projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
panfrost: Decode rendering block type
[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 * Copyright (C) 2019 Collabora
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 */
25
26 #include <panfrost-job.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <memory.h>
30 #include <stdbool.h>
31 #include <stdarg.h>
32 #include "decode.h"
33 #include "util/u_math.h"
34
35 #include "../pan_pretty_print.h"
36 #include "../midgard/disassemble.h"
37 #include "../bifrost/disassemble.h"
38 int pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost);
39
40 #define MEMORY_PROP(obj, p) {\
41 char *a = pointer_as_memory_reference(obj->p); \
42 pandecode_prop("%s = %s", #p, a); \
43 free(a); \
44 }
45
46 #define MEMORY_COMMENT(obj, p) {\
47 char *a = pointer_as_memory_reference(obj->p); \
48 pandecode_msg("%s = %s\n", #p, a); \
49 free(a); \
50 }
51
52 #define DYN_MEMORY_PROP(obj, no, p) { \
53 if (obj->p) \
54 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
55 }
56
57 /* Semantic logging type.
58 *
59 * Raw: for raw messages to be printed as is.
60 * Message: for helpful information to be commented out in replays.
61 * Property: for properties of a struct
62 *
63 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
64 */
65
66 enum pandecode_log_type {
67 PANDECODE_RAW,
68 PANDECODE_MESSAGE,
69 PANDECODE_PROPERTY
70 };
71
72 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
73 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
74 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
75
76 unsigned pandecode_indent = 0;
77
78 static void
79 pandecode_make_indent(void)
80 {
81 for (unsigned i = 0; i < pandecode_indent; ++i)
82 printf(" ");
83 }
84
85 static void
86 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
87 {
88 va_list ap;
89
90 pandecode_make_indent();
91
92 if (type == PANDECODE_MESSAGE)
93 printf("// ");
94 else if (type == PANDECODE_PROPERTY)
95 printf(".");
96
97 va_start(ap, format);
98 vprintf(format, ap);
99 va_end(ap);
100
101 if (type == PANDECODE_PROPERTY)
102 printf(",\n");
103 }
104
105 static void
106 pandecode_log_cont(const char *format, ...)
107 {
108 va_list ap;
109
110 va_start(ap, format);
111 vprintf(format, ap);
112 va_end(ap);
113 }
114
115 struct pandecode_flag_info {
116 u64 flag;
117 const char *name;
118 };
119
120 static void
121 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
122 u64 flags)
123 {
124 bool decodable_flags_found = false;
125
126 for (int i = 0; flag_info[i].name; i++) {
127 if ((flags & flag_info[i].flag) != flag_info[i].flag)
128 continue;
129
130 if (!decodable_flags_found) {
131 decodable_flags_found = true;
132 } else {
133 pandecode_log_cont(" | ");
134 }
135
136 pandecode_log_cont("%s", flag_info[i].name);
137
138 flags &= ~flag_info[i].flag;
139 }
140
141 if (decodable_flags_found) {
142 if (flags)
143 pandecode_log_cont(" | 0x%" PRIx64, flags);
144 } else {
145 pandecode_log_cont("0x%" PRIx64, flags);
146 }
147 }
148
149 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
150 static const struct pandecode_flag_info gl_enable_flag_info[] = {
151 FLAG_INFO(OCCLUSION_QUERY),
152 FLAG_INFO(OCCLUSION_PRECISE),
153 FLAG_INFO(FRONT_CCW_TOP),
154 FLAG_INFO(CULL_FACE_FRONT),
155 FLAG_INFO(CULL_FACE_BACK),
156 {}
157 };
158 #undef FLAG_INFO
159
160 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
161 static const struct pandecode_flag_info clear_flag_info[] = {
162 FLAG_INFO(FAST),
163 FLAG_INFO(SLOW),
164 FLAG_INFO(SLOW_STENCIL),
165 {}
166 };
167 #undef FLAG_INFO
168
169 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
170 static const struct pandecode_flag_info mask_flag_info[] = {
171 FLAG_INFO(R),
172 FLAG_INFO(G),
173 FLAG_INFO(B),
174 FLAG_INFO(A),
175 {}
176 };
177 #undef FLAG_INFO
178
179 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
180 static const struct pandecode_flag_info u3_flag_info[] = {
181 FLAG_INFO(HAS_MSAA),
182 FLAG_INFO(CAN_DISCARD),
183 FLAG_INFO(HAS_BLEND_SHADER),
184 FLAG_INFO(DEPTH_TEST),
185 {}
186 };
187
188 static const struct pandecode_flag_info u4_flag_info[] = {
189 FLAG_INFO(NO_MSAA),
190 FLAG_INFO(NO_DITHER),
191 FLAG_INFO(DEPTH_RANGE_A),
192 FLAG_INFO(DEPTH_RANGE_B),
193 FLAG_INFO(STENCIL_TEST),
194 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
195 {}
196 };
197 #undef FLAG_INFO
198
199 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
200 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
201 FLAG_INFO(MSAA_A),
202 FLAG_INFO(MSAA_B),
203 FLAG_INFO(MSAA_8),
204 {}
205 };
206 #undef FLAG_INFO
207
208 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
209 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
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(EARLY_Z),
227 FLAG_INFO(HELPER_INVOCATIONS),
228 FLAG_INFO(READS_TILEBUFFER),
229 FLAG_INFO(READS_ZS),
230 {}
231 };
232 #undef FLAG_INFO
233
234 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
235 static const struct pandecode_flag_info mfbd_flag_info [] = {
236 FLAG_INFO(DEPTH_WRITE),
237 FLAG_INFO(EXTRA),
238 {}
239 };
240 #undef FLAG_INFO
241
242
243 extern char *replace_fragment;
244 extern char *replace_vertex;
245
246 static char *
247 pandecode_job_type_name(enum mali_job_type type)
248 {
249 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
250
251 switch (type) {
252 DEFINE_CASE(NULL);
253 DEFINE_CASE(SET_VALUE);
254 DEFINE_CASE(CACHE_FLUSH);
255 DEFINE_CASE(COMPUTE);
256 DEFINE_CASE(VERTEX);
257 DEFINE_CASE(TILER);
258 DEFINE_CASE(FUSED);
259 DEFINE_CASE(FRAGMENT);
260
261 case JOB_NOT_STARTED:
262 return "NOT_STARTED";
263
264 default:
265 pandecode_log("Warning! Unknown job type %x\n", type);
266 return "!?!?!?";
267 }
268
269 #undef DEFINE_CASE
270 }
271
272 static char *
273 pandecode_draw_mode_name(enum mali_draw_mode mode)
274 {
275 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
276
277 switch (mode) {
278 DEFINE_CASE(DRAW_NONE);
279 DEFINE_CASE(POINTS);
280 DEFINE_CASE(LINES);
281 DEFINE_CASE(TRIANGLES);
282 DEFINE_CASE(TRIANGLE_STRIP);
283 DEFINE_CASE(TRIANGLE_FAN);
284 DEFINE_CASE(LINE_STRIP);
285 DEFINE_CASE(LINE_LOOP);
286 DEFINE_CASE(POLYGON);
287 DEFINE_CASE(QUADS);
288 DEFINE_CASE(QUAD_STRIP);
289
290 default:
291 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
292 }
293
294 #undef DEFINE_CASE
295 }
296
297 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
298 static char *
299 pandecode_func_name(enum mali_func mode)
300 {
301 switch (mode) {
302 DEFINE_CASE(NEVER);
303 DEFINE_CASE(LESS);
304 DEFINE_CASE(EQUAL);
305 DEFINE_CASE(LEQUAL);
306 DEFINE_CASE(GREATER);
307 DEFINE_CASE(NOTEQUAL);
308 DEFINE_CASE(GEQUAL);
309 DEFINE_CASE(ALWAYS);
310
311 default:
312 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
313 }
314 }
315 #undef DEFINE_CASE
316
317 /* Why is this duplicated? Who knows... */
318 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
319 static char *
320 pandecode_alt_func_name(enum mali_alt_func mode)
321 {
322 switch (mode) {
323 DEFINE_CASE(NEVER);
324 DEFINE_CASE(LESS);
325 DEFINE_CASE(EQUAL);
326 DEFINE_CASE(LEQUAL);
327 DEFINE_CASE(GREATER);
328 DEFINE_CASE(NOTEQUAL);
329 DEFINE_CASE(GEQUAL);
330 DEFINE_CASE(ALWAYS);
331
332 default:
333 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
334 }
335 }
336 #undef DEFINE_CASE
337
338 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
339 static char *
340 pandecode_stencil_op_name(enum mali_stencil_op op)
341 {
342 switch (op) {
343 DEFINE_CASE(KEEP);
344 DEFINE_CASE(REPLACE);
345 DEFINE_CASE(ZERO);
346 DEFINE_CASE(INVERT);
347 DEFINE_CASE(INCR_WRAP);
348 DEFINE_CASE(DECR_WRAP);
349 DEFINE_CASE(INCR);
350 DEFINE_CASE(DECR);
351
352 default:
353 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
354 }
355 }
356
357 #undef DEFINE_CASE
358
359 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
360 static char *pandecode_attr_mode_name(enum mali_attr_mode mode)
361 {
362 switch(mode) {
363 DEFINE_CASE(UNUSED);
364 DEFINE_CASE(LINEAR);
365 DEFINE_CASE(POT_DIVIDE);
366 DEFINE_CASE(MODULO);
367 DEFINE_CASE(NPOT_DIVIDE);
368 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
369 }
370 }
371
372 #undef DEFINE_CASE
373
374 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
375 static char *
376 pandecode_channel_name(enum mali_channel channel)
377 {
378 switch (channel) {
379 DEFINE_CASE(RED);
380 DEFINE_CASE(GREEN);
381 DEFINE_CASE(BLUE);
382 DEFINE_CASE(ALPHA);
383 DEFINE_CASE(ZERO);
384 DEFINE_CASE(ONE);
385 DEFINE_CASE(RESERVED_0);
386 DEFINE_CASE(RESERVED_1);
387
388 default:
389 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
390 }
391 }
392 #undef DEFINE_CASE
393
394 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
395 static char *
396 pandecode_wrap_mode_name(enum mali_wrap_mode op)
397 {
398 switch (op) {
399 DEFINE_CASE(REPEAT);
400 DEFINE_CASE(CLAMP_TO_EDGE);
401 DEFINE_CASE(CLAMP_TO_BORDER);
402 DEFINE_CASE(MIRRORED_REPEAT);
403
404 default:
405 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
406 }
407 }
408 #undef DEFINE_CASE
409
410 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
411 static char *
412 pandecode_texture_type(enum mali_texture_type type)
413 {
414 switch (type) {
415 DEFINE_CASE(1D);
416 DEFINE_CASE(2D);
417 DEFINE_CASE(3D);
418 DEFINE_CASE(CUBE);
419
420 default:
421 unreachable("Unknown case");
422 }
423 }
424 #undef DEFINE_CASE
425
426 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
427 static char *
428 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
429 {
430 switch (fmt) {
431 DEFINE_CASE(TILED);
432 DEFINE_CASE(UNKNOWN);
433 DEFINE_CASE(LINEAR);
434 DEFINE_CASE(AFBC);
435
436 default:
437 unreachable("Invalid case");
438 }
439 }
440 #undef DEFINE_CASE
441
442 static inline char *
443 pandecode_decode_fbd_type(enum mali_fbd_type type)
444 {
445 if (type == MALI_SFBD) return "SFBD";
446 else if (type == MALI_MFBD) return "MFBD";
447 else return "WATFBD /* XXX */";
448 }
449
450 static void
451 pandecode_replay_sfbd(uint64_t gpu_va, int job_no)
452 {
453 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
454 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
455
456 pandecode_log("struct mali_single_framebuffer framebuffer_%d = {\n", job_no);
457 pandecode_indent++;
458
459 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
460 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
461
462 pandecode_log(".format = ");
463 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
464 pandecode_log_cont(",\n");
465
466 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
467 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
468
469 MEMORY_PROP(s, framebuffer);
470 pandecode_prop("stride = %d", s->stride);
471
472 /* Earlier in the actual commandstream -- right before width -- but we
473 * delay to flow nicer */
474
475 pandecode_log(".clear_flags = ");
476 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
477 pandecode_log_cont(",\n");
478
479 if (s->depth_buffer | s->depth_buffer_enable) {
480 MEMORY_PROP(s, depth_buffer);
481 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
482 }
483
484 if (s->stencil_buffer | s->stencil_buffer_enable) {
485 MEMORY_PROP(s, stencil_buffer);
486 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
487 }
488
489 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
490 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
491 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
492 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
493 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
494 }
495
496 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
497 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
498 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
499 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
500 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
501 }
502
503 if (s->clear_stencil) {
504 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
505 }
506
507 MEMORY_PROP(s, unknown_address_0);
508 MEMORY_PROP(s, tiler_polygon_list);
509 MEMORY_PROP(s, tiler_polygon_list_body);
510
511 pandecode_prop("tiler_resolution_check = 0x%" PRIx32, s->tiler_resolution_check);
512 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, s->tiler_hierarchy_mask);
513 pandecode_prop("tiler_flags = 0x%" PRIx16, s->tiler_flags);
514
515 MEMORY_PROP(s, tiler_heap_free);
516 MEMORY_PROP(s, tiler_heap_end);
517
518 pandecode_indent--;
519 pandecode_log("};\n");
520
521 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
522 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
523 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
524 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
525
526 printf(".zero3 = {");
527
528 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
529 printf("%X, ", s->zero3[i]);
530
531 printf("},\n");
532
533 printf(".zero6 = {");
534
535 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
536 printf("%X, ", s->zero6[i]);
537
538 printf("},\n");
539 }
540
541 static void
542 pandecode_replay_swizzle(unsigned swizzle)
543 {
544 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
545 pandecode_channel_name((swizzle >> 0) & 0x7),
546 pandecode_channel_name((swizzle >> 3) & 0x7),
547 pandecode_channel_name((swizzle >> 6) & 0x7),
548 pandecode_channel_name((swizzle >> 9) & 0x7));
549 }
550
551 static void
552 pandecode_rt_format(struct mali_rt_format format)
553 {
554 pandecode_log(".format = {\n");
555 pandecode_indent++;
556
557 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
558 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
559 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
560
561 pandecode_prop("block = %s",
562 pandecode_mfbd_block_format(format.block));
563
564 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
565 MALI_NEGATIVE(format.nr_channels));
566
567 pandecode_log(".flags = ");
568 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
569 pandecode_log_cont(",\n");
570
571 pandecode_replay_swizzle(format.swizzle);
572
573 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
574
575 pandecode_indent--;
576 pandecode_log("},\n");
577 }
578
579 static void
580 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
581 {
582 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no);
583 pandecode_indent++;
584
585 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
586 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
587 struct pandecode_mapped_memory *mem =
588 pandecode_find_mapped_gpu_mem_containing(rt_va);
589 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
590
591 pandecode_log("{\n");
592 pandecode_indent++;
593
594 pandecode_rt_format(rt->format);
595
596 /* TODO: How the actual heck does AFBC enabling work here? */
597 if (0) {
598 pandecode_log(".afbc = {\n");
599 pandecode_indent++;
600
601 char *a = pointer_as_memory_reference(rt->afbc.metadata);
602 pandecode_prop("metadata = %s", a);
603 free(a);
604
605 pandecode_prop("stride = %d", rt->afbc.stride);
606 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
607
608 pandecode_indent--;
609 pandecode_log("},\n");
610 } else {
611 pandecode_log(".chunknown = {\n");
612 pandecode_indent++;
613
614 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
615
616 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
617 pandecode_prop("pointer = %s", a);
618 free(a);
619
620 pandecode_indent--;
621 pandecode_log("},\n");
622 }
623
624 MEMORY_PROP(rt, framebuffer);
625 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
626
627 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
628 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
629 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
630 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
631 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
632 }
633
634 if (rt->zero1 || rt->zero2 || rt->zero3) {
635 pandecode_msg("render target zeros tripped\n");
636 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
637 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
638 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
639 }
640
641 pandecode_indent--;
642 pandecode_log("},\n");
643 }
644
645 pandecode_indent--;
646 pandecode_log("};\n");
647 }
648
649 static void
650 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
651 {
652 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
653 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
654
655 if (fb->sample_locations) {
656 /* The blob stores all possible sample locations in a single buffer
657 * allocated on startup, and just switches the pointer when switching
658 * MSAA state. For now, we just put the data into the cmdstream, but we
659 * should do something like what the blob does with a real driver.
660 *
661 * There seem to be 32 slots for sample locations, followed by another
662 * 16. The second 16 is just the center location followed by 15 zeros
663 * in all the cases I've identified (maybe shader vs. depth/color
664 * samples?).
665 */
666
667 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
668
669 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
670
671 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
672 pandecode_indent++;
673
674 for (int i = 0; i < 32 + 16; i++) {
675 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
676 }
677
678 pandecode_indent--;
679 pandecode_log("};\n");
680 }
681
682 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no);
683 pandecode_indent++;
684
685 pandecode_prop("unk0 = 0x%x", fb->unk0);
686
687 if (fb->sample_locations)
688 pandecode_prop("sample_locations = sample_locations_%d", job_no);
689
690 /* Assume that unknown1 was emitted in the last job for
691 * now */
692 MEMORY_PROP(fb, unknown1);
693
694 pandecode_prop("tiler_polygon_list_size = 0x%x", fb->tiler_polygon_list_size);
695 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, fb->tiler_hierarchy_mask);
696 pandecode_prop("tiler_flags = 0x%" PRIx16, fb->tiler_flags);
697
698 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
699 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
700 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
701 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
702
703 pandecode_prop("unk1 = 0x%x", fb->unk1);
704 pandecode_prop("unk2 = 0x%x", fb->unk2);
705 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
706 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
707
708 pandecode_log(".mfbd_flags = ");
709 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
710 pandecode_log_cont(",\n");
711
712 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
713 pandecode_prop("clear_depth = %f", fb->clear_depth);
714
715 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
716 MEMORY_PROP(fb, scratchpad);
717 MEMORY_PROP(fb, tiler_polygon_list);
718 MEMORY_PROP(fb, tiler_polygon_list_body);
719 MEMORY_PROP(fb, tiler_heap_start);
720 MEMORY_PROP(fb, tiler_heap_end);
721
722 if (fb->zero3 || fb->zero4) {
723 pandecode_msg("framebuffer zeros tripped\n");
724 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
725 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
726 }
727
728 bool nonzero_weights = false;
729
730 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
731 nonzero_weights |= fb->tiler_weights[w] != 0x0;
732 }
733
734 if (nonzero_weights) {
735 pandecode_log(".tiler_weights = {");
736
737 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
738 pandecode_log("%d, ", fb->tiler_weights[w]);
739 }
740
741 pandecode_log("},");
742 }
743
744 pandecode_indent--;
745 pandecode_log("};\n");
746
747 gpu_va += sizeof(struct bifrost_framebuffer);
748
749 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
750 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
751 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
752
753 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no);
754 pandecode_indent++;
755
756 MEMORY_PROP(fbx, checksum);
757
758 if (fbx->checksum_stride)
759 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
760
761 pandecode_log(".flags = ");
762 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
763 pandecode_log_cont(",\n");
764
765 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
766 pandecode_log(".ds_afbc = {\n");
767 pandecode_indent++;
768
769 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
770 pandecode_prop("depth_stencil_afbc_stride = %d",
771 fbx->ds_afbc.depth_stencil_afbc_stride);
772 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
773
774 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
775 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
776 pandecode_prop("zero1 = 0x%" PRIx32,
777 fbx->ds_afbc.zero1);
778 pandecode_prop("padding = 0x%" PRIx64,
779 fbx->ds_afbc.padding);
780 }
781
782 pandecode_indent--;
783 pandecode_log("},\n");
784 } else {
785 pandecode_log(".ds_linear = {\n");
786 pandecode_indent++;
787
788 if (fbx->ds_linear.depth) {
789 MEMORY_PROP((&fbx->ds_linear), depth);
790 pandecode_prop("depth_stride = %d",
791 fbx->ds_linear.depth_stride);
792 }
793
794 if (fbx->ds_linear.stencil) {
795 MEMORY_PROP((&fbx->ds_linear), stencil);
796 pandecode_prop("stencil_stride = %d",
797 fbx->ds_linear.stencil_stride);
798 }
799
800 if (fbx->ds_linear.depth_stride_zero ||
801 fbx->ds_linear.stencil_stride_zero ||
802 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
803 pandecode_msg("Depth/stencil zeros tripped\n");
804 pandecode_prop("depth_stride_zero = 0x%x",
805 fbx->ds_linear.depth_stride_zero);
806 pandecode_prop("stencil_stride_zero = 0x%x",
807 fbx->ds_linear.stencil_stride_zero);
808 pandecode_prop("zero1 = 0x%" PRIx32,
809 fbx->ds_linear.zero1);
810 pandecode_prop("zero2 = 0x%" PRIx32,
811 fbx->ds_linear.zero2);
812 }
813
814 pandecode_indent--;
815 pandecode_log("},\n");
816 }
817
818 if (fbx->zero3 || fbx->zero4) {
819 pandecode_msg("fb_extra zeros tripped\n");
820 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
821 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
822 }
823
824 pandecode_indent--;
825 pandecode_log("};\n");
826
827 gpu_va += sizeof(struct bifrost_fb_extra);
828 }
829
830 if (with_render_targets)
831 pandecode_render_target(gpu_va, job_no, fb);
832 }
833
834 static void
835 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
836 mali_ptr addr, int job_no, char *suffix,
837 int count, bool varying)
838 {
839 char *prefix = varying ? "varyings" : "attributes";
840
841 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
842
843 char base[128];
844 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
845
846 for (int i = 0; i < count; ++i) {
847 enum mali_attr_mode mode = attr[i].elements & 7;
848
849 if (mode == MALI_ATTR_UNUSED)
850 continue;
851
852 mali_ptr raw_elements = attr[i].elements & ~7;
853
854 /* TODO: Do we maybe want to dump the attribute values
855 * themselves given the specified format? Or is that too hard?
856 * */
857
858 char *a = pointer_as_memory_reference(raw_elements);
859 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
860 free(a);
861 }
862
863 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
864 pandecode_indent++;
865
866 for (int i = 0; i < count; ++i) {
867 pandecode_log("{\n");
868 pandecode_indent++;
869
870 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(attr[i].elements & 7));
871 pandecode_prop("shift = %d", attr[i].shift);
872 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
873 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
874 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
875 pandecode_indent--;
876 pandecode_log("}, \n");
877
878 if ((attr[i].elements & 7) == MALI_ATTR_NPOT_DIVIDE) {
879 i++;
880 pandecode_log("{\n");
881 pandecode_indent++;
882 pandecode_prop("unk = 0x%x", attr[i].unk);
883 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
884 if (attr[i].zero != 0)
885 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
886 pandecode_prop("divisor = %d", attr[i].divisor);
887 pandecode_indent--;
888 pandecode_log("}, \n");
889 }
890
891 }
892
893 pandecode_indent--;
894 pandecode_log("};\n");
895 }
896
897 static mali_ptr
898 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
899 {
900 /* TODO: Decode flags */
901 mali_ptr shader_ptr = ptr & ~15;
902
903 char *a = pointer_as_memory_reference(shader_ptr);
904 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
905 free(a);
906
907 return shader_ptr;
908 }
909
910 static void
911 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
912 {
913 const char *func = pandecode_func_name(stencil->func);
914 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
915 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
916 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
917
918 if (stencil->zero)
919 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
920
921 pandecode_log(".stencil_%s = {\n", name);
922 pandecode_indent++;
923 pandecode_prop("ref = %d", stencil->ref);
924 pandecode_prop("mask = 0x%02X", stencil->mask);
925 pandecode_prop("func = %s", func);
926 pandecode_prop("sfail = %s", sfail);
927 pandecode_prop("dpfail = %s", dpfail);
928 pandecode_prop("dppass = %s", dppass);
929 pandecode_indent--;
930 pandecode_log("},\n");
931 }
932
933 static void
934 pandecode_replay_blend_equation(const struct mali_blend_equation *blend)
935 {
936 if (blend->zero1)
937 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
938
939 pandecode_log(".equation = {\n");
940 pandecode_indent++;
941
942 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
943 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
944
945 pandecode_log(".color_mask = ");
946 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
947 pandecode_log_cont(",\n");
948
949 pandecode_indent--;
950 pandecode_log("},\n");
951 }
952
953 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
954
955 static unsigned
956 decode_bifrost_constant(u16 constant)
957 {
958 float lo = (float) (constant & 0xFF);
959 float hi = (float) (constant >> 8);
960
961 return (hi / 255.0) + (lo / 65535.0);
962 }
963
964 static mali_ptr
965 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
966 {
967 struct bifrost_blend_rt *b =
968 ((struct bifrost_blend_rt *) descs) + rt_no;
969
970 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
971 pandecode_indent++;
972
973 pandecode_prop("flags = 0x%" PRIx16, b->flags);
974 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
975 b->constant, decode_bifrost_constant(b->constant));
976
977 /* TODO figure out blend shader enable bit */
978 pandecode_replay_blend_equation(&b->equation);
979 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
980 pandecode_prop("index = 0x%" PRIx16, b->index);
981 pandecode_prop("shader = 0x%" PRIx32, b->shader);
982
983 pandecode_indent--;
984 pandecode_log("},\n");
985
986 return 0;
987 }
988
989 static mali_ptr
990 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
991 {
992 pandecode_log(".blend = {\n");
993 pandecode_indent++;
994
995 if (is_shader) {
996 pandecode_replay_shader_address("shader", blend->shader);
997 } else {
998 pandecode_replay_blend_equation(&blend->equation);
999 pandecode_prop("constant = %f", blend->constant);
1000 }
1001
1002 pandecode_indent--;
1003 pandecode_log("},\n");
1004
1005 /* Return blend shader to disassemble if present */
1006 return is_shader ? (blend->shader & ~0xF) : 0;
1007 }
1008
1009 static mali_ptr
1010 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1011 {
1012 struct midgard_blend_rt *b =
1013 ((struct midgard_blend_rt *) descs) + rt_no;
1014
1015 /* Flags determine presence of blend shader */
1016 bool is_shader = (b->flags & 0xF) >= 0x2;
1017
1018 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1019 pandecode_indent++;
1020
1021 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1022
1023 mali_ptr shader = pandecode_midgard_blend(&b->blend, is_shader);
1024
1025 pandecode_indent--;
1026 pandecode_log("};\n");
1027
1028 return shader;
1029 }
1030
1031 static int
1032 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1033 {
1034 char base[128];
1035 char *prefix = varying ? "varying" : "attribute";
1036 unsigned max_index = 0;
1037 snprintf(base, sizeof(base), "%s_meta", prefix);
1038
1039 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1040 pandecode_indent++;
1041
1042 struct mali_attr_meta *attr_meta;
1043 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1044
1045 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1046
1047 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1048 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1049 sizeof(*attr_mem));
1050
1051 pandecode_log("{\n");
1052 pandecode_indent++;
1053 pandecode_prop("index = %d", attr_meta->index);
1054
1055 if (attr_meta->index > max_index)
1056 max_index = attr_meta->index;
1057 pandecode_replay_swizzle(attr_meta->swizzle);
1058 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
1059
1060 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1061 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1062 pandecode_prop("src_offset = 0x%" PRIx64, (u64) attr_meta->src_offset);
1063 pandecode_indent--;
1064 pandecode_log("},\n");
1065
1066 }
1067
1068 pandecode_indent--;
1069 pandecode_log("};\n");
1070
1071 return max_index;
1072 }
1073
1074 static void
1075 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1076 {
1077 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1078
1079 if (imem) {
1080 /* Indices are literally just a u32 array :) */
1081
1082 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1083
1084 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1085 pandecode_indent++;
1086
1087 for (unsigned i = 0; i < (index_count + 1); i += 3)
1088 pandecode_log("%d, %d, %d,\n",
1089 indices[i],
1090 indices[i + 1],
1091 indices[i + 2]);
1092
1093 pandecode_indent--;
1094 pandecode_log("};\n");
1095 }
1096 }
1097
1098 /* return bits [lo, hi) of word */
1099 static u32
1100 bits(u32 word, u32 lo, u32 hi)
1101 {
1102 if (hi - lo >= 32)
1103 return word; // avoid undefined behavior with the shift
1104
1105 return (word >> lo) & ((1 << (hi - lo)) - 1);
1106 }
1107
1108 static void
1109 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1110 {
1111 pandecode_log_cont("{\n");
1112 pandecode_indent++;
1113
1114 pandecode_prop("invocation_count = %" PRIx32, p->invocation_count);
1115 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1116 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1117 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1118 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1119 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1120 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1121
1122 /* Decode invocation_count. See the comment before the definition of
1123 * invocation_count for an explanation.
1124 */
1125 pandecode_msg("size: (%d, %d, %d)\n",
1126 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1127 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1128 bits(p->invocation_count, p->size_z_shift,
1129 p->workgroups_x_shift) + 1);
1130 pandecode_msg("workgroups: (%d, %d, %d)\n",
1131 bits(p->invocation_count, p->workgroups_x_shift,
1132 p->workgroups_y_shift) + 1,
1133 bits(p->invocation_count, p->workgroups_y_shift,
1134 p->workgroups_z_shift) + 1,
1135 bits(p->invocation_count, p->workgroups_z_shift,
1136 32) + 1);
1137
1138 /* TODO: Decode */
1139 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1140 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1141
1142 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
1143
1144 /* Index count only exists for tiler jobs anyway */
1145
1146 if (p->index_count)
1147 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1148
1149 pandecode_prop("negative_start = %d", p->negative_start);
1150
1151 DYN_MEMORY_PROP(p, job_no, indices);
1152
1153 if (p->zero1) {
1154 pandecode_msg("Zero tripped\n");
1155 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1156 }
1157
1158 pandecode_indent--;
1159 pandecode_log("},\n");
1160 }
1161
1162 static void
1163 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1164 {
1165 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1166
1167 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1168
1169 for (int i = 0; i < ubufs_count; i++) {
1170 mali_ptr ptr = ubufs[i].ptr << 2;
1171 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1172 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1173 char name[50];
1174 snprintf(name, sizeof(name), "ubuf_%d", i);
1175 /* The blob uses ubuf 0 to upload internal stuff and
1176 * uniforms that won't fit/are accessed indirectly, so
1177 * it puts it in the batchbuffer.
1178 */
1179 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1180 pandecode_indent++;
1181
1182 for (int j = 0; j <= ubufs[i].size; j++) {
1183 for (int k = 0; k < 4; k++) {
1184 if (k == 0)
1185 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1186 else
1187 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1188
1189 }
1190
1191 pandecode_log_cont("\n");
1192 }
1193
1194 pandecode_indent--;
1195 pandecode_log("};\n");
1196 }
1197
1198 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1199 job_no);
1200 pandecode_indent++;
1201
1202 for (int i = 0; i < ubufs_count; i++) {
1203 pandecode_log("{\n");
1204 pandecode_indent++;
1205 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1206 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1207 pandecode_indent--;
1208 pandecode_log("},\n");
1209 }
1210
1211 pandecode_indent--;
1212 pandecode_log("};\n");
1213 }
1214
1215 static void
1216 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1217 {
1218
1219 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1220
1221 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1222
1223 if (scratchpad->zero)
1224 pandecode_msg("XXX scratchpad zero tripped");
1225
1226 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no, suffix);
1227 pandecode_indent++;
1228
1229 pandecode_prop("flags = 0x%x", scratchpad->flags);
1230 MEMORY_PROP(scratchpad, gpu_scratchpad);
1231
1232 pandecode_indent--;
1233 pandecode_log("};\n");
1234 }
1235
1236 static void
1237 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1238 bool is_bifrost)
1239 {
1240 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1241 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1242
1243 /* Compute maximum possible size */
1244 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1245
1246 /* Print some boilerplate to clearly denote the assembly (which doesn't
1247 * obey indentation rules), and actually do the disassembly! */
1248
1249 printf("\n\n");
1250
1251 if (is_bifrost) {
1252 disassemble_bifrost(code, sz, false);
1253 } else {
1254 disassemble_midgard(code, sz);
1255 }
1256
1257 printf("\n\n");
1258 }
1259
1260 static void
1261 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1262 int job_no, enum mali_job_type job_type,
1263 char *suffix, bool is_bifrost)
1264 {
1265 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1266 struct pandecode_mapped_memory *attr_mem;
1267
1268 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1269 * are the only things actually needed from the FBD, vertex/tiler jobs
1270 * no longer reference the FBD -- instead, this field points to some
1271 * info about the scratchpad.
1272 */
1273 if (is_bifrost)
1274 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1275 else if (p->framebuffer & MALI_MFBD)
1276 pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1277 else
1278 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1279
1280 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1281 int texture_count = 0, sampler_count = 0;
1282
1283 if (shader_meta_ptr) {
1284 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1285 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1286
1287 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no, suffix);
1288 pandecode_indent++;
1289
1290 /* Save for dumps */
1291 attribute_count = s->attribute_count;
1292 varying_count = s->varying_count;
1293 texture_count = s->texture_count;
1294 sampler_count = s->sampler_count;
1295
1296 if (is_bifrost) {
1297 uniform_count = s->bifrost2.uniform_count;
1298 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1299 } else {
1300 uniform_count = s->midgard1.uniform_count;
1301 /* TODO figure this out */
1302 uniform_buffer_count = 1;
1303 }
1304
1305 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1306
1307 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1308 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1309 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1310 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1311
1312 if (is_bifrost) {
1313 pandecode_log(".bifrost1 = {\n");
1314 pandecode_indent++;
1315
1316 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1317 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1318
1319 pandecode_indent--;
1320 pandecode_log("},\n");
1321 } else {
1322 pandecode_log(".midgard1 = {\n");
1323 pandecode_indent++;
1324
1325 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1326 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1327
1328 pandecode_log(".unknown1 = ");
1329 pandecode_log_decoded_flags(shader_unknown1_flag_info, s->midgard1.unknown1);
1330 pandecode_log_cont(",\n");
1331
1332 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1333
1334 pandecode_indent--;
1335 pandecode_log("},\n");
1336 }
1337
1338 if (s->depth_units || s->depth_factor) {
1339 if (is_bifrost)
1340 pandecode_prop("depth_units = %f", s->depth_units);
1341 else
1342 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f);
1343
1344 pandecode_prop("depth_factor = %f", s->depth_factor);
1345 }
1346
1347 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1348 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1349
1350 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1351 invert_alpha_coverage ? "~" : "",
1352 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1353
1354 pandecode_log(".unknown2_3 = ");
1355
1356 int unknown2_3 = s->unknown2_3;
1357 int unknown2_4 = s->unknown2_4;
1358
1359 /* We're not quite sure what these flags mean without the depth test, if anything */
1360
1361 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1362 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1363 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1364
1365 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1366 }
1367
1368 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1369 pandecode_log_cont(",\n");
1370
1371 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1372 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1373
1374 pandecode_log(".unknown2_4 = ");
1375 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1376 pandecode_log_cont(",\n");
1377
1378 pandecode_replay_stencil("front", &s->stencil_front);
1379 pandecode_replay_stencil("back", &s->stencil_back);
1380
1381 if (is_bifrost) {
1382 pandecode_log(".bifrost2 = {\n");
1383 pandecode_indent++;
1384
1385 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1386 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1387 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1388 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1389
1390 pandecode_indent--;
1391 pandecode_log("},\n");
1392 } else {
1393 pandecode_log(".midgard2 = {\n");
1394 pandecode_indent++;
1395
1396 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1397 pandecode_indent--;
1398 pandecode_log("},\n");
1399 }
1400
1401 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1402
1403 if (!is_bifrost) {
1404 /* TODO: Blend shaders routing/disasm */
1405
1406 pandecode_midgard_blend(&s->blend, false);
1407 }
1408
1409 pandecode_indent--;
1410 pandecode_log("};\n");
1411
1412 /* MRT blend fields are used whenever MFBD is used, with
1413 * per-RT descriptors */
1414
1415 if (job_type == JOB_TYPE_TILER) {
1416 void* blend_base = (void *) (s + 1);
1417
1418 for (unsigned i = 0; i < 4; i++) {
1419 mali_ptr shader = 0;
1420
1421 if (is_bifrost)
1422 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1423 else
1424 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1425
1426 if (shader)
1427 pandecode_shader_disassemble(shader, job_no, job_type, false);
1428 }
1429 }
1430
1431 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1432 } else
1433 pandecode_msg("<no shader>\n");
1434
1435 if (p->viewport) {
1436 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1437 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1438
1439 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1440 pandecode_indent++;
1441
1442 pandecode_prop("clip_minx = %f", f->clip_minx);
1443 pandecode_prop("clip_miny = %f", f->clip_miny);
1444 pandecode_prop("clip_minz = %f", f->clip_minz);
1445 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1446 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1447 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1448
1449 /* Only the higher coordinates are MALI_POSITIVE scaled */
1450
1451 pandecode_prop("viewport0 = { %d, %d }",
1452 f->viewport0[0], f->viewport0[1]);
1453
1454 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1455 f->viewport1[0] + 1, f->viewport1[1] + 1);
1456
1457 pandecode_indent--;
1458 pandecode_log("};\n");
1459 }
1460
1461 if (p->attribute_meta) {
1462 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1463
1464 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1465 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1466 }
1467
1468 /* Varyings are encoded like attributes but not actually sent; we just
1469 * pass a zero buffer with the right stride/size set, (or whatever)
1470 * since the GPU will write to it itself */
1471
1472 if (p->varyings) {
1473 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1474
1475 /* Number of descriptors depends on whether there are
1476 * non-internal varyings */
1477
1478 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1479 }
1480
1481 if (p->varying_meta) {
1482 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1483 }
1484
1485 if (p->uniforms) {
1486 int rows = uniform_count, width = 4;
1487 size_t sz = rows * width * sizeof(float);
1488
1489 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1490 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1491 float *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1492
1493 pandecode_log("float uniforms_%d%s[] = {\n", job_no, suffix);
1494
1495 pandecode_indent++;
1496
1497 for (int row = 0; row < rows; row++) {
1498 for (int i = 0; i < width; i++)
1499 pandecode_log_cont("%ff, ", uniforms[i]);
1500
1501 pandecode_log_cont("\n");
1502
1503 uniforms += width;
1504 }
1505
1506 pandecode_indent--;
1507 pandecode_log("};\n");
1508 }
1509
1510 if (p->uniform_buffers) {
1511 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1512 }
1513
1514 if (p->texture_trampoline) {
1515 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1516
1517 if (mmem) {
1518 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1519
1520 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1521 pandecode_indent++;
1522
1523 for (int tex = 0; tex < texture_count; ++tex) {
1524 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1525 char *a = pointer_as_memory_reference(*u);
1526 pandecode_log("%s,\n", a);
1527 free(a);
1528 }
1529
1530 pandecode_indent--;
1531 pandecode_log("};\n");
1532
1533 /* Now, finally, descend down into the texture descriptor */
1534 for (int tex = 0; tex < texture_count; ++tex) {
1535 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1536 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1537
1538 if (tmem) {
1539 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1540
1541 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no, tex);
1542 pandecode_indent++;
1543
1544 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1545 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1546 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1547 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1548 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1549 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1550 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1551
1552 struct mali_texture_format f = t->format;
1553
1554 pandecode_log(".format = {\n");
1555 pandecode_indent++;
1556
1557 pandecode_replay_swizzle(f.swizzle);
1558 pandecode_prop("format = %s", pandecode_format_name(f.format));
1559 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1560 pandecode_prop("srgb = %" PRId32, f.srgb);
1561 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1562 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1563
1564 pandecode_indent--;
1565 pandecode_log("},\n");
1566
1567 pandecode_replay_swizzle(t->swizzle);
1568
1569 if (t->swizzle_zero) {
1570 /* Shouldn't happen */
1571 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1572 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1573 }
1574
1575 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1576
1577 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1578 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1579 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1580
1581 pandecode_log(".payload = {\n");
1582 pandecode_indent++;
1583
1584 /* A bunch of bitmap pointers follow.
1585 * We work out the correct number,
1586 * based on the mipmap/cubemap
1587 * properties, but dump extra
1588 * possibilities to futureproof */
1589
1590 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1591 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1592
1593 /* Miptree for each face */
1594 if (f.type == MALI_TEX_CUBE)
1595 bitmap_count *= 6;
1596
1597 /* Array of textures */
1598 bitmap_count *= MALI_NEGATIVE(t->array_size);
1599
1600 /* Stride for each element */
1601 if (manual_stride)
1602 bitmap_count *= 2;
1603
1604 /* Sanity check the size */
1605 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1606 assert (bitmap_count <= max_count);
1607
1608 /* Dump more to be safe, but not _that_ much more */
1609 int safe_count = MIN2(bitmap_count * 2, max_count);
1610
1611 for (int i = 0; i < safe_count; ++i) {
1612 char *prefix = (i >= bitmap_count) ? "// " : "";
1613
1614 /* How we dump depends if this is a stride or a pointer */
1615
1616 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1617 /* signed 32-bit snuck in as a 64-bit pointer */
1618 uint64_t stride_set = t->payload[i];
1619 uint32_t clamped_stride = stride_set;
1620 int32_t stride = clamped_stride;
1621 assert(stride_set == clamped_stride);
1622 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1623 } else {
1624 char *a = pointer_as_memory_reference(t->payload[i]);
1625 pandecode_log("%s%s, \n", prefix, a);
1626 free(a);
1627 }
1628 }
1629
1630 pandecode_indent--;
1631 pandecode_log("},\n");
1632
1633 pandecode_indent--;
1634 pandecode_log("};\n");
1635 }
1636 }
1637 }
1638 }
1639
1640 if (p->sampler_descriptor) {
1641 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1642
1643 if (smem) {
1644 struct mali_sampler_descriptor *s;
1645
1646 mali_ptr d = p->sampler_descriptor;
1647
1648 for (int i = 0; i < sampler_count; ++i) {
1649 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1650
1651 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1652 pandecode_indent++;
1653
1654 /* Only the lower two bits are understood right now; the rest we display as hex */
1655 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1656 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1657 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1658 s->filter_mode & ~3);
1659
1660 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1661 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1662
1663 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1664 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1665 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1666
1667 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1668
1669 if (s->zero || s->zero2) {
1670 pandecode_msg("Zero tripped\n");
1671 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1672 }
1673
1674 pandecode_prop("unknown2 = %d", s->unknown2);
1675
1676 pandecode_prop("border_color = { %f, %f, %f, %f }",
1677 s->border_color[0],
1678 s->border_color[1],
1679 s->border_color[2],
1680 s->border_color[3]);
1681
1682 pandecode_indent--;
1683 pandecode_log("};\n");
1684 }
1685 }
1686 }
1687 }
1688
1689 static void
1690 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1691 {
1692 pandecode_log_cont("{\n");
1693 pandecode_indent++;
1694
1695 MEMORY_PROP(p, position_varying);
1696 MEMORY_COMMENT(p, position_varying);
1697 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1698 MEMORY_COMMENT(p, uniform_buffers);
1699 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1700 MEMORY_COMMENT(p, texture_trampoline);
1701 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1702 MEMORY_COMMENT(p, sampler_descriptor);
1703 DYN_MEMORY_PROP(p, job_no, uniforms);
1704 MEMORY_COMMENT(p, uniforms);
1705 DYN_MEMORY_PROP(p, job_no, attributes);
1706 MEMORY_COMMENT(p, attributes);
1707 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1708 MEMORY_COMMENT(p, attribute_meta);
1709 DYN_MEMORY_PROP(p, job_no, varyings);
1710 MEMORY_COMMENT(p, varyings);
1711 DYN_MEMORY_PROP(p, job_no, varying_meta);
1712 MEMORY_COMMENT(p, varying_meta);
1713 DYN_MEMORY_PROP(p, job_no, viewport);
1714 MEMORY_COMMENT(p, viewport);
1715 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1716 MEMORY_COMMENT(p, occlusion_counter);
1717 MEMORY_COMMENT(p, framebuffer & ~1);
1718 pandecode_msg("%" PRIx64 "\n", p->viewport);
1719 pandecode_msg("%" PRIx64 "\n", p->framebuffer);
1720
1721 if (is_bifrost)
1722 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1723 else
1724 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1725
1726 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1727 pandecode_prop("flags = %d", p->flags);
1728
1729 pandecode_indent--;
1730 pandecode_log("},\n");
1731 }
1732
1733 static void
1734 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1735 {
1736 pandecode_log_cont("{\n");
1737 pandecode_indent++;
1738
1739 pandecode_prop("unk2 = 0x%x", v->unk2);
1740
1741 if (v->zero0 || v->zero1) {
1742 pandecode_msg("vertex only zero tripped");
1743 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1744 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1745 }
1746
1747 pandecode_indent--;
1748 pandecode_log("}\n");
1749 }
1750
1751 static void
1752 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1753 {
1754
1755 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1756 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1757
1758 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1759 pandecode_indent++;
1760
1761 if (h->zero) {
1762 pandecode_msg("tiler heap zero tripped\n");
1763 pandecode_prop("zero = 0x%x", h->zero);
1764 }
1765
1766 for (int i = 0; i < 12; i++) {
1767 if (h->zeros[i] != 0) {
1768 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1769 i, h->zeros[i]);
1770 }
1771 }
1772
1773 pandecode_prop("heap_size = 0x%x", h->heap_size);
1774 MEMORY_PROP(h, tiler_heap_start);
1775 MEMORY_PROP(h, tiler_heap_free);
1776
1777 /* this might point to the beginning of another buffer, when it's
1778 * really the end of the tiler heap buffer, so we have to be careful
1779 * here.
1780 */
1781 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1782 pandecode_prop("tiler_heap_end = %s + 1", a);
1783 free(a);
1784
1785 pandecode_indent--;
1786 pandecode_log("};\n");
1787 }
1788
1789 static void
1790 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1791 {
1792 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1793 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1794
1795 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1796
1797 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
1798 pandecode_indent++;
1799
1800 if (t->zero0 || t->zero1) {
1801 pandecode_msg("tiler meta zero tripped");
1802 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1803 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1804 }
1805
1806 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
1807 pandecode_prop("flags = 0x%" PRIx16, t->flags);
1808
1809 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1810 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
1811 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
1812
1813 for (int i = 0; i < 12; i++) {
1814 if (t->zeros[i] != 0) {
1815 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
1816 i, t->zeros[i]);
1817 }
1818 }
1819
1820 pandecode_indent--;
1821 pandecode_log("};\n");
1822 }
1823
1824 static void
1825 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
1826 {
1827 pandecode_log(".gl_enables = ");
1828
1829 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
1830
1831 pandecode_log_cont(",\n");
1832 }
1833
1834 static void
1835 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
1836 {
1837 pandecode_log(".primitive_size = {\n");
1838 pandecode_indent++;
1839
1840 if (constant) {
1841 pandecode_prop("constant = %f", u.constant);
1842 } else {
1843 MEMORY_PROP((&u), pointer);
1844 }
1845
1846 pandecode_indent--;
1847 pandecode_log("},\n");
1848 }
1849
1850 static void
1851 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
1852 {
1853 pandecode_log_cont("{\n");
1854 pandecode_indent++;
1855
1856 /* TODO: gl_PointSize on Bifrost */
1857 pandecode_replay_primitive_size(t->primitive_size, true);
1858
1859 DYN_MEMORY_PROP(t, job_no, tiler_meta);
1860 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
1861
1862 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
1863 || t->zero6 || t->zero7 || t->zero8) {
1864 pandecode_msg("tiler only zero tripped");
1865 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1866 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
1867 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
1868 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
1869 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
1870 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
1871 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
1872 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
1873 }
1874
1875 pandecode_indent--;
1876 pandecode_log("},\n");
1877 }
1878
1879 static int
1880 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
1881 const struct pandecode_mapped_memory *mem,
1882 mali_ptr payload, int job_no)
1883 {
1884 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
1885
1886 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
1887
1888 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
1889 pandecode_indent++;
1890
1891 pandecode_log(".prefix = ");
1892 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1893
1894 pandecode_log(".vertex = ");
1895 pandecode_replay_vertex_only_bfr(&v->vertex);
1896
1897 pandecode_log(".postfix = ");
1898 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
1899
1900 pandecode_indent--;
1901 pandecode_log("};\n");
1902
1903 return sizeof(*v);
1904 }
1905
1906 static int
1907 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
1908 const struct pandecode_mapped_memory *mem,
1909 mali_ptr payload, int job_no)
1910 {
1911 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
1912
1913 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
1914
1915 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
1916 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
1917
1918 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
1919 pandecode_indent++;
1920
1921 pandecode_log(".prefix = ");
1922 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
1923
1924 pandecode_log(".tiler = ");
1925 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
1926
1927 pandecode_log(".postfix = ");
1928 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
1929
1930 pandecode_indent--;
1931 pandecode_log("};\n");
1932
1933 return sizeof(*t);
1934 }
1935
1936 static int
1937 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
1938 const struct pandecode_mapped_memory *mem,
1939 mali_ptr payload, int job_no)
1940 {
1941 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
1942
1943 char *a = pointer_as_memory_reference(payload);
1944 pandecode_msg("vt payload: %s\n", a);
1945 free(a);
1946
1947 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
1948
1949 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
1950
1951 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
1952 pandecode_indent++;
1953
1954 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
1955 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
1956
1957 pandecode_log(".prefix = ");
1958 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
1959
1960 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
1961 pandecode_prop("draw_start = %d", v->draw_start);
1962
1963 #ifndef __LP64__
1964
1965 if (v->zero3) {
1966 pandecode_msg("Zero tripped\n");
1967 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
1968 }
1969
1970 #endif
1971
1972 if (v->zero5) {
1973 pandecode_msg("Zero tripped\n");
1974 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
1975 }
1976
1977 pandecode_log(".postfix = ");
1978 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
1979
1980 pandecode_indent--;
1981 pandecode_log("};\n");
1982
1983 return sizeof(*v);
1984 }
1985
1986 static int
1987 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
1988 mali_ptr payload, int job_no,
1989 bool is_bifrost)
1990 {
1991 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
1992
1993 bool fbd_dumped = false;
1994
1995 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
1996 /* Only SFBDs are understood, not MFBDs. We're speculating,
1997 * based on the versioning, kernel code, etc, that the
1998 * difference is between Single FrameBuffer Descriptor and
1999 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2000 * up with multi-framebuffer support being added (T7xx onwards,
2001 * including Gxx). In any event, there's some field shuffling
2002 * that we haven't looked into yet. */
2003
2004 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
2005 fbd_dumped = true;
2006 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2007 /* We don't know if Bifrost supports SFBD's at all, since the
2008 * driver never uses them. And the format is different from
2009 * Midgard anyways, due to the tiler heap and scratchpad being
2010 * moved out into separate structures, so it's not clear what a
2011 * Bifrost SFBD would even look like without getting an actual
2012 * trace, which appears impossible.
2013 */
2014
2015 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2016 fbd_dumped = true;
2017 }
2018
2019 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2020
2021 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no);
2022 pandecode_indent++;
2023
2024 /* See the comments by the macro definitions for mathematical context
2025 * on why this is so weird */
2026
2027 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2028 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2029
2030 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2031 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2032 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2033
2034 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2035 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2036 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2037
2038 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2039 * we have to fallback on offsets. */
2040
2041 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2042
2043 if (fbd_dumped)
2044 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2045 else
2046 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2047
2048 pandecode_indent--;
2049 pandecode_log("};\n");
2050
2051 return sizeof(*s);
2052 }
2053
2054 static int job_descriptor_number = 0;
2055
2056 int
2057 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
2058 {
2059 struct mali_job_descriptor_header *h;
2060
2061 int start_number = 0;
2062
2063 bool first = true;
2064 bool last_size;
2065
2066 do {
2067 struct pandecode_mapped_memory *mem =
2068 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2069
2070 void *payload;
2071
2072 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2073
2074 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2075 * high 32-bits of the 64-bit pointer are reused to store
2076 * something else.
2077 */
2078 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2079 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2080 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2081
2082 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2083 MALI_PAYLOAD_SIZE);
2084
2085 int job_no = job_descriptor_number++;
2086
2087 if (first)
2088 start_number = job_no;
2089
2090 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no);
2091 pandecode_indent++;
2092
2093 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
2094
2095 /* Save for next job fixing */
2096 last_size = h->job_descriptor_size;
2097
2098 if (h->job_descriptor_size)
2099 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2100
2101 if (h->exception_status)
2102 pandecode_prop("exception_status = %d", h->exception_status);
2103
2104 if (h->first_incomplete_task)
2105 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2106
2107 if (h->fault_pointer)
2108 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2109
2110 if (h->job_barrier)
2111 pandecode_prop("job_barrier = %d", h->job_barrier);
2112
2113 pandecode_prop("job_index = %d", h->job_index);
2114
2115 if (h->unknown_flags)
2116 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2117
2118 if (h->job_dependency_index_1)
2119 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2120
2121 if (h->job_dependency_index_2)
2122 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2123
2124 pandecode_indent--;
2125 pandecode_log("};\n");
2126
2127 /* Do not touch the field yet -- decode the payload first, and
2128 * don't touch that either. This is essential for the uploads
2129 * to occur in sequence and therefore be dynamically allocated
2130 * correctly. Do note the size, however, for that related
2131 * reason. */
2132
2133 switch (h->job_type) {
2134 case JOB_TYPE_SET_VALUE: {
2135 struct mali_payload_set_value *s = payload;
2136
2137 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no);
2138 pandecode_indent++;
2139 MEMORY_PROP(s, out);
2140 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2141 pandecode_indent--;
2142 pandecode_log("};\n");
2143
2144 break;
2145 }
2146
2147 case JOB_TYPE_TILER:
2148 case JOB_TYPE_VERTEX:
2149 case JOB_TYPE_COMPUTE:
2150 if (bifrost) {
2151 if (h->job_type == JOB_TYPE_TILER)
2152 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2153 else
2154 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2155 } else
2156 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2157
2158 break;
2159
2160 case JOB_TYPE_FRAGMENT:
2161 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2162 break;
2163
2164 default:
2165 break;
2166 }
2167
2168 /* Handle linkage */
2169
2170 if (!first) {
2171 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2172
2173 if (last_size)
2174 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2175 else
2176 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2177 }
2178
2179 first = false;
2180
2181 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2182
2183 return start_number;
2184 }