panfrost/midgard: Add ibitcount8 op
[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_umin = 0x61,
78 midgard_alu_op_imax = 0x62,
79 midgard_alu_op_umax = 0x63,
80 midgard_alu_op_iasr = 0x68,
81 midgard_alu_op_ilsr = 0x69,
82 midgard_alu_op_ishl = 0x6E,
83 midgard_alu_op_iand = 0x70,
84 midgard_alu_op_ior = 0x71,
85 midgard_alu_op_inot = 0x72,
86 midgard_alu_op_iandnot = 0x74, /* (a, b) -> a & ~b, used for not/b2f */
87 midgard_alu_op_ixor = 0x76,
88 midgard_alu_op_ilzcnt = 0x78, /* Number of zeroes on left. 31 - ilzcnt(x) = findMSB(x) */
89 midgard_alu_op_ibitcount8 = 0x7A, /* Counts bits in 8-bit increments */
90 midgard_alu_op_imov = 0x7B,
91 midgard_alu_op_iabs = 0x7C,
92 midgard_alu_op_feq = 0x80,
93 midgard_alu_op_fne = 0x81,
94 midgard_alu_op_flt = 0x82,
95 midgard_alu_op_fle = 0x83,
96 midgard_alu_op_fball_eq = 0x88,
97 midgard_alu_op_bball_eq = 0x89,
98 midgard_alu_op_fball_lt = 0x8A, /* all(lessThan(.., ..)) */
99 midgard_alu_op_fball_lte = 0x8B, /* all(lessThanEqual(.., ..)) */
100 midgard_alu_op_bbany_neq = 0x90, /* used for bvec4(1) */
101 midgard_alu_op_fbany_neq = 0x91, /* bvec4(0) also */
102 midgard_alu_op_fbany_lt = 0x92, /* any(lessThan(.., ..)) */
103 midgard_alu_op_fbany_lte = 0x93, /* any(lessThanEqual(.., ..)) */
104 midgard_alu_op_f2i = 0x99,
105 midgard_alu_op_f2u8 = 0x9C,
106 midgard_alu_op_f2u = 0x9D,
107
108 midgard_alu_op_ieq = 0xA0,
109 midgard_alu_op_ine = 0xA1,
110 midgard_alu_op_ult = 0xA2,
111 midgard_alu_op_ule = 0xA3,
112 midgard_alu_op_ilt = 0xA4,
113 midgard_alu_op_ile = 0xA5,
114 midgard_alu_op_iball_eq = 0xA8,
115 midgard_alu_op_ball = 0xA9,
116 midgard_alu_op_uball_lt = 0xAA,
117 midgard_alu_op_uball_lte = 0xAB,
118 midgard_alu_op_iball_lt = 0xAC,
119 midgard_alu_op_iball_lte = 0xAD,
120 midgard_alu_op_ibany_eq = 0xB0,
121 midgard_alu_op_ibany_neq = 0xB1,
122 midgard_alu_op_ubany_lt = 0xB2,
123 midgard_alu_op_ubany_lte = 0xB3,
124 midgard_alu_op_ibany_lt = 0xB4, /* any(lessThan(.., ..)) */
125 midgard_alu_op_ibany_lte = 0xB5, /* any(lessThanEqual(.., ..)) */
126 midgard_alu_op_i2f = 0xB8,
127 midgard_alu_op_u2f = 0xBC,
128 midgard_alu_op_icsel = 0xC1,
129 midgard_alu_op_fcsel_i = 0xC4,
130 midgard_alu_op_fcsel = 0xC5,
131 midgard_alu_op_fround = 0xC6,
132 midgard_alu_op_fatan_pt2 = 0xE8,
133 midgard_alu_op_fpow_pt1 = 0xEC,
134 midgard_alu_op_frcp = 0xF0,
135 midgard_alu_op_frsqrt = 0xF2,
136 midgard_alu_op_fsqrt = 0xF3,
137 midgard_alu_op_fexp2 = 0xF4,
138 midgard_alu_op_flog2 = 0xF5,
139 midgard_alu_op_fsin = 0xF6,
140 midgard_alu_op_fcos = 0xF7,
141 midgard_alu_op_fatan2_pt1 = 0xF9,
142 } midgard_alu_op;
143
144 typedef enum {
145 midgard_outmod_none = 0,
146 midgard_outmod_pos = 1,
147 midgard_outmod_int = 2,
148 midgard_outmod_sat = 3
149 } midgard_outmod;
150
151 typedef enum {
152 midgard_reg_mode_quarter = 0,
153 midgard_reg_mode_half = 1,
154 midgard_reg_mode_full = 2,
155 midgard_reg_mode_double = 3 /* TODO: verify */
156 } midgard_reg_mode;
157
158 typedef enum {
159 midgard_dest_override_lower = 0,
160 midgard_dest_override_upper = 1,
161 midgard_dest_override_none = 2
162 } midgard_dest_override;
163
164 typedef struct
165 __attribute__((__packed__))
166 {
167 bool abs : 1;
168 bool negate : 1;
169
170 /* replicate lower half if dest = half, or low/high half selection if
171 * dest = full
172 */
173 bool rep_low : 1;
174 bool rep_high : 1; /* unused if dest = full */
175 bool half : 1; /* only matters if dest = full */
176 unsigned swizzle : 8;
177 }
178 midgard_vector_alu_src;
179
180 typedef struct
181 __attribute__((__packed__))
182 {
183 midgard_alu_op op : 8;
184 midgard_reg_mode reg_mode : 2;
185 unsigned src1 : 13;
186 unsigned src2 : 13;
187 midgard_dest_override dest_override : 2;
188 midgard_outmod outmod : 2;
189 unsigned mask : 8;
190 }
191 midgard_vector_alu;
192
193 typedef struct
194 __attribute__((__packed__))
195 {
196 bool abs : 1;
197 bool negate : 1;
198 bool full : 1; /* 0 = half, 1 = full */
199 unsigned component : 3;
200 }
201 midgard_scalar_alu_src;
202
203 typedef struct
204 __attribute__((__packed__))
205 {
206 midgard_alu_op op : 8;
207 unsigned src1 : 6;
208 unsigned src2 : 11;
209 unsigned unknown : 1;
210 midgard_outmod outmod : 2;
211 bool output_full : 1;
212 unsigned output_component : 3;
213 }
214 midgard_scalar_alu;
215
216 typedef struct
217 __attribute__((__packed__))
218 {
219 unsigned src1_reg : 5;
220 unsigned src2_reg : 5;
221 unsigned out_reg : 5;
222 bool src2_imm : 1;
223 }
224 midgard_reg_info;
225
226 /* In addition to conditional branches and jumps (unconditional branches),
227 * Midgard implements a bit of fixed function functionality used in fragment
228 * shaders via specially crafted branches. These have special branch opcodes,
229 * which perform a fixed-function operation and/or use the results of a
230 * fixed-function operation as the branch condition. */
231
232 typedef enum {
233 /* Regular branches */
234 midgard_jmp_writeout_op_branch_uncond = 1,
235 midgard_jmp_writeout_op_branch_cond = 2,
236
237 /* In a fragment shader, execute a discard_if instruction, with the
238 * corresponding condition code. Terminates the shader, so generally
239 * set the branch target to out of the shader */
240 midgard_jmp_writeout_op_discard = 4,
241
242 /* Branch if the tilebuffer is not yet ready. At the beginning of a
243 * fragment shader that reads from the tile buffer, for instance via
244 * ARM_shader_framebuffer_fetch or EXT_pixel_local_storage, this branch
245 * operation should be used as a loop. An instruction like
246 * "br.tilebuffer.always -1" does the trick, corresponding to
247 * "while(!is_tilebuffer_ready) */
248 midgard_jmp_writeout_op_tilebuffer_pending = 6,
249
250 /* In a fragment shader, try to write out the value pushed to r0 to the
251 * tilebuffer, subject to unknown state in r1.z and r1.w. If this
252 * succeeds, the shader terminates. If it fails, it branches to the
253 * specified branch target. Generally, this should be used in a loop to
254 * itself, acting as "do { write(r0); } while(!write_successful);" */
255 midgard_jmp_writeout_op_writeout = 7,
256 } midgard_jmp_writeout_op;
257
258 typedef enum {
259 midgard_condition_write0 = 0,
260
261 /* These condition codes denote a conditional branch on FALSE and on
262 * TRUE respectively */
263 midgard_condition_false = 1,
264 midgard_condition_true = 2,
265
266 /* This condition code always branches. For a pure branch, the
267 * unconditional branch coding should be used instead, but for
268 * fixed-function branch opcodes, this is still useful */
269 midgard_condition_always = 3,
270 } midgard_condition;
271
272 typedef struct
273 __attribute__((__packed__))
274 {
275 midgard_jmp_writeout_op op : 3; /* == branch_uncond */
276 unsigned dest_tag : 4; /* tag of branch destination */
277 unsigned unknown : 2;
278 int offset : 7;
279 }
280 midgard_branch_uncond;
281
282 typedef struct
283 __attribute__((__packed__))
284 {
285 midgard_jmp_writeout_op op : 3; /* == branch_cond */
286 unsigned dest_tag : 4; /* tag of branch destination */
287 int offset : 7;
288 midgard_condition cond : 2;
289 }
290 midgard_branch_cond;
291
292 typedef struct
293 __attribute__((__packed__))
294 {
295 midgard_jmp_writeout_op op : 3; /* == branch_cond */
296 unsigned dest_tag : 4; /* tag of branch destination */
297 unsigned unknown : 2;
298 signed offset : 23;
299 unsigned cond : 16;
300 }
301 midgard_branch_extended;
302
303 typedef struct
304 __attribute__((__packed__))
305 {
306 midgard_jmp_writeout_op op : 3; /* == writeout */
307 unsigned unknown : 13;
308 }
309 midgard_writeout;
310
311 /*
312 * Load/store words
313 */
314
315 typedef enum {
316 midgard_op_ld_st_noop = 0x03,
317
318 /* Unclear why this is on the L/S unit, but (with an address of 0,
319 * appropriate swizzle, magic constant 0x24, and xy mask?) moves fp32 cube
320 * map coordinates in r27 to its cube map texture coordinate
321 * destination (e.g r29). 0x4 magic for loading from fp16 instead */
322
323 midgard_op_store_cubemap_coords = 0x0E,
324
325 midgard_op_load_attr_16 = 0x95,
326 midgard_op_load_attr_32 = 0x94,
327 midgard_op_load_vary_16 = 0x99,
328 midgard_op_load_vary_32 = 0x98,
329 midgard_op_load_color_buffer_16 = 0x9D,
330 midgard_op_load_color_buffer_8 = 0xBA,
331 midgard_op_load_uniform_16 = 0xAC,
332 midgard_op_load_uniform_32 = 0xB0,
333 midgard_op_store_vary_16 = 0xD5,
334 midgard_op_store_vary_32 = 0xD4
335 } midgard_load_store_op;
336
337 typedef enum {
338 midgard_interp_centroid = 1,
339 midgard_interp_default = 2
340 } midgard_interpolation;
341
342 typedef struct
343 __attribute__((__packed__))
344 {
345 unsigned zero1 : 4; /* Always zero */
346
347 /* Varying qualifiers, zero if not a varying */
348 unsigned flat : 1;
349 unsigned is_varying : 1; /* Always one for varying, but maybe something else? */
350 midgard_interpolation interpolation : 2;
351
352 unsigned zero2 : 2; /* Always zero */
353 }
354 midgard_varying_parameter;
355
356 typedef struct
357 __attribute__((__packed__))
358 {
359 midgard_load_store_op op : 8;
360 unsigned reg : 5;
361 unsigned mask : 4;
362 unsigned swizzle : 8;
363 unsigned unknown : 16;
364
365 unsigned varying_parameters : 10;
366
367 unsigned address : 9;
368 }
369 midgard_load_store_word;
370
371 typedef struct
372 __attribute__((__packed__))
373 {
374 unsigned type : 4;
375 unsigned next_type : 4;
376 uint64_t word1 : 60;
377 uint64_t word2 : 60;
378 }
379 midgard_load_store;
380
381 /* Texture pipeline results are in r28-r29 */
382 #define REG_TEX_BASE 28
383
384 /* Texture opcodes... maybe? */
385 #define TEXTURE_OP_NORMAL 0x11
386 #define TEXTURE_OP_TEXEL_FETCH 0x14
387
388 /* Texture format types, found in format */
389 #define TEXTURE_CUBE 0x00
390 #define TEXTURE_2D 0x02
391 #define TEXTURE_3D 0x03
392
393 typedef struct
394 __attribute__((__packed__))
395 {
396 unsigned type : 4;
397 unsigned next_type : 4;
398
399 unsigned op : 6;
400 unsigned shadow : 1;
401 unsigned unknown3 : 1;
402
403 /* A little obscure, but last is set for the last texture operation in
404 * a shader. cont appears to just be last's opposite (?). Yeah, I know,
405 * kind of funky.. BiOpen thinks it could do with memory hinting, or
406 * tile locking? */
407
408 unsigned cont : 1;
409 unsigned last : 1;
410
411 unsigned format : 5;
412 unsigned has_offset : 1;
413
414 /* Like in Bifrost */
415 unsigned filter : 1;
416
417 unsigned in_reg_select : 1;
418 unsigned in_reg_upper : 1;
419
420 unsigned in_reg_swizzle_left : 2;
421 unsigned in_reg_swizzle_right : 2;
422
423 unsigned unknown1 : 2;
424
425 unsigned unknown8 : 4;
426
427 unsigned out_full : 1;
428
429 /* Always 1 afaict... */
430 unsigned unknown7 : 2;
431
432 unsigned out_reg_select : 1;
433 unsigned out_upper : 1;
434
435 unsigned mask : 4;
436
437 unsigned unknown2 : 2;
438
439 unsigned swizzle : 8;
440 unsigned unknown4 : 8;
441
442 unsigned unknownA : 4;
443
444 unsigned offset_unknown1 : 1;
445 unsigned offset_reg_select : 1;
446 unsigned offset_reg_upper : 1;
447 unsigned offset_unknown4 : 1;
448 unsigned offset_unknown5 : 1;
449 unsigned offset_unknown6 : 1;
450 unsigned offset_unknown7 : 1;
451 unsigned offset_unknown8 : 1;
452 unsigned offset_unknown9 : 1;
453
454 unsigned unknownB : 3;
455
456 /* Texture bias or LOD, depending on whether it is executed in a
457 * fragment/vertex shader respectively. Compute as int(2^8 * biasf).
458 *
459 * For texel fetch, this is the LOD as is. */
460 unsigned bias : 8;
461
462 unsigned unknown9 : 8;
463
464 unsigned texture_handle : 16;
465 unsigned sampler_handle : 16;
466 }
467 midgard_texture_word;
468
469 /* Opcode name table */
470
471 static char *alu_opcode_names[256] = {
472 [midgard_alu_op_fadd] = "fadd",
473 [midgard_alu_op_fmul] = "fmul",
474 [midgard_alu_op_fmin] = "fmin",
475 [midgard_alu_op_fmax] = "fmax",
476 [midgard_alu_op_fmov] = "fmov",
477 [midgard_alu_op_froundeven] = "froundeven",
478 [midgard_alu_op_ftrunc] = "ftrunc",
479 [midgard_alu_op_ffloor] = "ffloor",
480 [midgard_alu_op_fceil] = "fceil",
481 [midgard_alu_op_ffma] = "ffma",
482 [midgard_alu_op_fdot3] = "fdot3",
483 [midgard_alu_op_fdot3r] = "fdot3r",
484 [midgard_alu_op_fdot4] = "fdot4",
485 [midgard_alu_op_freduce] = "freduce",
486 [midgard_alu_op_imin] = "imin",
487 [midgard_alu_op_umin] = "umin",
488 [midgard_alu_op_imax] = "imax",
489 [midgard_alu_op_umax] = "umax",
490 [midgard_alu_op_ishl] = "ishl",
491 [midgard_alu_op_iasr] = "iasr",
492 [midgard_alu_op_ilsr] = "ilsr",
493 [midgard_alu_op_iadd] = "iadd",
494 [midgard_alu_op_ishladd] = "ishladd",
495 [midgard_alu_op_isub] = "isub",
496 [midgard_alu_op_imul] = "imul",
497 [midgard_alu_op_imov] = "imov",
498 [midgard_alu_op_iabs] = "iabs",
499 [midgard_alu_op_iand] = "iand",
500 [midgard_alu_op_ior] = "ior",
501 [midgard_alu_op_inot] = "inot",
502 [midgard_alu_op_iandnot] = "iandnot",
503 [midgard_alu_op_ixor] = "ixor",
504 [midgard_alu_op_ilzcnt] = "ilzcnt",
505 [midgard_alu_op_ibitcount8] = "ibitcount8",
506 [midgard_alu_op_feq] = "feq",
507 [midgard_alu_op_fne] = "fne",
508 [midgard_alu_op_flt] = "flt",
509 [midgard_alu_op_fle] = "fle",
510 [midgard_alu_op_fball_eq] = "fball_eq",
511 [midgard_alu_op_fbany_neq] = "fbany_neq",
512 [midgard_alu_op_bball_eq] = "bball_eq",
513 [midgard_alu_op_fball_lt] = "fball_lt",
514 [midgard_alu_op_fball_lte] = "fball_lte",
515 [midgard_alu_op_bbany_neq] = "bbany_neq",
516 [midgard_alu_op_fbany_lt] = "fbany_lt",
517 [midgard_alu_op_fbany_lte] = "fbany_lte",
518 [midgard_alu_op_f2i] = "f2i",
519 [midgard_alu_op_f2u] = "f2u",
520 [midgard_alu_op_f2u8] = "f2u8",
521 [midgard_alu_op_ieq] = "ieq",
522 [midgard_alu_op_ine] = "ine",
523 [midgard_alu_op_ult] = "ult",
524 [midgard_alu_op_ule] = "ule",
525 [midgard_alu_op_ilt] = "ilt",
526 [midgard_alu_op_ile] = "ile",
527 [midgard_alu_op_iball_eq] = "iball_eq",
528 [midgard_alu_op_ball] = "ball",
529 [midgard_alu_op_uball_lt] = "uball_lt",
530 [midgard_alu_op_uball_lte] = "uball_lte",
531 [midgard_alu_op_iball_lt] = "iball_lt",
532 [midgard_alu_op_iball_lte] = "iball_lte",
533 [midgard_alu_op_iball_eq] = "iball_eq",
534 [midgard_alu_op_ibany_neq] = "ibany_neq",
535 [midgard_alu_op_ubany_lt] = "ubany_lt",
536 [midgard_alu_op_ubany_lte] = "ubany_lte",
537 [midgard_alu_op_ibany_lt] = "ibany_lt",
538 [midgard_alu_op_ibany_lte] = "ibany_lte",
539 [midgard_alu_op_i2f] = "i2f",
540 [midgard_alu_op_u2f] = "u2f",
541 [midgard_alu_op_icsel] = "icsel",
542 [midgard_alu_op_fcsel_i] = "fcsel_i",
543 [midgard_alu_op_fcsel] = "fcsel",
544 [midgard_alu_op_fround] = "fround",
545 [midgard_alu_op_fatan_pt2] = "fatan_pt2",
546 [midgard_alu_op_frcp] = "frcp",
547 [midgard_alu_op_frsqrt] = "frsqrt",
548 [midgard_alu_op_fsqrt] = "fsqrt",
549 [midgard_alu_op_fpow_pt1] = "fpow_pt1",
550 [midgard_alu_op_fexp2] = "fexp2",
551 [midgard_alu_op_flog2] = "flog2",
552 [midgard_alu_op_fsin] = "fsin",
553 [midgard_alu_op_fcos] = "fcos",
554 [midgard_alu_op_fatan2_pt1] = "fatan2_pt1"
555 };
556
557 static char *load_store_opcode_names[256] = {
558 [midgard_op_store_cubemap_coords] = "st_cubemap_coords",
559 [midgard_op_load_attr_16] = "ld_attr_16",
560 [midgard_op_load_attr_32] = "ld_attr_32",
561 [midgard_op_load_vary_16] = "ld_vary_16",
562 [midgard_op_load_vary_32] = "ld_vary_32",
563 [midgard_op_load_uniform_16] = "ld_uniform_16",
564 [midgard_op_load_uniform_32] = "ld_uniform_32",
565 [midgard_op_load_color_buffer_8] = "ld_color_buffer_8",
566 [midgard_op_load_color_buffer_16] = "ld_color_buffer_16",
567 [midgard_op_store_vary_16] = "st_vary_16",
568 [midgard_op_store_vary_32] = "st_vary_32"
569 };
570
571 #endif