projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
panfrost: Replay more varying buffers
[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
33 #include "../pan_pretty_print.h"
34 #include "../midgard/disassemble.h"
35 int pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost);
36
37 #define MEMORY_PROP(obj, p) {\
38 char *a = pointer_as_memory_reference(obj->p); \
39 pandecode_prop("%s = %s", #p, a); \
40 free(a); \
41 }
42
43 #define MEMORY_COMMENT(obj, p) {\
44 char *a = pointer_as_memory_reference(obj->p); \
45 pandecode_msg("%s = %s\n", #p, a); \
46 free(a); \
47 }
48
49 #define DYN_MEMORY_PROP(obj, no, p) { \
50 if (obj->p) \
51 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
52 }
53
54 /* Semantic logging type.
55 *
56 * Raw: for raw messages to be printed as is.
57 * Message: for helpful information to be commented out in replays.
58 * Property: for properties of a struct
59 *
60 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
61 */
62
63 enum pandecode_log_type {
64 PANDECODE_RAW,
65 PANDECODE_MESSAGE,
66 PANDECODE_PROPERTY
67 };
68
69 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
70 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
71 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
72
73 unsigned pandecode_indent = 0;
74
75 static void
76 pandecode_make_indent(void)
77 {
78 for (unsigned i = 0; i < pandecode_indent; ++i)
79 printf(" ");
80 }
81
82 static void
83 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
84 {
85 va_list ap;
86
87 pandecode_make_indent();
88
89 if (type == PANDECODE_MESSAGE)
90 printf("// ");
91 else if (type == PANDECODE_PROPERTY)
92 printf(".");
93
94 va_start(ap, format);
95 vprintf(format, ap);
96 va_end(ap);
97
98 if (type == PANDECODE_PROPERTY)
99 printf(",\n");
100 }
101
102 static void
103 pandecode_log_cont(const char *format, ...)
104 {
105 va_list ap;
106
107 va_start(ap, format);
108 vprintf(format, ap);
109 va_end(ap);
110 }
111
112 struct pandecode_flag_info {
113 u64 flag;
114 const char *name;
115 };
116
117 static void
118 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
119 u64 flags)
120 {
121 bool decodable_flags_found = false;
122
123 for (int i = 0; flag_info[i].name; i++) {
124 if ((flags & flag_info[i].flag) != flag_info[i].flag)
125 continue;
126
127 if (!decodable_flags_found) {
128 decodable_flags_found = true;
129 } else {
130 pandecode_log_cont(" | ");
131 }
132
133 pandecode_log_cont("%s", flag_info[i].name);
134
135 flags &= ~flag_info[i].flag;
136 }
137
138 if (decodable_flags_found) {
139 if (flags)
140 pandecode_log_cont(" | 0x%" PRIx64, flags);
141 } else {
142 pandecode_log_cont("0x%" PRIx64, flags);
143 }
144 }
145
146 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
147 static const struct pandecode_flag_info gl_enable_flag_info[] = {
148 FLAG_INFO(CULL_FACE_FRONT),
149 FLAG_INFO(CULL_FACE_BACK),
150 FLAG_INFO(OCCLUSION_QUERY),
151 FLAG_INFO(OCCLUSION_PRECISE),
152 {}
153 };
154 #undef FLAG_INFO
155
156 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
157 static const struct pandecode_flag_info clear_flag_info[] = {
158 FLAG_INFO(FAST),
159 FLAG_INFO(SLOW),
160 FLAG_INFO(SLOW_STENCIL),
161 {}
162 };
163 #undef FLAG_INFO
164
165 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
166 static const struct pandecode_flag_info mask_flag_info[] = {
167 FLAG_INFO(R),
168 FLAG_INFO(G),
169 FLAG_INFO(B),
170 FLAG_INFO(A),
171 {}
172 };
173 #undef FLAG_INFO
174
175 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
176 static const struct pandecode_flag_info u3_flag_info[] = {
177 FLAG_INFO(HAS_MSAA),
178 FLAG_INFO(CAN_DISCARD),
179 FLAG_INFO(HAS_BLEND_SHADER),
180 FLAG_INFO(DEPTH_TEST),
181 {}
182 };
183
184 static const struct pandecode_flag_info u4_flag_info[] = {
185 FLAG_INFO(NO_MSAA),
186 FLAG_INFO(NO_DITHER),
187 FLAG_INFO(DEPTH_RANGE_A),
188 FLAG_INFO(DEPTH_RANGE_B),
189 FLAG_INFO(STENCIL_TEST),
190 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
191 {}
192 };
193 #undef FLAG_INFO
194
195 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
196 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
197 FLAG_INFO(MSAA_A),
198 FLAG_INFO(MSAA_B),
199 FLAG_INFO(MSAA_8),
200 {}
201 };
202 #undef FLAG_INFO
203
204 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
205 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
206 FLAG_INFO(AFBC),
207 FLAG_INFO(MSAA),
208 {}
209 };
210 #undef FLAG_INFO
211
212 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
213 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
214 FLAG_INFO(PRESENT),
215 FLAG_INFO(AFBC),
216 FLAG_INFO(ZS),
217 {}
218 };
219 #undef FLAG_INFO
220
221 extern char *replace_fragment;
222 extern char *replace_vertex;
223
224 static char *
225 pandecode_job_type_name(enum mali_job_type type)
226 {
227 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
228
229 switch (type) {
230 DEFINE_CASE(NULL);
231 DEFINE_CASE(SET_VALUE);
232 DEFINE_CASE(CACHE_FLUSH);
233 DEFINE_CASE(COMPUTE);
234 DEFINE_CASE(VERTEX);
235 DEFINE_CASE(TILER);
236 DEFINE_CASE(FUSED);
237 DEFINE_CASE(FRAGMENT);
238
239 case JOB_NOT_STARTED:
240 return "NOT_STARTED";
241
242 default:
243 pandecode_log("Warning! Unknown job type %x\n", type);
244 return "!?!?!?";
245 }
246
247 #undef DEFINE_CASE
248 }
249
250 static char *
251 pandecode_draw_mode_name(enum mali_draw_mode mode)
252 {
253 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
254
255 switch (mode) {
256 DEFINE_CASE(DRAW_NONE);
257 DEFINE_CASE(POINTS);
258 DEFINE_CASE(LINES);
259 DEFINE_CASE(TRIANGLES);
260 DEFINE_CASE(TRIANGLE_STRIP);
261 DEFINE_CASE(TRIANGLE_FAN);
262 DEFINE_CASE(LINE_STRIP);
263 DEFINE_CASE(LINE_LOOP);
264 DEFINE_CASE(POLYGON);
265 DEFINE_CASE(QUADS);
266 DEFINE_CASE(QUAD_STRIP);
267
268 default:
269 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
270 }
271
272 #undef DEFINE_CASE
273 }
274
275 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
276 static char *
277 pandecode_func_name(enum mali_func mode)
278 {
279 switch (mode) {
280 DEFINE_CASE(NEVER);
281 DEFINE_CASE(LESS);
282 DEFINE_CASE(EQUAL);
283 DEFINE_CASE(LEQUAL);
284 DEFINE_CASE(GREATER);
285 DEFINE_CASE(NOTEQUAL);
286 DEFINE_CASE(GEQUAL);
287 DEFINE_CASE(ALWAYS);
288
289 default:
290 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
291 }
292 }
293 #undef DEFINE_CASE
294
295 /* Why is this duplicated? Who knows... */
296 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
297 static char *
298 pandecode_alt_func_name(enum mali_alt_func mode)
299 {
300 switch (mode) {
301 DEFINE_CASE(NEVER);
302 DEFINE_CASE(LESS);
303 DEFINE_CASE(EQUAL);
304 DEFINE_CASE(LEQUAL);
305 DEFINE_CASE(GREATER);
306 DEFINE_CASE(NOTEQUAL);
307 DEFINE_CASE(GEQUAL);
308 DEFINE_CASE(ALWAYS);
309
310 default:
311 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
312 }
313 }
314 #undef DEFINE_CASE
315
316 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
317 static char *
318 pandecode_stencil_op_name(enum mali_stencil_op op)
319 {
320 switch (op) {
321 DEFINE_CASE(KEEP);
322 DEFINE_CASE(REPLACE);
323 DEFINE_CASE(ZERO);
324 DEFINE_CASE(INVERT);
325 DEFINE_CASE(INCR_WRAP);
326 DEFINE_CASE(DECR_WRAP);
327 DEFINE_CASE(INCR);
328 DEFINE_CASE(DECR);
329
330 default:
331 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
332 }
333 }
334
335 #undef DEFINE_CASE
336
337 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
338 static char *pandecode_attr_mode_name(enum mali_attr_mode mode)
339 {
340 switch(mode) {
341 DEFINE_CASE(UNUSED);
342 DEFINE_CASE(LINEAR);
343 DEFINE_CASE(POT_DIVIDE);
344 DEFINE_CASE(MODULO);
345 DEFINE_CASE(NPOT_DIVIDE);
346 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
347 }
348 }
349
350 #undef DEFINE_CASE
351
352 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
353 static char *
354 pandecode_channel_name(enum mali_channel channel)
355 {
356 switch (channel) {
357 DEFINE_CASE(RED);
358 DEFINE_CASE(GREEN);
359 DEFINE_CASE(BLUE);
360 DEFINE_CASE(ALPHA);
361 DEFINE_CASE(ZERO);
362 DEFINE_CASE(ONE);
363 DEFINE_CASE(RESERVED_0);
364 DEFINE_CASE(RESERVED_1);
365
366 default:
367 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
368 }
369 }
370 #undef DEFINE_CASE
371
372 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
373 static char *
374 pandecode_wrap_mode_name(enum mali_wrap_mode op)
375 {
376 switch (op) {
377 DEFINE_CASE(REPEAT);
378 DEFINE_CASE(CLAMP_TO_EDGE);
379 DEFINE_CASE(CLAMP_TO_BORDER);
380 DEFINE_CASE(MIRRORED_REPEAT);
381
382 default:
383 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
384 }
385 }
386 #undef DEFINE_CASE
387
388 static inline char *
389 pandecode_decode_fbd_type(enum mali_fbd_type type)
390 {
391 if (type == MALI_SFBD) return "SFBD";
392 else if (type == MALI_MFBD) return "MFBD";
393 else return "WATFBD /* XXX */";
394 }
395
396 static void
397 pandecode_replay_sfbd(uint64_t gpu_va, int job_no)
398 {
399 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
400 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
401
402 pandecode_log("struct mali_single_framebuffer framebuffer_%d = {\n", job_no);
403 pandecode_indent++;
404
405 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
406 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
407
408 pandecode_log(".format = ");
409 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
410 pandecode_log_cont(",\n");
411
412 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
413 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
414
415 MEMORY_PROP(s, framebuffer);
416 pandecode_prop("stride = %d", s->stride);
417
418 /* Earlier in the actual commandstream -- right before width -- but we
419 * delay to flow nicer */
420
421 pandecode_log(".clear_flags = ");
422 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
423 pandecode_log_cont(",\n");
424
425 if (s->depth_buffer | s->depth_buffer_enable) {
426 MEMORY_PROP(s, depth_buffer);
427 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
428 }
429
430 if (s->stencil_buffer | s->stencil_buffer_enable) {
431 MEMORY_PROP(s, stencil_buffer);
432 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
433 }
434
435 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
436 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
437 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
438 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
439 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
440 }
441
442 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
443 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
444 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
445 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
446 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
447 }
448
449 if (s->clear_stencil) {
450 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
451 }
452
453 MEMORY_PROP(s, unknown_address_0);
454 MEMORY_PROP(s, unknown_address_1);
455 MEMORY_PROP(s, unknown_address_2);
456
457 pandecode_prop("resolution_check = 0x%" PRIx32, s->resolution_check);
458 pandecode_prop("tiler_flags = 0x%" PRIx32, s->tiler_flags);
459
460 MEMORY_PROP(s, tiler_heap_free);
461 MEMORY_PROP(s, tiler_heap_end);
462
463 pandecode_indent--;
464 pandecode_log("};\n");
465
466 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
467 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
468 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
469 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
470
471 printf(".zero3 = {");
472
473 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
474 printf("%X, ", s->zero3[i]);
475
476 printf("},\n");
477
478 printf(".zero6 = {");
479
480 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
481 printf("%X, ", s->zero6[i]);
482
483 printf("},\n");
484 }
485
486 static void
487 pandecode_replay_swizzle(unsigned swizzle)
488 {
489 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
490 pandecode_channel_name((swizzle >> 0) & 0x7),
491 pandecode_channel_name((swizzle >> 3) & 0x7),
492 pandecode_channel_name((swizzle >> 6) & 0x7),
493 pandecode_channel_name((swizzle >> 9) & 0x7));
494 }
495
496 static void
497 pandecode_rt_format(struct mali_rt_format format)
498 {
499 pandecode_log(".format = {\n");
500 pandecode_indent++;
501
502 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
503 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
504
505 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
506 MALI_NEGATIVE(format.nr_channels));
507
508 pandecode_log(".flags = ");
509 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
510 pandecode_log_cont(",\n");
511
512 pandecode_replay_swizzle(format.swizzle);
513
514 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
515
516 pandecode_indent--;
517 pandecode_log("},\n");
518 }
519
520 static void
521 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no)
522 {
523 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
524 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
525
526 if (fb->sample_locations) {
527 /* The blob stores all possible sample locations in a single buffer
528 * allocated on startup, and just switches the pointer when switching
529 * MSAA state. For now, we just put the data into the cmdstream, but we
530 * should do something like what the blob does with a real driver.
531 *
532 * There seem to be 32 slots for sample locations, followed by another
533 * 16. The second 16 is just the center location followed by 15 zeros
534 * in all the cases I've identified (maybe shader vs. depth/color
535 * samples?).
536 */
537
538 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
539
540 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
541
542 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
543 pandecode_indent++;
544
545 for (int i = 0; i < 32 + 16; i++) {
546 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
547 }
548
549 pandecode_indent--;
550 pandecode_log("};\n");
551 }
552
553 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no);
554 pandecode_indent++;
555
556 pandecode_prop("unk0 = 0x%x", fb->unk0);
557
558 if (fb->sample_locations)
559 pandecode_prop("sample_locations = sample_locations_%d", job_no);
560
561 /* Assume that unknown1 and tiler_meta were emitted in the last job for
562 * now */
563 /*pandecode_prop("unknown1 = unknown1_%d_p", job_no - 1);
564 pandecode_prop("tiler_meta = tiler_meta_%d_p", job_no - 1);*/
565 MEMORY_PROP(fb, unknown1);
566 MEMORY_PROP(fb, tiler_meta);
567
568 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
569 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
570 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
571 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
572
573 pandecode_prop("unk1 = 0x%x", fb->unk1);
574 pandecode_prop("unk2 = 0x%x", fb->unk2);
575 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
576 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
577
578 pandecode_prop("unk3 = 0x%x", fb->unk3);
579 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
580 pandecode_prop("clear_depth = %f", fb->clear_depth);
581
582 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
583 MEMORY_PROP(fb, scratchpad);
584 MEMORY_PROP(fb, tiler_scratch_start);
585 MEMORY_PROP(fb, tiler_scratch_middle);
586 MEMORY_PROP(fb, tiler_heap_start);
587 MEMORY_PROP(fb, tiler_heap_end);
588
589 if (fb->zero3 || fb->zero4 || fb->zero9 || fb->zero10 || fb->zero11 || fb->zero12) {
590 pandecode_msg("framebuffer zeros tripped\n");
591 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
592 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
593 pandecode_prop("zero9 = 0x%" PRIx64, fb->zero9);
594 pandecode_prop("zero10 = 0x%" PRIx64, fb->zero10);
595 pandecode_prop("zero11 = 0x%" PRIx64, fb->zero11);
596 pandecode_prop("zero12 = 0x%" PRIx64, fb->zero12);
597 }
598
599 pandecode_indent--;
600 pandecode_log("};\n");
601
602 gpu_va += sizeof(struct bifrost_framebuffer);
603
604 if (fb->unk3 & MALI_MFBD_EXTRA) {
605 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
606 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
607
608 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no);
609 pandecode_indent++;
610
611 MEMORY_PROP(fbx, checksum);
612
613 if (fbx->checksum_stride)
614 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
615
616 pandecode_log(".flags = ");
617 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
618 pandecode_log_cont(",\n");
619
620 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
621 pandecode_log(".ds_afbc = {\n");
622 pandecode_indent++;
623
624 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
625 pandecode_prop("depth_stencil_afbc_stride = %d",
626 fbx->ds_afbc.depth_stencil_afbc_stride);
627 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
628
629 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
630 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
631 pandecode_prop("zero1 = 0x%" PRIx32,
632 fbx->ds_afbc.zero1);
633 pandecode_prop("padding = 0x%" PRIx64,
634 fbx->ds_afbc.padding);
635 }
636
637 pandecode_indent--;
638 pandecode_log("},\n");
639 } else {
640 pandecode_log(".ds_linear = {\n");
641 pandecode_indent++;
642
643 if (fbx->ds_linear.depth) {
644 MEMORY_PROP((&fbx->ds_linear), depth);
645 pandecode_prop("depth_stride = %d",
646 fbx->ds_linear.depth_stride);
647 }
648
649 if (fbx->ds_linear.stencil) {
650 MEMORY_PROP((&fbx->ds_linear), stencil);
651 pandecode_prop("stencil_stride = %d",
652 fbx->ds_linear.stencil_stride);
653 }
654
655 if (fbx->ds_linear.depth_stride_zero ||
656 fbx->ds_linear.stencil_stride_zero ||
657 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
658 pandecode_msg("Depth/stencil zeros tripped\n");
659 pandecode_prop("depth_stride_zero = 0x%x",
660 fbx->ds_linear.depth_stride_zero);
661 pandecode_prop("stencil_stride_zero = 0x%x",
662 fbx->ds_linear.stencil_stride_zero);
663 pandecode_prop("zero1 = 0x%" PRIx32,
664 fbx->ds_linear.zero1);
665 pandecode_prop("zero2 = 0x%" PRIx32,
666 fbx->ds_linear.zero2);
667 }
668
669 pandecode_indent--;
670 pandecode_log("},\n");
671 }
672
673 if (fbx->zero3 || fbx->zero4) {
674 pandecode_msg("fb_extra zeros tripped\n");
675 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
676 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
677 }
678
679 pandecode_indent--;
680 pandecode_log("};\n");
681
682 gpu_va += sizeof(struct bifrost_fb_extra);
683 }
684
685 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no);
686 pandecode_indent++;
687
688 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
689 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
690 mem = pandecode_find_mapped_gpu_mem_containing(rt_va);
691 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
692
693 pandecode_log("{\n");
694 pandecode_indent++;
695
696 pandecode_rt_format(rt->format);
697
698 /* TODO: How the actual heck does AFBC enabling work here? */
699 if (0) {
700 pandecode_log(".afbc = {\n");
701 pandecode_indent++;
702
703 char *a = pointer_as_memory_reference(rt->afbc.metadata);
704 pandecode_prop("metadata = %s", a);
705 free(a);
706
707 pandecode_prop("stride = %d", rt->afbc.stride);
708 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
709
710 pandecode_indent--;
711 pandecode_log("},\n");
712 } else {
713 pandecode_log(".chunknown = {\n");
714 pandecode_indent++;
715
716 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
717
718 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
719 pandecode_prop("pointer = %s", a);
720 free(a);
721
722 pandecode_indent--;
723 pandecode_log("},\n");
724 }
725
726 MEMORY_PROP(rt, framebuffer);
727 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
728
729 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
730 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
731 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
732 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
733 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
734 }
735
736 if (rt->zero1 || rt->zero2 || rt->zero3) {
737 pandecode_msg("render target zeros tripped\n");
738 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
739 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
740 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
741 }
742
743 pandecode_indent--;
744 pandecode_log("},\n");
745 }
746
747 pandecode_indent--;
748 pandecode_log("};\n");
749 }
750
751 static void
752 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
753 mali_ptr addr, int job_no, char *suffix,
754 int count, bool varying)
755 {
756 char *prefix = varying ? "varyings" : "attributes";
757
758 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
759
760 char base[128];
761 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
762
763 for (int i = 0; i < count; ++i) {
764 enum mali_attr_mode mode = attr[i].elements & 7;
765
766 if (mode == MALI_ATTR_UNUSED)
767 continue;
768
769 mali_ptr raw_elements = attr[i].elements & ~7;
770
771 /* TODO: Do we maybe want to dump the attribute values
772 * themselves given the specified format? Or is that too hard?
773 * */
774
775 char *a = pointer_as_memory_reference(raw_elements);
776 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
777 free(a);
778 }
779
780 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
781 pandecode_indent++;
782
783 for (int i = 0; i < count; ++i) {
784 pandecode_log("{\n");
785 pandecode_indent++;
786
787 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(attr[i].elements & 7));
788 pandecode_prop("shift = %d", attr[i].shift);
789 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
790 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
791 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
792 pandecode_indent--;
793 pandecode_log("}, \n");
794
795 if ((attr[i].elements & 7) == MALI_ATTR_NPOT_DIVIDE) {
796 i++;
797 pandecode_log("{\n");
798 pandecode_indent++;
799 pandecode_prop("unk = 0x%x", attr[i].unk);
800 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
801 if (attr[i].zero != 0)
802 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
803 pandecode_prop("divisor = %d", attr[i].divisor);
804 pandecode_indent--;
805 pandecode_log("}, \n");
806 }
807
808 }
809
810 pandecode_indent--;
811 pandecode_log("};\n");
812 }
813
814 static mali_ptr
815 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
816 {
817 /* TODO: Decode flags */
818 mali_ptr shader_ptr = ptr & ~15;
819
820 char *a = pointer_as_memory_reference(shader_ptr);
821 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
822 free(a);
823
824 return shader_ptr;
825 }
826
827 static void
828 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
829 {
830 const char *func = pandecode_func_name(stencil->func);
831 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
832 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
833 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
834
835 if (stencil->zero)
836 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
837
838 pandecode_log(".stencil_%s = {\n", name);
839 pandecode_indent++;
840 pandecode_prop("ref = %d", stencil->ref);
841 pandecode_prop("mask = 0x%02X", stencil->mask);
842 pandecode_prop("func = %s", func);
843 pandecode_prop("sfail = %s", sfail);
844 pandecode_prop("dpfail = %s", dpfail);
845 pandecode_prop("dppass = %s", dppass);
846 pandecode_indent--;
847 pandecode_log("},\n");
848 }
849
850 static void
851 pandecode_replay_blend_equation(const struct mali_blend_equation *blend, const char *suffix)
852 {
853 if (blend->zero1)
854 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
855
856 pandecode_log(".blend_equation%s = {\n", suffix);
857 pandecode_indent++;
858
859 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
860 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
861
862 pandecode_log(".color_mask = ");
863 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
864 pandecode_log_cont(",\n");
865
866 pandecode_indent--;
867 pandecode_log("},\n");
868 }
869
870 static int
871 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
872 {
873 char base[128];
874 char *prefix = varying ? "varying" : "attribute";
875 unsigned max_index = 0;
876 snprintf(base, sizeof(base), "%s_meta", prefix);
877
878 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
879 pandecode_indent++;
880
881 struct mali_attr_meta *attr_meta;
882 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
883
884 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
885
886 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
887 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
888 sizeof(*attr_mem));
889
890 pandecode_log("{\n");
891 pandecode_indent++;
892 pandecode_prop("index = %d", attr_meta->index);
893
894 if (attr_meta->index > max_index)
895 max_index = attr_meta->index;
896 pandecode_replay_swizzle(attr_meta->swizzle);
897 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
898
899 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
900 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
901 pandecode_prop("src_offset = 0x%" PRIx64, (u64) attr_meta->src_offset);
902 pandecode_indent--;
903 pandecode_log("},\n");
904
905 }
906
907 pandecode_indent--;
908 pandecode_log("};\n");
909
910 return max_index;
911 }
912
913 static void
914 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
915 {
916 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
917
918 if (imem) {
919 /* Indices are literally just a u32 array :) */
920
921 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
922
923 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
924 pandecode_indent++;
925
926 for (unsigned i = 0; i < (index_count + 1); i += 3)
927 pandecode_log("%d, %d, %d,\n",
928 indices[i],
929 indices[i + 1],
930 indices[i + 2]);
931
932 pandecode_indent--;
933 pandecode_log("};\n");
934 }
935 }
936
937 /* return bits [lo, hi) of word */
938 static u32
939 bits(u32 word, u32 lo, u32 hi)
940 {
941 if (hi - lo >= 32)
942 return word; // avoid undefined behavior with the shift
943
944 return (word >> lo) & ((1 << (hi - lo)) - 1);
945 }
946
947 static void
948 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
949 {
950 pandecode_log_cont("{\n");
951 pandecode_indent++;
952
953 pandecode_prop("invocation_count = %" PRIx32, p->invocation_count);
954 pandecode_prop("size_y_shift = %d", p->size_y_shift);
955 pandecode_prop("size_z_shift = %d", p->size_z_shift);
956 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
957 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
958 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
959 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
960
961 /* Decode invocation_count. See the comment before the definition of
962 * invocation_count for an explanation.
963 */
964 pandecode_msg("size: (%d, %d, %d)\n",
965 bits(p->invocation_count, 0, p->size_y_shift) + 1,
966 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
967 bits(p->invocation_count, p->size_z_shift,
968 p->workgroups_x_shift) + 1);
969 pandecode_msg("workgroups: (%d, %d, %d)\n",
970 bits(p->invocation_count, p->workgroups_x_shift,
971 p->workgroups_y_shift) + 1,
972 bits(p->invocation_count, p->workgroups_y_shift,
973 p->workgroups_z_shift) + 1,
974 bits(p->invocation_count, p->workgroups_z_shift,
975 32) + 1);
976
977 /* TODO: Decode */
978 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
979 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
980
981 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
982
983 /* Index count only exists for tiler jobs anyway */
984
985 if (p->index_count)
986 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
987
988 DYN_MEMORY_PROP(p, job_no, indices);
989
990 if (p->zero1) {
991 pandecode_msg("Zero tripped\n");
992 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
993 }
994
995 pandecode_indent--;
996 pandecode_log("},\n");
997 }
998
999 static void
1000 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1001 {
1002 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1003
1004 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1005
1006 for (int i = 0; i < ubufs_count; i++) {
1007 mali_ptr ptr = ubufs[i].ptr << 2;
1008 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1009 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1010 char name[50];
1011 snprintf(name, sizeof(name), "ubuf_%d", i);
1012 /* The blob uses ubuf 0 to upload internal stuff and
1013 * uniforms that won't fit/are accessed indirectly, so
1014 * it puts it in the batchbuffer.
1015 */
1016 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1017 pandecode_indent++;
1018
1019 for (int j = 0; j <= ubufs[i].size; j++) {
1020 for (int k = 0; k < 4; k++) {
1021 if (k == 0)
1022 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1023 else
1024 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1025
1026 }
1027
1028 pandecode_log_cont("\n");
1029 }
1030
1031 pandecode_indent--;
1032 pandecode_log("};\n");
1033 }
1034
1035 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1036 job_no);
1037 pandecode_indent++;
1038
1039 for (int i = 0; i < ubufs_count; i++) {
1040 pandecode_log("{\n");
1041 pandecode_indent++;
1042 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1043 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1044 pandecode_indent--;
1045 pandecode_log("},\n");
1046 }
1047
1048 pandecode_indent--;
1049 pandecode_log("};\n");
1050 }
1051
1052 static void
1053 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1054 {
1055
1056 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1057
1058 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1059
1060 if (scratchpad->zero)
1061 pandecode_msg("XXX scratchpad zero tripped");
1062
1063 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no, suffix);
1064 pandecode_indent++;
1065
1066 pandecode_prop("flags = 0x%x", scratchpad->flags);
1067 MEMORY_PROP(scratchpad, gpu_scratchpad);
1068
1069 pandecode_indent--;
1070 pandecode_log("};\n");
1071 }
1072
1073 static void
1074 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1075 bool is_bifrost)
1076 {
1077 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1078 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1079
1080 /* Compute maximum possible size */
1081 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1082
1083 /* TODO: When Bifrost is upstreamed, disassemble that too */
1084 if (is_bifrost) {
1085 pandecode_msg("Bifrost disassembler not yet upstreamed");
1086 return;
1087 }
1088
1089 /* Print some boilerplate to clearly denote the assembly (which doesn't
1090 * obey indentation rules), and actually do the disassembly! */
1091
1092 printf("\n\n");
1093 disassemble_midgard(code, sz);
1094 printf("\n\n");
1095 }
1096
1097 static void
1098 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1099 int job_no, enum mali_job_type job_type,
1100 char *suffix, bool is_bifrost)
1101 {
1102 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1103 struct pandecode_mapped_memory *attr_mem;
1104
1105 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1106 * are the only things actually needed from the FBD, vertex/tiler jobs
1107 * no longer reference the FBD -- instead, this field points to some
1108 * info about the scratchpad.
1109 */
1110 if (is_bifrost)
1111 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1112 else if (p->framebuffer & MALI_MFBD)
1113 pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no);
1114 else
1115 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1116
1117 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1118 int texture_count = 0, sampler_count = 0;
1119
1120 if (shader_meta_ptr) {
1121 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1122 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1123
1124 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no, suffix);
1125 pandecode_indent++;
1126
1127 /* Save for dumps */
1128 attribute_count = s->attribute_count;
1129 varying_count = s->varying_count;
1130 texture_count = s->texture_count;
1131 sampler_count = s->sampler_count;
1132
1133 if (is_bifrost) {
1134 uniform_count = s->bifrost2.uniform_count;
1135 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1136 } else {
1137 uniform_count = s->midgard1.uniform_count;
1138 /* TODO figure this out */
1139 uniform_buffer_count = 1;
1140 }
1141
1142 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1143
1144 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1145 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1146 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1147 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1148
1149 if (is_bifrost) {
1150 pandecode_log(".bifrost1 = {\n");
1151 pandecode_indent++;
1152
1153 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1154 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1155
1156 pandecode_indent--;
1157 pandecode_log("},\n");
1158 } else {
1159 pandecode_log(".midgard1 = {\n");
1160 pandecode_indent++;
1161
1162 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1163 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1164 pandecode_prop("unknown1 = %s0x%" PRIx32,
1165 s->midgard1.unknown1 & MALI_NO_ALPHA_TO_COVERAGE ? "MALI_NO_ALPHA_TO_COVERAGE | " : "",
1166 s->midgard1.unknown1 & ~MALI_NO_ALPHA_TO_COVERAGE);
1167 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1168
1169 pandecode_indent--;
1170 pandecode_log("},\n");
1171 }
1172
1173 if (s->depth_units || s->depth_factor) {
1174 if (is_bifrost)
1175 pandecode_prop("depth_units = %f", s->depth_units);
1176 else
1177 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f);
1178
1179 pandecode_prop("depth_factor = %f", s->depth_factor);
1180 }
1181
1182 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1183 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1184
1185 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1186 invert_alpha_coverage ? "~" : "",
1187 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1188
1189 pandecode_log(".unknown2_3 = ");
1190
1191 int unknown2_3 = s->unknown2_3;
1192 int unknown2_4 = s->unknown2_4;
1193
1194 /* We're not quite sure what these flags mean without the depth test, if anything */
1195
1196 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1197 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1198 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1199
1200 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1201 }
1202
1203 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1204 pandecode_log_cont(",\n");
1205
1206 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1207 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1208
1209 pandecode_log(".unknown2_4 = ");
1210 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1211 pandecode_log_cont(",\n");
1212
1213 pandecode_replay_stencil("front", &s->stencil_front);
1214 pandecode_replay_stencil("back", &s->stencil_back);
1215
1216 if (is_bifrost) {
1217 pandecode_log(".bifrost2 = {\n");
1218 pandecode_indent++;
1219
1220 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1221 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1222 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1223 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1224
1225 pandecode_indent--;
1226 pandecode_log("},\n");
1227 } else {
1228 pandecode_log(".midgard2 = {\n");
1229 pandecode_indent++;
1230
1231 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1232 pandecode_indent--;
1233 pandecode_log("},\n");
1234 }
1235
1236 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1237
1238 bool blend_shader = false;
1239
1240 if (!is_bifrost) {
1241 if (s->unknown2_3 & MALI_HAS_BLEND_SHADER) {
1242 blend_shader = true;
1243 pandecode_replay_shader_address("blend_shader", s->blend_shader);
1244 } else {
1245 pandecode_replay_blend_equation(&s->blend_equation, "");
1246 }
1247 }
1248
1249 pandecode_indent--;
1250 pandecode_log("};\n");
1251
1252 /* MRT blend fields are used whenever MFBD is used */
1253
1254 if (job_type == JOB_TYPE_TILER) {
1255 pandecode_log("struct mali_blend_meta blend_meta_%d[] = {\n",
1256 job_no);
1257 pandecode_indent++;
1258
1259 int i;
1260
1261 for (i = 0; i < 4; i++) {
1262 const struct mali_blend_meta *b = &s->blend_meta[i];
1263 pandecode_log("{\n");
1264 pandecode_indent++;
1265
1266 #ifndef BIFROST
1267 pandecode_prop("unk1 = 0x%" PRIx64, b->unk1);
1268 pandecode_replay_blend_equation(&b->blend_equation_1, "_1");
1269 pandecode_replay_blend_equation(&b->blend_equation_2, "_2");
1270
1271 if (b->zero2) {
1272 pandecode_msg("blend zero tripped\n");
1273 pandecode_prop("zero2 = 0x%x", b->zero2);
1274 }
1275
1276 #else
1277
1278 pandecode_prop("unk1 = 0x%" PRIx32, b->unk1);
1279 /* TODO figure out blend shader enable bit */
1280 pandecode_replay_blend_equation(&b->blend_equation);
1281 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1282 pandecode_prop("index = 0x%" PRIx16, b->index);
1283 pandecode_prop("unk3 = 0x%" PRIx32, b->unk3);
1284 #endif
1285
1286 pandecode_indent--;
1287 pandecode_log("},\n");
1288
1289 #ifdef BIFROST
1290 if (b->unk2 == 3)
1291 break;
1292 #else
1293 /* TODO: What's this supposed to be? */
1294 if (b->unk1 & 0x200)
1295 break;
1296 #endif
1297
1298 }
1299
1300 pandecode_indent--;
1301 pandecode_log("};\n");
1302
1303 /* This needs to be uploaded right after the
1304 * shader_meta since it's technically part of the same
1305 * (variable-size) structure.
1306 */
1307 }
1308
1309 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1310
1311 if (!is_bifrost && blend_shader)
1312 pandecode_shader_disassemble(s->blend_shader & ~0xF, job_no, job_type, false);
1313
1314 } else
1315 pandecode_msg("<no shader>\n");
1316
1317 if (p->viewport) {
1318 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1319 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1320
1321 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1322 pandecode_indent++;
1323
1324 pandecode_prop("clip_minx = %f", f->clip_minx);
1325 pandecode_prop("clip_miny = %f", f->clip_miny);
1326 pandecode_prop("clip_minz = %f", f->clip_minz);
1327 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1328 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1329 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1330
1331 /* Only the higher coordinates are MALI_POSITIVE scaled */
1332
1333 pandecode_prop("viewport0 = { %d, %d }",
1334 f->viewport0[0], f->viewport0[1]);
1335
1336 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1337 f->viewport1[0] + 1, f->viewport1[1] + 1);
1338
1339 pandecode_indent--;
1340 pandecode_log("};\n");
1341 }
1342
1343 if (p->attribute_meta) {
1344 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1345
1346 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1347 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1348 }
1349
1350 /* Varyings are encoded like attributes but not actually sent; we just
1351 * pass a zero buffer with the right stride/size set, (or whatever)
1352 * since the GPU will write to it itself */
1353
1354 if (p->varyings) {
1355 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1356
1357 /* Number of descriptors depends on whether there are
1358 * non-internal varyings */
1359
1360 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1361 }
1362
1363 if (p->varying_meta) {
1364 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1365 }
1366
1367 if (p->uniforms) {
1368 int rows = uniform_count, width = 4;
1369 size_t sz = rows * width * sizeof(float);
1370
1371 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1372 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1373 float *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1374
1375 pandecode_log("float uniforms_%d%s[] = {\n", job_no, suffix);
1376
1377 pandecode_indent++;
1378
1379 for (int row = 0; row < rows; row++) {
1380 for (int i = 0; i < width; i++)
1381 pandecode_log_cont("%ff, ", uniforms[i]);
1382
1383 pandecode_log_cont("\n");
1384
1385 uniforms += width;
1386 }
1387
1388 pandecode_indent--;
1389 pandecode_log("};\n");
1390 }
1391
1392 if (p->uniform_buffers) {
1393 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1394 }
1395
1396 if (p->texture_trampoline) {
1397 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1398
1399 if (mmem) {
1400 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1401
1402 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1403 pandecode_indent++;
1404
1405 for (int tex = 0; tex < texture_count; ++tex) {
1406 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1407 char *a = pointer_as_memory_reference(*u);
1408 pandecode_log("%s,\n", a);
1409 free(a);
1410 }
1411
1412 pandecode_indent--;
1413 pandecode_log("};\n");
1414
1415 /* Now, finally, descend down into the texture descriptor */
1416 for (int tex = 0; tex < texture_count; ++tex) {
1417 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1418 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1419
1420 if (tmem) {
1421 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1422
1423 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
1424 pandecode_indent++;
1425
1426 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1427 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1428 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1429
1430 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1431 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1432 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1433
1434 struct mali_texture_format f = t->format;
1435
1436 pandecode_log(".format = {\n");
1437 pandecode_indent++;
1438
1439 pandecode_replay_swizzle(f.swizzle);
1440 pandecode_prop("format = %s", pandecode_format_name(f.format));
1441
1442 pandecode_prop("usage1 = 0x%" PRIx32, f.usage1);
1443 pandecode_prop("is_not_cubemap = %" PRId32, f.is_not_cubemap);
1444 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1445
1446 pandecode_indent--;
1447 pandecode_log("},\n");
1448
1449 pandecode_replay_swizzle(t->swizzle);
1450
1451 if (t->swizzle_zero) {
1452 /* Shouldn't happen */
1453 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1454 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1455 }
1456
1457 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1458
1459 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1460 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1461 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1462
1463 pandecode_log(".swizzled_bitmaps = {\n");
1464 pandecode_indent++;
1465
1466 int bitmap_count = 1 + t->nr_mipmap_levels + t->unknown3A;
1467 int max_count = sizeof(t->swizzled_bitmaps) / sizeof(t->swizzled_bitmaps[0]);
1468
1469 if (bitmap_count > max_count) {
1470 pandecode_msg("XXX: bitmap count tripped");
1471 bitmap_count = max_count;
1472 }
1473
1474 for (int i = 0; i < bitmap_count; ++i) {
1475 char *a = pointer_as_memory_reference(t->swizzled_bitmaps[i]);
1476 pandecode_log("%s, \n", a);
1477 free(a);
1478 }
1479
1480 pandecode_indent--;
1481 pandecode_log("},\n");
1482
1483 pandecode_indent--;
1484 pandecode_log("};\n");
1485 }
1486 }
1487 }
1488 }
1489
1490 if (p->sampler_descriptor) {
1491 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1492
1493 if (smem) {
1494 struct mali_sampler_descriptor *s;
1495
1496 mali_ptr d = p->sampler_descriptor;
1497
1498 for (int i = 0; i < sampler_count; ++i) {
1499 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1500
1501 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1502 pandecode_indent++;
1503
1504 /* Only the lower two bits are understood right now; the rest we display as hex */
1505 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1506 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1507 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1508 s->filter_mode & ~3);
1509
1510 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1511 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1512
1513 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1514 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1515 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1516
1517 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1518
1519 if (s->zero || s->zero2) {
1520 pandecode_msg("Zero tripped\n");
1521 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1522 }
1523
1524 pandecode_prop("unknown2 = %d", s->unknown2);
1525
1526 pandecode_prop("border_color = { %f, %f, %f, %f }",
1527 s->border_color[0],
1528 s->border_color[1],
1529 s->border_color[2],
1530 s->border_color[3]);
1531
1532 pandecode_indent--;
1533 pandecode_log("};\n");
1534 }
1535 }
1536 }
1537 }
1538
1539 static void
1540 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1541 {
1542 pandecode_log_cont("{\n");
1543 pandecode_indent++;
1544
1545 MEMORY_PROP(p, position_varying);
1546 MEMORY_COMMENT(p, position_varying);
1547 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1548 MEMORY_COMMENT(p, uniform_buffers);
1549 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1550 MEMORY_COMMENT(p, texture_trampoline);
1551 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1552 MEMORY_COMMENT(p, sampler_descriptor);
1553 DYN_MEMORY_PROP(p, job_no, uniforms);
1554 MEMORY_COMMENT(p, uniforms);
1555 DYN_MEMORY_PROP(p, job_no, attributes);
1556 MEMORY_COMMENT(p, attributes);
1557 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1558 MEMORY_COMMENT(p, attribute_meta);
1559 DYN_MEMORY_PROP(p, job_no, varyings);
1560 MEMORY_COMMENT(p, varyings);
1561 DYN_MEMORY_PROP(p, job_no, varying_meta);
1562 MEMORY_COMMENT(p, varying_meta);
1563 DYN_MEMORY_PROP(p, job_no, viewport);
1564 MEMORY_COMMENT(p, viewport);
1565 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1566 MEMORY_COMMENT(p, occlusion_counter);
1567 MEMORY_COMMENT(p, framebuffer & ~1);
1568 pandecode_msg("%" PRIx64 "\n", p->viewport);
1569 pandecode_msg("%" PRIx64 "\n", p->framebuffer);
1570
1571 if (is_bifrost)
1572 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1573 else
1574 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1575
1576 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1577 pandecode_prop("flags = %d", p->flags);
1578
1579 pandecode_indent--;
1580 pandecode_log("},\n");
1581 }
1582
1583 static void
1584 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1585 {
1586 pandecode_log_cont("{\n");
1587 pandecode_indent++;
1588
1589 pandecode_prop("unk2 = 0x%x", v->unk2);
1590
1591 if (v->zero0 || v->zero1) {
1592 pandecode_msg("vertex only zero tripped");
1593 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1594 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1595 }
1596
1597 pandecode_indent--;
1598 pandecode_log("}\n");
1599 }
1600
1601 static void
1602 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1603 {
1604
1605 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1606 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1607
1608 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1609 pandecode_indent++;
1610
1611 if (h->zero) {
1612 pandecode_msg("tiler heap zero tripped\n");
1613 pandecode_prop("zero = 0x%x", h->zero);
1614 }
1615
1616 for (int i = 0; i < 12; i++) {
1617 if (h->zeros[i] != 0) {
1618 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1619 i, h->zeros[i]);
1620 }
1621 }
1622
1623 pandecode_prop("heap_size = 0x%x", h->heap_size);
1624 MEMORY_PROP(h, tiler_heap_start);
1625 MEMORY_PROP(h, tiler_heap_free);
1626
1627 /* this might point to the beginning of another buffer, when it's
1628 * really the end of the tiler heap buffer, so we have to be careful
1629 * here.
1630 */
1631 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1632 pandecode_prop("tiler_heap_end = %s + 1", a);
1633 free(a);
1634
1635 pandecode_indent--;
1636 pandecode_log("};\n");
1637 }
1638
1639 static void
1640 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1641 {
1642 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1643 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1644
1645 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1646
1647 pandecode_log("struct mali_tiler_meta tiler_meta_%d = {\n", job_no);
1648 pandecode_indent++;
1649
1650 if (t->zero0 || t->zero1) {
1651 pandecode_msg("tiler meta zero tripped");
1652 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1653 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1654 }
1655
1656 pandecode_prop("unk = 0x%x", t->unk);
1657 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1658 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
1659 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
1660
1661 for (int i = 0; i < 12; i++) {
1662 if (t->zeros[i] != 0) {
1663 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
1664 i, t->zeros[i]);
1665 }
1666 }
1667
1668 pandecode_indent--;
1669 pandecode_log("};\n");
1670 }
1671
1672 static void
1673 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
1674 {
1675 pandecode_log(".gl_enables = ");
1676
1677 if (job_type == JOB_TYPE_TILER) {
1678 pandecode_log_cont("MALI_FRONT_FACE(MALI_%s) | ",
1679 gl_enables & MALI_FRONT_FACE(MALI_CW) ? "CW" : "CCW");
1680
1681 gl_enables &= ~(MALI_FRONT_FACE(1));
1682 }
1683
1684 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
1685
1686 pandecode_log_cont(",\n");
1687 }
1688
1689 static void
1690 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
1691 {
1692 pandecode_log(".primitive_size = {\n");
1693 pandecode_indent++;
1694
1695 if (constant) {
1696 pandecode_prop("constant = %f", u.constant);
1697 } else {
1698 MEMORY_PROP((&u), pointer);
1699 }
1700
1701 pandecode_indent--;
1702 pandecode_log("},\n");
1703 }
1704
1705 static void
1706 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
1707 {
1708 pandecode_log_cont("{\n");
1709 pandecode_indent++;
1710
1711 /* TODO: gl_PointSize on Bifrost */
1712 pandecode_replay_primitive_size(t->primitive_size, true);
1713
1714 DYN_MEMORY_PROP(t, job_no, tiler_meta);
1715 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
1716
1717 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
1718 || t->zero6 || t->zero7 || t->zero8) {
1719 pandecode_msg("tiler only zero tripped");
1720 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1721 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
1722 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
1723 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
1724 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
1725 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
1726 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
1727 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
1728 }
1729
1730 pandecode_indent--;
1731 pandecode_log("},\n");
1732 }
1733
1734 static int
1735 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
1736 const struct pandecode_mapped_memory *mem,
1737 mali_ptr payload, int job_no)
1738 {
1739 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
1740
1741 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
1742
1743 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
1744 pandecode_indent++;
1745
1746 pandecode_log(".prefix = ");
1747 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1748
1749 pandecode_log(".vertex = ");
1750 pandecode_replay_vertex_only_bfr(&v->vertex);
1751
1752 pandecode_log(".postfix = ");
1753 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
1754
1755 pandecode_indent--;
1756 pandecode_log("};\n");
1757
1758 return sizeof(*v);
1759 }
1760
1761 static int
1762 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
1763 const struct pandecode_mapped_memory *mem,
1764 mali_ptr payload, int job_no)
1765 {
1766 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
1767
1768 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
1769
1770 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
1771 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
1772
1773 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
1774 pandecode_indent++;
1775
1776 pandecode_log(".prefix = ");
1777 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
1778
1779 pandecode_log(".tiler = ");
1780 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
1781
1782 pandecode_log(".postfix = ");
1783 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
1784
1785 pandecode_indent--;
1786 pandecode_log("};\n");
1787
1788 return sizeof(*t);
1789 }
1790
1791 static int
1792 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
1793 const struct pandecode_mapped_memory *mem,
1794 mali_ptr payload, int job_no)
1795 {
1796 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
1797
1798 char *a = pointer_as_memory_reference(payload);
1799 pandecode_msg("vt payload: %s\n", a);
1800 free(a);
1801
1802 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
1803
1804 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
1805
1806 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
1807 pandecode_indent++;
1808
1809 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
1810 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
1811
1812 pandecode_log(".prefix = ");
1813 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1814
1815 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
1816 pandecode_prop("draw_start = %d", v->draw_start);
1817
1818 #ifndef __LP64__
1819
1820 if (v->zero3) {
1821 pandecode_msg("Zero tripped\n");
1822 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
1823 }
1824
1825 #endif
1826
1827 if (v->zero5) {
1828 pandecode_msg("Zero tripped\n");
1829 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
1830 }
1831
1832 pandecode_log(".postfix = ");
1833 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
1834
1835 pandecode_indent--;
1836 pandecode_log("};\n");
1837
1838 return sizeof(*v);
1839 }
1840
1841 static int
1842 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
1843 mali_ptr payload, int job_no,
1844 bool is_bifrost)
1845 {
1846 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
1847
1848 bool fbd_dumped = false;
1849
1850 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
1851 /* Only SFBDs are understood, not MFBDs. We're speculating,
1852 * based on the versioning, kernel code, etc, that the
1853 * difference is between Single FrameBuffer Descriptor and
1854 * Multiple FrmaeBuffer Descriptor; the change apparently lines
1855 * up with multi-framebuffer support being added (T7xx onwards,
1856 * including Gxx). In any event, there's some field shuffling
1857 * that we haven't looked into yet. */
1858
1859 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
1860 fbd_dumped = true;
1861 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
1862 /* We don't know if Bifrost supports SFBD's at all, since the
1863 * driver never uses them. And the format is different from
1864 * Midgard anyways, due to the tiler heap and scratchpad being
1865 * moved out into separate structures, so it's not clear what a
1866 * Bifrost SFBD would even look like without getting an actual
1867 * trace, which appears impossible.
1868 */
1869
1870 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no);
1871 fbd_dumped = true;
1872 }
1873
1874 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
1875
1876 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no);
1877 pandecode_indent++;
1878
1879 /* See the comments by the macro definitions for mathematical context
1880 * on why this is so weird */
1881
1882 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
1883 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
1884
1885 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
1886 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
1887 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
1888
1889 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
1890 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
1891 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
1892
1893 /* If the FBD was just decoded, we can refer to it by pointer. If not,
1894 * we have to fallback on offsets. */
1895
1896 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
1897
1898 if (fbd_dumped)
1899 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
1900 else
1901 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
1902
1903 pandecode_indent--;
1904 pandecode_log("};\n");
1905
1906 return sizeof(*s);
1907 }
1908
1909 static int job_descriptor_number = 0;
1910
1911 int
1912 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
1913 {
1914 struct mali_job_descriptor_header *h;
1915
1916 int start_number = 0;
1917
1918 bool first = true;
1919 bool last_size;
1920
1921 do {
1922 struct pandecode_mapped_memory *mem =
1923 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
1924
1925 void *payload;
1926
1927 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
1928
1929 /* On Midgard, for 32-bit jobs except for fragment jobs, the
1930 * high 32-bits of the 64-bit pointer are reused to store
1931 * something else.
1932 */
1933 int offset = h->job_descriptor_size == MALI_JOB_32 &&
1934 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
1935 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
1936
1937 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
1938 MALI_PAYLOAD_SIZE);
1939
1940 int job_no = job_descriptor_number++;
1941
1942 if (first)
1943 start_number = job_no;
1944
1945 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no);
1946 pandecode_indent++;
1947
1948 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
1949
1950 /* Save for next job fixing */
1951 last_size = h->job_descriptor_size;
1952
1953 if (h->job_descriptor_size)
1954 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
1955
1956 if (h->exception_status)
1957 pandecode_prop("exception_status = %d", h->exception_status);
1958
1959 if (h->first_incomplete_task)
1960 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
1961
1962 if (h->fault_pointer)
1963 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
1964
1965 if (h->job_barrier)
1966 pandecode_prop("job_barrier = %d", h->job_barrier);
1967
1968 pandecode_prop("job_index = %d", h->job_index);
1969
1970 if (h->unknown_flags)
1971 pandecode_prop("unknown_flags = %d", h->unknown_flags);
1972
1973 if (h->job_dependency_index_1)
1974 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
1975
1976 if (h->job_dependency_index_2)
1977 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
1978
1979 pandecode_indent--;
1980 pandecode_log("};\n");
1981
1982 /* Do not touch the field yet -- decode the payload first, and
1983 * don't touch that either. This is essential for the uploads
1984 * to occur in sequence and therefore be dynamically allocated
1985 * correctly. Do note the size, however, for that related
1986 * reason. */
1987
1988 switch (h->job_type) {
1989 case JOB_TYPE_SET_VALUE: {
1990 struct mali_payload_set_value *s = payload;
1991
1992 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no);
1993 pandecode_indent++;
1994 MEMORY_PROP(s, out);
1995 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
1996 pandecode_indent--;
1997 pandecode_log("};\n");
1998
1999 break;
2000 }
2001
2002 case JOB_TYPE_TILER:
2003 case JOB_TYPE_VERTEX:
2004 case JOB_TYPE_COMPUTE:
2005 if (bifrost) {
2006 if (h->job_type == JOB_TYPE_TILER)
2007 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2008 else
2009 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2010 } else
2011 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2012
2013 break;
2014
2015 case JOB_TYPE_FRAGMENT:
2016 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2017 break;
2018
2019 default:
2020 break;
2021 }
2022
2023 /* Handle linkage */
2024
2025 if (!first) {
2026 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2027
2028 if (last_size)
2029 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2030 else
2031 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2032 }
2033
2034 first = false;
2035
2036 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2037
2038 return start_number;
2039 }