3d80b4e6bae90d00b22532d5688fd8192049bf0d
[mesa.git] / src / intel / common / gen_batch_decoder.c
1 /*
2 * Copyright © 2017 Intel Corporation
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "common/gen_decoder.h"
25 #include "gen_disasm.h"
26 #include "util/macros.h"
27 #include "main/macros.h" /* Needed for ROUND_DOWN_TO */
28
29 #include <string.h>
30
31 void
32 gen_batch_decode_ctx_init(struct gen_batch_decode_ctx *ctx,
33 const struct gen_device_info *devinfo,
34 FILE *fp, enum gen_batch_decode_flags flags,
35 const char *xml_path,
36 struct gen_batch_decode_bo (*get_bo)(void *,
37 bool,
38 uint64_t),
39 unsigned (*get_state_size)(void *, uint64_t,
40 uint64_t),
41 void *user_data)
42 {
43 memset(ctx, 0, sizeof(*ctx));
44
45 ctx->get_bo = get_bo;
46 ctx->get_state_size = get_state_size;
47 ctx->user_data = user_data;
48 ctx->fp = fp;
49 ctx->flags = flags;
50 ctx->max_vbo_decoded_lines = -1; /* No limit! */
51 ctx->engine = I915_ENGINE_CLASS_RENDER;
52
53 if (xml_path == NULL)
54 ctx->spec = gen_spec_load(devinfo);
55 else
56 ctx->spec = gen_spec_load_from_path(devinfo, xml_path);
57 ctx->disasm = gen_disasm_create(devinfo);
58 }
59
60 void
61 gen_batch_decode_ctx_finish(struct gen_batch_decode_ctx *ctx)
62 {
63 gen_spec_destroy(ctx->spec);
64 gen_disasm_destroy(ctx->disasm);
65 }
66
67 #define CSI "\e["
68 #define RED_COLOR CSI "31m"
69 #define BLUE_HEADER CSI "0;44m"
70 #define GREEN_HEADER CSI "1;42m"
71 #define NORMAL CSI "0m"
72
73 static void
74 ctx_print_group(struct gen_batch_decode_ctx *ctx,
75 struct gen_group *group,
76 uint64_t address, const void *map)
77 {
78 gen_print_group(ctx->fp, group, address, map, 0,
79 (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) != 0);
80 }
81
82 static struct gen_batch_decode_bo
83 ctx_get_bo(struct gen_batch_decode_ctx *ctx, bool ppgtt, uint64_t addr)
84 {
85 if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0)) {
86 /* On Broadwell and above, we have 48-bit addresses which consume two
87 * dwords. Some packets require that these get stored in a "canonical
88 * form" which means that bit 47 is sign-extended through the upper
89 * bits. In order to correctly handle those aub dumps, we need to mask
90 * off the top 16 bits.
91 */
92 addr &= (~0ull >> 16);
93 }
94
95 struct gen_batch_decode_bo bo = ctx->get_bo(ctx->user_data, ppgtt, addr);
96
97 if (gen_spec_get_gen(ctx->spec) >= gen_make_gen(8,0))
98 bo.addr &= (~0ull >> 16);
99
100 /* We may actually have an offset into the bo */
101 if (bo.map != NULL) {
102 assert(bo.addr <= addr);
103 uint64_t offset = addr - bo.addr;
104 bo.map += offset;
105 bo.addr += offset;
106 bo.size -= offset;
107 }
108
109 return bo;
110 }
111
112 static int
113 update_count(struct gen_batch_decode_ctx *ctx,
114 uint64_t address,
115 uint64_t base_address,
116 unsigned element_dwords,
117 unsigned guess)
118 {
119 unsigned size = 0;
120
121 if (ctx->get_state_size)
122 size = ctx->get_state_size(ctx->user_data, address, base_address);
123
124 if (size > 0)
125 return size / (sizeof(uint32_t) * element_dwords);
126
127 /* In the absence of any information, just guess arbitrarily. */
128 return guess;
129 }
130
131 static void
132 ctx_disassemble_program(struct gen_batch_decode_ctx *ctx,
133 uint32_t ksp, const char *type)
134 {
135 uint64_t addr = ctx->instruction_base + ksp;
136 struct gen_batch_decode_bo bo = ctx_get_bo(ctx, true, addr);
137 if (!bo.map)
138 return;
139
140 fprintf(ctx->fp, "\nReferenced %s:\n", type);
141 gen_disasm_disassemble(ctx->disasm, bo.map, 0, ctx->fp);
142 }
143
144 /* Heuristic to determine whether a uint32_t is probably actually a float
145 * (http://stackoverflow.com/a/2953466)
146 */
147
148 static bool
149 probably_float(uint32_t bits)
150 {
151 int exp = ((bits & 0x7f800000U) >> 23) - 127;
152 uint32_t mant = bits & 0x007fffff;
153
154 /* +- 0.0 */
155 if (exp == -127 && mant == 0)
156 return true;
157
158 /* +- 1 billionth to 1 billion */
159 if (-30 <= exp && exp <= 30)
160 return true;
161
162 /* some value with only a few binary digits */
163 if ((mant & 0x0000ffff) == 0)
164 return true;
165
166 return false;
167 }
168
169 static void
170 ctx_print_buffer(struct gen_batch_decode_ctx *ctx,
171 struct gen_batch_decode_bo bo,
172 uint32_t read_length,
173 uint32_t pitch,
174 int max_lines)
175 {
176 const uint32_t *dw_end =
177 bo.map + ROUND_DOWN_TO(MIN2(bo.size, read_length), 4);
178
179 int column_count = 0, pitch_col_count = 0, line_count = -1;
180 for (const uint32_t *dw = bo.map; dw < dw_end; dw++) {
181 if (pitch_col_count * 4 == pitch || column_count == 8) {
182 fprintf(ctx->fp, "\n");
183 column_count = 0;
184 if (pitch_col_count * 4 == pitch)
185 pitch_col_count = 0;
186 line_count++;
187
188 if (max_lines >= 0 && line_count >= max_lines)
189 break;
190 }
191 fprintf(ctx->fp, column_count == 0 ? " " : " ");
192
193 if ((ctx->flags & GEN_BATCH_DECODE_FLOATS) && probably_float(*dw))
194 fprintf(ctx->fp, " %8.2f", *(float *) dw);
195 else
196 fprintf(ctx->fp, " 0x%08x", *dw);
197
198 column_count++;
199 pitch_col_count++;
200 }
201 fprintf(ctx->fp, "\n");
202 }
203
204 static struct gen_group *
205 gen_ctx_find_instruction(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
206 {
207 return gen_spec_find_instruction(ctx->spec, ctx->engine, p);
208 }
209
210 static void
211 handle_state_base_address(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
212 {
213 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
214
215 struct gen_field_iterator iter;
216 gen_field_iterator_init(&iter, inst, p, 0, false);
217
218 uint64_t surface_base = 0, dynamic_base = 0, instruction_base = 0;
219 bool surface_modify = 0, dynamic_modify = 0, instruction_modify = 0;
220
221 while (gen_field_iterator_next(&iter)) {
222 if (strcmp(iter.name, "Surface State Base Address") == 0) {
223 surface_base = iter.raw_value;
224 } else if (strcmp(iter.name, "Dynamic State Base Address") == 0) {
225 dynamic_base = iter.raw_value;
226 } else if (strcmp(iter.name, "Instruction Base Address") == 0) {
227 instruction_base = iter.raw_value;
228 } else if (strcmp(iter.name, "Surface State Base Address Modify Enable") == 0) {
229 surface_modify = iter.raw_value;
230 } else if (strcmp(iter.name, "Dynamic State Base Address Modify Enable") == 0) {
231 dynamic_modify = iter.raw_value;
232 } else if (strcmp(iter.name, "Instruction Base Address Modify Enable") == 0) {
233 instruction_modify = iter.raw_value;
234 }
235 }
236
237 if (dynamic_modify)
238 ctx->dynamic_base = dynamic_base;
239
240 if (surface_modify)
241 ctx->surface_base = surface_base;
242
243 if (instruction_modify)
244 ctx->instruction_base = instruction_base;
245 }
246
247 static void
248 dump_binding_table(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
249 {
250 struct gen_group *strct =
251 gen_spec_find_struct(ctx->spec, "RENDER_SURFACE_STATE");
252 if (strct == NULL) {
253 fprintf(ctx->fp, "did not find RENDER_SURFACE_STATE info\n");
254 return;
255 }
256
257 if (count < 0) {
258 count = update_count(ctx, ctx->surface_base + offset,
259 ctx->surface_base, 1, 8);
260 }
261
262 if (offset % 32 != 0 || offset >= UINT16_MAX) {
263 fprintf(ctx->fp, " invalid binding table pointer\n");
264 return;
265 }
266
267 struct gen_batch_decode_bo bind_bo =
268 ctx_get_bo(ctx, true, ctx->surface_base + offset);
269
270 if (bind_bo.map == NULL) {
271 fprintf(ctx->fp, " binding table unavailable\n");
272 return;
273 }
274
275 const uint32_t *pointers = bind_bo.map;
276 for (int i = 0; i < count; i++) {
277 if (pointers[i] == 0)
278 continue;
279
280 uint64_t addr = ctx->surface_base + pointers[i];
281 struct gen_batch_decode_bo bo = ctx_get_bo(ctx, true, addr);
282 uint32_t size = strct->dw_length * 4;
283
284 if (pointers[i] % 32 != 0 ||
285 addr < bo.addr || addr + size >= bo.addr + bo.size) {
286 fprintf(ctx->fp, "pointer %u: 0x%08x <not valid>\n", i, pointers[i]);
287 continue;
288 }
289
290 fprintf(ctx->fp, "pointer %u: 0x%08x\n", i, pointers[i]);
291 ctx_print_group(ctx, strct, addr, bo.map + (addr - bo.addr));
292 }
293 }
294
295 static void
296 dump_samplers(struct gen_batch_decode_ctx *ctx, uint32_t offset, int count)
297 {
298 struct gen_group *strct = gen_spec_find_struct(ctx->spec, "SAMPLER_STATE");
299 uint64_t state_addr = ctx->dynamic_base + offset;
300
301 if (count < 0) {
302 count = update_count(ctx, state_addr, ctx->dynamic_base,
303 strct->dw_length, 4);
304 }
305
306 struct gen_batch_decode_bo bo = ctx_get_bo(ctx, true, state_addr);
307 const void *state_map = bo.map;
308
309 if (state_map == NULL) {
310 fprintf(ctx->fp, " samplers unavailable\n");
311 return;
312 }
313
314 if (offset % 32 != 0 || state_addr - bo.addr >= bo.size) {
315 fprintf(ctx->fp, " invalid sampler state pointer\n");
316 return;
317 }
318
319 for (int i = 0; i < count; i++) {
320 fprintf(ctx->fp, "sampler state %d\n", i);
321 ctx_print_group(ctx, strct, state_addr, state_map);
322 state_addr += 16;
323 state_map += 16;
324 }
325 }
326
327 static void
328 handle_media_interface_descriptor_load(struct gen_batch_decode_ctx *ctx,
329 const uint32_t *p)
330 {
331 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
332 struct gen_group *desc =
333 gen_spec_find_struct(ctx->spec, "INTERFACE_DESCRIPTOR_DATA");
334
335 struct gen_field_iterator iter;
336 gen_field_iterator_init(&iter, inst, p, 0, false);
337 uint32_t descriptor_offset = 0;
338 int descriptor_count = 0;
339 while (gen_field_iterator_next(&iter)) {
340 if (strcmp(iter.name, "Interface Descriptor Data Start Address") == 0) {
341 descriptor_offset = strtol(iter.value, NULL, 16);
342 } else if (strcmp(iter.name, "Interface Descriptor Total Length") == 0) {
343 descriptor_count =
344 strtol(iter.value, NULL, 16) / (desc->dw_length * 4);
345 }
346 }
347
348 uint64_t desc_addr = ctx->dynamic_base + descriptor_offset;
349 struct gen_batch_decode_bo bo = ctx_get_bo(ctx, true, desc_addr);
350 const void *desc_map = bo.map;
351
352 if (desc_map == NULL) {
353 fprintf(ctx->fp, " interface descriptors unavailable\n");
354 return;
355 }
356
357 for (int i = 0; i < descriptor_count; i++) {
358 fprintf(ctx->fp, "descriptor %d: %08x\n", i, descriptor_offset);
359
360 ctx_print_group(ctx, desc, desc_addr, desc_map);
361
362 gen_field_iterator_init(&iter, desc, desc_map, 0, false);
363 uint64_t ksp = 0;
364 uint32_t sampler_offset = 0, sampler_count = 0;
365 uint32_t binding_table_offset = 0, binding_entry_count = 0;
366 while (gen_field_iterator_next(&iter)) {
367 if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
368 ksp = strtoll(iter.value, NULL, 16);
369 } else if (strcmp(iter.name, "Sampler State Pointer") == 0) {
370 sampler_offset = strtol(iter.value, NULL, 16);
371 } else if (strcmp(iter.name, "Sampler Count") == 0) {
372 sampler_count = strtol(iter.value, NULL, 10);
373 } else if (strcmp(iter.name, "Binding Table Pointer") == 0) {
374 binding_table_offset = strtol(iter.value, NULL, 16);
375 } else if (strcmp(iter.name, "Binding Table Entry Count") == 0) {
376 binding_entry_count = strtol(iter.value, NULL, 10);
377 }
378 }
379
380 ctx_disassemble_program(ctx, ksp, "compute shader");
381 fprintf(ctx->fp, "\n");
382
383 dump_samplers(ctx, sampler_offset, sampler_count);
384 dump_binding_table(ctx, binding_table_offset, binding_entry_count);
385
386 desc_map += desc->dw_length;
387 desc_addr += desc->dw_length * 4;
388 }
389 }
390
391 static void
392 handle_3dstate_vertex_buffers(struct gen_batch_decode_ctx *ctx,
393 const uint32_t *p)
394 {
395 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
396 struct gen_group *vbs = gen_spec_find_struct(ctx->spec, "VERTEX_BUFFER_STATE");
397
398 struct gen_batch_decode_bo vb = {};
399 uint32_t vb_size = 0;
400 int index = -1;
401 int pitch = -1;
402 bool ready = false;
403
404 struct gen_field_iterator iter;
405 gen_field_iterator_init(&iter, inst, p, 0, false);
406 while (gen_field_iterator_next(&iter)) {
407 if (iter.struct_desc != vbs)
408 continue;
409
410 struct gen_field_iterator vbs_iter;
411 gen_field_iterator_init(&vbs_iter, vbs, &iter.p[iter.start_bit / 32], 0, false);
412 while (gen_field_iterator_next(&vbs_iter)) {
413 if (strcmp(vbs_iter.name, "Vertex Buffer Index") == 0) {
414 index = vbs_iter.raw_value;
415 } else if (strcmp(vbs_iter.name, "Buffer Pitch") == 0) {
416 pitch = vbs_iter.raw_value;
417 } else if (strcmp(vbs_iter.name, "Buffer Starting Address") == 0) {
418 vb = ctx_get_bo(ctx, true, vbs_iter.raw_value);
419 } else if (strcmp(vbs_iter.name, "Buffer Size") == 0) {
420 vb_size = vbs_iter.raw_value;
421 ready = true;
422 } else if (strcmp(vbs_iter.name, "End Address") == 0) {
423 if (vb.map && vbs_iter.raw_value >= vb.addr)
424 vb_size = (vbs_iter.raw_value + 1) - vb.addr;
425 else
426 vb_size = 0;
427 ready = true;
428 }
429
430 if (!ready)
431 continue;
432
433 fprintf(ctx->fp, "vertex buffer %d, size %d\n", index, vb_size);
434
435 if (vb.map == NULL) {
436 fprintf(ctx->fp, " buffer contents unavailable\n");
437 continue;
438 }
439
440 if (vb.map == 0 || vb_size == 0)
441 continue;
442
443 ctx_print_buffer(ctx, vb, vb_size, pitch, ctx->max_vbo_decoded_lines);
444
445 vb.map = NULL;
446 vb_size = 0;
447 index = -1;
448 pitch = -1;
449 ready = false;
450 }
451 }
452 }
453
454 static void
455 handle_3dstate_index_buffer(struct gen_batch_decode_ctx *ctx,
456 const uint32_t *p)
457 {
458 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
459
460 struct gen_batch_decode_bo ib = {};
461 uint32_t ib_size = 0;
462 uint32_t format = 0;
463
464 struct gen_field_iterator iter;
465 gen_field_iterator_init(&iter, inst, p, 0, false);
466 while (gen_field_iterator_next(&iter)) {
467 if (strcmp(iter.name, "Index Format") == 0) {
468 format = iter.raw_value;
469 } else if (strcmp(iter.name, "Buffer Starting Address") == 0) {
470 ib = ctx_get_bo(ctx, true, iter.raw_value);
471 } else if (strcmp(iter.name, "Buffer Size") == 0) {
472 ib_size = iter.raw_value;
473 }
474 }
475
476 if (ib.map == NULL) {
477 fprintf(ctx->fp, " buffer contents unavailable\n");
478 return;
479 }
480
481 const void *m = ib.map;
482 const void *ib_end = ib.map + MIN2(ib.size, ib_size);
483 for (int i = 0; m < ib_end && i < 10; i++) {
484 switch (format) {
485 case 0:
486 fprintf(ctx->fp, "%3d ", *(uint8_t *)m);
487 m += 1;
488 break;
489 case 1:
490 fprintf(ctx->fp, "%3d ", *(uint16_t *)m);
491 m += 2;
492 break;
493 case 2:
494 fprintf(ctx->fp, "%3d ", *(uint32_t *)m);
495 m += 4;
496 break;
497 }
498 }
499
500 if (m < ib_end)
501 fprintf(ctx->fp, "...");
502 fprintf(ctx->fp, "\n");
503 }
504
505 static void
506 decode_single_ksp(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
507 {
508 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
509
510 uint64_t ksp = 0;
511 bool is_simd8 = false; /* vertex shaders on Gen8+ only */
512 bool is_enabled = true;
513
514 struct gen_field_iterator iter;
515 gen_field_iterator_init(&iter, inst, p, 0, false);
516 while (gen_field_iterator_next(&iter)) {
517 if (strcmp(iter.name, "Kernel Start Pointer") == 0) {
518 ksp = iter.raw_value;
519 } else if (strcmp(iter.name, "SIMD8 Dispatch Enable") == 0) {
520 is_simd8 = iter.raw_value;
521 } else if (strcmp(iter.name, "Dispatch Mode") == 0) {
522 is_simd8 = strcmp(iter.value, "SIMD8") == 0;
523 } else if (strcmp(iter.name, "Dispatch Enable") == 0) {
524 is_simd8 = strcmp(iter.value, "SIMD8") == 0;
525 } else if (strcmp(iter.name, "Enable") == 0) {
526 is_enabled = iter.raw_value;
527 }
528 }
529
530 const char *type =
531 strcmp(inst->name, "VS_STATE") == 0 ? "vertex shader" :
532 strcmp(inst->name, "GS_STATE") == 0 ? "geometry shader" :
533 strcmp(inst->name, "SF_STATE") == 0 ? "strips and fans shader" :
534 strcmp(inst->name, "CLIP_STATE") == 0 ? "clip shader" :
535 strcmp(inst->name, "3DSTATE_DS") == 0 ? "tessellation evaluation shader" :
536 strcmp(inst->name, "3DSTATE_HS") == 0 ? "tessellation control shader" :
537 strcmp(inst->name, "3DSTATE_VS") == 0 ? (is_simd8 ? "SIMD8 vertex shader" : "vec4 vertex shader") :
538 strcmp(inst->name, "3DSTATE_GS") == 0 ? (is_simd8 ? "SIMD8 geometry shader" : "vec4 geometry shader") :
539 NULL;
540
541 if (is_enabled) {
542 ctx_disassemble_program(ctx, ksp, type);
543 fprintf(ctx->fp, "\n");
544 }
545 }
546
547 static void
548 decode_ps_kernels(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
549 {
550 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
551
552 uint64_t ksp[3] = {0, 0, 0};
553 bool enabled[3] = {false, false, false};
554
555 struct gen_field_iterator iter;
556 gen_field_iterator_init(&iter, inst, p, 0, false);
557 while (gen_field_iterator_next(&iter)) {
558 if (strncmp(iter.name, "Kernel Start Pointer ",
559 strlen("Kernel Start Pointer ")) == 0) {
560 int idx = iter.name[strlen("Kernel Start Pointer ")] - '0';
561 ksp[idx] = strtol(iter.value, NULL, 16);
562 } else if (strcmp(iter.name, "8 Pixel Dispatch Enable") == 0) {
563 enabled[0] = strcmp(iter.value, "true") == 0;
564 } else if (strcmp(iter.name, "16 Pixel Dispatch Enable") == 0) {
565 enabled[1] = strcmp(iter.value, "true") == 0;
566 } else if (strcmp(iter.name, "32 Pixel Dispatch Enable") == 0) {
567 enabled[2] = strcmp(iter.value, "true") == 0;
568 }
569 }
570
571 /* Reorder KSPs to be [8, 16, 32] instead of the hardware order. */
572 if (enabled[0] + enabled[1] + enabled[2] == 1) {
573 if (enabled[1]) {
574 ksp[1] = ksp[0];
575 ksp[0] = 0;
576 } else if (enabled[2]) {
577 ksp[2] = ksp[0];
578 ksp[0] = 0;
579 }
580 } else {
581 uint64_t tmp = ksp[1];
582 ksp[1] = ksp[2];
583 ksp[2] = tmp;
584 }
585
586 if (enabled[0])
587 ctx_disassemble_program(ctx, ksp[0], "SIMD8 fragment shader");
588 if (enabled[1])
589 ctx_disassemble_program(ctx, ksp[1], "SIMD16 fragment shader");
590 if (enabled[2])
591 ctx_disassemble_program(ctx, ksp[2], "SIMD32 fragment shader");
592
593 if (enabled[0] || enabled[1] || enabled[2])
594 fprintf(ctx->fp, "\n");
595 }
596
597 static void
598 decode_3dstate_constant_all(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
599 {
600 struct gen_group *inst =
601 gen_spec_find_instruction(ctx->spec, ctx->engine, p);
602 struct gen_group *body =
603 gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_ALL_DATA");
604
605 uint32_t read_length[4];
606 struct gen_batch_decode_bo buffer[4];
607 memset(buffer, 0, sizeof(buffer));
608
609 struct gen_field_iterator outer;
610 gen_field_iterator_init(&outer, inst, p, 0, false);
611 int idx = 0;
612 while (gen_field_iterator_next(&outer)) {
613 if (outer.struct_desc != body)
614 continue;
615
616 struct gen_field_iterator iter;
617 gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
618 0, false);
619 while (gen_field_iterator_next(&iter)) {
620 if (!strcmp(iter.name, "Pointer To Constant Buffer")) {
621 buffer[idx] = ctx_get_bo(ctx, true, iter.raw_value);
622 } else if (!strcmp(iter.name, "Constant Buffer Read Length")) {
623 read_length[idx] = iter.raw_value;
624 }
625 }
626 idx++;
627 }
628
629 for (int i = 0; i < 4; i++) {
630 if (read_length[i] == 0 || buffer[i].map == NULL)
631 continue;
632
633 unsigned size = read_length[i] * 32;
634 fprintf(ctx->fp, "constant buffer %d, size %u\n", i, size);
635
636 ctx_print_buffer(ctx, buffer[i], size, 0, -1);
637 }
638 }
639
640 static void
641 decode_3dstate_constant(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
642 {
643 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
644 struct gen_group *body =
645 gen_spec_find_struct(ctx->spec, "3DSTATE_CONSTANT_BODY");
646
647 uint32_t read_length[4] = {0};
648 uint64_t read_addr[4];
649
650 struct gen_field_iterator outer;
651 gen_field_iterator_init(&outer, inst, p, 0, false);
652 while (gen_field_iterator_next(&outer)) {
653 if (outer.struct_desc != body)
654 continue;
655
656 struct gen_field_iterator iter;
657 gen_field_iterator_init(&iter, body, &outer.p[outer.start_bit / 32],
658 0, false);
659
660 while (gen_field_iterator_next(&iter)) {
661 int idx;
662 if (sscanf(iter.name, "Read Length[%d]", &idx) == 1) {
663 read_length[idx] = iter.raw_value;
664 } else if (sscanf(iter.name, "Buffer[%d]", &idx) == 1) {
665 read_addr[idx] = iter.raw_value;
666 }
667 }
668
669 for (int i = 0; i < 4; i++) {
670 if (read_length[i] == 0)
671 continue;
672
673 struct gen_batch_decode_bo buffer = ctx_get_bo(ctx, true, read_addr[i]);
674 if (!buffer.map) {
675 fprintf(ctx->fp, "constant buffer %d unavailable\n", i);
676 continue;
677 }
678
679 unsigned size = read_length[i] * 32;
680 fprintf(ctx->fp, "constant buffer %d, size %u\n", i, size);
681
682 ctx_print_buffer(ctx, buffer, size, 0, -1);
683 }
684 }
685 }
686
687 static void
688 decode_gen6_3dstate_binding_table_pointers(struct gen_batch_decode_ctx *ctx,
689 const uint32_t *p)
690 {
691 fprintf(ctx->fp, "VS Binding Table:\n");
692 dump_binding_table(ctx, p[1], -1);
693
694 fprintf(ctx->fp, "GS Binding Table:\n");
695 dump_binding_table(ctx, p[2], -1);
696
697 fprintf(ctx->fp, "PS Binding Table:\n");
698 dump_binding_table(ctx, p[3], -1);
699 }
700
701 static void
702 decode_3dstate_binding_table_pointers(struct gen_batch_decode_ctx *ctx,
703 const uint32_t *p)
704 {
705 dump_binding_table(ctx, p[1], -1);
706 }
707
708 static void
709 decode_3dstate_sampler_state_pointers(struct gen_batch_decode_ctx *ctx,
710 const uint32_t *p)
711 {
712 dump_samplers(ctx, p[1], -1);
713 }
714
715 static void
716 decode_3dstate_sampler_state_pointers_gen6(struct gen_batch_decode_ctx *ctx,
717 const uint32_t *p)
718 {
719 dump_samplers(ctx, p[1], -1);
720 dump_samplers(ctx, p[2], -1);
721 dump_samplers(ctx, p[3], -1);
722 }
723
724 static bool
725 str_ends_with(const char *str, const char *end)
726 {
727 int offset = strlen(str) - strlen(end);
728 if (offset < 0)
729 return false;
730
731 return strcmp(str + offset, end) == 0;
732 }
733
734 static void
735 decode_dynamic_state_pointers(struct gen_batch_decode_ctx *ctx,
736 const char *struct_type, const uint32_t *p,
737 int count)
738 {
739 struct gen_group *inst = gen_ctx_find_instruction(ctx, p);
740
741 uint32_t state_offset = 0;
742
743 struct gen_field_iterator iter;
744 gen_field_iterator_init(&iter, inst, p, 0, false);
745 while (gen_field_iterator_next(&iter)) {
746 if (str_ends_with(iter.name, "Pointer")) {
747 state_offset = iter.raw_value;
748 break;
749 }
750 }
751
752 uint64_t state_addr = ctx->dynamic_base + state_offset;
753 struct gen_batch_decode_bo bo = ctx_get_bo(ctx, true, state_addr);
754 const void *state_map = bo.map;
755
756 if (state_map == NULL) {
757 fprintf(ctx->fp, " dynamic %s state unavailable\n", struct_type);
758 return;
759 }
760
761 struct gen_group *state = gen_spec_find_struct(ctx->spec, struct_type);
762 if (strcmp(struct_type, "BLEND_STATE") == 0) {
763 /* Blend states are different from the others because they have a header
764 * struct called BLEND_STATE which is followed by a variable number of
765 * BLEND_STATE_ENTRY structs.
766 */
767 fprintf(ctx->fp, "%s\n", struct_type);
768 ctx_print_group(ctx, state, state_addr, state_map);
769
770 state_addr += state->dw_length * 4;
771 state_map += state->dw_length * 4;
772
773 struct_type = "BLEND_STATE_ENTRY";
774 state = gen_spec_find_struct(ctx->spec, struct_type);
775 }
776
777 count = update_count(ctx, ctx->dynamic_base + state_offset,
778 ctx->dynamic_base, state->dw_length, count);
779
780 for (int i = 0; i < count; i++) {
781 fprintf(ctx->fp, "%s %d\n", struct_type, i);
782 ctx_print_group(ctx, state, state_addr, state_map);
783
784 state_addr += state->dw_length * 4;
785 state_map += state->dw_length * 4;
786 }
787 }
788
789 static void
790 decode_3dstate_viewport_state_pointers_cc(struct gen_batch_decode_ctx *ctx,
791 const uint32_t *p)
792 {
793 decode_dynamic_state_pointers(ctx, "CC_VIEWPORT", p, 4);
794 }
795
796 static void
797 decode_3dstate_viewport_state_pointers_sf_clip(struct gen_batch_decode_ctx *ctx,
798 const uint32_t *p)
799 {
800 decode_dynamic_state_pointers(ctx, "SF_CLIP_VIEWPORT", p, 4);
801 }
802
803 static void
804 decode_3dstate_blend_state_pointers(struct gen_batch_decode_ctx *ctx,
805 const uint32_t *p)
806 {
807 decode_dynamic_state_pointers(ctx, "BLEND_STATE", p, 1);
808 }
809
810 static void
811 decode_3dstate_cc_state_pointers(struct gen_batch_decode_ctx *ctx,
812 const uint32_t *p)
813 {
814 decode_dynamic_state_pointers(ctx, "COLOR_CALC_STATE", p, 1);
815 }
816
817 static void
818 decode_3dstate_scissor_state_pointers(struct gen_batch_decode_ctx *ctx,
819 const uint32_t *p)
820 {
821 decode_dynamic_state_pointers(ctx, "SCISSOR_RECT", p, 1);
822 }
823
824 static void
825 decode_3dstate_slice_table_state_pointers(struct gen_batch_decode_ctx *ctx,
826 const uint32_t *p)
827 {
828 decode_dynamic_state_pointers(ctx, "SLICE_HASH_TABLE", p, 1);
829 }
830
831 static void
832 decode_load_register_imm(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
833 {
834 struct gen_group *reg = gen_spec_find_register(ctx->spec, p[1]);
835
836 if (reg != NULL) {
837 fprintf(ctx->fp, "register %s (0x%x): 0x%x\n",
838 reg->name, reg->register_offset, p[2]);
839 ctx_print_group(ctx, reg, reg->register_offset, &p[2]);
840 }
841 }
842
843 static void
844 decode_vs_state(struct gen_batch_decode_ctx *ctx, uint32_t offset)
845 {
846 struct gen_group *strct =
847 gen_spec_find_struct(ctx->spec, "VS_STATE");
848 if (strct == NULL) {
849 fprintf(ctx->fp, "did not find VS_STATE info\n");
850 return;
851 }
852
853 struct gen_batch_decode_bo bind_bo =
854 ctx_get_bo(ctx, true, offset);
855
856 if (bind_bo.map == NULL) {
857 fprintf(ctx->fp, " vs state unavailable\n");
858 return;
859 }
860
861 ctx_print_group(ctx, strct, offset, bind_bo.map);
862 }
863
864
865 static void
866 decode_clip_state(struct gen_batch_decode_ctx *ctx, uint32_t offset)
867 {
868 struct gen_group *strct =
869 gen_spec_find_struct(ctx->spec, "CLIP_STATE");
870 if (strct == NULL) {
871 fprintf(ctx->fp, "did not find CLIP_STATE info\n");
872 return;
873 }
874
875 struct gen_batch_decode_bo bind_bo =
876 ctx_get_bo(ctx, true, offset);
877
878 if (bind_bo.map == NULL) {
879 fprintf(ctx->fp, " clip state unavailable\n");
880 return;
881 }
882
883 ctx_print_group(ctx, strct, offset, bind_bo.map);
884
885 struct gen_group *vp_strct =
886 gen_spec_find_struct(ctx->spec, "CLIP_VIEWPORT");
887 if (vp_strct == NULL) {
888 fprintf(ctx->fp, "did not find CLIP_VIEWPORT info\n");
889 return;
890 }
891 uint32_t clip_vp_offset = ((uint32_t *)bind_bo.map)[6] & ~0x3;
892 struct gen_batch_decode_bo vp_bo =
893 ctx_get_bo(ctx, true, clip_vp_offset);
894 if (vp_bo.map == NULL) {
895 fprintf(ctx->fp, " clip vp state unavailable\n");
896 return;
897 }
898 ctx_print_group(ctx, vp_strct, clip_vp_offset, vp_bo.map);
899 }
900
901 static void
902 decode_sf_state(struct gen_batch_decode_ctx *ctx, uint32_t offset)
903 {
904 struct gen_group *strct =
905 gen_spec_find_struct(ctx->spec, "SF_STATE");
906 if (strct == NULL) {
907 fprintf(ctx->fp, "did not find SF_STATE info\n");
908 return;
909 }
910
911 struct gen_batch_decode_bo bind_bo =
912 ctx_get_bo(ctx, true, offset);
913
914 if (bind_bo.map == NULL) {
915 fprintf(ctx->fp, " sf state unavailable\n");
916 return;
917 }
918
919 ctx_print_group(ctx, strct, offset, bind_bo.map);
920
921 struct gen_group *vp_strct =
922 gen_spec_find_struct(ctx->spec, "SF_VIEWPORT");
923 if (vp_strct == NULL) {
924 fprintf(ctx->fp, "did not find SF_VIEWPORT info\n");
925 return;
926 }
927
928 uint32_t sf_vp_offset = ((uint32_t *)bind_bo.map)[5] & ~0x3;
929 struct gen_batch_decode_bo vp_bo =
930 ctx_get_bo(ctx, true, sf_vp_offset);
931 if (vp_bo.map == NULL) {
932 fprintf(ctx->fp, " sf vp state unavailable\n");
933 return;
934 }
935 ctx_print_group(ctx, vp_strct, sf_vp_offset, vp_bo.map);
936 }
937
938 static void
939 decode_wm_state(struct gen_batch_decode_ctx *ctx, uint32_t offset)
940 {
941 struct gen_group *strct =
942 gen_spec_find_struct(ctx->spec, "WM_STATE");
943 if (strct == NULL) {
944 fprintf(ctx->fp, "did not find WM_STATE info\n");
945 return;
946 }
947
948 struct gen_batch_decode_bo bind_bo =
949 ctx_get_bo(ctx, true, offset);
950
951 if (bind_bo.map == NULL) {
952 fprintf(ctx->fp, " wm state unavailable\n");
953 return;
954 }
955
956 ctx_print_group(ctx, strct, offset, bind_bo.map);
957 }
958
959 static void
960 decode_cc_state(struct gen_batch_decode_ctx *ctx, uint32_t offset)
961 {
962 struct gen_group *strct =
963 gen_spec_find_struct(ctx->spec, "COLOR_CALC_STATE");
964 if (strct == NULL) {
965 fprintf(ctx->fp, "did not find COLOR_CALC_STATE info\n");
966 return;
967 }
968
969 struct gen_batch_decode_bo bind_bo =
970 ctx_get_bo(ctx, true, offset);
971
972 if (bind_bo.map == NULL) {
973 fprintf(ctx->fp, " cc state unavailable\n");
974 return;
975 }
976
977 ctx_print_group(ctx, strct, offset, bind_bo.map);
978
979 struct gen_group *vp_strct =
980 gen_spec_find_struct(ctx->spec, "CC_VIEWPORT");
981 if (vp_strct == NULL) {
982 fprintf(ctx->fp, "did not find CC_VIEWPORT info\n");
983 return;
984 }
985 uint32_t cc_vp_offset = ((uint32_t *)bind_bo.map)[4] & ~0x3;
986 struct gen_batch_decode_bo vp_bo =
987 ctx_get_bo(ctx, true, cc_vp_offset);
988 if (vp_bo.map == NULL) {
989 fprintf(ctx->fp, " cc vp state unavailable\n");
990 return;
991 }
992 ctx_print_group(ctx, vp_strct, cc_vp_offset, vp_bo.map);
993 }
994 static void
995 decode_pipelined_pointers(struct gen_batch_decode_ctx *ctx, const uint32_t *p)
996 {
997 fprintf(ctx->fp, "VS State Table:\n");
998 decode_vs_state(ctx, p[1]);
999 fprintf(ctx->fp, "Clip State Table:\n");
1000 decode_clip_state(ctx, p[3] & ~1);
1001 fprintf(ctx->fp, "SF State Table:\n");
1002 decode_sf_state(ctx, p[4]);
1003 fprintf(ctx->fp, "WM State Table:\n");
1004 decode_wm_state(ctx, p[5]);
1005 fprintf(ctx->fp, "CC State Table:\n");
1006 decode_cc_state(ctx, p[6]);
1007 }
1008
1009 struct custom_decoder {
1010 const char *cmd_name;
1011 void (*decode)(struct gen_batch_decode_ctx *ctx, const uint32_t *p);
1012 } custom_decoders[] = {
1013 { "STATE_BASE_ADDRESS", handle_state_base_address },
1014 { "MEDIA_INTERFACE_DESCRIPTOR_LOAD", handle_media_interface_descriptor_load },
1015 { "3DSTATE_VERTEX_BUFFERS", handle_3dstate_vertex_buffers },
1016 { "3DSTATE_INDEX_BUFFER", handle_3dstate_index_buffer },
1017 { "3DSTATE_VS", decode_single_ksp },
1018 { "3DSTATE_GS", decode_single_ksp },
1019 { "3DSTATE_DS", decode_single_ksp },
1020 { "3DSTATE_HS", decode_single_ksp },
1021 { "3DSTATE_PS", decode_ps_kernels },
1022 { "3DSTATE_WM", decode_ps_kernels },
1023 { "3DSTATE_CONSTANT_VS", decode_3dstate_constant },
1024 { "3DSTATE_CONSTANT_GS", decode_3dstate_constant },
1025 { "3DSTATE_CONSTANT_PS", decode_3dstate_constant },
1026 { "3DSTATE_CONSTANT_HS", decode_3dstate_constant },
1027 { "3DSTATE_CONSTANT_DS", decode_3dstate_constant },
1028 { "3DSTATE_CONSTANT_ALL", decode_3dstate_constant_all },
1029
1030 { "3DSTATE_BINDING_TABLE_POINTERS", decode_gen6_3dstate_binding_table_pointers },
1031 { "3DSTATE_BINDING_TABLE_POINTERS_VS", decode_3dstate_binding_table_pointers },
1032 { "3DSTATE_BINDING_TABLE_POINTERS_HS", decode_3dstate_binding_table_pointers },
1033 { "3DSTATE_BINDING_TABLE_POINTERS_DS", decode_3dstate_binding_table_pointers },
1034 { "3DSTATE_BINDING_TABLE_POINTERS_GS", decode_3dstate_binding_table_pointers },
1035 { "3DSTATE_BINDING_TABLE_POINTERS_PS", decode_3dstate_binding_table_pointers },
1036
1037 { "3DSTATE_SAMPLER_STATE_POINTERS_VS", decode_3dstate_sampler_state_pointers },
1038 { "3DSTATE_SAMPLER_STATE_POINTERS_HS", decode_3dstate_sampler_state_pointers },
1039 { "3DSTATE_SAMPLER_STATE_POINTERS_DS", decode_3dstate_sampler_state_pointers },
1040 { "3DSTATE_SAMPLER_STATE_POINTERS_GS", decode_3dstate_sampler_state_pointers },
1041 { "3DSTATE_SAMPLER_STATE_POINTERS_PS", decode_3dstate_sampler_state_pointers },
1042 { "3DSTATE_SAMPLER_STATE_POINTERS", decode_3dstate_sampler_state_pointers_gen6 },
1043
1044 { "3DSTATE_VIEWPORT_STATE_POINTERS_CC", decode_3dstate_viewport_state_pointers_cc },
1045 { "3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP", decode_3dstate_viewport_state_pointers_sf_clip },
1046 { "3DSTATE_BLEND_STATE_POINTERS", decode_3dstate_blend_state_pointers },
1047 { "3DSTATE_CC_STATE_POINTERS", decode_3dstate_cc_state_pointers },
1048 { "3DSTATE_SCISSOR_STATE_POINTERS", decode_3dstate_scissor_state_pointers },
1049 { "3DSTATE_SLICE_TABLE_STATE_POINTERS", decode_3dstate_slice_table_state_pointers },
1050 { "MI_LOAD_REGISTER_IMM", decode_load_register_imm },
1051 { "3DSTATE_PIPELINED_POINTERS", decode_pipelined_pointers }
1052 };
1053
1054 void
1055 gen_print_batch(struct gen_batch_decode_ctx *ctx,
1056 const uint32_t *batch, uint32_t batch_size,
1057 uint64_t batch_addr, bool from_ring)
1058 {
1059 const uint32_t *p, *end = batch + batch_size / sizeof(uint32_t);
1060 int length;
1061 struct gen_group *inst;
1062 const char *reset_color = ctx->flags & GEN_BATCH_DECODE_IN_COLOR ? NORMAL : "";
1063
1064 if (ctx->n_batch_buffer_start >= 100) {
1065 fprintf(ctx->fp, "%s0x%08"PRIx64": Max batch buffer jumps exceeded%s\n",
1066 (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) ? RED_COLOR : "",
1067 (ctx->flags & GEN_BATCH_DECODE_OFFSETS) ? batch_addr : 0,
1068 reset_color);
1069 return;
1070 }
1071
1072 ctx->n_batch_buffer_start++;
1073
1074 for (p = batch; p < end; p += length) {
1075 inst = gen_ctx_find_instruction(ctx, p);
1076 length = gen_group_get_length(inst, p);
1077 assert(inst == NULL || length > 0);
1078 length = MAX2(1, length);
1079
1080 uint64_t offset;
1081 if (ctx->flags & GEN_BATCH_DECODE_OFFSETS)
1082 offset = batch_addr + ((char *)p - (char *)batch);
1083 else
1084 offset = 0;
1085
1086 if (inst == NULL) {
1087 fprintf(ctx->fp, "%s0x%08"PRIx64": unknown instruction %08x%s\n",
1088 (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) ? RED_COLOR : "",
1089 offset, p[0], reset_color);
1090 continue;
1091 }
1092
1093 const char *color;
1094 const char *inst_name = gen_group_get_name(inst);
1095 if (ctx->flags & GEN_BATCH_DECODE_IN_COLOR) {
1096 reset_color = NORMAL;
1097 if (ctx->flags & GEN_BATCH_DECODE_FULL) {
1098 if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0 ||
1099 strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0)
1100 color = GREEN_HEADER;
1101 else
1102 color = BLUE_HEADER;
1103 } else {
1104 color = NORMAL;
1105 }
1106 } else {
1107 color = "";
1108 reset_color = "";
1109 }
1110
1111 fprintf(ctx->fp, "%s0x%08"PRIx64": 0x%08x: %-80s%s\n",
1112 color, offset, p[0], inst_name, reset_color);
1113
1114 if (ctx->flags & GEN_BATCH_DECODE_FULL) {
1115 ctx_print_group(ctx, inst, offset, p);
1116
1117 for (int i = 0; i < ARRAY_SIZE(custom_decoders); i++) {
1118 if (strcmp(inst_name, custom_decoders[i].cmd_name) == 0) {
1119 custom_decoders[i].decode(ctx, p);
1120 break;
1121 }
1122 }
1123 }
1124
1125 if (strcmp(inst_name, "MI_BATCH_BUFFER_START") == 0) {
1126 uint64_t next_batch_addr = 0;
1127 bool ppgtt = false;
1128 bool second_level = false;
1129 struct gen_field_iterator iter;
1130 gen_field_iterator_init(&iter, inst, p, 0, false);
1131 while (gen_field_iterator_next(&iter)) {
1132 if (strcmp(iter.name, "Batch Buffer Start Address") == 0) {
1133 next_batch_addr = iter.raw_value;
1134 } else if (strcmp(iter.name, "Second Level Batch Buffer") == 0) {
1135 second_level = iter.raw_value;
1136 } else if (strcmp(iter.name, "Address Space Indicator") == 0) {
1137 ppgtt = iter.raw_value;
1138 }
1139 }
1140
1141 struct gen_batch_decode_bo next_batch = ctx_get_bo(ctx, ppgtt, next_batch_addr);
1142
1143 if (next_batch.map == NULL) {
1144 fprintf(ctx->fp, "Secondary batch at 0x%08"PRIx64" unavailable\n",
1145 next_batch_addr);
1146 } else {
1147 gen_print_batch(ctx, next_batch.map, next_batch.size,
1148 next_batch.addr, false);
1149 }
1150 if (second_level) {
1151 /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
1152 * like a subroutine call. Commands that come afterwards get
1153 * processed once the 2nd level batch buffer returns with
1154 * MI_BATCH_BUFFER_END.
1155 */
1156 continue;
1157 } else if (!from_ring) {
1158 /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
1159 * like a goto. Nothing after it will ever get processed. In
1160 * order to prevent the recursion from growing, we just reset the
1161 * loop and continue;
1162 */
1163 break;
1164 }
1165 } else if (strcmp(inst_name, "MI_BATCH_BUFFER_END") == 0) {
1166 break;
1167 }
1168 }
1169
1170 ctx->n_batch_buffer_start--;
1171 }