pan/decode: Don't print zero exception_status
[mesa.git] / src / panfrost / pandecode / decode.c
1 /*
2 * Copyright (C) 2017-2019 Alyssa Rosenzweig
3 * Copyright (C) 2017-2019 Connor Abbott
4 * Copyright (C) 2019 Collabora, Ltd.
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/macros.h"
34 #include "util/u_math.h"
35
36 #include "pan_pretty_print.h"
37 #include "midgard/disassemble.h"
38 #include "bifrost/disassemble.h"
39
40 #include "pan_encoder.h"
41
42 int pandecode_jc(mali_ptr jc_gpu_va, bool bifrost);
43
44 #define MEMORY_PROP(obj, p) {\
45 if (obj->p) { \
46 char *a = pointer_as_memory_reference(obj->p); \
47 pandecode_prop("%s = %s", #p, a); \
48 free(a); \
49 } \
50 }
51
52 #define MEMORY_PROP_DIR(obj, p) {\
53 if (obj.p) { \
54 char *a = pointer_as_memory_reference(obj.p); \
55 pandecode_prop("%s = %s", #p, a); \
56 free(a); \
57 } \
58 }
59
60 /* Semantic logging type.
61 *
62 * Raw: for raw messages to be printed as is.
63 * Message: for helpful information to be commented out in replays.
64 * Property: for properties of a struct
65 *
66 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
67 */
68
69 enum pandecode_log_type {
70 PANDECODE_RAW,
71 PANDECODE_MESSAGE,
72 PANDECODE_PROPERTY
73 };
74
75 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
76 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
77 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
78
79 unsigned pandecode_indent = 0;
80
81 static void
82 pandecode_make_indent(void)
83 {
84 for (unsigned i = 0; i < pandecode_indent; ++i)
85 printf(" ");
86 }
87
88 static void
89 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
90 {
91 va_list ap;
92
93 pandecode_make_indent();
94
95 if (type == PANDECODE_MESSAGE)
96 printf("// ");
97 else if (type == PANDECODE_PROPERTY)
98 printf(".");
99
100 va_start(ap, format);
101 vprintf(format, ap);
102 va_end(ap);
103
104 if (type == PANDECODE_PROPERTY)
105 printf(",\n");
106 }
107
108 static void
109 pandecode_log_cont(const char *format, ...)
110 {
111 va_list ap;
112
113 va_start(ap, format);
114 vprintf(format, ap);
115 va_end(ap);
116 }
117
118 struct pandecode_flag_info {
119 u64 flag;
120 const char *name;
121 };
122
123 static void
124 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
125 u64 flags)
126 {
127 bool decodable_flags_found = false;
128
129 for (int i = 0; flag_info[i].name; i++) {
130 if ((flags & flag_info[i].flag) != flag_info[i].flag)
131 continue;
132
133 if (!decodable_flags_found) {
134 decodable_flags_found = true;
135 } else {
136 pandecode_log_cont(" | ");
137 }
138
139 pandecode_log_cont("%s", flag_info[i].name);
140
141 flags &= ~flag_info[i].flag;
142 }
143
144 if (decodable_flags_found) {
145 if (flags)
146 pandecode_log_cont(" | 0x%" PRIx64, flags);
147 } else {
148 pandecode_log_cont("0x%" PRIx64, flags);
149 }
150 }
151
152 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
153 static const struct pandecode_flag_info gl_enable_flag_info[] = {
154 FLAG_INFO(OCCLUSION_QUERY),
155 FLAG_INFO(OCCLUSION_PRECISE),
156 FLAG_INFO(FRONT_CCW_TOP),
157 FLAG_INFO(CULL_FACE_FRONT),
158 FLAG_INFO(CULL_FACE_BACK),
159 {}
160 };
161 #undef FLAG_INFO
162
163 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
164 static const struct pandecode_flag_info clear_flag_info[] = {
165 FLAG_INFO(FAST),
166 FLAG_INFO(SLOW),
167 FLAG_INFO(SLOW_STENCIL),
168 {}
169 };
170 #undef FLAG_INFO
171
172 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
173 static const struct pandecode_flag_info mask_flag_info[] = {
174 FLAG_INFO(R),
175 FLAG_INFO(G),
176 FLAG_INFO(B),
177 FLAG_INFO(A),
178 {}
179 };
180 #undef FLAG_INFO
181
182 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
183 static const struct pandecode_flag_info u3_flag_info[] = {
184 FLAG_INFO(HAS_MSAA),
185 FLAG_INFO(CAN_DISCARD),
186 FLAG_INFO(HAS_BLEND_SHADER),
187 FLAG_INFO(DEPTH_TEST),
188 {}
189 };
190
191 static const struct pandecode_flag_info u4_flag_info[] = {
192 FLAG_INFO(NO_MSAA),
193 FLAG_INFO(NO_DITHER),
194 FLAG_INFO(DEPTH_RANGE_A),
195 FLAG_INFO(DEPTH_RANGE_B),
196 FLAG_INFO(STENCIL_TEST),
197 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
198 {}
199 };
200 #undef FLAG_INFO
201
202 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
203 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
204 FLAG_INFO(MSAA_A),
205 FLAG_INFO(MSAA_B),
206 FLAG_INFO(MSAA_8),
207 {}
208 };
209 #undef FLAG_INFO
210
211 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
212 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
213 FLAG_INFO(MSAA),
214 FLAG_INFO(SRGB),
215 {}
216 };
217 #undef FLAG_INFO
218
219 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
220 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
221 FLAG_INFO(PRESENT),
222 FLAG_INFO(AFBC),
223 FLAG_INFO(ZS),
224 {}
225 };
226 #undef FLAG_INFO
227
228 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
229 static const struct pandecode_flag_info shader_midgard1_flag_info [] = {
230 FLAG_INFO(EARLY_Z),
231 FLAG_INFO(HELPER_INVOCATIONS),
232 FLAG_INFO(READS_TILEBUFFER),
233 FLAG_INFO(READS_ZS),
234 {}
235 };
236 #undef FLAG_INFO
237
238 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
239 static const struct pandecode_flag_info mfbd_flag_info [] = {
240 FLAG_INFO(DEPTH_WRITE),
241 FLAG_INFO(EXTRA),
242 {}
243 };
244 #undef FLAG_INFO
245
246 #define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
247 static const struct pandecode_flag_info sampler_flag_info [] = {
248 FLAG_INFO(MAG_NEAREST),
249 FLAG_INFO(MIN_NEAREST),
250 FLAG_INFO(MIP_LINEAR_1),
251 FLAG_INFO(MIP_LINEAR_2),
252 FLAG_INFO(NORM_COORDS),
253 {}
254 };
255 #undef FLAG_INFO
256
257 extern char *replace_fragment;
258 extern char *replace_vertex;
259
260 static char *
261 pandecode_job_type(enum mali_job_type type)
262 {
263 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
264
265 switch (type) {
266 DEFINE_CASE(NULL);
267 DEFINE_CASE(SET_VALUE);
268 DEFINE_CASE(CACHE_FLUSH);
269 DEFINE_CASE(COMPUTE);
270 DEFINE_CASE(VERTEX);
271 DEFINE_CASE(TILER);
272 DEFINE_CASE(FUSED);
273 DEFINE_CASE(FRAGMENT);
274
275 case JOB_NOT_STARTED:
276 return "NOT_STARTED";
277
278 default:
279 pandecode_log("Warning! Unknown job type %x\n", type);
280 return "!?!?!?";
281 }
282
283 #undef DEFINE_CASE
284 }
285
286 static char *
287 pandecode_draw_mode(enum mali_draw_mode mode)
288 {
289 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
290
291 switch (mode) {
292 DEFINE_CASE(DRAW_NONE);
293 DEFINE_CASE(POINTS);
294 DEFINE_CASE(LINES);
295 DEFINE_CASE(TRIANGLES);
296 DEFINE_CASE(TRIANGLE_STRIP);
297 DEFINE_CASE(TRIANGLE_FAN);
298 DEFINE_CASE(LINE_STRIP);
299 DEFINE_CASE(LINE_LOOP);
300 DEFINE_CASE(POLYGON);
301 DEFINE_CASE(QUADS);
302 DEFINE_CASE(QUAD_STRIP);
303
304 default:
305 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
306 }
307
308 #undef DEFINE_CASE
309 }
310
311 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
312 static char *
313 pandecode_func(enum mali_func mode)
314 {
315 switch (mode) {
316 DEFINE_CASE(NEVER);
317 DEFINE_CASE(LESS);
318 DEFINE_CASE(EQUAL);
319 DEFINE_CASE(LEQUAL);
320 DEFINE_CASE(GREATER);
321 DEFINE_CASE(NOTEQUAL);
322 DEFINE_CASE(GEQUAL);
323 DEFINE_CASE(ALWAYS);
324
325 default:
326 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
327 }
328 }
329 #undef DEFINE_CASE
330
331 /* Why is this duplicated? Who knows... */
332 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
333 static char *
334 pandecode_alt_func(enum mali_alt_func mode)
335 {
336 switch (mode) {
337 DEFINE_CASE(NEVER);
338 DEFINE_CASE(LESS);
339 DEFINE_CASE(EQUAL);
340 DEFINE_CASE(LEQUAL);
341 DEFINE_CASE(GREATER);
342 DEFINE_CASE(NOTEQUAL);
343 DEFINE_CASE(GEQUAL);
344 DEFINE_CASE(ALWAYS);
345
346 default:
347 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
348 }
349 }
350 #undef DEFINE_CASE
351
352 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
353 static char *
354 pandecode_stencil_op(enum mali_stencil_op op)
355 {
356 switch (op) {
357 DEFINE_CASE(KEEP);
358 DEFINE_CASE(REPLACE);
359 DEFINE_CASE(ZERO);
360 DEFINE_CASE(INVERT);
361 DEFINE_CASE(INCR_WRAP);
362 DEFINE_CASE(DECR_WRAP);
363 DEFINE_CASE(INCR);
364 DEFINE_CASE(DECR);
365
366 default:
367 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
368 }
369 }
370
371 #undef DEFINE_CASE
372
373 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
374 static char *pandecode_attr_mode(enum mali_attr_mode mode)
375 {
376 switch(mode) {
377 DEFINE_CASE(UNUSED);
378 DEFINE_CASE(LINEAR);
379 DEFINE_CASE(POT_DIVIDE);
380 DEFINE_CASE(MODULO);
381 DEFINE_CASE(NPOT_DIVIDE);
382 DEFINE_CASE(IMAGE);
383 DEFINE_CASE(INTERNAL);
384 default:
385 return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
386 }
387 }
388
389 #undef DEFINE_CASE
390
391 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
392 static char *
393 pandecode_channel(enum mali_channel channel)
394 {
395 switch (channel) {
396 DEFINE_CASE(RED);
397 DEFINE_CASE(GREEN);
398 DEFINE_CASE(BLUE);
399 DEFINE_CASE(ALPHA);
400 DEFINE_CASE(ZERO);
401 DEFINE_CASE(ONE);
402 DEFINE_CASE(RESERVED_0);
403 DEFINE_CASE(RESERVED_1);
404
405 default:
406 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
407 }
408 }
409 #undef DEFINE_CASE
410
411 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
412 static char *
413 pandecode_wrap_mode(enum mali_wrap_mode op)
414 {
415 switch (op) {
416 DEFINE_CASE(REPEAT);
417 DEFINE_CASE(CLAMP_TO_EDGE);
418 DEFINE_CASE(CLAMP_TO_BORDER);
419 DEFINE_CASE(MIRRORED_REPEAT);
420
421 default:
422 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
423 }
424 }
425 #undef DEFINE_CASE
426
427 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
428 static char *
429 pandecode_texture_type(enum mali_texture_type type)
430 {
431 switch (type) {
432 DEFINE_CASE(1D);
433 DEFINE_CASE(2D);
434 DEFINE_CASE(3D);
435 DEFINE_CASE(CUBE);
436
437 default:
438 unreachable("Unknown case");
439 }
440 }
441 #undef DEFINE_CASE
442
443 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
444 static char *
445 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
446 {
447 switch (fmt) {
448 DEFINE_CASE(TILED);
449 DEFINE_CASE(UNKNOWN);
450 DEFINE_CASE(LINEAR);
451 DEFINE_CASE(AFBC);
452
453 default:
454 unreachable("Invalid case");
455 }
456 }
457 #undef DEFINE_CASE
458
459 #define DEFINE_CASE(name) case MALI_EXCEPTION_ACCESS_## name: return ""#name
460 static char *
461 pandecode_exception_access(enum mali_exception_access fmt)
462 {
463 switch (fmt) {
464 DEFINE_CASE(NONE);
465 DEFINE_CASE(EXECUTE);
466 DEFINE_CASE(READ);
467 DEFINE_CASE(WRITE);
468
469 default:
470 unreachable("Invalid case");
471 }
472 }
473 #undef DEFINE_CASE
474
475 /* Midgard's tiler descriptor is embedded within the
476 * larger FBD */
477
478 static void
479 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor *t)
480 {
481 pandecode_log(".tiler = {\n");
482 pandecode_indent++;
483
484 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
485 pandecode_prop("flags = 0x%" PRIx16, t->flags);
486 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
487
488 MEMORY_PROP(t, polygon_list);
489 MEMORY_PROP(t, polygon_list_body);
490
491 MEMORY_PROP(t, heap_start);
492
493 if (t->heap_start == t->heap_end) {
494 /* Print identically to show symmetry for empty tiler heaps */
495 MEMORY_PROP(t, heap_end);
496 } else {
497 /* Points to the end of a buffer */
498 char *a = pointer_as_memory_reference(t->heap_end - 1);
499 pandecode_prop("heap_end = %s + 1", a);
500 free(a);
501 }
502
503 bool nonzero_weights = false;
504
505 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
506 nonzero_weights |= t->weights[w] != 0x0;
507 }
508
509 if (nonzero_weights) {
510 pandecode_log(".weights = {");
511
512 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
513 pandecode_log("%d, ", t->weights[w]);
514 }
515
516 pandecode_log("},");
517 }
518
519 pandecode_indent--;
520 pandecode_log("}\n");
521 }
522
523 static void
524 pandecode_sfbd(uint64_t gpu_va, int job_no)
525 {
526 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
527 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
528
529 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
530 pandecode_indent++;
531
532 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
533 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
534
535 pandecode_log(".format = ");
536 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
537 pandecode_log_cont(",\n");
538
539 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
540 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
541
542 MEMORY_PROP(s, framebuffer);
543 pandecode_prop("stride = %d", s->stride);
544
545 /* Earlier in the actual commandstream -- right before width -- but we
546 * delay to flow nicer */
547
548 pandecode_log(".clear_flags = ");
549 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
550 pandecode_log_cont(",\n");
551
552 if (s->depth_buffer | s->depth_buffer_enable) {
553 MEMORY_PROP(s, depth_buffer);
554 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
555 }
556
557 if (s->stencil_buffer | s->stencil_buffer_enable) {
558 MEMORY_PROP(s, stencil_buffer);
559 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
560 }
561
562 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
563 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
564 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
565 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
566 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
567 }
568
569 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
570 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
571 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
572 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
573 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
574 }
575
576 if (s->clear_stencil) {
577 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
578 }
579
580 MEMORY_PROP(s, unknown_address_0);
581 const struct midgard_tiler_descriptor t = s->tiler;
582 pandecode_midgard_tiler_descriptor(&t);
583
584 pandecode_indent--;
585 pandecode_log("};\n");
586
587 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
588 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
589 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
590 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
591
592 printf(".zero3 = {");
593
594 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
595 printf("%X, ", s->zero3[i]);
596
597 printf("},\n");
598
599 printf(".zero6 = {");
600
601 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
602 printf("%X, ", s->zero6[i]);
603
604 printf("},\n");
605 }
606
607 static void
608 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
609 {
610 pandecode_log(".unknown%d = {", name);
611
612 for (int i = 0; i < count; ++i)
613 printf("%X, ", slide[i]);
614
615 pandecode_log("},\n");
616 }
617
618 #define SHORT_SLIDE(num) \
619 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
620
621 static void
622 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
623 {
624 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
625 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
626
627 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
628 pandecode_indent++;
629
630 SHORT_SLIDE(1);
631
632 pandecode_indent--;
633 printf("},\n");
634 }
635
636 static void
637 pandecode_swizzle(unsigned swizzle)
638 {
639 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
640 pandecode_channel((swizzle >> 0) & 0x7),
641 pandecode_channel((swizzle >> 3) & 0x7),
642 pandecode_channel((swizzle >> 6) & 0x7),
643 pandecode_channel((swizzle >> 9) & 0x7));
644 }
645
646 static void
647 pandecode_rt_format(struct mali_rt_format format)
648 {
649 pandecode_log(".format = {\n");
650 pandecode_indent++;
651
652 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
653 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
654 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
655
656 pandecode_prop("block = %s",
657 pandecode_mfbd_block_format(format.block));
658
659 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
660 MALI_NEGATIVE(format.nr_channels));
661
662 pandecode_log(".flags = ");
663 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
664 pandecode_log_cont(",\n");
665
666 pandecode_swizzle(format.swizzle);
667
668 pandecode_prop("no_preload = 0x%" PRIx32, format.no_preload);
669
670 if (format.zero)
671 pandecode_prop("zero = 0x%" PRIx32, format.zero);
672
673 pandecode_indent--;
674 pandecode_log("},\n");
675 }
676
677 static void
678 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
679 {
680 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
681 pandecode_indent++;
682
683 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
684 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
685 struct pandecode_mapped_memory *mem =
686 pandecode_find_mapped_gpu_mem_containing(rt_va);
687 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
688
689 pandecode_log("{\n");
690 pandecode_indent++;
691
692 pandecode_rt_format(rt->format);
693
694 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
695 pandecode_log(".afbc = {\n");
696 pandecode_indent++;
697
698 char *a = pointer_as_memory_reference(rt->afbc.metadata);
699 pandecode_prop("metadata = %s", a);
700 free(a);
701
702 pandecode_prop("stride = %d", rt->afbc.stride);
703 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
704
705 pandecode_indent--;
706 pandecode_log("},\n");
707 } else {
708 pandecode_log(".chunknown = {\n");
709 pandecode_indent++;
710
711 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
712
713 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
714 pandecode_prop("pointer = %s", a);
715 free(a);
716
717 pandecode_indent--;
718 pandecode_log("},\n");
719 }
720
721 MEMORY_PROP(rt, framebuffer);
722 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
723
724 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
725 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
726 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
727 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
728 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
729 }
730
731 if (rt->zero1 || rt->zero2 || rt->zero3) {
732 pandecode_msg("render target zeros tripped\n");
733 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
734 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
735 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
736 }
737
738 pandecode_indent--;
739 pandecode_log("},\n");
740 }
741
742 pandecode_indent--;
743 pandecode_log("};\n");
744 }
745
746 static unsigned
747 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
748 {
749 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
750 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
751
752 if (fb->sample_locations) {
753 /* The blob stores all possible sample locations in a single buffer
754 * allocated on startup, and just switches the pointer when switching
755 * MSAA state. For now, we just put the data into the cmdstream, but we
756 * should do something like what the blob does with a real driver.
757 *
758 * There seem to be 32 slots for sample locations, followed by another
759 * 16. The second 16 is just the center location followed by 15 zeros
760 * in all the cases I've identified (maybe shader vs. depth/color
761 * samples?).
762 */
763
764 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
765
766 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
767
768 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
769 pandecode_indent++;
770
771 for (int i = 0; i < 32 + 16; i++) {
772 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
773 }
774
775 pandecode_indent--;
776 pandecode_log("};\n");
777 }
778
779 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
780 pandecode_indent++;
781
782 pandecode_prop("unk0 = 0x%x", fb->unk0);
783
784 if (fb->sample_locations)
785 pandecode_prop("sample_locations = sample_locations_%d", job_no);
786
787 /* Assume that unknown1 was emitted in the last job for
788 * now */
789 MEMORY_PROP(fb, unknown1);
790
791 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
792 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
793 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
794 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
795
796 pandecode_prop("unk1 = 0x%x", fb->unk1);
797 pandecode_prop("unk2 = 0x%x", fb->unk2);
798 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
799 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
800
801 pandecode_log(".mfbd_flags = ");
802 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
803 pandecode_log_cont(",\n");
804
805 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
806 pandecode_prop("clear_depth = %f", fb->clear_depth);
807
808 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
809 MEMORY_PROP(fb, scratchpad);
810 const struct midgard_tiler_descriptor t = fb->tiler;
811 pandecode_midgard_tiler_descriptor(&t);
812
813 if (fb->zero3 || fb->zero4) {
814 pandecode_msg("framebuffer zeros tripped\n");
815 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
816 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
817 }
818
819 pandecode_indent--;
820 pandecode_log("};\n");
821
822 gpu_va += sizeof(struct bifrost_framebuffer);
823
824 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
825 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
826 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
827
828 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
829 pandecode_indent++;
830
831 MEMORY_PROP(fbx, checksum);
832
833 if (fbx->checksum_stride)
834 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
835
836 pandecode_log(".flags = ");
837 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
838 pandecode_log_cont(",\n");
839
840 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
841 pandecode_log(".ds_afbc = {\n");
842 pandecode_indent++;
843
844 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
845 pandecode_prop("depth_stencil_afbc_stride = %d",
846 fbx->ds_afbc.depth_stencil_afbc_stride);
847 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
848
849 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
850 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
851 pandecode_prop("zero1 = 0x%" PRIx32,
852 fbx->ds_afbc.zero1);
853 pandecode_prop("padding = 0x%" PRIx64,
854 fbx->ds_afbc.padding);
855 }
856
857 pandecode_indent--;
858 pandecode_log("},\n");
859 } else {
860 pandecode_log(".ds_linear = {\n");
861 pandecode_indent++;
862
863 if (fbx->ds_linear.depth) {
864 MEMORY_PROP_DIR(fbx->ds_linear, depth);
865 pandecode_prop("depth_stride = %d",
866 fbx->ds_linear.depth_stride);
867 }
868
869 if (fbx->ds_linear.stencil) {
870 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
871 pandecode_prop("stencil_stride = %d",
872 fbx->ds_linear.stencil_stride);
873 }
874
875 if (fbx->ds_linear.depth_stride_zero ||
876 fbx->ds_linear.stencil_stride_zero ||
877 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
878 pandecode_msg("Depth/stencil zeros tripped\n");
879 pandecode_prop("depth_stride_zero = 0x%x",
880 fbx->ds_linear.depth_stride_zero);
881 pandecode_prop("stencil_stride_zero = 0x%x",
882 fbx->ds_linear.stencil_stride_zero);
883 pandecode_prop("zero1 = 0x%" PRIx32,
884 fbx->ds_linear.zero1);
885 pandecode_prop("zero2 = 0x%" PRIx32,
886 fbx->ds_linear.zero2);
887 }
888
889 pandecode_indent--;
890 pandecode_log("},\n");
891 }
892
893 if (fbx->zero3 || fbx->zero4) {
894 pandecode_msg("fb_extra zeros tripped\n");
895 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
896 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
897 }
898
899 pandecode_indent--;
900 pandecode_log("};\n");
901
902 gpu_va += sizeof(struct bifrost_fb_extra);
903 }
904
905 if (with_render_targets)
906 pandecode_render_target(gpu_va, job_no, fb);
907
908 /* Passback the render target count */
909 return MALI_NEGATIVE(fb->rt_count_1);
910 }
911
912 /* Just add a comment decoding the shift/odd fields forming the padded vertices
913 * count */
914
915 static void
916 pandecode_padded_vertices(unsigned shift, unsigned k)
917 {
918 unsigned odd = 2*k + 1;
919 unsigned pot = 1 << shift;
920 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
921 }
922
923 /* Given a magic divisor, recover what we were trying to divide by.
924 *
925 * Let m represent the magic divisor. By definition, m is an element on Z, whre
926 * 0 <= m < 2^N, for N bits in m.
927 *
928 * Let q represent the number we would like to divide by.
929 *
930 * By definition of a magic divisor for N-bit unsigned integers (a number you
931 * multiply by to magically get division), m is a number such that:
932 *
933 * (m * x) & (2^N - 1) = floor(x/q).
934 * for all x on Z where 0 <= x < 2^N
935 *
936 * Ignore the case where any of the above values equals zero; it is irrelevant
937 * for our purposes (instanced arrays).
938 *
939 * Choose x = q. Then:
940 *
941 * (m * x) & (2^N - 1) = floor(x/q).
942 * (m * q) & (2^N - 1) = floor(q/q).
943 *
944 * floor(q/q) = floor(1) = 1, therefore:
945 *
946 * (m * q) & (2^N - 1) = 1
947 *
948 * Recall the identity that the bitwise AND of one less than a power-of-two
949 * equals the modulo with that power of two, i.e. for all x:
950 *
951 * x & (2^N - 1) = x % N
952 *
953 * Therefore:
954 *
955 * mq % (2^N) = 1
956 *
957 * By definition, a modular multiplicative inverse of a number m is the number
958 * q such that with respect to a modulos M:
959 *
960 * mq % M = 1
961 *
962 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
963 *
964 */
965
966 static void
967 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
968 {
969 #if 0
970 /* Compute the modular inverse of `magic` with respect to 2^(32 -
971 * shift) the most lame way possible... just repeatedly add.
972 * Asymptoptically slow but nobody cares in practice, unless you have
973 * massive numbers of vertices or high divisors. */
974
975 unsigned inverse = 0;
976
977 /* Magic implicitly has the highest bit set */
978 magic |= (1 << 31);
979
980 /* Depending on rounding direction */
981 if (extra)
982 magic++;
983
984 for (;;) {
985 uint32_t product = magic * inverse;
986
987 if (shift) {
988 product >>= shift;
989 }
990
991 if (product == 1)
992 break;
993
994 ++inverse;
995 }
996
997 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
998
999 /* Recall we're supposed to divide by (gl_level_divisor *
1000 * padded_num_vertices) */
1001
1002 unsigned padded_num_vertices = inverse / orig_divisor;
1003
1004 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
1005 #endif
1006 }
1007
1008 static void
1009 pandecode_attributes(const struct pandecode_mapped_memory *mem,
1010 mali_ptr addr, int job_no, char *suffix,
1011 int count, bool varying)
1012 {
1013 char *prefix = varying ? "varyings" : "attributes";
1014
1015 if (!addr) {
1016 pandecode_msg("no %s\n", prefix);
1017 return;
1018 }
1019
1020 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
1021
1022 char base[128];
1023 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
1024
1025 for (int i = 0; i < count; ++i) {
1026 enum mali_attr_mode mode = attr[i].elements & 7;
1027
1028 if (mode == MALI_ATTR_UNUSED)
1029 continue;
1030
1031 mali_ptr raw_elements = attr[i].elements & ~7;
1032
1033 /* TODO: Do we maybe want to dump the attribute values
1034 * themselves given the specified format? Or is that too hard?
1035 * */
1036
1037 char *a = pointer_as_memory_reference(raw_elements);
1038 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
1039 free(a);
1040 }
1041
1042 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1043 pandecode_indent++;
1044
1045 for (int i = 0; i < count; ++i) {
1046 pandecode_log("{\n");
1047 pandecode_indent++;
1048
1049 unsigned mode = attr[i].elements & 7;
1050 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode(mode));
1051 pandecode_prop("shift = %d", attr[i].shift);
1052 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1053 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1054 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1055
1056 /* Decode further where possible */
1057
1058 if (mode == MALI_ATTR_MODULO) {
1059 pandecode_padded_vertices(
1060 attr[i].shift,
1061 attr[i].extra_flags);
1062 }
1063
1064 pandecode_indent--;
1065 pandecode_log("}, \n");
1066
1067 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1068 i++;
1069 pandecode_log("{\n");
1070 pandecode_indent++;
1071 pandecode_prop("unk = 0x%x", attr[i].unk);
1072 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1073 if (attr[i].zero != 0)
1074 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1075 pandecode_prop("divisor = %d", attr[i].divisor);
1076 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1077 pandecode_indent--;
1078 pandecode_log("}, \n");
1079 }
1080
1081 }
1082
1083 pandecode_indent--;
1084 pandecode_log("};\n");
1085 }
1086
1087 static mali_ptr
1088 pandecode_shader_address(const char *name, mali_ptr ptr)
1089 {
1090 /* TODO: Decode flags */
1091 mali_ptr shader_ptr = ptr & ~15;
1092
1093 char *a = pointer_as_memory_reference(shader_ptr);
1094 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1095 free(a);
1096
1097 return shader_ptr;
1098 }
1099
1100 static bool
1101 all_zero(unsigned *buffer, unsigned count)
1102 {
1103 for (unsigned i = 0; i < count; ++i) {
1104 if (buffer[i])
1105 return false;
1106 }
1107
1108 return true;
1109 }
1110
1111 static void
1112 pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
1113 {
1114 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1115 return;
1116
1117 const char *func = pandecode_func(stencil->func);
1118 const char *sfail = pandecode_stencil_op(stencil->sfail);
1119 const char *dpfail = pandecode_stencil_op(stencil->dpfail);
1120 const char *dppass = pandecode_stencil_op(stencil->dppass);
1121
1122 if (stencil->zero)
1123 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1124
1125 pandecode_log(".stencil_%s = {\n", name);
1126 pandecode_indent++;
1127 pandecode_prop("ref = %d", stencil->ref);
1128 pandecode_prop("mask = 0x%02X", stencil->mask);
1129 pandecode_prop("func = %s", func);
1130 pandecode_prop("sfail = %s", sfail);
1131 pandecode_prop("dpfail = %s", dpfail);
1132 pandecode_prop("dppass = %s", dppass);
1133 pandecode_indent--;
1134 pandecode_log("},\n");
1135 }
1136
1137 static void
1138 pandecode_blend_equation(const struct mali_blend_equation *blend)
1139 {
1140 if (blend->zero1)
1141 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1142
1143 pandecode_log(".equation = {\n");
1144 pandecode_indent++;
1145
1146 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1147 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1148
1149 pandecode_log(".color_mask = ");
1150 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1151 pandecode_log_cont(",\n");
1152
1153 pandecode_indent--;
1154 pandecode_log("},\n");
1155 }
1156
1157 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1158
1159 static unsigned
1160 decode_bifrost_constant(u16 constant)
1161 {
1162 float lo = (float) (constant & 0xFF);
1163 float hi = (float) (constant >> 8);
1164
1165 return (hi / 255.0) + (lo / 65535.0);
1166 }
1167
1168 static mali_ptr
1169 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1170 {
1171 struct bifrost_blend_rt *b =
1172 ((struct bifrost_blend_rt *) descs) + rt_no;
1173
1174 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1175 pandecode_indent++;
1176
1177 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1178 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1179 b->constant, decode_bifrost_constant(b->constant));
1180
1181 /* TODO figure out blend shader enable bit */
1182 pandecode_blend_equation(&b->equation);
1183 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1184 pandecode_prop("index = 0x%" PRIx16, b->index);
1185 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1186
1187 pandecode_indent--;
1188 pandecode_log("},\n");
1189
1190 return 0;
1191 }
1192
1193 static mali_ptr
1194 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1195 {
1196 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1197 return 0;
1198
1199 pandecode_log(".blend = {\n");
1200 pandecode_indent++;
1201
1202 if (is_shader) {
1203 pandecode_shader_address("shader", blend->shader);
1204 } else {
1205 pandecode_blend_equation(&blend->equation);
1206 pandecode_prop("constant = %f", blend->constant);
1207 }
1208
1209 pandecode_indent--;
1210 pandecode_log("},\n");
1211
1212 /* Return blend shader to disassemble if present */
1213 return is_shader ? (blend->shader & ~0xF) : 0;
1214 }
1215
1216 static mali_ptr
1217 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1218 {
1219 struct midgard_blend_rt *b =
1220 ((struct midgard_blend_rt *) descs) + rt_no;
1221
1222 /* Flags determine presence of blend shader */
1223 bool is_shader = (b->flags & 0xF) >= 0x2;
1224
1225 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1226 pandecode_indent++;
1227
1228 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1229
1230 union midgard_blend blend = b->blend;
1231 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1232
1233 pandecode_indent--;
1234 pandecode_log("};\n");
1235
1236 return shader;
1237 }
1238
1239 static int
1240 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1241 {
1242 char base[128];
1243 char *prefix = varying ? "varying" : "attribute";
1244 unsigned max_index = 0;
1245 snprintf(base, sizeof(base), "%s_meta", prefix);
1246
1247 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1248 pandecode_indent++;
1249
1250 struct mali_attr_meta *attr_meta;
1251 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1252
1253 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1254
1255 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1256 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1257 sizeof(*attr_mem));
1258
1259 pandecode_log("{\n");
1260 pandecode_indent++;
1261 pandecode_prop("index = %d", attr_meta->index);
1262
1263 if (attr_meta->index > max_index)
1264 max_index = attr_meta->index;
1265 pandecode_swizzle(attr_meta->swizzle);
1266 pandecode_prop("format = %s", pandecode_format(attr_meta->format));
1267
1268 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1269 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1270 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1271 pandecode_indent--;
1272 pandecode_log("},\n");
1273
1274 }
1275
1276 pandecode_indent--;
1277 pandecode_log("};\n");
1278
1279 return count ? (max_index + 1) : 0;
1280 }
1281
1282 static void
1283 pandecode_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1284 {
1285 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1286
1287 if (imem) {
1288 /* Indices are literally just a u32 array :) */
1289
1290 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1291
1292 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1293 pandecode_indent++;
1294
1295 for (unsigned i = 0; i < (index_count + 1); i += 3)
1296 pandecode_log("%d, %d, %d,\n",
1297 indices[i],
1298 indices[i + 1],
1299 indices[i + 2]);
1300
1301 pandecode_indent--;
1302 pandecode_log("};\n");
1303 }
1304 }
1305
1306 /* return bits [lo, hi) of word */
1307 static u32
1308 bits(u32 word, u32 lo, u32 hi)
1309 {
1310 if (hi - lo >= 32)
1311 return word; // avoid undefined behavior with the shift
1312
1313 return (word >> lo) & ((1 << (hi - lo)) - 1);
1314 }
1315
1316 static void
1317 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no, bool noninstanced)
1318 {
1319 pandecode_log_cont("{\n");
1320 pandecode_indent++;
1321
1322 /* Decode invocation_count. See the comment before the definition of
1323 * invocation_count for an explanation.
1324 */
1325
1326 unsigned size_x = bits(p->invocation_count, 0, p->size_y_shift) + 1;
1327 unsigned size_y = bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1;
1328 unsigned size_z = bits(p->invocation_count, p->size_z_shift, p->workgroups_x_shift) + 1;
1329
1330 unsigned groups_x = bits(p->invocation_count, p->workgroups_x_shift, p->workgroups_y_shift) + 1;
1331 unsigned groups_y = bits(p->invocation_count, p->workgroups_y_shift, p->workgroups_z_shift) + 1;
1332 unsigned groups_z = bits(p->invocation_count, p->workgroups_z_shift, 32) + 1;
1333
1334 /* Even though we have this decoded, we want to ensure that the
1335 * representation is "unique" so we don't lose anything by printing only
1336 * the final result. More specifically, we need to check that we were
1337 * passed something in canonical form, since the definition per the
1338 * hardware is inherently not unique. How? Well, take the resulting
1339 * decode and pack it ourselves! If it is bit exact with what we
1340 * decoded, we're good to go. */
1341
1342 struct mali_vertex_tiler_prefix ref;
1343 panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, noninstanced);
1344
1345 bool canonical =
1346 (p->invocation_count == ref.invocation_count) &&
1347 (p->size_y_shift == ref.size_y_shift) &&
1348 (p->size_z_shift == ref.size_z_shift) &&
1349 (p->workgroups_x_shift == ref.workgroups_x_shift) &&
1350 (p->workgroups_y_shift == ref.workgroups_y_shift) &&
1351 (p->workgroups_z_shift == ref.workgroups_z_shift) &&
1352 (p->workgroups_x_shift_2 == ref.workgroups_x_shift_2);
1353
1354 if (!canonical) {
1355 pandecode_msg("XXX: non-canonical workgroups packing\n");
1356 pandecode_msg("expected: %X, %d, %d, %d, %d, %d\n",
1357 ref.invocation_count,
1358 ref.size_y_shift,
1359 ref.size_z_shift,
1360 ref.workgroups_x_shift,
1361 ref.workgroups_y_shift,
1362 ref.workgroups_z_shift,
1363 ref.workgroups_x_shift_2);
1364
1365 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1366 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1367 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1368 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1369 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1370 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1371 pandecode_prop("workgroups_x_shift_2 = %d", p->workgroups_x_shift_2);
1372 }
1373
1374 /* Regardless, print the decode */
1375 pandecode_msg("size (%d, %d, %d), count (%d, %d, %d)\n",
1376 size_x, size_y, size_z,
1377 groups_x, groups_y, groups_z);
1378
1379 /* TODO: Decode */
1380 if (p->unknown_draw)
1381 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1382
1383 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1384
1385 if (p->draw_mode != MALI_DRAW_NONE)
1386 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
1387
1388 /* Index count only exists for tiler jobs anyway */
1389
1390 if (p->index_count)
1391 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1392
1393 if (p->offset_bias_correction)
1394 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1395
1396 if (p->zero1) {
1397 pandecode_msg("Zero tripped\n");
1398 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1399 }
1400
1401 pandecode_indent--;
1402 pandecode_log("},\n");
1403 }
1404
1405 static void
1406 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1407 {
1408 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1409
1410 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1411
1412 for (int i = 0; i < ubufs_count; i++) {
1413 mali_ptr ptr = ubufs[i].ptr << 2;
1414 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1415 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1416 char name[50];
1417 snprintf(name, sizeof(name), "ubuf_%d", i);
1418 /* The blob uses ubuf 0 to upload internal stuff and
1419 * uniforms that won't fit/are accessed indirectly, so
1420 * it puts it in the batchbuffer.
1421 */
1422 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1423 pandecode_indent++;
1424
1425 for (int j = 0; j <= ubufs[i].size; j++) {
1426 for (int k = 0; k < 4; k++) {
1427 if (k == 0)
1428 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1429 else
1430 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1431
1432 }
1433
1434 pandecode_log_cont("\n");
1435 }
1436
1437 pandecode_indent--;
1438 pandecode_log("};\n");
1439 }
1440
1441 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64"_%d[] = {\n",
1442 pubufs, job_no);
1443 pandecode_indent++;
1444
1445 for (int i = 0; i < ubufs_count; i++) {
1446 pandecode_log("{\n");
1447 pandecode_indent++;
1448 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1449 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1450 pandecode_indent--;
1451 pandecode_log("},\n");
1452 }
1453
1454 pandecode_indent--;
1455 pandecode_log("};\n");
1456 }
1457
1458 static void
1459 pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1460 {
1461
1462 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1463
1464 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1465
1466 if (scratchpad->zero)
1467 pandecode_msg("XXX scratchpad zero tripped");
1468
1469 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1470 pandecode_indent++;
1471
1472 pandecode_prop("flags = 0x%x", scratchpad->flags);
1473 MEMORY_PROP(scratchpad, gpu_scratchpad);
1474
1475 pandecode_indent--;
1476 pandecode_log("};\n");
1477 }
1478
1479 static unsigned shader_id = 0;
1480
1481 static void
1482 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1483 bool is_bifrost, unsigned nr_regs)
1484 {
1485 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1486 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1487
1488 /* Compute maximum possible size */
1489 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1490
1491 /* Print some boilerplate to clearly denote the assembly (which doesn't
1492 * obey indentation rules), and actually do the disassembly! */
1493
1494 printf("\n\n");
1495
1496 char prefix[512];
1497
1498 snprintf(prefix, sizeof(prefix) - 1, "shader%d - %s shader: ",
1499 shader_id++,
1500 (type == JOB_TYPE_TILER) ? "FRAGMENT" : "VERTEX");
1501
1502 if (is_bifrost) {
1503 disassemble_bifrost(code, sz, false);
1504 } else {
1505 disassemble_midgard(code, sz, true, nr_regs, prefix);
1506 }
1507
1508 printf("\n\n");
1509 }
1510
1511 static void
1512 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1513 int job_no, enum mali_job_type job_type,
1514 char *suffix, bool is_bifrost)
1515 {
1516 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1517 struct pandecode_mapped_memory *attr_mem;
1518
1519 unsigned rt_count = 1;
1520
1521 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1522 * are the only things actually needed from the FBD, vertex/tiler jobs
1523 * no longer reference the FBD -- instead, this field points to some
1524 * info about the scratchpad.
1525 */
1526 if (is_bifrost)
1527 pandecode_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1528 else if (p->framebuffer & MALI_MFBD)
1529 rt_count = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1530 else if (job_type == JOB_TYPE_COMPUTE)
1531 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1532 else
1533 pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1534
1535 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1536 int texture_count = 0, sampler_count = 0;
1537
1538 if (shader_meta_ptr) {
1539 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1540 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1541
1542 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1543 pandecode_indent++;
1544
1545 /* Save for dumps */
1546 attribute_count = s->attribute_count;
1547 varying_count = s->varying_count;
1548 texture_count = s->texture_count;
1549 sampler_count = s->sampler_count;
1550
1551 if (is_bifrost) {
1552 uniform_count = s->bifrost2.uniform_count;
1553 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1554 } else {
1555 uniform_count = s->midgard1.uniform_count;
1556 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1557 }
1558
1559 mali_ptr shader_ptr = pandecode_shader_address("shader", s->shader);
1560
1561 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1562 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1563 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1564 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1565
1566 unsigned nr_registers = 0;
1567
1568 if (is_bifrost) {
1569 pandecode_log(".bifrost1 = {\n");
1570 pandecode_indent++;
1571
1572 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1573 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1574
1575 pandecode_indent--;
1576 pandecode_log("},\n");
1577 } else {
1578 pandecode_log(".midgard1 = {\n");
1579 pandecode_indent++;
1580
1581 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1582 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1583 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1584 nr_registers = s->midgard1.work_count;
1585
1586 pandecode_log(".flags = ");
1587 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1588 pandecode_log_cont(",\n");
1589
1590 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1591
1592 pandecode_indent--;
1593 pandecode_log("},\n");
1594 }
1595
1596 if (s->depth_units || s->depth_factor) {
1597 pandecode_prop("depth_factor = %f", s->depth_factor);
1598 pandecode_prop("depth_units = %f", s->depth_units);
1599 }
1600
1601 if (s->alpha_coverage) {
1602 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1603 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1604
1605 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1606 invert_alpha_coverage ? "~" : "",
1607 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1608 }
1609
1610 if (s->unknown2_3 || s->unknown2_4) {
1611 pandecode_log(".unknown2_3 = ");
1612
1613 int unknown2_3 = s->unknown2_3;
1614 int unknown2_4 = s->unknown2_4;
1615
1616 /* We're not quite sure what these flags mean without the depth test, if anything */
1617
1618 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1619 const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
1620 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1621
1622 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1623 }
1624
1625 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1626 pandecode_log_cont(",\n");
1627
1628 pandecode_log(".unknown2_4 = ");
1629 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1630 pandecode_log_cont(",\n");
1631 }
1632
1633 if (s->stencil_mask_front || s->stencil_mask_back) {
1634 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1635 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1636 }
1637
1638 pandecode_stencil("front", &s->stencil_front);
1639 pandecode_stencil("back", &s->stencil_back);
1640
1641 if (is_bifrost) {
1642 pandecode_log(".bifrost2 = {\n");
1643 pandecode_indent++;
1644
1645 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1646 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1647 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1648 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1649
1650 pandecode_indent--;
1651 pandecode_log("},\n");
1652 } else if (s->midgard2.unknown2_7) {
1653 pandecode_log(".midgard2 = {\n");
1654 pandecode_indent++;
1655
1656 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1657 pandecode_indent--;
1658 pandecode_log("},\n");
1659 }
1660
1661 if (s->unknown2_8)
1662 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1663
1664 if (!is_bifrost) {
1665 /* TODO: Blend shaders routing/disasm */
1666
1667 union midgard_blend blend = s->blend;
1668 pandecode_midgard_blend(&blend, false);
1669 }
1670
1671 pandecode_indent--;
1672 pandecode_log("};\n");
1673
1674 /* MRT blend fields are used whenever MFBD is used, with
1675 * per-RT descriptors */
1676
1677 if (job_type == JOB_TYPE_TILER) {
1678 void* blend_base = (void *) (s + 1);
1679
1680 for (unsigned i = 0; i < rt_count; i++) {
1681 mali_ptr shader = 0;
1682
1683 if (is_bifrost)
1684 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1685 else
1686 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1687
1688 if (shader & ~0xF)
1689 pandecode_shader_disassemble(shader, job_no, job_type, false, 0);
1690 }
1691 }
1692
1693 if (shader_ptr & ~0xF)
1694 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost, nr_registers);
1695 } else
1696 pandecode_msg("<no shader>\n");
1697
1698 if (p->viewport) {
1699 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1700 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1701
1702 pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
1703 pandecode_indent++;
1704
1705 pandecode_prop("clip_minx = %f", f->clip_minx);
1706 pandecode_prop("clip_miny = %f", f->clip_miny);
1707 pandecode_prop("clip_minz = %f", f->clip_minz);
1708 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1709 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1710 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1711
1712 /* Only the higher coordinates are MALI_POSITIVE scaled */
1713
1714 pandecode_prop("viewport0 = { %d, %d }",
1715 f->viewport0[0], f->viewport0[1]);
1716
1717 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1718 f->viewport1[0] + 1, f->viewport1[1] + 1);
1719
1720 pandecode_indent--;
1721 pandecode_log("};\n");
1722 }
1723
1724 if (p->attribute_meta) {
1725 unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1726
1727 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1728 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1729 }
1730
1731 /* Varyings are encoded like attributes but not actually sent; we just
1732 * pass a zero buffer with the right stride/size set, (or whatever)
1733 * since the GPU will write to it itself */
1734
1735 if (p->varying_meta) {
1736 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1737 }
1738
1739 if (p->varyings) {
1740 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1741
1742 /* Number of descriptors depends on whether there are
1743 * non-internal varyings */
1744
1745 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
1746 }
1747
1748 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1749
1750 if (p->uniforms && !is_compute) {
1751 int rows = uniform_count, width = 4;
1752 size_t sz = rows * width * sizeof(float);
1753
1754 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1755 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1756 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1757
1758 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1759
1760 pandecode_indent++;
1761
1762 for (int row = 0; row < rows; row++) {
1763 for (int i = 0; i < width; i++) {
1764 u32 v = uniforms[i];
1765 float f;
1766 memcpy(&f, &v, sizeof(v));
1767 pandecode_log_cont("%X /* %f */, ", v, f);
1768 }
1769
1770 pandecode_log_cont("\n");
1771
1772 uniforms += width;
1773 }
1774
1775 pandecode_indent--;
1776 pandecode_log("};\n");
1777 } else if (p->uniforms) {
1778 int rows = uniform_count * 2;
1779 size_t sz = rows * sizeof(mali_ptr);
1780
1781 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1782 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1783 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1784
1785 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1786
1787 pandecode_indent++;
1788
1789 for (int row = 0; row < rows; row++) {
1790 char *a = pointer_as_memory_reference(uniforms[row]);
1791 pandecode_log("%s,\n", a);
1792 free(a);
1793 }
1794
1795 pandecode_indent--;
1796 pandecode_log("};\n");
1797
1798 }
1799
1800 if (p->uniform_buffers) {
1801 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1802 }
1803
1804 if (p->texture_trampoline) {
1805 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1806
1807 if (mmem) {
1808 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1809
1810 pandecode_log("uint64_t texture_trampoline_%"PRIx64"_%d[] = {\n", p->texture_trampoline, job_no);
1811 pandecode_indent++;
1812
1813 for (int tex = 0; tex < texture_count; ++tex) {
1814 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1815 char *a = pointer_as_memory_reference(*u);
1816 pandecode_log("%s,\n", a);
1817 free(a);
1818 }
1819
1820 pandecode_indent--;
1821 pandecode_log("};\n");
1822
1823 /* Now, finally, descend down into the texture descriptor */
1824 for (int tex = 0; tex < texture_count; ++tex) {
1825 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1826 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1827
1828 if (tmem) {
1829 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1830
1831 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1832 pandecode_indent++;
1833
1834 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1835 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1836 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1837 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1838 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1839 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1840 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1841
1842 struct mali_texture_format f = t->format;
1843
1844 pandecode_log(".format = {\n");
1845 pandecode_indent++;
1846
1847 pandecode_swizzle(f.swizzle);
1848 pandecode_prop("format = %s", pandecode_format(f.format));
1849 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1850 pandecode_prop("srgb = %" PRId32, f.srgb);
1851 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1852 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1853
1854 pandecode_indent--;
1855 pandecode_log("},\n");
1856
1857 pandecode_swizzle(t->swizzle);
1858
1859 if (t->swizzle_zero) {
1860 /* Shouldn't happen */
1861 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1862 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1863 }
1864
1865 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1866
1867 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1868 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1869 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1870
1871 pandecode_log(".payload = {\n");
1872 pandecode_indent++;
1873
1874 /* A bunch of bitmap pointers follow.
1875 * We work out the correct number,
1876 * based on the mipmap/cubemap
1877 * properties, but dump extra
1878 * possibilities to futureproof */
1879
1880 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1881 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1882
1883 /* Miptree for each face */
1884 if (f.type == MALI_TEX_CUBE)
1885 bitmap_count *= 6;
1886
1887 /* Array of textures */
1888 bitmap_count *= MALI_NEGATIVE(t->array_size);
1889
1890 /* Stride for each element */
1891 if (manual_stride)
1892 bitmap_count *= 2;
1893
1894 /* Sanity check the size */
1895 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1896 assert (bitmap_count <= max_count);
1897
1898 /* Dump more to be safe, but not _that_ much more */
1899 int safe_count = MIN2(bitmap_count * 2, max_count);
1900
1901 for (int i = 0; i < safe_count; ++i) {
1902 char *prefix = (i >= bitmap_count) ? "// " : "";
1903
1904 /* How we dump depends if this is a stride or a pointer */
1905
1906 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1907 /* signed 32-bit snuck in as a 64-bit pointer */
1908 uint64_t stride_set = t->payload[i];
1909 uint32_t clamped_stride = stride_set;
1910 int32_t stride = clamped_stride;
1911 assert(stride_set == clamped_stride);
1912 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1913 } else {
1914 char *a = pointer_as_memory_reference(t->payload[i]);
1915 pandecode_log("%s%s, \n", prefix, a);
1916 free(a);
1917 }
1918 }
1919
1920 pandecode_indent--;
1921 pandecode_log("},\n");
1922
1923 pandecode_indent--;
1924 pandecode_log("};\n");
1925 }
1926 }
1927 }
1928 }
1929
1930 if (p->sampler_descriptor) {
1931 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1932
1933 if (smem) {
1934 struct mali_sampler_descriptor *s;
1935
1936 mali_ptr d = p->sampler_descriptor;
1937
1938 for (int i = 0; i < sampler_count; ++i) {
1939 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1940
1941 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", d + sizeof(*s) * i, job_no, i);
1942 pandecode_indent++;
1943
1944 pandecode_log(".filter_mode = ");
1945 pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
1946 pandecode_log_cont(",\n");
1947
1948 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1949 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1950
1951 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
1952 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
1953 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
1954
1955 pandecode_prop("compare_func = %s", pandecode_alt_func(s->compare_func));
1956
1957 if (s->zero || s->zero2) {
1958 pandecode_msg("Zero tripped\n");
1959 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1960 }
1961
1962 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1963
1964 pandecode_prop("border_color = { %f, %f, %f, %f }",
1965 s->border_color[0],
1966 s->border_color[1],
1967 s->border_color[2],
1968 s->border_color[3]);
1969
1970 pandecode_indent--;
1971 pandecode_log("};\n");
1972 }
1973 }
1974 }
1975 }
1976
1977 static void
1978 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1979 {
1980 if (!(p->position_varying || p->occlusion_counter || p->flags))
1981 return;
1982
1983 pandecode_log(".postfix = {\n");
1984 pandecode_indent++;
1985
1986 MEMORY_PROP(p, position_varying);
1987 MEMORY_PROP(p, occlusion_counter);
1988
1989 if (p->flags)
1990 pandecode_prop("flags = %d", p->flags);
1991
1992 pandecode_indent--;
1993 pandecode_log("},\n");
1994 }
1995
1996 static void
1997 pandecode_vertex_only_bfr(struct bifrost_vertex_only *v)
1998 {
1999 pandecode_log_cont("{\n");
2000 pandecode_indent++;
2001
2002 pandecode_prop("unk2 = 0x%x", v->unk2);
2003
2004 if (v->zero0 || v->zero1) {
2005 pandecode_msg("vertex only zero tripped");
2006 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
2007 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
2008 }
2009
2010 pandecode_indent--;
2011 pandecode_log("}\n");
2012 }
2013
2014 static void
2015 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
2016 {
2017
2018 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2019 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
2020
2021 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
2022 pandecode_indent++;
2023
2024 if (h->zero) {
2025 pandecode_msg("tiler heap zero tripped\n");
2026 pandecode_prop("zero = 0x%x", h->zero);
2027 }
2028
2029 for (int i = 0; i < 12; i++) {
2030 if (h->zeros[i] != 0) {
2031 pandecode_msg("tiler heap zero %d tripped, value %x\n",
2032 i, h->zeros[i]);
2033 }
2034 }
2035
2036 pandecode_prop("heap_size = 0x%x", h->heap_size);
2037 MEMORY_PROP(h, tiler_heap_start);
2038 MEMORY_PROP(h, tiler_heap_free);
2039
2040 /* this might point to the beginning of another buffer, when it's
2041 * really the end of the tiler heap buffer, so we have to be careful
2042 * here. but for zero length, we need the same pointer.
2043 */
2044
2045 if (h->tiler_heap_end == h->tiler_heap_start) {
2046 MEMORY_PROP(h, tiler_heap_start);
2047 } else {
2048 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2049 pandecode_prop("tiler_heap_end = %s + 1", a);
2050 free(a);
2051 }
2052
2053 pandecode_indent--;
2054 pandecode_log("};\n");
2055 }
2056
2057 static void
2058 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2059 {
2060 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2061 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2062
2063 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2064
2065 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
2066 pandecode_indent++;
2067
2068 if (t->zero0 || t->zero1) {
2069 pandecode_msg("tiler meta zero tripped");
2070 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2071 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2072 }
2073
2074 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2075 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2076
2077 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2078 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2079
2080 for (int i = 0; i < 12; i++) {
2081 if (t->zeros[i] != 0) {
2082 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2083 i, t->zeros[i]);
2084 }
2085 }
2086
2087 pandecode_indent--;
2088 pandecode_log("};\n");
2089 }
2090
2091 static void
2092 pandecode_gl_enables(uint32_t gl_enables, int job_type)
2093 {
2094 pandecode_log(".gl_enables = ");
2095
2096 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2097
2098 pandecode_log_cont(",\n");
2099 }
2100
2101 static void
2102 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2103 {
2104 if (u.pointer == 0x0)
2105 return;
2106
2107 pandecode_log(".primitive_size = {\n");
2108 pandecode_indent++;
2109
2110 if (constant) {
2111 pandecode_prop("constant = %f", u.constant);
2112 } else {
2113 MEMORY_PROP((&u), pointer);
2114 }
2115
2116 pandecode_indent--;
2117 pandecode_log("},\n");
2118 }
2119
2120 static void
2121 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2122 {
2123 pandecode_log_cont("{\n");
2124 pandecode_indent++;
2125
2126 /* TODO: gl_PointSize on Bifrost */
2127 pandecode_primitive_size(t->primitive_size, true);
2128
2129 pandecode_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2130
2131 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2132 || t->zero6 || t->zero7 || t->zero8) {
2133 pandecode_msg("tiler only zero tripped");
2134 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2135 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2136 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2137 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2138 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2139 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2140 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2141 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2142 }
2143
2144 pandecode_indent--;
2145 pandecode_log("},\n");
2146 }
2147
2148 static int
2149 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2150 const struct pandecode_mapped_memory *mem,
2151 mali_ptr payload, int job_no)
2152 {
2153 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2154
2155 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2156
2157 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2158 pandecode_indent++;
2159
2160 pandecode_log(".prefix = ");
2161 pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
2162
2163 pandecode_log(".vertex = ");
2164 pandecode_vertex_only_bfr(&v->vertex);
2165
2166 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2167
2168 pandecode_indent--;
2169 pandecode_log("};\n");
2170
2171 return sizeof(*v);
2172 }
2173
2174 static int
2175 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2176 const struct pandecode_mapped_memory *mem,
2177 mali_ptr payload, int job_no)
2178 {
2179 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2180
2181 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2182
2183 pandecode_indices(t->prefix.indices, t->prefix.index_count, job_no);
2184 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2185
2186 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2187 pandecode_indent++;
2188
2189 pandecode_log(".prefix = ");
2190 pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
2191
2192 pandecode_log(".tiler = ");
2193 pandecode_tiler_only_bfr(&t->tiler, job_no);
2194
2195 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2196
2197 pandecode_indent--;
2198 pandecode_log("};\n");
2199
2200 return sizeof(*t);
2201 }
2202
2203 static int
2204 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2205 const struct pandecode_mapped_memory *mem,
2206 mali_ptr payload, int job_no)
2207 {
2208 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2209
2210 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2211
2212 pandecode_indices(v->prefix.indices, v->prefix.index_count, job_no);
2213
2214 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2215 pandecode_indent++;
2216
2217 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2218 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2219
2220 bool instanced = v->instance_shift || v->instance_odd;
2221 bool is_graphics = (h->job_type == JOB_TYPE_VERTEX) || (h->job_type == JOB_TYPE_TILER);
2222
2223 pandecode_log(".prefix = ");
2224 pandecode_vertex_tiler_prefix(&v->prefix, job_no, !instanced && is_graphics);
2225
2226 pandecode_gl_enables(v->gl_enables, h->job_type);
2227
2228 if (v->instance_shift || v->instance_odd) {
2229 pandecode_prop("instance_shift = 0x%d /* %d */",
2230 v->instance_shift, 1 << v->instance_shift);
2231 pandecode_prop("instance_odd = 0x%X /* %d */",
2232 v->instance_odd, (2 * v->instance_odd) + 1);
2233
2234 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2235 }
2236
2237 if (v->offset_start)
2238 pandecode_prop("offset_start = %d", v->offset_start);
2239
2240 if (v->zero5) {
2241 pandecode_msg("Zero tripped\n");
2242 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2243 }
2244
2245 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2246
2247 pandecode_indent--;
2248 pandecode_log("};\n");
2249
2250 return sizeof(*v);
2251 }
2252
2253 static int
2254 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2255 mali_ptr payload, int job_no,
2256 bool is_bifrost)
2257 {
2258 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2259
2260 bool fbd_dumped = false;
2261
2262 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2263 /* Only SFBDs are understood, not MFBDs. We're speculating,
2264 * based on the versioning, kernel code, etc, that the
2265 * difference is between Single FrameBuffer Descriptor and
2266 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2267 * up with multi-framebuffer support being added (T7xx onwards,
2268 * including Gxx). In any event, there's some field shuffling
2269 * that we haven't looked into yet. */
2270
2271 pandecode_sfbd(s->framebuffer & FBD_MASK, job_no);
2272 fbd_dumped = true;
2273 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2274 /* We don't know if Bifrost supports SFBD's at all, since the
2275 * driver never uses them. And the format is different from
2276 * Midgard anyways, due to the tiler heap and scratchpad being
2277 * moved out into separate structures, so it's not clear what a
2278 * Bifrost SFBD would even look like without getting an actual
2279 * trace, which appears impossible.
2280 */
2281
2282 pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2283 fbd_dumped = true;
2284 }
2285
2286 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2287 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2288 pandecode_indent++;
2289
2290 /* See the comments by the macro definitions for mathematical context
2291 * on why this is so weird */
2292
2293 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2294 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2295
2296 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2297 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2298 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2299
2300 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2301 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2302 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2303
2304 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2305 * we have to fallback on offsets. */
2306
2307 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2308
2309 /* TODO: Decode */
2310 unsigned extra_flags = (s->framebuffer & ~FBD_MASK) & ~MALI_MFBD;
2311
2312 if (fbd_dumped)
2313 pandecode_prop("framebuffer = framebuffer_%d_p | %s | 0x%X", job_no,
2314 fbd_type, extra_flags);
2315 else
2316 pandecode_prop("framebuffer = %s | %s | 0x%X", pointer_as_memory_reference(p),
2317 fbd_type, extra_flags);
2318
2319 pandecode_indent--;
2320 pandecode_log("};\n");
2321
2322 return sizeof(*s);
2323 }
2324
2325 static int job_descriptor_number = 0;
2326
2327 int
2328 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
2329 {
2330 struct mali_job_descriptor_header *h;
2331
2332 int start_number = 0;
2333
2334 bool first = true;
2335 bool last_size;
2336
2337 do {
2338 struct pandecode_mapped_memory *mem =
2339 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2340
2341 void *payload;
2342
2343 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2344
2345 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2346 * high 32-bits of the 64-bit pointer are reused to store
2347 * something else.
2348 */
2349 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2350 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2351 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2352
2353 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2354 MALI_PAYLOAD_SIZE);
2355
2356 int job_no = job_descriptor_number++;
2357
2358 if (first)
2359 start_number = job_no;
2360
2361 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2362 pandecode_indent++;
2363
2364 pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
2365
2366 /* Save for next job fixing */
2367 last_size = h->job_descriptor_size;
2368
2369 if (h->job_descriptor_size)
2370 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2371
2372 if (h->exception_status && h->exception_status != 0x1)
2373 pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
2374 h->exception_status,
2375 (h->exception_status >> 16) & 0xFFFF,
2376 pandecode_exception_access((h->exception_status >> 8) & 0x3),
2377 h->exception_status & 0xFF);
2378
2379 if (h->first_incomplete_task)
2380 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2381
2382 if (h->fault_pointer)
2383 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2384
2385 if (h->job_barrier)
2386 pandecode_prop("job_barrier = %d", h->job_barrier);
2387
2388 pandecode_prop("job_index = %d", h->job_index);
2389
2390 if (h->unknown_flags)
2391 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2392
2393 if (h->job_dependency_index_1)
2394 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2395
2396 if (h->job_dependency_index_2)
2397 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2398
2399 pandecode_indent--;
2400 pandecode_log("};\n");
2401
2402 /* Do not touch the field yet -- decode the payload first, and
2403 * don't touch that either. This is essential for the uploads
2404 * to occur in sequence and therefore be dynamically allocated
2405 * correctly. Do note the size, however, for that related
2406 * reason. */
2407
2408 switch (h->job_type) {
2409 case JOB_TYPE_SET_VALUE: {
2410 struct mali_payload_set_value *s = payload;
2411 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2412 pandecode_indent++;
2413 MEMORY_PROP(s, out);
2414 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2415 pandecode_indent--;
2416 pandecode_log("};\n");
2417
2418 break;
2419 }
2420
2421 case JOB_TYPE_TILER:
2422 case JOB_TYPE_VERTEX:
2423 case JOB_TYPE_COMPUTE:
2424 if (bifrost) {
2425 if (h->job_type == JOB_TYPE_TILER)
2426 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
2427 else
2428 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
2429 } else
2430 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2431
2432 break;
2433
2434 case JOB_TYPE_FRAGMENT:
2435 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost);
2436 break;
2437
2438 default:
2439 break;
2440 }
2441
2442 /* Handle linkage */
2443
2444 if (!first) {
2445 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2446
2447 if (last_size)
2448 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2449 else
2450 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2451 }
2452
2453 first = false;
2454
2455 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2456
2457 return start_number;
2458 }