projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
pan/decode: Print "just right" count of texture pointers
[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(
480 const struct midgard_tiler_descriptor *t,
481 unsigned width,
482 unsigned height)
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
490 MEMORY_PROP(t, polygon_list);
491
492 /* The body is offset from the base of the polygon list */
493 assert(t->polygon_list_body > t->polygon_list);
494 unsigned body_offset = t->polygon_list_body - t->polygon_list;
495
496 /* It needs to fit inside the reported size */
497 assert(t->polygon_list_size >= body_offset);
498
499 /* Check that we fit */
500 struct pandecode_mapped_memory *plist =
501 pandecode_find_mapped_gpu_mem_containing(t->polygon_list);
502
503 assert(t->polygon_list_size <= plist->length);
504
505 /* Now that we've sanity checked, we'll try to calculate the sizes
506 * ourselves for comparison */
507
508 unsigned ref_header = panfrost_tiler_header_size(width, height, t->hierarchy_mask);
509 unsigned ref_body = panfrost_tiler_body_size(width, height, t->hierarchy_mask);
510 unsigned ref_size = ref_header + ref_body;
511
512 if (!((ref_header == body_offset) && (ref_size == t->polygon_list_size))) {
513 pandecode_msg("XXX: bad polygon list size (expected %d / 0x%x)\n",
514 ref_header, ref_size);
515 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
516 pandecode_msg("body offset %d\n", body_offset);
517 }
518
519 /* The tiler heap has a start and end specified, so check that
520 * everything fits in a contiguous BO (otherwise, we risk out-of-bounds
521 * reads) */
522
523 MEMORY_PROP(t, heap_start);
524 assert(t->heap_end >= t->heap_start);
525
526 struct pandecode_mapped_memory *heap =
527 pandecode_find_mapped_gpu_mem_containing(t->heap_start);
528
529 unsigned heap_size = t->heap_end - t->heap_start;
530 assert(heap_size <= heap->length);
531
532 pandecode_msg("heap size %d\n", heap_size);
533
534 bool nonzero_weights = false;
535
536 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
537 nonzero_weights |= t->weights[w] != 0x0;
538 }
539
540 if (nonzero_weights) {
541 pandecode_log(".weights = {");
542
543 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
544 pandecode_log("%d, ", t->weights[w]);
545 }
546
547 pandecode_log("},");
548 }
549
550 pandecode_indent--;
551 pandecode_log("}\n");
552 }
553
554 static void
555 pandecode_sfbd(uint64_t gpu_va, int job_no)
556 {
557 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
558 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
559
560 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
561 pandecode_indent++;
562
563 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
564 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
565
566 pandecode_log(".format = ");
567 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
568 pandecode_log_cont(",\n");
569
570 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
571 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
572
573 MEMORY_PROP(s, framebuffer);
574 pandecode_prop("stride = %d", s->stride);
575
576 /* Earlier in the actual commandstream -- right before width -- but we
577 * delay to flow nicer */
578
579 pandecode_log(".clear_flags = ");
580 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
581 pandecode_log_cont(",\n");
582
583 if (s->depth_buffer | s->depth_buffer_enable) {
584 MEMORY_PROP(s, depth_buffer);
585 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
586 }
587
588 if (s->stencil_buffer | s->stencil_buffer_enable) {
589 MEMORY_PROP(s, stencil_buffer);
590 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
591 }
592
593 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
594 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
595 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
596 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
597 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
598 }
599
600 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
601 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
602 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
603 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
604 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
605 }
606
607 if (s->clear_stencil) {
608 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
609 }
610
611 MEMORY_PROP(s, unknown_address_0);
612 const struct midgard_tiler_descriptor t = s->tiler;
613 pandecode_midgard_tiler_descriptor(&t, s->width + 1, s->height + 1);
614
615 pandecode_indent--;
616 pandecode_log("};\n");
617
618 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
619 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
620 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
621 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
622
623 printf(".zero3 = {");
624
625 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
626 printf("%X, ", s->zero3[i]);
627
628 printf("},\n");
629
630 printf(".zero6 = {");
631
632 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
633 printf("%X, ", s->zero6[i]);
634
635 printf("},\n");
636 }
637
638 static void
639 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
640 {
641 pandecode_log(".unknown%d = {", name);
642
643 for (int i = 0; i < count; ++i)
644 printf("%X, ", slide[i]);
645
646 pandecode_log("},\n");
647 }
648
649 #define SHORT_SLIDE(num) \
650 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
651
652 static void
653 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
654 {
655 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
656 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
657
658 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
659 pandecode_indent++;
660
661 SHORT_SLIDE(1);
662
663 pandecode_indent--;
664 printf("},\n");
665 }
666
667 static void
668 pandecode_swizzle(unsigned swizzle)
669 {
670 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
671 pandecode_channel((swizzle >> 0) & 0x7),
672 pandecode_channel((swizzle >> 3) & 0x7),
673 pandecode_channel((swizzle >> 6) & 0x7),
674 pandecode_channel((swizzle >> 9) & 0x7));
675 }
676
677 static void
678 pandecode_rt_format(struct mali_rt_format format)
679 {
680 pandecode_log(".format = {\n");
681 pandecode_indent++;
682
683 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
684 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
685 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
686
687 pandecode_prop("block = %s",
688 pandecode_mfbd_block_format(format.block));
689
690 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
691 MALI_NEGATIVE(format.nr_channels));
692
693 pandecode_log(".flags = ");
694 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
695 pandecode_log_cont(",\n");
696
697 pandecode_swizzle(format.swizzle);
698
699 pandecode_prop("no_preload = 0x%" PRIx32, format.no_preload);
700
701 if (format.zero)
702 pandecode_prop("zero = 0x%" PRIx32, format.zero);
703
704 pandecode_indent--;
705 pandecode_log("},\n");
706 }
707
708 static void
709 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
710 {
711 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
712 pandecode_indent++;
713
714 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
715 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
716 struct pandecode_mapped_memory *mem =
717 pandecode_find_mapped_gpu_mem_containing(rt_va);
718 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
719
720 pandecode_log("{\n");
721 pandecode_indent++;
722
723 pandecode_rt_format(rt->format);
724
725 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
726 pandecode_log(".afbc = {\n");
727 pandecode_indent++;
728
729 char *a = pointer_as_memory_reference(rt->afbc.metadata);
730 pandecode_prop("metadata = %s", a);
731 free(a);
732
733 pandecode_prop("stride = %d", rt->afbc.stride);
734 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
735
736 pandecode_indent--;
737 pandecode_log("},\n");
738 } else {
739 pandecode_log(".chunknown = {\n");
740 pandecode_indent++;
741
742 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
743
744 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
745 pandecode_prop("pointer = %s", a);
746 free(a);
747
748 pandecode_indent--;
749 pandecode_log("},\n");
750 }
751
752 MEMORY_PROP(rt, framebuffer);
753 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
754
755 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
756 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
757 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
758 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
759 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
760 }
761
762 if (rt->zero1 || rt->zero2 || rt->zero3) {
763 pandecode_msg("render target zeros tripped\n");
764 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
765 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
766 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
767 }
768
769 pandecode_indent--;
770 pandecode_log("},\n");
771 }
772
773 pandecode_indent--;
774 pandecode_log("};\n");
775 }
776
777 static unsigned
778 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
779 {
780 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
781 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
782
783 if (fb->sample_locations) {
784 /* The blob stores all possible sample locations in a single buffer
785 * allocated on startup, and just switches the pointer when switching
786 * MSAA state. For now, we just put the data into the cmdstream, but we
787 * should do something like what the blob does with a real driver.
788 *
789 * There seem to be 32 slots for sample locations, followed by another
790 * 16. The second 16 is just the center location followed by 15 zeros
791 * in all the cases I've identified (maybe shader vs. depth/color
792 * samples?).
793 */
794
795 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
796
797 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
798
799 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
800 pandecode_indent++;
801
802 for (int i = 0; i < 32 + 16; i++) {
803 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
804 }
805
806 pandecode_indent--;
807 pandecode_log("};\n");
808 }
809
810 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
811 pandecode_indent++;
812
813 pandecode_prop("unk0 = 0x%x", fb->unk0);
814
815 if (fb->sample_locations)
816 pandecode_prop("sample_locations = sample_locations_%d", job_no);
817
818 /* Assume that unknown1 was emitted in the last job for
819 * now */
820 MEMORY_PROP(fb, unknown1);
821
822 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
823 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
824 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
825 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
826
827 pandecode_prop("unk1 = 0x%x", fb->unk1);
828 pandecode_prop("unk2 = 0x%x", fb->unk2);
829 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
830 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
831
832 pandecode_log(".mfbd_flags = ");
833 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
834 pandecode_log_cont(",\n");
835
836 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
837 pandecode_prop("clear_depth = %f", fb->clear_depth);
838
839 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
840 MEMORY_PROP(fb, scratchpad);
841 const struct midgard_tiler_descriptor t = fb->tiler;
842 pandecode_midgard_tiler_descriptor(&t, fb->width1 + 1, fb->height1 + 1);
843
844 if (fb->zero3 || fb->zero4) {
845 pandecode_msg("framebuffer zeros tripped\n");
846 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
847 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
848 }
849
850 pandecode_indent--;
851 pandecode_log("};\n");
852
853 gpu_va += sizeof(struct bifrost_framebuffer);
854
855 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
856 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
857 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
858
859 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
860 pandecode_indent++;
861
862 MEMORY_PROP(fbx, checksum);
863
864 if (fbx->checksum_stride)
865 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
866
867 pandecode_log(".flags = ");
868 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
869 pandecode_log_cont(",\n");
870
871 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
872 pandecode_log(".ds_afbc = {\n");
873 pandecode_indent++;
874
875 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
876 pandecode_prop("depth_stencil_afbc_stride = %d",
877 fbx->ds_afbc.depth_stencil_afbc_stride);
878 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
879
880 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
881 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
882 pandecode_prop("zero1 = 0x%" PRIx32,
883 fbx->ds_afbc.zero1);
884 pandecode_prop("padding = 0x%" PRIx64,
885 fbx->ds_afbc.padding);
886 }
887
888 pandecode_indent--;
889 pandecode_log("},\n");
890 } else {
891 pandecode_log(".ds_linear = {\n");
892 pandecode_indent++;
893
894 if (fbx->ds_linear.depth) {
895 MEMORY_PROP_DIR(fbx->ds_linear, depth);
896 pandecode_prop("depth_stride = %d",
897 fbx->ds_linear.depth_stride);
898 }
899
900 if (fbx->ds_linear.stencil) {
901 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
902 pandecode_prop("stencil_stride = %d",
903 fbx->ds_linear.stencil_stride);
904 }
905
906 if (fbx->ds_linear.depth_stride_zero ||
907 fbx->ds_linear.stencil_stride_zero ||
908 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
909 pandecode_msg("Depth/stencil zeros tripped\n");
910 pandecode_prop("depth_stride_zero = 0x%x",
911 fbx->ds_linear.depth_stride_zero);
912 pandecode_prop("stencil_stride_zero = 0x%x",
913 fbx->ds_linear.stencil_stride_zero);
914 pandecode_prop("zero1 = 0x%" PRIx32,
915 fbx->ds_linear.zero1);
916 pandecode_prop("zero2 = 0x%" PRIx32,
917 fbx->ds_linear.zero2);
918 }
919
920 pandecode_indent--;
921 pandecode_log("},\n");
922 }
923
924 if (fbx->zero3 || fbx->zero4) {
925 pandecode_msg("fb_extra zeros tripped\n");
926 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
927 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
928 }
929
930 pandecode_indent--;
931 pandecode_log("};\n");
932
933 gpu_va += sizeof(struct bifrost_fb_extra);
934 }
935
936 if (with_render_targets)
937 pandecode_render_target(gpu_va, job_no, fb);
938
939 /* Passback the render target count */
940 return MALI_NEGATIVE(fb->rt_count_1);
941 }
942
943 /* Just add a comment decoding the shift/odd fields forming the padded vertices
944 * count */
945
946 static void
947 pandecode_padded_vertices(unsigned shift, unsigned k)
948 {
949 unsigned odd = 2*k + 1;
950 unsigned pot = 1 << shift;
951 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
952 }
953
954 /* Given a magic divisor, recover what we were trying to divide by.
955 *
956 * Let m represent the magic divisor. By definition, m is an element on Z, whre
957 * 0 <= m < 2^N, for N bits in m.
958 *
959 * Let q represent the number we would like to divide by.
960 *
961 * By definition of a magic divisor for N-bit unsigned integers (a number you
962 * multiply by to magically get division), m is a number such that:
963 *
964 * (m * x) & (2^N - 1) = floor(x/q).
965 * for all x on Z where 0 <= x < 2^N
966 *
967 * Ignore the case where any of the above values equals zero; it is irrelevant
968 * for our purposes (instanced arrays).
969 *
970 * Choose x = q. Then:
971 *
972 * (m * x) & (2^N - 1) = floor(x/q).
973 * (m * q) & (2^N - 1) = floor(q/q).
974 *
975 * floor(q/q) = floor(1) = 1, therefore:
976 *
977 * (m * q) & (2^N - 1) = 1
978 *
979 * Recall the identity that the bitwise AND of one less than a power-of-two
980 * equals the modulo with that power of two, i.e. for all x:
981 *
982 * x & (2^N - 1) = x % N
983 *
984 * Therefore:
985 *
986 * mq % (2^N) = 1
987 *
988 * By definition, a modular multiplicative inverse of a number m is the number
989 * q such that with respect to a modulos M:
990 *
991 * mq % M = 1
992 *
993 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
994 *
995 */
996
997 static void
998 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
999 {
1000 #if 0
1001 /* Compute the modular inverse of `magic` with respect to 2^(32 -
1002 * shift) the most lame way possible... just repeatedly add.
1003 * Asymptoptically slow but nobody cares in practice, unless you have
1004 * massive numbers of vertices or high divisors. */
1005
1006 unsigned inverse = 0;
1007
1008 /* Magic implicitly has the highest bit set */
1009 magic |= (1 << 31);
1010
1011 /* Depending on rounding direction */
1012 if (extra)
1013 magic++;
1014
1015 for (;;) {
1016 uint32_t product = magic * inverse;
1017
1018 if (shift) {
1019 product >>= shift;
1020 }
1021
1022 if (product == 1)
1023 break;
1024
1025 ++inverse;
1026 }
1027
1028 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
1029
1030 /* Recall we're supposed to divide by (gl_level_divisor *
1031 * padded_num_vertices) */
1032
1033 unsigned padded_num_vertices = inverse / orig_divisor;
1034
1035 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
1036 #endif
1037 }
1038
1039 static void
1040 pandecode_attributes(const struct pandecode_mapped_memory *mem,
1041 mali_ptr addr, int job_no, char *suffix,
1042 int count, bool varying)
1043 {
1044 char *prefix = varying ? "varyings" : "attributes";
1045
1046 if (!addr) {
1047 pandecode_msg("no %s\n", prefix);
1048 return;
1049 }
1050
1051 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
1052
1053 char base[128];
1054 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
1055
1056 for (int i = 0; i < count; ++i) {
1057 enum mali_attr_mode mode = attr[i].elements & 7;
1058
1059 if (mode == MALI_ATTR_UNUSED)
1060 continue;
1061
1062 mali_ptr raw_elements = attr[i].elements & ~7;
1063
1064 /* TODO: Do we maybe want to dump the attribute values
1065 * themselves given the specified format? Or is that too hard?
1066 * */
1067
1068 char *a = pointer_as_memory_reference(raw_elements);
1069 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
1070 free(a);
1071 }
1072
1073 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1074 pandecode_indent++;
1075
1076 for (int i = 0; i < count; ++i) {
1077 pandecode_log("{\n");
1078 pandecode_indent++;
1079
1080 unsigned mode = attr[i].elements & 7;
1081 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode(mode));
1082 pandecode_prop("shift = %d", attr[i].shift);
1083 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1084 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1085 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1086
1087 /* Decode further where possible */
1088
1089 if (mode == MALI_ATTR_MODULO) {
1090 pandecode_padded_vertices(
1091 attr[i].shift,
1092 attr[i].extra_flags);
1093 }
1094
1095 pandecode_indent--;
1096 pandecode_log("}, \n");
1097
1098 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1099 i++;
1100 pandecode_log("{\n");
1101 pandecode_indent++;
1102 pandecode_prop("unk = 0x%x", attr[i].unk);
1103 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1104 if (attr[i].zero != 0)
1105 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1106 pandecode_prop("divisor = %d", attr[i].divisor);
1107 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1108 pandecode_indent--;
1109 pandecode_log("}, \n");
1110 }
1111
1112 }
1113
1114 pandecode_indent--;
1115 pandecode_log("};\n");
1116 }
1117
1118 static mali_ptr
1119 pandecode_shader_address(const char *name, mali_ptr ptr)
1120 {
1121 /* TODO: Decode flags */
1122 mali_ptr shader_ptr = ptr & ~15;
1123
1124 char *a = pointer_as_memory_reference(shader_ptr);
1125 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1126 free(a);
1127
1128 return shader_ptr;
1129 }
1130
1131 static bool
1132 all_zero(unsigned *buffer, unsigned count)
1133 {
1134 for (unsigned i = 0; i < count; ++i) {
1135 if (buffer[i])
1136 return false;
1137 }
1138
1139 return true;
1140 }
1141
1142 static void
1143 pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
1144 {
1145 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1146 return;
1147
1148 const char *func = pandecode_func(stencil->func);
1149 const char *sfail = pandecode_stencil_op(stencil->sfail);
1150 const char *dpfail = pandecode_stencil_op(stencil->dpfail);
1151 const char *dppass = pandecode_stencil_op(stencil->dppass);
1152
1153 if (stencil->zero)
1154 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1155
1156 pandecode_log(".stencil_%s = {\n", name);
1157 pandecode_indent++;
1158 pandecode_prop("ref = %d", stencil->ref);
1159 pandecode_prop("mask = 0x%02X", stencil->mask);
1160 pandecode_prop("func = %s", func);
1161 pandecode_prop("sfail = %s", sfail);
1162 pandecode_prop("dpfail = %s", dpfail);
1163 pandecode_prop("dppass = %s", dppass);
1164 pandecode_indent--;
1165 pandecode_log("},\n");
1166 }
1167
1168 static void
1169 pandecode_blend_equation(const struct mali_blend_equation *blend)
1170 {
1171 if (blend->zero1)
1172 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1173
1174 pandecode_log(".equation = {\n");
1175 pandecode_indent++;
1176
1177 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1178 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1179
1180 pandecode_log(".color_mask = ");
1181 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1182 pandecode_log_cont(",\n");
1183
1184 pandecode_indent--;
1185 pandecode_log("},\n");
1186 }
1187
1188 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1189
1190 static unsigned
1191 decode_bifrost_constant(u16 constant)
1192 {
1193 float lo = (float) (constant & 0xFF);
1194 float hi = (float) (constant >> 8);
1195
1196 return (hi / 255.0) + (lo / 65535.0);
1197 }
1198
1199 static mali_ptr
1200 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1201 {
1202 struct bifrost_blend_rt *b =
1203 ((struct bifrost_blend_rt *) descs) + rt_no;
1204
1205 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1206 pandecode_indent++;
1207
1208 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1209 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1210 b->constant, decode_bifrost_constant(b->constant));
1211
1212 /* TODO figure out blend shader enable bit */
1213 pandecode_blend_equation(&b->equation);
1214 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1215 pandecode_prop("index = 0x%" PRIx16, b->index);
1216 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1217
1218 pandecode_indent--;
1219 pandecode_log("},\n");
1220
1221 return 0;
1222 }
1223
1224 static mali_ptr
1225 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1226 {
1227 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1228 return 0;
1229
1230 pandecode_log(".blend = {\n");
1231 pandecode_indent++;
1232
1233 if (is_shader) {
1234 pandecode_shader_address("shader", blend->shader);
1235 } else {
1236 pandecode_blend_equation(&blend->equation);
1237 pandecode_prop("constant = %f", blend->constant);
1238 }
1239
1240 pandecode_indent--;
1241 pandecode_log("},\n");
1242
1243 /* Return blend shader to disassemble if present */
1244 return is_shader ? (blend->shader & ~0xF) : 0;
1245 }
1246
1247 static mali_ptr
1248 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1249 {
1250 struct midgard_blend_rt *b =
1251 ((struct midgard_blend_rt *) descs) + rt_no;
1252
1253 /* Flags determine presence of blend shader */
1254 bool is_shader = (b->flags & 0xF) >= 0x2;
1255
1256 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1257 pandecode_indent++;
1258
1259 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1260
1261 union midgard_blend blend = b->blend;
1262 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1263
1264 pandecode_indent--;
1265 pandecode_log("};\n");
1266
1267 return shader;
1268 }
1269
1270 static int
1271 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1272 {
1273 char base[128];
1274 char *prefix = varying ? "varying" : "attribute";
1275 unsigned max_index = 0;
1276 snprintf(base, sizeof(base), "%s_meta", prefix);
1277
1278 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1279 pandecode_indent++;
1280
1281 struct mali_attr_meta *attr_meta;
1282 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1283
1284 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1285
1286 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1287 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1288 sizeof(*attr_mem));
1289
1290 pandecode_log("{\n");
1291 pandecode_indent++;
1292 pandecode_prop("index = %d", attr_meta->index);
1293
1294 if (attr_meta->index > max_index)
1295 max_index = attr_meta->index;
1296 pandecode_swizzle(attr_meta->swizzle);
1297 pandecode_prop("format = %s", pandecode_format(attr_meta->format));
1298
1299 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1300 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1301 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1302 pandecode_indent--;
1303 pandecode_log("},\n");
1304
1305 }
1306
1307 pandecode_indent--;
1308 pandecode_log("};\n");
1309
1310 return count ? (max_index + 1) : 0;
1311 }
1312
1313 static void
1314 pandecode_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1315 {
1316 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1317
1318 if (imem) {
1319 /* Indices are literally just a u32 array :) */
1320
1321 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1322
1323 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1324 pandecode_indent++;
1325
1326 for (unsigned i = 0; i < (index_count + 1); i += 3)
1327 pandecode_log("%d, %d, %d,\n",
1328 indices[i],
1329 indices[i + 1],
1330 indices[i + 2]);
1331
1332 pandecode_indent--;
1333 pandecode_log("};\n");
1334 }
1335 }
1336
1337 /* return bits [lo, hi) of word */
1338 static u32
1339 bits(u32 word, u32 lo, u32 hi)
1340 {
1341 if (hi - lo >= 32)
1342 return word; // avoid undefined behavior with the shift
1343
1344 return (word >> lo) & ((1 << (hi - lo)) - 1);
1345 }
1346
1347 static void
1348 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no, bool noninstanced)
1349 {
1350 pandecode_log_cont("{\n");
1351 pandecode_indent++;
1352
1353 /* Decode invocation_count. See the comment before the definition of
1354 * invocation_count for an explanation.
1355 */
1356
1357 unsigned size_x = bits(p->invocation_count, 0, p->size_y_shift) + 1;
1358 unsigned size_y = bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1;
1359 unsigned size_z = bits(p->invocation_count, p->size_z_shift, p->workgroups_x_shift) + 1;
1360
1361 unsigned groups_x = bits(p->invocation_count, p->workgroups_x_shift, p->workgroups_y_shift) + 1;
1362 unsigned groups_y = bits(p->invocation_count, p->workgroups_y_shift, p->workgroups_z_shift) + 1;
1363 unsigned groups_z = bits(p->invocation_count, p->workgroups_z_shift, 32) + 1;
1364
1365 /* Even though we have this decoded, we want to ensure that the
1366 * representation is "unique" so we don't lose anything by printing only
1367 * the final result. More specifically, we need to check that we were
1368 * passed something in canonical form, since the definition per the
1369 * hardware is inherently not unique. How? Well, take the resulting
1370 * decode and pack it ourselves! If it is bit exact with what we
1371 * decoded, we're good to go. */
1372
1373 struct mali_vertex_tiler_prefix ref;
1374 panfrost_pack_work_groups_compute(&ref, groups_x, groups_y, groups_z, size_x, size_y, size_z, noninstanced);
1375
1376 bool canonical =
1377 (p->invocation_count == ref.invocation_count) &&
1378 (p->size_y_shift == ref.size_y_shift) &&
1379 (p->size_z_shift == ref.size_z_shift) &&
1380 (p->workgroups_x_shift == ref.workgroups_x_shift) &&
1381 (p->workgroups_y_shift == ref.workgroups_y_shift) &&
1382 (p->workgroups_z_shift == ref.workgroups_z_shift) &&
1383 (p->workgroups_x_shift_2 == ref.workgroups_x_shift_2);
1384
1385 if (!canonical) {
1386 pandecode_msg("XXX: non-canonical workgroups packing\n");
1387 pandecode_msg("expected: %X, %d, %d, %d, %d, %d\n",
1388 ref.invocation_count,
1389 ref.size_y_shift,
1390 ref.size_z_shift,
1391 ref.workgroups_x_shift,
1392 ref.workgroups_y_shift,
1393 ref.workgroups_z_shift,
1394 ref.workgroups_x_shift_2);
1395
1396 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1397 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1398 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1399 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1400 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1401 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1402 pandecode_prop("workgroups_x_shift_2 = %d", p->workgroups_x_shift_2);
1403 }
1404
1405 /* Regardless, print the decode */
1406 pandecode_msg("size (%d, %d, %d), count (%d, %d, %d)\n",
1407 size_x, size_y, size_z,
1408 groups_x, groups_y, groups_z);
1409
1410 /* TODO: Decode */
1411 if (p->unknown_draw)
1412 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1413
1414 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1415
1416 if (p->draw_mode != MALI_DRAW_NONE)
1417 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
1418
1419 /* Index count only exists for tiler jobs anyway */
1420
1421 if (p->index_count)
1422 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1423
1424 if (p->offset_bias_correction)
1425 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1426
1427 if (p->zero1) {
1428 pandecode_msg("Zero tripped\n");
1429 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1430 }
1431
1432 pandecode_indent--;
1433 pandecode_log("},\n");
1434 }
1435
1436 static void
1437 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1438 {
1439 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1440
1441 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1442
1443 for (int i = 0; i < ubufs_count; i++) {
1444 mali_ptr ptr = ubufs[i].ptr << 2;
1445 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1446 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1447 char name[50];
1448 snprintf(name, sizeof(name), "ubuf_%d", i);
1449 /* The blob uses ubuf 0 to upload internal stuff and
1450 * uniforms that won't fit/are accessed indirectly, so
1451 * it puts it in the batchbuffer.
1452 */
1453 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1454 pandecode_indent++;
1455
1456 for (int j = 0; j <= ubufs[i].size; j++) {
1457 for (int k = 0; k < 4; k++) {
1458 if (k == 0)
1459 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1460 else
1461 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1462
1463 }
1464
1465 pandecode_log_cont("\n");
1466 }
1467
1468 pandecode_indent--;
1469 pandecode_log("};\n");
1470 }
1471
1472 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64"_%d[] = {\n",
1473 pubufs, job_no);
1474 pandecode_indent++;
1475
1476 for (int i = 0; i < ubufs_count; i++) {
1477 pandecode_log("{\n");
1478 pandecode_indent++;
1479 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1480 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1481 pandecode_indent--;
1482 pandecode_log("},\n");
1483 }
1484
1485 pandecode_indent--;
1486 pandecode_log("};\n");
1487 }
1488
1489 static void
1490 pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1491 {
1492
1493 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1494
1495 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1496
1497 if (scratchpad->zero)
1498 pandecode_msg("XXX scratchpad zero tripped");
1499
1500 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1501 pandecode_indent++;
1502
1503 pandecode_prop("flags = 0x%x", scratchpad->flags);
1504 MEMORY_PROP(scratchpad, gpu_scratchpad);
1505
1506 pandecode_indent--;
1507 pandecode_log("};\n");
1508 }
1509
1510 static unsigned shader_id = 0;
1511
1512 static void
1513 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1514 bool is_bifrost, unsigned nr_regs)
1515 {
1516 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1517 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1518
1519 /* Compute maximum possible size */
1520 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1521
1522 /* Print some boilerplate to clearly denote the assembly (which doesn't
1523 * obey indentation rules), and actually do the disassembly! */
1524
1525 printf("\n\n");
1526
1527 char prefix[512];
1528
1529 snprintf(prefix, sizeof(prefix) - 1, "shader%d - %s shader: ",
1530 shader_id++,
1531 (type == JOB_TYPE_TILER) ? "FRAGMENT" : "VERTEX");
1532
1533 if (is_bifrost) {
1534 disassemble_bifrost(code, sz, false);
1535 } else {
1536 disassemble_midgard(code, sz, true, nr_regs, prefix);
1537 }
1538
1539 printf("\n\n");
1540 }
1541
1542 static void
1543 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1544 int job_no, enum mali_job_type job_type,
1545 char *suffix, bool is_bifrost)
1546 {
1547 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1548 struct pandecode_mapped_memory *attr_mem;
1549
1550 unsigned rt_count = 1;
1551
1552 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1553 * are the only things actually needed from the FBD, vertex/tiler jobs
1554 * no longer reference the FBD -- instead, this field points to some
1555 * info about the scratchpad.
1556 */
1557 if (is_bifrost)
1558 pandecode_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1559 else if (p->framebuffer & MALI_MFBD)
1560 rt_count = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1561 else if (job_type == JOB_TYPE_COMPUTE)
1562 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1563 else
1564 pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1565
1566 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1567 int texture_count = 0, sampler_count = 0;
1568
1569 if (shader_meta_ptr) {
1570 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1571 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1572
1573 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1574 pandecode_indent++;
1575
1576 /* Save for dumps */
1577 attribute_count = s->attribute_count;
1578 varying_count = s->varying_count;
1579 texture_count = s->texture_count;
1580 sampler_count = s->sampler_count;
1581
1582 if (is_bifrost) {
1583 uniform_count = s->bifrost2.uniform_count;
1584 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1585 } else {
1586 uniform_count = s->midgard1.uniform_count;
1587 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1588 }
1589
1590 mali_ptr shader_ptr = pandecode_shader_address("shader", s->shader);
1591
1592 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1593 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1594 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1595 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1596
1597 unsigned nr_registers = 0;
1598
1599 if (is_bifrost) {
1600 pandecode_log(".bifrost1 = {\n");
1601 pandecode_indent++;
1602
1603 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1604 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1605
1606 pandecode_indent--;
1607 pandecode_log("},\n");
1608 } else {
1609 pandecode_log(".midgard1 = {\n");
1610 pandecode_indent++;
1611
1612 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1613 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1614 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1615 nr_registers = s->midgard1.work_count;
1616
1617 pandecode_log(".flags = ");
1618 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1619 pandecode_log_cont(",\n");
1620
1621 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1622
1623 pandecode_indent--;
1624 pandecode_log("},\n");
1625 }
1626
1627 if (s->depth_units || s->depth_factor) {
1628 pandecode_prop("depth_factor = %f", s->depth_factor);
1629 pandecode_prop("depth_units = %f", s->depth_units);
1630 }
1631
1632 if (s->alpha_coverage) {
1633 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1634 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1635
1636 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1637 invert_alpha_coverage ? "~" : "",
1638 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1639 }
1640
1641 if (s->unknown2_3 || s->unknown2_4) {
1642 pandecode_log(".unknown2_3 = ");
1643
1644 int unknown2_3 = s->unknown2_3;
1645 int unknown2_4 = s->unknown2_4;
1646
1647 /* We're not quite sure what these flags mean without the depth test, if anything */
1648
1649 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1650 const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
1651 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1652
1653 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1654 }
1655
1656 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1657 pandecode_log_cont(",\n");
1658
1659 pandecode_log(".unknown2_4 = ");
1660 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1661 pandecode_log_cont(",\n");
1662 }
1663
1664 if (s->stencil_mask_front || s->stencil_mask_back) {
1665 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1666 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1667 }
1668
1669 pandecode_stencil("front", &s->stencil_front);
1670 pandecode_stencil("back", &s->stencil_back);
1671
1672 if (is_bifrost) {
1673 pandecode_log(".bifrost2 = {\n");
1674 pandecode_indent++;
1675
1676 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1677 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1678 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1679 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1680
1681 pandecode_indent--;
1682 pandecode_log("},\n");
1683 } else if (s->midgard2.unknown2_7) {
1684 pandecode_log(".midgard2 = {\n");
1685 pandecode_indent++;
1686
1687 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1688 pandecode_indent--;
1689 pandecode_log("},\n");
1690 }
1691
1692 if (s->unknown2_8)
1693 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1694
1695 if (!is_bifrost) {
1696 /* TODO: Blend shaders routing/disasm */
1697
1698 union midgard_blend blend = s->blend;
1699 pandecode_midgard_blend(&blend, false);
1700 }
1701
1702 pandecode_indent--;
1703 pandecode_log("};\n");
1704
1705 /* MRT blend fields are used whenever MFBD is used, with
1706 * per-RT descriptors */
1707
1708 if (job_type == JOB_TYPE_TILER) {
1709 void* blend_base = (void *) (s + 1);
1710
1711 for (unsigned i = 0; i < rt_count; i++) {
1712 mali_ptr shader = 0;
1713
1714 if (is_bifrost)
1715 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1716 else
1717 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1718
1719 if (shader & ~0xF)
1720 pandecode_shader_disassemble(shader, job_no, job_type, false, 0);
1721 }
1722 }
1723
1724 if (shader_ptr & ~0xF)
1725 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost, nr_registers);
1726 } else
1727 pandecode_msg("<no shader>\n");
1728
1729 if (p->viewport) {
1730 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1731 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1732
1733 pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
1734 pandecode_indent++;
1735
1736 pandecode_prop("clip_minx = %f", f->clip_minx);
1737 pandecode_prop("clip_miny = %f", f->clip_miny);
1738 pandecode_prop("clip_minz = %f", f->clip_minz);
1739 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1740 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1741 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1742
1743 /* Only the higher coordinates are MALI_POSITIVE scaled */
1744
1745 pandecode_prop("viewport0 = { %d, %d }",
1746 f->viewport0[0], f->viewport0[1]);
1747
1748 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1749 f->viewport1[0] + 1, f->viewport1[1] + 1);
1750
1751 pandecode_indent--;
1752 pandecode_log("};\n");
1753 }
1754
1755 if (p->attribute_meta) {
1756 unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1757
1758 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1759 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1760 }
1761
1762 /* Varyings are encoded like attributes but not actually sent; we just
1763 * pass a zero buffer with the right stride/size set, (or whatever)
1764 * since the GPU will write to it itself */
1765
1766 if (p->varying_meta) {
1767 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1768 }
1769
1770 if (p->varyings) {
1771 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1772
1773 /* Number of descriptors depends on whether there are
1774 * non-internal varyings */
1775
1776 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
1777 }
1778
1779 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1780
1781 if (p->uniforms && !is_compute) {
1782 int rows = uniform_count, width = 4;
1783 size_t sz = rows * width * sizeof(float);
1784
1785 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1786 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1787 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1788
1789 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1790
1791 pandecode_indent++;
1792
1793 for (int row = 0; row < rows; row++) {
1794 for (int i = 0; i < width; i++) {
1795 u32 v = uniforms[i];
1796 float f;
1797 memcpy(&f, &v, sizeof(v));
1798 pandecode_log_cont("%X /* %f */, ", v, f);
1799 }
1800
1801 pandecode_log_cont("\n");
1802
1803 uniforms += width;
1804 }
1805
1806 pandecode_indent--;
1807 pandecode_log("};\n");
1808 } else if (p->uniforms) {
1809 int rows = uniform_count * 2;
1810 size_t sz = rows * sizeof(mali_ptr);
1811
1812 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1813 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1814 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1815
1816 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1817
1818 pandecode_indent++;
1819
1820 for (int row = 0; row < rows; row++) {
1821 char *a = pointer_as_memory_reference(uniforms[row]);
1822 pandecode_log("%s,\n", a);
1823 free(a);
1824 }
1825
1826 pandecode_indent--;
1827 pandecode_log("};\n");
1828
1829 }
1830
1831 if (p->uniform_buffers) {
1832 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1833 }
1834
1835 if (p->texture_trampoline) {
1836 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1837
1838 if (mmem) {
1839 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1840
1841 pandecode_log("uint64_t texture_trampoline_%"PRIx64"_%d[] = {\n", p->texture_trampoline, job_no);
1842 pandecode_indent++;
1843
1844 for (int tex = 0; tex < texture_count; ++tex) {
1845 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1846 char *a = pointer_as_memory_reference(*u);
1847 pandecode_log("%s,\n", a);
1848 free(a);
1849 }
1850
1851 pandecode_indent--;
1852 pandecode_log("};\n");
1853
1854 /* Now, finally, descend down into the texture descriptor */
1855 for (int tex = 0; tex < texture_count; ++tex) {
1856 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1857 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1858
1859 if (tmem) {
1860 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1861
1862 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1863 pandecode_indent++;
1864
1865 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1866 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1867 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1868 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1869 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1870 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1871 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1872
1873 struct mali_texture_format f = t->format;
1874
1875 pandecode_log(".format = {\n");
1876 pandecode_indent++;
1877
1878 pandecode_swizzle(f.swizzle);
1879 pandecode_prop("format = %s", pandecode_format(f.format));
1880 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1881 pandecode_prop("srgb = %" PRId32, f.srgb);
1882 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1883 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1884
1885 pandecode_indent--;
1886 pandecode_log("},\n");
1887
1888 pandecode_swizzle(t->swizzle);
1889
1890 if (t->swizzle_zero) {
1891 /* Shouldn't happen */
1892 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1893 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1894 }
1895
1896 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1897
1898 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1899 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1900 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1901
1902 pandecode_log(".payload = {\n");
1903 pandecode_indent++;
1904
1905 /* A bunch of bitmap pointers follow.
1906 * We work out the correct number,
1907 * based on the mipmap/cubemap
1908 * properties, but dump extra
1909 * possibilities to futureproof */
1910
1911 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1912 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1913
1914 /* Miptree for each face */
1915 if (f.type == MALI_TEX_CUBE)
1916 bitmap_count *= 6;
1917
1918 /* Array of textures */
1919 bitmap_count *= MALI_NEGATIVE(t->array_size);
1920
1921 /* Stride for each element */
1922 if (manual_stride)
1923 bitmap_count *= 2;
1924
1925 /* Sanity check the size */
1926 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1927 assert (bitmap_count <= max_count);
1928
1929 for (int i = 0; i < bitmap_count; ++i) {
1930 /* How we dump depends if this is a stride or a pointer */
1931
1932 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1933 /* signed 32-bit snuck in as a 64-bit pointer */
1934 uint64_t stride_set = t->payload[i];
1935 uint32_t clamped_stride = stride_set;
1936 int32_t stride = clamped_stride;
1937 assert(stride_set == clamped_stride);
1938 pandecode_log("(mali_ptr) %d /* stride */, \n", stride);
1939 } else {
1940 char *a = pointer_as_memory_reference(t->payload[i]);
1941 pandecode_log("%s, \n", a);
1942 free(a);
1943 }
1944 }
1945
1946 pandecode_indent--;
1947 pandecode_log("},\n");
1948
1949 pandecode_indent--;
1950 pandecode_log("};\n");
1951 }
1952 }
1953 }
1954 }
1955
1956 if (p->sampler_descriptor) {
1957 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1958
1959 if (smem) {
1960 struct mali_sampler_descriptor *s;
1961
1962 mali_ptr d = p->sampler_descriptor;
1963
1964 for (int i = 0; i < sampler_count; ++i) {
1965 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1966
1967 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", d + sizeof(*s) * i, job_no, i);
1968 pandecode_indent++;
1969
1970 pandecode_log(".filter_mode = ");
1971 pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
1972 pandecode_log_cont(",\n");
1973
1974 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1975 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1976
1977 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
1978 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
1979 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
1980
1981 pandecode_prop("compare_func = %s", pandecode_alt_func(s->compare_func));
1982
1983 if (s->zero || s->zero2) {
1984 pandecode_msg("Zero tripped\n");
1985 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1986 }
1987
1988 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1989
1990 pandecode_prop("border_color = { %f, %f, %f, %f }",
1991 s->border_color[0],
1992 s->border_color[1],
1993 s->border_color[2],
1994 s->border_color[3]);
1995
1996 pandecode_indent--;
1997 pandecode_log("};\n");
1998 }
1999 }
2000 }
2001 }
2002
2003 static void
2004 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
2005 {
2006 if (!(p->position_varying || p->occlusion_counter || p->flags))
2007 return;
2008
2009 pandecode_log(".postfix = {\n");
2010 pandecode_indent++;
2011
2012 MEMORY_PROP(p, position_varying);
2013 MEMORY_PROP(p, occlusion_counter);
2014
2015 if (p->flags)
2016 pandecode_prop("flags = %d", p->flags);
2017
2018 pandecode_indent--;
2019 pandecode_log("},\n");
2020 }
2021
2022 static void
2023 pandecode_vertex_only_bfr(struct bifrost_vertex_only *v)
2024 {
2025 pandecode_log_cont("{\n");
2026 pandecode_indent++;
2027
2028 pandecode_prop("unk2 = 0x%x", v->unk2);
2029
2030 if (v->zero0 || v->zero1) {
2031 pandecode_msg("vertex only zero tripped");
2032 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
2033 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
2034 }
2035
2036 pandecode_indent--;
2037 pandecode_log("}\n");
2038 }
2039
2040 static void
2041 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
2042 {
2043
2044 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2045 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
2046
2047 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
2048 pandecode_indent++;
2049
2050 if (h->zero) {
2051 pandecode_msg("tiler heap zero tripped\n");
2052 pandecode_prop("zero = 0x%x", h->zero);
2053 }
2054
2055 for (int i = 0; i < 12; i++) {
2056 if (h->zeros[i] != 0) {
2057 pandecode_msg("tiler heap zero %d tripped, value %x\n",
2058 i, h->zeros[i]);
2059 }
2060 }
2061
2062 pandecode_prop("heap_size = 0x%x", h->heap_size);
2063 MEMORY_PROP(h, tiler_heap_start);
2064 MEMORY_PROP(h, tiler_heap_free);
2065
2066 /* this might point to the beginning of another buffer, when it's
2067 * really the end of the tiler heap buffer, so we have to be careful
2068 * here. but for zero length, we need the same pointer.
2069 */
2070
2071 if (h->tiler_heap_end == h->tiler_heap_start) {
2072 MEMORY_PROP(h, tiler_heap_start);
2073 } else {
2074 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2075 pandecode_prop("tiler_heap_end = %s + 1", a);
2076 free(a);
2077 }
2078
2079 pandecode_indent--;
2080 pandecode_log("};\n");
2081 }
2082
2083 static void
2084 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2085 {
2086 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2087 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2088
2089 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2090
2091 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
2092 pandecode_indent++;
2093
2094 if (t->zero0 || t->zero1) {
2095 pandecode_msg("tiler meta zero tripped");
2096 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2097 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2098 }
2099
2100 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2101 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2102
2103 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2104 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2105
2106 for (int i = 0; i < 12; i++) {
2107 if (t->zeros[i] != 0) {
2108 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2109 i, t->zeros[i]);
2110 }
2111 }
2112
2113 pandecode_indent--;
2114 pandecode_log("};\n");
2115 }
2116
2117 static void
2118 pandecode_gl_enables(uint32_t gl_enables, int job_type)
2119 {
2120 pandecode_log(".gl_enables = ");
2121
2122 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2123
2124 pandecode_log_cont(",\n");
2125 }
2126
2127 static void
2128 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2129 {
2130 if (u.pointer == 0x0)
2131 return;
2132
2133 pandecode_log(".primitive_size = {\n");
2134 pandecode_indent++;
2135
2136 if (constant) {
2137 pandecode_prop("constant = %f", u.constant);
2138 } else {
2139 MEMORY_PROP((&u), pointer);
2140 }
2141
2142 pandecode_indent--;
2143 pandecode_log("},\n");
2144 }
2145
2146 static void
2147 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2148 {
2149 pandecode_log_cont("{\n");
2150 pandecode_indent++;
2151
2152 /* TODO: gl_PointSize on Bifrost */
2153 pandecode_primitive_size(t->primitive_size, true);
2154
2155 pandecode_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2156
2157 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2158 || t->zero6 || t->zero7 || t->zero8) {
2159 pandecode_msg("tiler only zero tripped");
2160 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2161 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2162 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2163 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2164 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2165 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2166 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2167 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2168 }
2169
2170 pandecode_indent--;
2171 pandecode_log("},\n");
2172 }
2173
2174 static int
2175 pandecode_vertex_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_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2180
2181 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2182
2183 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2184 pandecode_indent++;
2185
2186 pandecode_log(".prefix = ");
2187 pandecode_vertex_tiler_prefix(&v->prefix, job_no, false);
2188
2189 pandecode_log(".vertex = ");
2190 pandecode_vertex_only_bfr(&v->vertex);
2191
2192 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2193
2194 pandecode_indent--;
2195 pandecode_log("};\n");
2196
2197 return sizeof(*v);
2198 }
2199
2200 static int
2201 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2202 const struct pandecode_mapped_memory *mem,
2203 mali_ptr payload, int job_no)
2204 {
2205 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2206
2207 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2208
2209 pandecode_indices(t->prefix.indices, t->prefix.index_count, job_no);
2210 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2211
2212 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2213 pandecode_indent++;
2214
2215 pandecode_log(".prefix = ");
2216 pandecode_vertex_tiler_prefix(&t->prefix, job_no, false);
2217
2218 pandecode_log(".tiler = ");
2219 pandecode_tiler_only_bfr(&t->tiler, job_no);
2220
2221 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2222
2223 pandecode_indent--;
2224 pandecode_log("};\n");
2225
2226 return sizeof(*t);
2227 }
2228
2229 static int
2230 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2231 const struct pandecode_mapped_memory *mem,
2232 mali_ptr payload, int job_no)
2233 {
2234 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2235
2236 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2237
2238 pandecode_indices(v->prefix.indices, v->prefix.index_count, job_no);
2239
2240 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2241 pandecode_indent++;
2242
2243 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2244 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2245
2246 bool instanced = v->instance_shift || v->instance_odd;
2247 bool is_graphics = (h->job_type == JOB_TYPE_VERTEX) || (h->job_type == JOB_TYPE_TILER);
2248
2249 pandecode_log(".prefix = ");
2250 pandecode_vertex_tiler_prefix(&v->prefix, job_no, !instanced && is_graphics);
2251
2252 pandecode_gl_enables(v->gl_enables, h->job_type);
2253
2254 if (v->instance_shift || v->instance_odd) {
2255 pandecode_prop("instance_shift = 0x%d /* %d */",
2256 v->instance_shift, 1 << v->instance_shift);
2257 pandecode_prop("instance_odd = 0x%X /* %d */",
2258 v->instance_odd, (2 * v->instance_odd) + 1);
2259
2260 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2261 }
2262
2263 if (v->offset_start)
2264 pandecode_prop("offset_start = %d", v->offset_start);
2265
2266 if (v->zero5) {
2267 pandecode_msg("Zero tripped\n");
2268 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2269 }
2270
2271 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2272
2273 pandecode_indent--;
2274 pandecode_log("};\n");
2275
2276 return sizeof(*v);
2277 }
2278
2279 static int
2280 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2281 mali_ptr payload, int job_no,
2282 bool is_bifrost)
2283 {
2284 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2285
2286 bool fbd_dumped = false;
2287
2288 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2289 /* Only SFBDs are understood, not MFBDs. We're speculating,
2290 * based on the versioning, kernel code, etc, that the
2291 * difference is between Single FrameBuffer Descriptor and
2292 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2293 * up with multi-framebuffer support being added (T7xx onwards,
2294 * including Gxx). In any event, there's some field shuffling
2295 * that we haven't looked into yet. */
2296
2297 pandecode_sfbd(s->framebuffer & FBD_MASK, job_no);
2298 fbd_dumped = true;
2299 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2300 /* We don't know if Bifrost supports SFBD's at all, since the
2301 * driver never uses them. And the format is different from
2302 * Midgard anyways, due to the tiler heap and scratchpad being
2303 * moved out into separate structures, so it's not clear what a
2304 * Bifrost SFBD would even look like without getting an actual
2305 * trace, which appears impossible.
2306 */
2307
2308 pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2309 fbd_dumped = true;
2310 }
2311
2312 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2313 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2314 pandecode_indent++;
2315
2316 /* See the comments by the macro definitions for mathematical context
2317 * on why this is so weird */
2318
2319 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2320 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2321
2322 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2323 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2324 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2325
2326 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2327 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2328 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2329
2330 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2331 * we have to fallback on offsets. */
2332
2333 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2334
2335 /* TODO: Decode */
2336 unsigned extra_flags = (s->framebuffer & ~FBD_MASK) & ~MALI_MFBD;
2337
2338 if (fbd_dumped)
2339 pandecode_prop("framebuffer = framebuffer_%d_p | %s | 0x%X", job_no,
2340 fbd_type, extra_flags);
2341 else
2342 pandecode_prop("framebuffer = %s | %s | 0x%X", pointer_as_memory_reference(p),
2343 fbd_type, extra_flags);
2344
2345 pandecode_indent--;
2346 pandecode_log("};\n");
2347
2348 return sizeof(*s);
2349 }
2350
2351 static int job_descriptor_number = 0;
2352
2353 int
2354 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
2355 {
2356 struct mali_job_descriptor_header *h;
2357
2358 int start_number = 0;
2359
2360 bool first = true;
2361 bool last_size;
2362
2363 do {
2364 struct pandecode_mapped_memory *mem =
2365 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2366
2367 void *payload;
2368
2369 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2370
2371 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2372 * high 32-bits of the 64-bit pointer are reused to store
2373 * something else.
2374 */
2375 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2376 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2377 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2378
2379 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2380 MALI_PAYLOAD_SIZE);
2381
2382 int job_no = job_descriptor_number++;
2383
2384 if (first)
2385 start_number = job_no;
2386
2387 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2388 pandecode_indent++;
2389
2390 pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
2391
2392 /* Save for next job fixing */
2393 last_size = h->job_descriptor_size;
2394
2395 if (h->job_descriptor_size)
2396 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2397
2398 if (h->exception_status && h->exception_status != 0x1)
2399 pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
2400 h->exception_status,
2401 (h->exception_status >> 16) & 0xFFFF,
2402 pandecode_exception_access((h->exception_status >> 8) & 0x3),
2403 h->exception_status & 0xFF);
2404
2405 if (h->first_incomplete_task)
2406 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2407
2408 if (h->fault_pointer)
2409 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2410
2411 if (h->job_barrier)
2412 pandecode_prop("job_barrier = %d", h->job_barrier);
2413
2414 pandecode_prop("job_index = %d", h->job_index);
2415
2416 if (h->unknown_flags)
2417 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2418
2419 if (h->job_dependency_index_1)
2420 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2421
2422 if (h->job_dependency_index_2)
2423 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2424
2425 pandecode_indent--;
2426 pandecode_log("};\n");
2427
2428 /* Do not touch the field yet -- decode the payload first, and
2429 * don't touch that either. This is essential for the uploads
2430 * to occur in sequence and therefore be dynamically allocated
2431 * correctly. Do note the size, however, for that related
2432 * reason. */
2433
2434 switch (h->job_type) {
2435 case JOB_TYPE_SET_VALUE: {
2436 struct mali_payload_set_value *s = payload;
2437 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2438 pandecode_indent++;
2439 MEMORY_PROP(s, out);
2440 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2441 pandecode_indent--;
2442 pandecode_log("};\n");
2443
2444 break;
2445 }
2446
2447 case JOB_TYPE_TILER:
2448 case JOB_TYPE_VERTEX:
2449 case JOB_TYPE_COMPUTE:
2450 if (bifrost) {
2451 if (h->job_type == JOB_TYPE_TILER)
2452 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
2453 else
2454 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
2455 } else
2456 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2457
2458 break;
2459
2460 case JOB_TYPE_FRAGMENT:
2461 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost);
2462 break;
2463
2464 default:
2465 break;
2466 }
2467
2468 /* Handle linkage */
2469
2470 if (!first) {
2471 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2472
2473 if (last_size)
2474 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2475 else
2476 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2477 }
2478
2479 first = false;
2480
2481 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2482
2483 return start_number;
2484 }