intel/fs: Implement nir_intrinsic_global_atomic_*
[mesa.git] / src / intel / compiler / brw_disasm.c
1 /*
2 * Copyright © 2008 Keith Packard
3 *
4 * Permission to use, copy, modify, distribute, and sell this software and its
5 * documentation for any purpose is hereby granted without fee, provided that
6 * the above copyright notice appear in all copies and that both that copyright
7 * notice and this permission notice appear in supporting documentation, and
8 * that the name of the copyright holders not be used in advertising or
9 * publicity pertaining to distribution of the software without specific,
10 * written prior permission. The copyright holders make no representations
11 * about the suitability of this software for any purpose. It is provided "as
12 * is" without express or implied warranty.
13 *
14 * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
15 * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO
16 * EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY SPECIAL, INDIRECT OR
17 * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
18 * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
19 * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
20 * OF THIS SOFTWARE.
21 */
22
23 #include <stdio.h>
24 #include <string.h>
25 #include <stdarg.h>
26
27 #include "brw_eu_defines.h"
28 #include "brw_inst.h"
29 #include "brw_shader.h"
30 #include "brw_reg.h"
31 #include "brw_inst.h"
32 #include "brw_eu.h"
33 #include "util/half_float.h"
34
35 static bool
36 has_jip(const struct gen_device_info *devinfo, enum opcode opcode)
37 {
38 if (devinfo->gen < 6)
39 return false;
40
41 return opcode == BRW_OPCODE_IF ||
42 opcode == BRW_OPCODE_ELSE ||
43 opcode == BRW_OPCODE_ENDIF ||
44 opcode == BRW_OPCODE_WHILE ||
45 opcode == BRW_OPCODE_BREAK ||
46 opcode == BRW_OPCODE_CONTINUE ||
47 opcode == BRW_OPCODE_HALT;
48 }
49
50 static bool
51 has_uip(const struct gen_device_info *devinfo, enum opcode opcode)
52 {
53 if (devinfo->gen < 6)
54 return false;
55
56 return (devinfo->gen >= 7 && opcode == BRW_OPCODE_IF) ||
57 (devinfo->gen >= 8 && opcode == BRW_OPCODE_ELSE) ||
58 opcode == BRW_OPCODE_BREAK ||
59 opcode == BRW_OPCODE_CONTINUE ||
60 opcode == BRW_OPCODE_HALT;
61 }
62
63 static bool
64 has_branch_ctrl(const struct gen_device_info *devinfo, enum opcode opcode)
65 {
66 if (devinfo->gen < 8)
67 return false;
68
69 return opcode == BRW_OPCODE_IF ||
70 opcode == BRW_OPCODE_ELSE;
71 /* opcode == BRW_OPCODE_GOTO; */
72 }
73
74 static bool
75 is_logic_instruction(unsigned opcode)
76 {
77 return opcode == BRW_OPCODE_AND ||
78 opcode == BRW_OPCODE_NOT ||
79 opcode == BRW_OPCODE_OR ||
80 opcode == BRW_OPCODE_XOR;
81 }
82
83 static bool
84 is_send(unsigned opcode)
85 {
86 return opcode == BRW_OPCODE_SEND ||
87 opcode == BRW_OPCODE_SENDC ||
88 opcode == BRW_OPCODE_SENDS ||
89 opcode == BRW_OPCODE_SENDSC;
90 }
91
92 static bool
93 is_split_send(UNUSED const struct gen_device_info *devinfo, unsigned opcode)
94 {
95 return opcode == BRW_OPCODE_SENDS ||
96 opcode == BRW_OPCODE_SENDSC;
97 }
98
99 const char *const conditional_modifier[16] = {
100 [BRW_CONDITIONAL_NONE] = "",
101 [BRW_CONDITIONAL_Z] = ".z",
102 [BRW_CONDITIONAL_NZ] = ".nz",
103 [BRW_CONDITIONAL_G] = ".g",
104 [BRW_CONDITIONAL_GE] = ".ge",
105 [BRW_CONDITIONAL_L] = ".l",
106 [BRW_CONDITIONAL_LE] = ".le",
107 [BRW_CONDITIONAL_R] = ".r",
108 [BRW_CONDITIONAL_O] = ".o",
109 [BRW_CONDITIONAL_U] = ".u",
110 };
111
112 static const char *const m_negate[2] = {
113 [0] = "",
114 [1] = "-",
115 };
116
117 static const char *const _abs[2] = {
118 [0] = "",
119 [1] = "(abs)",
120 };
121
122 static const char *const m_bitnot[2] = { "", "~" };
123
124 static const char *const vert_stride[16] = {
125 [0] = "0",
126 [1] = "1",
127 [2] = "2",
128 [3] = "4",
129 [4] = "8",
130 [5] = "16",
131 [6] = "32",
132 [15] = "VxH",
133 };
134
135 static const char *const width[8] = {
136 [0] = "1",
137 [1] = "2",
138 [2] = "4",
139 [3] = "8",
140 [4] = "16",
141 };
142
143 static const char *const horiz_stride[4] = {
144 [0] = "0",
145 [1] = "1",
146 [2] = "2",
147 [3] = "4"
148 };
149
150 static const char *const chan_sel[4] = {
151 [0] = "x",
152 [1] = "y",
153 [2] = "z",
154 [3] = "w",
155 };
156
157 static const char *const debug_ctrl[2] = {
158 [0] = "",
159 [1] = ".breakpoint"
160 };
161
162 static const char *const saturate[2] = {
163 [0] = "",
164 [1] = ".sat"
165 };
166
167 static const char *const cmpt_ctrl[2] = {
168 [0] = "",
169 [1] = "compacted"
170 };
171
172 static const char *const accwr[2] = {
173 [0] = "",
174 [1] = "AccWrEnable"
175 };
176
177 static const char *const branch_ctrl[2] = {
178 [0] = "",
179 [1] = "BranchCtrl"
180 };
181
182 static const char *const wectrl[2] = {
183 [0] = "",
184 [1] = "WE_all"
185 };
186
187 static const char *const exec_size[8] = {
188 [0] = "1",
189 [1] = "2",
190 [2] = "4",
191 [3] = "8",
192 [4] = "16",
193 [5] = "32"
194 };
195
196 static const char *const pred_inv[2] = {
197 [0] = "+",
198 [1] = "-"
199 };
200
201 const char *const pred_ctrl_align16[16] = {
202 [1] = "",
203 [2] = ".x",
204 [3] = ".y",
205 [4] = ".z",
206 [5] = ".w",
207 [6] = ".any4h",
208 [7] = ".all4h",
209 };
210
211 static const char *const pred_ctrl_align1[16] = {
212 [BRW_PREDICATE_NORMAL] = "",
213 [BRW_PREDICATE_ALIGN1_ANYV] = ".anyv",
214 [BRW_PREDICATE_ALIGN1_ALLV] = ".allv",
215 [BRW_PREDICATE_ALIGN1_ANY2H] = ".any2h",
216 [BRW_PREDICATE_ALIGN1_ALL2H] = ".all2h",
217 [BRW_PREDICATE_ALIGN1_ANY4H] = ".any4h",
218 [BRW_PREDICATE_ALIGN1_ALL4H] = ".all4h",
219 [BRW_PREDICATE_ALIGN1_ANY8H] = ".any8h",
220 [BRW_PREDICATE_ALIGN1_ALL8H] = ".all8h",
221 [BRW_PREDICATE_ALIGN1_ANY16H] = ".any16h",
222 [BRW_PREDICATE_ALIGN1_ALL16H] = ".all16h",
223 [BRW_PREDICATE_ALIGN1_ANY32H] = ".any32h",
224 [BRW_PREDICATE_ALIGN1_ALL32H] = ".all32h",
225 };
226
227 static const char *const thread_ctrl[4] = {
228 [BRW_THREAD_NORMAL] = "",
229 [BRW_THREAD_ATOMIC] = "atomic",
230 [BRW_THREAD_SWITCH] = "switch",
231 };
232
233 static const char *const compr_ctrl[4] = {
234 [0] = "",
235 [1] = "sechalf",
236 [2] = "compr",
237 [3] = "compr4",
238 };
239
240 static const char *const dep_ctrl[4] = {
241 [0] = "",
242 [1] = "NoDDClr",
243 [2] = "NoDDChk",
244 [3] = "NoDDClr,NoDDChk",
245 };
246
247 static const char *const mask_ctrl[4] = {
248 [0] = "",
249 [1] = "nomask",
250 };
251
252 static const char *const access_mode[2] = {
253 [0] = "align1",
254 [1] = "align16",
255 };
256
257 static const char *const reg_file[4] = {
258 [0] = "A",
259 [1] = "g",
260 [2] = "m",
261 [3] = "imm",
262 };
263
264 static const char *const writemask[16] = {
265 [0x0] = ".",
266 [0x1] = ".x",
267 [0x2] = ".y",
268 [0x3] = ".xy",
269 [0x4] = ".z",
270 [0x5] = ".xz",
271 [0x6] = ".yz",
272 [0x7] = ".xyz",
273 [0x8] = ".w",
274 [0x9] = ".xw",
275 [0xa] = ".yw",
276 [0xb] = ".xyw",
277 [0xc] = ".zw",
278 [0xd] = ".xzw",
279 [0xe] = ".yzw",
280 [0xf] = "",
281 };
282
283 static const char *const end_of_thread[2] = {
284 [0] = "",
285 [1] = "EOT"
286 };
287
288 /* SFIDs on Gen4-5 */
289 static const char *const gen4_sfid[16] = {
290 [BRW_SFID_NULL] = "null",
291 [BRW_SFID_MATH] = "math",
292 [BRW_SFID_SAMPLER] = "sampler",
293 [BRW_SFID_MESSAGE_GATEWAY] = "gateway",
294 [BRW_SFID_DATAPORT_READ] = "read",
295 [BRW_SFID_DATAPORT_WRITE] = "write",
296 [BRW_SFID_URB] = "urb",
297 [BRW_SFID_THREAD_SPAWNER] = "thread_spawner",
298 [BRW_SFID_VME] = "vme",
299 };
300
301 static const char *const gen6_sfid[16] = {
302 [BRW_SFID_NULL] = "null",
303 [BRW_SFID_MATH] = "math",
304 [BRW_SFID_SAMPLER] = "sampler",
305 [BRW_SFID_MESSAGE_GATEWAY] = "gateway",
306 [BRW_SFID_URB] = "urb",
307 [BRW_SFID_THREAD_SPAWNER] = "thread_spawner",
308 [GEN6_SFID_DATAPORT_SAMPLER_CACHE] = "dp_sampler",
309 [GEN6_SFID_DATAPORT_RENDER_CACHE] = "render",
310 [GEN6_SFID_DATAPORT_CONSTANT_CACHE] = "const",
311 [GEN7_SFID_DATAPORT_DATA_CACHE] = "data",
312 [GEN7_SFID_PIXEL_INTERPOLATOR] = "pixel interp",
313 [HSW_SFID_DATAPORT_DATA_CACHE_1] = "dp data 1",
314 [HSW_SFID_CRE] = "cre",
315 };
316
317 static const char *const gen7_gateway_subfuncid[8] = {
318 [BRW_MESSAGE_GATEWAY_SFID_OPEN_GATEWAY] = "open",
319 [BRW_MESSAGE_GATEWAY_SFID_CLOSE_GATEWAY] = "close",
320 [BRW_MESSAGE_GATEWAY_SFID_FORWARD_MSG] = "forward msg",
321 [BRW_MESSAGE_GATEWAY_SFID_GET_TIMESTAMP] = "get timestamp",
322 [BRW_MESSAGE_GATEWAY_SFID_BARRIER_MSG] = "barrier msg",
323 [BRW_MESSAGE_GATEWAY_SFID_UPDATE_GATEWAY_STATE] = "update state",
324 [BRW_MESSAGE_GATEWAY_SFID_MMIO_READ_WRITE] = "mmio read/write",
325 };
326
327 static const char *const gen4_dp_read_port_msg_type[4] = {
328 [0b00] = "OWord Block Read",
329 [0b01] = "OWord Dual Block Read",
330 [0b10] = "Media Block Read",
331 [0b11] = "DWord Scattered Read",
332 };
333
334 static const char *const g45_dp_read_port_msg_type[8] = {
335 [0b000] = "OWord Block Read",
336 [0b010] = "OWord Dual Block Read",
337 [0b100] = "Media Block Read",
338 [0b110] = "DWord Scattered Read",
339 [0b001] = "Render Target UNORM Read",
340 [0b011] = "AVC Loop Filter Read",
341 };
342
343 static const char *const dp_write_port_msg_type[8] = {
344 [0b000] = "OWord block write",
345 [0b001] = "OWord dual block write",
346 [0b010] = "media block write",
347 [0b011] = "DWord scattered write",
348 [0b100] = "RT write",
349 [0b101] = "streamed VB write",
350 [0b110] = "RT UNORM write", /* G45+ */
351 [0b111] = "flush render cache",
352 };
353
354 static const char *const dp_rc_msg_type_gen6[16] = {
355 [BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ] = "OWORD block read",
356 [GEN6_DATAPORT_READ_MESSAGE_RENDER_UNORM_READ] = "RT UNORM read",
357 [GEN6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ] = "OWORD dual block read",
358 [GEN6_DATAPORT_READ_MESSAGE_MEDIA_BLOCK_READ] = "media block read",
359 [GEN6_DATAPORT_READ_MESSAGE_OWORD_UNALIGN_BLOCK_READ] =
360 "OWORD unaligned block read",
361 [GEN6_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ] = "DWORD scattered read",
362 [GEN6_DATAPORT_WRITE_MESSAGE_DWORD_ATOMIC_WRITE] = "DWORD atomic write",
363 [GEN6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE] = "OWORD block write",
364 [GEN6_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE] =
365 "OWORD dual block write",
366 [GEN6_DATAPORT_WRITE_MESSAGE_MEDIA_BLOCK_WRITE] = "media block write",
367 [GEN6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE] =
368 "DWORD scattered write",
369 [GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE] = "RT write",
370 [GEN6_DATAPORT_WRITE_MESSAGE_STREAMED_VB_WRITE] = "streamed VB write",
371 [GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_UNORM_WRITE] = "RT UNORM write",
372 };
373
374 static const char *const dp_rc_msg_type_gen7[16] = {
375 [GEN7_DATAPORT_RC_MEDIA_BLOCK_READ] = "media block read",
376 [GEN7_DATAPORT_RC_TYPED_SURFACE_READ] = "typed surface read",
377 [GEN7_DATAPORT_RC_TYPED_ATOMIC_OP] = "typed atomic op",
378 [GEN7_DATAPORT_RC_MEMORY_FENCE] = "memory fence",
379 [GEN7_DATAPORT_RC_MEDIA_BLOCK_WRITE] = "media block write",
380 [GEN7_DATAPORT_RC_RENDER_TARGET_WRITE] = "RT write",
381 [GEN7_DATAPORT_RC_TYPED_SURFACE_WRITE] = "typed surface write"
382 };
383
384 static const char *const dp_rc_msg_type_gen9[16] = {
385 [GEN9_DATAPORT_RC_RENDER_TARGET_WRITE] = "RT write",
386 [GEN9_DATAPORT_RC_RENDER_TARGET_READ] = "RT read"
387 };
388
389 static const char *const *
390 dp_rc_msg_type(const struct gen_device_info *devinfo)
391 {
392 return (devinfo->gen >= 9 ? dp_rc_msg_type_gen9 :
393 devinfo->gen >= 7 ? dp_rc_msg_type_gen7 :
394 devinfo->gen >= 6 ? dp_rc_msg_type_gen6 :
395 dp_write_port_msg_type);
396 }
397
398 static const char *const m_rt_write_subtype[] = {
399 [0b000] = "SIMD16",
400 [0b001] = "SIMD16/RepData",
401 [0b010] = "SIMD8/DualSrcLow",
402 [0b011] = "SIMD8/DualSrcHigh",
403 [0b100] = "SIMD8",
404 [0b101] = "SIMD8/ImageWrite", /* Gen6+ */
405 [0b111] = "SIMD16/RepData-111", /* no idea how this is different than 1 */
406 };
407
408 static const char *const dp_dc0_msg_type_gen7[16] = {
409 [GEN7_DATAPORT_DC_OWORD_BLOCK_READ] = "DC OWORD block read",
410 [GEN7_DATAPORT_DC_UNALIGNED_OWORD_BLOCK_READ] =
411 "DC unaligned OWORD block read",
412 [GEN7_DATAPORT_DC_OWORD_DUAL_BLOCK_READ] = "DC OWORD dual block read",
413 [GEN7_DATAPORT_DC_DWORD_SCATTERED_READ] = "DC DWORD scattered read",
414 [GEN7_DATAPORT_DC_BYTE_SCATTERED_READ] = "DC byte scattered read",
415 [GEN7_DATAPORT_DC_UNTYPED_SURFACE_READ] = "DC untyped surface read",
416 [GEN7_DATAPORT_DC_UNTYPED_ATOMIC_OP] = "DC untyped atomic",
417 [GEN7_DATAPORT_DC_MEMORY_FENCE] = "DC mfence",
418 [GEN7_DATAPORT_DC_OWORD_BLOCK_WRITE] = "DC OWORD block write",
419 [GEN7_DATAPORT_DC_OWORD_DUAL_BLOCK_WRITE] = "DC OWORD dual block write",
420 [GEN7_DATAPORT_DC_DWORD_SCATTERED_WRITE] = "DC DWORD scatterd write",
421 [GEN7_DATAPORT_DC_BYTE_SCATTERED_WRITE] = "DC byte scattered write",
422 [GEN7_DATAPORT_DC_UNTYPED_SURFACE_WRITE] = "DC untyped surface write",
423 };
424
425 static const char *const dp_dc1_msg_type_hsw[32] = {
426 [HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_READ] = "untyped surface read",
427 [HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP] = "DC untyped atomic op",
428 [HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP_SIMD4X2] =
429 "DC untyped 4x2 atomic op",
430 [HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_READ] = "DC media block read",
431 [HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ] = "DC typed surface read",
432 [HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP] = "DC typed atomic",
433 [HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2] = "DC typed 4x2 atomic op",
434 [HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_WRITE] = "DC untyped surface write",
435 [HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_WRITE] = "DC media block write",
436 [HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP] = "DC atomic counter op",
437 [HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP_SIMD4X2] =
438 "DC 4x2 atomic counter op",
439 [HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE] = "DC typed surface write",
440 [GEN9_DATAPORT_DC_PORT1_A64_SCATTERED_READ] = "DC A64 scattered read",
441 [GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_READ] = "DC A64 untyped surface read",
442 [GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_OP] = "DC A64 untyped atomic op",
443 [GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_WRITE] = "DC A64 untyped surface write",
444 [GEN8_DATAPORT_DC_PORT1_A64_SCATTERED_WRITE] = "DC A64 scattered write",
445 [GEN9_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_FLOAT_OP] =
446 "DC untyped atomic float op",
447 [GEN9_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_FLOAT_OP] =
448 "DC A64 untyped atomic float op",
449 };
450
451 static const char *const aop[16] = {
452 [BRW_AOP_AND] = "and",
453 [BRW_AOP_OR] = "or",
454 [BRW_AOP_XOR] = "xor",
455 [BRW_AOP_MOV] = "mov",
456 [BRW_AOP_INC] = "inc",
457 [BRW_AOP_DEC] = "dec",
458 [BRW_AOP_ADD] = "add",
459 [BRW_AOP_SUB] = "sub",
460 [BRW_AOP_REVSUB] = "revsub",
461 [BRW_AOP_IMAX] = "imax",
462 [BRW_AOP_IMIN] = "imin",
463 [BRW_AOP_UMAX] = "umax",
464 [BRW_AOP_UMIN] = "umin",
465 [BRW_AOP_CMPWR] = "cmpwr",
466 [BRW_AOP_PREDEC] = "predec",
467 };
468
469 static const char *const aop_float[4] = {
470 [BRW_AOP_FMAX] = "fmax",
471 [BRW_AOP_FMIN] = "fmin",
472 [BRW_AOP_FCMPWR] = "fcmpwr",
473 };
474
475 static const char * const pixel_interpolator_msg_types[4] = {
476 [GEN7_PIXEL_INTERPOLATOR_LOC_SHARED_OFFSET] = "per_message_offset",
477 [GEN7_PIXEL_INTERPOLATOR_LOC_SAMPLE] = "sample_position",
478 [GEN7_PIXEL_INTERPOLATOR_LOC_CENTROID] = "centroid",
479 [GEN7_PIXEL_INTERPOLATOR_LOC_PER_SLOT_OFFSET] = "per_slot_offset",
480 };
481
482 static const char *const math_function[16] = {
483 [BRW_MATH_FUNCTION_INV] = "inv",
484 [BRW_MATH_FUNCTION_LOG] = "log",
485 [BRW_MATH_FUNCTION_EXP] = "exp",
486 [BRW_MATH_FUNCTION_SQRT] = "sqrt",
487 [BRW_MATH_FUNCTION_RSQ] = "rsq",
488 [BRW_MATH_FUNCTION_SIN] = "sin",
489 [BRW_MATH_FUNCTION_COS] = "cos",
490 [BRW_MATH_FUNCTION_SINCOS] = "sincos",
491 [BRW_MATH_FUNCTION_FDIV] = "fdiv",
492 [BRW_MATH_FUNCTION_POW] = "pow",
493 [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER] = "intdivmod",
494 [BRW_MATH_FUNCTION_INT_DIV_QUOTIENT] = "intdiv",
495 [BRW_MATH_FUNCTION_INT_DIV_REMAINDER] = "intmod",
496 [GEN8_MATH_FUNCTION_INVM] = "invm",
497 [GEN8_MATH_FUNCTION_RSQRTM] = "rsqrtm",
498 };
499
500 static const char *const math_saturate[2] = {
501 [0] = "",
502 [1] = "sat"
503 };
504
505 static const char *const math_signed[2] = {
506 [0] = "",
507 [1] = "signed"
508 };
509
510 static const char *const math_scalar[2] = {
511 [0] = "",
512 [1] = "scalar"
513 };
514
515 static const char *const math_precision[2] = {
516 [0] = "",
517 [1] = "partial_precision"
518 };
519
520 static const char *const gen5_urb_opcode[] = {
521 [0] = "urb_write",
522 [1] = "ff_sync",
523 };
524
525 static const char *const gen7_urb_opcode[] = {
526 [BRW_URB_OPCODE_WRITE_HWORD] = "write HWord",
527 [BRW_URB_OPCODE_WRITE_OWORD] = "write OWord",
528 [BRW_URB_OPCODE_READ_HWORD] = "read HWord",
529 [BRW_URB_OPCODE_READ_OWORD] = "read OWord",
530 [GEN7_URB_OPCODE_ATOMIC_MOV] = "atomic mov", /* Gen7+ */
531 [GEN7_URB_OPCODE_ATOMIC_INC] = "atomic inc", /* Gen7+ */
532 [GEN8_URB_OPCODE_ATOMIC_ADD] = "atomic add", /* Gen8+ */
533 [GEN8_URB_OPCODE_SIMD8_WRITE] = "SIMD8 write", /* Gen8+ */
534 [GEN8_URB_OPCODE_SIMD8_READ] = "SIMD8 read", /* Gen8+ */
535 /* [9-15] - reserved */
536 };
537
538 static const char *const urb_swizzle[4] = {
539 [BRW_URB_SWIZZLE_NONE] = "",
540 [BRW_URB_SWIZZLE_INTERLEAVE] = "interleave",
541 [BRW_URB_SWIZZLE_TRANSPOSE] = "transpose",
542 };
543
544 static const char *const urb_allocate[2] = {
545 [0] = "",
546 [1] = "allocate"
547 };
548
549 static const char *const urb_used[2] = {
550 [0] = "",
551 [1] = "used"
552 };
553
554 static const char *const urb_complete[2] = {
555 [0] = "",
556 [1] = "complete"
557 };
558
559 static const char *const gen5_sampler_msg_type[] = {
560 [GEN5_SAMPLER_MESSAGE_SAMPLE] = "sample",
561 [GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS] = "sample_b",
562 [GEN5_SAMPLER_MESSAGE_SAMPLE_LOD] = "sample_l",
563 [GEN5_SAMPLER_MESSAGE_SAMPLE_COMPARE] = "sample_c",
564 [GEN5_SAMPLER_MESSAGE_SAMPLE_DERIVS] = "sample_d",
565 [GEN5_SAMPLER_MESSAGE_SAMPLE_BIAS_COMPARE] = "sample_b_c",
566 [GEN5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE] = "sample_l_c",
567 [GEN5_SAMPLER_MESSAGE_SAMPLE_LD] = "ld",
568 [GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4] = "gather4",
569 [GEN5_SAMPLER_MESSAGE_LOD] = "lod",
570 [GEN5_SAMPLER_MESSAGE_SAMPLE_RESINFO] = "resinfo",
571 [GEN6_SAMPLER_MESSAGE_SAMPLE_SAMPLEINFO] = "sampleinfo",
572 [GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C] = "gather4_c",
573 [GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO] = "gather4_po",
574 [GEN7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C] = "gather4_po_c",
575 [HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE] = "sample_d_c",
576 [GEN9_SAMPLER_MESSAGE_SAMPLE_LZ] = "sample_lz",
577 [GEN9_SAMPLER_MESSAGE_SAMPLE_C_LZ] = "sample_c_lz",
578 [GEN9_SAMPLER_MESSAGE_SAMPLE_LD_LZ] = "ld_lz",
579 [GEN9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W] = "ld2dms_w",
580 [GEN7_SAMPLER_MESSAGE_SAMPLE_LD_MCS] = "ld_mcs",
581 [GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DMS] = "ld2dms",
582 [GEN7_SAMPLER_MESSAGE_SAMPLE_LD2DSS] = "ld2dss",
583 };
584
585 static const char *const gen5_sampler_simd_mode[4] = {
586 [BRW_SAMPLER_SIMD_MODE_SIMD4X2] = "SIMD4x2",
587 [BRW_SAMPLER_SIMD_MODE_SIMD8] = "SIMD8",
588 [BRW_SAMPLER_SIMD_MODE_SIMD16] = "SIMD16",
589 [BRW_SAMPLER_SIMD_MODE_SIMD32_64] = "SIMD32/64",
590 };
591
592 static const char *const sampler_target_format[4] = {
593 [0] = "F",
594 [2] = "UD",
595 [3] = "D"
596 };
597
598
599 static int column;
600
601 static int
602 string(FILE *file, const char *string)
603 {
604 fputs(string, file);
605 column += strlen(string);
606 return 0;
607 }
608
609 static int
610 format(FILE *f, const char *format, ...) PRINTFLIKE(2, 3);
611
612 static int
613 format(FILE *f, const char *format, ...)
614 {
615 char buf[1024];
616 va_list args;
617 va_start(args, format);
618
619 vsnprintf(buf, sizeof(buf) - 1, format, args);
620 va_end(args);
621 string(f, buf);
622 return 0;
623 }
624
625 static int
626 newline(FILE *f)
627 {
628 putc('\n', f);
629 column = 0;
630 return 0;
631 }
632
633 static int
634 pad(FILE *f, int c)
635 {
636 do
637 string(f, " ");
638 while (column < c);
639 return 0;
640 }
641
642 static int
643 control(FILE *file, const char *name, const char *const ctrl[],
644 unsigned id, int *space)
645 {
646 if (!ctrl[id]) {
647 fprintf(file, "*** invalid %s value %d ", name, id);
648 return 1;
649 }
650 if (ctrl[id][0]) {
651 if (space && *space)
652 string(file, " ");
653 string(file, ctrl[id]);
654 if (space)
655 *space = 1;
656 }
657 return 0;
658 }
659
660 static int
661 print_opcode(FILE *file, const struct gen_device_info *devinfo,
662 enum opcode id)
663 {
664 const struct opcode_desc *desc = brw_opcode_desc(devinfo, id);
665 if (!desc) {
666 format(file, "*** invalid opcode value %d ", id);
667 return 1;
668 }
669 string(file, desc->name);
670 return 0;
671 }
672
673 static int
674 reg(FILE *file, unsigned _reg_file, unsigned _reg_nr)
675 {
676 int err = 0;
677
678 /* Clear the Compr4 instruction compression bit. */
679 if (_reg_file == BRW_MESSAGE_REGISTER_FILE)
680 _reg_nr &= ~BRW_MRF_COMPR4;
681
682 if (_reg_file == BRW_ARCHITECTURE_REGISTER_FILE) {
683 switch (_reg_nr & 0xf0) {
684 case BRW_ARF_NULL:
685 string(file, "null");
686 break;
687 case BRW_ARF_ADDRESS:
688 format(file, "a%d", _reg_nr & 0x0f);
689 break;
690 case BRW_ARF_ACCUMULATOR:
691 format(file, "acc%d", _reg_nr & 0x0f);
692 break;
693 case BRW_ARF_FLAG:
694 format(file, "f%d", _reg_nr & 0x0f);
695 break;
696 case BRW_ARF_MASK:
697 format(file, "mask%d", _reg_nr & 0x0f);
698 break;
699 case BRW_ARF_MASK_STACK:
700 format(file, "msd%d", _reg_nr & 0x0f);
701 break;
702 case BRW_ARF_STATE:
703 format(file, "sr%d", _reg_nr & 0x0f);
704 break;
705 case BRW_ARF_CONTROL:
706 format(file, "cr%d", _reg_nr & 0x0f);
707 break;
708 case BRW_ARF_NOTIFICATION_COUNT:
709 format(file, "n%d", _reg_nr & 0x0f);
710 break;
711 case BRW_ARF_IP:
712 string(file, "ip");
713 return -1;
714 break;
715 case BRW_ARF_TDR:
716 format(file, "tdr0");
717 return -1;
718 case BRW_ARF_TIMESTAMP:
719 format(file, "tm%d", _reg_nr & 0x0f);
720 break;
721 default:
722 format(file, "ARF%d", _reg_nr);
723 break;
724 }
725 } else {
726 err |= control(file, "src reg file", reg_file, _reg_file, NULL);
727 format(file, "%d", _reg_nr);
728 }
729 return err;
730 }
731
732 static int
733 dest(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
734 {
735 enum brw_reg_type type = brw_inst_dst_type(devinfo, inst);
736 unsigned elem_size = brw_reg_type_to_size(type);
737 int err = 0;
738
739 if (is_split_send(devinfo, brw_inst_opcode(devinfo, inst))) {
740 /* These are fixed for split sends */
741 type = BRW_REGISTER_TYPE_UD;
742 elem_size = 4;
743 if (brw_inst_dst_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
744 err |= reg(file, brw_inst_send_dst_reg_file(devinfo, inst),
745 brw_inst_dst_da_reg_nr(devinfo, inst));
746 unsigned subreg_nr = brw_inst_dst_da16_subreg_nr(devinfo, inst);
747 if (subreg_nr)
748 format(file, ".%u", subreg_nr);
749 string(file, brw_reg_type_to_letters(type));
750 } else {
751 string(file, "g[a0");
752 if (brw_inst_dst_ia_subreg_nr(devinfo, inst))
753 format(file, ".%"PRIu64, brw_inst_dst_ia_subreg_nr(devinfo, inst) /
754 elem_size);
755 if (brw_inst_send_dst_ia16_addr_imm(devinfo, inst))
756 format(file, " %d", brw_inst_send_dst_ia16_addr_imm(devinfo, inst));
757 string(file, "]<");
758 string(file, brw_reg_type_to_letters(type));
759 }
760 } else if (brw_inst_access_mode(devinfo, inst) == BRW_ALIGN_1) {
761 if (brw_inst_dst_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
762 err |= reg(file, brw_inst_dst_reg_file(devinfo, inst),
763 brw_inst_dst_da_reg_nr(devinfo, inst));
764 if (err == -1)
765 return 0;
766 if (brw_inst_dst_da1_subreg_nr(devinfo, inst))
767 format(file, ".%"PRIu64, brw_inst_dst_da1_subreg_nr(devinfo, inst) /
768 elem_size);
769 string(file, "<");
770 err |= control(file, "horiz stride", horiz_stride,
771 brw_inst_dst_hstride(devinfo, inst), NULL);
772 string(file, ">");
773 string(file, brw_reg_type_to_letters(type));
774 } else {
775 string(file, "g[a0");
776 if (brw_inst_dst_ia_subreg_nr(devinfo, inst))
777 format(file, ".%"PRIu64, brw_inst_dst_ia_subreg_nr(devinfo, inst) /
778 elem_size);
779 if (brw_inst_dst_ia1_addr_imm(devinfo, inst))
780 format(file, " %d", brw_inst_dst_ia1_addr_imm(devinfo, inst));
781 string(file, "]<");
782 err |= control(file, "horiz stride", horiz_stride,
783 brw_inst_dst_hstride(devinfo, inst), NULL);
784 string(file, ">");
785 string(file, brw_reg_type_to_letters(type));
786 }
787 } else {
788 if (brw_inst_dst_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
789 err |= reg(file, brw_inst_dst_reg_file(devinfo, inst),
790 brw_inst_dst_da_reg_nr(devinfo, inst));
791 if (err == -1)
792 return 0;
793 if (brw_inst_dst_da16_subreg_nr(devinfo, inst))
794 format(file, ".%u", 16 / elem_size);
795 string(file, "<1>");
796 err |= control(file, "writemask", writemask,
797 brw_inst_da16_writemask(devinfo, inst), NULL);
798 string(file, brw_reg_type_to_letters(type));
799 } else {
800 err = 1;
801 string(file, "Indirect align16 address mode not supported");
802 }
803 }
804
805 return 0;
806 }
807
808 static int
809 dest_3src(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
810 {
811 bool is_align1 = brw_inst_3src_access_mode(devinfo, inst) == BRW_ALIGN_1;
812 int err = 0;
813 uint32_t reg_file;
814 unsigned subreg_nr;
815 enum brw_reg_type type;
816
817 if (is_align1 && brw_inst_3src_a1_dst_reg_file(devinfo, inst))
818 reg_file = BRW_ARCHITECTURE_REGISTER_FILE;
819 else if (devinfo->gen == 6 && brw_inst_3src_a16_dst_reg_file(devinfo, inst))
820 reg_file = BRW_MESSAGE_REGISTER_FILE;
821 else
822 reg_file = BRW_GENERAL_REGISTER_FILE;
823
824 err |= reg(file, reg_file, brw_inst_3src_dst_reg_nr(devinfo, inst));
825 if (err == -1)
826 return 0;
827
828 if (is_align1) {
829 type = brw_inst_3src_a1_dst_type(devinfo, inst);
830 subreg_nr = brw_inst_3src_a1_dst_subreg_nr(devinfo, inst);
831 } else {
832 type = brw_inst_3src_a16_dst_type(devinfo, inst);
833 subreg_nr = brw_inst_3src_a16_dst_subreg_nr(devinfo, inst) * 4;
834 }
835 subreg_nr /= brw_reg_type_to_size(type);
836
837 if (subreg_nr)
838 format(file, ".%u", subreg_nr);
839 string(file, "<1>");
840
841 if (!is_align1) {
842 err |= control(file, "writemask", writemask,
843 brw_inst_3src_a16_dst_writemask(devinfo, inst), NULL);
844 }
845 string(file, brw_reg_type_to_letters(type));
846
847 return 0;
848 }
849
850 static int
851 src_align1_region(FILE *file,
852 unsigned _vert_stride, unsigned _width,
853 unsigned _horiz_stride)
854 {
855 int err = 0;
856 string(file, "<");
857 err |= control(file, "vert stride", vert_stride, _vert_stride, NULL);
858 string(file, ",");
859 err |= control(file, "width", width, _width, NULL);
860 string(file, ",");
861 err |= control(file, "horiz_stride", horiz_stride, _horiz_stride, NULL);
862 string(file, ">");
863 return err;
864 }
865
866 static int
867 src_da1(FILE *file,
868 const struct gen_device_info *devinfo,
869 unsigned opcode,
870 enum brw_reg_type type, unsigned _reg_file,
871 unsigned _vert_stride, unsigned _width, unsigned _horiz_stride,
872 unsigned reg_num, unsigned sub_reg_num, unsigned __abs,
873 unsigned _negate)
874 {
875 int err = 0;
876
877 if (devinfo->gen >= 8 && is_logic_instruction(opcode))
878 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
879 else
880 err |= control(file, "negate", m_negate, _negate, NULL);
881
882 err |= control(file, "abs", _abs, __abs, NULL);
883
884 err |= reg(file, _reg_file, reg_num);
885 if (err == -1)
886 return 0;
887 if (sub_reg_num) {
888 unsigned elem_size = brw_reg_type_to_size(type);
889 format(file, ".%d", sub_reg_num / elem_size); /* use formal style like spec */
890 }
891 src_align1_region(file, _vert_stride, _width, _horiz_stride);
892 string(file, brw_reg_type_to_letters(type));
893 return err;
894 }
895
896 static int
897 src_ia1(FILE *file,
898 const struct gen_device_info *devinfo,
899 unsigned opcode,
900 enum brw_reg_type type,
901 int _addr_imm,
902 unsigned _addr_subreg_nr,
903 unsigned _negate,
904 unsigned __abs,
905 unsigned _horiz_stride, unsigned _width, unsigned _vert_stride)
906 {
907 int err = 0;
908
909 if (devinfo->gen >= 8 && is_logic_instruction(opcode))
910 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
911 else
912 err |= control(file, "negate", m_negate, _negate, NULL);
913
914 err |= control(file, "abs", _abs, __abs, NULL);
915
916 string(file, "g[a0");
917 if (_addr_subreg_nr)
918 format(file, ".%d", _addr_subreg_nr);
919 if (_addr_imm)
920 format(file, " %d", _addr_imm);
921 string(file, "]");
922 src_align1_region(file, _vert_stride, _width, _horiz_stride);
923 string(file, brw_reg_type_to_letters(type));
924 return err;
925 }
926
927 static int
928 src_swizzle(FILE *file, unsigned swiz)
929 {
930 unsigned x = BRW_GET_SWZ(swiz, BRW_CHANNEL_X);
931 unsigned y = BRW_GET_SWZ(swiz, BRW_CHANNEL_Y);
932 unsigned z = BRW_GET_SWZ(swiz, BRW_CHANNEL_Z);
933 unsigned w = BRW_GET_SWZ(swiz, BRW_CHANNEL_W);
934 int err = 0;
935
936 if (x == y && x == z && x == w) {
937 string(file, ".");
938 err |= control(file, "channel select", chan_sel, x, NULL);
939 } else if (swiz != BRW_SWIZZLE_XYZW) {
940 string(file, ".");
941 err |= control(file, "channel select", chan_sel, x, NULL);
942 err |= control(file, "channel select", chan_sel, y, NULL);
943 err |= control(file, "channel select", chan_sel, z, NULL);
944 err |= control(file, "channel select", chan_sel, w, NULL);
945 }
946 return err;
947 }
948
949 static int
950 src_da16(FILE *file,
951 const struct gen_device_info *devinfo,
952 unsigned opcode,
953 enum brw_reg_type type,
954 unsigned _reg_file,
955 unsigned _vert_stride,
956 unsigned _reg_nr,
957 unsigned _subreg_nr,
958 unsigned __abs,
959 unsigned _negate,
960 unsigned swz_x, unsigned swz_y, unsigned swz_z, unsigned swz_w)
961 {
962 int err = 0;
963
964 if (devinfo->gen >= 8 && is_logic_instruction(opcode))
965 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
966 else
967 err |= control(file, "negate", m_negate, _negate, NULL);
968
969 err |= control(file, "abs", _abs, __abs, NULL);
970
971 err |= reg(file, _reg_file, _reg_nr);
972 if (err == -1)
973 return 0;
974 if (_subreg_nr) {
975 unsigned elem_size = brw_reg_type_to_size(type);
976
977 /* bit4 for subreg number byte addressing. Make this same meaning as
978 in da1 case, so output looks consistent. */
979 format(file, ".%d", 16 / elem_size);
980 }
981 string(file, "<");
982 err |= control(file, "vert stride", vert_stride, _vert_stride, NULL);
983 string(file, ">");
984 err |= src_swizzle(file, BRW_SWIZZLE4(swz_x, swz_y, swz_z, swz_w));
985 string(file, brw_reg_type_to_letters(type));
986 return err;
987 }
988
989 static enum brw_vertical_stride
990 vstride_from_align1_3src_vstride(enum gen10_align1_3src_vertical_stride vstride)
991 {
992 switch (vstride) {
993 case BRW_ALIGN1_3SRC_VERTICAL_STRIDE_0: return BRW_VERTICAL_STRIDE_0;
994 case BRW_ALIGN1_3SRC_VERTICAL_STRIDE_2: return BRW_VERTICAL_STRIDE_2;
995 case BRW_ALIGN1_3SRC_VERTICAL_STRIDE_4: return BRW_VERTICAL_STRIDE_4;
996 case BRW_ALIGN1_3SRC_VERTICAL_STRIDE_8: return BRW_VERTICAL_STRIDE_8;
997 default:
998 unreachable("not reached");
999 }
1000 }
1001
1002 static enum brw_horizontal_stride
1003 hstride_from_align1_3src_hstride(enum gen10_align1_3src_src_horizontal_stride hstride)
1004 {
1005 switch (hstride) {
1006 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_0: return BRW_HORIZONTAL_STRIDE_0;
1007 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_1: return BRW_HORIZONTAL_STRIDE_1;
1008 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_2: return BRW_HORIZONTAL_STRIDE_2;
1009 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_4: return BRW_HORIZONTAL_STRIDE_4;
1010 default:
1011 unreachable("not reached");
1012 }
1013 }
1014
1015 static enum brw_vertical_stride
1016 vstride_from_align1_3src_hstride(enum gen10_align1_3src_src_horizontal_stride hstride)
1017 {
1018 switch (hstride) {
1019 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_0: return BRW_VERTICAL_STRIDE_0;
1020 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_1: return BRW_VERTICAL_STRIDE_1;
1021 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_2: return BRW_VERTICAL_STRIDE_2;
1022 case BRW_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_4: return BRW_VERTICAL_STRIDE_4;
1023 default:
1024 unreachable("not reached");
1025 }
1026 }
1027
1028 /* From "GEN10 Regioning Rules for Align1 Ternary Operations" in the
1029 * "Register Region Restrictions" documentation
1030 */
1031 static enum brw_width
1032 implied_width(enum brw_vertical_stride _vert_stride,
1033 enum brw_horizontal_stride _horiz_stride)
1034 {
1035 /* "1. Width is 1 when Vertical and Horizontal Strides are both zero." */
1036 if (_vert_stride == BRW_VERTICAL_STRIDE_0 &&
1037 _horiz_stride == BRW_HORIZONTAL_STRIDE_0) {
1038 return BRW_WIDTH_1;
1039
1040 /* "2. Width is equal to vertical stride when Horizontal Stride is zero." */
1041 } else if (_horiz_stride == BRW_HORIZONTAL_STRIDE_0) {
1042 switch (_vert_stride) {
1043 case BRW_VERTICAL_STRIDE_2: return BRW_WIDTH_2;
1044 case BRW_VERTICAL_STRIDE_4: return BRW_WIDTH_4;
1045 case BRW_VERTICAL_STRIDE_8: return BRW_WIDTH_8;
1046 case BRW_VERTICAL_STRIDE_0:
1047 default:
1048 unreachable("not reached");
1049 }
1050
1051 } else {
1052 /* FINISHME: Implement these: */
1053
1054 /* "3. Width is equal to Vertical Stride/Horizontal Stride when both
1055 * Strides are non-zero.
1056 *
1057 * 4. Vertical Stride must not be zero if Horizontal Stride is non-zero.
1058 * This implies Vertical Stride is always greater than Horizontal
1059 * Stride."
1060 *
1061 * Given these statements and the knowledge that the stride and width
1062 * values are encoded in logarithmic form, we can perform the division
1063 * by just subtracting.
1064 */
1065 return _vert_stride - _horiz_stride;
1066 }
1067 }
1068
1069 static int
1070 src0_3src(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1071 {
1072 int err = 0;
1073 unsigned reg_nr, subreg_nr;
1074 enum brw_reg_file _file;
1075 enum brw_reg_type type;
1076 enum brw_vertical_stride _vert_stride;
1077 enum brw_width _width;
1078 enum brw_horizontal_stride _horiz_stride;
1079 bool is_scalar_region;
1080 bool is_align1 = brw_inst_3src_access_mode(devinfo, inst) == BRW_ALIGN_1;
1081
1082 if (is_align1) {
1083 if (brw_inst_3src_a1_src0_reg_file(devinfo, inst) ==
1084 BRW_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1085 _file = BRW_GENERAL_REGISTER_FILE;
1086 reg_nr = brw_inst_3src_src0_reg_nr(devinfo, inst);
1087 subreg_nr = brw_inst_3src_a1_src0_subreg_nr(devinfo, inst);
1088 type = brw_inst_3src_a1_src0_type(devinfo, inst);
1089 } else if (brw_inst_3src_a1_src0_type(devinfo, inst) ==
1090 BRW_REGISTER_TYPE_NF) {
1091 _file = BRW_ARCHITECTURE_REGISTER_FILE;
1092 reg_nr = brw_inst_3src_src0_reg_nr(devinfo, inst);
1093 subreg_nr = brw_inst_3src_a1_src0_subreg_nr(devinfo, inst);
1094 type = brw_inst_3src_a1_src0_type(devinfo, inst);
1095 } else {
1096 _file = BRW_IMMEDIATE_VALUE;
1097 uint16_t imm_val = brw_inst_3src_a1_src0_imm(devinfo, inst);
1098 enum brw_reg_type type = brw_inst_3src_a1_src0_type(devinfo, inst);
1099
1100 if (type == BRW_REGISTER_TYPE_W) {
1101 format(file, "%dW", imm_val);
1102 } else if (type == BRW_REGISTER_TYPE_UW) {
1103 format(file, "0x%04xUW", imm_val);
1104 } else if (type == BRW_REGISTER_TYPE_HF) {
1105 format(file, "%-gF", _mesa_half_to_float(imm_val));
1106 }
1107 return 0;
1108 }
1109
1110 _vert_stride = vstride_from_align1_3src_vstride(
1111 brw_inst_3src_a1_src0_vstride(devinfo, inst));
1112 _horiz_stride = hstride_from_align1_3src_hstride(
1113 brw_inst_3src_a1_src0_hstride(devinfo, inst));
1114 _width = implied_width(_vert_stride, _horiz_stride);
1115 } else {
1116 _file = BRW_GENERAL_REGISTER_FILE;
1117 reg_nr = brw_inst_3src_src0_reg_nr(devinfo, inst);
1118 subreg_nr = brw_inst_3src_a16_src0_subreg_nr(devinfo, inst) * 4;
1119 type = brw_inst_3src_a16_src_type(devinfo, inst);
1120
1121 if (brw_inst_3src_a16_src0_rep_ctrl(devinfo, inst)) {
1122 _vert_stride = BRW_VERTICAL_STRIDE_0;
1123 _width = BRW_WIDTH_1;
1124 _horiz_stride = BRW_HORIZONTAL_STRIDE_0;
1125 } else {
1126 _vert_stride = BRW_VERTICAL_STRIDE_4;
1127 _width = BRW_WIDTH_4;
1128 _horiz_stride = BRW_HORIZONTAL_STRIDE_1;
1129 }
1130 }
1131 is_scalar_region = _vert_stride == BRW_VERTICAL_STRIDE_0 &&
1132 _width == BRW_WIDTH_1 &&
1133 _horiz_stride == BRW_HORIZONTAL_STRIDE_0;
1134
1135 subreg_nr /= brw_reg_type_to_size(type);
1136
1137 err |= control(file, "negate", m_negate,
1138 brw_inst_3src_src0_negate(devinfo, inst), NULL);
1139 err |= control(file, "abs", _abs, brw_inst_3src_src0_abs(devinfo, inst), NULL);
1140
1141 err |= reg(file, _file, reg_nr);
1142 if (err == -1)
1143 return 0;
1144 if (subreg_nr || is_scalar_region)
1145 format(file, ".%d", subreg_nr);
1146 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1147 if (!is_scalar_region && !is_align1)
1148 err |= src_swizzle(file, brw_inst_3src_a16_src0_swizzle(devinfo, inst));
1149 string(file, brw_reg_type_to_letters(type));
1150 return err;
1151 }
1152
1153 static int
1154 src1_3src(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1155 {
1156 int err = 0;
1157 unsigned reg_nr, subreg_nr;
1158 enum brw_reg_file _file;
1159 enum brw_reg_type type;
1160 enum brw_vertical_stride _vert_stride;
1161 enum brw_width _width;
1162 enum brw_horizontal_stride _horiz_stride;
1163 bool is_scalar_region;
1164 bool is_align1 = brw_inst_3src_access_mode(devinfo, inst) == BRW_ALIGN_1;
1165
1166 if (is_align1) {
1167 if (brw_inst_3src_a1_src1_reg_file(devinfo, inst) ==
1168 BRW_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1169 _file = BRW_GENERAL_REGISTER_FILE;
1170 } else {
1171 _file = BRW_ARCHITECTURE_REGISTER_FILE;
1172 }
1173
1174 reg_nr = brw_inst_3src_src1_reg_nr(devinfo, inst);
1175 subreg_nr = brw_inst_3src_a1_src1_subreg_nr(devinfo, inst);
1176 type = brw_inst_3src_a1_src1_type(devinfo, inst);
1177
1178 _vert_stride = vstride_from_align1_3src_vstride(
1179 brw_inst_3src_a1_src1_vstride(devinfo, inst));
1180 _horiz_stride = hstride_from_align1_3src_hstride(
1181 brw_inst_3src_a1_src1_hstride(devinfo, inst));
1182 _width = implied_width(_vert_stride, _horiz_stride);
1183 } else {
1184 _file = BRW_GENERAL_REGISTER_FILE;
1185 reg_nr = brw_inst_3src_src1_reg_nr(devinfo, inst);
1186 subreg_nr = brw_inst_3src_a16_src1_subreg_nr(devinfo, inst) * 4;
1187 type = brw_inst_3src_a16_src_type(devinfo, inst);
1188
1189 if (brw_inst_3src_a16_src1_rep_ctrl(devinfo, inst)) {
1190 _vert_stride = BRW_VERTICAL_STRIDE_0;
1191 _width = BRW_WIDTH_1;
1192 _horiz_stride = BRW_HORIZONTAL_STRIDE_0;
1193 } else {
1194 _vert_stride = BRW_VERTICAL_STRIDE_4;
1195 _width = BRW_WIDTH_4;
1196 _horiz_stride = BRW_HORIZONTAL_STRIDE_1;
1197 }
1198 }
1199 is_scalar_region = _vert_stride == BRW_VERTICAL_STRIDE_0 &&
1200 _width == BRW_WIDTH_1 &&
1201 _horiz_stride == BRW_HORIZONTAL_STRIDE_0;
1202
1203 subreg_nr /= brw_reg_type_to_size(type);
1204
1205 err |= control(file, "negate", m_negate,
1206 brw_inst_3src_src1_negate(devinfo, inst), NULL);
1207 err |= control(file, "abs", _abs, brw_inst_3src_src1_abs(devinfo, inst), NULL);
1208
1209 err |= reg(file, _file, reg_nr);
1210 if (err == -1)
1211 return 0;
1212 if (subreg_nr || is_scalar_region)
1213 format(file, ".%d", subreg_nr);
1214 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1215 if (!is_scalar_region && !is_align1)
1216 err |= src_swizzle(file, brw_inst_3src_a16_src1_swizzle(devinfo, inst));
1217 string(file, brw_reg_type_to_letters(type));
1218 return err;
1219 }
1220
1221 static int
1222 src2_3src(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1223 {
1224 int err = 0;
1225 unsigned reg_nr, subreg_nr;
1226 enum brw_reg_file _file;
1227 enum brw_reg_type type;
1228 enum brw_vertical_stride _vert_stride;
1229 enum brw_width _width;
1230 enum brw_horizontal_stride _horiz_stride;
1231 bool is_scalar_region;
1232 bool is_align1 = brw_inst_3src_access_mode(devinfo, inst) == BRW_ALIGN_1;
1233
1234 if (is_align1) {
1235 if (brw_inst_3src_a1_src2_reg_file(devinfo, inst) ==
1236 BRW_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1237 _file = BRW_GENERAL_REGISTER_FILE;
1238 reg_nr = brw_inst_3src_src2_reg_nr(devinfo, inst);
1239 subreg_nr = brw_inst_3src_a1_src2_subreg_nr(devinfo, inst);
1240 type = brw_inst_3src_a1_src2_type(devinfo, inst);
1241 } else {
1242 _file = BRW_IMMEDIATE_VALUE;
1243 uint16_t imm_val = brw_inst_3src_a1_src2_imm(devinfo, inst);
1244 enum brw_reg_type type = brw_inst_3src_a1_src2_type(devinfo, inst);
1245
1246 if (type == BRW_REGISTER_TYPE_W) {
1247 format(file, "%dW", imm_val);
1248 } else if (type == BRW_REGISTER_TYPE_UW) {
1249 format(file, "0x%04xUW", imm_val);
1250 } else if (type == BRW_REGISTER_TYPE_HF) {
1251 format(file, "%-gF", _mesa_half_to_float(imm_val));
1252 }
1253 return 0;
1254 }
1255
1256 /* FINISHME: No vertical stride on src2. Is using the hstride in place
1257 * correct? Doesn't seem like it, since there's hstride=1 but
1258 * no vstride=1.
1259 */
1260 _vert_stride = vstride_from_align1_3src_hstride(
1261 brw_inst_3src_a1_src2_hstride(devinfo, inst));
1262 _horiz_stride = hstride_from_align1_3src_hstride(
1263 brw_inst_3src_a1_src2_hstride(devinfo, inst));
1264 _width = implied_width(_vert_stride, _horiz_stride);
1265 } else {
1266 _file = BRW_GENERAL_REGISTER_FILE;
1267 reg_nr = brw_inst_3src_src2_reg_nr(devinfo, inst);
1268 subreg_nr = brw_inst_3src_a16_src2_subreg_nr(devinfo, inst) * 4;
1269 type = brw_inst_3src_a16_src_type(devinfo, inst);
1270
1271 if (brw_inst_3src_a16_src2_rep_ctrl(devinfo, inst)) {
1272 _vert_stride = BRW_VERTICAL_STRIDE_0;
1273 _width = BRW_WIDTH_1;
1274 _horiz_stride = BRW_HORIZONTAL_STRIDE_0;
1275 } else {
1276 _vert_stride = BRW_VERTICAL_STRIDE_4;
1277 _width = BRW_WIDTH_4;
1278 _horiz_stride = BRW_HORIZONTAL_STRIDE_1;
1279 }
1280 }
1281 is_scalar_region = _vert_stride == BRW_VERTICAL_STRIDE_0 &&
1282 _width == BRW_WIDTH_1 &&
1283 _horiz_stride == BRW_HORIZONTAL_STRIDE_0;
1284
1285 subreg_nr /= brw_reg_type_to_size(type);
1286
1287 err |= control(file, "negate", m_negate,
1288 brw_inst_3src_src2_negate(devinfo, inst), NULL);
1289 err |= control(file, "abs", _abs, brw_inst_3src_src2_abs(devinfo, inst), NULL);
1290
1291 err |= reg(file, _file, reg_nr);
1292 if (err == -1)
1293 return 0;
1294 if (subreg_nr || is_scalar_region)
1295 format(file, ".%d", subreg_nr);
1296 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1297 if (!is_scalar_region && !is_align1)
1298 err |= src_swizzle(file, brw_inst_3src_a16_src2_swizzle(devinfo, inst));
1299 string(file, brw_reg_type_to_letters(type));
1300 return err;
1301 }
1302
1303 static int
1304 imm(FILE *file, const struct gen_device_info *devinfo, enum brw_reg_type type,
1305 const brw_inst *inst)
1306 {
1307 switch (type) {
1308 case BRW_REGISTER_TYPE_UQ:
1309 format(file, "0x%016"PRIx64"UQ", brw_inst_imm_uq(devinfo, inst));
1310 break;
1311 case BRW_REGISTER_TYPE_Q:
1312 format(file, "%"PRId64"Q", brw_inst_imm_uq(devinfo, inst));
1313 break;
1314 case BRW_REGISTER_TYPE_UD:
1315 format(file, "0x%08xUD", brw_inst_imm_ud(devinfo, inst));
1316 break;
1317 case BRW_REGISTER_TYPE_D:
1318 format(file, "%dD", brw_inst_imm_d(devinfo, inst));
1319 break;
1320 case BRW_REGISTER_TYPE_UW:
1321 format(file, "0x%04xUW", (uint16_t) brw_inst_imm_ud(devinfo, inst));
1322 break;
1323 case BRW_REGISTER_TYPE_W:
1324 format(file, "%dW", (int16_t) brw_inst_imm_d(devinfo, inst));
1325 break;
1326 case BRW_REGISTER_TYPE_UV:
1327 format(file, "0x%08xUV", brw_inst_imm_ud(devinfo, inst));
1328 break;
1329 case BRW_REGISTER_TYPE_VF:
1330 format(file, "0x%"PRIx64"VF", brw_inst_bits(inst, 127, 96));
1331 pad(file, 48);
1332 format(file, "/* [%-gF, %-gF, %-gF, %-gF]VF */",
1333 brw_vf_to_float(brw_inst_imm_ud(devinfo, inst)),
1334 brw_vf_to_float(brw_inst_imm_ud(devinfo, inst) >> 8),
1335 brw_vf_to_float(brw_inst_imm_ud(devinfo, inst) >> 16),
1336 brw_vf_to_float(brw_inst_imm_ud(devinfo, inst) >> 24));
1337 break;
1338 case BRW_REGISTER_TYPE_V:
1339 format(file, "0x%08xV", brw_inst_imm_ud(devinfo, inst));
1340 break;
1341 case BRW_REGISTER_TYPE_F:
1342 format(file, "0x%"PRIx64"F", brw_inst_bits(inst, 127, 96));
1343 pad(file, 48);
1344 format(file, " /* %-gF */", brw_inst_imm_f(devinfo, inst));
1345 break;
1346 case BRW_REGISTER_TYPE_DF:
1347 format(file, "0x%016"PRIx64"DF", brw_inst_bits(inst, 127, 64));
1348 pad(file, 48);
1349 format(file, "/* %-gDF */", brw_inst_imm_df(devinfo, inst));
1350 break;
1351 case BRW_REGISTER_TYPE_HF:
1352 string(file, "Half Float IMM");
1353 break;
1354 case BRW_REGISTER_TYPE_NF:
1355 case BRW_REGISTER_TYPE_UB:
1356 case BRW_REGISTER_TYPE_B:
1357 format(file, "*** invalid immediate type %d ", type);
1358 }
1359 return 0;
1360 }
1361
1362 static int
1363 src_sends_da(FILE *file,
1364 const struct gen_device_info *devinfo,
1365 enum brw_reg_type type,
1366 unsigned _reg_nr,
1367 unsigned _reg_subnr)
1368 {
1369 int err = 0;
1370
1371 err |= reg(file, BRW_GENERAL_REGISTER_FILE, _reg_nr);
1372 if (err == -1)
1373 return 0;
1374 if (_reg_subnr)
1375 format(file, ".1");
1376 string(file, brw_reg_type_to_letters(type));
1377
1378 return err;
1379 }
1380
1381 static int
1382 src_sends_ia(FILE *file,
1383 const struct gen_device_info *devinfo,
1384 enum brw_reg_type type,
1385 int _addr_imm,
1386 unsigned _addr_subreg_nr)
1387 {
1388 string(file, "g[a0");
1389 if (_addr_subreg_nr)
1390 format(file, ".1");
1391 if (_addr_imm)
1392 format(file, " %d", _addr_imm);
1393 string(file, "]");
1394 string(file, brw_reg_type_to_letters(type));
1395
1396 return 0;
1397 }
1398
1399 static int
1400 src0(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1401 {
1402 if (is_split_send(devinfo, brw_inst_opcode(devinfo, inst))) {
1403 if (brw_inst_send_src0_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
1404 return src_sends_da(file,
1405 devinfo,
1406 BRW_REGISTER_TYPE_UD,
1407 brw_inst_src0_da_reg_nr(devinfo, inst),
1408 brw_inst_src0_da16_subreg_nr(devinfo, inst));
1409 } else {
1410 return src_sends_ia(file,
1411 devinfo,
1412 BRW_REGISTER_TYPE_UD,
1413 brw_inst_send_src0_ia16_addr_imm(devinfo, inst),
1414 brw_inst_src0_ia_subreg_nr(devinfo, inst));
1415 }
1416 } else if (brw_inst_src0_reg_file(devinfo, inst) == BRW_IMMEDIATE_VALUE) {
1417 return imm(file, devinfo, brw_inst_src0_type(devinfo, inst), inst);
1418 } else if (brw_inst_access_mode(devinfo, inst) == BRW_ALIGN_1) {
1419 if (brw_inst_src0_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
1420 return src_da1(file,
1421 devinfo,
1422 brw_inst_opcode(devinfo, inst),
1423 brw_inst_src0_type(devinfo, inst),
1424 brw_inst_src0_reg_file(devinfo, inst),
1425 brw_inst_src0_vstride(devinfo, inst),
1426 brw_inst_src0_width(devinfo, inst),
1427 brw_inst_src0_hstride(devinfo, inst),
1428 brw_inst_src0_da_reg_nr(devinfo, inst),
1429 brw_inst_src0_da1_subreg_nr(devinfo, inst),
1430 brw_inst_src0_abs(devinfo, inst),
1431 brw_inst_src0_negate(devinfo, inst));
1432 } else {
1433 return src_ia1(file,
1434 devinfo,
1435 brw_inst_opcode(devinfo, inst),
1436 brw_inst_src0_type(devinfo, inst),
1437 brw_inst_src0_ia1_addr_imm(devinfo, inst),
1438 brw_inst_src0_ia_subreg_nr(devinfo, inst),
1439 brw_inst_src0_negate(devinfo, inst),
1440 brw_inst_src0_abs(devinfo, inst),
1441 brw_inst_src0_hstride(devinfo, inst),
1442 brw_inst_src0_width(devinfo, inst),
1443 brw_inst_src0_vstride(devinfo, inst));
1444 }
1445 } else {
1446 if (brw_inst_src0_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
1447 return src_da16(file,
1448 devinfo,
1449 brw_inst_opcode(devinfo, inst),
1450 brw_inst_src0_type(devinfo, inst),
1451 brw_inst_src0_reg_file(devinfo, inst),
1452 brw_inst_src0_vstride(devinfo, inst),
1453 brw_inst_src0_da_reg_nr(devinfo, inst),
1454 brw_inst_src0_da16_subreg_nr(devinfo, inst),
1455 brw_inst_src0_abs(devinfo, inst),
1456 brw_inst_src0_negate(devinfo, inst),
1457 brw_inst_src0_da16_swiz_x(devinfo, inst),
1458 brw_inst_src0_da16_swiz_y(devinfo, inst),
1459 brw_inst_src0_da16_swiz_z(devinfo, inst),
1460 brw_inst_src0_da16_swiz_w(devinfo, inst));
1461 } else {
1462 string(file, "Indirect align16 address mode not supported");
1463 return 1;
1464 }
1465 }
1466 }
1467
1468 static int
1469 src1(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1470 {
1471 if (is_split_send(devinfo, brw_inst_opcode(devinfo, inst))) {
1472 return src_sends_da(file,
1473 devinfo,
1474 BRW_REGISTER_TYPE_UD,
1475 brw_inst_send_src1_reg_nr(devinfo, inst),
1476 0 /* subreg_nr */);
1477 } else if (brw_inst_src1_reg_file(devinfo, inst) == BRW_IMMEDIATE_VALUE) {
1478 return imm(file, devinfo, brw_inst_src1_type(devinfo, inst), inst);
1479 } else if (brw_inst_access_mode(devinfo, inst) == BRW_ALIGN_1) {
1480 if (brw_inst_src1_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
1481 return src_da1(file,
1482 devinfo,
1483 brw_inst_opcode(devinfo, inst),
1484 brw_inst_src1_type(devinfo, inst),
1485 brw_inst_src1_reg_file(devinfo, inst),
1486 brw_inst_src1_vstride(devinfo, inst),
1487 brw_inst_src1_width(devinfo, inst),
1488 brw_inst_src1_hstride(devinfo, inst),
1489 brw_inst_src1_da_reg_nr(devinfo, inst),
1490 brw_inst_src1_da1_subreg_nr(devinfo, inst),
1491 brw_inst_src1_abs(devinfo, inst),
1492 brw_inst_src1_negate(devinfo, inst));
1493 } else {
1494 return src_ia1(file,
1495 devinfo,
1496 brw_inst_opcode(devinfo, inst),
1497 brw_inst_src1_type(devinfo, inst),
1498 brw_inst_src1_ia1_addr_imm(devinfo, inst),
1499 brw_inst_src1_ia_subreg_nr(devinfo, inst),
1500 brw_inst_src1_negate(devinfo, inst),
1501 brw_inst_src1_abs(devinfo, inst),
1502 brw_inst_src1_hstride(devinfo, inst),
1503 brw_inst_src1_width(devinfo, inst),
1504 brw_inst_src1_vstride(devinfo, inst));
1505 }
1506 } else {
1507 if (brw_inst_src1_address_mode(devinfo, inst) == BRW_ADDRESS_DIRECT) {
1508 return src_da16(file,
1509 devinfo,
1510 brw_inst_opcode(devinfo, inst),
1511 brw_inst_src1_type(devinfo, inst),
1512 brw_inst_src1_reg_file(devinfo, inst),
1513 brw_inst_src1_vstride(devinfo, inst),
1514 brw_inst_src1_da_reg_nr(devinfo, inst),
1515 brw_inst_src1_da16_subreg_nr(devinfo, inst),
1516 brw_inst_src1_abs(devinfo, inst),
1517 brw_inst_src1_negate(devinfo, inst),
1518 brw_inst_src1_da16_swiz_x(devinfo, inst),
1519 brw_inst_src1_da16_swiz_y(devinfo, inst),
1520 brw_inst_src1_da16_swiz_z(devinfo, inst),
1521 brw_inst_src1_da16_swiz_w(devinfo, inst));
1522 } else {
1523 string(file, "Indirect align16 address mode not supported");
1524 return 1;
1525 }
1526 }
1527 }
1528
1529 static int
1530 qtr_ctrl(FILE *file, const struct gen_device_info *devinfo, const brw_inst *inst)
1531 {
1532 int qtr_ctl = brw_inst_qtr_control(devinfo, inst);
1533 int exec_size = 1 << brw_inst_exec_size(devinfo, inst);
1534 const unsigned nib_ctl = devinfo->gen < 7 ? 0 :
1535 brw_inst_nib_control(devinfo, inst);
1536
1537 if (exec_size < 8 || nib_ctl) {
1538 format(file, " %dN", qtr_ctl * 2 + nib_ctl + 1);
1539 } else if (exec_size == 8) {
1540 switch (qtr_ctl) {
1541 case 0:
1542 string(file, " 1Q");
1543 break;
1544 case 1:
1545 string(file, " 2Q");
1546 break;
1547 case 2:
1548 string(file, " 3Q");
1549 break;
1550 case 3:
1551 string(file, " 4Q");
1552 break;
1553 }
1554 } else if (exec_size == 16) {
1555 if (qtr_ctl < 2)
1556 string(file, " 1H");
1557 else
1558 string(file, " 2H");
1559 }
1560 return 0;
1561 }
1562
1563 #ifdef DEBUG
1564 static __attribute__((__unused__)) int
1565 brw_disassemble_imm(const struct gen_device_info *devinfo,
1566 uint32_t dw3, uint32_t dw2, uint32_t dw1, uint32_t dw0)
1567 {
1568 brw_inst inst;
1569 inst.data[0] = (((uint64_t) dw1) << 32) | ((uint64_t) dw0);
1570 inst.data[1] = (((uint64_t) dw3) << 32) | ((uint64_t) dw2);
1571 return brw_disassemble_inst(stderr, devinfo, &inst, false);
1572 }
1573 #endif
1574
1575 int
1576 brw_disassemble_inst(FILE *file, const struct gen_device_info *devinfo,
1577 const brw_inst *inst, bool is_compacted)
1578 {
1579 int err = 0;
1580 int space = 0;
1581
1582 const enum opcode opcode = brw_inst_opcode(devinfo, inst);
1583 const struct opcode_desc *desc = brw_opcode_desc(devinfo, opcode);
1584
1585 if (brw_inst_pred_control(devinfo, inst)) {
1586 string(file, "(");
1587 err |= control(file, "predicate inverse", pred_inv,
1588 brw_inst_pred_inv(devinfo, inst), NULL);
1589 format(file, "f%"PRIu64".%"PRIu64,
1590 devinfo->gen >= 7 ? brw_inst_flag_reg_nr(devinfo, inst) : 0,
1591 brw_inst_flag_subreg_nr(devinfo, inst));
1592 if (brw_inst_access_mode(devinfo, inst) == BRW_ALIGN_1) {
1593 err |= control(file, "predicate control align1", pred_ctrl_align1,
1594 brw_inst_pred_control(devinfo, inst), NULL);
1595 } else {
1596 err |= control(file, "predicate control align16", pred_ctrl_align16,
1597 brw_inst_pred_control(devinfo, inst), NULL);
1598 }
1599 string(file, ") ");
1600 }
1601
1602 err |= print_opcode(file, devinfo, opcode);
1603 err |= control(file, "saturate", saturate, brw_inst_saturate(devinfo, inst),
1604 NULL);
1605
1606 err |= control(file, "debug control", debug_ctrl,
1607 brw_inst_debug_control(devinfo, inst), NULL);
1608
1609 if (opcode == BRW_OPCODE_MATH) {
1610 string(file, " ");
1611 err |= control(file, "function", math_function,
1612 brw_inst_math_function(devinfo, inst), NULL);
1613 } else if (!is_send(opcode)) {
1614 err |= control(file, "conditional modifier", conditional_modifier,
1615 brw_inst_cond_modifier(devinfo, inst), NULL);
1616
1617 /* If we're using the conditional modifier, print which flags reg is
1618 * used for it. Note that on gen6+, the embedded-condition SEL and
1619 * control flow doesn't update flags.
1620 */
1621 if (brw_inst_cond_modifier(devinfo, inst) &&
1622 (devinfo->gen < 6 || (opcode != BRW_OPCODE_SEL &&
1623 opcode != BRW_OPCODE_CSEL &&
1624 opcode != BRW_OPCODE_IF &&
1625 opcode != BRW_OPCODE_WHILE))) {
1626 format(file, ".f%"PRIu64".%"PRIu64,
1627 devinfo->gen >= 7 ? brw_inst_flag_reg_nr(devinfo, inst) : 0,
1628 brw_inst_flag_subreg_nr(devinfo, inst));
1629 }
1630 }
1631
1632 if (opcode != BRW_OPCODE_NOP && opcode != BRW_OPCODE_NENOP) {
1633 string(file, "(");
1634 err |= control(file, "execution size", exec_size,
1635 brw_inst_exec_size(devinfo, inst), NULL);
1636 string(file, ")");
1637 }
1638
1639 if (opcode == BRW_OPCODE_SEND && devinfo->gen < 6)
1640 format(file, " %"PRIu64, brw_inst_base_mrf(devinfo, inst));
1641
1642 if (has_uip(devinfo, opcode)) {
1643 /* Instructions that have UIP also have JIP. */
1644 pad(file, 16);
1645 format(file, "JIP: %d", brw_inst_jip(devinfo, inst));
1646 pad(file, 32);
1647 format(file, "UIP: %d", brw_inst_uip(devinfo, inst));
1648 } else if (has_jip(devinfo, opcode)) {
1649 pad(file, 16);
1650 if (devinfo->gen >= 7) {
1651 format(file, "JIP: %d", brw_inst_jip(devinfo, inst));
1652 } else {
1653 format(file, "JIP: %d", brw_inst_gen6_jump_count(devinfo, inst));
1654 }
1655 } else if (devinfo->gen < 6 && (opcode == BRW_OPCODE_BREAK ||
1656 opcode == BRW_OPCODE_CONTINUE ||
1657 opcode == BRW_OPCODE_ELSE)) {
1658 pad(file, 16);
1659 format(file, "Jump: %d", brw_inst_gen4_jump_count(devinfo, inst));
1660 pad(file, 32);
1661 format(file, "Pop: %"PRIu64, brw_inst_gen4_pop_count(devinfo, inst));
1662 } else if (devinfo->gen < 6 && (opcode == BRW_OPCODE_IF ||
1663 opcode == BRW_OPCODE_IFF ||
1664 opcode == BRW_OPCODE_HALT)) {
1665 pad(file, 16);
1666 format(file, "Jump: %d", brw_inst_gen4_jump_count(devinfo, inst));
1667 } else if (devinfo->gen < 6 && opcode == BRW_OPCODE_ENDIF) {
1668 pad(file, 16);
1669 format(file, "Pop: %"PRIu64, brw_inst_gen4_pop_count(devinfo, inst));
1670 } else if (opcode == BRW_OPCODE_JMPI) {
1671 pad(file, 16);
1672 err |= src1(file, devinfo, inst);
1673 } else if (desc && desc->nsrc == 3) {
1674 pad(file, 16);
1675 err |= dest_3src(file, devinfo, inst);
1676
1677 pad(file, 32);
1678 err |= src0_3src(file, devinfo, inst);
1679
1680 pad(file, 48);
1681 err |= src1_3src(file, devinfo, inst);
1682
1683 pad(file, 64);
1684 err |= src2_3src(file, devinfo, inst);
1685 } else if (desc) {
1686 if (desc->ndst > 0) {
1687 pad(file, 16);
1688 err |= dest(file, devinfo, inst);
1689 }
1690
1691 if (desc->nsrc > 0) {
1692 pad(file, 32);
1693 err |= src0(file, devinfo, inst);
1694 }
1695
1696 if (desc->nsrc > 1) {
1697 pad(file, 48);
1698 err |= src1(file, devinfo, inst);
1699 }
1700 }
1701
1702 if (is_send(opcode)) {
1703 enum brw_message_target sfid = brw_inst_sfid(devinfo, inst);
1704
1705 bool has_imm_desc = false, has_imm_ex_desc = false;
1706 uint32_t imm_desc = 0, imm_ex_desc = 0;
1707 if (is_split_send(devinfo, opcode)) {
1708 pad(file, 64);
1709 if (brw_inst_send_sel_reg32_desc(devinfo, inst)) {
1710 /* show the indirect descriptor source */
1711 err |= src_sends_ia(file, devinfo, BRW_REGISTER_TYPE_UD, 0, 0);
1712 } else {
1713 has_imm_desc = true;
1714 imm_desc = brw_inst_send_desc(devinfo, inst);
1715 fprintf(file, "0x%08"PRIx32, imm_desc);
1716 }
1717
1718 pad(file, 80);
1719 if (brw_inst_send_sel_reg32_ex_desc(devinfo, inst)) {
1720 /* show the indirect descriptor source */
1721 err |= src_sends_ia(file, devinfo, BRW_REGISTER_TYPE_UD, 0,
1722 brw_inst_send_ex_desc_ia_subreg_nr(devinfo, inst));
1723 } else {
1724 has_imm_ex_desc = true;
1725 imm_ex_desc = brw_inst_send_ex_desc(devinfo, inst);
1726 fprintf(file, "0x%08"PRIx32, imm_ex_desc);
1727 }
1728 } else {
1729 if (brw_inst_src1_reg_file(devinfo, inst) != BRW_IMMEDIATE_VALUE) {
1730 /* show the indirect descriptor source */
1731 pad(file, 48);
1732 err |= src1(file, devinfo, inst);
1733 pad(file, 64);
1734 } else {
1735 has_imm_desc = true;
1736 imm_desc = brw_inst_send_desc(devinfo, inst);
1737 pad(file, 48);
1738 }
1739
1740 /* Print message descriptor as immediate source */
1741 fprintf(file, "0x%08"PRIx64, inst->data[1] >> 32);
1742 }
1743
1744 newline(file);
1745 pad(file, 16);
1746 space = 0;
1747
1748 fprintf(file, " ");
1749 err |= control(file, "SFID", devinfo->gen >= 6 ? gen6_sfid : gen4_sfid,
1750 sfid, &space);
1751 string(file, " MsgDesc:");
1752
1753 if (!has_imm_desc) {
1754 format(file, " indirect");
1755 } else {
1756 switch (sfid) {
1757 case BRW_SFID_MATH:
1758 err |= control(file, "math function", math_function,
1759 brw_inst_math_msg_function(devinfo, inst), &space);
1760 err |= control(file, "math saturate", math_saturate,
1761 brw_inst_math_msg_saturate(devinfo, inst), &space);
1762 err |= control(file, "math signed", math_signed,
1763 brw_inst_math_msg_signed_int(devinfo, inst), &space);
1764 err |= control(file, "math scalar", math_scalar,
1765 brw_inst_math_msg_data_type(devinfo, inst), &space);
1766 err |= control(file, "math precision", math_precision,
1767 brw_inst_math_msg_precision(devinfo, inst), &space);
1768 break;
1769 case BRW_SFID_SAMPLER:
1770 if (devinfo->gen >= 5) {
1771 err |= control(file, "sampler message", gen5_sampler_msg_type,
1772 brw_sampler_desc_msg_type(devinfo, imm_desc),
1773 &space);
1774 err |= control(file, "sampler simd mode", gen5_sampler_simd_mode,
1775 brw_sampler_desc_simd_mode(devinfo, imm_desc),
1776 &space);
1777 format(file, " Surface = %u Sampler = %u",
1778 brw_sampler_desc_binding_table_index(devinfo, imm_desc),
1779 brw_sampler_desc_sampler(devinfo, imm_desc));
1780 } else {
1781 format(file, " (%u, %u, %u, ",
1782 brw_sampler_desc_binding_table_index(devinfo, imm_desc),
1783 brw_sampler_desc_sampler(devinfo, imm_desc),
1784 brw_sampler_desc_msg_type(devinfo, imm_desc));
1785 if (!devinfo->is_g4x) {
1786 err |= control(file, "sampler target format",
1787 sampler_target_format,
1788 brw_sampler_desc_return_format(devinfo, imm_desc),
1789 NULL);
1790 }
1791 string(file, ")");
1792 }
1793 break;
1794 case GEN6_SFID_DATAPORT_SAMPLER_CACHE:
1795 case GEN6_SFID_DATAPORT_CONSTANT_CACHE:
1796 /* aka BRW_SFID_DATAPORT_READ on Gen4-5 */
1797 if (devinfo->gen >= 6) {
1798 format(file, " (%u, %u, %u, %u)",
1799 brw_dp_desc_binding_table_index(devinfo, imm_desc),
1800 brw_dp_desc_msg_control(devinfo, imm_desc),
1801 brw_dp_desc_msg_type(devinfo, imm_desc),
1802 devinfo->gen >= 7 ? 0u :
1803 brw_dp_write_desc_write_commit(devinfo, imm_desc));
1804 } else {
1805 bool is_965 = devinfo->gen == 4 && !devinfo->is_g4x;
1806 err |= control(file, "DP read message type",
1807 is_965 ? gen4_dp_read_port_msg_type :
1808 g45_dp_read_port_msg_type,
1809 brw_dp_read_desc_msg_type(devinfo, imm_desc),
1810 &space);
1811
1812 format(file, " MsgCtrl = 0x%u",
1813 brw_dp_read_desc_msg_control(devinfo, imm_desc));
1814
1815 format(file, " Surface = %u",
1816 brw_dp_desc_binding_table_index(devinfo, imm_desc));
1817 }
1818 break;
1819
1820 case GEN6_SFID_DATAPORT_RENDER_CACHE: {
1821 /* aka BRW_SFID_DATAPORT_WRITE on Gen4-5 */
1822 unsigned msg_type = brw_dp_write_desc_msg_type(devinfo, imm_desc);
1823
1824 err |= control(file, "DP rc message type",
1825 dp_rc_msg_type(devinfo), msg_type, &space);
1826
1827 bool is_rt_write = msg_type ==
1828 (devinfo->gen >= 6 ? GEN6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE
1829 : BRW_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE);
1830
1831 if (is_rt_write) {
1832 err |= control(file, "RT message type", m_rt_write_subtype,
1833 brw_inst_rt_message_type(devinfo, inst), &space);
1834 if (devinfo->gen >= 6 && brw_inst_rt_slot_group(devinfo, inst))
1835 string(file, " Hi");
1836 if (brw_dp_write_desc_last_render_target(devinfo, imm_desc))
1837 string(file, " LastRT");
1838 if (devinfo->gen < 7 &&
1839 brw_dp_write_desc_write_commit(devinfo, imm_desc))
1840 string(file, " WriteCommit");
1841 } else {
1842 format(file, " MsgCtrl = 0x%u",
1843 brw_dp_write_desc_msg_control(devinfo, imm_desc));
1844 }
1845
1846 format(file, " Surface = %u",
1847 brw_dp_desc_binding_table_index(devinfo, imm_desc));
1848 break;
1849 }
1850
1851 case BRW_SFID_URB: {
1852 unsigned opcode = brw_inst_urb_opcode(devinfo, inst);
1853
1854 format(file, " %"PRIu64, brw_inst_urb_global_offset(devinfo, inst));
1855
1856 space = 1;
1857
1858 err |= control(file, "urb opcode",
1859 devinfo->gen >= 7 ? gen7_urb_opcode
1860 : gen5_urb_opcode,
1861 opcode, &space);
1862
1863 if (devinfo->gen >= 7 &&
1864 brw_inst_urb_per_slot_offset(devinfo, inst)) {
1865 string(file, " per-slot");
1866 }
1867
1868 if (opcode == GEN8_URB_OPCODE_SIMD8_WRITE ||
1869 opcode == GEN8_URB_OPCODE_SIMD8_READ) {
1870 if (brw_inst_urb_channel_mask_present(devinfo, inst))
1871 string(file, " masked");
1872 } else {
1873 err |= control(file, "urb swizzle", urb_swizzle,
1874 brw_inst_urb_swizzle_control(devinfo, inst),
1875 &space);
1876 }
1877
1878 if (devinfo->gen < 7) {
1879 err |= control(file, "urb allocate", urb_allocate,
1880 brw_inst_urb_allocate(devinfo, inst), &space);
1881 err |= control(file, "urb used", urb_used,
1882 brw_inst_urb_used(devinfo, inst), &space);
1883 }
1884 if (devinfo->gen < 8) {
1885 err |= control(file, "urb complete", urb_complete,
1886 brw_inst_urb_complete(devinfo, inst), &space);
1887 }
1888 break;
1889 }
1890 case BRW_SFID_THREAD_SPAWNER:
1891 break;
1892
1893 case BRW_SFID_MESSAGE_GATEWAY:
1894 format(file, " (%s)",
1895 gen7_gateway_subfuncid[brw_inst_gateway_subfuncid(devinfo, inst)]);
1896 break;
1897
1898 case GEN7_SFID_DATAPORT_DATA_CACHE:
1899 if (devinfo->gen >= 7) {
1900 format(file, " (");
1901
1902 err |= control(file, "DP DC0 message type",
1903 dp_dc0_msg_type_gen7,
1904 brw_dp_desc_msg_type(devinfo, imm_desc), &space);
1905
1906 format(file, ", %u, ",
1907 brw_dp_desc_binding_table_index(devinfo, imm_desc));
1908
1909 switch (brw_inst_dp_msg_type(devinfo, inst)) {
1910 case GEN7_DATAPORT_DC_UNTYPED_ATOMIC_OP:
1911 control(file, "atomic op", aop,
1912 brw_dp_desc_msg_control(devinfo, imm_desc) & 0xf,
1913 &space);
1914 break;
1915 default:
1916 format(file, "%u",
1917 brw_dp_desc_msg_control(devinfo, imm_desc));
1918 }
1919 format(file, ")");
1920 break;
1921 }
1922 /* FALLTHROUGH */
1923
1924 case HSW_SFID_DATAPORT_DATA_CACHE_1: {
1925 if (devinfo->gen >= 7) {
1926 format(file, " (");
1927
1928 unsigned msg_ctrl = brw_dp_desc_msg_control(devinfo, imm_desc);
1929
1930 err |= control(file, "DP DC1 message type",
1931 dp_dc1_msg_type_hsw,
1932 brw_dp_desc_msg_type(devinfo, imm_desc), &space);
1933
1934 format(file, ", Surface = %u, ",
1935 brw_dp_desc_binding_table_index(devinfo, imm_desc));
1936
1937 switch (brw_inst_dp_msg_type(devinfo, inst)) {
1938 case HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP:
1939 case HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP:
1940 case HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP:
1941 format(file, "SIMD%d,", (msg_ctrl & (1 << 4)) ? 8 : 16);
1942 /* fallthrough */
1943 case HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP_SIMD4X2:
1944 case HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2:
1945 case HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP_SIMD4X2:
1946 case GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_OP:
1947 control(file, "atomic op", aop, msg_ctrl & 0xf, &space);
1948 break;
1949 case HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_READ:
1950 case HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_WRITE:
1951 case HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ:
1952 case HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE:
1953 case GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_WRITE:
1954 case GEN8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_READ: {
1955 static const char *simd_modes[] = { "4x2", "16", "8" };
1956 format(file, "SIMD%s, Mask = 0x%x",
1957 simd_modes[msg_ctrl >> 4], msg_ctrl & 0xf);
1958 break;
1959 }
1960 case GEN9_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_FLOAT_OP:
1961 case GEN9_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_FLOAT_OP:
1962 format(file, "SIMD%d,", (msg_ctrl & (1 << 4)) ? 8 : 16);
1963 control(file, "atomic float op", aop_float, msg_ctrl & 0xf,
1964 &space);
1965 break;
1966 default:
1967 format(file, "0x%x", msg_ctrl);
1968 }
1969 format(file, ")");
1970 break;
1971 }
1972 /* FALLTHROUGH */
1973 }
1974
1975 case GEN7_SFID_PIXEL_INTERPOLATOR:
1976 if (devinfo->gen >= 7) {
1977 format(file, " (%s, %s, 0x%02"PRIx64")",
1978 brw_inst_pi_nopersp(devinfo, inst) ? "linear" : "persp",
1979 pixel_interpolator_msg_types[brw_inst_pi_message_type(devinfo, inst)],
1980 brw_inst_pi_message_data(devinfo, inst));
1981 break;
1982 }
1983 /* FALLTHROUGH */
1984
1985 default:
1986 format(file, "unsupported shared function ID %d", sfid);
1987 break;
1988 }
1989
1990 if (space)
1991 string(file, " ");
1992 }
1993 if (has_imm_desc)
1994 format(file, "mlen %u", brw_message_desc_mlen(devinfo, imm_desc));
1995 if (has_imm_ex_desc) {
1996 format(file, " ex_mlen %u",
1997 brw_message_ex_desc_ex_mlen(devinfo, imm_ex_desc));
1998 }
1999 if (has_imm_desc)
2000 format(file, " rlen %u", brw_message_desc_rlen(devinfo, imm_desc));
2001 }
2002 pad(file, 64);
2003 if (opcode != BRW_OPCODE_NOP && opcode != BRW_OPCODE_NENOP) {
2004 string(file, "{");
2005 space = 1;
2006 err |= control(file, "access mode", access_mode,
2007 brw_inst_access_mode(devinfo, inst), &space);
2008 if (devinfo->gen >= 6) {
2009 err |= control(file, "write enable control", wectrl,
2010 brw_inst_mask_control(devinfo, inst), &space);
2011 } else {
2012 err |= control(file, "mask control", mask_ctrl,
2013 brw_inst_mask_control(devinfo, inst), &space);
2014 }
2015 err |= control(file, "dependency control", dep_ctrl,
2016 ((brw_inst_no_dd_check(devinfo, inst) << 1) |
2017 brw_inst_no_dd_clear(devinfo, inst)), &space);
2018
2019 if (devinfo->gen >= 6)
2020 err |= qtr_ctrl(file, devinfo, inst);
2021 else {
2022 if (brw_inst_qtr_control(devinfo, inst) == BRW_COMPRESSION_COMPRESSED &&
2023 desc && desc->ndst > 0 &&
2024 brw_inst_dst_reg_file(devinfo, inst) == BRW_MESSAGE_REGISTER_FILE &&
2025 brw_inst_dst_da_reg_nr(devinfo, inst) & BRW_MRF_COMPR4) {
2026 format(file, " compr4");
2027 } else {
2028 err |= control(file, "compression control", compr_ctrl,
2029 brw_inst_qtr_control(devinfo, inst), &space);
2030 }
2031 }
2032
2033 err |= control(file, "compaction", cmpt_ctrl, is_compacted, &space);
2034 err |= control(file, "thread control", thread_ctrl,
2035 brw_inst_thread_control(devinfo, inst), &space);
2036 if (has_branch_ctrl(devinfo, opcode)) {
2037 err |= control(file, "branch ctrl", branch_ctrl,
2038 brw_inst_branch_control(devinfo, inst), &space);
2039 } else if (devinfo->gen >= 6) {
2040 err |= control(file, "acc write control", accwr,
2041 brw_inst_acc_wr_control(devinfo, inst), &space);
2042 }
2043 if (is_send(opcode))
2044 err |= control(file, "end of thread", end_of_thread,
2045 brw_inst_eot(devinfo, inst), &space);
2046 if (space)
2047 string(file, " ");
2048 string(file, "}");
2049 }
2050 string(file, ";");
2051 newline(file);
2052 return err;
2053 }