pan/bi: Pack FMA SEL16
[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_MSCALE (0x50 >> 3)
116
117 struct bifrost_fma_mscale {
118 unsigned src0 : 3;
119 unsigned src1 : 3;
120 unsigned src2 : 3;
121 unsigned src3 : 3;
122
123 /* If mscale_mode is set - an MSCALE specific mode. If it is not set, a
124 * regular outmod */
125 unsigned mode : 2;
126 unsigned mscale_mode : 1;
127
128 unsigned src0_abs : 1;
129 unsigned src1_neg : 1;
130 unsigned src2_neg : 1;
131 unsigned op : 5;
132 } __attribute__((packed));
133
134 #define BIFROST_ADD_OP_BLEND (0x1952c)
135 #define BIFROST_ADD_OP_FRCP_FAST_F32 (0x0cc00)
136 #define BIFROST_ADD_OP_FRCP_FAST_F16_X (0x0ce10)
137 #define BIFROST_ADD_OP_FRCP_FAST_F16_Y (0x0ce30)
138 #define BIFROST_ADD_OP_FRSQ_FAST_F32 (0x0cc20)
139 #define BIFROST_ADD_OP_FRSQ_FAST_F16_X (0x0ce50)
140 #define BIFROST_ADD_OP_FRSQ_FAST_F16_Y (0x0ce70)
141 #define BIFROST_ADD_OP_LOG2_HELP (0x0cc68)
142 #define BIFROST_ADD_OP_FEXP2_FAST (0x0cd58)
143
144 struct bifrost_add_inst {
145 unsigned src0 : 3;
146 unsigned op : 17;
147 } __attribute__((packed));
148
149 #define BIFROST_ADD_OP_LD_UBO_1 (0x0c1a0 >> 3)
150 #define BIFROST_ADD_OP_LD_UBO_2 (0x0c1e0 >> 3)
151 #define BIFROST_ADD_OP_LD_UBO_3 (0x0caa0 >> 3)
152 #define BIFROST_ADD_OP_LD_UBO_4 (0x0c220 >> 3)
153 #define BIFROST_ADD_SEL_16(swiz) ((0xea60 >> 3) | (swiz))
154
155 struct bifrost_add_2src {
156 unsigned src0 : 3;
157 unsigned src1 : 3;
158 unsigned op : 14;
159 } __attribute__((packed));
160
161 #define BIFROST_ADD_OP_FMAX32 (0x00)
162 #define BIFROST_ADD_OP_FMIN32 (0x01)
163 #define BIFROST_ADD_OP_FADD32 (0x02)
164
165 #define BIFROST_ADD_OP_FADD16 (0x0A)
166
167 struct bifrost_add_faddmin {
168 unsigned src0 : 3;
169 unsigned src1 : 3;
170 unsigned src1_abs : 1;
171 unsigned src0_neg : 1;
172 unsigned src1_neg : 1;
173 unsigned select : 2; /* swizzle_0 for fp16 */
174 unsigned outmod : 2; /* swizzle_1 for fp16 */
175 unsigned mode : 2;
176 unsigned src0_abs : 1;
177 unsigned op : 4;
178 } __attribute__((packed));
179
180 #define BIFROST_ADD_OP_FMAX16 (0x10)
181 #define BIFROST_ADD_OP_FMIN16 (0x12)
182
183 struct bifrost_add_fmin16 {
184 unsigned src0 : 3;
185 unsigned src1 : 3;
186 /* abs2 inferred as with FMA */
187 unsigned abs1 : 1;
188 unsigned src0_neg : 1;
189 unsigned src1_neg : 1;
190 unsigned src0_swizzle : 2;
191 unsigned src1_swizzle : 2;
192 unsigned mode : 2;
193 unsigned op : 5;
194 } __attribute__((packed));
195
196 #define BIFROST_ADD_OP_ST_VAR (0x19300 >> 8)
197
198 struct bifrost_st_vary {
199 unsigned src0 : 3;
200 unsigned src1 : 3;
201 unsigned src2 : 3;
202 unsigned channels : 2;
203 unsigned op : 9;
204 } __attribute__((packed));
205
206 #define BIFROST_ADD_OP_ATEST (0xc8f)
207
208 struct bifrost_add_atest {
209 /* gl_SampleMask (R60) */
210 unsigned src0 : 3;
211
212 /* Alpha value */
213 unsigned src1 : 3;
214
215 /* If half, X/Y select. If !half, always set */
216 unsigned component : 1;
217 unsigned half : 1;
218
219 unsigned op : 12;
220 } __attribute__((packed));
221
222 enum bifrost_outmod {
223 BIFROST_NONE = 0x0,
224 BIFROST_POS = 0x1,
225 BIFROST_SAT_SIGNED = 0x2,
226 BIFROST_SAT = 0x3,
227 };
228
229 enum bifrost_roundmode {
230 BIFROST_RTE = 0x0, /* round to even */
231 BIFROST_RTP = 0x1, /* round to positive */
232 BIFROST_RTN = 0x2, /* round to negative */
233 BIFROST_RTZ = 0x3 /* round to zero */
234 };
235
236 /* NONE: Same as fmax() and fmin() -- return the other
237 * number if any number is NaN. Also always return +0 if
238 * one argument is +0 and the other is -0.
239 *
240 * NAN_WINS: Instead of never returning a NaN, always return
241 * one. The "greater"/"lesser" NaN is always returned, first
242 * by checking the sign and then the mantissa bits.
243 *
244 * SRC1_WINS: For max, implement src0 > src1 ? src0 : src1.
245 * For min, implement src0 < src1 ? src0 : src1. This
246 * includes handling NaN's and signedness of 0 differently
247 * from above, since +0 and -0 compare equal and comparisons
248 * always return false for NaN's. As a result, this mode is
249 * *not* commutative.
250 *
251 * SRC0_WINS: For max, implement src0 < src1 ? src1 : src0
252 * For min, implement src0 > src1 ? src1 : src0
253 */
254
255
256 enum bifrost_minmax_mode {
257 BIFROST_MINMAX_NONE = 0x0,
258 BIFROST_NAN_WINS = 0x1,
259 BIFROST_SRC1_WINS = 0x2,
260 BIFROST_SRC0_WINS = 0x3,
261 };
262
263 #define BIFROST_FMA_OP_FADD32 (0x58 >> 2)
264 #define BIFROST_FMA_OP_FMAX32 (0x40 >> 2)
265 #define BIFROST_FMA_OP_FMIN32 (0x44 >> 2)
266
267 struct bifrost_fma_add {
268 unsigned src0 : 3;
269 unsigned src1 : 3;
270 unsigned src1_abs : 1;
271 unsigned src0_neg : 1;
272 unsigned src1_neg : 1;
273 unsigned unk : 3;
274 unsigned src0_abs : 1;
275 enum bifrost_roundmode roundmode : 2;
276 enum bifrost_outmod outmod : 2;
277 unsigned op : 6;
278 } __attribute__((packed));
279
280 #define BIFROST_FMA_OP_FMAX16 (0xC0 >> 2)
281 #define BIFROST_FMA_OP_FMIN16 (0xCC >> 2)
282 #define BIFROST_FMA_OP_FADD16 (0xD8 >> 2)
283
284 struct bifrost_fma_add_minmax16 {
285 unsigned src0 : 3;
286 unsigned src1 : 3;
287 /* abs2 inferred as (src1 < src0) */
288 unsigned abs1 : 1;
289 unsigned src0_neg : 1;
290 unsigned src1_neg : 1;
291 unsigned src0_swizzle : 2;
292 unsigned src1_swizzle : 2;
293 unsigned mode : 2;
294 enum bifrost_outmod outmod : 2;
295 /* roundmode for add, min/max mode for min/max */
296 unsigned op : 6;
297 } __attribute__((packed));
298
299 #define BIFROST_FMA_OP_FMA (0x00)
300
301 struct bifrost_fma_fma {
302 unsigned src0 : 3;
303 unsigned src1 : 3;
304 unsigned src2 : 3;
305 unsigned src_expand : 3;
306 unsigned src0_abs : 1;
307 enum bifrost_roundmode roundmode : 2;
308 enum bifrost_outmod outmod : 2;
309 unsigned src0_neg : 1; /* 14 */
310 unsigned src2_neg : 1;
311 unsigned src1_abs : 1;
312 unsigned src2_abs : 1; /* 17 */
313 unsigned op : 2;
314 } __attribute__((packed));
315
316 #define BIFROST_FMA_OP_FMA16 (0x2)
317
318 struct bifrost_fma_fma16 {
319 unsigned src0 : 3;
320 unsigned src1 : 3;
321 unsigned src2 : 3;
322 unsigned swizzle_0 : 2;
323 unsigned swizzle_1 : 2;
324 enum bifrost_roundmode roundmode : 2;
325 enum bifrost_outmod outmod : 2;
326 unsigned src0_neg : 1;
327 unsigned src2_neg : 1;
328 unsigned swizzle_2 : 2;
329 unsigned op : 2;
330 } __attribute__((packed));
331
332 enum bifrost_csel_cond {
333 BIFROST_FEQ_F = 0x0,
334 BIFROST_FGT_F = 0x1,
335 BIFROST_FGE_F = 0x2,
336 BIFROST_IEQ_F = 0x3,
337 BIFROST_IGT_I = 0x4,
338 BIFROST_IGE_I = 0x5,
339 BIFROST_UGT_I = 0x6,
340 BIFROST_UGE_I = 0x7
341 };
342
343 #define BIFROST_FMA_OP_CSEL4 (0x5c)
344 #define BIFROST_FMA_OP_CSEL4_V16 (0xdc)
345
346 struct bifrost_csel4 {
347 unsigned src0 : 3;
348 unsigned src1 : 3;
349 unsigned src2 : 3;
350 unsigned src3 : 3;
351 enum bifrost_csel_cond cond : 3;
352 unsigned op : 8;
353 } __attribute__((packed));
354
355 struct bifrost_shift_fma {
356 unsigned src0 : 3;
357 unsigned src1 : 3;
358 unsigned src2 : 3;
359 unsigned half : 3; /* 000 for i32, 100 for i8, 111 for v2i16 */
360 unsigned unk : 1; /* always set? */
361 unsigned invert_1 : 1; /* Inverts sources to combining op */
362 /* For XOR, switches RSHIFT to LSHIFT since only one invert needed */
363 unsigned invert_2 : 1;
364 unsigned op : 8;
365 } __attribute__((packed));
366
367 struct bifrost_shift_add {
368 unsigned src0 : 3;
369 unsigned src1 : 3;
370 unsigned src2 : 3;
371 unsigned zero : 2;
372
373 unsigned invert_1 : 1;
374 unsigned invert_2 : 1;
375
376 unsigned op : 7;
377 } __attribute__((packed));
378
379 /* Two sources for vectorization */
380 #define BIFROST_FMA_FLOAT32_TO_16 (0xdd000 >> 3)
381 #define BIFROST_ADD_FLOAT32_TO_16 (0x0EC00 >> 3)
382
383 enum bifrost_convert_mode {
384 BIFROST_CONV_UNK0 = 0,
385 BIFROST_CONV_F32_TO_I32 = 1,
386 BIFROST_CONV_F16_TO_I16 = 2,
387 BIFROST_CONV_I32_TO_F32 = 3,
388 BIFROST_CONV_I16_TO_X32 = 4,
389 BIFROST_CONV_F16_TO_F32 = 5,
390 BIFROST_CONV_I16_TO_F16 = 6,
391 BIFROST_CONV_UNK7 = 7
392 };
393
394 /* i16 to x32 */
395 #define BIFROST_CONVERT_4(is_unsigned, component, to_float) \
396 ((is_unsigned & 1) | ((component & 1) << 1) | ((to_float & 1) << 2) | \
397 ((0x3) << 3) | ((4) << 5) | 0x100)
398
399 /* f16 to f32 */
400 #define BIFROST_CONVERT_5(component) \
401 ((component & 1) | ((1) << 1) | ((5) << 5) | 0x100)
402
403 /* Other conversions */
404 #define BIFROST_CONVERT(is_unsigned, roundmode, swizzle, mode) \
405 ((is_unsigned & 1) | ((roundmode & 3) << 1) | ((swizzle & 3) << 3) | ((mode & 7) << 5))
406
407 #define BIFROST_FMA_CONVERT (0xe0000)
408 #define BIFROST_ADD_CONVERT (0x07800)
409
410 enum bifrost_ldst_type {
411 BIFROST_LDST_F16 = 0,
412 BIFROST_LDST_F32 = 1,
413 BIFROST_LDST_I32 = 2,
414 BIFROST_LDST_U32 = 3
415 };
416
417 #define BIFROST_ADD_OP_LD_VAR_ADDR (0x18000 >> 10)
418
419 struct bifrost_ld_var_addr {
420 unsigned src0 : 3;
421 unsigned src1 : 3;
422 unsigned location : 5;
423 enum bifrost_ldst_type type : 2;
424 unsigned op : 7;
425 } __attribute__((packed));
426
427 #define BIFROST_ADD_OP_LD_ATTR (0x08000 >> 12)
428
429 struct bifrost_ld_attr {
430 unsigned src0 : 3;
431 unsigned src1 : 3;
432 unsigned location : 5;
433 unsigned channels : 2; /* MALI_POSITIVE */
434 enum bifrost_ldst_type type : 2;
435 unsigned op : 5;
436 } __attribute__((packed));
437
438 enum bifrost_interp_mode {
439 BIFROST_INTERP_PER_FRAG = 0x0,
440 BIFROST_INTERP_CENTROID = 0x1,
441 BIFROST_INTERP_DEFAULT = 0x2,
442 BIFROST_INTERP_EXPLICIT = 0x3
443 };
444
445 #define BIFROST_ADD_OP_LD_VAR_16 (0x1a << 1)
446 #define BIFROST_ADD_OP_LD_VAR_32 (0x0a << 1)
447
448 struct bifrost_ld_var {
449 unsigned src0 : 3;
450
451 /* If top two bits set, indirect with src in bottom three */
452 unsigned addr : 5;
453
454 unsigned channels : 2; /* MALI_POSITIVE */
455 enum bifrost_interp_mode interp_mode : 2;
456 unsigned reuse : 1;
457 unsigned flat : 1;
458 unsigned op : 6;
459 } __attribute__((packed));
460
461 struct bifrost_tex_ctrl {
462 unsigned sampler_index : 4; // also used to signal indirects
463 unsigned tex_index : 7;
464 bool no_merge_index : 1; // whether to merge (direct) sampler & texture indices
465 bool filter : 1; // use the usual filtering pipeline (0 for texelFetch & textureGather)
466 unsigned unk0 : 2;
467 bool texel_offset : 1; // *Offset()
468 bool is_shadow : 1;
469 bool is_array : 1;
470 unsigned tex_type : 2; // 2D, 3D, Cube, Buffer
471 bool compute_lod : 1; // 0 for *Lod()
472 bool not_supply_lod : 1; // 0 for *Lod() or when a bias is applied
473 bool calc_gradients : 1; // 0 for *Grad()
474 unsigned unk1 : 1;
475 unsigned result_type : 4; // integer, unsigned, float TODO: why is this 4 bits?
476 unsigned unk2 : 4;
477 } __attribute__((packed));
478
479 struct bifrost_dual_tex_ctrl {
480 unsigned sampler_index0 : 2;
481 unsigned unk0 : 2;
482 unsigned tex_index0 : 2;
483 unsigned sampler_index1 : 2;
484 unsigned tex_index1 : 2;
485 unsigned unk1 : 22;
486 } __attribute__((packed));
487
488 #define BIFROST_ADD_OP_TEX_COMPACT_F32 (0x0b000 >> 10)
489 #define BIFROST_ADD_OP_TEX_COMPACT_F16 (0x1b000 >> 10)
490
491 struct bifrost_tex_compact {
492 unsigned src0 : 3;
493 unsigned src1 : 3;
494 unsigned tex_index : 3;
495 unsigned unknown : 1;
496 unsigned sampler_index : 3;
497 unsigned op : 7;
498 } __attribute__((packed));
499
500 enum branch_bit_size {
501 BR_SIZE_32 = 0,
502 BR_SIZE_16XX = 1,
503 BR_SIZE_16YY = 2,
504 // For the above combinations of bitsize and location, an extra bit is
505 // encoded via comparing the sources. The only possible source of ambiguity
506 // would be if the sources were the same, but then the branch condition
507 // would be always true or always false anyways, so we can ignore it. But
508 // this no longer works when comparing the y component to the x component,
509 // since it's valid to compare the y component of a source against its own
510 // x component. Instead, the extra bit is encoded via an extra bitsize.
511 BR_SIZE_16YX0 = 3,
512 BR_SIZE_16YX1 = 4,
513 BR_SIZE_32_AND_16X = 5,
514 BR_SIZE_32_AND_16Y = 6,
515 // Used for comparisons with zero and always-true, see below. I think this
516 // only works for integer comparisons.
517 BR_SIZE_ZERO = 7,
518 };
519
520 enum bifrost_reg_write_unit {
521 REG_WRITE_NONE = 0, // don't write
522 REG_WRITE_TWO, // write using reg2
523 REG_WRITE_THREE, // write using reg3
524 };
525
526 struct bifrost_regs {
527 unsigned uniform_const : 8;
528 unsigned reg2 : 6;
529 unsigned reg3 : 6;
530 unsigned reg0 : 5;
531 unsigned reg1 : 6;
532 unsigned ctrl : 4;
533 } __attribute__((packed));
534
535 enum bifrost_branch_cond {
536 BR_COND_LT = 0,
537 BR_COND_LE = 1,
538 BR_COND_GE = 2,
539 BR_COND_GT = 3,
540 // Equal vs. not-equal determined by src0/src1 comparison
541 BR_COND_EQ = 4,
542 // floating-point comparisons
543 // Becomes UNE when you flip the arguments
544 BR_COND_OEQ = 5,
545 // TODO what happens when you flip the arguments?
546 BR_COND_OGT = 6,
547 BR_COND_OLT = 7,
548 };
549
550 enum bifrost_branch_code {
551 BR_ALWAYS = 63,
552 };
553
554 struct bifrost_branch {
555 unsigned src0 : 3;
556
557 /* For BR_SIZE_ZERO, upper two bits become ctrl */
558 unsigned src1 : 3;
559
560 /* Offset source -- always uniform/const but
561 * theoretically could support indirect jumps? */
562 unsigned src2 : 3;
563
564 enum bifrost_branch_cond cond : 3;
565 enum branch_bit_size size : 3;
566
567 unsigned op : 5;
568 };
569
570 /* Clause packing */
571
572 #define BIFROST_FMA_NOP (0x701960 | BIFROST_SRC_STAGE)
573 #define BIFROST_ADD_NOP (0x3D960 | BIFROST_SRC_STAGE)
574
575 struct bifrost_fmt1 {
576 unsigned ins_0 : 3;
577 unsigned tag : 5;
578 uint64_t ins_1 : 64;
579 unsigned ins_2 : 11;
580 uint64_t header : 45;
581 } __attribute__((packed));
582
583 #define BIFROST_FMT1_INSTRUCTIONS 0b00101
584 #define BIFROST_FMT1_FINAL 0b01001
585 #define BIFROST_FMT1_CONSTANTS 0b00001
586
587 #define BIFROST_FMTC_CONSTANTS 0b0011
588 #define BIFROST_FMTC_FINAL 0b0111
589
590 struct bifrost_fmt_constant {
591 unsigned pos : 4;
592 unsigned tag : 4;
593 uint64_t imm_1 : 60;
594 uint64_t imm_2 : 60;
595 } __attribute__((packed));
596
597 enum bifrost_reg_control {
598 BIFROST_WRITE_FMA_P2 = 1,
599 BIFROST_WRITE_FMA_P2_READ_P3 = 2,
600 BIFROST_FIRST_WRITE_FMA_P2_READ_P3 = 3,
601 BIFROST_READ_P3 = 4,
602 BIFROST_WRITE_ADD_P2 = 5,
603 BIFROST_WRITE_ADD_P2_READ_P3 = 6,
604 BIFROST_WRITE_ADD_P2_FMA_P3 = 7,
605
606 BIFROST_FIRST_NONE = 8,
607 BIFROST_FIRST_WRITE_FMA_P2 = 9,
608 /* INSTR_INVALID_ENC */
609 BIFROST_REG_NONE = 11,
610 BIFROST_FIRST_READ_P3 = 12,
611 BIFROST_FIRST_WRITE_ADD_P2 = 13,
612 BIFROST_FIRST_WRITE_ADD_P2_READ_P3 = 14,
613 BIFROST_FIRST_WRITE_ADD_P2_FMA_P3 = 15
614 };
615
616 #endif