3)
164 #define BIFROST_ADD_SEL_16(swiz) ((0xea60 >> 3) | (swiz))
165
166 struct bifrost_add_2src {
167 unsigned src0 : 3;
168 unsigned src1 : 3;
169 unsigned op : 14;
170 } __attribute__((packed));
171
172 #define BIFROST_ADD_OP_FMAX32 (0x00)
173 #define BIFROST_ADD_OP_FMIN32 (0x01)
174 #define BIFROST_ADD_OP_FADD32 (0x02)
175
176 #define BIFROST_ADD_OP_FADD16 (0x0A)
177
178 struct bifrost_add_faddmin {
179 unsigned src0 : 3;
180 unsigned src1 : 3;
181 unsigned src1_abs : 1;
182 unsigned src0_neg : 1;
183 unsigned src1_neg : 1;
184 unsigned select : 2; /* swizzle_0 for fp16 */
185 unsigned outmod : 2; /* swizzle_1 for fp16 */
186 unsigned mode : 2;
187 unsigned src0_abs : 1;
188 unsigned op : 4;
189 } __attribute__((packed));
190
191 #define BIFROST_ADD_OP_FMAX16 (0x10)
192 #define BIFROST_ADD_OP_FMIN16 (0x12)
193
194 struct bifrost_add_fmin16 {
195 unsigned src0 : 3;
196 unsigned src1 : 3;
197 /* abs2 inferred as with FMA */
198 unsigned abs1 : 1;
199 unsigned src0_neg : 1;
200 unsigned src1_neg : 1;
201 unsigned src0_swizzle : 2;
202 unsigned src1_swizzle : 2;
203 unsigned mode : 2;
204 unsigned op : 5;
205 } __attribute__((packed));
206
207 #define BIFROST_ADD_OP_ST_VAR (0x19300 >> 8)
208
209 struct bifrost_st_vary {
210 unsigned src0 : 3;
211 unsigned src1 : 3;
212 unsigned src2 : 3;
213 unsigned channels : 2;
214 unsigned op : 9;
215 } __attribute__((packed));
216
217 #define BIFROST_ADD_OP_ATEST (0xc8f)
218
219 struct bifrost_add_atest {
220 /* gl_SampleMask (R60) */
221 unsigned src0 : 3;
222
223 /* Alpha value */
224 unsigned src1 : 3;
225
226 /* If half, X/Y select. If !half, always set */
227 unsigned component : 1;
228 unsigned half : 1;
229
230 unsigned op : 12;
231 } __attribute__((packed));
232
233 enum bifrost_outmod {
234 BIFROST_NONE = 0x0,
235 BIFROST_POS = 0x1,
236 BIFROST_SAT_SIGNED = 0x2,
237 BIFROST_SAT = 0x3,
238 };
239
240 enum bifrost_roundmode {
241 BIFROST_RTE = 0x0, /* round to even */
242 BIFROST_RTP = 0x1, /* round to positive */
243 BIFROST_RTN = 0x2, /* round to negative */
244 BIFROST_RTZ = 0x3 /* round to zero */
245 };
246
247 /* NONE: Same as fmax() and fmin() -- return the other
248 * number if any number is NaN. Also always return +0 if
249 * one argument is +0 and the other is -0.
250 *
251 * NAN_WINS: Instead of never returning a NaN, always return
252 * one. The "greater"/"lesser" NaN is always returned, first
253 * by checking the sign and then the mantissa bits.
254 *
255 * SRC1_WINS: For max, implement src0 > src1 ? src0 : src1.
256 * For min, implement src0 < src1 ? src0 : src1. This
257 * includes handling NaN's and signedness of 0 differently
258 * from above, since +0 and -0 compare equal and comparisons
259 * always return false for NaN's. As a result, this mode is
260 * *not* commutative.
261 *
262 * SRC0_WINS: For max, implement src0 < src1 ? src1 : src0
263 * For min, implement src0 > src1 ? src1 : src0
264 */
265
266
267 enum bifrost_minmax_mode {
268 BIFROST_MINMAX_NONE = 0x0,
269 BIFROST_NAN_WINS = 0x1,
270 BIFROST_SRC1_WINS = 0x2,
271 BIFROST_SRC0_WINS = 0x3,
272 };
273
274 #define BIFROST_FMA_OP_FADD32 (0x58 >> 2)
275 #define BIFROST_FMA_OP_FMAX32 (0x40 >> 2)
276 #define BIFROST_FMA_OP_FMIN32 (0x44 >> 2)
277
278 struct bifrost_fma_add {
279 unsigned src0 : 3;
280 unsigned src1 : 3;
281 unsigned src1_abs : 1;
282 unsigned src0_neg : 1;
283 unsigned src1_neg : 1;
284 unsigned unk : 3;
285 unsigned src0_abs : 1;
286 enum bifrost_roundmode roundmode : 2;
287 enum bifrost_outmod outmod : 2;
288 unsigned op : 6;
289 } __attribute__((packed));
290
291 #define BIFROST_FMA_OP_FMAX16 (0xC0 >> 2)
292 #define BIFROST_FMA_OP_FMIN16 (0xCC >> 2)
293 #define BIFROST_FMA_OP_FADD16 (0xD8 >> 2)
294
295 struct bifrost_fma_add_minmax16 {
296 unsigned src0 : 3;
297 unsigned src1 : 3;
298 /* abs2 inferred as (src1 < src0) */
299 unsigned abs1 : 1;
300 unsigned src0_neg : 1;
301 unsigned src1_neg : 1;
302 unsigned src0_swizzle : 2;
303 unsigned src1_swizzle : 2;
304 unsigned mode : 2;
305 enum bifrost_outmod outmod : 2;
306 /* roundmode for add, min/max mode for min/max */
307 unsigned op : 6;
308 } __attribute__((packed));
309
310 #define BIFROST_FMA_OP_FMA (0x00)
311
312 struct bifrost_fma_fma {
313 unsigned src0 : 3;
314 unsigned src1 : 3;
315 unsigned src2 : 3;
316 unsigned src_expand : 3;
317 unsigned src0_abs : 1;
318 enum bifrost_roundmode roundmode : 2;
319 enum bifrost_outmod outmod : 2;
320 unsigned src0_neg : 1; /* 14 */
321 unsigned src2_neg : 1;
322 unsigned src1_abs : 1;
323 unsigned src2_abs : 1; /* 17 */
324 unsigned op : 2;
325 } __attribute__((packed));
326
327 #define BIFROST_FMA_OP_FMA16 (0x2)
328
329 struct bifrost_fma_fma16 {
330 unsigned src0 : 3;
331 unsigned src1 : 3;
332 unsigned src2 : 3;
333 unsigned swizzle_0 : 2;
334 unsigned swizzle_1 : 2;
335 enum bifrost_roundmode roundmode : 2;
336 enum bifrost_outmod outmod : 2;
337 unsigned src0_neg : 1;
338 unsigned src2_neg : 1;
339 unsigned swizzle_2 : 2;
340 unsigned op : 2;
341 } __attribute__((packed));
342
343 enum bifrost_csel_cond {
344 BIFROST_FEQ_F = 0x0,
345 BIFROST_FGT_F = 0x1,
346 BIFROST_FGE_F = 0x2,
347 BIFROST_IEQ_F = 0x3,
348 BIFROST_IGT_I = 0x4,
349 BIFROST_IGE_I = 0x5,
350 BIFROST_UGT_I = 0x6,
351 BIFROST_UGE_I = 0x7
352 };
353
354 #define BIFROST_FMA_OP_CSEL4 (0x5c)
355 #define BIFROST_FMA_OP_CSEL4_V16 (0xdc)
356
357 struct bifrost_csel4 {
358 unsigned src0 : 3;
359 unsigned src1 : 3;
360 unsigned src2 : 3;
361 unsigned src3 : 3;
362 enum bifrost_csel_cond cond : 3;
363 unsigned op : 8;
364 } __attribute__((packed));
365
366 #define BIFROST_FMA_OP_RSHIFT_NAND (0x60000 >> 12)
367 #define BIFROST_FMA_OP_RSHIFT_AND (0x61000 >> 12)
368 #define BIFROST_FMA_OP_LSHIFT_NAND (0x62000 >> 12)
369 #define BIFROST_FMA_OP_LSHIFT_AND (0x63000 >> 12)
370 #define BIFROST_FMA_OP_RSHIFT_XOR (0x64000 >> 12)
371 #define BIFROST_FMA_OP_LSHIFT_ADD_32 (0x65200 >> 6)
372 #define BIFROST_FMA_OP_LSHIFT_SUB_32 (0x65600 >> 6)
373 #define BIFROST_FMA_OP_LSHIFT_RSUB_32 (0x65a00 >> 6)
374 #define BIFROST_FMA_OP_RSHIFT_ADD_32 (0x65e00 >> 6)
375 #define BIFROST_FMA_OP_RSHIFT_SUB_32 (0x66200 >> 6)
376 #define BIFROST_FMA_OP_RSHIFT_RSUB_32 (0x66600 >> 6)
377
378 struct bifrost_shift_fma {
379 unsigned src0 : 3;
380 unsigned src1 : 3;
381 unsigned src2 : 3;
382 unsigned half : 3;
383 unsigned unk : 1; /* always set? */
384 unsigned invert_1 : 1; /* Inverts sources to combining op */
385 /* For XOR, switches RSHIFT to LSHIFT since only one invert needed */
386 unsigned invert_2 : 1;
387 unsigned op : 8;
388 } __attribute__((packed));
389
390 struct bifrost_shift_add {
391 unsigned src0 : 3;
392 unsigned src1 : 3;
393 unsigned src2 : 3;
394 unsigned zero : 2;
395
396 unsigned invert_1 : 1;
397 unsigned invert_2 : 1;
398
399 unsigned op : 7;
400 } __attribute__((packed));
401
402 enum bifrost_fcmp_cond {
403 BIFROST_OEQ = 0,
404 BIFROST_OGT = 1,
405 BIFROST_OGE = 2,
406 BIFROST_UNE = 3,
407 BIFROST_OLT = 4,
408 BIFROST_OLE = 5,
409 };
410
411 #define BIFROST_FMA_OP_FCMP_GL (0x48000 >> 13)
412 #define BIFROST_FMA_OP_FCMP_D3D (0x4c000 >> 13)
413
414 struct bifrost_fma_fcmp {
415 unsigned src0 : 3;
416 unsigned src1 : 3;
417 unsigned src1_abs : 1;
418 unsigned unk1 : 1;
419 unsigned src1_neg : 1;
420 unsigned src_expand : 3;
421 unsigned src0_abs : 1;
422 enum bifrost_fcmp_cond cond : 3;
423 unsigned op : 7;
424 } __attribute__((packed));
425
426 struct bifrost_add_fcmp {
427 unsigned src0 : 3;
428 unsigned src1 : 3;
429 enum bifrost_fcmp_cond cond : 3;
430 unsigned src_expand : 2;
431 unsigned src0_abs : 1;
432 unsigned src1_abs : 1;
433 unsigned src1_neg : 1;
434 unsigned op : 6;
435 } __attribute__((packed));
436
437 #define BIFROST_FMA_OP_FCMP_GL_16 (0xc8000 >> 13)
438 #define BIFROST_FMA_OP_FCMP_D3D_16 (0xcc000 >> 13)
439
440 struct bifrost_fma_fcmp16 {
441 unsigned src0 : 3;
442 unsigned src1 : 3;
443
444 /* abs2 inferred */
445 unsigned abs1 : 1;
446 unsigned unk : 2;
447
448 unsigned src0_swizzle : 2;
449 unsigned src1_swizzle : 2;
450
451 enum bifrost_fcmp_cond cond : 3;
452 unsigned op : 7;
453 } __attribute__((packed));
454
455 struct bifrost_add_fcmp16 {
456 unsigned src0 : 3;
457 unsigned src1 : 3;
458 enum bifrost_fcmp_cond cond : 3;
459
460 unsigned src0_swizzle : 2;
461 unsigned src1_swizzle : 2;
462
463 /* No abs mods */
464 unsigned src0_neg : 1;
465
466 unsigned op : 6;
467 } __attribute__((packed));
468
469 enum bifrost_icmp_cond {
470 BIFROST_ICMP_IGT = 0,
471 BIFROST_ICMP_IGE = 1,
472 BIFROST_ICMP_UGT = 2,
473 BIFROST_ICMP_UGE = 3,
474 BIFROST_ICMP_EQ = 4,
475 BIFROST_ICMP_NEQ = 5,
476 };
477
478 struct bifrost_fma_icmp32 {
479 unsigned src0 : 3;
480 unsigned src1 : 3;
481 enum bifrost_icmp_cond cond : 3;
482 unsigned unk1 : 1; /* set */
483 unsigned d3d : 1;
484 unsigned op : 12;
485 } __attribute__((packed));
486
487 struct bifrost_fma_icmp16 {
488 unsigned src0 : 3;
489 unsigned src1 : 3;
490 unsigned unk : 5; /* 11010 */
491 enum bifrost_icmp_cond cond : 3;
492 unsigned op : 9;
493 } __attribute__((packed));
494
495 struct bifrost_add_icmp {
496 unsigned src0 : 3;
497 unsigned src1 : 3;
498 enum bifrost_icmp_cond cond : 3;
499 unsigned sz : 1; /* 1 for 32, 0 for 8 */
500 unsigned d3d : 1;
501 unsigned op : 9;
502 } __attribute__((packed));
503
504 /* Two sources for vectorization */
505 #define BIFROST_FMA_FLOAT32_TO_16 (0xdd000 >> 3)
506 #define BIFROST_ADD_FLOAT32_TO_16 (0x0EC00 >> 3)
507
508 enum bifrost_convert_mode {
509 BIFROST_CONV_UNK0 = 0,
510 BIFROST_CONV_F32_TO_I32 = 1,
511 BIFROST_CONV_F16_TO_I16 = 2,
512 BIFROST_CONV_I32_TO_F32 = 3,
513 BIFROST_CONV_I16_TO_X32 = 4,
514 BIFROST_CONV_F16_TO_F32 = 5,
515 BIFROST_CONV_I16_TO_F16 = 6,
516 BIFROST_CONV_UNK7 = 7
517 };
518
519 /* i16 to x32 */
520 #define BIFROST_CONVERT_4(is_unsigned, component, to_float) \
521 ((is_unsigned & 1) | ((component & 1) << 1) | ((to_float & 1) << 2) | \
522 ((0x3) << 3) | ((4) << 5) | 0x100)
523
524 /* f16 to f32 */
525 #define BIFROST_CONVERT_5(component) \
526 ((component & 1) | ((1) << 1) | ((5) << 5) | 0x100)
527
528 /* Other conversions */
529 #define BIFROST_CONVERT(is_unsigned, roundmode, swizzle, mode) \
530 ((is_unsigned & 1) | ((roundmode & 3) << 1) | ((swizzle & 3) << 3) | ((mode & 7) << 5))
531
532 #define BIFROST_FMA_CONVERT (0xe0000)
533 #define BIFROST_ADD_CONVERT (0x07800)
534
535 enum bifrost_ldst_type {
536 BIFROST_LDST_F16 = 0,
537 BIFROST_LDST_F32 = 1,
538 BIFROST_LDST_I32 = 2,
539 BIFROST_LDST_U32 = 3
540 };
541
542 #define BIFROST_ADD_OP_LD_VAR_ADDR (0x18000 >> 10)
543
544 struct bifrost_ld_var_addr {
545 unsigned src0 : 3;
546 unsigned src1 : 3;
547 unsigned location : 5;
548 enum bifrost_ldst_type type : 2;
549 unsigned op : 7;
550 } __attribute__((packed));
551
552 #define BIFROST_ADD_OP_LD_ATTR (0x08000 >> 12)
553
554 struct bifrost_ld_attr {
555 unsigned src0 : 3;
556 unsigned src1 : 3;
557 unsigned location : 5;
558 unsigned channels : 2; /* MALI_POSITIVE */
559 enum bifrost_ldst_type type : 2;
560 unsigned op : 5;
561 } __attribute__((packed));
562
563 enum bifrost_interp_mode {
564 BIFROST_INTERP_PER_FRAG = 0x0,
565 BIFROST_INTERP_CENTROID = 0x1,
566 BIFROST_INTERP_DEFAULT = 0x2,
567 BIFROST_INTERP_EXPLICIT = 0x3
568 };
569
570 #define BIFROST_ADD_OP_LD_VAR_16 (0x1a << 1)
571 #define BIFROST_ADD_OP_LD_VAR_32 (0x0a << 1)
572
573 struct bifrost_ld_var {
574 unsigned src0 : 3;
575
576 /* If top two bits set, indirect with src in bottom three */
577 unsigned addr : 5;
578
579 unsigned channels : 2; /* MALI_POSITIVE */
580 enum bifrost_interp_mode interp_mode : 2;
581 unsigned reuse : 1;
582 unsigned flat : 1;
583 unsigned op : 6;
584 } __attribute__((packed));
585
586 struct bifrost_tex_ctrl {
587 unsigned sampler_index : 4; // also used to signal indirects
588 unsigned tex_index : 7;
589 bool no_merge_index : 1; // whether to merge (direct) sampler & texture indices
590 bool filter : 1; // use the usual filtering pipeline (0 for texelFetch & textureGather)
591 unsigned unk0 : 2;
592 bool texel_offset : 1; // *Offset()
593 bool is_shadow : 1;
594 bool is_array : 1;
595 unsigned tex_type : 2; // 2D, 3D, Cube, Buffer
596 bool compute_lod : 1; // 0 for *Lod()
597 bool not_supply_lod : 1; // 0 for *Lod() or when a bias is applied
598 bool calc_gradients : 1; // 0 for *Grad()
599 unsigned unk1 : 1;
600 unsigned result_type : 4; // integer, unsigned, float TODO: why is this 4 bits?
601 unsigned unk2 : 4;
602 } __attribute__((packed));
603
604 struct bifrost_dual_tex_ctrl {
605 unsigned sampler_index0 : 2;
606 unsigned unk0 : 2;
607 unsigned tex_index0 : 2;
608 unsigned sampler_index1 : 2;
609 unsigned tex_index1 : 2;
610 unsigned unk1 : 22;
611 } __attribute__((packed));
612
613 #define BIFROST_ADD_OP_TEX_COMPACT_F32 (0x0b000 >> 10)
614 #define BIFROST_ADD_OP_TEX_COMPACT_F16 (0x1b000 >> 10)
615
616 struct bifrost_tex_compact {
617 unsigned src0 : 3;
618 unsigned src1 : 3;
619 unsigned tex_index : 3;
620 unsigned unknown : 1;
621 unsigned sampler_index : 3;
622 unsigned op : 7;
623 } __attribute__((packed));
624
625 enum branch_bit_size {
626 BR_SIZE_32 = 0,
627 BR_SIZE_16XX = 1,
628 BR_SIZE_16YY = 2,
629 // For the above combinations of bitsize and location, an extra bit is
630 // encoded via comparing the sources. The only possible source of ambiguity
631 // would be if the sources were the same, but then the branch condition
632 // would be always true or always false anyways, so we can ignore it. But
633 // this no longer works when comparing the y component to the x component,
634 // since it's valid to compare the y component of a source against its own
635 // x component. Instead, the extra bit is encoded via an extra bitsize.
636 BR_SIZE_16YX0 = 3,
637 BR_SIZE_16YX1 = 4,
638 BR_SIZE_32_AND_16X = 5,
639 BR_SIZE_32_AND_16Y = 6,
640 // Used for comparisons with zero and always-true, see below. I think this
641 // only works for integer comparisons.
642 BR_SIZE_ZERO = 7,
643 };
644
645 enum bifrost_reg_write_unit {
646 REG_WRITE_NONE = 0, // don't write
647 REG_WRITE_TWO, // write using reg2
648 REG_WRITE_THREE, // write using reg3
649 };
650
651 struct bifrost_regs {
652 unsigned uniform_const : 8;
653 unsigned reg2 : 6;
654 unsigned reg3 : 6;
655 unsigned reg0 : 5;
656 unsigned reg1 : 6;
657 unsigned ctrl : 4;
658 } __attribute__((packed));
659
660 enum bifrost_branch_cond {
661 BR_COND_LT = 0,
662 BR_COND_LE = 1,
663 BR_COND_GE = 2,
664 BR_COND_GT = 3,
665 // Equal vs. not-equal determined by src0/src1 comparison
666 BR_COND_EQ = 4,
667 // floating-point comparisons
668 // Becomes UNE when you flip the arguments
669 BR_COND_OEQ = 5,
670 // TODO what happens when you flip the arguments?
671 BR_COND_OGT = 6,
672 BR_COND_OLT = 7,
673 };
674
675 enum bifrost_branch_code {
676 BR_ALWAYS = 63,
677 };
678
679 struct bifrost_branch {
680 unsigned src0 : 3;
681
682 /* For BR_SIZE_ZERO, upper two bits become ctrl */
683 unsigned src1 : 3;
684
685 /* Offset source -- always uniform/const but
686 * theoretically could support indirect jumps? */
687 unsigned src2 : 3;
688
689 enum bifrost_branch_cond cond : 3;
690 enum branch_bit_size size : 3;
691
692 unsigned op : 5;
693 };
694
695 /* Clause packing */
696
697 #define BIFROST_FMA_NOP (0x701960 | BIFROST_SRC_STAGE)
698 #define BIFROST_ADD_NOP (0x3D960 | BIFROST_SRC_STAGE)
699
700 struct bifrost_fmt1 {
701 unsigned ins_0 : 3;
702 unsigned tag : 5;
703 uint64_t ins_1 : 64;
704 unsigned ins_2 : 11;
705 uint64_t header : 45;
706 } __attribute__((packed));
707
708 #define BIFROST_FMT1_INSTRUCTIONS 0b00101
709 #define BIFROST_FMT1_FINAL 0b01001
710 #define BIFROST_FMT1_CONSTANTS 0b00001
711
712 #define BIFROST_FMTC_CONSTANTS 0b0011
713 #define BIFROST_FMTC_FINAL 0b0111
714
715 struct bifrost_fmt_constant {
716 unsigned pos : 4;
717 unsigned tag : 4;
718 uint64_t imm_1 : 60;
719 uint64_t imm_2 : 60;
720 } __attribute__((packed));
721
722 enum bifrost_reg_control {
723 BIFROST_WRITE_FMA_P2 = 1,
724 BIFROST_WRITE_FMA_P2_READ_P3 = 2,
725 BIFROST_FIRST_WRITE_FMA_P2_READ_P3 = 3,
726 BIFROST_READ_P3 = 4,
727 BIFROST_WRITE_ADD_P2 = 5,
728 BIFROST_WRITE_ADD_P2_READ_P3 = 6,
729 BIFROST_WRITE_ADD_P2_FMA_P3 = 7,
730
731 BIFROST_FIRST_NONE = 8,
732 BIFROST_FIRST_WRITE_FMA_P2 = 9,
733 /* INSTR_INVALID_ENC */
734 BIFROST_REG_NONE = 11,
735 BIFROST_FIRST_READ_P3 = 12,
736 BIFROST_FIRST_WRITE_ADD_P2 = 13,
737 BIFROST_FIRST_WRITE_ADD_P2_READ_P3 = 14,
738 BIFROST_FIRST_WRITE_ADD_P2_FMA_P3 = 15
739 };
740
741 #endif