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