1 /*
2 * Copyright © 2008 Keith Packard
3 * Copyright © 2014 Intel Corporation
4 *
5 * Permission to use, copy, modify, distribute, and sell this software and its
6 * documentation for any purpose is hereby granted without fee, provided that
7 * the above copyright notice appear in all copies and that both that copyright
8 * notice and this permission notice appear in supporting documentation, and
9 * that the name of the copyright holders not be used in advertising or
10 * publicity pertaining to distribution of the software without specific,
11 * written prior permission. The copyright holders make no representations
12 * about the suitability of this software for any purpose. It is provided "as
13 * is" without express or implied warranty.
14 *
15 * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
16 * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS, IN NO
17 * EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY SPECIAL, INDIRECT OR
18 * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
19 * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
20 * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
21 * OF THIS SOFTWARE.
22 */
23
24 #include <stdarg.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28
29 #include "elk_disasm.h"
30 #include "elk_disasm_info.h"
31 #include "elk_eu_defines.h"
32 #include "elk_eu.h"
33 #include "elk_inst.h"
34 #include "elk_isa_info.h"
35 #include "elk_reg.h"
36 #include "elk_shader.h"
37 #include "util/half_float.h"
38
39 bool
elk_has_jip(const struct intel_device_info * devinfo,enum elk_opcode opcode)40 elk_has_jip(const struct intel_device_info *devinfo, enum elk_opcode opcode)
41 {
42 if (devinfo->ver < 6)
43 return false;
44
45 return opcode == ELK_OPCODE_IF ||
46 opcode == ELK_OPCODE_ELSE ||
47 opcode == ELK_OPCODE_ENDIF ||
48 opcode == ELK_OPCODE_WHILE ||
49 opcode == ELK_OPCODE_BREAK ||
50 opcode == ELK_OPCODE_CONTINUE ||
51 opcode == ELK_OPCODE_HALT;
52 }
53
54 bool
elk_has_uip(const struct intel_device_info * devinfo,enum elk_opcode opcode)55 elk_has_uip(const struct intel_device_info *devinfo, enum elk_opcode opcode)
56 {
57 if (devinfo->ver < 6)
58 return false;
59
60 return (devinfo->ver >= 7 && opcode == ELK_OPCODE_IF) ||
61 (devinfo->ver >= 8 && opcode == ELK_OPCODE_ELSE) ||
62 opcode == ELK_OPCODE_BREAK ||
63 opcode == ELK_OPCODE_CONTINUE ||
64 opcode == ELK_OPCODE_HALT;
65 }
66
67 static bool
has_branch_ctrl(const struct intel_device_info * devinfo,enum elk_opcode opcode)68 has_branch_ctrl(const struct intel_device_info *devinfo, enum elk_opcode opcode)
69 {
70 if (devinfo->ver < 8)
71 return false;
72
73 return opcode == ELK_OPCODE_IF ||
74 opcode == ELK_OPCODE_ELSE;
75 /* opcode == ELK_OPCODE_GOTO; */
76 }
77
78 static bool
is_logic_instruction(unsigned opcode)79 is_logic_instruction(unsigned opcode)
80 {
81 return opcode == ELK_OPCODE_AND ||
82 opcode == ELK_OPCODE_NOT ||
83 opcode == ELK_OPCODE_OR ||
84 opcode == ELK_OPCODE_XOR;
85 }
86
87 static bool
is_send(unsigned opcode)88 is_send(unsigned opcode)
89 {
90 return opcode == ELK_OPCODE_SEND ||
91 opcode == ELK_OPCODE_SENDC ||
92 opcode == ELK_OPCODE_SENDS ||
93 opcode == ELK_OPCODE_SENDSC;
94 }
95
96 static bool
is_split_send(UNUSED const struct intel_device_info * devinfo,unsigned opcode)97 is_split_send(UNUSED const struct intel_device_info *devinfo, unsigned opcode)
98 {
99 if (devinfo->ver >= 12)
100 return is_send(opcode);
101 else
102 return opcode == ELK_OPCODE_SENDS ||
103 opcode == ELK_OPCODE_SENDSC;
104 }
105
106 const char *const elk_conditional_modifier[16] = {
107 [ELK_CONDITIONAL_NONE] = "",
108 [ELK_CONDITIONAL_Z] = ".z",
109 [ELK_CONDITIONAL_NZ] = ".nz",
110 [ELK_CONDITIONAL_G] = ".g",
111 [ELK_CONDITIONAL_GE] = ".ge",
112 [ELK_CONDITIONAL_L] = ".l",
113 [ELK_CONDITIONAL_LE] = ".le",
114 [ELK_CONDITIONAL_R] = ".r",
115 [ELK_CONDITIONAL_O] = ".o",
116 [ELK_CONDITIONAL_U] = ".u",
117 };
118
119 static const char *const m_negate[2] = {
120 [0] = "",
121 [1] = "-",
122 };
123
124 static const char *const _abs[2] = {
125 [0] = "",
126 [1] = "(abs)",
127 };
128
129 static const char *const m_bitnot[2] = { "", "~" };
130
131 static const char *const vert_stride[16] = {
132 [0] = "0",
133 [1] = "1",
134 [2] = "2",
135 [3] = "4",
136 [4] = "8",
137 [5] = "16",
138 [6] = "32",
139 [15] = "VxH",
140 };
141
142 static const char *const width[8] = {
143 [0] = "1",
144 [1] = "2",
145 [2] = "4",
146 [3] = "8",
147 [4] = "16",
148 };
149
150 static const char *const horiz_stride[4] = {
151 [0] = "0",
152 [1] = "1",
153 [2] = "2",
154 [3] = "4"
155 };
156
157 static const char *const chan_sel[4] = {
158 [0] = "x",
159 [1] = "y",
160 [2] = "z",
161 [3] = "w",
162 };
163
164 static const char *const debug_ctrl[2] = {
165 [0] = "",
166 [1] = ".breakpoint"
167 };
168
169 static const char *const saturate[2] = {
170 [0] = "",
171 [1] = ".sat"
172 };
173
174 static const char *const cmpt_ctrl[2] = {
175 [0] = "",
176 [1] = "compacted"
177 };
178
179 static const char *const accwr[2] = {
180 [0] = "",
181 [1] = "AccWrEnable"
182 };
183
184 static const char *const branch_ctrl[2] = {
185 [0] = "",
186 [1] = "BranchCtrl"
187 };
188
189 static const char *const wectrl[2] = {
190 [0] = "",
191 [1] = "WE_all"
192 };
193
194 static const char *const exec_size[8] = {
195 [0] = "1",
196 [1] = "2",
197 [2] = "4",
198 [3] = "8",
199 [4] = "16",
200 [5] = "32"
201 };
202
203 static const char *const pred_inv[2] = {
204 [0] = "+",
205 [1] = "-"
206 };
207
208 const char *const elk_pred_ctrl_align16[16] = {
209 [1] = "",
210 [2] = ".x",
211 [3] = ".y",
212 [4] = ".z",
213 [5] = ".w",
214 [6] = ".any4h",
215 [7] = ".all4h",
216 };
217
218 static const char *const pred_ctrl_align1[16] = {
219 [ELK_PREDICATE_NORMAL] = "",
220 [ELK_PREDICATE_ALIGN1_ANYV] = ".anyv",
221 [ELK_PREDICATE_ALIGN1_ALLV] = ".allv",
222 [ELK_PREDICATE_ALIGN1_ANY2H] = ".any2h",
223 [ELK_PREDICATE_ALIGN1_ALL2H] = ".all2h",
224 [ELK_PREDICATE_ALIGN1_ANY4H] = ".any4h",
225 [ELK_PREDICATE_ALIGN1_ALL4H] = ".all4h",
226 [ELK_PREDICATE_ALIGN1_ANY8H] = ".any8h",
227 [ELK_PREDICATE_ALIGN1_ALL8H] = ".all8h",
228 [ELK_PREDICATE_ALIGN1_ANY16H] = ".any16h",
229 [ELK_PREDICATE_ALIGN1_ALL16H] = ".all16h",
230 [ELK_PREDICATE_ALIGN1_ANY32H] = ".any32h",
231 [ELK_PREDICATE_ALIGN1_ALL32H] = ".all32h",
232 };
233
234 static const char *const xe2_pred_ctrl[4] = {
235 [ELK_PREDICATE_NORMAL] = "",
236 [XE2_PREDICATE_ANY] = ".any",
237 [XE2_PREDICATE_ALL] = ".all",
238 };
239
240 static const char *const thread_ctrl[4] = {
241 [ELK_THREAD_NORMAL] = "",
242 [ELK_THREAD_ATOMIC] = "atomic",
243 [ELK_THREAD_SWITCH] = "switch",
244 };
245
246 static const char *const compr_ctrl[4] = {
247 [0] = "",
248 [1] = "sechalf",
249 [2] = "compr",
250 [3] = "compr4",
251 };
252
253 static const char *const dep_ctrl[4] = {
254 [0] = "",
255 [1] = "NoDDClr",
256 [2] = "NoDDChk",
257 [3] = "NoDDClr,NoDDChk",
258 };
259
260 static const char *const mask_ctrl[4] = {
261 [0] = "",
262 [1] = "nomask",
263 };
264
265 static const char *const access_mode[2] = {
266 [0] = "align1",
267 [1] = "align16",
268 };
269
270 static const char *const reg_file[4] = {
271 [0] = "A",
272 [1] = "g",
273 [2] = "m",
274 [3] = "imm",
275 };
276
277 static const char *const writemask[16] = {
278 [0x0] = ".",
279 [0x1] = ".x",
280 [0x2] = ".y",
281 [0x3] = ".xy",
282 [0x4] = ".z",
283 [0x5] = ".xz",
284 [0x6] = ".yz",
285 [0x7] = ".xyz",
286 [0x8] = ".w",
287 [0x9] = ".xw",
288 [0xa] = ".yw",
289 [0xb] = ".xyw",
290 [0xc] = ".zw",
291 [0xd] = ".xzw",
292 [0xe] = ".yzw",
293 [0xf] = "",
294 };
295
296 static const char *const end_of_thread[2] = {
297 [0] = "",
298 [1] = "EOT"
299 };
300
301 /* SFIDs on Gfx4-5 */
302 static const char *const gfx4_sfid[16] = {
303 [ELK_SFID_NULL] = "null",
304 [ELK_SFID_MATH] = "math",
305 [ELK_SFID_SAMPLER] = "sampler",
306 [ELK_SFID_MESSAGE_GATEWAY] = "gateway",
307 [ELK_SFID_DATAPORT_READ] = "read",
308 [ELK_SFID_DATAPORT_WRITE] = "write",
309 [ELK_SFID_URB] = "urb",
310 [ELK_SFID_THREAD_SPAWNER] = "thread_spawner",
311 [ELK_SFID_VME] = "vme",
312 };
313
314 static const char *const gfx6_sfid[16] = {
315 [ELK_SFID_NULL] = "null",
316 [ELK_SFID_MATH] = "math",
317 [ELK_SFID_SAMPLER] = "sampler",
318 [ELK_SFID_MESSAGE_GATEWAY] = "gateway",
319 [ELK_SFID_URB] = "urb",
320 [ELK_SFID_THREAD_SPAWNER] = "thread_spawner",
321 [GFX6_SFID_DATAPORT_SAMPLER_CACHE] = "dp_sampler",
322 [GFX6_SFID_DATAPORT_RENDER_CACHE] = "render",
323 [GFX6_SFID_DATAPORT_CONSTANT_CACHE] = "const",
324 [GFX7_SFID_DATAPORT_DATA_CACHE] = "data",
325 [GFX7_SFID_PIXEL_INTERPOLATOR] = "pixel interp",
326 [HSW_SFID_DATAPORT_DATA_CACHE_1] = "dp data 1",
327 [HSW_SFID_CRE] = "cre",
328 [GFX12_SFID_SLM] = "slm",
329 [GFX12_SFID_TGM] = "tgm",
330 [GFX12_SFID_UGM] = "ugm",
331 };
332
333 static const char *const gfx7_gateway_subfuncid[8] = {
334 [ELK_MESSAGE_GATEWAY_SFID_OPEN_GATEWAY] = "open",
335 [ELK_MESSAGE_GATEWAY_SFID_CLOSE_GATEWAY] = "close",
336 [ELK_MESSAGE_GATEWAY_SFID_FORWARD_MSG] = "forward msg",
337 [ELK_MESSAGE_GATEWAY_SFID_GET_TIMESTAMP] = "get timestamp",
338 [ELK_MESSAGE_GATEWAY_SFID_BARRIER_MSG] = "barrier msg",
339 [ELK_MESSAGE_GATEWAY_SFID_UPDATE_GATEWAY_STATE] = "update state",
340 [ELK_MESSAGE_GATEWAY_SFID_MMIO_READ_WRITE] = "mmio read/write",
341 };
342
343 static const char *const gfx4_dp_read_port_msg_type[4] = {
344 [0b00] = "OWord Block Read",
345 [0b01] = "OWord Dual Block Read",
346 [0b10] = "Media Block Read",
347 [0b11] = "DWord Scattered Read",
348 };
349
350 static const char *const g45_dp_read_port_msg_type[8] = {
351 [0b000] = "OWord Block Read",
352 [0b010] = "OWord Dual Block Read",
353 [0b100] = "Media Block Read",
354 [0b110] = "DWord Scattered Read",
355 [0b001] = "Render Target UNORM Read",
356 [0b011] = "AVC Loop Filter Read",
357 };
358
359 static const char *const dp_write_port_msg_type[8] = {
360 [0b000] = "OWord block write",
361 [0b001] = "OWord dual block write",
362 [0b010] = "media block write",
363 [0b011] = "DWord scattered write",
364 [0b100] = "RT write",
365 [0b101] = "streamed VB write",
366 [0b110] = "RT UNORM write", /* G45+ */
367 [0b111] = "flush render cache",
368 };
369
370 static const char *const dp_rc_msg_type_gfx6[16] = {
371 [ELK_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ] = "OWORD block read",
372 [GFX6_DATAPORT_READ_MESSAGE_RENDER_UNORM_READ] = "RT UNORM read",
373 [GFX6_DATAPORT_READ_MESSAGE_OWORD_DUAL_BLOCK_READ] = "OWORD dual block read",
374 [GFX6_DATAPORT_READ_MESSAGE_MEDIA_BLOCK_READ] = "media block read",
375 [GFX6_DATAPORT_READ_MESSAGE_OWORD_UNALIGN_BLOCK_READ] =
376 "OWORD unaligned block read",
377 [GFX6_DATAPORT_READ_MESSAGE_DWORD_SCATTERED_READ] = "DWORD scattered read",
378 [GFX6_DATAPORT_WRITE_MESSAGE_DWORD_ATOMIC_WRITE] = "DWORD atomic write",
379 [GFX6_DATAPORT_WRITE_MESSAGE_OWORD_BLOCK_WRITE] = "OWORD block write",
380 [GFX6_DATAPORT_WRITE_MESSAGE_OWORD_DUAL_BLOCK_WRITE] =
381 "OWORD dual block write",
382 [GFX6_DATAPORT_WRITE_MESSAGE_MEDIA_BLOCK_WRITE] = "media block write",
383 [GFX6_DATAPORT_WRITE_MESSAGE_DWORD_SCATTERED_WRITE] =
384 "DWORD scattered write",
385 [GFX6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE] = "RT write",
386 [GFX6_DATAPORT_WRITE_MESSAGE_STREAMED_VB_WRITE] = "streamed VB write",
387 [GFX6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_UNORM_WRITE] = "RT UNORM write",
388 };
389
390 static const char *const dp_rc_msg_type_gfx7[16] = {
391 [GFX7_DATAPORT_RC_MEDIA_BLOCK_READ] = "media block read",
392 [GFX7_DATAPORT_RC_TYPED_SURFACE_READ] = "typed surface read",
393 [GFX7_DATAPORT_RC_TYPED_ATOMIC_OP] = "typed atomic op",
394 [GFX7_DATAPORT_RC_MEMORY_FENCE] = "memory fence",
395 [GFX7_DATAPORT_RC_MEDIA_BLOCK_WRITE] = "media block write",
396 [GFX7_DATAPORT_RC_RENDER_TARGET_WRITE] = "RT write",
397 [GFX7_DATAPORT_RC_TYPED_SURFACE_WRITE] = "typed surface write"
398 };
399
400 static const char *const dp_rc_msg_type_gfx9[16] = {
401 [GFX9_DATAPORT_RC_RENDER_TARGET_WRITE] = "RT write",
402 [GFX9_DATAPORT_RC_RENDER_TARGET_READ] = "RT read"
403 };
404
405 static const char *const *
dp_rc_msg_type(const struct intel_device_info * devinfo)406 dp_rc_msg_type(const struct intel_device_info *devinfo)
407 {
408 return (devinfo->ver >= 9 ? dp_rc_msg_type_gfx9 :
409 devinfo->ver >= 7 ? dp_rc_msg_type_gfx7 :
410 devinfo->ver >= 6 ? dp_rc_msg_type_gfx6 :
411 dp_write_port_msg_type);
412 }
413
414 static const char *const m_rt_write_subtype[] = {
415 [0b000] = "SIMD16",
416 [0b001] = "SIMD16/RepData",
417 [0b010] = "SIMD8/DualSrcLow",
418 [0b011] = "SIMD8/DualSrcHigh",
419 [0b100] = "SIMD8",
420 [0b101] = "SIMD8/ImageWrite", /* Gfx6+ */
421 [0b111] = "SIMD16/RepData-111", /* no idea how this is different than 1 */
422 };
423
424 static const char *const dp_dc0_msg_type_gfx7[16] = {
425 [GFX7_DATAPORT_DC_OWORD_BLOCK_READ] = "DC OWORD block read",
426 [GFX7_DATAPORT_DC_UNALIGNED_OWORD_BLOCK_READ] =
427 "DC unaligned OWORD block read",
428 [GFX7_DATAPORT_DC_OWORD_DUAL_BLOCK_READ] = "DC OWORD dual block read",
429 [GFX7_DATAPORT_DC_DWORD_SCATTERED_READ] = "DC DWORD scattered read",
430 [GFX7_DATAPORT_DC_BYTE_SCATTERED_READ] = "DC byte scattered read",
431 [GFX7_DATAPORT_DC_UNTYPED_SURFACE_READ] = "DC untyped surface read",
432 [GFX7_DATAPORT_DC_UNTYPED_ATOMIC_OP] = "DC untyped atomic",
433 [GFX7_DATAPORT_DC_MEMORY_FENCE] = "DC mfence",
434 [GFX7_DATAPORT_DC_OWORD_BLOCK_WRITE] = "DC OWORD block write",
435 [GFX7_DATAPORT_DC_OWORD_DUAL_BLOCK_WRITE] = "DC OWORD dual block write",
436 [GFX7_DATAPORT_DC_DWORD_SCATTERED_WRITE] = "DC DWORD scatterd write",
437 [GFX7_DATAPORT_DC_BYTE_SCATTERED_WRITE] = "DC byte scattered write",
438 [GFX7_DATAPORT_DC_UNTYPED_SURFACE_WRITE] = "DC untyped surface write",
439 };
440
441 static const char *const dp_oword_block_rw[8] = {
442 [ELK_DATAPORT_OWORD_BLOCK_1_OWORDLOW] = "1-low",
443 [ELK_DATAPORT_OWORD_BLOCK_1_OWORDHIGH] = "1-high",
444 [ELK_DATAPORT_OWORD_BLOCK_2_OWORDS] = "2",
445 [ELK_DATAPORT_OWORD_BLOCK_4_OWORDS] = "4",
446 [ELK_DATAPORT_OWORD_BLOCK_8_OWORDS] = "8",
447 };
448
449 static const char *const dp_dc1_msg_type_hsw[32] = {
450 [HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_READ] = "untyped surface read",
451 [HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP] = "DC untyped atomic op",
452 [HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP_SIMD4X2] =
453 "DC untyped 4x2 atomic op",
454 [HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_READ] = "DC media block read",
455 [HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ] = "DC typed surface read",
456 [HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP] = "DC typed atomic",
457 [HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2] = "DC typed 4x2 atomic op",
458 [HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_WRITE] = "DC untyped surface write",
459 [HSW_DATAPORT_DC_PORT1_MEDIA_BLOCK_WRITE] = "DC media block write",
460 [HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP] = "DC atomic counter op",
461 [HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP_SIMD4X2] =
462 "DC 4x2 atomic counter op",
463 [HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE] = "DC typed surface write",
464 [GFX9_DATAPORT_DC_PORT1_A64_SCATTERED_READ] = "DC A64 scattered read",
465 [GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_READ] = "DC A64 untyped surface read",
466 [GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_OP] = "DC A64 untyped atomic op",
467 [GFX9_DATAPORT_DC_PORT1_A64_OWORD_BLOCK_READ] = "DC A64 oword block read",
468 [GFX9_DATAPORT_DC_PORT1_A64_OWORD_BLOCK_WRITE] = "DC A64 oword block write",
469 [GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_WRITE] = "DC A64 untyped surface write",
470 [GFX8_DATAPORT_DC_PORT1_A64_SCATTERED_WRITE] = "DC A64 scattered write",
471 [GFX9_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_FLOAT_OP] =
472 "DC untyped atomic float op",
473 [GFX9_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_FLOAT_OP] =
474 "DC A64 untyped atomic float op",
475 [GFX12_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_HALF_INT_OP] =
476 "DC A64 untyped atomic half-integer op",
477 [GFX12_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_HALF_FLOAT_OP] =
478 "DC A64 untyped atomic half-float op",
479 };
480
481 static const char *const aop[16] = {
482 [ELK_AOP_AND] = "and",
483 [ELK_AOP_OR] = "or",
484 [ELK_AOP_XOR] = "xor",
485 [ELK_AOP_MOV] = "mov",
486 [ELK_AOP_INC] = "inc",
487 [ELK_AOP_DEC] = "dec",
488 [ELK_AOP_ADD] = "add",
489 [ELK_AOP_SUB] = "sub",
490 [ELK_AOP_REVSUB] = "revsub",
491 [ELK_AOP_IMAX] = "imax",
492 [ELK_AOP_IMIN] = "imin",
493 [ELK_AOP_UMAX] = "umax",
494 [ELK_AOP_UMIN] = "umin",
495 [ELK_AOP_CMPWR] = "cmpwr",
496 [ELK_AOP_PREDEC] = "predec",
497 };
498
499 static const char *const aop_float[5] = {
500 [ELK_AOP_FMAX] = "fmax",
501 [ELK_AOP_FMIN] = "fmin",
502 [ELK_AOP_FCMPWR] = "fcmpwr",
503 [ELK_AOP_FADD] = "fadd",
504 };
505
506 static const char * const pixel_interpolator_msg_types[4] = {
507 [GFX7_PIXEL_INTERPOLATOR_LOC_SHARED_OFFSET] = "per_message_offset",
508 [GFX7_PIXEL_INTERPOLATOR_LOC_SAMPLE] = "sample_position",
509 [GFX7_PIXEL_INTERPOLATOR_LOC_CENTROID] = "centroid",
510 [GFX7_PIXEL_INTERPOLATOR_LOC_PER_SLOT_OFFSET] = "per_slot_offset",
511 };
512
513 static const char *const math_function[16] = {
514 [ELK_MATH_FUNCTION_INV] = "inv",
515 [ELK_MATH_FUNCTION_LOG] = "log",
516 [ELK_MATH_FUNCTION_EXP] = "exp",
517 [ELK_MATH_FUNCTION_SQRT] = "sqrt",
518 [ELK_MATH_FUNCTION_RSQ] = "rsq",
519 [ELK_MATH_FUNCTION_SIN] = "sin",
520 [ELK_MATH_FUNCTION_COS] = "cos",
521 [ELK_MATH_FUNCTION_SINCOS] = "sincos",
522 [ELK_MATH_FUNCTION_FDIV] = "fdiv",
523 [ELK_MATH_FUNCTION_POW] = "pow",
524 [ELK_MATH_FUNCTION_INT_DIV_QUOTIENT_AND_REMAINDER] = "intdivmod",
525 [ELK_MATH_FUNCTION_INT_DIV_QUOTIENT] = "intdiv",
526 [ELK_MATH_FUNCTION_INT_DIV_REMAINDER] = "intmod",
527 [GFX8_MATH_FUNCTION_INVM] = "invm",
528 [GFX8_MATH_FUNCTION_RSQRTM] = "rsqrtm",
529 };
530
531 static const char *const sync_function[16] = {
532 [TGL_SYNC_NOP] = "nop",
533 [TGL_SYNC_ALLRD] = "allrd",
534 [TGL_SYNC_ALLWR] = "allwr",
535 [TGL_SYNC_FENCE] = "fence",
536 [TGL_SYNC_BAR] = "bar",
537 [TGL_SYNC_HOST] = "host",
538 };
539
540 static const char *const math_saturate[2] = {
541 [0] = "",
542 [1] = "sat"
543 };
544
545 static const char *const math_signed[2] = {
546 [0] = "",
547 [1] = "signed"
548 };
549
550 static const char *const math_scalar[2] = {
551 [0] = "",
552 [1] = "scalar"
553 };
554
555 static const char *const math_precision[2] = {
556 [0] = "",
557 [1] = "partial_precision"
558 };
559
560 static const char *const gfx5_urb_opcode[] = {
561 [0] = "urb_write",
562 [1] = "ff_sync",
563 };
564
565 static const char *const gfx7_urb_opcode[] = {
566 [ELK_URB_OPCODE_WRITE_HWORD] = "write HWord",
567 [ELK_URB_OPCODE_WRITE_OWORD] = "write OWord",
568 [ELK_URB_OPCODE_READ_HWORD] = "read HWord",
569 [ELK_URB_OPCODE_READ_OWORD] = "read OWord",
570 [GFX7_URB_OPCODE_ATOMIC_MOV] = "atomic mov", /* Gfx7+ */
571 [GFX7_URB_OPCODE_ATOMIC_INC] = "atomic inc", /* Gfx7+ */
572 [GFX8_URB_OPCODE_ATOMIC_ADD] = "atomic add", /* Gfx8+ */
573 [GFX8_URB_OPCODE_SIMD8_WRITE] = "SIMD8 write", /* Gfx8+ */
574 [GFX8_URB_OPCODE_SIMD8_READ] = "SIMD8 read", /* Gfx8+ */
575 [GFX125_URB_OPCODE_FENCE] = "fence", /* Gfx12.5+ */
576 /* [10-15] - reserved */
577 };
578
579 static const char *const urb_swizzle[4] = {
580 [ELK_URB_SWIZZLE_NONE] = "",
581 [ELK_URB_SWIZZLE_INTERLEAVE] = "interleave",
582 [ELK_URB_SWIZZLE_TRANSPOSE] = "transpose",
583 };
584
585 static const char *const urb_allocate[2] = {
586 [0] = "",
587 [1] = "allocate"
588 };
589
590 static const char *const urb_used[2] = {
591 [0] = "",
592 [1] = "used"
593 };
594
595 static const char *const urb_complete[2] = {
596 [0] = "",
597 [1] = "complete"
598 };
599
600 static const char *const gfx5_sampler_msg_type[] = {
601 [GFX5_SAMPLER_MESSAGE_SAMPLE] = "sample",
602 [GFX5_SAMPLER_MESSAGE_SAMPLE_BIAS] = "sample_b",
603 [GFX5_SAMPLER_MESSAGE_SAMPLE_LOD] = "sample_l",
604 [GFX5_SAMPLER_MESSAGE_SAMPLE_COMPARE] = "sample_c",
605 [GFX5_SAMPLER_MESSAGE_SAMPLE_DERIVS] = "sample_d",
606 [GFX5_SAMPLER_MESSAGE_SAMPLE_BIAS_COMPARE] = "sample_b_c",
607 [GFX5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE] = "sample_l_c",
608 [GFX5_SAMPLER_MESSAGE_SAMPLE_LD] = "ld",
609 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4] = "gather4",
610 [GFX5_SAMPLER_MESSAGE_LOD] = "lod",
611 [GFX5_SAMPLER_MESSAGE_SAMPLE_RESINFO] = "resinfo",
612 [GFX6_SAMPLER_MESSAGE_SAMPLE_SAMPLEINFO] = "sampleinfo",
613 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C] = "gather4_c",
614 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO] = "gather4_po",
615 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO_C] = "gather4_po_c",
616 [HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE] = "sample_d_c",
617 [GFX9_SAMPLER_MESSAGE_SAMPLE_LZ] = "sample_lz",
618 [GFX9_SAMPLER_MESSAGE_SAMPLE_C_LZ] = "sample_c_lz",
619 [GFX9_SAMPLER_MESSAGE_SAMPLE_LD_LZ] = "ld_lz",
620 [GFX9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W] = "ld2dms_w",
621 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD_MCS] = "ld_mcs",
622 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD2DMS] = "ld2dms",
623 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD2DSS] = "ld2dss",
624 };
625
626 static const char *const xe2_sampler_msg_type[] = {
627 [GFX5_SAMPLER_MESSAGE_SAMPLE] = "sample",
628 [GFX5_SAMPLER_MESSAGE_SAMPLE_BIAS] = "sample_b",
629 [GFX5_SAMPLER_MESSAGE_SAMPLE_LOD] = "sample_l",
630 [GFX5_SAMPLER_MESSAGE_SAMPLE_COMPARE] = "sample_c",
631 [GFX5_SAMPLER_MESSAGE_SAMPLE_DERIVS] = "sample_d",
632 [GFX5_SAMPLER_MESSAGE_SAMPLE_BIAS_COMPARE] = "sample_b_c",
633 [GFX5_SAMPLER_MESSAGE_SAMPLE_LOD_COMPARE] = "sample_l_c",
634 [GFX5_SAMPLER_MESSAGE_SAMPLE_LD] = "ld",
635 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4] = "gather4",
636 [GFX5_SAMPLER_MESSAGE_LOD] = "lod",
637 [GFX5_SAMPLER_MESSAGE_SAMPLE_RESINFO] = "resinfo",
638 [GFX6_SAMPLER_MESSAGE_SAMPLE_SAMPLEINFO] = "sampleinfo",
639 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4_C] = "gather4_c",
640 [GFX7_SAMPLER_MESSAGE_SAMPLE_GATHER4_PO] = "gather4_po",
641 [XE2_SAMPLER_MESSAGE_SAMPLE_MLOD] = "sample_mlod",
642 [XE2_SAMPLER_MESSAGE_SAMPLE_COMPARE_MLOD] = "sample_c_mlod",
643 [HSW_SAMPLER_MESSAGE_SAMPLE_DERIV_COMPARE] = "sample_d_c",
644 [GFX9_SAMPLER_MESSAGE_SAMPLE_LZ] = "sample_lz",
645 [GFX9_SAMPLER_MESSAGE_SAMPLE_C_LZ] = "sample_c_lz",
646 [GFX9_SAMPLER_MESSAGE_SAMPLE_LD_LZ] = "ld_lz",
647 [GFX9_SAMPLER_MESSAGE_SAMPLE_LD2DMS_W] = "ld2dms_w",
648 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD_MCS] = "ld_mcs",
649 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD2DMS] = "ld2dms",
650 [GFX7_SAMPLER_MESSAGE_SAMPLE_LD2DSS] = "ld2dss",
651 };
652
653 static const char *const gfx5_sampler_simd_mode[7] = {
654 [ELK_SAMPLER_SIMD_MODE_SIMD4X2] = "SIMD4x2",
655 [ELK_SAMPLER_SIMD_MODE_SIMD8] = "SIMD8",
656 [ELK_SAMPLER_SIMD_MODE_SIMD16] = "SIMD16",
657 [ELK_SAMPLER_SIMD_MODE_SIMD32_64] = "SIMD32/64",
658 [GFX10_SAMPLER_SIMD_MODE_SIMD8H] = "SIMD8H",
659 [GFX10_SAMPLER_SIMD_MODE_SIMD16H] = "SIMD16H",
660 };
661
662 static const char *const xe2_sampler_simd_mode[7] = {
663 [XE2_SAMPLER_SIMD_MODE_SIMD16] = "SIMD16",
664 [XE2_SAMPLER_SIMD_MODE_SIMD32] = "SIMD32",
665 [XE2_SAMPLER_SIMD_MODE_SIMD16H] = "SIMD16H",
666 [XE2_SAMPLER_SIMD_MODE_SIMD32H] = "SIMD32H",
667 };
668
669 static const char *const sampler_target_format[4] = {
670 [0] = "F",
671 [2] = "UD",
672 [3] = "D"
673 };
674
675 static const char *const lsc_operation[] = {
676 [LSC_OP_LOAD] = "load",
677 [LSC_OP_LOAD_CMASK] = "load_cmask",
678 [LSC_OP_STORE] = "store",
679 [LSC_OP_STORE_CMASK] = "store_cmask",
680 [LSC_OP_FENCE] = "fence",
681 [LSC_OP_ATOMIC_INC] = "atomic_inc",
682 [LSC_OP_ATOMIC_DEC] = "atomic_dec",
683 [LSC_OP_ATOMIC_LOAD] = "atomic_load",
684 [LSC_OP_ATOMIC_STORE] = "atomic_store",
685 [LSC_OP_ATOMIC_ADD] = "atomic_add",
686 [LSC_OP_ATOMIC_SUB] = "atomic_sub",
687 [LSC_OP_ATOMIC_MIN] = "atomic_min",
688 [LSC_OP_ATOMIC_MAX] = "atomic_max",
689 [LSC_OP_ATOMIC_UMIN] = "atomic_umin",
690 [LSC_OP_ATOMIC_UMAX] = "atomic_umax",
691 [LSC_OP_ATOMIC_CMPXCHG] = "atomic_cmpxchg",
692 [LSC_OP_ATOMIC_FADD] = "atomic_fadd",
693 [LSC_OP_ATOMIC_FSUB] = "atomic_fsub",
694 [LSC_OP_ATOMIC_FMIN] = "atomic_fmin",
695 [LSC_OP_ATOMIC_FMAX] = "atomic_fmax",
696 [LSC_OP_ATOMIC_FCMPXCHG] = "atomic_fcmpxchg",
697 [LSC_OP_ATOMIC_AND] = "atomic_and",
698 [LSC_OP_ATOMIC_OR] = "atomic_or",
699 [LSC_OP_ATOMIC_XOR] = "atomic_xor",
700 };
701
702 static const char *const lsc_addr_surface_type[] = {
703 [LSC_ADDR_SURFTYPE_FLAT] = "flat",
704 [LSC_ADDR_SURFTYPE_BSS] = "bss",
705 [LSC_ADDR_SURFTYPE_SS] = "ss",
706 [LSC_ADDR_SURFTYPE_BTI] = "bti",
707 };
708
709 static const char* const lsc_fence_scope[] = {
710 [LSC_FENCE_THREADGROUP] = "threadgroup",
711 [LSC_FENCE_LOCAL] = "local",
712 [LSC_FENCE_TILE] = "tile",
713 [LSC_FENCE_GPU] = "gpu",
714 [LSC_FENCE_ALL_GPU] = "all_gpu",
715 [LSC_FENCE_SYSTEM_RELEASE] = "system_release",
716 [LSC_FENCE_SYSTEM_ACQUIRE] = "system_acquire",
717 };
718
719 static const char* const lsc_flush_type[] = {
720 [LSC_FLUSH_TYPE_NONE] = "none",
721 [LSC_FLUSH_TYPE_EVICT] = "evict",
722 [LSC_FLUSH_TYPE_INVALIDATE] = "invalidate",
723 [LSC_FLUSH_TYPE_DISCARD] = "discard",
724 [LSC_FLUSH_TYPE_CLEAN] = "clean",
725 [LSC_FLUSH_TYPE_L3ONLY] = "l3only",
726 [LSC_FLUSH_TYPE_NONE_6] = "none_6",
727 };
728
729 static const char* const lsc_addr_size[] = {
730 [LSC_ADDR_SIZE_A16] = "a16",
731 [LSC_ADDR_SIZE_A32] = "a32",
732 [LSC_ADDR_SIZE_A64] = "a64",
733 };
734
735 static const char* const lsc_backup_fence_routing[] = {
736 [LSC_NORMAL_ROUTING] = "normal_routing",
737 [LSC_ROUTE_TO_LSC] = "route_to_lsc",
738 };
739
740 static const char* const lsc_data_size[] = {
741 [LSC_DATA_SIZE_D8] = "d8",
742 [LSC_DATA_SIZE_D16] = "d16",
743 [LSC_DATA_SIZE_D32] = "d32",
744 [LSC_DATA_SIZE_D64] = "d64",
745 [LSC_DATA_SIZE_D8U32] = "d8u32",
746 [LSC_DATA_SIZE_D16U32] = "d16u32",
747 [LSC_DATA_SIZE_D16BF32] = "d16bf32",
748 };
749
750 static const char* const lsc_vect_size_str[] = {
751 [LSC_VECT_SIZE_V1] = "V1",
752 [LSC_VECT_SIZE_V2] = "V2",
753 [LSC_VECT_SIZE_V3] = "V3",
754 [LSC_VECT_SIZE_V4] = "V4",
755 [LSC_VECT_SIZE_V8] = "V8",
756 [LSC_VECT_SIZE_V16] = "V16",
757 [LSC_VECT_SIZE_V32] = "V32",
758 [LSC_VECT_SIZE_V64] = "V64",
759 };
760
761 static const char* const lsc_cmask_str[] = {
762 [LSC_CMASK_X] = "x",
763 [LSC_CMASK_Y] = "y",
764 [LSC_CMASK_XY] = "xy",
765 [LSC_CMASK_Z] = "z",
766 [LSC_CMASK_XZ] = "xz",
767 [LSC_CMASK_YZ] = "yz",
768 [LSC_CMASK_XYZ] = "xyz",
769 [LSC_CMASK_W] = "w",
770 [LSC_CMASK_XW] = "xw",
771 [LSC_CMASK_YW] = "yw",
772 [LSC_CMASK_XYW] = "xyw",
773 [LSC_CMASK_ZW] = "zw",
774 [LSC_CMASK_XZW] = "xzw",
775 [LSC_CMASK_YZW] = "yzw",
776 [LSC_CMASK_XYZW] = "xyzw",
777 };
778
779 static const char* const lsc_cache_load[] = {
780 [LSC_CACHE_LOAD_L1STATE_L3MOCS] = "L1STATE_L3MOCS",
781 [LSC_CACHE_LOAD_L1UC_L3UC] = "L1UC_L3UC",
782 [LSC_CACHE_LOAD_L1UC_L3C] = "L1UC_L3C",
783 [LSC_CACHE_LOAD_L1C_L3UC] = "L1C_L3UC",
784 [LSC_CACHE_LOAD_L1C_L3C] = "L1C_L3C",
785 [LSC_CACHE_LOAD_L1S_L3UC] = "L1S_L3UC",
786 [LSC_CACHE_LOAD_L1S_L3C] = "L1S_L3C",
787 [LSC_CACHE_LOAD_L1IAR_L3C] = "L1IAR_L3C",
788 };
789
790 static const char* const lsc_cache_store[] = {
791 [LSC_CACHE_STORE_L1STATE_L3MOCS] = "L1STATE_L3MOCS",
792 [LSC_CACHE_STORE_L1UC_L3UC] = "L1UC_L3UC",
793 [LSC_CACHE_STORE_L1UC_L3WB] = "L1UC_L3WB",
794 [LSC_CACHE_STORE_L1WT_L3UC] = "L1WT_L3UC",
795 [LSC_CACHE_STORE_L1WT_L3WB] = "L1WT_L3WB",
796 [LSC_CACHE_STORE_L1S_L3UC] = "L1S_L3UC",
797 [LSC_CACHE_STORE_L1S_L3WB] = "L1S_L3WB",
798 [LSC_CACHE_STORE_L1WB_L3WB] = "L1WB_L3WB",
799 };
800
801 static const char* const xe2_lsc_cache_load[] = {
802 [XE2_LSC_CACHE_LOAD_L1STATE_L3MOCS] = "L1STATE_L3MOCS",
803 [XE2_LSC_CACHE_LOAD_L1UC_L3UC] = "L1UC_L3UC",
804 [XE2_LSC_CACHE_LOAD_L1UC_L3C] = "L1UC_L3C",
805 [XE2_LSC_CACHE_LOAD_L1UC_L3CC] = "L1UC_L3CC",
806 [XE2_LSC_CACHE_LOAD_L1C_L3UC] = "L1C_L3UC",
807 [XE2_LSC_CACHE_LOAD_L1C_L3C] = "L1C_L3C",
808 [XE2_LSC_CACHE_LOAD_L1C_L3CC] = "L1C_L3CC",
809 [XE2_LSC_CACHE_LOAD_L1S_L3UC] = "L1S_L3UC",
810 [XE2_LSC_CACHE_LOAD_L1S_L3C] = "L1S_L3C",
811 [XE2_LSC_CACHE_LOAD_L1IAR_L3IAR] = "L1IAR_L3IAR",
812 };
813
814 static const char* const xe2_lsc_cache_store[] = {
815 [XE2_LSC_CACHE_STORE_L1STATE_L3MOCS] = "L1STATE_L3MOCS",
816 [XE2_LSC_CACHE_STORE_L1UC_L3UC] = "L1UC_L3UC",
817 [XE2_LSC_CACHE_STORE_L1UC_L3WB] = "L1UC_L3WB",
818 [XE2_LSC_CACHE_STORE_L1WT_L3UC] = "L1WT_L3UC",
819 [XE2_LSC_CACHE_STORE_L1WT_L3WB] = "L1WT_L3WB",
820 [XE2_LSC_CACHE_STORE_L1S_L3UC] = "L1S_L3UC",
821 [XE2_LSC_CACHE_STORE_L1S_L3WB] = "L1S_L3WB",
822 [XE2_LSC_CACHE_STORE_L1WB_L3WB] = "L1WB_L3WB",
823 };
824
825 static const char* const dpas_systolic_depth[4] = {
826 [0] = "16",
827 [1] = "2",
828 [2] = "4",
829 [3] = "8"
830 };
831
832 static int column;
833
834 static int
string(FILE * file,const char * string)835 string(FILE *file, const char *string)
836 {
837 fputs(string, file);
838 column += strlen(string);
839 return 0;
840 }
841
842 static int
843 format(FILE *f, const char *format, ...) PRINTFLIKE(2, 3);
844
845 static int
format(FILE * f,const char * format,...)846 format(FILE *f, const char *format, ...)
847 {
848 char buf[1024];
849 va_list args;
850 va_start(args, format);
851
852 vsnprintf(buf, sizeof(buf) - 1, format, args);
853 va_end(args);
854 string(f, buf);
855 return 0;
856 }
857
858 static int
newline(FILE * f)859 newline(FILE *f)
860 {
861 putc('\n', f);
862 column = 0;
863 return 0;
864 }
865
866 static int
pad(FILE * f,int c)867 pad(FILE *f, int c)
868 {
869 do
870 string(f, " ");
871 while (column < c);
872 return 0;
873 }
874
875 static int
control(FILE * file,const char * name,const char * const ctrl[],unsigned id,int * space)876 control(FILE *file, const char *name, const char *const ctrl[],
877 unsigned id, int *space)
878 {
879 if (!ctrl[id]) {
880 fprintf(file, "*** invalid %s value %d ", name, id);
881 return 1;
882 }
883 if (ctrl[id][0]) {
884 if (space && *space)
885 string(file, " ");
886 string(file, ctrl[id]);
887 if (space)
888 *space = 1;
889 }
890 return 0;
891 }
892
893 static int
print_opcode(FILE * file,const struct elk_isa_info * isa,enum elk_opcode id)894 print_opcode(FILE *file, const struct elk_isa_info *isa,
895 enum elk_opcode id)
896 {
897 const struct elk_opcode_desc *desc = elk_opcode_desc(isa, id);
898 if (!desc) {
899 format(file, "*** invalid opcode value %d ", id);
900 return 1;
901 }
902 string(file, desc->name);
903 return 0;
904 }
905
906 static int
reg(FILE * file,unsigned _reg_file,unsigned _reg_nr)907 reg(FILE *file, unsigned _reg_file, unsigned _reg_nr)
908 {
909 int err = 0;
910
911 /* Clear the Compr4 instruction compression bit. */
912 if (_reg_file == ELK_MESSAGE_REGISTER_FILE)
913 _reg_nr &= ~ELK_MRF_COMPR4;
914
915 if (_reg_file == ELK_ARCHITECTURE_REGISTER_FILE) {
916 switch (_reg_nr & 0xf0) {
917 case ELK_ARF_NULL:
918 string(file, "null");
919 break;
920 case ELK_ARF_ADDRESS:
921 format(file, "a%d", _reg_nr & 0x0f);
922 break;
923 case ELK_ARF_ACCUMULATOR:
924 format(file, "acc%d", _reg_nr & 0x0f);
925 break;
926 case ELK_ARF_FLAG:
927 format(file, "f%d", _reg_nr & 0x0f);
928 break;
929 case ELK_ARF_MASK:
930 format(file, "mask%d", _reg_nr & 0x0f);
931 break;
932 case ELK_ARF_MASK_STACK:
933 format(file, "ms%d", _reg_nr & 0x0f);
934 break;
935 case ELK_ARF_MASK_STACK_DEPTH:
936 format(file, "msd%d", _reg_nr & 0x0f);
937 break;
938 case ELK_ARF_STATE:
939 format(file, "sr%d", _reg_nr & 0x0f);
940 break;
941 case ELK_ARF_CONTROL:
942 format(file, "cr%d", _reg_nr & 0x0f);
943 break;
944 case ELK_ARF_NOTIFICATION_COUNT:
945 format(file, "n%d", _reg_nr & 0x0f);
946 break;
947 case ELK_ARF_IP:
948 string(file, "ip");
949 return -1;
950 break;
951 case ELK_ARF_TDR:
952 format(file, "tdr0");
953 return -1;
954 case ELK_ARF_TIMESTAMP:
955 format(file, "tm%d", _reg_nr & 0x0f);
956 break;
957 default:
958 format(file, "ARF%d", _reg_nr);
959 break;
960 }
961 } else {
962 err |= control(file, "src reg file", reg_file, _reg_file, NULL);
963 format(file, "%d", _reg_nr);
964 }
965 return err;
966 }
967
968 static int
dest(FILE * file,const struct elk_isa_info * isa,const elk_inst * inst)969 dest(FILE *file, const struct elk_isa_info *isa, const elk_inst *inst)
970 {
971 const struct intel_device_info *devinfo = isa->devinfo;
972 enum elk_reg_type type = elk_inst_dst_type(devinfo, inst);
973 unsigned elem_size = elk_reg_type_to_size(type);
974 int err = 0;
975
976 if (is_split_send(devinfo, elk_inst_opcode(isa, inst))) {
977 /* These are fixed for split sends */
978 type = ELK_REGISTER_TYPE_UD;
979 elem_size = 4;
980 if (devinfo->ver >= 12) {
981 err |= reg(file, elk_inst_send_dst_reg_file(devinfo, inst),
982 elk_inst_dst_da_reg_nr(devinfo, inst));
983 string(file, elk_reg_type_to_letters(type));
984 } else if (elk_inst_dst_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
985 err |= reg(file, elk_inst_send_dst_reg_file(devinfo, inst),
986 elk_inst_dst_da_reg_nr(devinfo, inst));
987 unsigned subreg_nr = elk_inst_dst_da16_subreg_nr(devinfo, inst);
988 if (subreg_nr)
989 format(file, ".%u", subreg_nr);
990 string(file, elk_reg_type_to_letters(type));
991 } else {
992 string(file, "g[a0");
993 if (elk_inst_dst_ia_subreg_nr(devinfo, inst))
994 format(file, ".%"PRIu64, elk_inst_dst_ia_subreg_nr(devinfo, inst) /
995 elem_size);
996 if (elk_inst_send_dst_ia16_addr_imm(devinfo, inst))
997 format(file, " %d", elk_inst_send_dst_ia16_addr_imm(devinfo, inst));
998 string(file, "]<");
999 string(file, elk_reg_type_to_letters(type));
1000 }
1001 } else if (elk_inst_access_mode(devinfo, inst) == ELK_ALIGN_1) {
1002 if (elk_inst_dst_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1003 err |= reg(file, elk_inst_dst_reg_file(devinfo, inst),
1004 elk_inst_dst_da_reg_nr(devinfo, inst));
1005 if (err == -1)
1006 return 0;
1007 if (elk_inst_dst_da1_subreg_nr(devinfo, inst))
1008 format(file, ".%"PRIu64, elk_inst_dst_da1_subreg_nr(devinfo, inst) /
1009 elem_size);
1010 string(file, "<");
1011 err |= control(file, "horiz stride", horiz_stride,
1012 elk_inst_dst_hstride(devinfo, inst), NULL);
1013 string(file, ">");
1014 string(file, elk_reg_type_to_letters(type));
1015 } else {
1016 string(file, "g[a0");
1017 if (elk_inst_dst_ia_subreg_nr(devinfo, inst))
1018 format(file, ".%"PRIu64, elk_inst_dst_ia_subreg_nr(devinfo, inst) /
1019 elem_size);
1020 if (elk_inst_dst_ia1_addr_imm(devinfo, inst))
1021 format(file, " %d", elk_inst_dst_ia1_addr_imm(devinfo, inst));
1022 string(file, "]<");
1023 err |= control(file, "horiz stride", horiz_stride,
1024 elk_inst_dst_hstride(devinfo, inst), NULL);
1025 string(file, ">");
1026 string(file, elk_reg_type_to_letters(type));
1027 }
1028 } else {
1029 if (elk_inst_dst_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1030 err |= reg(file, elk_inst_dst_reg_file(devinfo, inst),
1031 elk_inst_dst_da_reg_nr(devinfo, inst));
1032 if (err == -1)
1033 return 0;
1034 if (elk_inst_dst_da16_subreg_nr(devinfo, inst))
1035 format(file, ".%u", 16 / elem_size);
1036 string(file, "<1>");
1037 err |= control(file, "writemask", writemask,
1038 elk_inst_da16_writemask(devinfo, inst), NULL);
1039 string(file, elk_reg_type_to_letters(type));
1040 } else {
1041 err = 1;
1042 string(file, "Indirect align16 address mode not supported");
1043 }
1044 }
1045
1046 return 0;
1047 }
1048
1049 static int
dest_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1050 dest_3src(FILE *file, const struct intel_device_info *devinfo,
1051 const elk_inst *inst)
1052 {
1053 bool is_align1 = elk_inst_3src_access_mode(devinfo, inst) == ELK_ALIGN_1;
1054 int err = 0;
1055 uint32_t reg_file;
1056 unsigned subreg_nr;
1057 enum elk_reg_type type;
1058
1059 if (devinfo->ver < 10 && is_align1)
1060 return 0;
1061
1062 if (devinfo->ver == 6 && elk_inst_3src_a16_dst_reg_file(devinfo, inst))
1063 reg_file = ELK_MESSAGE_REGISTER_FILE;
1064 else if (devinfo->ver >= 12)
1065 reg_file = elk_inst_3src_a1_dst_reg_file(devinfo, inst);
1066 else if (is_align1 && elk_inst_3src_a1_dst_reg_file(devinfo, inst))
1067 reg_file = ELK_ARCHITECTURE_REGISTER_FILE;
1068 else
1069 reg_file = ELK_GENERAL_REGISTER_FILE;
1070
1071 err |= reg(file, reg_file, elk_inst_3src_dst_reg_nr(devinfo, inst));
1072 if (err == -1)
1073 return 0;
1074
1075 if (is_align1) {
1076 type = elk_inst_3src_a1_dst_type(devinfo, inst);
1077 subreg_nr = elk_inst_3src_a1_dst_subreg_nr(devinfo, inst);
1078 } else {
1079 type = elk_inst_3src_a16_dst_type(devinfo, inst);
1080 subreg_nr = elk_inst_3src_a16_dst_subreg_nr(devinfo, inst) * 4;
1081 }
1082 subreg_nr /= elk_reg_type_to_size(type);
1083
1084 if (subreg_nr)
1085 format(file, ".%u", subreg_nr);
1086 string(file, "<1>");
1087
1088 if (!is_align1) {
1089 err |= control(file, "writemask", writemask,
1090 elk_inst_3src_a16_dst_writemask(devinfo, inst), NULL);
1091 }
1092 string(file, elk_reg_type_to_letters(type));
1093
1094 return 0;
1095 }
1096
1097 static int
dest_dpas_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1098 dest_dpas_3src(FILE *file, const struct intel_device_info *devinfo,
1099 const elk_inst *inst)
1100 {
1101 uint32_t reg_file = elk_inst_dpas_3src_dst_reg_file(devinfo, inst);
1102
1103 if (reg(file, reg_file, elk_inst_dpas_3src_dst_reg_nr(devinfo, inst)) == -1)
1104 return 0;
1105
1106 enum elk_reg_type type = elk_inst_dpas_3src_dst_type(devinfo, inst);
1107 unsigned subreg_nr = elk_inst_dpas_3src_dst_subreg_nr(devinfo, inst);
1108
1109 if (subreg_nr)
1110 format(file, ".%u", subreg_nr);
1111 string(file, "<1>");
1112
1113 string(file, elk_reg_type_to_letters(type));
1114
1115 return 0;
1116 }
1117
1118 static int
src_align1_region(FILE * file,unsigned _vert_stride,unsigned _width,unsigned _horiz_stride)1119 src_align1_region(FILE *file,
1120 unsigned _vert_stride, unsigned _width,
1121 unsigned _horiz_stride)
1122 {
1123 int err = 0;
1124 string(file, "<");
1125 err |= control(file, "vert stride", vert_stride, _vert_stride, NULL);
1126 string(file, ",");
1127 err |= control(file, "width", width, _width, NULL);
1128 string(file, ",");
1129 err |= control(file, "horiz_stride", horiz_stride, _horiz_stride, NULL);
1130 string(file, ">");
1131 return err;
1132 }
1133
1134 static int
src_da1(FILE * file,const struct intel_device_info * devinfo,unsigned opcode,enum elk_reg_type type,unsigned _reg_file,unsigned _vert_stride,unsigned _width,unsigned _horiz_stride,unsigned reg_num,unsigned sub_reg_num,unsigned __abs,unsigned _negate)1135 src_da1(FILE *file,
1136 const struct intel_device_info *devinfo,
1137 unsigned opcode,
1138 enum elk_reg_type type, unsigned _reg_file,
1139 unsigned _vert_stride, unsigned _width, unsigned _horiz_stride,
1140 unsigned reg_num, unsigned sub_reg_num, unsigned __abs,
1141 unsigned _negate)
1142 {
1143 int err = 0;
1144
1145 if (devinfo->ver >= 8 && is_logic_instruction(opcode))
1146 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
1147 else
1148 err |= control(file, "negate", m_negate, _negate, NULL);
1149
1150 err |= control(file, "abs", _abs, __abs, NULL);
1151
1152 err |= reg(file, _reg_file, reg_num);
1153 if (err == -1)
1154 return 0;
1155 if (sub_reg_num) {
1156 unsigned elem_size = elk_reg_type_to_size(type);
1157 format(file, ".%d", sub_reg_num / elem_size); /* use formal style like spec */
1158 }
1159 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1160 string(file, elk_reg_type_to_letters(type));
1161 return err;
1162 }
1163
1164 static int
src_ia1(FILE * file,const struct intel_device_info * devinfo,unsigned opcode,enum elk_reg_type type,int _addr_imm,unsigned _addr_subreg_nr,unsigned _negate,unsigned __abs,unsigned _horiz_stride,unsigned _width,unsigned _vert_stride)1165 src_ia1(FILE *file,
1166 const struct intel_device_info *devinfo,
1167 unsigned opcode,
1168 enum elk_reg_type type,
1169 int _addr_imm,
1170 unsigned _addr_subreg_nr,
1171 unsigned _negate,
1172 unsigned __abs,
1173 unsigned _horiz_stride, unsigned _width, unsigned _vert_stride)
1174 {
1175 int err = 0;
1176
1177 if (devinfo->ver >= 8 && is_logic_instruction(opcode))
1178 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
1179 else
1180 err |= control(file, "negate", m_negate, _negate, NULL);
1181
1182 err |= control(file, "abs", _abs, __abs, NULL);
1183
1184 string(file, "g[a0");
1185 if (_addr_subreg_nr)
1186 format(file, ".%d", _addr_subreg_nr);
1187 if (_addr_imm)
1188 format(file, " %d", _addr_imm);
1189 string(file, "]");
1190 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1191 string(file, elk_reg_type_to_letters(type));
1192 return err;
1193 }
1194
1195 static int
src_swizzle(FILE * file,unsigned swiz)1196 src_swizzle(FILE *file, unsigned swiz)
1197 {
1198 unsigned x = ELK_GET_SWZ(swiz, ELK_CHANNEL_X);
1199 unsigned y = ELK_GET_SWZ(swiz, ELK_CHANNEL_Y);
1200 unsigned z = ELK_GET_SWZ(swiz, ELK_CHANNEL_Z);
1201 unsigned w = ELK_GET_SWZ(swiz, ELK_CHANNEL_W);
1202 int err = 0;
1203
1204 if (x == y && x == z && x == w) {
1205 string(file, ".");
1206 err |= control(file, "channel select", chan_sel, x, NULL);
1207 } else if (swiz != ELK_SWIZZLE_XYZW) {
1208 string(file, ".");
1209 err |= control(file, "channel select", chan_sel, x, NULL);
1210 err |= control(file, "channel select", chan_sel, y, NULL);
1211 err |= control(file, "channel select", chan_sel, z, NULL);
1212 err |= control(file, "channel select", chan_sel, w, NULL);
1213 }
1214 return err;
1215 }
1216
1217 static int
src_da16(FILE * file,const struct intel_device_info * devinfo,unsigned opcode,enum elk_reg_type type,unsigned _reg_file,unsigned _vert_stride,unsigned _reg_nr,unsigned _subreg_nr,unsigned __abs,unsigned _negate,unsigned swz_x,unsigned swz_y,unsigned swz_z,unsigned swz_w)1218 src_da16(FILE *file,
1219 const struct intel_device_info *devinfo,
1220 unsigned opcode,
1221 enum elk_reg_type type,
1222 unsigned _reg_file,
1223 unsigned _vert_stride,
1224 unsigned _reg_nr,
1225 unsigned _subreg_nr,
1226 unsigned __abs,
1227 unsigned _negate,
1228 unsigned swz_x, unsigned swz_y, unsigned swz_z, unsigned swz_w)
1229 {
1230 int err = 0;
1231
1232 if (devinfo->ver >= 8 && is_logic_instruction(opcode))
1233 err |= control(file, "bitnot", m_bitnot, _negate, NULL);
1234 else
1235 err |= control(file, "negate", m_negate, _negate, NULL);
1236
1237 err |= control(file, "abs", _abs, __abs, NULL);
1238
1239 err |= reg(file, _reg_file, _reg_nr);
1240 if (err == -1)
1241 return 0;
1242 if (_subreg_nr) {
1243 unsigned elem_size = elk_reg_type_to_size(type);
1244
1245 /* bit4 for subreg number byte addressing. Make this same meaning as
1246 in da1 case, so output looks consistent. */
1247 format(file, ".%d", 16 / elem_size);
1248 }
1249 string(file, "<");
1250 err |= control(file, "vert stride", vert_stride, _vert_stride, NULL);
1251 string(file, ">");
1252 err |= src_swizzle(file, ELK_SWIZZLE4(swz_x, swz_y, swz_z, swz_w));
1253 string(file, elk_reg_type_to_letters(type));
1254 return err;
1255 }
1256
1257 static enum elk_vertical_stride
vstride_from_align1_3src_vstride(const struct intel_device_info * devinfo,enum gfx10_align1_3src_vertical_stride vstride)1258 vstride_from_align1_3src_vstride(const struct intel_device_info *devinfo,
1259 enum gfx10_align1_3src_vertical_stride vstride)
1260 {
1261 switch (vstride) {
1262 case ELK_ALIGN1_3SRC_VERTICAL_STRIDE_0: return ELK_VERTICAL_STRIDE_0;
1263 case ELK_ALIGN1_3SRC_VERTICAL_STRIDE_2:
1264 if (devinfo->ver >= 12)
1265 return ELK_VERTICAL_STRIDE_1;
1266 else
1267 return ELK_VERTICAL_STRIDE_2;
1268 case ELK_ALIGN1_3SRC_VERTICAL_STRIDE_4: return ELK_VERTICAL_STRIDE_4;
1269 case ELK_ALIGN1_3SRC_VERTICAL_STRIDE_8: return ELK_VERTICAL_STRIDE_8;
1270 default:
1271 unreachable("not reached");
1272 }
1273 }
1274
1275 static enum elk_horizontal_stride
hstride_from_align1_3src_hstride(enum gfx10_align1_3src_src_horizontal_stride hstride)1276 hstride_from_align1_3src_hstride(enum gfx10_align1_3src_src_horizontal_stride hstride)
1277 {
1278 switch (hstride) {
1279 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_0: return ELK_HORIZONTAL_STRIDE_0;
1280 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_1: return ELK_HORIZONTAL_STRIDE_1;
1281 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_2: return ELK_HORIZONTAL_STRIDE_2;
1282 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_4: return ELK_HORIZONTAL_STRIDE_4;
1283 default:
1284 unreachable("not reached");
1285 }
1286 }
1287
1288 static enum elk_vertical_stride
vstride_from_align1_3src_hstride(enum gfx10_align1_3src_src_horizontal_stride hstride)1289 vstride_from_align1_3src_hstride(enum gfx10_align1_3src_src_horizontal_stride hstride)
1290 {
1291 switch (hstride) {
1292 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_0: return ELK_VERTICAL_STRIDE_0;
1293 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_1: return ELK_VERTICAL_STRIDE_1;
1294 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_2: return ELK_VERTICAL_STRIDE_2;
1295 case ELK_ALIGN1_3SRC_SRC_HORIZONTAL_STRIDE_4: return ELK_VERTICAL_STRIDE_4;
1296 default:
1297 unreachable("not reached");
1298 }
1299 }
1300
1301 /* From "GFX10 Regioning Rules for Align1 Ternary Operations" in the
1302 * "Register Region Restrictions" documentation
1303 */
1304 static enum elk_width
implied_width(enum elk_vertical_stride _vert_stride,enum elk_horizontal_stride _horiz_stride)1305 implied_width(enum elk_vertical_stride _vert_stride,
1306 enum elk_horizontal_stride _horiz_stride)
1307 {
1308 /* "1. Width is 1 when Vertical and Horizontal Strides are both zero." */
1309 if (_vert_stride == ELK_VERTICAL_STRIDE_0 &&
1310 _horiz_stride == ELK_HORIZONTAL_STRIDE_0) {
1311 return ELK_WIDTH_1;
1312
1313 /* "2. Width is equal to vertical stride when Horizontal Stride is zero." */
1314 } else if (_horiz_stride == ELK_HORIZONTAL_STRIDE_0) {
1315 switch (_vert_stride) {
1316 case ELK_VERTICAL_STRIDE_1: return ELK_WIDTH_1;
1317 case ELK_VERTICAL_STRIDE_2: return ELK_WIDTH_2;
1318 case ELK_VERTICAL_STRIDE_4: return ELK_WIDTH_4;
1319 case ELK_VERTICAL_STRIDE_8: return ELK_WIDTH_8;
1320 case ELK_VERTICAL_STRIDE_0:
1321 default:
1322 unreachable("not reached");
1323 }
1324
1325 } else {
1326 /* FINISHME: Implement these: */
1327
1328 /* "3. Width is equal to Vertical Stride/Horizontal Stride when both
1329 * Strides are non-zero.
1330 *
1331 * 4. Vertical Stride must not be zero if Horizontal Stride is non-zero.
1332 * This implies Vertical Stride is always greater than Horizontal
1333 * Stride."
1334 *
1335 * Given these statements and the knowledge that the stride and width
1336 * values are encoded in logarithmic form, we can perform the division
1337 * by just subtracting.
1338 */
1339 return _vert_stride - _horiz_stride;
1340 }
1341 }
1342
1343 static int
src0_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1344 src0_3src(FILE *file, const struct intel_device_info *devinfo,
1345 const elk_inst *inst)
1346 {
1347 int err = 0;
1348 unsigned reg_nr, subreg_nr;
1349 enum elk_reg_file _file;
1350 enum elk_reg_type type;
1351 enum elk_vertical_stride _vert_stride;
1352 enum elk_width _width;
1353 enum elk_horizontal_stride _horiz_stride;
1354 bool is_scalar_region;
1355 bool is_align1 = elk_inst_3src_access_mode(devinfo, inst) == ELK_ALIGN_1;
1356
1357 if (devinfo->ver < 10 && is_align1)
1358 return 0;
1359
1360 if (is_align1) {
1361 if (devinfo->ver >= 12 && !elk_inst_3src_a1_src0_is_imm(devinfo, inst)) {
1362 _file = elk_inst_3src_a1_src0_reg_file(devinfo, inst);
1363 } else if (elk_inst_3src_a1_src0_reg_file(devinfo, inst) ==
1364 ELK_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1365 _file = ELK_GENERAL_REGISTER_FILE;
1366 } else if (elk_inst_3src_a1_src0_type(devinfo, inst) ==
1367 ELK_REGISTER_TYPE_NF) {
1368 _file = ELK_ARCHITECTURE_REGISTER_FILE;
1369 } else {
1370 _file = ELK_IMMEDIATE_VALUE;
1371 uint16_t imm_val = elk_inst_3src_a1_src0_imm(devinfo, inst);
1372 enum elk_reg_type type = elk_inst_3src_a1_src0_type(devinfo, inst);
1373
1374 if (type == ELK_REGISTER_TYPE_W) {
1375 format(file, "%dW", imm_val);
1376 } else if (type == ELK_REGISTER_TYPE_UW) {
1377 format(file, "0x%04xUW", imm_val);
1378 } else if (type == ELK_REGISTER_TYPE_HF) {
1379 format(file, "0x%04xHF", imm_val);
1380 }
1381 return 0;
1382 }
1383
1384 reg_nr = elk_inst_3src_src0_reg_nr(devinfo, inst);
1385 subreg_nr = elk_inst_3src_a1_src0_subreg_nr(devinfo, inst);
1386 type = elk_inst_3src_a1_src0_type(devinfo, inst);
1387 _vert_stride = vstride_from_align1_3src_vstride(
1388 devinfo, elk_inst_3src_a1_src0_vstride(devinfo, inst));
1389 _horiz_stride = hstride_from_align1_3src_hstride(
1390 elk_inst_3src_a1_src0_hstride(devinfo, inst));
1391 _width = implied_width(_vert_stride, _horiz_stride);
1392 } else {
1393 _file = ELK_GENERAL_REGISTER_FILE;
1394 reg_nr = elk_inst_3src_src0_reg_nr(devinfo, inst);
1395 subreg_nr = elk_inst_3src_a16_src0_subreg_nr(devinfo, inst) * 4;
1396 type = elk_inst_3src_a16_src_type(devinfo, inst);
1397
1398 if (elk_inst_3src_a16_src0_rep_ctrl(devinfo, inst)) {
1399 _vert_stride = ELK_VERTICAL_STRIDE_0;
1400 _width = ELK_WIDTH_1;
1401 _horiz_stride = ELK_HORIZONTAL_STRIDE_0;
1402 } else {
1403 _vert_stride = ELK_VERTICAL_STRIDE_4;
1404 _width = ELK_WIDTH_4;
1405 _horiz_stride = ELK_HORIZONTAL_STRIDE_1;
1406 }
1407 }
1408 is_scalar_region = _vert_stride == ELK_VERTICAL_STRIDE_0 &&
1409 _width == ELK_WIDTH_1 &&
1410 _horiz_stride == ELK_HORIZONTAL_STRIDE_0;
1411
1412 subreg_nr /= elk_reg_type_to_size(type);
1413
1414 err |= control(file, "negate", m_negate,
1415 elk_inst_3src_src0_negate(devinfo, inst), NULL);
1416 err |= control(file, "abs", _abs, elk_inst_3src_src0_abs(devinfo, inst), NULL);
1417
1418 err |= reg(file, _file, reg_nr);
1419 if (err == -1)
1420 return 0;
1421 if (subreg_nr || is_scalar_region)
1422 format(file, ".%d", subreg_nr);
1423 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1424 if (!is_scalar_region && !is_align1)
1425 err |= src_swizzle(file, elk_inst_3src_a16_src0_swizzle(devinfo, inst));
1426 string(file, elk_reg_type_to_letters(type));
1427 return err;
1428 }
1429
1430 static int
src1_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1431 src1_3src(FILE *file, const struct intel_device_info *devinfo,
1432 const elk_inst *inst)
1433 {
1434 int err = 0;
1435 unsigned reg_nr, subreg_nr;
1436 enum elk_reg_file _file;
1437 enum elk_reg_type type;
1438 enum elk_vertical_stride _vert_stride;
1439 enum elk_width _width;
1440 enum elk_horizontal_stride _horiz_stride;
1441 bool is_scalar_region;
1442 bool is_align1 = elk_inst_3src_access_mode(devinfo, inst) == ELK_ALIGN_1;
1443
1444 if (devinfo->ver < 10 && is_align1)
1445 return 0;
1446
1447 if (is_align1) {
1448 if (devinfo->ver >= 12) {
1449 _file = elk_inst_3src_a1_src1_reg_file(devinfo, inst);
1450 } else if (elk_inst_3src_a1_src1_reg_file(devinfo, inst) ==
1451 ELK_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1452 _file = ELK_GENERAL_REGISTER_FILE;
1453 } else {
1454 _file = ELK_ARCHITECTURE_REGISTER_FILE;
1455 }
1456
1457 reg_nr = elk_inst_3src_src1_reg_nr(devinfo, inst);
1458 subreg_nr = elk_inst_3src_a1_src1_subreg_nr(devinfo, inst);
1459 type = elk_inst_3src_a1_src1_type(devinfo, inst);
1460
1461 _vert_stride = vstride_from_align1_3src_vstride(
1462 devinfo, elk_inst_3src_a1_src1_vstride(devinfo, inst));
1463 _horiz_stride = hstride_from_align1_3src_hstride(
1464 elk_inst_3src_a1_src1_hstride(devinfo, inst));
1465 _width = implied_width(_vert_stride, _horiz_stride);
1466 } else {
1467 _file = ELK_GENERAL_REGISTER_FILE;
1468 reg_nr = elk_inst_3src_src1_reg_nr(devinfo, inst);
1469 subreg_nr = elk_inst_3src_a16_src1_subreg_nr(devinfo, inst) * 4;
1470 type = elk_inst_3src_a16_src_type(devinfo, inst);
1471
1472 if (elk_inst_3src_a16_src1_rep_ctrl(devinfo, inst)) {
1473 _vert_stride = ELK_VERTICAL_STRIDE_0;
1474 _width = ELK_WIDTH_1;
1475 _horiz_stride = ELK_HORIZONTAL_STRIDE_0;
1476 } else {
1477 _vert_stride = ELK_VERTICAL_STRIDE_4;
1478 _width = ELK_WIDTH_4;
1479 _horiz_stride = ELK_HORIZONTAL_STRIDE_1;
1480 }
1481 }
1482 is_scalar_region = _vert_stride == ELK_VERTICAL_STRIDE_0 &&
1483 _width == ELK_WIDTH_1 &&
1484 _horiz_stride == ELK_HORIZONTAL_STRIDE_0;
1485
1486 subreg_nr /= elk_reg_type_to_size(type);
1487
1488 err |= control(file, "negate", m_negate,
1489 elk_inst_3src_src1_negate(devinfo, inst), NULL);
1490 err |= control(file, "abs", _abs, elk_inst_3src_src1_abs(devinfo, inst), NULL);
1491
1492 err |= reg(file, _file, reg_nr);
1493 if (err == -1)
1494 return 0;
1495 if (subreg_nr || is_scalar_region)
1496 format(file, ".%d", subreg_nr);
1497 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1498 if (!is_scalar_region && !is_align1)
1499 err |= src_swizzle(file, elk_inst_3src_a16_src1_swizzle(devinfo, inst));
1500 string(file, elk_reg_type_to_letters(type));
1501 return err;
1502 }
1503
1504 static int
src2_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1505 src2_3src(FILE *file, const struct intel_device_info *devinfo,
1506 const elk_inst *inst)
1507 {
1508 int err = 0;
1509 unsigned reg_nr, subreg_nr;
1510 enum elk_reg_file _file;
1511 enum elk_reg_type type;
1512 enum elk_vertical_stride _vert_stride;
1513 enum elk_width _width;
1514 enum elk_horizontal_stride _horiz_stride;
1515 bool is_scalar_region;
1516 bool is_align1 = elk_inst_3src_access_mode(devinfo, inst) == ELK_ALIGN_1;
1517
1518 if (devinfo->ver < 10 && is_align1)
1519 return 0;
1520
1521 if (is_align1) {
1522 if (devinfo->ver >= 12 && !elk_inst_3src_a1_src2_is_imm(devinfo, inst)) {
1523 _file = elk_inst_3src_a1_src2_reg_file(devinfo, inst);
1524 } else if (elk_inst_3src_a1_src2_reg_file(devinfo, inst) ==
1525 ELK_ALIGN1_3SRC_GENERAL_REGISTER_FILE) {
1526 _file = ELK_GENERAL_REGISTER_FILE;
1527 } else {
1528 _file = ELK_IMMEDIATE_VALUE;
1529 uint16_t imm_val = elk_inst_3src_a1_src2_imm(devinfo, inst);
1530 enum elk_reg_type type = elk_inst_3src_a1_src2_type(devinfo, inst);
1531
1532 if (type == ELK_REGISTER_TYPE_W) {
1533 format(file, "%dW", imm_val);
1534 } else if (type == ELK_REGISTER_TYPE_UW) {
1535 format(file, "0x%04xUW", imm_val);
1536 } else if (type == ELK_REGISTER_TYPE_HF) {
1537 format(file, "0x%04xHF", imm_val);
1538 }
1539 return 0;
1540 }
1541
1542 reg_nr = elk_inst_3src_src2_reg_nr(devinfo, inst);
1543 subreg_nr = elk_inst_3src_a1_src2_subreg_nr(devinfo, inst);
1544 type = elk_inst_3src_a1_src2_type(devinfo, inst);
1545 /* FINISHME: No vertical stride on src2. Is using the hstride in place
1546 * correct? Doesn't seem like it, since there's hstride=1 but
1547 * no vstride=1.
1548 */
1549 _vert_stride = vstride_from_align1_3src_hstride(
1550 elk_inst_3src_a1_src2_hstride(devinfo, inst));
1551 _horiz_stride = hstride_from_align1_3src_hstride(
1552 elk_inst_3src_a1_src2_hstride(devinfo, inst));
1553 _width = implied_width(_vert_stride, _horiz_stride);
1554 } else {
1555 _file = ELK_GENERAL_REGISTER_FILE;
1556 reg_nr = elk_inst_3src_src2_reg_nr(devinfo, inst);
1557 subreg_nr = elk_inst_3src_a16_src2_subreg_nr(devinfo, inst) * 4;
1558 type = elk_inst_3src_a16_src_type(devinfo, inst);
1559
1560 if (elk_inst_3src_a16_src2_rep_ctrl(devinfo, inst)) {
1561 _vert_stride = ELK_VERTICAL_STRIDE_0;
1562 _width = ELK_WIDTH_1;
1563 _horiz_stride = ELK_HORIZONTAL_STRIDE_0;
1564 } else {
1565 _vert_stride = ELK_VERTICAL_STRIDE_4;
1566 _width = ELK_WIDTH_4;
1567 _horiz_stride = ELK_HORIZONTAL_STRIDE_1;
1568 }
1569 }
1570 is_scalar_region = _vert_stride == ELK_VERTICAL_STRIDE_0 &&
1571 _width == ELK_WIDTH_1 &&
1572 _horiz_stride == ELK_HORIZONTAL_STRIDE_0;
1573
1574 subreg_nr /= elk_reg_type_to_size(type);
1575
1576 err |= control(file, "negate", m_negate,
1577 elk_inst_3src_src2_negate(devinfo, inst), NULL);
1578 err |= control(file, "abs", _abs, elk_inst_3src_src2_abs(devinfo, inst), NULL);
1579
1580 err |= reg(file, _file, reg_nr);
1581 if (err == -1)
1582 return 0;
1583 if (subreg_nr || is_scalar_region)
1584 format(file, ".%d", subreg_nr);
1585 src_align1_region(file, _vert_stride, _width, _horiz_stride);
1586 if (!is_scalar_region && !is_align1)
1587 err |= src_swizzle(file, elk_inst_3src_a16_src2_swizzle(devinfo, inst));
1588 string(file, elk_reg_type_to_letters(type));
1589 return err;
1590 }
1591
1592 static int
src0_dpas_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1593 src0_dpas_3src(FILE *file, const struct intel_device_info *devinfo,
1594 const elk_inst *inst)
1595 {
1596 uint32_t reg_file = elk_inst_dpas_3src_src0_reg_file(devinfo, inst);
1597
1598 if (reg(file, reg_file, elk_inst_dpas_3src_src0_reg_nr(devinfo, inst)) == -1)
1599 return 0;
1600
1601 unsigned subreg_nr = elk_inst_dpas_3src_src0_subreg_nr(devinfo, inst);
1602 enum elk_reg_type type = elk_inst_dpas_3src_src0_type(devinfo, inst);
1603
1604 if (subreg_nr)
1605 format(file, ".%d", subreg_nr);
1606 src_align1_region(file, 1, 1, 0);
1607
1608 string(file, elk_reg_type_to_letters(type));
1609
1610 return 0;
1611 }
1612
1613 static int
src1_dpas_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1614 src1_dpas_3src(FILE *file, const struct intel_device_info *devinfo,
1615 const elk_inst *inst)
1616 {
1617 uint32_t reg_file = elk_inst_dpas_3src_src1_reg_file(devinfo, inst);
1618
1619 if (reg(file, reg_file, elk_inst_dpas_3src_src1_reg_nr(devinfo, inst)) == -1)
1620 return 0;
1621
1622 unsigned subreg_nr = elk_inst_dpas_3src_src1_subreg_nr(devinfo, inst);
1623 enum elk_reg_type type = elk_inst_dpas_3src_src1_type(devinfo, inst);
1624
1625 if (subreg_nr)
1626 format(file, ".%d", subreg_nr);
1627 src_align1_region(file, 1, 1, 0);
1628
1629 string(file, elk_reg_type_to_letters(type));
1630
1631 return 0;
1632 }
1633
1634 static int
src2_dpas_3src(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1635 src2_dpas_3src(FILE *file, const struct intel_device_info *devinfo,
1636 const elk_inst *inst)
1637 {
1638 uint32_t reg_file = elk_inst_dpas_3src_src2_reg_file(devinfo, inst);
1639
1640 if (reg(file, reg_file, elk_inst_dpas_3src_src2_reg_nr(devinfo, inst)) == -1)
1641 return 0;
1642
1643 unsigned subreg_nr = elk_inst_dpas_3src_src2_subreg_nr(devinfo, inst);
1644 enum elk_reg_type type = elk_inst_dpas_3src_src2_type(devinfo, inst);
1645
1646 if (subreg_nr)
1647 format(file, ".%d", subreg_nr);
1648 src_align1_region(file, 1, 1, 0);
1649
1650 string(file, elk_reg_type_to_letters(type));
1651
1652 return 0;
1653 }
1654
1655 static int
imm(FILE * file,const struct elk_isa_info * isa,enum elk_reg_type type,const elk_inst * inst)1656 imm(FILE *file, const struct elk_isa_info *isa, enum elk_reg_type type,
1657 const elk_inst *inst)
1658 {
1659 const struct intel_device_info *devinfo = isa->devinfo;
1660
1661 switch (type) {
1662 case ELK_REGISTER_TYPE_UQ:
1663 format(file, "0x%016"PRIx64"UQ", elk_inst_imm_uq(devinfo, inst));
1664 break;
1665 case ELK_REGISTER_TYPE_Q:
1666 format(file, "0x%016"PRIx64"Q", elk_inst_imm_uq(devinfo, inst));
1667 break;
1668 case ELK_REGISTER_TYPE_UD:
1669 format(file, "0x%08xUD", elk_inst_imm_ud(devinfo, inst));
1670 break;
1671 case ELK_REGISTER_TYPE_D:
1672 format(file, "%dD", elk_inst_imm_d(devinfo, inst));
1673 break;
1674 case ELK_REGISTER_TYPE_UW:
1675 format(file, "0x%04xUW", (uint16_t) elk_inst_imm_ud(devinfo, inst));
1676 break;
1677 case ELK_REGISTER_TYPE_W:
1678 format(file, "%dW", (int16_t) elk_inst_imm_d(devinfo, inst));
1679 break;
1680 case ELK_REGISTER_TYPE_UV:
1681 format(file, "0x%08xUV", elk_inst_imm_ud(devinfo, inst));
1682 break;
1683 case ELK_REGISTER_TYPE_VF:
1684 format(file, "0x%"PRIx64"VF", elk_inst_bits(inst, 127, 96));
1685 pad(file, 48);
1686 format(file, "/* [%-gF, %-gF, %-gF, %-gF]VF */",
1687 elk_vf_to_float(elk_inst_imm_ud(devinfo, inst)),
1688 elk_vf_to_float(elk_inst_imm_ud(devinfo, inst) >> 8),
1689 elk_vf_to_float(elk_inst_imm_ud(devinfo, inst) >> 16),
1690 elk_vf_to_float(elk_inst_imm_ud(devinfo, inst) >> 24));
1691 break;
1692 case ELK_REGISTER_TYPE_V:
1693 format(file, "0x%08xV", elk_inst_imm_ud(devinfo, inst));
1694 break;
1695 case ELK_REGISTER_TYPE_F:
1696 /* The DIM instruction's src0 uses an F type but contains a
1697 * 64-bit immediate
1698 */
1699 if (elk_inst_opcode(isa, inst) == ELK_OPCODE_DIM) {
1700 format(file, "0x%"PRIx64"F", elk_inst_bits(inst, 127, 64));
1701 pad(file, 48);
1702 format(file, "/* %-gF */", elk_inst_imm_df(devinfo, inst));
1703 } else {
1704 format(file, "0x%"PRIx64"F", elk_inst_bits(inst, 127, 96));
1705 pad(file, 48);
1706 format(file, " /* %-gF */", elk_inst_imm_f(devinfo, inst));
1707 }
1708 break;
1709 case ELK_REGISTER_TYPE_DF:
1710 format(file, "0x%016"PRIx64"DF", elk_inst_imm_uq(devinfo, inst));
1711 pad(file, 48);
1712 format(file, "/* %-gDF */", elk_inst_imm_df(devinfo, inst));
1713 break;
1714 case ELK_REGISTER_TYPE_HF:
1715 format(file, "0x%04xHF",
1716 (uint16_t) elk_inst_imm_ud(devinfo, inst));
1717 pad(file, 48);
1718 format(file, "/* %-gHF */",
1719 _mesa_half_to_float((uint16_t) elk_inst_imm_ud(devinfo, inst)));
1720 break;
1721 case ELK_REGISTER_TYPE_NF:
1722 case ELK_REGISTER_TYPE_UB:
1723 case ELK_REGISTER_TYPE_B:
1724 format(file, "*** invalid immediate type %d ", type);
1725 }
1726 return 0;
1727 }
1728
1729 static int
src_sends_da(FILE * file,const struct intel_device_info * devinfo,enum elk_reg_type type,enum elk_reg_file _reg_file,unsigned _reg_nr,unsigned _reg_subnr)1730 src_sends_da(FILE *file,
1731 const struct intel_device_info *devinfo,
1732 enum elk_reg_type type,
1733 enum elk_reg_file _reg_file,
1734 unsigned _reg_nr,
1735 unsigned _reg_subnr)
1736 {
1737 int err = 0;
1738
1739 err |= reg(file, _reg_file, _reg_nr);
1740 if (err == -1)
1741 return 0;
1742 if (_reg_subnr)
1743 format(file, ".1");
1744 string(file, elk_reg_type_to_letters(type));
1745
1746 return err;
1747 }
1748
1749 static int
src_sends_ia(FILE * file,const struct intel_device_info * devinfo,enum elk_reg_type type,int _addr_imm,unsigned _addr_subreg_nr)1750 src_sends_ia(FILE *file,
1751 const struct intel_device_info *devinfo,
1752 enum elk_reg_type type,
1753 int _addr_imm,
1754 unsigned _addr_subreg_nr)
1755 {
1756 string(file, "g[a0");
1757 if (_addr_subreg_nr)
1758 format(file, ".1");
1759 if (_addr_imm)
1760 format(file, " %d", _addr_imm);
1761 string(file, "]");
1762 string(file, elk_reg_type_to_letters(type));
1763
1764 return 0;
1765 }
1766
1767 static int
src_send_desc_ia(FILE * file,const struct intel_device_info * devinfo,unsigned _addr_subreg_nr)1768 src_send_desc_ia(FILE *file,
1769 const struct intel_device_info *devinfo,
1770 unsigned _addr_subreg_nr)
1771 {
1772 string(file, "a0");
1773 if (_addr_subreg_nr)
1774 format(file, ".%d", _addr_subreg_nr);
1775 format(file, "<0>UD");
1776
1777 return 0;
1778 }
1779
1780 static int
src0(FILE * file,const struct elk_isa_info * isa,const elk_inst * inst)1781 src0(FILE *file, const struct elk_isa_info *isa, const elk_inst *inst)
1782 {
1783 const struct intel_device_info *devinfo = isa->devinfo;
1784
1785 if (is_split_send(devinfo, elk_inst_opcode(isa, inst))) {
1786 if (devinfo->ver >= 12) {
1787 return src_sends_da(file,
1788 devinfo,
1789 ELK_REGISTER_TYPE_UD,
1790 elk_inst_send_src0_reg_file(devinfo, inst),
1791 elk_inst_src0_da_reg_nr(devinfo, inst),
1792 0);
1793 } else if (elk_inst_send_src0_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1794 return src_sends_da(file,
1795 devinfo,
1796 ELK_REGISTER_TYPE_UD,
1797 ELK_GENERAL_REGISTER_FILE,
1798 elk_inst_src0_da_reg_nr(devinfo, inst),
1799 elk_inst_src0_da16_subreg_nr(devinfo, inst));
1800 } else {
1801 return src_sends_ia(file,
1802 devinfo,
1803 ELK_REGISTER_TYPE_UD,
1804 elk_inst_send_src0_ia16_addr_imm(devinfo, inst),
1805 elk_inst_src0_ia_subreg_nr(devinfo, inst));
1806 }
1807 } else if (elk_inst_src0_reg_file(devinfo, inst) == ELK_IMMEDIATE_VALUE) {
1808 return imm(file, isa, elk_inst_src0_type(devinfo, inst), inst);
1809 } else if (elk_inst_access_mode(devinfo, inst) == ELK_ALIGN_1) {
1810 if (elk_inst_src0_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1811 return src_da1(file,
1812 devinfo,
1813 elk_inst_opcode(isa, inst),
1814 elk_inst_src0_type(devinfo, inst),
1815 elk_inst_src0_reg_file(devinfo, inst),
1816 elk_inst_src0_vstride(devinfo, inst),
1817 elk_inst_src0_width(devinfo, inst),
1818 elk_inst_src0_hstride(devinfo, inst),
1819 elk_inst_src0_da_reg_nr(devinfo, inst),
1820 elk_inst_src0_da1_subreg_nr(devinfo, inst),
1821 elk_inst_src0_abs(devinfo, inst),
1822 elk_inst_src0_negate(devinfo, inst));
1823 } else {
1824 return src_ia1(file,
1825 devinfo,
1826 elk_inst_opcode(isa, inst),
1827 elk_inst_src0_type(devinfo, inst),
1828 elk_inst_src0_ia1_addr_imm(devinfo, inst),
1829 elk_inst_src0_ia_subreg_nr(devinfo, inst),
1830 elk_inst_src0_negate(devinfo, inst),
1831 elk_inst_src0_abs(devinfo, inst),
1832 elk_inst_src0_hstride(devinfo, inst),
1833 elk_inst_src0_width(devinfo, inst),
1834 elk_inst_src0_vstride(devinfo, inst));
1835 }
1836 } else {
1837 if (elk_inst_src0_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1838 return src_da16(file,
1839 devinfo,
1840 elk_inst_opcode(isa, inst),
1841 elk_inst_src0_type(devinfo, inst),
1842 elk_inst_src0_reg_file(devinfo, inst),
1843 elk_inst_src0_vstride(devinfo, inst),
1844 elk_inst_src0_da_reg_nr(devinfo, inst),
1845 elk_inst_src0_da16_subreg_nr(devinfo, inst),
1846 elk_inst_src0_abs(devinfo, inst),
1847 elk_inst_src0_negate(devinfo, inst),
1848 elk_inst_src0_da16_swiz_x(devinfo, inst),
1849 elk_inst_src0_da16_swiz_y(devinfo, inst),
1850 elk_inst_src0_da16_swiz_z(devinfo, inst),
1851 elk_inst_src0_da16_swiz_w(devinfo, inst));
1852 } else {
1853 string(file, "Indirect align16 address mode not supported");
1854 return 1;
1855 }
1856 }
1857 }
1858
1859 static int
src1(FILE * file,const struct elk_isa_info * isa,const elk_inst * inst)1860 src1(FILE *file, const struct elk_isa_info *isa, const elk_inst *inst)
1861 {
1862 const struct intel_device_info *devinfo = isa->devinfo;
1863
1864 if (is_split_send(devinfo, elk_inst_opcode(isa, inst))) {
1865 return src_sends_da(file,
1866 devinfo,
1867 ELK_REGISTER_TYPE_UD,
1868 elk_inst_send_src1_reg_file(devinfo, inst),
1869 elk_inst_send_src1_reg_nr(devinfo, inst),
1870 0 /* subreg_nr */);
1871 } else if (elk_inst_src1_reg_file(devinfo, inst) == ELK_IMMEDIATE_VALUE) {
1872 return imm(file, isa, elk_inst_src1_type(devinfo, inst), inst);
1873 } else if (elk_inst_access_mode(devinfo, inst) == ELK_ALIGN_1) {
1874 if (elk_inst_src1_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1875 return src_da1(file,
1876 devinfo,
1877 elk_inst_opcode(isa, inst),
1878 elk_inst_src1_type(devinfo, inst),
1879 elk_inst_src1_reg_file(devinfo, inst),
1880 elk_inst_src1_vstride(devinfo, inst),
1881 elk_inst_src1_width(devinfo, inst),
1882 elk_inst_src1_hstride(devinfo, inst),
1883 elk_inst_src1_da_reg_nr(devinfo, inst),
1884 elk_inst_src1_da1_subreg_nr(devinfo, inst),
1885 elk_inst_src1_abs(devinfo, inst),
1886 elk_inst_src1_negate(devinfo, inst));
1887 } else {
1888 return src_ia1(file,
1889 devinfo,
1890 elk_inst_opcode(isa, inst),
1891 elk_inst_src1_type(devinfo, inst),
1892 elk_inst_src1_ia1_addr_imm(devinfo, inst),
1893 elk_inst_src1_ia_subreg_nr(devinfo, inst),
1894 elk_inst_src1_negate(devinfo, inst),
1895 elk_inst_src1_abs(devinfo, inst),
1896 elk_inst_src1_hstride(devinfo, inst),
1897 elk_inst_src1_width(devinfo, inst),
1898 elk_inst_src1_vstride(devinfo, inst));
1899 }
1900 } else {
1901 if (elk_inst_src1_address_mode(devinfo, inst) == ELK_ADDRESS_DIRECT) {
1902 return src_da16(file,
1903 devinfo,
1904 elk_inst_opcode(isa, inst),
1905 elk_inst_src1_type(devinfo, inst),
1906 elk_inst_src1_reg_file(devinfo, inst),
1907 elk_inst_src1_vstride(devinfo, inst),
1908 elk_inst_src1_da_reg_nr(devinfo, inst),
1909 elk_inst_src1_da16_subreg_nr(devinfo, inst),
1910 elk_inst_src1_abs(devinfo, inst),
1911 elk_inst_src1_negate(devinfo, inst),
1912 elk_inst_src1_da16_swiz_x(devinfo, inst),
1913 elk_inst_src1_da16_swiz_y(devinfo, inst),
1914 elk_inst_src1_da16_swiz_z(devinfo, inst),
1915 elk_inst_src1_da16_swiz_w(devinfo, inst));
1916 } else {
1917 string(file, "Indirect align16 address mode not supported");
1918 return 1;
1919 }
1920 }
1921 }
1922
1923 static int
qtr_ctrl(FILE * file,const struct intel_device_info * devinfo,const elk_inst * inst)1924 qtr_ctrl(FILE *file, const struct intel_device_info *devinfo,
1925 const elk_inst *inst)
1926 {
1927 int qtr_ctl = elk_inst_qtr_control(devinfo, inst);
1928 int exec_size = 1 << elk_inst_exec_size(devinfo, inst);
1929 const unsigned nib_ctl = devinfo->ver < 7 || devinfo->ver >= 20 ? 0 :
1930 elk_inst_nib_control(devinfo, inst);
1931
1932 if (exec_size < 8 || nib_ctl) {
1933 format(file, " %dN", qtr_ctl * 2 + nib_ctl + 1);
1934 } else if (exec_size == 8) {
1935 switch (qtr_ctl) {
1936 case 0:
1937 string(file, " 1Q");
1938 break;
1939 case 1:
1940 string(file, " 2Q");
1941 break;
1942 case 2:
1943 string(file, " 3Q");
1944 break;
1945 case 3:
1946 string(file, " 4Q");
1947 break;
1948 }
1949 } else if (exec_size == 16) {
1950 if (qtr_ctl < 2)
1951 string(file, " 1H");
1952 else
1953 string(file, " 2H");
1954 }
1955 return 0;
1956 }
1957
1958 static bool
inst_has_type(const struct elk_isa_info * isa,const elk_inst * inst,enum elk_reg_type type)1959 inst_has_type(const struct elk_isa_info *isa,
1960 const elk_inst *inst,
1961 enum elk_reg_type type)
1962 {
1963 const struct intel_device_info *devinfo = isa->devinfo;
1964 const unsigned num_sources = elk_num_sources_from_inst(isa, inst);
1965
1966 if (elk_inst_dst_type(devinfo, inst) == type)
1967 return true;
1968
1969 if (num_sources >= 3) {
1970 if (elk_inst_3src_access_mode(devinfo, inst) == ELK_ALIGN_1)
1971 return elk_inst_3src_a1_src0_type(devinfo, inst) == type ||
1972 elk_inst_3src_a1_src1_type(devinfo, inst) == type ||
1973 elk_inst_3src_a1_src2_type(devinfo, inst) == type;
1974 else
1975 return elk_inst_3src_a16_src_type(devinfo, inst) == type;
1976 } else if (num_sources == 2) {
1977 return elk_inst_src0_type(devinfo, inst) == type ||
1978 elk_inst_src1_type(devinfo, inst) == type;
1979 } else {
1980 return elk_inst_src0_type(devinfo, inst) == type;
1981 }
1982 }
1983
1984 static int
swsb(FILE * file,const struct elk_isa_info * isa,const elk_inst * inst)1985 swsb(FILE *file, const struct elk_isa_info *isa, const elk_inst *inst)
1986 {
1987 const struct intel_device_info *devinfo = isa->devinfo;
1988 const enum elk_opcode opcode = elk_inst_opcode(isa, inst);
1989 const uint32_t x = elk_inst_swsb(devinfo, inst);
1990 const bool is_unordered =
1991 opcode == ELK_OPCODE_SEND || opcode == ELK_OPCODE_SENDC ||
1992 opcode == ELK_OPCODE_MATH || opcode == ELK_OPCODE_DPAS ||
1993 (devinfo->has_64bit_float_via_math_pipe &&
1994 inst_has_type(isa, inst, ELK_REGISTER_TYPE_DF));
1995 const struct tgl_swsb swsb = tgl_swsb_decode(devinfo, is_unordered, x);
1996 if (swsb.regdist)
1997 format(file, " %s@%d",
1998 (swsb.pipe == TGL_PIPE_FLOAT ? "F" :
1999 swsb.pipe == TGL_PIPE_INT ? "I" :
2000 swsb.pipe == TGL_PIPE_LONG ? "L" :
2001 swsb.pipe == TGL_PIPE_ALL ? "A" : "" ),
2002 swsb.regdist);
2003 if (swsb.mode)
2004 format(file, " $%d%s", swsb.sbid,
2005 (swsb.mode & TGL_SBID_SET ? "" :
2006 swsb.mode & TGL_SBID_DST ? ".dst" : ".src"));
2007 return 0;
2008 }
2009
2010 #ifdef DEBUG
2011 static __attribute__((__unused__)) int
elk_disassemble_imm(const struct elk_isa_info * isa,uint32_t dw3,uint32_t dw2,uint32_t dw1,uint32_t dw0)2012 elk_disassemble_imm(const struct elk_isa_info *isa,
2013 uint32_t dw3, uint32_t dw2, uint32_t dw1, uint32_t dw0)
2014 {
2015 elk_inst inst;
2016 inst.data[0] = (((uint64_t) dw1) << 32) | ((uint64_t) dw0);
2017 inst.data[1] = (((uint64_t) dw3) << 32) | ((uint64_t) dw2);
2018 return elk_disassemble_inst(stderr, isa, &inst, false, 0, NULL);
2019 }
2020 #endif
2021
2022 static void
write_label(FILE * file,const struct intel_device_info * devinfo,const struct elk_label * root_label,int offset,int jump)2023 write_label(FILE *file, const struct intel_device_info *devinfo,
2024 const struct elk_label *root_label,
2025 int offset, int jump)
2026 {
2027 if (root_label != NULL) {
2028 int to_bytes_scale = sizeof(elk_inst) / elk_jump_scale(devinfo);
2029 const struct elk_label *label =
2030 elk_find_label(root_label, offset + jump * to_bytes_scale);
2031 if (label != NULL) {
2032 format(file, " LABEL%d", label->number);
2033 }
2034 }
2035 }
2036
2037 static void
lsc_disassemble_ex_desc(const struct intel_device_info * devinfo,uint32_t imm_desc,uint32_t imm_ex_desc,FILE * file)2038 lsc_disassemble_ex_desc(const struct intel_device_info *devinfo,
2039 uint32_t imm_desc,
2040 uint32_t imm_ex_desc,
2041 FILE *file)
2042 {
2043 const unsigned addr_type = lsc_msg_desc_addr_type(devinfo, imm_desc);
2044 switch (addr_type) {
2045 case LSC_ADDR_SURFTYPE_FLAT:
2046 format(file, " base_offset %u ",
2047 lsc_flat_ex_desc_base_offset(devinfo, imm_ex_desc));
2048 break;
2049 case LSC_ADDR_SURFTYPE_BSS:
2050 case LSC_ADDR_SURFTYPE_SS:
2051 format(file, " surface_state_index %u ",
2052 lsc_bss_ex_desc_index(devinfo, imm_ex_desc));
2053 break;
2054 case LSC_ADDR_SURFTYPE_BTI:
2055 format(file, " BTI %u ",
2056 lsc_bti_ex_desc_index(devinfo, imm_ex_desc));
2057 format(file, " base_offset %u ",
2058 lsc_bti_ex_desc_base_offset(devinfo, imm_ex_desc));
2059 break;
2060 default:
2061 format(file, "unsupported address surface type %d", addr_type);
2062 break;
2063 }
2064 }
2065
2066 static inline bool
elk_sfid_is_lsc(unsigned sfid)2067 elk_sfid_is_lsc(unsigned sfid)
2068 {
2069 switch (sfid) {
2070 case GFX12_SFID_UGM:
2071 case GFX12_SFID_SLM:
2072 case GFX12_SFID_TGM:
2073 return true;
2074 default:
2075 break;
2076 }
2077
2078 return false;
2079 }
2080
2081 int
elk_disassemble_inst(FILE * file,const struct elk_isa_info * isa,const elk_inst * inst,bool is_compacted,int offset,const struct elk_label * root_label)2082 elk_disassemble_inst(FILE *file, const struct elk_isa_info *isa,
2083 const elk_inst *inst, bool is_compacted,
2084 int offset, const struct elk_label *root_label)
2085 {
2086 const struct intel_device_info *devinfo = isa->devinfo;
2087
2088 int err = 0;
2089 int space = 0;
2090
2091 const enum elk_opcode opcode = elk_inst_opcode(isa, inst);
2092 const struct elk_opcode_desc *desc = elk_opcode_desc(isa, opcode);
2093
2094 if (elk_inst_pred_control(devinfo, inst)) {
2095 string(file, "(");
2096 err |= control(file, "predicate inverse", pred_inv,
2097 elk_inst_pred_inv(devinfo, inst), NULL);
2098 format(file, "f%"PRIu64".%"PRIu64,
2099 devinfo->ver >= 7 ? elk_inst_flag_reg_nr(devinfo, inst) : 0,
2100 elk_inst_flag_subreg_nr(devinfo, inst));
2101 if (devinfo->ver >= 20) {
2102 err |= control(file, "predicate control", xe2_pred_ctrl,
2103 elk_inst_pred_control(devinfo, inst), NULL);
2104 } else if (elk_inst_access_mode(devinfo, inst) == ELK_ALIGN_1) {
2105 err |= control(file, "predicate control align1", pred_ctrl_align1,
2106 elk_inst_pred_control(devinfo, inst), NULL);
2107 } else {
2108 err |= control(file, "predicate control align16", elk_pred_ctrl_align16,
2109 elk_inst_pred_control(devinfo, inst), NULL);
2110 }
2111 string(file, ") ");
2112 }
2113
2114 err |= print_opcode(file, isa, opcode);
2115
2116 if (!is_send(opcode))
2117 err |= control(file, "saturate", saturate, elk_inst_saturate(devinfo, inst),
2118 NULL);
2119
2120 err |= control(file, "debug control", debug_ctrl,
2121 elk_inst_debug_control(devinfo, inst), NULL);
2122
2123 if (opcode == ELK_OPCODE_MATH) {
2124 string(file, " ");
2125 err |= control(file, "function", math_function,
2126 elk_inst_math_function(devinfo, inst), NULL);
2127
2128 } else if (opcode == ELK_OPCODE_SYNC) {
2129 string(file, " ");
2130 err |= control(file, "function", sync_function,
2131 elk_inst_cond_modifier(devinfo, inst), NULL);
2132
2133 } else if (opcode == ELK_OPCODE_DPAS) {
2134 string(file, ".");
2135
2136 err |= control(file, "systolic depth", dpas_systolic_depth,
2137 elk_inst_dpas_3src_sdepth(devinfo, inst), NULL);
2138
2139 const unsigned rcount = elk_inst_dpas_3src_rcount(devinfo, inst) + 1;
2140
2141 format(file, "x%d", rcount);
2142 } else if (!is_send(opcode) &&
2143 (devinfo->ver < 12 ||
2144 elk_inst_src0_reg_file(devinfo, inst) != ELK_IMMEDIATE_VALUE ||
2145 type_sz(elk_inst_src0_type(devinfo, inst)) < 8)) {
2146 err |= control(file, "conditional modifier", elk_conditional_modifier,
2147 elk_inst_cond_modifier(devinfo, inst), NULL);
2148
2149 /* If we're using the conditional modifier, print which flags reg is
2150 * used for it. Note that on gfx6+, the embedded-condition SEL and
2151 * control flow doesn't update flags.
2152 */
2153 if (elk_inst_cond_modifier(devinfo, inst) &&
2154 (devinfo->ver < 6 || (opcode != ELK_OPCODE_SEL &&
2155 opcode != ELK_OPCODE_CSEL &&
2156 opcode != ELK_OPCODE_IF &&
2157 opcode != ELK_OPCODE_WHILE))) {
2158 format(file, ".f%"PRIu64".%"PRIu64,
2159 devinfo->ver >= 7 ? elk_inst_flag_reg_nr(devinfo, inst) : 0,
2160 elk_inst_flag_subreg_nr(devinfo, inst));
2161 }
2162 }
2163
2164 if (opcode != ELK_OPCODE_NOP && opcode != ELK_OPCODE_NENOP) {
2165 string(file, "(");
2166 err |= control(file, "execution size", exec_size,
2167 elk_inst_exec_size(devinfo, inst), NULL);
2168 string(file, ")");
2169 }
2170
2171 if (opcode == ELK_OPCODE_SEND && devinfo->ver < 6)
2172 format(file, " %"PRIu64, elk_inst_base_mrf(devinfo, inst));
2173
2174 if (elk_has_uip(devinfo, opcode)) {
2175 /* Instructions that have UIP also have JIP. */
2176 pad(file, 16);
2177 string(file, "JIP: ");
2178 write_label(file, devinfo, root_label, offset, elk_inst_jip(devinfo, inst));
2179
2180 pad(file, 38);
2181 string(file, "UIP: ");
2182 write_label(file, devinfo, root_label, offset, elk_inst_uip(devinfo, inst));
2183 } else if (elk_has_jip(devinfo, opcode)) {
2184 int jip;
2185 if (devinfo->ver >= 7) {
2186 jip = elk_inst_jip(devinfo, inst);
2187 } else {
2188 jip = elk_inst_gfx6_jump_count(devinfo, inst);
2189 }
2190
2191 pad(file, 16);
2192 string(file, "JIP: ");
2193 write_label(file, devinfo, root_label, offset, jip);
2194 } else if (devinfo->ver < 6 && (opcode == ELK_OPCODE_BREAK ||
2195 opcode == ELK_OPCODE_CONTINUE ||
2196 opcode == ELK_OPCODE_ELSE)) {
2197 pad(file, 16);
2198 format(file, "Jump: %d", elk_inst_gfx4_jump_count(devinfo, inst));
2199 pad(file, 32);
2200 format(file, "Pop: %"PRIu64, elk_inst_gfx4_pop_count(devinfo, inst));
2201 } else if (devinfo->ver < 6 && (opcode == ELK_OPCODE_IF ||
2202 opcode == ELK_OPCODE_IFF ||
2203 opcode == ELK_OPCODE_HALT ||
2204 opcode == ELK_OPCODE_WHILE)) {
2205 pad(file, 16);
2206 format(file, "Jump: %d", elk_inst_gfx4_jump_count(devinfo, inst));
2207 } else if (devinfo->ver < 6 && opcode == ELK_OPCODE_ENDIF) {
2208 pad(file, 16);
2209 format(file, "Pop: %"PRIu64, elk_inst_gfx4_pop_count(devinfo, inst));
2210 } else if (opcode == ELK_OPCODE_JMPI) {
2211 pad(file, 16);
2212 err |= src1(file, isa, inst);
2213 } else if (opcode == ELK_OPCODE_DPAS) {
2214 pad(file, 16);
2215 err |= dest_dpas_3src(file, devinfo, inst);
2216
2217 pad(file, 32);
2218 err |= src0_dpas_3src(file, devinfo, inst);
2219
2220 pad(file, 48);
2221 err |= src1_dpas_3src(file, devinfo, inst);
2222
2223 pad(file, 64);
2224 err |= src2_dpas_3src(file, devinfo, inst);
2225
2226 } else if (desc && desc->nsrc == 3) {
2227 pad(file, 16);
2228 err |= dest_3src(file, devinfo, inst);
2229
2230 pad(file, 32);
2231 err |= src0_3src(file, devinfo, inst);
2232
2233 pad(file, 48);
2234 err |= src1_3src(file, devinfo, inst);
2235
2236 pad(file, 64);
2237 err |= src2_3src(file, devinfo, inst);
2238 } else if (desc) {
2239 if (desc->ndst > 0) {
2240 pad(file, 16);
2241 err |= dest(file, isa, inst);
2242 }
2243
2244 if (desc->nsrc > 0) {
2245 pad(file, 32);
2246 err |= src0(file, isa, inst);
2247 }
2248
2249 if (desc->nsrc > 1) {
2250 pad(file, 48);
2251 err |= src1(file, isa, inst);
2252 }
2253 }
2254
2255 if (is_send(opcode)) {
2256 enum elk_message_target sfid = elk_inst_sfid(devinfo, inst);
2257
2258 bool has_imm_desc = false, has_imm_ex_desc = false;
2259 uint32_t imm_desc = 0, imm_ex_desc = 0;
2260 if (is_split_send(devinfo, opcode)) {
2261 pad(file, 64);
2262 if (elk_inst_send_sel_reg32_desc(devinfo, inst)) {
2263 /* show the indirect descriptor source */
2264 err |= src_send_desc_ia(file, devinfo, 0);
2265 } else {
2266 has_imm_desc = true;
2267 imm_desc = elk_inst_send_desc(devinfo, inst);
2268 fprintf(file, "0x%08"PRIx32, imm_desc);
2269 }
2270
2271 pad(file, 80);
2272 if (elk_inst_send_sel_reg32_ex_desc(devinfo, inst)) {
2273 /* show the indirect descriptor source */
2274 err |= src_send_desc_ia(file, devinfo,
2275 elk_inst_send_ex_desc_ia_subreg_nr(devinfo, inst));
2276 } else {
2277 has_imm_ex_desc = true;
2278 imm_ex_desc = elk_inst_sends_ex_desc(devinfo, inst);
2279 fprintf(file, "0x%08"PRIx32, imm_ex_desc);
2280 }
2281 } else {
2282 if (elk_inst_src1_reg_file(devinfo, inst) != ELK_IMMEDIATE_VALUE) {
2283 /* show the indirect descriptor source */
2284 pad(file, 48);
2285 err |= src1(file, isa, inst);
2286 pad(file, 64);
2287 } else {
2288 has_imm_desc = true;
2289 imm_desc = elk_inst_send_desc(devinfo, inst);
2290 pad(file, 48);
2291 }
2292
2293 /* Print message descriptor as immediate source */
2294 fprintf(file, "0x%08"PRIx64, inst->data[1] >> 32);
2295 }
2296
2297 newline(file);
2298 pad(file, 16);
2299 space = 0;
2300
2301 fprintf(file, " ");
2302 err |= control(file, "SFID", devinfo->ver >= 6 ? gfx6_sfid : gfx4_sfid,
2303 sfid, &space);
2304 string(file, " MsgDesc:");
2305
2306 if (!has_imm_desc) {
2307 format(file, " indirect");
2308 } else {
2309 bool unsupported = false;
2310 switch (sfid) {
2311 case ELK_SFID_MATH:
2312 err |= control(file, "math function", math_function,
2313 elk_inst_math_msg_function(devinfo, inst), &space);
2314 err |= control(file, "math saturate", math_saturate,
2315 elk_inst_math_msg_saturate(devinfo, inst), &space);
2316 err |= control(file, "math signed", math_signed,
2317 elk_inst_math_msg_signed_int(devinfo, inst), &space);
2318 err |= control(file, "math scalar", math_scalar,
2319 elk_inst_math_msg_data_type(devinfo, inst), &space);
2320 err |= control(file, "math precision", math_precision,
2321 elk_inst_math_msg_precision(devinfo, inst), &space);
2322 break;
2323 case ELK_SFID_SAMPLER:
2324 if (devinfo->ver >= 20) {
2325 err |= control(file, "sampler message", xe2_sampler_msg_type,
2326 elk_sampler_desc_msg_type(devinfo, imm_desc),
2327 &space);
2328 err |= control(file, "sampler simd mode", xe2_sampler_simd_mode,
2329 elk_sampler_desc_simd_mode(devinfo, imm_desc),
2330 &space);
2331 if (elk_sampler_desc_return_format(devinfo, imm_desc)) {
2332 string(file, " HP");
2333 }
2334 format(file, " Surface = %u Sampler = %u",
2335 elk_sampler_desc_binding_table_index(devinfo, imm_desc),
2336 elk_sampler_desc_sampler(devinfo, imm_desc));
2337 } else if (devinfo->ver >= 5) {
2338 err |= control(file, "sampler message", gfx5_sampler_msg_type,
2339 elk_sampler_desc_msg_type(devinfo, imm_desc),
2340 &space);
2341 err |= control(file, "sampler simd mode", gfx5_sampler_simd_mode,
2342 elk_sampler_desc_simd_mode(devinfo, imm_desc),
2343 &space);
2344 if (devinfo->ver >= 8 &&
2345 elk_sampler_desc_return_format(devinfo, imm_desc)) {
2346 string(file, " HP");
2347 }
2348 format(file, " Surface = %u Sampler = %u",
2349 elk_sampler_desc_binding_table_index(devinfo, imm_desc),
2350 elk_sampler_desc_sampler(devinfo, imm_desc));
2351 } else {
2352 format(file, " (bti %u, sampler %u, msg_type %u, ",
2353 elk_sampler_desc_binding_table_index(devinfo, imm_desc),
2354 elk_sampler_desc_sampler(devinfo, imm_desc),
2355 elk_sampler_desc_msg_type(devinfo, imm_desc));
2356 if (devinfo->verx10 != 45) {
2357 err |= control(file, "sampler target format",
2358 sampler_target_format,
2359 elk_sampler_desc_return_format(devinfo, imm_desc),
2360 NULL);
2361 }
2362 string(file, ")");
2363 }
2364 break;
2365 case GFX6_SFID_DATAPORT_SAMPLER_CACHE:
2366 case GFX6_SFID_DATAPORT_CONSTANT_CACHE:
2367 /* aka ELK_SFID_DATAPORT_READ on Gfx4-5 */
2368 if (devinfo->ver >= 6) {
2369 format(file, " (bti %u, msg_ctrl %u, msg_type %u, write_commit %u)",
2370 elk_dp_desc_binding_table_index(devinfo, imm_desc),
2371 elk_dp_desc_msg_control(devinfo, imm_desc),
2372 elk_dp_desc_msg_type(devinfo, imm_desc),
2373 devinfo->ver >= 7 ? 0u :
2374 elk_dp_write_desc_write_commit(devinfo, imm_desc));
2375 } else {
2376 bool is_965 = devinfo->verx10 == 40;
2377 err |= control(file, "DP read message type",
2378 is_965 ? gfx4_dp_read_port_msg_type :
2379 g45_dp_read_port_msg_type,
2380 elk_dp_read_desc_msg_type(devinfo, imm_desc),
2381 &space);
2382
2383 format(file, " MsgCtrl = 0x%u",
2384 elk_dp_read_desc_msg_control(devinfo, imm_desc));
2385
2386 format(file, " Surface = %u",
2387 elk_dp_desc_binding_table_index(devinfo, imm_desc));
2388 }
2389 break;
2390
2391 case GFX6_SFID_DATAPORT_RENDER_CACHE: {
2392 /* aka ELK_SFID_DATAPORT_WRITE on Gfx4-5 */
2393 unsigned msg_type = elk_fb_write_desc_msg_type(devinfo, imm_desc);
2394
2395 err |= control(file, "DP rc message type",
2396 dp_rc_msg_type(devinfo), msg_type, &space);
2397
2398 bool is_rt_write = msg_type ==
2399 (devinfo->ver >= 6 ? GFX6_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE
2400 : ELK_DATAPORT_WRITE_MESSAGE_RENDER_TARGET_WRITE);
2401
2402 if (is_rt_write) {
2403 err |= control(file, "RT message type", m_rt_write_subtype,
2404 elk_inst_rt_message_type(devinfo, inst), &space);
2405 if (devinfo->ver >= 6 && elk_inst_rt_slot_group(devinfo, inst))
2406 string(file, " Hi");
2407 if (elk_fb_write_desc_last_render_target(devinfo, imm_desc))
2408 string(file, " LastRT");
2409 if (devinfo->ver >= 10 &&
2410 elk_fb_write_desc_coarse_write(devinfo, imm_desc))
2411 string(file, " CoarseWrite");
2412 if (devinfo->ver < 7 &&
2413 elk_fb_write_desc_write_commit(devinfo, imm_desc))
2414 string(file, " WriteCommit");
2415 } else {
2416 format(file, " MsgCtrl = 0x%u",
2417 elk_fb_write_desc_msg_control(devinfo, imm_desc));
2418 }
2419
2420 format(file, " Surface = %u",
2421 elk_fb_desc_binding_table_index(devinfo, imm_desc));
2422 break;
2423 }
2424
2425 case ELK_SFID_URB: {
2426 if (devinfo->ver >= 20) {
2427 format(file, " (");
2428 const enum elk_lsc_opcode op = lsc_msg_desc_opcode(devinfo, imm_desc);
2429 err |= control(file, "operation", lsc_operation,
2430 op, &space);
2431 format(file, ",");
2432 err |= control(file, "addr_size", lsc_addr_size,
2433 lsc_msg_desc_addr_size(devinfo, imm_desc),
2434 &space);
2435
2436 format(file, ",");
2437 err |= control(file, "data_size", lsc_data_size,
2438 lsc_msg_desc_data_size(devinfo, imm_desc),
2439 &space);
2440 format(file, ",");
2441 if (elk_lsc_opcode_has_cmask(op)) {
2442 err |= control(file, "component_mask",
2443 lsc_cmask_str,
2444 lsc_msg_desc_cmask(devinfo, imm_desc),
2445 &space);
2446 } else {
2447 err |= control(file, "vector_size",
2448 lsc_vect_size_str,
2449 lsc_msg_desc_vect_size(devinfo, imm_desc),
2450 &space);
2451 if (lsc_msg_desc_transpose(devinfo, imm_desc))
2452 format(file, ", transpose");
2453 }
2454 switch(op) {
2455 case LSC_OP_LOAD_CMASK:
2456 case LSC_OP_LOAD:
2457 format(file, ",");
2458 err |= control(file, "cache_load",
2459 lsc_cache_load,
2460 lsc_msg_desc_cache_ctrl(devinfo, imm_desc),
2461 &space);
2462 break;
2463 default:
2464 format(file, ",");
2465 err |= control(file, "cache_store",
2466 lsc_cache_store,
2467 lsc_msg_desc_cache_ctrl(devinfo, imm_desc),
2468 &space);
2469 break;
2470 }
2471
2472 format(file, " dst_len = %u,", lsc_msg_desc_dest_len(devinfo, imm_desc));
2473 format(file, " src0_len = %u,", lsc_msg_desc_src0_len(devinfo, imm_desc));
2474 format(file, " src1_len = %d", elk_message_ex_desc_ex_mlen(devinfo, imm_ex_desc));
2475 err |= control(file, "address_type", lsc_addr_surface_type,
2476 lsc_msg_desc_addr_type(devinfo, imm_desc), &space);
2477 format(file, " )");
2478 } else {
2479 unsigned urb_opcode = elk_inst_urb_opcode(devinfo, inst);
2480
2481 format(file, " offset %"PRIu64, elk_inst_urb_global_offset(devinfo, inst));
2482
2483 space = 1;
2484
2485 err |= control(file, "urb opcode",
2486 devinfo->ver >= 7 ? gfx7_urb_opcode
2487 : gfx5_urb_opcode,
2488 urb_opcode, &space);
2489
2490 if (devinfo->ver >= 7 &&
2491 elk_inst_urb_per_slot_offset(devinfo, inst)) {
2492 string(file, " per-slot");
2493 }
2494
2495 if (urb_opcode == GFX8_URB_OPCODE_SIMD8_WRITE ||
2496 urb_opcode == GFX8_URB_OPCODE_SIMD8_READ) {
2497 if (elk_inst_urb_channel_mask_present(devinfo, inst))
2498 string(file, " masked");
2499 } else if (urb_opcode != GFX125_URB_OPCODE_FENCE) {
2500 err |= control(file, "urb swizzle", urb_swizzle,
2501 elk_inst_urb_swizzle_control(devinfo, inst),
2502 &space);
2503 }
2504
2505 if (devinfo->ver < 7) {
2506 err |= control(file, "urb allocate", urb_allocate,
2507 elk_inst_urb_allocate(devinfo, inst), &space);
2508 err |= control(file, "urb used", urb_used,
2509 elk_inst_urb_used(devinfo, inst), &space);
2510 }
2511 if (devinfo->ver < 8) {
2512 err |= control(file, "urb complete", urb_complete,
2513 elk_inst_urb_complete(devinfo, inst), &space);
2514 }
2515 }
2516 break;
2517 }
2518 case ELK_SFID_THREAD_SPAWNER:
2519 break;
2520
2521 case ELK_SFID_MESSAGE_GATEWAY:
2522 format(file, " (%s)",
2523 gfx7_gateway_subfuncid[elk_inst_gateway_subfuncid(devinfo, inst)]);
2524 break;
2525
2526 case GFX12_SFID_SLM:
2527 case GFX12_SFID_TGM:
2528 case GFX12_SFID_UGM: {
2529 assert(devinfo->has_lsc);
2530 format(file, " (");
2531 const enum elk_lsc_opcode op = lsc_msg_desc_opcode(devinfo, imm_desc);
2532 err |= control(file, "operation", lsc_operation,
2533 op, &space);
2534 format(file, ",");
2535 err |= control(file, "addr_size", lsc_addr_size,
2536 lsc_msg_desc_addr_size(devinfo, imm_desc),
2537 &space);
2538
2539 if (op == LSC_OP_FENCE) {
2540 format(file, ",");
2541 err |= control(file, "scope", lsc_fence_scope,
2542 lsc_fence_msg_desc_scope(devinfo, imm_desc),
2543 &space);
2544 format(file, ",");
2545 err |= control(file, "flush_type", lsc_flush_type,
2546 lsc_fence_msg_desc_flush_type(devinfo, imm_desc),
2547 &space);
2548 format(file, ",");
2549 err |= control(file, "backup_mode_fence_routing",
2550 lsc_backup_fence_routing,
2551 lsc_fence_msg_desc_backup_routing(devinfo, imm_desc),
2552 &space);
2553 } else {
2554 format(file, ",");
2555 err |= control(file, "data_size", lsc_data_size,
2556 lsc_msg_desc_data_size(devinfo, imm_desc),
2557 &space);
2558 format(file, ",");
2559 if (elk_lsc_opcode_has_cmask(op)) {
2560 err |= control(file, "component_mask",
2561 lsc_cmask_str,
2562 lsc_msg_desc_cmask(devinfo, imm_desc),
2563 &space);
2564 } else {
2565 err |= control(file, "vector_size",
2566 lsc_vect_size_str,
2567 lsc_msg_desc_vect_size(devinfo, imm_desc),
2568 &space);
2569 if (lsc_msg_desc_transpose(devinfo, imm_desc))
2570 format(file, ", transpose");
2571 }
2572 switch(op) {
2573 case LSC_OP_LOAD_CMASK:
2574 case LSC_OP_LOAD:
2575 format(file, ",");
2576 err |= control(file, "cache_load",
2577 devinfo->ver >= 20 ?
2578 xe2_lsc_cache_load :
2579 lsc_cache_load,
2580 lsc_msg_desc_cache_ctrl(devinfo, imm_desc),
2581 &space);
2582 break;
2583 default:
2584 format(file, ",");
2585 err |= control(file, "cache_store",
2586 devinfo->ver >= 20 ?
2587 xe2_lsc_cache_store :
2588 lsc_cache_store,
2589 lsc_msg_desc_cache_ctrl(devinfo, imm_desc),
2590 &space);
2591 break;
2592 }
2593 }
2594 format(file, " dst_len = %u,", lsc_msg_desc_dest_len(devinfo, imm_desc));
2595 format(file, " src0_len = %u,", lsc_msg_desc_src0_len(devinfo, imm_desc));
2596
2597 if (!elk_inst_send_sel_reg32_ex_desc(devinfo, inst))
2598 format(file, " src1_len = %d",
2599 elk_message_ex_desc_ex_mlen(devinfo, imm_ex_desc));
2600
2601 err |= control(file, "address_type", lsc_addr_surface_type,
2602 lsc_msg_desc_addr_type(devinfo, imm_desc), &space);
2603 format(file, " )");
2604 break;
2605 }
2606
2607 case GFX7_SFID_DATAPORT_DATA_CACHE:
2608 if (devinfo->ver >= 7) {
2609 format(file, " (");
2610 space = 0;
2611
2612 err |= control(file, "DP DC0 message type",
2613 dp_dc0_msg_type_gfx7,
2614 elk_dp_desc_msg_type(devinfo, imm_desc), &space);
2615
2616 format(file, ", bti %u, ",
2617 elk_dp_desc_binding_table_index(devinfo, imm_desc));
2618
2619 switch (elk_inst_dp_msg_type(devinfo, inst)) {
2620 case GFX7_DATAPORT_DC_UNTYPED_ATOMIC_OP:
2621 control(file, "atomic op", aop,
2622 elk_dp_desc_msg_control(devinfo, imm_desc) & 0xf,
2623 &space);
2624 break;
2625 case GFX7_DATAPORT_DC_OWORD_BLOCK_READ:
2626 case GFX7_DATAPORT_DC_OWORD_BLOCK_WRITE: {
2627 unsigned msg_ctrl = elk_dp_desc_msg_control(devinfo, imm_desc);
2628 assert(dp_oword_block_rw[msg_ctrl & 7]);
2629 format(file, "owords = %s, aligned = %d",
2630 dp_oword_block_rw[msg_ctrl & 7], (msg_ctrl >> 3) & 3);
2631 break;
2632 }
2633 default:
2634 format(file, "%u",
2635 elk_dp_desc_msg_control(devinfo, imm_desc));
2636 }
2637 format(file, ")");
2638 } else {
2639 unsupported = true;
2640 }
2641 break;
2642
2643 case HSW_SFID_DATAPORT_DATA_CACHE_1: {
2644 if (devinfo->ver >= 7) {
2645 format(file, " (");
2646 space = 0;
2647
2648 unsigned msg_ctrl = elk_dp_desc_msg_control(devinfo, imm_desc);
2649
2650 err |= control(file, "DP DC1 message type",
2651 dp_dc1_msg_type_hsw,
2652 elk_dp_desc_msg_type(devinfo, imm_desc), &space);
2653
2654 format(file, ", Surface = %u, ",
2655 elk_dp_desc_binding_table_index(devinfo, imm_desc));
2656
2657 switch (elk_inst_dp_msg_type(devinfo, inst)) {
2658 case HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP:
2659 case HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP:
2660 case HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP:
2661 format(file, "SIMD%d,", (msg_ctrl & (1 << 4)) ? 8 : 16);
2662 FALLTHROUGH;
2663 case HSW_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_OP_SIMD4X2:
2664 case HSW_DATAPORT_DC_PORT1_TYPED_ATOMIC_OP_SIMD4X2:
2665 case HSW_DATAPORT_DC_PORT1_ATOMIC_COUNTER_OP_SIMD4X2:
2666 case GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_OP:
2667 case GFX12_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_HALF_INT_OP:
2668 control(file, "atomic op", aop, msg_ctrl & 0xf, &space);
2669 break;
2670 case HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_READ:
2671 case HSW_DATAPORT_DC_PORT1_UNTYPED_SURFACE_WRITE:
2672 case HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_READ:
2673 case HSW_DATAPORT_DC_PORT1_TYPED_SURFACE_WRITE:
2674 case GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_WRITE:
2675 case GFX8_DATAPORT_DC_PORT1_A64_UNTYPED_SURFACE_READ: {
2676 static const char *simd_modes[] = { "4x2", "16", "8" };
2677 format(file, "SIMD%s, Mask = 0x%x",
2678 simd_modes[msg_ctrl >> 4], msg_ctrl & 0xf);
2679 break;
2680 }
2681 case GFX9_DATAPORT_DC_PORT1_UNTYPED_ATOMIC_FLOAT_OP:
2682 case GFX9_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_FLOAT_OP:
2683 case GFX12_DATAPORT_DC_PORT1_A64_UNTYPED_ATOMIC_HALF_FLOAT_OP:
2684 format(file, "SIMD%d,", (msg_ctrl & (1 << 4)) ? 8 : 16);
2685 control(file, "atomic float op", aop_float, msg_ctrl & 0xf,
2686 &space);
2687 break;
2688 case GFX9_DATAPORT_DC_PORT1_A64_OWORD_BLOCK_WRITE:
2689 case GFX9_DATAPORT_DC_PORT1_A64_OWORD_BLOCK_READ:
2690 assert(dp_oword_block_rw[msg_ctrl & 7]);
2691 format(file, "owords = %s, aligned = %d",
2692 dp_oword_block_rw[msg_ctrl & 7], (msg_ctrl >> 3) & 3);
2693 break;
2694 default:
2695 format(file, "0x%x", msg_ctrl);
2696 }
2697 format(file, ")");
2698 } else {
2699 unsupported = true;
2700 }
2701 break;
2702 }
2703
2704 case GFX7_SFID_PIXEL_INTERPOLATOR:
2705 if (devinfo->ver >= 7) {
2706 format(file, " (%s, %s, 0x%02"PRIx64")",
2707 elk_inst_pi_nopersp(devinfo, inst) ? "linear" : "persp",
2708 pixel_interpolator_msg_types[elk_inst_pi_message_type(devinfo, inst)],
2709 elk_inst_pi_message_data(devinfo, inst));
2710 } else {
2711 unsupported = true;
2712 }
2713 break;
2714
2715 default:
2716 unsupported = true;
2717 break;
2718 }
2719
2720 if (unsupported)
2721 format(file, "unsupported shared function ID %d", sfid);
2722
2723 if (space)
2724 string(file, " ");
2725 }
2726 if (devinfo->verx10 >= 125 &&
2727 elk_inst_send_sel_reg32_ex_desc(devinfo, inst) &&
2728 elk_inst_send_ex_bso(devinfo, inst)) {
2729 format(file, " src1_len = %u",
2730 (unsigned) elk_inst_send_src1_len(devinfo, inst));
2731
2732 format(file, " ex_bso");
2733 }
2734 if (elk_sfid_is_lsc(sfid) ||
2735 (sfid == ELK_SFID_URB && devinfo->ver >= 20)) {
2736 lsc_disassemble_ex_desc(devinfo, imm_desc, imm_ex_desc, file);
2737 } else {
2738 if (has_imm_desc)
2739 format(file, " mlen %u", elk_message_desc_mlen(devinfo, imm_desc));
2740 if (has_imm_ex_desc) {
2741 format(file, " ex_mlen %u",
2742 elk_message_ex_desc_ex_mlen(devinfo, imm_ex_desc));
2743 }
2744 if (has_imm_desc)
2745 format(file, " rlen %u", elk_message_desc_rlen(devinfo, imm_desc));
2746 }
2747 }
2748 pad(file, 64);
2749 if (opcode != ELK_OPCODE_NOP && opcode != ELK_OPCODE_NENOP) {
2750 string(file, "{");
2751 space = 1;
2752 err |= control(file, "access mode", access_mode,
2753 elk_inst_access_mode(devinfo, inst), &space);
2754 if (devinfo->ver >= 6) {
2755 err |= control(file, "write enable control", wectrl,
2756 elk_inst_mask_control(devinfo, inst), &space);
2757 } else {
2758 err |= control(file, "mask control", mask_ctrl,
2759 elk_inst_mask_control(devinfo, inst), &space);
2760 }
2761
2762 if (devinfo->ver < 12) {
2763 err |= control(file, "dependency control", dep_ctrl,
2764 ((elk_inst_no_dd_check(devinfo, inst) << 1) |
2765 elk_inst_no_dd_clear(devinfo, inst)), &space);
2766 }
2767
2768 if (devinfo->ver >= 6)
2769 err |= qtr_ctrl(file, devinfo, inst);
2770 else {
2771 if (elk_inst_qtr_control(devinfo, inst) == ELK_COMPRESSION_COMPRESSED &&
2772 desc && desc->ndst > 0 &&
2773 elk_inst_dst_reg_file(devinfo, inst) == ELK_MESSAGE_REGISTER_FILE &&
2774 elk_inst_dst_da_reg_nr(devinfo, inst) & ELK_MRF_COMPR4) {
2775 format(file, " compr4");
2776 } else {
2777 err |= control(file, "compression control", compr_ctrl,
2778 elk_inst_qtr_control(devinfo, inst), &space);
2779 }
2780 }
2781
2782 if (devinfo->ver >= 12)
2783 err |= swsb(file, isa, inst);
2784
2785 err |= control(file, "compaction", cmpt_ctrl, is_compacted, &space);
2786 err |= control(file, "thread control", thread_ctrl,
2787 (devinfo->ver >= 12 ? elk_inst_atomic_control(devinfo, inst) :
2788 elk_inst_thread_control(devinfo, inst)),
2789 &space);
2790 if (has_branch_ctrl(devinfo, opcode)) {
2791 err |= control(file, "branch ctrl", branch_ctrl,
2792 elk_inst_branch_control(devinfo, inst), &space);
2793 } else if (devinfo->ver >= 6 && devinfo->ver < 20) {
2794 err |= control(file, "acc write control", accwr,
2795 elk_inst_acc_wr_control(devinfo, inst), &space);
2796 }
2797 if (is_send(opcode))
2798 err |= control(file, "end of thread", end_of_thread,
2799 elk_inst_eot(devinfo, inst), &space);
2800 if (space)
2801 string(file, " ");
2802 string(file, "}");
2803 }
2804 string(file, ";");
2805 newline(file);
2806 return err;
2807 }
2808
2809 int
elk_disassemble_find_end(const struct elk_isa_info * isa,const void * assembly,int start)2810 elk_disassemble_find_end(const struct elk_isa_info *isa,
2811 const void *assembly, int start)
2812 {
2813 const struct intel_device_info *devinfo = isa->devinfo;
2814 int offset = start;
2815
2816 /* This loop exits when send-with-EOT or when opcode is 0 */
2817 while (true) {
2818 const elk_inst *insn = assembly + offset;
2819
2820 if (elk_inst_cmpt_control(devinfo, insn)) {
2821 offset += 8;
2822 } else {
2823 offset += 16;
2824 }
2825
2826 /* Simplistic, but efficient way to terminate disasm */
2827 uint32_t opcode = elk_inst_opcode(isa, insn);
2828 if (opcode == 0 || (is_send(opcode) && elk_inst_eot(devinfo, insn))) {
2829 break;
2830 }
2831 }
2832
2833 return offset;
2834 }
2835
2836 void
elk_disassemble_with_errors(const struct elk_isa_info * isa,const void * assembly,int start,FILE * out)2837 elk_disassemble_with_errors(const struct elk_isa_info *isa,
2838 const void *assembly, int start, FILE *out)
2839 {
2840 int end = elk_disassemble_find_end(isa, assembly, start);
2841
2842 /* Make a dummy disasm structure that elk_validate_instructions
2843 * can work from.
2844 */
2845 struct elk_disasm_info *elk_disasm_info = elk_disasm_initialize(isa, NULL);
2846 elk_disasm_new_inst_group(elk_disasm_info, start);
2847 elk_disasm_new_inst_group(elk_disasm_info, end);
2848
2849 elk_validate_instructions(isa, assembly, start, end, elk_disasm_info);
2850
2851 void *mem_ctx = ralloc_context(NULL);
2852 const struct elk_label *root_label =
2853 elk_label_assembly(isa, assembly, start, end, mem_ctx);
2854
2855 foreach_list_typed(struct inst_group, group, link,
2856 &elk_disasm_info->group_list) {
2857 struct exec_node *next_node = exec_node_get_next(&group->link);
2858 if (exec_node_is_tail_sentinel(next_node))
2859 break;
2860
2861 struct inst_group *next =
2862 exec_node_data(struct inst_group, next_node, link);
2863
2864 int start_offset = group->offset;
2865 int end_offset = next->offset;
2866
2867 elk_disassemble(isa, assembly, start_offset, end_offset,
2868 root_label, out);
2869
2870 if (group->error) {
2871 fputs(group->error, out);
2872 }
2873 }
2874
2875 ralloc_free(mem_ctx);
2876 ralloc_free(elk_disasm_info);
2877 }
2878