panfrost/midgard: Add more ball/bany, iabs ops
[mesa.git] / src / gallium / drivers / panfrost / midgard / midgard.h
1 /* Author(s):
2 * Connor Abbott
3 * Alyssa Rosenzweig
4 *
5 * Copyright (c) 2013 Connor Abbott (connor@abbott.cx)
6 * Copyright (c) 2018 Alyssa Rosenzweig (alyssa@rosenzweig.io)
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
21 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 */
26
27 #ifndef __midgard_h__
28 #define __midgard_h__
29
30 #include <stdint.h>
31 #include <stdbool.h>
32
33 #define MIDGARD_DBG_MSGS 0x0001
34 #define MIDGARD_DBG_SHADERS 0x0002
35
36 extern int midgard_debug;
37
38 typedef enum {
39 midgard_word_type_alu,
40 midgard_word_type_load_store,
41 midgard_word_type_texture,
42 midgard_word_type_unknown
43 } midgard_word_type;
44
45 typedef enum {
46 midgard_alu_vmul,
47 midgard_alu_sadd,
48 midgard_alu_smul,
49 midgard_alu_vadd,
50 midgard_alu_lut
51 } midgard_alu;
52
53 /*
54 * ALU words
55 */
56
57 typedef enum {
58 midgard_alu_op_fadd = 0x10,
59 midgard_alu_op_fmul = 0x14,
60 midgard_alu_op_fmin = 0x28,
61 midgard_alu_op_fmax = 0x2C,
62 midgard_alu_op_fmov = 0x30,
63 midgard_alu_op_froundeven = 0x34,
64 midgard_alu_op_ftrunc = 0x35,
65 midgard_alu_op_ffloor = 0x36,
66 midgard_alu_op_fceil = 0x37,
67 midgard_alu_op_ffma = 0x38,
68 midgard_alu_op_fdot3 = 0x3C,
69 midgard_alu_op_fdot3r = 0x3D,
70 midgard_alu_op_fdot4 = 0x3E,
71 midgard_alu_op_freduce = 0x3F,
72 midgard_alu_op_iadd = 0x40,
73 midgard_alu_op_ishladd = 0x41,
74 midgard_alu_op_isub = 0x46,
75 midgard_alu_op_imul = 0x58,
76 midgard_alu_op_imin = 0x60,
77 midgard_alu_op_imax = 0x62,
78 midgard_alu_op_iasr = 0x68,
79 midgard_alu_op_ilsr = 0x69,
80 midgard_alu_op_ishl = 0x6E,
81 midgard_alu_op_iand = 0x70,
82 midgard_alu_op_ior = 0x71,
83 midgard_alu_op_inot = 0x72,
84 midgard_alu_op_iandnot = 0x74, /* (a, b) -> a & ~b, used for not/b2f */
85 midgard_alu_op_ixor = 0x76,
86 midgard_alu_op_imov = 0x7B,
87 midgard_alu_op_iabs = 0x7C,
88 midgard_alu_op_feq = 0x80,
89 midgard_alu_op_fne = 0x81,
90 midgard_alu_op_flt = 0x82,
91 midgard_alu_op_fle = 0x83,
92 midgard_alu_op_fball_eq = 0x88,
93 midgard_alu_op_bball_eq = 0x89,
94 midgard_alu_op_fball_lt = 0x8A, /* all(lessThan(.., ..)) */
95 midgard_alu_op_fball_lte = 0x8B, /* all(lessThanEqual(.., ..)) */
96 midgard_alu_op_bbany_neq = 0x90, /* used for bvec4(1) */
97 midgard_alu_op_fbany_neq = 0x91, /* bvec4(0) also */
98 midgard_alu_op_fbany_lt = 0x92, /* any(lessThan(.., ..)) */
99 midgard_alu_op_fbany_lte = 0x93, /* any(lessThanEqual(.., ..)) */
100 midgard_alu_op_f2i = 0x99,
101 midgard_alu_op_f2u8 = 0x9C,
102 midgard_alu_op_f2u = 0x9D,
103
104 midgard_alu_op_ieq = 0xA0,
105 midgard_alu_op_ine = 0xA1,
106 midgard_alu_op_ilt = 0xA4,
107 midgard_alu_op_ile = 0xA5,
108 midgard_alu_op_iball_eq = 0xA8,
109 midgard_alu_op_ball = 0xA9,
110 midgard_alu_op_uball_lt = 0xAA,
111 midgard_alu_op_uball_lte = 0xAB,
112 midgard_alu_op_iball_lt = 0xAC,
113 midgard_alu_op_iball_lte = 0xAD,
114 midgard_alu_op_ibany_eq = 0xB0,
115 midgard_alu_op_ibany_neq = 0xB1,
116 midgard_alu_op_ubany_lt = 0xB2,
117 midgard_alu_op_ubany_lte = 0xB3,
118 midgard_alu_op_ibany_lt = 0xB4, /* any(lessThan(.., ..)) */
119 midgard_alu_op_ibany_lte = 0xB5, /* any(lessThanEqual(.., ..)) */
120 midgard_alu_op_i2f = 0xB8,
121 midgard_alu_op_u2f = 0xBC,
122 midgard_alu_op_icsel = 0xC1,
123 midgard_alu_op_fcsel_i = 0xC4,
124 midgard_alu_op_fcsel = 0xC5,
125 midgard_alu_op_fround = 0xC6,
126 midgard_alu_op_fatan_pt2 = 0xE8,
127 midgard_alu_op_fpow = 0xEC,
128 midgard_alu_op_frcp = 0xF0,
129 midgard_alu_op_frsqrt = 0xF2,
130 midgard_alu_op_fsqrt = 0xF3,
131 midgard_alu_op_fexp2 = 0xF4,
132 midgard_alu_op_flog2 = 0xF5,
133 midgard_alu_op_fsin = 0xF6,
134 midgard_alu_op_fcos = 0xF7,
135 midgard_alu_op_fatan2_pt1 = 0xF9,
136 } midgard_alu_op;
137
138 typedef enum {
139 midgard_outmod_none = 0,
140 midgard_outmod_pos = 1,
141 midgard_outmod_int = 2,
142 midgard_outmod_sat = 3
143 } midgard_outmod;
144
145 typedef enum {
146 midgard_reg_mode_quarter = 0,
147 midgard_reg_mode_half = 1,
148 midgard_reg_mode_full = 2,
149 midgard_reg_mode_double = 3 /* TODO: verify */
150 } midgard_reg_mode;
151
152 typedef enum {
153 midgard_dest_override_lower = 0,
154 midgard_dest_override_upper = 1,
155 midgard_dest_override_none = 2
156 } midgard_dest_override;
157
158 typedef struct
159 __attribute__((__packed__))
160 {
161 bool abs : 1;
162 bool negate : 1;
163
164 /* replicate lower half if dest = half, or low/high half selection if
165 * dest = full
166 */
167 bool rep_low : 1;
168 bool rep_high : 1; /* unused if dest = full */
169 bool half : 1; /* only matters if dest = full */
170 unsigned swizzle : 8;
171 }
172 midgard_vector_alu_src;
173
174 typedef struct
175 __attribute__((__packed__))
176 {
177 midgard_alu_op op : 8;
178 midgard_reg_mode reg_mode : 2;
179 unsigned src1 : 13;
180 unsigned src2 : 13;
181 midgard_dest_override dest_override : 2;
182 midgard_outmod outmod : 2;
183 unsigned mask : 8;
184 }
185 midgard_vector_alu;
186
187 typedef struct
188 __attribute__((__packed__))
189 {
190 bool abs : 1;
191 bool negate : 1;
192 bool full : 1; /* 0 = half, 1 = full */
193 unsigned component : 3;
194 }
195 midgard_scalar_alu_src;
196
197 typedef struct
198 __attribute__((__packed__))
199 {
200 midgard_alu_op op : 8;
201 unsigned src1 : 6;
202 unsigned src2 : 11;
203 unsigned unknown : 1;
204 midgard_outmod outmod : 2;
205 bool output_full : 1;
206 unsigned output_component : 3;
207 }
208 midgard_scalar_alu;
209
210 typedef struct
211 __attribute__((__packed__))
212 {
213 unsigned src1_reg : 5;
214 unsigned src2_reg : 5;
215 unsigned out_reg : 5;
216 bool src2_imm : 1;
217 }
218 midgard_reg_info;
219
220 typedef enum {
221 midgard_jmp_writeout_op_branch_uncond = 1,
222 midgard_jmp_writeout_op_branch_cond = 2,
223 midgard_jmp_writeout_op_discard = 4,
224 midgard_jmp_writeout_op_writeout = 7,
225 } midgard_jmp_writeout_op;
226
227 typedef enum {
228 midgard_condition_write0 = 0,
229 midgard_condition_false = 1,
230 midgard_condition_true = 2,
231 midgard_condition_always = 3, /* Special for writeout/uncond discard */
232 } midgard_condition;
233
234 typedef struct
235 __attribute__((__packed__))
236 {
237 midgard_jmp_writeout_op op : 3; /* == branch_uncond */
238 unsigned dest_tag : 4; /* tag of branch destination */
239 unsigned unknown : 2;
240 int offset : 7;
241 }
242 midgard_branch_uncond;
243
244 typedef struct
245 __attribute__((__packed__))
246 {
247 midgard_jmp_writeout_op op : 3; /* == branch_cond */
248 unsigned dest_tag : 4; /* tag of branch destination */
249 int offset : 7;
250 midgard_condition cond : 2;
251 }
252 midgard_branch_cond;
253
254 typedef struct
255 __attribute__((__packed__))
256 {
257 midgard_jmp_writeout_op op : 3; /* == branch_cond */
258 unsigned dest_tag : 4; /* tag of branch destination */
259 unsigned unknown : 2;
260 signed offset : 23;
261 unsigned cond : 16;
262 }
263 midgard_branch_extended;
264
265 typedef struct
266 __attribute__((__packed__))
267 {
268 midgard_jmp_writeout_op op : 3; /* == writeout */
269 unsigned unknown : 13;
270 }
271 midgard_writeout;
272
273 /*
274 * Load/store words
275 */
276
277 typedef enum {
278 midgard_op_ld_st_noop = 0x03,
279 midgard_op_load_attr_16 = 0x95,
280 midgard_op_load_attr_32 = 0x94,
281 midgard_op_load_vary_16 = 0x99,
282 midgard_op_load_vary_32 = 0x98,
283 midgard_op_load_color_buffer_16 = 0x9D,
284 midgard_op_load_color_buffer_8 = 0xBA,
285 midgard_op_load_uniform_16 = 0xAC,
286 midgard_op_load_uniform_32 = 0xB0,
287 midgard_op_store_vary_16 = 0xD5,
288 midgard_op_store_vary_32 = 0xD4
289 } midgard_load_store_op;
290
291 typedef enum {
292 midgard_interp_centroid = 1,
293 midgard_interp_default = 2
294 } midgard_interpolation;
295
296 typedef struct
297 __attribute__((__packed__))
298 {
299 unsigned zero1 : 4; /* Always zero */
300
301 /* Varying qualifiers, zero if not a varying */
302 unsigned flat : 1;
303 unsigned is_varying : 1; /* Always one for varying, but maybe something else? */
304 midgard_interpolation interpolation : 2;
305
306 unsigned zero2 : 2; /* Always zero */
307 }
308 midgard_varying_parameter;
309
310 typedef struct
311 __attribute__((__packed__))
312 {
313 midgard_load_store_op op : 8;
314 unsigned reg : 5;
315 unsigned mask : 4;
316 unsigned swizzle : 8;
317 unsigned unknown : 16;
318
319 unsigned varying_parameters : 10;
320
321 unsigned address : 9;
322 }
323 midgard_load_store_word;
324
325 typedef struct
326 __attribute__((__packed__))
327 {
328 unsigned type : 4;
329 unsigned next_type : 4;
330 uint64_t word1 : 60;
331 uint64_t word2 : 60;
332 }
333 midgard_load_store;
334
335 /* Texture pipeline results are in r28-r29 */
336 #define REG_TEX_BASE 28
337
338 /* Texture opcodes... maybe? */
339 #define TEXTURE_OP_NORMAL 0x11
340 #define TEXTURE_OP_TEXEL_FETCH 0x14
341
342 /* Texture format types, found in format */
343 #define TEXTURE_CUBE 0x00
344 #define TEXTURE_2D 0x02
345 #define TEXTURE_3D 0x03
346
347 typedef struct
348 __attribute__((__packed__))
349 {
350 unsigned type : 4;
351 unsigned next_type : 4;
352
353 unsigned op : 6;
354 unsigned shadow : 1;
355 unsigned unknown3 : 1;
356
357 /* A little obscure, but last is set for the last texture operation in
358 * a shader. cont appears to just be last's opposite (?). Yeah, I know,
359 * kind of funky.. BiOpen thinks it could do with memory hinting, or
360 * tile locking? */
361
362 unsigned cont : 1;
363 unsigned last : 1;
364
365 unsigned format : 5;
366 unsigned has_offset : 1;
367
368 /* Like in Bifrost */
369 unsigned filter : 1;
370
371 unsigned in_reg_select : 1;
372 unsigned in_reg_upper : 1;
373
374 unsigned in_reg_swizzle_left : 2;
375 unsigned in_reg_swizzle_right : 2;
376
377 unsigned unknown1 : 2;
378
379 unsigned unknown8 : 4;
380
381 unsigned out_full : 1;
382
383 /* Always 1 afaict... */
384 unsigned unknown7 : 2;
385
386 unsigned out_reg_select : 1;
387 unsigned out_upper : 1;
388
389 unsigned mask : 4;
390
391 unsigned unknown2 : 2;
392
393 unsigned swizzle : 8;
394 unsigned unknown4 : 8;
395
396 unsigned unknownA : 4;
397
398 unsigned offset_unknown1 : 1;
399 unsigned offset_reg_select : 1;
400 unsigned offset_reg_upper : 1;
401 unsigned offset_unknown4 : 1;
402 unsigned offset_unknown5 : 1;
403 unsigned offset_unknown6 : 1;
404 unsigned offset_unknown7 : 1;
405 unsigned offset_unknown8 : 1;
406 unsigned offset_unknown9 : 1;
407
408 unsigned unknownB : 3;
409
410 /* Texture bias or LOD, depending on whether it is executed in a
411 * fragment/vertex shader respectively. Compute as int(2^8 * biasf).
412 *
413 * For texel fetch, this is the LOD as is. */
414 unsigned bias : 8;
415
416 unsigned unknown9 : 8;
417
418 unsigned texture_handle : 16;
419 unsigned sampler_handle : 16;
420 }
421 midgard_texture_word;
422
423 /* Opcode name table */
424
425 static char *alu_opcode_names[256] = {
426 [midgard_alu_op_fadd] = "fadd",
427 [midgard_alu_op_fmul] = "fmul",
428 [midgard_alu_op_fmin] = "fmin",
429 [midgard_alu_op_fmax] = "fmax",
430 [midgard_alu_op_fmov] = "fmov",
431 [midgard_alu_op_froundeven] = "froundeven",
432 [midgard_alu_op_ftrunc] = "ftrunc",
433 [midgard_alu_op_ffloor] = "ffloor",
434 [midgard_alu_op_fceil] = "fceil",
435 [midgard_alu_op_ffma] = "ffma",
436 [midgard_alu_op_fdot3] = "fdot3",
437 [midgard_alu_op_fdot3r] = "fdot3r",
438 [midgard_alu_op_fdot4] = "fdot4",
439 [midgard_alu_op_freduce] = "freduce",
440 [midgard_alu_op_imin] = "imin",
441 [midgard_alu_op_imax] = "imax",
442 [midgard_alu_op_ishl] = "ishl",
443 [midgard_alu_op_iasr] = "iasr",
444 [midgard_alu_op_ilsr] = "ilsr",
445 [midgard_alu_op_iadd] = "iadd",
446 [midgard_alu_op_ishladd] = "ishladd",
447 [midgard_alu_op_isub] = "isub",
448 [midgard_alu_op_imul] = "imul",
449 [midgard_alu_op_imov] = "imov",
450 [midgard_alu_op_iabs] = "iabs",
451 [midgard_alu_op_iand] = "iand",
452 [midgard_alu_op_ior] = "ior",
453 [midgard_alu_op_inot] = "inot",
454 [midgard_alu_op_iandnot] = "iandnot",
455 [midgard_alu_op_ixor] = "ixor",
456 [midgard_alu_op_feq] = "feq",
457 [midgard_alu_op_fne] = "fne",
458 [midgard_alu_op_flt] = "flt",
459 [midgard_alu_op_fle] = "fle",
460 [midgard_alu_op_fball_eq] = "fball_eq",
461 [midgard_alu_op_fbany_neq] = "fbany_neq",
462 [midgard_alu_op_bball_eq] = "bball_eq",
463 [midgard_alu_op_fball_lt] = "fball_lt",
464 [midgard_alu_op_fball_lte] = "fball_lte",
465 [midgard_alu_op_bbany_neq] = "bbany_neq",
466 [midgard_alu_op_fbany_lt] = "fbany_lt",
467 [midgard_alu_op_fbany_lte] = "fbany_lte",
468 [midgard_alu_op_f2i] = "f2i",
469 [midgard_alu_op_f2u] = "f2u",
470 [midgard_alu_op_f2u8] = "f2u8",
471 [midgard_alu_op_ieq] = "ieq",
472 [midgard_alu_op_ine] = "ine",
473 [midgard_alu_op_ilt] = "ilt",
474 [midgard_alu_op_ile] = "ile",
475 [midgard_alu_op_iball_eq] = "iball_eq",
476 [midgard_alu_op_ball] = "ball",
477 [midgard_alu_op_uball_lt] = "uball_lt",
478 [midgard_alu_op_uball_lte] = "uball_lte",
479 [midgard_alu_op_iball_lt] = "iball_lt",
480 [midgard_alu_op_iball_lte] = "iball_lte",
481 [midgard_alu_op_iball_eq] = "iball_eq",
482 [midgard_alu_op_ibany_neq] = "ibany_neq",
483 [midgard_alu_op_ubany_lt] = "ubany_lt",
484 [midgard_alu_op_ubany_lte] = "ubany_lte",
485 [midgard_alu_op_ibany_lt] = "ibany_lt",
486 [midgard_alu_op_ibany_lte] = "ibany_lte",
487 [midgard_alu_op_i2f] = "i2f",
488 [midgard_alu_op_u2f] = "u2f",
489 [midgard_alu_op_icsel] = "icsel",
490 [midgard_alu_op_fcsel_i] = "fcsel_i",
491 [midgard_alu_op_fcsel] = "fcsel",
492 [midgard_alu_op_fround] = "fround",
493 [midgard_alu_op_fatan_pt2] = "fatan_pt2",
494 [midgard_alu_op_frcp] = "frcp",
495 [midgard_alu_op_frsqrt] = "frsqrt",
496 [midgard_alu_op_fsqrt] = "fsqrt",
497 [midgard_alu_op_fpow] = "fpow",
498 [midgard_alu_op_fexp2] = "fexp2",
499 [midgard_alu_op_flog2] = "flog2",
500 [midgard_alu_op_fsin] = "fsin",
501 [midgard_alu_op_fcos] = "fcos",
502 [midgard_alu_op_fatan2_pt1] = "fatan2_pt1"
503 };
504
505 static char *load_store_opcode_names[256] = {
506 [midgard_op_load_attr_16] = "ld_attr_16",
507 [midgard_op_load_attr_32] = "ld_attr_32",
508 [midgard_op_load_vary_16] = "ld_vary_16",
509 [midgard_op_load_vary_32] = "ld_vary_32",
510 [midgard_op_load_uniform_16] = "ld_uniform_16",
511 [midgard_op_load_uniform_32] = "ld_uniform_32",
512 [midgard_op_load_color_buffer_8] = "ld_color_buffer_8",
513 [midgard_op_load_color_buffer_16] = "ld_color_buffer_16",
514 [midgard_op_store_vary_16] = "st_vary_16",
515 [midgard_op_store_vary_32] = "st_vary_32"
516 };
517
518 #endif