projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
panfrost: Rename tiler fields per tiler research
[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 "decode.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, tiler_polygon_list);
467 MEMORY_PROP(s, tiler_polygon_list_body);
468
469 pandecode_prop("tiler_resolution_check = 0x%" PRIx32, s->tiler_resolution_check);
470 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, s->tiler_hierarchy_mask);
471 pandecode_prop("tiler_flags = 0x%" PRIx16, s->tiler_flags);
472
473 MEMORY_PROP(s, tiler_heap_free);
474 MEMORY_PROP(s, tiler_heap_end);
475
476 pandecode_indent--;
477 pandecode_log("};\n");
478
479 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
480 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
481 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
482 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
483
484 printf(".zero3 = {");
485
486 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
487 printf("%X, ", s->zero3[i]);
488
489 printf("},\n");
490
491 printf(".zero6 = {");
492
493 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
494 printf("%X, ", s->zero6[i]);
495
496 printf("},\n");
497 }
498
499 static void
500 pandecode_replay_swizzle(unsigned swizzle)
501 {
502 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
503 pandecode_channel_name((swizzle >> 0) & 0x7),
504 pandecode_channel_name((swizzle >> 3) & 0x7),
505 pandecode_channel_name((swizzle >> 6) & 0x7),
506 pandecode_channel_name((swizzle >> 9) & 0x7));
507 }
508
509 static void
510 pandecode_rt_format(struct mali_rt_format format)
511 {
512 pandecode_log(".format = {\n");
513 pandecode_indent++;
514
515 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
516 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
517
518 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
519 MALI_NEGATIVE(format.nr_channels));
520
521 pandecode_log(".flags = ");
522 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
523 pandecode_log_cont(",\n");
524
525 pandecode_replay_swizzle(format.swizzle);
526
527 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
528
529 pandecode_indent--;
530 pandecode_log("},\n");
531 }
532
533 static void
534 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
535 {
536 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no);
537 pandecode_indent++;
538
539 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
540 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
541 struct pandecode_mapped_memory *mem =
542 pandecode_find_mapped_gpu_mem_containing(rt_va);
543 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
544
545 pandecode_log("{\n");
546 pandecode_indent++;
547
548 pandecode_rt_format(rt->format);
549
550 /* TODO: How the actual heck does AFBC enabling work here? */
551 if (0) {
552 pandecode_log(".afbc = {\n");
553 pandecode_indent++;
554
555 char *a = pointer_as_memory_reference(rt->afbc.metadata);
556 pandecode_prop("metadata = %s", a);
557 free(a);
558
559 pandecode_prop("stride = %d", rt->afbc.stride);
560 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
561
562 pandecode_indent--;
563 pandecode_log("},\n");
564 } else {
565 pandecode_log(".chunknown = {\n");
566 pandecode_indent++;
567
568 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
569
570 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
571 pandecode_prop("pointer = %s", a);
572 free(a);
573
574 pandecode_indent--;
575 pandecode_log("},\n");
576 }
577
578 MEMORY_PROP(rt, framebuffer);
579 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
580
581 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
582 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
583 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
584 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
585 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
586 }
587
588 if (rt->zero1 || rt->zero2 || rt->zero3) {
589 pandecode_msg("render target zeros tripped\n");
590 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
591 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
592 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
593 }
594
595 pandecode_indent--;
596 pandecode_log("},\n");
597 }
598
599 pandecode_indent--;
600 pandecode_log("};\n");
601 }
602
603 static void
604 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
605 {
606 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
607 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
608
609 if (fb->sample_locations) {
610 /* The blob stores all possible sample locations in a single buffer
611 * allocated on startup, and just switches the pointer when switching
612 * MSAA state. For now, we just put the data into the cmdstream, but we
613 * should do something like what the blob does with a real driver.
614 *
615 * There seem to be 32 slots for sample locations, followed by another
616 * 16. The second 16 is just the center location followed by 15 zeros
617 * in all the cases I've identified (maybe shader vs. depth/color
618 * samples?).
619 */
620
621 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
622
623 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
624
625 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
626 pandecode_indent++;
627
628 for (int i = 0; i < 32 + 16; i++) {
629 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
630 }
631
632 pandecode_indent--;
633 pandecode_log("};\n");
634 }
635
636 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no);
637 pandecode_indent++;
638
639 pandecode_prop("unk0 = 0x%x", fb->unk0);
640
641 if (fb->sample_locations)
642 pandecode_prop("sample_locations = sample_locations_%d", job_no);
643
644 /* Assume that unknown1 was emitted in the last job for
645 * now */
646 MEMORY_PROP(fb, unknown1);
647
648 pandecode_prop("tiler_polygon_list_size = 0x%x", fb->tiler_polygon_list_size);
649 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, fb->tiler_hierarchy_mask);
650 pandecode_prop("tiler_flags = 0x%" PRIx16, fb->tiler_flags);
651
652 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
653 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
654 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
655 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
656
657 pandecode_prop("unk1 = 0x%x", fb->unk1);
658 pandecode_prop("unk2 = 0x%x", fb->unk2);
659 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
660 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
661
662 pandecode_prop("unk3 = 0x%x", fb->unk3);
663 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
664 pandecode_prop("clear_depth = %f", fb->clear_depth);
665
666 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
667 MEMORY_PROP(fb, scratchpad);
668 MEMORY_PROP(fb, tiler_polygon_list);
669 MEMORY_PROP(fb, tiler_polygon_list_body);
670 MEMORY_PROP(fb, tiler_heap_start);
671 MEMORY_PROP(fb, tiler_heap_end);
672
673 if (fb->zero3 || fb->zero4) {
674 pandecode_msg("framebuffer zeros tripped\n");
675 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
676 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
677 }
678
679 bool nonzero_weights = false;
680
681 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
682 nonzero_weights |= fb->tiler_weights[w] != 0x0;
683 }
684
685 if (nonzero_weights) {
686 pandecode_log(".tiler_weights = {");
687
688 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
689 pandecode_log("%d, ", fb->tiler_weights[w]);
690 }
691
692 pandecode_log("},");
693 }
694
695 pandecode_indent--;
696 pandecode_log("};\n");
697
698 gpu_va += sizeof(struct bifrost_framebuffer);
699
700 if ((fb->unk3 & MALI_MFBD_EXTRA) && with_render_targets) {
701 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
702 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
703
704 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no);
705 pandecode_indent++;
706
707 MEMORY_PROP(fbx, checksum);
708
709 if (fbx->checksum_stride)
710 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
711
712 pandecode_log(".flags = ");
713 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
714 pandecode_log_cont(",\n");
715
716 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
717 pandecode_log(".ds_afbc = {\n");
718 pandecode_indent++;
719
720 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
721 pandecode_prop("depth_stencil_afbc_stride = %d",
722 fbx->ds_afbc.depth_stencil_afbc_stride);
723 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
724
725 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
726 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
727 pandecode_prop("zero1 = 0x%" PRIx32,
728 fbx->ds_afbc.zero1);
729 pandecode_prop("padding = 0x%" PRIx64,
730 fbx->ds_afbc.padding);
731 }
732
733 pandecode_indent--;
734 pandecode_log("},\n");
735 } else {
736 pandecode_log(".ds_linear = {\n");
737 pandecode_indent++;
738
739 if (fbx->ds_linear.depth) {
740 MEMORY_PROP((&fbx->ds_linear), depth);
741 pandecode_prop("depth_stride = %d",
742 fbx->ds_linear.depth_stride);
743 }
744
745 if (fbx->ds_linear.stencil) {
746 MEMORY_PROP((&fbx->ds_linear), stencil);
747 pandecode_prop("stencil_stride = %d",
748 fbx->ds_linear.stencil_stride);
749 }
750
751 if (fbx->ds_linear.depth_stride_zero ||
752 fbx->ds_linear.stencil_stride_zero ||
753 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
754 pandecode_msg("Depth/stencil zeros tripped\n");
755 pandecode_prop("depth_stride_zero = 0x%x",
756 fbx->ds_linear.depth_stride_zero);
757 pandecode_prop("stencil_stride_zero = 0x%x",
758 fbx->ds_linear.stencil_stride_zero);
759 pandecode_prop("zero1 = 0x%" PRIx32,
760 fbx->ds_linear.zero1);
761 pandecode_prop("zero2 = 0x%" PRIx32,
762 fbx->ds_linear.zero2);
763 }
764
765 pandecode_indent--;
766 pandecode_log("},\n");
767 }
768
769 if (fbx->zero3 || fbx->zero4) {
770 pandecode_msg("fb_extra zeros tripped\n");
771 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
772 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
773 }
774
775 pandecode_indent--;
776 pandecode_log("};\n");
777
778 gpu_va += sizeof(struct bifrost_fb_extra);
779 }
780
781 if (with_render_targets)
782 pandecode_render_target(gpu_va, job_no, fb);
783 }
784
785 static void
786 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
787 mali_ptr addr, int job_no, char *suffix,
788 int count, bool varying)
789 {
790 char *prefix = varying ? "varyings" : "attributes";
791
792 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
793
794 char base[128];
795 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
796
797 for (int i = 0; i < count; ++i) {
798 enum mali_attr_mode mode = attr[i].elements & 7;
799
800 if (mode == MALI_ATTR_UNUSED)
801 continue;
802
803 mali_ptr raw_elements = attr[i].elements & ~7;
804
805 /* TODO: Do we maybe want to dump the attribute values
806 * themselves given the specified format? Or is that too hard?
807 * */
808
809 char *a = pointer_as_memory_reference(raw_elements);
810 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
811 free(a);
812 }
813
814 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
815 pandecode_indent++;
816
817 for (int i = 0; i < count; ++i) {
818 pandecode_log("{\n");
819 pandecode_indent++;
820
821 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(attr[i].elements & 7));
822 pandecode_prop("shift = %d", attr[i].shift);
823 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
824 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
825 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
826 pandecode_indent--;
827 pandecode_log("}, \n");
828
829 if ((attr[i].elements & 7) == MALI_ATTR_NPOT_DIVIDE) {
830 i++;
831 pandecode_log("{\n");
832 pandecode_indent++;
833 pandecode_prop("unk = 0x%x", attr[i].unk);
834 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
835 if (attr[i].zero != 0)
836 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
837 pandecode_prop("divisor = %d", attr[i].divisor);
838 pandecode_indent--;
839 pandecode_log("}, \n");
840 }
841
842 }
843
844 pandecode_indent--;
845 pandecode_log("};\n");
846 }
847
848 static mali_ptr
849 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
850 {
851 /* TODO: Decode flags */
852 mali_ptr shader_ptr = ptr & ~15;
853
854 char *a = pointer_as_memory_reference(shader_ptr);
855 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
856 free(a);
857
858 return shader_ptr;
859 }
860
861 static void
862 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
863 {
864 const char *func = pandecode_func_name(stencil->func);
865 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
866 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
867 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
868
869 if (stencil->zero)
870 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
871
872 pandecode_log(".stencil_%s = {\n", name);
873 pandecode_indent++;
874 pandecode_prop("ref = %d", stencil->ref);
875 pandecode_prop("mask = 0x%02X", stencil->mask);
876 pandecode_prop("func = %s", func);
877 pandecode_prop("sfail = %s", sfail);
878 pandecode_prop("dpfail = %s", dpfail);
879 pandecode_prop("dppass = %s", dppass);
880 pandecode_indent--;
881 pandecode_log("},\n");
882 }
883
884 static void
885 pandecode_replay_blend_equation(const struct mali_blend_equation *blend)
886 {
887 if (blend->zero1)
888 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
889
890 pandecode_log(".equation = {\n");
891 pandecode_indent++;
892
893 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
894 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
895
896 pandecode_log(".color_mask = ");
897 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
898 pandecode_log_cont(",\n");
899
900 pandecode_indent--;
901 pandecode_log("},\n");
902 }
903
904 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
905
906 static unsigned
907 decode_bifrost_constant(u16 constant)
908 {
909 float lo = (float) (constant & 0xFF);
910 float hi = (float) (constant >> 8);
911
912 return (hi / 255.0) + (lo / 65535.0);
913 }
914
915 static mali_ptr
916 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
917 {
918 struct bifrost_blend_rt *b =
919 ((struct bifrost_blend_rt *) descs) + rt_no;
920
921 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
922 pandecode_indent++;
923
924 pandecode_prop("flags = 0x%" PRIx16, b->flags);
925 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
926 b->constant, decode_bifrost_constant(b->constant));
927
928 /* TODO figure out blend shader enable bit */
929 pandecode_replay_blend_equation(&b->equation);
930 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
931 pandecode_prop("index = 0x%" PRIx16, b->index);
932 pandecode_prop("shader = 0x%" PRIx32, b->shader);
933
934 pandecode_indent--;
935 pandecode_log("},\n");
936
937 return 0;
938 }
939
940 static mali_ptr
941 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
942 {
943 pandecode_log(".blend = {\n");
944 pandecode_indent++;
945
946 if (is_shader) {
947 pandecode_replay_shader_address("shader", blend->shader);
948 } else {
949 pandecode_replay_blend_equation(&blend->equation);
950 pandecode_prop("constant = %f", blend->constant);
951 }
952
953 pandecode_indent--;
954 pandecode_log("},\n");
955
956 /* Return blend shader to disassemble if present */
957 return is_shader ? (blend->shader & ~0xF) : 0;
958 }
959
960 static mali_ptr
961 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
962 {
963 struct midgard_blend_rt *b =
964 ((struct midgard_blend_rt *) descs) + rt_no;
965
966 /* Flags determine presence of blend shader */
967 bool is_shader = (b->flags & 0xF) >= 0x2;
968
969 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
970 pandecode_indent++;
971
972 pandecode_prop("flags = 0x%" PRIx64, b->flags);
973
974 mali_ptr shader = pandecode_midgard_blend(&b->blend, is_shader);
975
976 pandecode_indent--;
977 pandecode_log("};\n");
978
979 return shader;
980 }
981
982 static int
983 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
984 {
985 char base[128];
986 char *prefix = varying ? "varying" : "attribute";
987 unsigned max_index = 0;
988 snprintf(base, sizeof(base), "%s_meta", prefix);
989
990 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
991 pandecode_indent++;
992
993 struct mali_attr_meta *attr_meta;
994 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
995
996 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
997
998 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
999 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1000 sizeof(*attr_mem));
1001
1002 pandecode_log("{\n");
1003 pandecode_indent++;
1004 pandecode_prop("index = %d", attr_meta->index);
1005
1006 if (attr_meta->index > max_index)
1007 max_index = attr_meta->index;
1008 pandecode_replay_swizzle(attr_meta->swizzle);
1009 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
1010
1011 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1012 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1013 pandecode_prop("src_offset = 0x%" PRIx64, (u64) attr_meta->src_offset);
1014 pandecode_indent--;
1015 pandecode_log("},\n");
1016
1017 }
1018
1019 pandecode_indent--;
1020 pandecode_log("};\n");
1021
1022 return max_index;
1023 }
1024
1025 static void
1026 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1027 {
1028 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1029
1030 if (imem) {
1031 /* Indices are literally just a u32 array :) */
1032
1033 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1034
1035 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1036 pandecode_indent++;
1037
1038 for (unsigned i = 0; i < (index_count + 1); i += 3)
1039 pandecode_log("%d, %d, %d,\n",
1040 indices[i],
1041 indices[i + 1],
1042 indices[i + 2]);
1043
1044 pandecode_indent--;
1045 pandecode_log("};\n");
1046 }
1047 }
1048
1049 /* return bits [lo, hi) of word */
1050 static u32
1051 bits(u32 word, u32 lo, u32 hi)
1052 {
1053 if (hi - lo >= 32)
1054 return word; // avoid undefined behavior with the shift
1055
1056 return (word >> lo) & ((1 << (hi - lo)) - 1);
1057 }
1058
1059 static void
1060 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1061 {
1062 pandecode_log_cont("{\n");
1063 pandecode_indent++;
1064
1065 pandecode_prop("invocation_count = %" PRIx32, p->invocation_count);
1066 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1067 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1068 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1069 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1070 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1071 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1072
1073 /* Decode invocation_count. See the comment before the definition of
1074 * invocation_count for an explanation.
1075 */
1076 pandecode_msg("size: (%d, %d, %d)\n",
1077 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1078 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1079 bits(p->invocation_count, p->size_z_shift,
1080 p->workgroups_x_shift) + 1);
1081 pandecode_msg("workgroups: (%d, %d, %d)\n",
1082 bits(p->invocation_count, p->workgroups_x_shift,
1083 p->workgroups_y_shift) + 1,
1084 bits(p->invocation_count, p->workgroups_y_shift,
1085 p->workgroups_z_shift) + 1,
1086 bits(p->invocation_count, p->workgroups_z_shift,
1087 32) + 1);
1088
1089 /* TODO: Decode */
1090 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1091 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1092
1093 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
1094
1095 /* Index count only exists for tiler jobs anyway */
1096
1097 if (p->index_count)
1098 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1099
1100 pandecode_prop("negative_start = %d", p->negative_start);
1101
1102 DYN_MEMORY_PROP(p, job_no, indices);
1103
1104 if (p->zero1) {
1105 pandecode_msg("Zero tripped\n");
1106 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1107 }
1108
1109 pandecode_indent--;
1110 pandecode_log("},\n");
1111 }
1112
1113 static void
1114 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1115 {
1116 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1117
1118 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1119
1120 for (int i = 0; i < ubufs_count; i++) {
1121 mali_ptr ptr = ubufs[i].ptr << 2;
1122 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1123 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1124 char name[50];
1125 snprintf(name, sizeof(name), "ubuf_%d", i);
1126 /* The blob uses ubuf 0 to upload internal stuff and
1127 * uniforms that won't fit/are accessed indirectly, so
1128 * it puts it in the batchbuffer.
1129 */
1130 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1131 pandecode_indent++;
1132
1133 for (int j = 0; j <= ubufs[i].size; j++) {
1134 for (int k = 0; k < 4; k++) {
1135 if (k == 0)
1136 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1137 else
1138 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1139
1140 }
1141
1142 pandecode_log_cont("\n");
1143 }
1144
1145 pandecode_indent--;
1146 pandecode_log("};\n");
1147 }
1148
1149 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1150 job_no);
1151 pandecode_indent++;
1152
1153 for (int i = 0; i < ubufs_count; i++) {
1154 pandecode_log("{\n");
1155 pandecode_indent++;
1156 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1157 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1158 pandecode_indent--;
1159 pandecode_log("},\n");
1160 }
1161
1162 pandecode_indent--;
1163 pandecode_log("};\n");
1164 }
1165
1166 static void
1167 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1168 {
1169
1170 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1171
1172 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1173
1174 if (scratchpad->zero)
1175 pandecode_msg("XXX scratchpad zero tripped");
1176
1177 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no, suffix);
1178 pandecode_indent++;
1179
1180 pandecode_prop("flags = 0x%x", scratchpad->flags);
1181 MEMORY_PROP(scratchpad, gpu_scratchpad);
1182
1183 pandecode_indent--;
1184 pandecode_log("};\n");
1185 }
1186
1187 static void
1188 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1189 bool is_bifrost)
1190 {
1191 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1192 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1193
1194 /* Compute maximum possible size */
1195 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1196
1197 /* Print some boilerplate to clearly denote the assembly (which doesn't
1198 * obey indentation rules), and actually do the disassembly! */
1199
1200 printf("\n\n");
1201
1202 if (is_bifrost) {
1203 disassemble_bifrost(code, sz, false);
1204 } else {
1205 disassemble_midgard(code, sz);
1206 }
1207
1208 printf("\n\n");
1209 }
1210
1211 static void
1212 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1213 int job_no, enum mali_job_type job_type,
1214 char *suffix, bool is_bifrost)
1215 {
1216 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1217 struct pandecode_mapped_memory *attr_mem;
1218
1219 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1220 * are the only things actually needed from the FBD, vertex/tiler jobs
1221 * no longer reference the FBD -- instead, this field points to some
1222 * info about the scratchpad.
1223 */
1224 if (is_bifrost)
1225 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1226 else if (p->framebuffer & MALI_MFBD)
1227 pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1228 else
1229 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1230
1231 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1232 int texture_count = 0, sampler_count = 0;
1233
1234 if (shader_meta_ptr) {
1235 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1236 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1237
1238 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no, suffix);
1239 pandecode_indent++;
1240
1241 /* Save for dumps */
1242 attribute_count = s->attribute_count;
1243 varying_count = s->varying_count;
1244 texture_count = s->texture_count;
1245 sampler_count = s->sampler_count;
1246
1247 if (is_bifrost) {
1248 uniform_count = s->bifrost2.uniform_count;
1249 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1250 } else {
1251 uniform_count = s->midgard1.uniform_count;
1252 /* TODO figure this out */
1253 uniform_buffer_count = 1;
1254 }
1255
1256 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1257
1258 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1259 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1260 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1261 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1262
1263 if (is_bifrost) {
1264 pandecode_log(".bifrost1 = {\n");
1265 pandecode_indent++;
1266
1267 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1268 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1269
1270 pandecode_indent--;
1271 pandecode_log("},\n");
1272 } else {
1273 pandecode_log(".midgard1 = {\n");
1274 pandecode_indent++;
1275
1276 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1277 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1278
1279 pandecode_log(".unknown1 = ");
1280 pandecode_log_decoded_flags(shader_unknown1_flag_info, s->midgard1.unknown1);
1281 pandecode_log_cont(",\n");
1282
1283 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1284
1285 pandecode_indent--;
1286 pandecode_log("},\n");
1287 }
1288
1289 if (s->depth_units || s->depth_factor) {
1290 if (is_bifrost)
1291 pandecode_prop("depth_units = %f", s->depth_units);
1292 else
1293 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f);
1294
1295 pandecode_prop("depth_factor = %f", s->depth_factor);
1296 }
1297
1298 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1299 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1300
1301 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1302 invert_alpha_coverage ? "~" : "",
1303 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1304
1305 pandecode_log(".unknown2_3 = ");
1306
1307 int unknown2_3 = s->unknown2_3;
1308 int unknown2_4 = s->unknown2_4;
1309
1310 /* We're not quite sure what these flags mean without the depth test, if anything */
1311
1312 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1313 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1314 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1315
1316 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1317 }
1318
1319 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1320 pandecode_log_cont(",\n");
1321
1322 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1323 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1324
1325 pandecode_log(".unknown2_4 = ");
1326 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1327 pandecode_log_cont(",\n");
1328
1329 pandecode_replay_stencil("front", &s->stencil_front);
1330 pandecode_replay_stencil("back", &s->stencil_back);
1331
1332 if (is_bifrost) {
1333 pandecode_log(".bifrost2 = {\n");
1334 pandecode_indent++;
1335
1336 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1337 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1338 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1339 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1340
1341 pandecode_indent--;
1342 pandecode_log("},\n");
1343 } else {
1344 pandecode_log(".midgard2 = {\n");
1345 pandecode_indent++;
1346
1347 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1348 pandecode_indent--;
1349 pandecode_log("},\n");
1350 }
1351
1352 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1353
1354 if (!is_bifrost) {
1355 /* TODO: Blend shaders routing/disasm */
1356
1357 pandecode_midgard_blend(&s->blend, false);
1358 }
1359
1360 pandecode_indent--;
1361 pandecode_log("};\n");
1362
1363 /* MRT blend fields are used whenever MFBD is used, with
1364 * per-RT descriptors */
1365
1366 if (job_type == JOB_TYPE_TILER) {
1367 void* blend_base = (void *) (s + 1);
1368
1369 for (unsigned i = 0; i < 4; i++) {
1370 mali_ptr shader = 0;
1371
1372 if (is_bifrost)
1373 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1374 else
1375 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1376
1377 if (shader)
1378 pandecode_shader_disassemble(shader, job_no, job_type, false);
1379 }
1380 }
1381
1382 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1383 } else
1384 pandecode_msg("<no shader>\n");
1385
1386 if (p->viewport) {
1387 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1388 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1389
1390 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1391 pandecode_indent++;
1392
1393 pandecode_prop("clip_minx = %f", f->clip_minx);
1394 pandecode_prop("clip_miny = %f", f->clip_miny);
1395 pandecode_prop("clip_minz = %f", f->clip_minz);
1396 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1397 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1398 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1399
1400 /* Only the higher coordinates are MALI_POSITIVE scaled */
1401
1402 pandecode_prop("viewport0 = { %d, %d }",
1403 f->viewport0[0], f->viewport0[1]);
1404
1405 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1406 f->viewport1[0] + 1, f->viewport1[1] + 1);
1407
1408 pandecode_indent--;
1409 pandecode_log("};\n");
1410 }
1411
1412 if (p->attribute_meta) {
1413 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1414
1415 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1416 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1417 }
1418
1419 /* Varyings are encoded like attributes but not actually sent; we just
1420 * pass a zero buffer with the right stride/size set, (or whatever)
1421 * since the GPU will write to it itself */
1422
1423 if (p->varyings) {
1424 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1425
1426 /* Number of descriptors depends on whether there are
1427 * non-internal varyings */
1428
1429 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1430 }
1431
1432 if (p->varying_meta) {
1433 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1434 }
1435
1436 if (p->uniforms) {
1437 int rows = uniform_count, width = 4;
1438 size_t sz = rows * width * sizeof(float);
1439
1440 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1441 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1442 float *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1443
1444 pandecode_log("float uniforms_%d%s[] = {\n", job_no, suffix);
1445
1446 pandecode_indent++;
1447
1448 for (int row = 0; row < rows; row++) {
1449 for (int i = 0; i < width; i++)
1450 pandecode_log_cont("%ff, ", uniforms[i]);
1451
1452 pandecode_log_cont("\n");
1453
1454 uniforms += width;
1455 }
1456
1457 pandecode_indent--;
1458 pandecode_log("};\n");
1459 }
1460
1461 if (p->uniform_buffers) {
1462 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1463 }
1464
1465 if (p->texture_trampoline) {
1466 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1467
1468 if (mmem) {
1469 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1470
1471 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1472 pandecode_indent++;
1473
1474 for (int tex = 0; tex < texture_count; ++tex) {
1475 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1476 char *a = pointer_as_memory_reference(*u);
1477 pandecode_log("%s,\n", a);
1478 free(a);
1479 }
1480
1481 pandecode_indent--;
1482 pandecode_log("};\n");
1483
1484 /* Now, finally, descend down into the texture descriptor */
1485 for (int tex = 0; tex < texture_count; ++tex) {
1486 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1487 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1488
1489 if (tmem) {
1490 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1491
1492 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
1493 pandecode_indent++;
1494
1495 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1496 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1497 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1498
1499 pandecode_prop("unknown1 = %" PRId16, t->unknown1);
1500 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1501 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1502 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1503
1504 struct mali_texture_format f = t->format;
1505
1506 pandecode_log(".format = {\n");
1507 pandecode_indent++;
1508
1509 pandecode_replay_swizzle(f.swizzle);
1510 pandecode_prop("format = %s", pandecode_format_name(f.format));
1511
1512 pandecode_prop("usage1 = 0x%" PRIx32, f.usage1);
1513 pandecode_prop("is_not_cubemap = %" PRId32, f.is_not_cubemap);
1514 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1515
1516 pandecode_indent--;
1517 pandecode_log("},\n");
1518
1519 pandecode_replay_swizzle(t->swizzle);
1520
1521 if (t->swizzle_zero) {
1522 /* Shouldn't happen */
1523 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1524 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1525 }
1526
1527 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1528
1529 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1530 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1531 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1532
1533 pandecode_log(".payload = {\n");
1534 pandecode_indent++;
1535
1536 /* A bunch of bitmap pointers follow.
1537 * We work out the correct number,
1538 * based on the mipmap/cubemap
1539 * properties, but dump extra
1540 * possibilities to futureproof */
1541
1542 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1543 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1544
1545 /* Miptree for each face */
1546 if (!f.is_not_cubemap)
1547 bitmap_count *= 6;
1548
1549 /* Stride for each element */
1550 if (manual_stride)
1551 bitmap_count *= 2;
1552
1553 /* Sanity check the size */
1554 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1555 assert (bitmap_count <= max_count);
1556
1557 /* Dump more to be safe, but not _that_ much more */
1558 int safe_count = MIN2(bitmap_count * 2, max_count);
1559
1560 for (int i = 0; i < safe_count; ++i) {
1561 char *prefix = (i >= bitmap_count) ? "// " : "";
1562
1563 /* How we dump depends if this is a stride or a pointer */
1564
1565 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1566 /* signed 32-bit snuck in as a 64-bit pointer */
1567 uint64_t stride_set = t->payload[i];
1568 uint32_t clamped_stride = stride_set;
1569 int32_t stride = clamped_stride;
1570 assert(stride_set == clamped_stride);
1571 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1572 } else {
1573 char *a = pointer_as_memory_reference(t->payload[i]);
1574 pandecode_log("%s%s, \n", prefix, a);
1575 free(a);
1576 }
1577 }
1578
1579 pandecode_indent--;
1580 pandecode_log("},\n");
1581
1582 pandecode_indent--;
1583 pandecode_log("};\n");
1584 }
1585 }
1586 }
1587 }
1588
1589 if (p->sampler_descriptor) {
1590 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1591
1592 if (smem) {
1593 struct mali_sampler_descriptor *s;
1594
1595 mali_ptr d = p->sampler_descriptor;
1596
1597 for (int i = 0; i < sampler_count; ++i) {
1598 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1599
1600 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1601 pandecode_indent++;
1602
1603 /* Only the lower two bits are understood right now; the rest we display as hex */
1604 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1605 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1606 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1607 s->filter_mode & ~3);
1608
1609 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1610 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1611
1612 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1613 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1614 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1615
1616 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1617
1618 if (s->zero || s->zero2) {
1619 pandecode_msg("Zero tripped\n");
1620 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1621 }
1622
1623 pandecode_prop("unknown2 = %d", s->unknown2);
1624
1625 pandecode_prop("border_color = { %f, %f, %f, %f }",
1626 s->border_color[0],
1627 s->border_color[1],
1628 s->border_color[2],
1629 s->border_color[3]);
1630
1631 pandecode_indent--;
1632 pandecode_log("};\n");
1633 }
1634 }
1635 }
1636 }
1637
1638 static void
1639 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1640 {
1641 pandecode_log_cont("{\n");
1642 pandecode_indent++;
1643
1644 MEMORY_PROP(p, position_varying);
1645 MEMORY_COMMENT(p, position_varying);
1646 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1647 MEMORY_COMMENT(p, uniform_buffers);
1648 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1649 MEMORY_COMMENT(p, texture_trampoline);
1650 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1651 MEMORY_COMMENT(p, sampler_descriptor);
1652 DYN_MEMORY_PROP(p, job_no, uniforms);
1653 MEMORY_COMMENT(p, uniforms);
1654 DYN_MEMORY_PROP(p, job_no, attributes);
1655 MEMORY_COMMENT(p, attributes);
1656 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1657 MEMORY_COMMENT(p, attribute_meta);
1658 DYN_MEMORY_PROP(p, job_no, varyings);
1659 MEMORY_COMMENT(p, varyings);
1660 DYN_MEMORY_PROP(p, job_no, varying_meta);
1661 MEMORY_COMMENT(p, varying_meta);
1662 DYN_MEMORY_PROP(p, job_no, viewport);
1663 MEMORY_COMMENT(p, viewport);
1664 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1665 MEMORY_COMMENT(p, occlusion_counter);
1666 MEMORY_COMMENT(p, framebuffer & ~1);
1667 pandecode_msg("%" PRIx64 "\n", p->viewport);
1668 pandecode_msg("%" PRIx64 "\n", p->framebuffer);
1669
1670 if (is_bifrost)
1671 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1672 else
1673 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1674
1675 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1676 pandecode_prop("flags = %d", p->flags);
1677
1678 pandecode_indent--;
1679 pandecode_log("},\n");
1680 }
1681
1682 static void
1683 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1684 {
1685 pandecode_log_cont("{\n");
1686 pandecode_indent++;
1687
1688 pandecode_prop("unk2 = 0x%x", v->unk2);
1689
1690 if (v->zero0 || v->zero1) {
1691 pandecode_msg("vertex only zero tripped");
1692 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1693 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1694 }
1695
1696 pandecode_indent--;
1697 pandecode_log("}\n");
1698 }
1699
1700 static void
1701 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1702 {
1703
1704 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1705 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1706
1707 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1708 pandecode_indent++;
1709
1710 if (h->zero) {
1711 pandecode_msg("tiler heap zero tripped\n");
1712 pandecode_prop("zero = 0x%x", h->zero);
1713 }
1714
1715 for (int i = 0; i < 12; i++) {
1716 if (h->zeros[i] != 0) {
1717 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1718 i, h->zeros[i]);
1719 }
1720 }
1721
1722 pandecode_prop("heap_size = 0x%x", h->heap_size);
1723 MEMORY_PROP(h, tiler_heap_start);
1724 MEMORY_PROP(h, tiler_heap_free);
1725
1726 /* this might point to the beginning of another buffer, when it's
1727 * really the end of the tiler heap buffer, so we have to be careful
1728 * here.
1729 */
1730 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1731 pandecode_prop("tiler_heap_end = %s + 1", a);
1732 free(a);
1733
1734 pandecode_indent--;
1735 pandecode_log("};\n");
1736 }
1737
1738 static void
1739 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1740 {
1741 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1742 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1743
1744 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1745
1746 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
1747 pandecode_indent++;
1748
1749 if (t->zero0 || t->zero1) {
1750 pandecode_msg("tiler meta zero tripped");
1751 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1752 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1753 }
1754
1755 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
1756 pandecode_prop("flags = 0x%" PRIx16, t->flags);
1757
1758 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1759 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
1760 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
1761
1762 for (int i = 0; i < 12; i++) {
1763 if (t->zeros[i] != 0) {
1764 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
1765 i, t->zeros[i]);
1766 }
1767 }
1768
1769 pandecode_indent--;
1770 pandecode_log("};\n");
1771 }
1772
1773 static void
1774 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
1775 {
1776 pandecode_log(".gl_enables = ");
1777
1778 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
1779
1780 pandecode_log_cont(",\n");
1781 }
1782
1783 static void
1784 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
1785 {
1786 pandecode_log(".primitive_size = {\n");
1787 pandecode_indent++;
1788
1789 if (constant) {
1790 pandecode_prop("constant = %f", u.constant);
1791 } else {
1792 MEMORY_PROP((&u), pointer);
1793 }
1794
1795 pandecode_indent--;
1796 pandecode_log("},\n");
1797 }
1798
1799 static void
1800 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
1801 {
1802 pandecode_log_cont("{\n");
1803 pandecode_indent++;
1804
1805 /* TODO: gl_PointSize on Bifrost */
1806 pandecode_replay_primitive_size(t->primitive_size, true);
1807
1808 DYN_MEMORY_PROP(t, job_no, tiler_meta);
1809 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
1810
1811 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
1812 || t->zero6 || t->zero7 || t->zero8) {
1813 pandecode_msg("tiler only zero tripped");
1814 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1815 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
1816 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
1817 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
1818 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
1819 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
1820 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
1821 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
1822 }
1823
1824 pandecode_indent--;
1825 pandecode_log("},\n");
1826 }
1827
1828 static int
1829 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
1830 const struct pandecode_mapped_memory *mem,
1831 mali_ptr payload, int job_no)
1832 {
1833 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
1834
1835 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
1836
1837 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
1838 pandecode_indent++;
1839
1840 pandecode_log(".prefix = ");
1841 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1842
1843 pandecode_log(".vertex = ");
1844 pandecode_replay_vertex_only_bfr(&v->vertex);
1845
1846 pandecode_log(".postfix = ");
1847 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
1848
1849 pandecode_indent--;
1850 pandecode_log("};\n");
1851
1852 return sizeof(*v);
1853 }
1854
1855 static int
1856 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
1857 const struct pandecode_mapped_memory *mem,
1858 mali_ptr payload, int job_no)
1859 {
1860 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
1861
1862 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
1863
1864 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
1865 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
1866
1867 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
1868 pandecode_indent++;
1869
1870 pandecode_log(".prefix = ");
1871 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
1872
1873 pandecode_log(".tiler = ");
1874 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
1875
1876 pandecode_log(".postfix = ");
1877 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
1878
1879 pandecode_indent--;
1880 pandecode_log("};\n");
1881
1882 return sizeof(*t);
1883 }
1884
1885 static int
1886 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
1887 const struct pandecode_mapped_memory *mem,
1888 mali_ptr payload, int job_no)
1889 {
1890 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
1891
1892 char *a = pointer_as_memory_reference(payload);
1893 pandecode_msg("vt payload: %s\n", a);
1894 free(a);
1895
1896 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
1897
1898 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
1899
1900 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
1901 pandecode_indent++;
1902
1903 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
1904 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
1905
1906 pandecode_log(".prefix = ");
1907 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1908
1909 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
1910 pandecode_prop("draw_start = %d", v->draw_start);
1911
1912 #ifndef __LP64__
1913
1914 if (v->zero3) {
1915 pandecode_msg("Zero tripped\n");
1916 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
1917 }
1918
1919 #endif
1920
1921 if (v->zero5) {
1922 pandecode_msg("Zero tripped\n");
1923 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
1924 }
1925
1926 pandecode_log(".postfix = ");
1927 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
1928
1929 pandecode_indent--;
1930 pandecode_log("};\n");
1931
1932 return sizeof(*v);
1933 }
1934
1935 static int
1936 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
1937 mali_ptr payload, int job_no,
1938 bool is_bifrost)
1939 {
1940 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
1941
1942 bool fbd_dumped = false;
1943
1944 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
1945 /* Only SFBDs are understood, not MFBDs. We're speculating,
1946 * based on the versioning, kernel code, etc, that the
1947 * difference is between Single FrameBuffer Descriptor and
1948 * Multiple FrmaeBuffer Descriptor; the change apparently lines
1949 * up with multi-framebuffer support being added (T7xx onwards,
1950 * including Gxx). In any event, there's some field shuffling
1951 * that we haven't looked into yet. */
1952
1953 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
1954 fbd_dumped = true;
1955 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
1956 /* We don't know if Bifrost supports SFBD's at all, since the
1957 * driver never uses them. And the format is different from
1958 * Midgard anyways, due to the tiler heap and scratchpad being
1959 * moved out into separate structures, so it's not clear what a
1960 * Bifrost SFBD would even look like without getting an actual
1961 * trace, which appears impossible.
1962 */
1963
1964 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
1965 fbd_dumped = true;
1966 }
1967
1968 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
1969
1970 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no);
1971 pandecode_indent++;
1972
1973 /* See the comments by the macro definitions for mathematical context
1974 * on why this is so weird */
1975
1976 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
1977 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
1978
1979 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
1980 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
1981 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
1982
1983 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
1984 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
1985 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
1986
1987 /* If the FBD was just decoded, we can refer to it by pointer. If not,
1988 * we have to fallback on offsets. */
1989
1990 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
1991
1992 if (fbd_dumped)
1993 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
1994 else
1995 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
1996
1997 pandecode_indent--;
1998 pandecode_log("};\n");
1999
2000 return sizeof(*s);
2001 }
2002
2003 static int job_descriptor_number = 0;
2004
2005 int
2006 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
2007 {
2008 struct mali_job_descriptor_header *h;
2009
2010 int start_number = 0;
2011
2012 bool first = true;
2013 bool last_size;
2014
2015 do {
2016 struct pandecode_mapped_memory *mem =
2017 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2018
2019 void *payload;
2020
2021 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2022
2023 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2024 * high 32-bits of the 64-bit pointer are reused to store
2025 * something else.
2026 */
2027 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2028 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2029 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2030
2031 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2032 MALI_PAYLOAD_SIZE);
2033
2034 int job_no = job_descriptor_number++;
2035
2036 if (first)
2037 start_number = job_no;
2038
2039 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no);
2040 pandecode_indent++;
2041
2042 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
2043
2044 /* Save for next job fixing */
2045 last_size = h->job_descriptor_size;
2046
2047 if (h->job_descriptor_size)
2048 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2049
2050 if (h->exception_status)
2051 pandecode_prop("exception_status = %d", h->exception_status);
2052
2053 if (h->first_incomplete_task)
2054 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2055
2056 if (h->fault_pointer)
2057 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2058
2059 if (h->job_barrier)
2060 pandecode_prop("job_barrier = %d", h->job_barrier);
2061
2062 pandecode_prop("job_index = %d", h->job_index);
2063
2064 if (h->unknown_flags)
2065 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2066
2067 if (h->job_dependency_index_1)
2068 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2069
2070 if (h->job_dependency_index_2)
2071 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2072
2073 pandecode_indent--;
2074 pandecode_log("};\n");
2075
2076 /* Do not touch the field yet -- decode the payload first, and
2077 * don't touch that either. This is essential for the uploads
2078 * to occur in sequence and therefore be dynamically allocated
2079 * correctly. Do note the size, however, for that related
2080 * reason. */
2081
2082 switch (h->job_type) {
2083 case JOB_TYPE_SET_VALUE: {
2084 struct mali_payload_set_value *s = payload;
2085
2086 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no);
2087 pandecode_indent++;
2088 MEMORY_PROP(s, out);
2089 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2090 pandecode_indent--;
2091 pandecode_log("};\n");
2092
2093 break;
2094 }
2095
2096 case JOB_TYPE_TILER:
2097 case JOB_TYPE_VERTEX:
2098 case JOB_TYPE_COMPUTE:
2099 if (bifrost) {
2100 if (h->job_type == JOB_TYPE_TILER)
2101 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2102 else
2103 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2104 } else
2105 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2106
2107 break;
2108
2109 case JOB_TYPE_FRAGMENT:
2110 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2111 break;
2112
2113 default:
2114 break;
2115 }
2116
2117 /* Handle linkage */
2118
2119 if (!first) {
2120 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2121
2122 if (last_size)
2123 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2124 else
2125 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2126 }
2127
2128 first = false;
2129
2130 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2131
2132 return start_number;
2133 }