projects / mesa.git / blob
? search:


dac27c366848a4c01bf4e7d7d3cc18e8145aba4d
[mesa.git] / src / gallium / drivers / panfrost / pandecode / decode.c
1 /*
2 * Copyright (C) 2017-2019 Alyssa Rosenzweig
3 * Copyright (C) 2017-2019 Connor Abbott
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 */
24
25 #include <panfrost-job.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <memory.h>
29 #include <stdbool.h>
30 #include <stdarg.h>
31 #include "mmap.h"
32 #include "util/u_math.h"
33
34 #include "../pan_pretty_print.h"
35 #include "../midgard/disassemble.h"
36 #include "../bifrost/disassemble.h"
37 int pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost);
38
39 #define MEMORY_PROP(obj, p) {\
40 char *a = pointer_as_memory_reference(obj->p); \
41 pandecode_prop("%s = %s", #p, a); \
42 free(a); \
43 }
44
45 #define MEMORY_COMMENT(obj, p) {\
46 char *a = pointer_as_memory_reference(obj->p); \
47 pandecode_msg("%s = %s\n", #p, a); \
48 free(a); \
49 }
50
51 #define DYN_MEMORY_PROP(obj, no, p) { \
52 if (obj->p) \
53 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
54 }
55
56 /* Semantic logging type.
57 *
58 * Raw: for raw messages to be printed as is.
59 * Message: for helpful information to be commented out in replays.
60 * Property: for properties of a struct
61 *
62 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
63 */
64
65 enum pandecode_log_type {
66 PANDECODE_RAW,
67 PANDECODE_MESSAGE,
68 PANDECODE_PROPERTY
69 };
70
71 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
72 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
73 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
74
75 unsigned pandecode_indent = 0;
76
77 static void
78 pandecode_make_indent(void)
79 {
80 for (unsigned i = 0; i < pandecode_indent; ++i)
81 printf(" ");
82 }
83
84 static void
85 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
86 {
87 va_list ap;
88
89 pandecode_make_indent();
90
91 if (type == PANDECODE_MESSAGE)
92 printf("// ");
93 else if (type == PANDECODE_PROPERTY)
94 printf(".");
95
96 va_start(ap, format);
97 vprintf(format, ap);
98 va_end(ap);
99
100 if (type == PANDECODE_PROPERTY)
101 printf(",\n");
102 }
103
104 static void
105 pandecode_log_cont(const char *format, ...)
106 {
107 va_list ap;
108
109 va_start(ap, format);
110 vprintf(format, ap);
111 va_end(ap);
112 }
113
114 struct pandecode_flag_info {
115 u64 flag;
116 const char *name;
117 };
118
119 static void
120 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
121 u64 flags)
122 {
123 bool decodable_flags_found = false;
124
125 for (int i = 0; flag_info[i].name; i++) {
126 if ((flags & flag_info[i].flag) != flag_info[i].flag)
127 continue;
128
129 if (!decodable_flags_found) {
130 decodable_flags_found = true;
131 } else {
132 pandecode_log_cont(" | ");
133 }
134
135 pandecode_log_cont("%s", flag_info[i].name);
136
137 flags &= ~flag_info[i].flag;
138 }
139
140 if (decodable_flags_found) {
141 if (flags)
142 pandecode_log_cont(" | 0x%" PRIx64, flags);
143 } else {
144 pandecode_log_cont("0x%" PRIx64, flags);
145 }
146 }
147
148 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
149 static const struct pandecode_flag_info gl_enable_flag_info[] = {
150 FLAG_INFO(OCCLUSION_QUERY),
151 FLAG_INFO(OCCLUSION_PRECISE),
152 FLAG_INFO(FRONT_CCW_TOP),
153 FLAG_INFO(CULL_FACE_FRONT),
154 FLAG_INFO(CULL_FACE_BACK),
155 {}
156 };
157 #undef FLAG_INFO
158
159 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
160 static const struct pandecode_flag_info clear_flag_info[] = {
161 FLAG_INFO(FAST),
162 FLAG_INFO(SLOW),
163 FLAG_INFO(SLOW_STENCIL),
164 {}
165 };
166 #undef FLAG_INFO
167
168 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
169 static const struct pandecode_flag_info mask_flag_info[] = {
170 FLAG_INFO(R),
171 FLAG_INFO(G),
172 FLAG_INFO(B),
173 FLAG_INFO(A),
174 {}
175 };
176 #undef FLAG_INFO
177
178 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
179 static const struct pandecode_flag_info u3_flag_info[] = {
180 FLAG_INFO(HAS_MSAA),
181 FLAG_INFO(CAN_DISCARD),
182 FLAG_INFO(HAS_BLEND_SHADER),
183 FLAG_INFO(DEPTH_TEST),
184 {}
185 };
186
187 static const struct pandecode_flag_info u4_flag_info[] = {
188 FLAG_INFO(NO_MSAA),
189 FLAG_INFO(NO_DITHER),
190 FLAG_INFO(DEPTH_RANGE_A),
191 FLAG_INFO(DEPTH_RANGE_B),
192 FLAG_INFO(STENCIL_TEST),
193 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
194 {}
195 };
196 #undef FLAG_INFO
197
198 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
199 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
200 FLAG_INFO(MSAA_A),
201 FLAG_INFO(MSAA_B),
202 FLAG_INFO(MSAA_8),
203 {}
204 };
205 #undef FLAG_INFO
206
207 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
208 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
209 FLAG_INFO(AFBC),
210 FLAG_INFO(MSAA),
211 {}
212 };
213 #undef FLAG_INFO
214
215 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
216 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
217 FLAG_INFO(PRESENT),
218 FLAG_INFO(AFBC),
219 FLAG_INFO(ZS),
220 {}
221 };
222 #undef FLAG_INFO
223
224 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
225 static const struct pandecode_flag_info shader_unknown1_flag_info [] = {
226 FLAG_INFO(NO_ALPHA_TO_COVERAGE),
227 FLAG_INFO(READS_TILEBUFFER),
228 FLAG_INFO(READS_ZS),
229 {}
230 };
231 #undef FLAG_INFO
232
233 extern char *replace_fragment;
234 extern char *replace_vertex;
235
236 static char *
237 pandecode_job_type_name(enum mali_job_type type)
238 {
239 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
240
241 switch (type) {
242 DEFINE_CASE(NULL);
243 DEFINE_CASE(SET_VALUE);
244 DEFINE_CASE(CACHE_FLUSH);
245 DEFINE_CASE(COMPUTE);
246 DEFINE_CASE(VERTEX);
247 DEFINE_CASE(TILER);
248 DEFINE_CASE(FUSED);
249 DEFINE_CASE(FRAGMENT);
250
251 case JOB_NOT_STARTED:
252 return "NOT_STARTED";
253
254 default:
255 pandecode_log("Warning! Unknown job type %x\n", type);
256 return "!?!?!?";
257 }
258
259 #undef DEFINE_CASE
260 }
261
262 static char *
263 pandecode_draw_mode_name(enum mali_draw_mode mode)
264 {
265 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
266
267 switch (mode) {
268 DEFINE_CASE(DRAW_NONE);
269 DEFINE_CASE(POINTS);
270 DEFINE_CASE(LINES);
271 DEFINE_CASE(TRIANGLES);
272 DEFINE_CASE(TRIANGLE_STRIP);
273 DEFINE_CASE(TRIANGLE_FAN);
274 DEFINE_CASE(LINE_STRIP);
275 DEFINE_CASE(LINE_LOOP);
276 DEFINE_CASE(POLYGON);
277 DEFINE_CASE(QUADS);
278 DEFINE_CASE(QUAD_STRIP);
279
280 default:
281 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
282 }
283
284 #undef DEFINE_CASE
285 }
286
287 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
288 static char *
289 pandecode_func_name(enum mali_func mode)
290 {
291 switch (mode) {
292 DEFINE_CASE(NEVER);
293 DEFINE_CASE(LESS);
294 DEFINE_CASE(EQUAL);
295 DEFINE_CASE(LEQUAL);
296 DEFINE_CASE(GREATER);
297 DEFINE_CASE(NOTEQUAL);
298 DEFINE_CASE(GEQUAL);
299 DEFINE_CASE(ALWAYS);
300
301 default:
302 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
303 }
304 }
305 #undef DEFINE_CASE
306
307 /* Why is this duplicated? Who knows... */
308 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
309 static char *
310 pandecode_alt_func_name(enum mali_alt_func mode)
311 {
312 switch (mode) {
313 DEFINE_CASE(NEVER);
314 DEFINE_CASE(LESS);
315 DEFINE_CASE(EQUAL);
316 DEFINE_CASE(LEQUAL);
317 DEFINE_CASE(GREATER);
318 DEFINE_CASE(NOTEQUAL);
319 DEFINE_CASE(GEQUAL);
320 DEFINE_CASE(ALWAYS);
321
322 default:
323 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
324 }
325 }
326 #undef DEFINE_CASE
327
328 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
329 static char *
330 pandecode_stencil_op_name(enum mali_stencil_op op)
331 {
332 switch (op) {
333 DEFINE_CASE(KEEP);
334 DEFINE_CASE(REPLACE);
335 DEFINE_CASE(ZERO);
336 DEFINE_CASE(INVERT);
337 DEFINE_CASE(INCR_WRAP);
338 DEFINE_CASE(DECR_WRAP);
339 DEFINE_CASE(INCR);
340 DEFINE_CASE(DECR);
341
342 default:
343 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
344 }
345 }
346
347 #undef DEFINE_CASE
348
349 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
350 static char *pandecode_attr_mode_name(enum mali_attr_mode mode)
351 {
352 switch(mode) {
353 DEFINE_CASE(UNUSED);
354 DEFINE_CASE(LINEAR);
355 DEFINE_CASE(POT_DIVIDE);
356 DEFINE_CASE(MODULO);
357 DEFINE_CASE(NPOT_DIVIDE);
358 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
359 }
360 }
361
362 #undef DEFINE_CASE
363
364 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
365 static char *
366 pandecode_channel_name(enum mali_channel channel)
367 {
368 switch (channel) {
369 DEFINE_CASE(RED);
370 DEFINE_CASE(GREEN);
371 DEFINE_CASE(BLUE);
372 DEFINE_CASE(ALPHA);
373 DEFINE_CASE(ZERO);
374 DEFINE_CASE(ONE);
375 DEFINE_CASE(RESERVED_0);
376 DEFINE_CASE(RESERVED_1);
377
378 default:
379 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
380 }
381 }
382 #undef DEFINE_CASE
383
384 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
385 static char *
386 pandecode_wrap_mode_name(enum mali_wrap_mode op)
387 {
388 switch (op) {
389 DEFINE_CASE(REPEAT);
390 DEFINE_CASE(CLAMP_TO_EDGE);
391 DEFINE_CASE(CLAMP_TO_BORDER);
392 DEFINE_CASE(MIRRORED_REPEAT);
393
394 default:
395 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
396 }
397 }
398 #undef DEFINE_CASE
399
400 static inline char *
401 pandecode_decode_fbd_type(enum mali_fbd_type type)
402 {
403 if (type == MALI_SFBD) return "SFBD";
404 else if (type == MALI_MFBD) return "MFBD";
405 else return "WATFBD /* XXX */";
406 }
407
408 static void
409 pandecode_replay_sfbd(uint64_t gpu_va, int job_no)
410 {
411 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
412 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
413
414 pandecode_log("struct mali_single_framebuffer framebuffer_%d = {\n", job_no);
415 pandecode_indent++;
416
417 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
418 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
419
420 pandecode_log(".format = ");
421 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
422 pandecode_log_cont(",\n");
423
424 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
425 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
426
427 MEMORY_PROP(s, framebuffer);
428 pandecode_prop("stride = %d", s->stride);
429
430 /* Earlier in the actual commandstream -- right before width -- but we
431 * delay to flow nicer */
432
433 pandecode_log(".clear_flags = ");
434 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
435 pandecode_log_cont(",\n");
436
437 if (s->depth_buffer | s->depth_buffer_enable) {
438 MEMORY_PROP(s, depth_buffer);
439 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
440 }
441
442 if (s->stencil_buffer | s->stencil_buffer_enable) {
443 MEMORY_PROP(s, stencil_buffer);
444 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
445 }
446
447 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
448 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
449 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
450 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
451 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
452 }
453
454 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
455 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
456 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
457 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
458 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
459 }
460
461 if (s->clear_stencil) {
462 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
463 }
464
465 MEMORY_PROP(s, unknown_address_0);
466 MEMORY_PROP(s, unknown_address_1);
467 MEMORY_PROP(s, unknown_address_2);
468
469 pandecode_prop("resolution_check = 0x%" PRIx32, s->resolution_check);
470 pandecode_prop("tiler_flags = 0x%" PRIx32, s->tiler_flags);
471
472 MEMORY_PROP(s, tiler_heap_free);
473 MEMORY_PROP(s, tiler_heap_end);
474
475 pandecode_indent--;
476 pandecode_log("};\n");
477
478 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
479 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
480 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
481 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
482
483 printf(".zero3 = {");
484
485 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
486 printf("%X, ", s->zero3[i]);
487
488 printf("},\n");
489
490 printf(".zero6 = {");
491
492 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
493 printf("%X, ", s->zero6[i]);
494
495 printf("},\n");
496 }
497
498 static void
499 pandecode_replay_swizzle(unsigned swizzle)
500 {
501 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
502 pandecode_channel_name((swizzle >> 0) & 0x7),
503 pandecode_channel_name((swizzle >> 3) & 0x7),
504 pandecode_channel_name((swizzle >> 6) & 0x7),
505 pandecode_channel_name((swizzle >> 9) & 0x7));
506 }
507
508 static void
509 pandecode_rt_format(struct mali_rt_format format)
510 {
511 pandecode_log(".format = {\n");
512 pandecode_indent++;
513
514 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
515 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
516
517 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
518 MALI_NEGATIVE(format.nr_channels));
519
520 pandecode_log(".flags = ");
521 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
522 pandecode_log_cont(",\n");
523
524 pandecode_replay_swizzle(format.swizzle);
525
526 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
527
528 pandecode_indent--;
529 pandecode_log("},\n");
530 }
531
532 static void
533 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no)
534 {
535 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
536 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
537
538 if (fb->sample_locations) {
539 /* The blob stores all possible sample locations in a single buffer
540 * allocated on startup, and just switches the pointer when switching
541 * MSAA state. For now, we just put the data into the cmdstream, but we
542 * should do something like what the blob does with a real driver.
543 *
544 * There seem to be 32 slots for sample locations, followed by another
545 * 16. The second 16 is just the center location followed by 15 zeros
546 * in all the cases I've identified (maybe shader vs. depth/color
547 * samples?).
548 */
549
550 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
551
552 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
553
554 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
555 pandecode_indent++;
556
557 for (int i = 0; i < 32 + 16; i++) {
558 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
559 }
560
561 pandecode_indent--;
562 pandecode_log("};\n");
563 }
564
565 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no);
566 pandecode_indent++;
567
568 pandecode_prop("unk0 = 0x%x", fb->unk0);
569
570 if (fb->sample_locations)
571 pandecode_prop("sample_locations = sample_locations_%d", job_no);
572
573 /* Assume that unknown1 and tiler_meta were emitted in the last job for
574 * now */
575 /*pandecode_prop("unknown1 = unknown1_%d_p", job_no - 1);
576 pandecode_prop("tiler_meta = tiler_meta_%d_p", job_no - 1);*/
577 MEMORY_PROP(fb, unknown1);
578 MEMORY_PROP(fb, tiler_meta);
579
580 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
581 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
582 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
583 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
584
585 pandecode_prop("unk1 = 0x%x", fb->unk1);
586 pandecode_prop("unk2 = 0x%x", fb->unk2);
587 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
588 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
589
590 pandecode_prop("unk3 = 0x%x", fb->unk3);
591 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
592 pandecode_prop("clear_depth = %f", fb->clear_depth);
593
594 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
595 MEMORY_PROP(fb, scratchpad);
596 MEMORY_PROP(fb, tiler_scratch_start);
597 MEMORY_PROP(fb, tiler_scratch_middle);
598 MEMORY_PROP(fb, tiler_heap_start);
599 MEMORY_PROP(fb, tiler_heap_end);
600
601 if (fb->zero3 || fb->zero4 || fb->zero9 || fb->zero10 || fb->zero11 || fb->zero12) {
602 pandecode_msg("framebuffer zeros tripped\n");
603 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
604 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
605 pandecode_prop("zero9 = 0x%" PRIx64, fb->zero9);
606 pandecode_prop("zero10 = 0x%" PRIx64, fb->zero10);
607 pandecode_prop("zero11 = 0x%" PRIx64, fb->zero11);
608 pandecode_prop("zero12 = 0x%" PRIx64, fb->zero12);
609 }
610
611 pandecode_indent--;
612 pandecode_log("};\n");
613
614 gpu_va += sizeof(struct bifrost_framebuffer);
615
616 if (fb->unk3 & MALI_MFBD_EXTRA) {
617 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
618 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
619
620 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no);
621 pandecode_indent++;
622
623 MEMORY_PROP(fbx, checksum);
624
625 if (fbx->checksum_stride)
626 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
627
628 pandecode_log(".flags = ");
629 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
630 pandecode_log_cont(",\n");
631
632 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
633 pandecode_log(".ds_afbc = {\n");
634 pandecode_indent++;
635
636 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
637 pandecode_prop("depth_stencil_afbc_stride = %d",
638 fbx->ds_afbc.depth_stencil_afbc_stride);
639 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
640
641 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
642 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
643 pandecode_prop("zero1 = 0x%" PRIx32,
644 fbx->ds_afbc.zero1);
645 pandecode_prop("padding = 0x%" PRIx64,
646 fbx->ds_afbc.padding);
647 }
648
649 pandecode_indent--;
650 pandecode_log("},\n");
651 } else {
652 pandecode_log(".ds_linear = {\n");
653 pandecode_indent++;
654
655 if (fbx->ds_linear.depth) {
656 MEMORY_PROP((&fbx->ds_linear), depth);
657 pandecode_prop("depth_stride = %d",
658 fbx->ds_linear.depth_stride);
659 }
660
661 if (fbx->ds_linear.stencil) {
662 MEMORY_PROP((&fbx->ds_linear), stencil);
663 pandecode_prop("stencil_stride = %d",
664 fbx->ds_linear.stencil_stride);
665 }
666
667 if (fbx->ds_linear.depth_stride_zero ||
668 fbx->ds_linear.stencil_stride_zero ||
669 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
670 pandecode_msg("Depth/stencil zeros tripped\n");
671 pandecode_prop("depth_stride_zero = 0x%x",
672 fbx->ds_linear.depth_stride_zero);
673 pandecode_prop("stencil_stride_zero = 0x%x",
674 fbx->ds_linear.stencil_stride_zero);
675 pandecode_prop("zero1 = 0x%" PRIx32,
676 fbx->ds_linear.zero1);
677 pandecode_prop("zero2 = 0x%" PRIx32,
678 fbx->ds_linear.zero2);
679 }
680
681 pandecode_indent--;
682 pandecode_log("},\n");
683 }
684
685 if (fbx->zero3 || fbx->zero4) {
686 pandecode_msg("fb_extra zeros tripped\n");
687 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
688 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
689 }
690
691 pandecode_indent--;
692 pandecode_log("};\n");
693
694 gpu_va += sizeof(struct bifrost_fb_extra);
695 }
696
697 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no);
698 pandecode_indent++;
699
700 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
701 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
702 mem = pandecode_find_mapped_gpu_mem_containing(rt_va);
703 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
704
705 pandecode_log("{\n");
706 pandecode_indent++;
707
708 pandecode_rt_format(rt->format);
709
710 /* TODO: How the actual heck does AFBC enabling work here? */
711 if (0) {
712 pandecode_log(".afbc = {\n");
713 pandecode_indent++;
714
715 char *a = pointer_as_memory_reference(rt->afbc.metadata);
716 pandecode_prop("metadata = %s", a);
717 free(a);
718
719 pandecode_prop("stride = %d", rt->afbc.stride);
720 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
721
722 pandecode_indent--;
723 pandecode_log("},\n");
724 } else {
725 pandecode_log(".chunknown = {\n");
726 pandecode_indent++;
727
728 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
729
730 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
731 pandecode_prop("pointer = %s", a);
732 free(a);
733
734 pandecode_indent--;
735 pandecode_log("},\n");
736 }
737
738 MEMORY_PROP(rt, framebuffer);
739 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
740
741 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
742 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
743 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
744 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
745 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
746 }
747
748 if (rt->zero1 || rt->zero2 || rt->zero3) {
749 pandecode_msg("render target zeros tripped\n");
750 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
751 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
752 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
753 }
754
755 pandecode_indent--;
756 pandecode_log("},\n");
757 }
758
759 pandecode_indent--;
760 pandecode_log("};\n");
761 }
762
763 static void
764 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
765 mali_ptr addr, int job_no, char *suffix,
766 int count, bool varying)
767 {
768 char *prefix = varying ? "varyings" : "attributes";
769
770 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
771
772 char base[128];
773 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
774
775 for (int i = 0; i < count; ++i) {
776 enum mali_attr_mode mode = attr[i].elements & 7;
777
778 if (mode == MALI_ATTR_UNUSED)
779 continue;
780
781 mali_ptr raw_elements = attr[i].elements & ~7;
782
783 /* TODO: Do we maybe want to dump the attribute values
784 * themselves given the specified format? Or is that too hard?
785 * */
786
787 char *a = pointer_as_memory_reference(raw_elements);
788 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
789 free(a);
790 }
791
792 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
793 pandecode_indent++;
794
795 for (int i = 0; i < count; ++i) {
796 pandecode_log("{\n");
797 pandecode_indent++;
798
799 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(attr[i].elements & 7));
800 pandecode_prop("shift = %d", attr[i].shift);
801 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
802 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
803 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
804 pandecode_indent--;
805 pandecode_log("}, \n");
806
807 if ((attr[i].elements & 7) == MALI_ATTR_NPOT_DIVIDE) {
808 i++;
809 pandecode_log("{\n");
810 pandecode_indent++;
811 pandecode_prop("unk = 0x%x", attr[i].unk);
812 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
813 if (attr[i].zero != 0)
814 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
815 pandecode_prop("divisor = %d", attr[i].divisor);
816 pandecode_indent--;
817 pandecode_log("}, \n");
818 }
819
820 }
821
822 pandecode_indent--;
823 pandecode_log("};\n");
824 }
825
826 static mali_ptr
827 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
828 {
829 /* TODO: Decode flags */
830 mali_ptr shader_ptr = ptr & ~15;
831
832 char *a = pointer_as_memory_reference(shader_ptr);
833 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
834 free(a);
835
836 return shader_ptr;
837 }
838
839 static void
840 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
841 {
842 const char *func = pandecode_func_name(stencil->func);
843 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
844 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
845 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
846
847 if (stencil->zero)
848 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
849
850 pandecode_log(".stencil_%s = {\n", name);
851 pandecode_indent++;
852 pandecode_prop("ref = %d", stencil->ref);
853 pandecode_prop("mask = 0x%02X", stencil->mask);
854 pandecode_prop("func = %s", func);
855 pandecode_prop("sfail = %s", sfail);
856 pandecode_prop("dpfail = %s", dpfail);
857 pandecode_prop("dppass = %s", dppass);
858 pandecode_indent--;
859 pandecode_log("},\n");
860 }
861
862 static void
863 pandecode_replay_blend_equation(const struct mali_blend_equation *blend)
864 {
865 if (blend->zero1)
866 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
867
868 pandecode_log(".equation = {\n");
869 pandecode_indent++;
870
871 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
872 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
873
874 pandecode_log(".color_mask = ");
875 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
876 pandecode_log_cont(",\n");
877
878 pandecode_indent--;
879 pandecode_log("},\n");
880 }
881
882 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
883
884 static unsigned
885 decode_bifrost_constant(u16 constant)
886 {
887 float lo = (float) (constant & 0xFF);
888 float hi = (float) (constant >> 8);
889
890 return (hi / 255.0) + (lo / 65535.0);
891 }
892
893 static mali_ptr
894 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
895 {
896 struct bifrost_blend_rt *b =
897 ((struct bifrost_blend_rt *) descs) + rt_no;
898
899 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
900 pandecode_indent++;
901
902 pandecode_prop("flags = 0x%" PRIx16, b->flags);
903 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
904 b->constant, decode_bifrost_constant(b->constant));
905
906 /* TODO figure out blend shader enable bit */
907 pandecode_replay_blend_equation(&b->equation);
908 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
909 pandecode_prop("index = 0x%" PRIx16, b->index);
910 pandecode_prop("shader = 0x%" PRIx32, b->shader);
911
912 pandecode_indent--;
913 pandecode_log("},\n");
914
915 return 0;
916 }
917
918 static mali_ptr
919 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
920 {
921 pandecode_log(".blend = {\n");
922 pandecode_indent++;
923
924 if (is_shader) {
925 pandecode_replay_shader_address("shader", blend->shader);
926 } else {
927 pandecode_replay_blend_equation(&blend->equation);
928 pandecode_prop("constant = %f", blend->constant);
929 }
930
931 pandecode_indent--;
932 pandecode_log("},\n");
933
934 /* Return blend shader to disassemble if present */
935 return is_shader ? (blend->shader & ~0xF) : 0;
936 }
937
938 static mali_ptr
939 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
940 {
941 struct midgard_blend_rt *b =
942 ((struct midgard_blend_rt *) descs) + rt_no;
943
944 /* Flags determine presence of blend shader */
945 bool is_shader = (b->flags & 0xF) >= 0x2;
946
947 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
948 pandecode_indent++;
949
950 pandecode_prop("flags = 0x%" PRIx64, b->flags);
951
952 mali_ptr shader = pandecode_midgard_blend(&b->blend, is_shader);
953
954 pandecode_indent--;
955 pandecode_log("};\n");
956
957 return shader;
958 }
959
960 static int
961 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
962 {
963 char base[128];
964 char *prefix = varying ? "varying" : "attribute";
965 unsigned max_index = 0;
966 snprintf(base, sizeof(base), "%s_meta", prefix);
967
968 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
969 pandecode_indent++;
970
971 struct mali_attr_meta *attr_meta;
972 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
973
974 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
975
976 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
977 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
978 sizeof(*attr_mem));
979
980 pandecode_log("{\n");
981 pandecode_indent++;
982 pandecode_prop("index = %d", attr_meta->index);
983
984 if (attr_meta->index > max_index)
985 max_index = attr_meta->index;
986 pandecode_replay_swizzle(attr_meta->swizzle);
987 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
988
989 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
990 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
991 pandecode_prop("src_offset = 0x%" PRIx64, (u64) attr_meta->src_offset);
992 pandecode_indent--;
993 pandecode_log("},\n");
994
995 }
996
997 pandecode_indent--;
998 pandecode_log("};\n");
999
1000 return max_index;
1001 }
1002
1003 static void
1004 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1005 {
1006 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1007
1008 if (imem) {
1009 /* Indices are literally just a u32 array :) */
1010
1011 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1012
1013 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1014 pandecode_indent++;
1015
1016 for (unsigned i = 0; i < (index_count + 1); i += 3)
1017 pandecode_log("%d, %d, %d,\n",
1018 indices[i],
1019 indices[i + 1],
1020 indices[i + 2]);
1021
1022 pandecode_indent--;
1023 pandecode_log("};\n");
1024 }
1025 }
1026
1027 /* return bits [lo, hi) of word */
1028 static u32
1029 bits(u32 word, u32 lo, u32 hi)
1030 {
1031 if (hi - lo >= 32)
1032 return word; // avoid undefined behavior with the shift
1033
1034 return (word >> lo) & ((1 << (hi - lo)) - 1);
1035 }
1036
1037 static void
1038 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1039 {
1040 pandecode_log_cont("{\n");
1041 pandecode_indent++;
1042
1043 pandecode_prop("invocation_count = %" PRIx32, p->invocation_count);
1044 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1045 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1046 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1047 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1048 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1049 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1050
1051 /* Decode invocation_count. See the comment before the definition of
1052 * invocation_count for an explanation.
1053 */
1054 pandecode_msg("size: (%d, %d, %d)\n",
1055 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1056 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1057 bits(p->invocation_count, p->size_z_shift,
1058 p->workgroups_x_shift) + 1);
1059 pandecode_msg("workgroups: (%d, %d, %d)\n",
1060 bits(p->invocation_count, p->workgroups_x_shift,
1061 p->workgroups_y_shift) + 1,
1062 bits(p->invocation_count, p->workgroups_y_shift,
1063 p->workgroups_z_shift) + 1,
1064 bits(p->invocation_count, p->workgroups_z_shift,
1065 32) + 1);
1066
1067 /* TODO: Decode */
1068 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1069 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1070
1071 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
1072
1073 /* Index count only exists for tiler jobs anyway */
1074
1075 if (p->index_count)
1076 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1077
1078 pandecode_prop("negative_start = %d", p->negative_start);
1079
1080 DYN_MEMORY_PROP(p, job_no, indices);
1081
1082 if (p->zero1) {
1083 pandecode_msg("Zero tripped\n");
1084 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1085 }
1086
1087 pandecode_indent--;
1088 pandecode_log("},\n");
1089 }
1090
1091 static void
1092 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1093 {
1094 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1095
1096 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1097
1098 for (int i = 0; i < ubufs_count; i++) {
1099 mali_ptr ptr = ubufs[i].ptr << 2;
1100 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1101 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1102 char name[50];
1103 snprintf(name, sizeof(name), "ubuf_%d", i);
1104 /* The blob uses ubuf 0 to upload internal stuff and
1105 * uniforms that won't fit/are accessed indirectly, so
1106 * it puts it in the batchbuffer.
1107 */
1108 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1109 pandecode_indent++;
1110
1111 for (int j = 0; j <= ubufs[i].size; j++) {
1112 for (int k = 0; k < 4; k++) {
1113 if (k == 0)
1114 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1115 else
1116 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1117
1118 }
1119
1120 pandecode_log_cont("\n");
1121 }
1122
1123 pandecode_indent--;
1124 pandecode_log("};\n");
1125 }
1126
1127 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1128 job_no);
1129 pandecode_indent++;
1130
1131 for (int i = 0; i < ubufs_count; i++) {
1132 pandecode_log("{\n");
1133 pandecode_indent++;
1134 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1135 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1136 pandecode_indent--;
1137 pandecode_log("},\n");
1138 }
1139
1140 pandecode_indent--;
1141 pandecode_log("};\n");
1142 }
1143
1144 static void
1145 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1146 {
1147
1148 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1149
1150 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1151
1152 if (scratchpad->zero)
1153 pandecode_msg("XXX scratchpad zero tripped");
1154
1155 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no, suffix);
1156 pandecode_indent++;
1157
1158 pandecode_prop("flags = 0x%x", scratchpad->flags);
1159 MEMORY_PROP(scratchpad, gpu_scratchpad);
1160
1161 pandecode_indent--;
1162 pandecode_log("};\n");
1163 }
1164
1165 static void
1166 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1167 bool is_bifrost)
1168 {
1169 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1170 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1171
1172 /* Compute maximum possible size */
1173 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1174
1175 /* Print some boilerplate to clearly denote the assembly (which doesn't
1176 * obey indentation rules), and actually do the disassembly! */
1177
1178 printf("\n\n");
1179
1180 if (is_bifrost) {
1181 disassemble_bifrost(code, sz, false);
1182 } else {
1183 disassemble_midgard(code, sz);
1184 }
1185
1186 printf("\n\n");
1187 }
1188
1189 static void
1190 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1191 int job_no, enum mali_job_type job_type,
1192 char *suffix, bool is_bifrost)
1193 {
1194 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1195 struct pandecode_mapped_memory *attr_mem;
1196
1197 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1198 * are the only things actually needed from the FBD, vertex/tiler jobs
1199 * no longer reference the FBD -- instead, this field points to some
1200 * info about the scratchpad.
1201 */
1202 if (is_bifrost)
1203 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1204 else if (p->framebuffer & MALI_MFBD)
1205 pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no);
1206 else
1207 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1208
1209 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1210 int texture_count = 0, sampler_count = 0;
1211
1212 if (shader_meta_ptr) {
1213 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1214 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1215
1216 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no, suffix);
1217 pandecode_indent++;
1218
1219 /* Save for dumps */
1220 attribute_count = s->attribute_count;
1221 varying_count = s->varying_count;
1222 texture_count = s->texture_count;
1223 sampler_count = s->sampler_count;
1224
1225 if (is_bifrost) {
1226 uniform_count = s->bifrost2.uniform_count;
1227 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1228 } else {
1229 uniform_count = s->midgard1.uniform_count;
1230 /* TODO figure this out */
1231 uniform_buffer_count = 1;
1232 }
1233
1234 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1235
1236 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1237 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1238 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1239 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1240
1241 if (is_bifrost) {
1242 pandecode_log(".bifrost1 = {\n");
1243 pandecode_indent++;
1244
1245 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1246 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1247
1248 pandecode_indent--;
1249 pandecode_log("},\n");
1250 } else {
1251 pandecode_log(".midgard1 = {\n");
1252 pandecode_indent++;
1253
1254 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1255 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1256
1257 pandecode_log(".unknown1 = ");
1258 pandecode_log_decoded_flags(shader_unknown1_flag_info, s->midgard1.unknown1);
1259 pandecode_log_cont(",\n");
1260
1261 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1262
1263 pandecode_indent--;
1264 pandecode_log("},\n");
1265 }
1266
1267 if (s->depth_units || s->depth_factor) {
1268 if (is_bifrost)
1269 pandecode_prop("depth_units = %f", s->depth_units);
1270 else
1271 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f);
1272
1273 pandecode_prop("depth_factor = %f", s->depth_factor);
1274 }
1275
1276 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1277 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1278
1279 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1280 invert_alpha_coverage ? "~" : "",
1281 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1282
1283 pandecode_log(".unknown2_3 = ");
1284
1285 int unknown2_3 = s->unknown2_3;
1286 int unknown2_4 = s->unknown2_4;
1287
1288 /* We're not quite sure what these flags mean without the depth test, if anything */
1289
1290 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1291 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1292 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1293
1294 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1295 }
1296
1297 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1298 pandecode_log_cont(",\n");
1299
1300 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1301 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1302
1303 pandecode_log(".unknown2_4 = ");
1304 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1305 pandecode_log_cont(",\n");
1306
1307 pandecode_replay_stencil("front", &s->stencil_front);
1308 pandecode_replay_stencil("back", &s->stencil_back);
1309
1310 if (is_bifrost) {
1311 pandecode_log(".bifrost2 = {\n");
1312 pandecode_indent++;
1313
1314 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1315 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1316 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1317 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1318
1319 pandecode_indent--;
1320 pandecode_log("},\n");
1321 } else {
1322 pandecode_log(".midgard2 = {\n");
1323 pandecode_indent++;
1324
1325 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1326 pandecode_indent--;
1327 pandecode_log("},\n");
1328 }
1329
1330 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1331
1332 if (!is_bifrost) {
1333 /* TODO: Blend shaders routing/disasm */
1334
1335 pandecode_midgard_blend(&s->blend, false);
1336 }
1337
1338 pandecode_indent--;
1339 pandecode_log("};\n");
1340
1341 /* MRT blend fields are used whenever MFBD is used, with
1342 * per-RT descriptors */
1343
1344 if (job_type == JOB_TYPE_TILER) {
1345 void* blend_base = (void *) (s + 1);
1346
1347 for (unsigned i = 0; i < 4; i++) {
1348 mali_ptr shader = 0;
1349
1350 if (is_bifrost)
1351 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1352 else
1353 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1354
1355 if (shader)
1356 pandecode_shader_disassemble(shader, job_no, job_type, false);
1357 }
1358 }
1359
1360 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1361 } else
1362 pandecode_msg("<no shader>\n");
1363
1364 if (p->viewport) {
1365 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1366 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1367
1368 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1369 pandecode_indent++;
1370
1371 pandecode_prop("clip_minx = %f", f->clip_minx);
1372 pandecode_prop("clip_miny = %f", f->clip_miny);
1373 pandecode_prop("clip_minz = %f", f->clip_minz);
1374 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1375 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1376 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1377
1378 /* Only the higher coordinates are MALI_POSITIVE scaled */
1379
1380 pandecode_prop("viewport0 = { %d, %d }",
1381 f->viewport0[0], f->viewport0[1]);
1382
1383 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1384 f->viewport1[0] + 1, f->viewport1[1] + 1);
1385
1386 pandecode_indent--;
1387 pandecode_log("};\n");
1388 }
1389
1390 if (p->attribute_meta) {
1391 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1392
1393 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1394 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1395 }
1396
1397 /* Varyings are encoded like attributes but not actually sent; we just
1398 * pass a zero buffer with the right stride/size set, (or whatever)
1399 * since the GPU will write to it itself */
1400
1401 if (p->varyings) {
1402 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1403
1404 /* Number of descriptors depends on whether there are
1405 * non-internal varyings */
1406
1407 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1408 }
1409
1410 if (p->varying_meta) {
1411 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1412 }
1413
1414 if (p->uniforms) {
1415 int rows = uniform_count, width = 4;
1416 size_t sz = rows * width * sizeof(float);
1417
1418 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1419 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1420 float *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1421
1422 pandecode_log("float uniforms_%d%s[] = {\n", job_no, suffix);
1423
1424 pandecode_indent++;
1425
1426 for (int row = 0; row < rows; row++) {
1427 for (int i = 0; i < width; i++)
1428 pandecode_log_cont("%ff, ", uniforms[i]);
1429
1430 pandecode_log_cont("\n");
1431
1432 uniforms += width;
1433 }
1434
1435 pandecode_indent--;
1436 pandecode_log("};\n");
1437 }
1438
1439 if (p->uniform_buffers) {
1440 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1441 }
1442
1443 if (p->texture_trampoline) {
1444 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1445
1446 if (mmem) {
1447 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1448
1449 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1450 pandecode_indent++;
1451
1452 for (int tex = 0; tex < texture_count; ++tex) {
1453 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1454 char *a = pointer_as_memory_reference(*u);
1455 pandecode_log("%s,\n", a);
1456 free(a);
1457 }
1458
1459 pandecode_indent--;
1460 pandecode_log("};\n");
1461
1462 /* Now, finally, descend down into the texture descriptor */
1463 for (int tex = 0; tex < texture_count; ++tex) {
1464 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1465 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1466
1467 if (tmem) {
1468 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1469
1470 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
1471 pandecode_indent++;
1472
1473 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1474 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1475 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1476
1477 pandecode_prop("unknown1 = %" PRId16, t->unknown1);
1478 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1479 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1480 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1481
1482 struct mali_texture_format f = t->format;
1483
1484 pandecode_log(".format = {\n");
1485 pandecode_indent++;
1486
1487 pandecode_replay_swizzle(f.swizzle);
1488 pandecode_prop("format = %s", pandecode_format_name(f.format));
1489
1490 pandecode_prop("usage1 = 0x%" PRIx32, f.usage1);
1491 pandecode_prop("is_not_cubemap = %" PRId32, f.is_not_cubemap);
1492 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1493
1494 pandecode_indent--;
1495 pandecode_log("},\n");
1496
1497 pandecode_replay_swizzle(t->swizzle);
1498
1499 if (t->swizzle_zero) {
1500 /* Shouldn't happen */
1501 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1502 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1503 }
1504
1505 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1506
1507 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1508 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1509 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1510
1511 pandecode_log(".swizzled_bitmaps = {\n");
1512 pandecode_indent++;
1513
1514 /* A bunch of bitmap pointers follow.
1515 * We work out the correct number,
1516 * based on the mipmap/cubemap
1517 * properties, but dump extra
1518 * possibilities to futureproof */
1519
1520 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1521
1522 if (!f.is_not_cubemap) {
1523 /* Miptree for each face */
1524 bitmap_count *= 6;
1525 }
1526
1527 if (f.usage2 & MALI_TEX_MANUAL_STRIDE) {
1528 /* Stride for each... what exactly? TODO More traces */
1529
1530 if (bitmap_count > 1) {
1531 pandecode_msg("Manual stride with mip/cubemaps, decode uncertain");
1532 }
1533
1534 /* This is a guess, we've only
1535 * seen for 1-level non-mip 2D
1536 * */
1537
1538 bitmap_count += 1;
1539 }
1540
1541 int max_count = sizeof(t->swizzled_bitmaps) / sizeof(t->swizzled_bitmaps[0]);
1542
1543 if (bitmap_count > max_count) {
1544 pandecode_msg("XXX: bitmap count tripped");
1545 bitmap_count = max_count;
1546 }
1547
1548 /* Dump more to be safe, but not _that_ much more */
1549 int safe_count = MIN2(bitmap_count * 2, max_count);
1550
1551 for (int i = 0; i < safe_count; ++i) {
1552 char *prefix = (i >= bitmap_count) ? "// " : "";
1553
1554 /* How we dump depends if this is a stride or a pointer */
1555
1556 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && ((i + 1) == bitmap_count)) {
1557 /* signed 32-bit snuck in as a 64-bit pointer */
1558 uint64_t stride_set = t->swizzled_bitmaps[i];
1559 uint32_t clamped_stride = stride_set;
1560 int32_t stride = clamped_stride;
1561 assert(stride_set == clamped_stride);
1562 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1563 } else {
1564 char *a = pointer_as_memory_reference(t->swizzled_bitmaps[i]);
1565 pandecode_log("%s%s, \n", prefix, a);
1566 free(a);
1567 }
1568 }
1569
1570 pandecode_indent--;
1571 pandecode_log("},\n");
1572
1573 pandecode_indent--;
1574 pandecode_log("};\n");
1575 }
1576 }
1577 }
1578 }
1579
1580 if (p->sampler_descriptor) {
1581 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1582
1583 if (smem) {
1584 struct mali_sampler_descriptor *s;
1585
1586 mali_ptr d = p->sampler_descriptor;
1587
1588 for (int i = 0; i < sampler_count; ++i) {
1589 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1590
1591 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1592 pandecode_indent++;
1593
1594 /* Only the lower two bits are understood right now; the rest we display as hex */
1595 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1596 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1597 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1598 s->filter_mode & ~3);
1599
1600 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1601 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1602
1603 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1604 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1605 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1606
1607 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1608
1609 if (s->zero || s->zero2) {
1610 pandecode_msg("Zero tripped\n");
1611 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1612 }
1613
1614 pandecode_prop("unknown2 = %d", s->unknown2);
1615
1616 pandecode_prop("border_color = { %f, %f, %f, %f }",
1617 s->border_color[0],
1618 s->border_color[1],
1619 s->border_color[2],
1620 s->border_color[3]);
1621
1622 pandecode_indent--;
1623 pandecode_log("};\n");
1624 }
1625 }
1626 }
1627 }
1628
1629 static void
1630 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1631 {
1632 pandecode_log_cont("{\n");
1633 pandecode_indent++;
1634
1635 MEMORY_PROP(p, position_varying);
1636 MEMORY_COMMENT(p, position_varying);
1637 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1638 MEMORY_COMMENT(p, uniform_buffers);
1639 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1640 MEMORY_COMMENT(p, texture_trampoline);
1641 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1642 MEMORY_COMMENT(p, sampler_descriptor);
1643 DYN_MEMORY_PROP(p, job_no, uniforms);
1644 MEMORY_COMMENT(p, uniforms);
1645 DYN_MEMORY_PROP(p, job_no, attributes);
1646 MEMORY_COMMENT(p, attributes);
1647 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1648 MEMORY_COMMENT(p, attribute_meta);
1649 DYN_MEMORY_PROP(p, job_no, varyings);
1650 MEMORY_COMMENT(p, varyings);
1651 DYN_MEMORY_PROP(p, job_no, varying_meta);
1652 MEMORY_COMMENT(p, varying_meta);
1653 DYN_MEMORY_PROP(p, job_no, viewport);
1654 MEMORY_COMMENT(p, viewport);
1655 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1656 MEMORY_COMMENT(p, occlusion_counter);
1657 MEMORY_COMMENT(p, framebuffer & ~1);
1658 pandecode_msg("%" PRIx64 "\n", p->viewport);
1659 pandecode_msg("%" PRIx64 "\n", p->framebuffer);
1660
1661 if (is_bifrost)
1662 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1663 else
1664 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1665
1666 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1667 pandecode_prop("flags = %d", p->flags);
1668
1669 pandecode_indent--;
1670 pandecode_log("},\n");
1671 }
1672
1673 static void
1674 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1675 {
1676 pandecode_log_cont("{\n");
1677 pandecode_indent++;
1678
1679 pandecode_prop("unk2 = 0x%x", v->unk2);
1680
1681 if (v->zero0 || v->zero1) {
1682 pandecode_msg("vertex only zero tripped");
1683 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1684 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1685 }
1686
1687 pandecode_indent--;
1688 pandecode_log("}\n");
1689 }
1690
1691 static void
1692 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1693 {
1694
1695 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1696 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1697
1698 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1699 pandecode_indent++;
1700
1701 if (h->zero) {
1702 pandecode_msg("tiler heap zero tripped\n");
1703 pandecode_prop("zero = 0x%x", h->zero);
1704 }
1705
1706 for (int i = 0; i < 12; i++) {
1707 if (h->zeros[i] != 0) {
1708 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1709 i, h->zeros[i]);
1710 }
1711 }
1712
1713 pandecode_prop("heap_size = 0x%x", h->heap_size);
1714 MEMORY_PROP(h, tiler_heap_start);
1715 MEMORY_PROP(h, tiler_heap_free);
1716
1717 /* this might point to the beginning of another buffer, when it's
1718 * really the end of the tiler heap buffer, so we have to be careful
1719 * here.
1720 */
1721 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1722 pandecode_prop("tiler_heap_end = %s + 1", a);
1723 free(a);
1724
1725 pandecode_indent--;
1726 pandecode_log("};\n");
1727 }
1728
1729 static void
1730 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1731 {
1732 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1733 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1734
1735 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1736
1737 pandecode_log("struct mali_tiler_meta tiler_meta_%d = {\n", job_no);
1738 pandecode_indent++;
1739
1740 if (t->zero0 || t->zero1) {
1741 pandecode_msg("tiler meta zero tripped");
1742 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1743 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1744 }
1745
1746 pandecode_prop("unk = 0x%x", t->unk);
1747 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1748 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
1749 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
1750
1751 for (int i = 0; i < 12; i++) {
1752 if (t->zeros[i] != 0) {
1753 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
1754 i, t->zeros[i]);
1755 }
1756 }
1757
1758 pandecode_indent--;
1759 pandecode_log("};\n");
1760 }
1761
1762 static void
1763 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
1764 {
1765 pandecode_log(".gl_enables = ");
1766
1767 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
1768
1769 pandecode_log_cont(",\n");
1770 }
1771
1772 static void
1773 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
1774 {
1775 pandecode_log(".primitive_size = {\n");
1776 pandecode_indent++;
1777
1778 if (constant) {
1779 pandecode_prop("constant = %f", u.constant);
1780 } else {
1781 MEMORY_PROP((&u), pointer);
1782 }
1783
1784 pandecode_indent--;
1785 pandecode_log("},\n");
1786 }
1787
1788 static void
1789 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
1790 {
1791 pandecode_log_cont("{\n");
1792 pandecode_indent++;
1793
1794 /* TODO: gl_PointSize on Bifrost */
1795 pandecode_replay_primitive_size(t->primitive_size, true);
1796
1797 DYN_MEMORY_PROP(t, job_no, tiler_meta);
1798 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
1799
1800 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
1801 || t->zero6 || t->zero7 || t->zero8) {
1802 pandecode_msg("tiler only zero tripped");
1803 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1804 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
1805 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
1806 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
1807 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
1808 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
1809 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
1810 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
1811 }
1812
1813 pandecode_indent--;
1814 pandecode_log("},\n");
1815 }
1816
1817 static int
1818 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
1819 const struct pandecode_mapped_memory *mem,
1820 mali_ptr payload, int job_no)
1821 {
1822 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
1823
1824 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
1825
1826 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
1827 pandecode_indent++;
1828
1829 pandecode_log(".prefix = ");
1830 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1831
1832 pandecode_log(".vertex = ");
1833 pandecode_replay_vertex_only_bfr(&v->vertex);
1834
1835 pandecode_log(".postfix = ");
1836 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
1837
1838 pandecode_indent--;
1839 pandecode_log("};\n");
1840
1841 return sizeof(*v);
1842 }
1843
1844 static int
1845 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
1846 const struct pandecode_mapped_memory *mem,
1847 mali_ptr payload, int job_no)
1848 {
1849 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
1850
1851 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
1852
1853 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
1854 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
1855
1856 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
1857 pandecode_indent++;
1858
1859 pandecode_log(".prefix = ");
1860 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
1861
1862 pandecode_log(".tiler = ");
1863 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
1864
1865 pandecode_log(".postfix = ");
1866 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
1867
1868 pandecode_indent--;
1869 pandecode_log("};\n");
1870
1871 return sizeof(*t);
1872 }
1873
1874 static int
1875 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
1876 const struct pandecode_mapped_memory *mem,
1877 mali_ptr payload, int job_no)
1878 {
1879 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
1880
1881 char *a = pointer_as_memory_reference(payload);
1882 pandecode_msg("vt payload: %s\n", a);
1883 free(a);
1884
1885 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
1886
1887 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
1888
1889 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
1890 pandecode_indent++;
1891
1892 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
1893 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
1894
1895 pandecode_log(".prefix = ");
1896 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1897
1898 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
1899 pandecode_prop("draw_start = %d", v->draw_start);
1900
1901 #ifndef __LP64__
1902
1903 if (v->zero3) {
1904 pandecode_msg("Zero tripped\n");
1905 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
1906 }
1907
1908 #endif
1909
1910 if (v->zero5) {
1911 pandecode_msg("Zero tripped\n");
1912 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
1913 }
1914
1915 pandecode_log(".postfix = ");
1916 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
1917
1918 pandecode_indent--;
1919 pandecode_log("};\n");
1920
1921 return sizeof(*v);
1922 }
1923
1924 static int
1925 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
1926 mali_ptr payload, int job_no,
1927 bool is_bifrost)
1928 {
1929 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
1930
1931 bool fbd_dumped = false;
1932
1933 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
1934 /* Only SFBDs are understood, not MFBDs. We're speculating,
1935 * based on the versioning, kernel code, etc, that the
1936 * difference is between Single FrameBuffer Descriptor and
1937 * Multiple FrmaeBuffer Descriptor; the change apparently lines
1938 * up with multi-framebuffer support being added (T7xx onwards,
1939 * including Gxx). In any event, there's some field shuffling
1940 * that we haven't looked into yet. */
1941
1942 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
1943 fbd_dumped = true;
1944 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
1945 /* We don't know if Bifrost supports SFBD's at all, since the
1946 * driver never uses them. And the format is different from
1947 * Midgard anyways, due to the tiler heap and scratchpad being
1948 * moved out into separate structures, so it's not clear what a
1949 * Bifrost SFBD would even look like without getting an actual
1950 * trace, which appears impossible.
1951 */
1952
1953 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no);
1954 fbd_dumped = true;
1955 }
1956
1957 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
1958
1959 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no);
1960 pandecode_indent++;
1961
1962 /* See the comments by the macro definitions for mathematical context
1963 * on why this is so weird */
1964
1965 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
1966 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
1967
1968 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
1969 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
1970 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
1971
1972 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
1973 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
1974 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
1975
1976 /* If the FBD was just decoded, we can refer to it by pointer. If not,
1977 * we have to fallback on offsets. */
1978
1979 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
1980
1981 if (fbd_dumped)
1982 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
1983 else
1984 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
1985
1986 pandecode_indent--;
1987 pandecode_log("};\n");
1988
1989 return sizeof(*s);
1990 }
1991
1992 static int job_descriptor_number = 0;
1993
1994 int
1995 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
1996 {
1997 struct mali_job_descriptor_header *h;
1998
1999 int start_number = 0;
2000
2001 bool first = true;
2002 bool last_size;
2003
2004 do {
2005 struct pandecode_mapped_memory *mem =
2006 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2007
2008 void *payload;
2009
2010 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2011
2012 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2013 * high 32-bits of the 64-bit pointer are reused to store
2014 * something else.
2015 */
2016 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2017 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2018 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2019
2020 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2021 MALI_PAYLOAD_SIZE);
2022
2023 int job_no = job_descriptor_number++;
2024
2025 if (first)
2026 start_number = job_no;
2027
2028 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no);
2029 pandecode_indent++;
2030
2031 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
2032
2033 /* Save for next job fixing */
2034 last_size = h->job_descriptor_size;
2035
2036 if (h->job_descriptor_size)
2037 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2038
2039 if (h->exception_status)
2040 pandecode_prop("exception_status = %d", h->exception_status);
2041
2042 if (h->first_incomplete_task)
2043 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2044
2045 if (h->fault_pointer)
2046 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2047
2048 if (h->job_barrier)
2049 pandecode_prop("job_barrier = %d", h->job_barrier);
2050
2051 pandecode_prop("job_index = %d", h->job_index);
2052
2053 if (h->unknown_flags)
2054 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2055
2056 if (h->job_dependency_index_1)
2057 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2058
2059 if (h->job_dependency_index_2)
2060 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2061
2062 pandecode_indent--;
2063 pandecode_log("};\n");
2064
2065 /* Do not touch the field yet -- decode the payload first, and
2066 * don't touch that either. This is essential for the uploads
2067 * to occur in sequence and therefore be dynamically allocated
2068 * correctly. Do note the size, however, for that related
2069 * reason. */
2070
2071 switch (h->job_type) {
2072 case JOB_TYPE_SET_VALUE: {
2073 struct mali_payload_set_value *s = payload;
2074
2075 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no);
2076 pandecode_indent++;
2077 MEMORY_PROP(s, out);
2078 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2079 pandecode_indent--;
2080 pandecode_log("};\n");
2081
2082 break;
2083 }
2084
2085 case JOB_TYPE_TILER:
2086 case JOB_TYPE_VERTEX:
2087 case JOB_TYPE_COMPUTE:
2088 if (bifrost) {
2089 if (h->job_type == JOB_TYPE_TILER)
2090 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2091 else
2092 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2093 } else
2094 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2095
2096 break;
2097
2098 case JOB_TYPE_FRAGMENT:
2099 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2100 break;
2101
2102 default:
2103 break;
2104 }
2105
2106 /* Handle linkage */
2107
2108 if (!first) {
2109 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2110
2111 if (last_size)
2112 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2113 else
2114 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2115 }
2116
2117 first = false;
2118
2119 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2120
2121 return start_number;
2122 }