projects / mesa.git / blob
? search:


771383ce86689a355d45b9ed0d9e6bde43c177fb
[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 /* Midgard's tiler descriptor is embedded within the
463 * larger FBD */
464
465 static void
466 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor *t)
467 {
468 pandecode_log(".tiler = {\n");
469 pandecode_indent++;
470
471 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
472 pandecode_prop("flags = 0x%" PRIx16, t->flags);
473 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
474
475 MEMORY_PROP(t, polygon_list);
476 MEMORY_PROP(t, polygon_list_body);
477
478 MEMORY_PROP(t, heap_start);
479
480 if (t->heap_start == t->heap_end) {
481 /* Print identically to show symmetry for empty tiler heaps */
482 MEMORY_PROP(t, heap_start);
483 } else {
484 /* Points to the end of a buffer */
485 char *a = pointer_as_memory_reference(t->heap_end - 1);
486 pandecode_prop("heap_end = %s + 1", a);
487 free(a);
488 }
489
490 bool nonzero_weights = false;
491
492 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
493 nonzero_weights |= t->weights[w] != 0x0;
494 }
495
496 if (nonzero_weights) {
497 pandecode_log(".weights = {");
498
499 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
500 pandecode_log("%d, ", t->weights[w]);
501 }
502
503 pandecode_log("},");
504 }
505
506 pandecode_indent--;
507 pandecode_log("}\n");
508 }
509
510 static void
511 pandecode_sfbd(uint64_t gpu_va, int job_no)
512 {
513 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
514 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
515
516 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
517 pandecode_indent++;
518
519 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
520 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
521
522 pandecode_log(".format = ");
523 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
524 pandecode_log_cont(",\n");
525
526 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
527 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
528
529 MEMORY_PROP(s, framebuffer);
530 pandecode_prop("stride = %d", s->stride);
531
532 /* Earlier in the actual commandstream -- right before width -- but we
533 * delay to flow nicer */
534
535 pandecode_log(".clear_flags = ");
536 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
537 pandecode_log_cont(",\n");
538
539 if (s->depth_buffer | s->depth_buffer_enable) {
540 MEMORY_PROP(s, depth_buffer);
541 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
542 }
543
544 if (s->stencil_buffer | s->stencil_buffer_enable) {
545 MEMORY_PROP(s, stencil_buffer);
546 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
547 }
548
549 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
550 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
551 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
552 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
553 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
554 }
555
556 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
557 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
558 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
559 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
560 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
561 }
562
563 if (s->clear_stencil) {
564 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
565 }
566
567 MEMORY_PROP(s, unknown_address_0);
568 const struct midgard_tiler_descriptor t = s->tiler;
569 pandecode_midgard_tiler_descriptor(&t);
570
571 pandecode_indent--;
572 pandecode_log("};\n");
573
574 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
575 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
576 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
577 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
578
579 printf(".zero3 = {");
580
581 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
582 printf("%X, ", s->zero3[i]);
583
584 printf("},\n");
585
586 printf(".zero6 = {");
587
588 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
589 printf("%X, ", s->zero6[i]);
590
591 printf("},\n");
592 }
593
594 static void
595 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
596 {
597 pandecode_log(".unknown%d = {", name);
598
599 for (int i = 0; i < count; ++i)
600 printf("%X, ", slide[i]);
601
602 pandecode_log("},\n");
603 }
604
605 #define SHORT_SLIDE(num) \
606 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
607
608 static void
609 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
610 {
611 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
612 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
613
614 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
615 pandecode_indent++;
616
617 SHORT_SLIDE(1);
618
619 pandecode_indent--;
620 printf("},\n");
621 }
622
623 static void
624 pandecode_swizzle(unsigned swizzle)
625 {
626 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
627 pandecode_channel((swizzle >> 0) & 0x7),
628 pandecode_channel((swizzle >> 3) & 0x7),
629 pandecode_channel((swizzle >> 6) & 0x7),
630 pandecode_channel((swizzle >> 9) & 0x7));
631 }
632
633 static void
634 pandecode_rt_format(struct mali_rt_format format)
635 {
636 pandecode_log(".format = {\n");
637 pandecode_indent++;
638
639 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
640 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
641 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
642
643 pandecode_prop("block = %s",
644 pandecode_mfbd_block_format(format.block));
645
646 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
647 MALI_NEGATIVE(format.nr_channels));
648
649 pandecode_log(".flags = ");
650 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
651 pandecode_log_cont(",\n");
652
653 pandecode_swizzle(format.swizzle);
654
655 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
656
657 pandecode_indent--;
658 pandecode_log("},\n");
659 }
660
661 static void
662 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
663 {
664 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
665 pandecode_indent++;
666
667 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
668 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
669 struct pandecode_mapped_memory *mem =
670 pandecode_find_mapped_gpu_mem_containing(rt_va);
671 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
672
673 pandecode_log("{\n");
674 pandecode_indent++;
675
676 pandecode_rt_format(rt->format);
677
678 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
679 pandecode_log(".afbc = {\n");
680 pandecode_indent++;
681
682 char *a = pointer_as_memory_reference(rt->afbc.metadata);
683 pandecode_prop("metadata = %s", a);
684 free(a);
685
686 pandecode_prop("stride = %d", rt->afbc.stride);
687 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
688
689 pandecode_indent--;
690 pandecode_log("},\n");
691 } else {
692 pandecode_log(".chunknown = {\n");
693 pandecode_indent++;
694
695 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
696
697 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
698 pandecode_prop("pointer = %s", a);
699 free(a);
700
701 pandecode_indent--;
702 pandecode_log("},\n");
703 }
704
705 MEMORY_PROP(rt, framebuffer);
706 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
707
708 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
709 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
710 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
711 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
712 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
713 }
714
715 if (rt->zero1 || rt->zero2 || rt->zero3) {
716 pandecode_msg("render target zeros tripped\n");
717 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
718 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
719 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
720 }
721
722 pandecode_indent--;
723 pandecode_log("},\n");
724 }
725
726 pandecode_indent--;
727 pandecode_log("};\n");
728 }
729
730 static unsigned
731 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
732 {
733 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
734 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
735
736 if (fb->sample_locations) {
737 /* The blob stores all possible sample locations in a single buffer
738 * allocated on startup, and just switches the pointer when switching
739 * MSAA state. For now, we just put the data into the cmdstream, but we
740 * should do something like what the blob does with a real driver.
741 *
742 * There seem to be 32 slots for sample locations, followed by another
743 * 16. The second 16 is just the center location followed by 15 zeros
744 * in all the cases I've identified (maybe shader vs. depth/color
745 * samples?).
746 */
747
748 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
749
750 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
751
752 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
753 pandecode_indent++;
754
755 for (int i = 0; i < 32 + 16; i++) {
756 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
757 }
758
759 pandecode_indent--;
760 pandecode_log("};\n");
761 }
762
763 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
764 pandecode_indent++;
765
766 pandecode_prop("unk0 = 0x%x", fb->unk0);
767
768 if (fb->sample_locations)
769 pandecode_prop("sample_locations = sample_locations_%d", job_no);
770
771 /* Assume that unknown1 was emitted in the last job for
772 * now */
773 MEMORY_PROP(fb, unknown1);
774
775 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
776 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
777 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
778 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
779
780 pandecode_prop("unk1 = 0x%x", fb->unk1);
781 pandecode_prop("unk2 = 0x%x", fb->unk2);
782 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
783 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
784
785 pandecode_log(".mfbd_flags = ");
786 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
787 pandecode_log_cont(",\n");
788
789 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
790 pandecode_prop("clear_depth = %f", fb->clear_depth);
791
792 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
793 MEMORY_PROP(fb, scratchpad);
794 const struct midgard_tiler_descriptor t = fb->tiler;
795 pandecode_midgard_tiler_descriptor(&t);
796
797 if (fb->zero3 || fb->zero4) {
798 pandecode_msg("framebuffer zeros tripped\n");
799 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
800 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
801 }
802
803 pandecode_indent--;
804 pandecode_log("};\n");
805
806 gpu_va += sizeof(struct bifrost_framebuffer);
807
808 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
809 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
810 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
811
812 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
813 pandecode_indent++;
814
815 MEMORY_PROP(fbx, checksum);
816
817 if (fbx->checksum_stride)
818 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
819
820 pandecode_log(".flags = ");
821 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
822 pandecode_log_cont(",\n");
823
824 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
825 pandecode_log(".ds_afbc = {\n");
826 pandecode_indent++;
827
828 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
829 pandecode_prop("depth_stencil_afbc_stride = %d",
830 fbx->ds_afbc.depth_stencil_afbc_stride);
831 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
832
833 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
834 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
835 pandecode_prop("zero1 = 0x%" PRIx32,
836 fbx->ds_afbc.zero1);
837 pandecode_prop("padding = 0x%" PRIx64,
838 fbx->ds_afbc.padding);
839 }
840
841 pandecode_indent--;
842 pandecode_log("},\n");
843 } else {
844 pandecode_log(".ds_linear = {\n");
845 pandecode_indent++;
846
847 if (fbx->ds_linear.depth) {
848 MEMORY_PROP_DIR(fbx->ds_linear, depth);
849 pandecode_prop("depth_stride = %d",
850 fbx->ds_linear.depth_stride);
851 }
852
853 if (fbx->ds_linear.stencil) {
854 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
855 pandecode_prop("stencil_stride = %d",
856 fbx->ds_linear.stencil_stride);
857 }
858
859 if (fbx->ds_linear.depth_stride_zero ||
860 fbx->ds_linear.stencil_stride_zero ||
861 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
862 pandecode_msg("Depth/stencil zeros tripped\n");
863 pandecode_prop("depth_stride_zero = 0x%x",
864 fbx->ds_linear.depth_stride_zero);
865 pandecode_prop("stencil_stride_zero = 0x%x",
866 fbx->ds_linear.stencil_stride_zero);
867 pandecode_prop("zero1 = 0x%" PRIx32,
868 fbx->ds_linear.zero1);
869 pandecode_prop("zero2 = 0x%" PRIx32,
870 fbx->ds_linear.zero2);
871 }
872
873 pandecode_indent--;
874 pandecode_log("},\n");
875 }
876
877 if (fbx->zero3 || fbx->zero4) {
878 pandecode_msg("fb_extra zeros tripped\n");
879 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
880 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
881 }
882
883 pandecode_indent--;
884 pandecode_log("};\n");
885
886 gpu_va += sizeof(struct bifrost_fb_extra);
887 }
888
889 if (with_render_targets)
890 pandecode_render_target(gpu_va, job_no, fb);
891
892 /* Passback the render target count */
893 return MALI_NEGATIVE(fb->rt_count_1);
894 }
895
896 /* Just add a comment decoding the shift/odd fields forming the padded vertices
897 * count */
898
899 static void
900 pandecode_padded_vertices(unsigned shift, unsigned k)
901 {
902 unsigned odd = 2*k + 1;
903 unsigned pot = 1 << shift;
904 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
905 }
906
907 /* Given a magic divisor, recover what we were trying to divide by.
908 *
909 * Let m represent the magic divisor. By definition, m is an element on Z, whre
910 * 0 <= m < 2^N, for N bits in m.
911 *
912 * Let q represent the number we would like to divide by.
913 *
914 * By definition of a magic divisor for N-bit unsigned integers (a number you
915 * multiply by to magically get division), m is a number such that:
916 *
917 * (m * x) & (2^N - 1) = floor(x/q).
918 * for all x on Z where 0 <= x < 2^N
919 *
920 * Ignore the case where any of the above values equals zero; it is irrelevant
921 * for our purposes (instanced arrays).
922 *
923 * Choose x = q. Then:
924 *
925 * (m * x) & (2^N - 1) = floor(x/q).
926 * (m * q) & (2^N - 1) = floor(q/q).
927 *
928 * floor(q/q) = floor(1) = 1, therefore:
929 *
930 * (m * q) & (2^N - 1) = 1
931 *
932 * Recall the identity that the bitwise AND of one less than a power-of-two
933 * equals the modulo with that power of two, i.e. for all x:
934 *
935 * x & (2^N - 1) = x % N
936 *
937 * Therefore:
938 *
939 * mq % (2^N) = 1
940 *
941 * By definition, a modular multiplicative inverse of a number m is the number
942 * q such that with respect to a modulos M:
943 *
944 * mq % M = 1
945 *
946 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
947 *
948 */
949
950 static void
951 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
952 {
953 #if 0
954 /* Compute the modular inverse of `magic` with respect to 2^(32 -
955 * shift) the most lame way possible... just repeatedly add.
956 * Asymptoptically slow but nobody cares in practice, unless you have
957 * massive numbers of vertices or high divisors. */
958
959 unsigned inverse = 0;
960
961 /* Magic implicitly has the highest bit set */
962 magic |= (1 << 31);
963
964 /* Depending on rounding direction */
965 if (extra)
966 magic++;
967
968 for (;;) {
969 uint32_t product = magic * inverse;
970
971 if (shift) {
972 product >>= shift;
973 }
974
975 if (product == 1)
976 break;
977
978 ++inverse;
979 }
980
981 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
982
983 /* Recall we're supposed to divide by (gl_level_divisor *
984 * padded_num_vertices) */
985
986 unsigned padded_num_vertices = inverse / orig_divisor;
987
988 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
989 #endif
990 }
991
992 static void
993 pandecode_attributes(const struct pandecode_mapped_memory *mem,
994 mali_ptr addr, int job_no, char *suffix,
995 int count, bool varying)
996 {
997 char *prefix = varying ? "varyings" : "attributes";
998
999 if (!addr) {
1000 pandecode_msg("no %s\n", prefix);
1001 return;
1002 }
1003
1004 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
1005
1006 char base[128];
1007 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
1008
1009 for (int i = 0; i < count; ++i) {
1010 enum mali_attr_mode mode = attr[i].elements & 7;
1011
1012 if (mode == MALI_ATTR_UNUSED)
1013 continue;
1014
1015 mali_ptr raw_elements = attr[i].elements & ~7;
1016
1017 /* TODO: Do we maybe want to dump the attribute values
1018 * themselves given the specified format? Or is that too hard?
1019 * */
1020
1021 char *a = pointer_as_memory_reference(raw_elements);
1022 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
1023 free(a);
1024 }
1025
1026 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1027 pandecode_indent++;
1028
1029 for (int i = 0; i < count; ++i) {
1030 pandecode_log("{\n");
1031 pandecode_indent++;
1032
1033 unsigned mode = attr[i].elements & 7;
1034 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode(mode));
1035 pandecode_prop("shift = %d", attr[i].shift);
1036 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1037 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1038 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1039
1040 /* Decode further where possible */
1041
1042 if (mode == MALI_ATTR_MODULO) {
1043 pandecode_padded_vertices(
1044 attr[i].shift,
1045 attr[i].extra_flags);
1046 }
1047
1048 pandecode_indent--;
1049 pandecode_log("}, \n");
1050
1051 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1052 i++;
1053 pandecode_log("{\n");
1054 pandecode_indent++;
1055 pandecode_prop("unk = 0x%x", attr[i].unk);
1056 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1057 if (attr[i].zero != 0)
1058 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1059 pandecode_prop("divisor = %d", attr[i].divisor);
1060 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1061 pandecode_indent--;
1062 pandecode_log("}, \n");
1063 }
1064
1065 }
1066
1067 pandecode_indent--;
1068 pandecode_log("};\n");
1069 }
1070
1071 static mali_ptr
1072 pandecode_shader_address(const char *name, mali_ptr ptr)
1073 {
1074 /* TODO: Decode flags */
1075 mali_ptr shader_ptr = ptr & ~15;
1076
1077 char *a = pointer_as_memory_reference(shader_ptr);
1078 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1079 free(a);
1080
1081 return shader_ptr;
1082 }
1083
1084 static bool
1085 all_zero(unsigned *buffer, unsigned count)
1086 {
1087 for (unsigned i = 0; i < count; ++i) {
1088 if (buffer[i])
1089 return false;
1090 }
1091
1092 return true;
1093 }
1094
1095 static void
1096 pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
1097 {
1098 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1099 return;
1100
1101 const char *func = pandecode_func(stencil->func);
1102 const char *sfail = pandecode_stencil_op(stencil->sfail);
1103 const char *dpfail = pandecode_stencil_op(stencil->dpfail);
1104 const char *dppass = pandecode_stencil_op(stencil->dppass);
1105
1106 if (stencil->zero)
1107 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1108
1109 pandecode_log(".stencil_%s = {\n", name);
1110 pandecode_indent++;
1111 pandecode_prop("ref = %d", stencil->ref);
1112 pandecode_prop("mask = 0x%02X", stencil->mask);
1113 pandecode_prop("func = %s", func);
1114 pandecode_prop("sfail = %s", sfail);
1115 pandecode_prop("dpfail = %s", dpfail);
1116 pandecode_prop("dppass = %s", dppass);
1117 pandecode_indent--;
1118 pandecode_log("},\n");
1119 }
1120
1121 static void
1122 pandecode_blend_equation(const struct mali_blend_equation *blend)
1123 {
1124 if (blend->zero1)
1125 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1126
1127 pandecode_log(".equation = {\n");
1128 pandecode_indent++;
1129
1130 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1131 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1132
1133 pandecode_log(".color_mask = ");
1134 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1135 pandecode_log_cont(",\n");
1136
1137 pandecode_indent--;
1138 pandecode_log("},\n");
1139 }
1140
1141 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1142
1143 static unsigned
1144 decode_bifrost_constant(u16 constant)
1145 {
1146 float lo = (float) (constant & 0xFF);
1147 float hi = (float) (constant >> 8);
1148
1149 return (hi / 255.0) + (lo / 65535.0);
1150 }
1151
1152 static mali_ptr
1153 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1154 {
1155 struct bifrost_blend_rt *b =
1156 ((struct bifrost_blend_rt *) descs) + rt_no;
1157
1158 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1159 pandecode_indent++;
1160
1161 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1162 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1163 b->constant, decode_bifrost_constant(b->constant));
1164
1165 /* TODO figure out blend shader enable bit */
1166 pandecode_blend_equation(&b->equation);
1167 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1168 pandecode_prop("index = 0x%" PRIx16, b->index);
1169 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1170
1171 pandecode_indent--;
1172 pandecode_log("},\n");
1173
1174 return 0;
1175 }
1176
1177 static mali_ptr
1178 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1179 {
1180 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1181 return 0;
1182
1183 pandecode_log(".blend = {\n");
1184 pandecode_indent++;
1185
1186 if (is_shader) {
1187 pandecode_shader_address("shader", blend->shader);
1188 } else {
1189 pandecode_blend_equation(&blend->equation);
1190 pandecode_prop("constant = %f", blend->constant);
1191 }
1192
1193 pandecode_indent--;
1194 pandecode_log("},\n");
1195
1196 /* Return blend shader to disassemble if present */
1197 return is_shader ? (blend->shader & ~0xF) : 0;
1198 }
1199
1200 static mali_ptr
1201 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1202 {
1203 struct midgard_blend_rt *b =
1204 ((struct midgard_blend_rt *) descs) + rt_no;
1205
1206 /* Flags determine presence of blend shader */
1207 bool is_shader = (b->flags & 0xF) >= 0x2;
1208
1209 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1210 pandecode_indent++;
1211
1212 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1213
1214 union midgard_blend blend = b->blend;
1215 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1216
1217 pandecode_indent--;
1218 pandecode_log("};\n");
1219
1220 return shader;
1221 }
1222
1223 static int
1224 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1225 {
1226 char base[128];
1227 char *prefix = varying ? "varying" : "attribute";
1228 unsigned max_index = 0;
1229 snprintf(base, sizeof(base), "%s_meta", prefix);
1230
1231 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1232 pandecode_indent++;
1233
1234 struct mali_attr_meta *attr_meta;
1235 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1236
1237 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1238
1239 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1240 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1241 sizeof(*attr_mem));
1242
1243 pandecode_log("{\n");
1244 pandecode_indent++;
1245 pandecode_prop("index = %d", attr_meta->index);
1246
1247 if (attr_meta->index > max_index)
1248 max_index = attr_meta->index;
1249 pandecode_swizzle(attr_meta->swizzle);
1250 pandecode_prop("format = %s", pandecode_format(attr_meta->format));
1251
1252 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1253 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1254 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1255 pandecode_indent--;
1256 pandecode_log("},\n");
1257
1258 }
1259
1260 pandecode_indent--;
1261 pandecode_log("};\n");
1262
1263 return count ? (max_index + 1) : 0;
1264 }
1265
1266 static void
1267 pandecode_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1268 {
1269 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1270
1271 if (imem) {
1272 /* Indices are literally just a u32 array :) */
1273
1274 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1275
1276 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1277 pandecode_indent++;
1278
1279 for (unsigned i = 0; i < (index_count + 1); i += 3)
1280 pandecode_log("%d, %d, %d,\n",
1281 indices[i],
1282 indices[i + 1],
1283 indices[i + 2]);
1284
1285 pandecode_indent--;
1286 pandecode_log("};\n");
1287 }
1288 }
1289
1290 /* return bits [lo, hi) of word */
1291 static u32
1292 bits(u32 word, u32 lo, u32 hi)
1293 {
1294 if (hi - lo >= 32)
1295 return word; // avoid undefined behavior with the shift
1296
1297 return (word >> lo) & ((1 << (hi - lo)) - 1);
1298 }
1299
1300 static void
1301 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1302 {
1303 pandecode_log_cont("{\n");
1304 pandecode_indent++;
1305
1306 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1307 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1308 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1309 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1310 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1311 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1312 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1313
1314 /* Decode invocation_count. See the comment before the definition of
1315 * invocation_count for an explanation.
1316 */
1317 pandecode_msg("size: (%d, %d, %d)\n",
1318 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1319 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1320 bits(p->invocation_count, p->size_z_shift,
1321 p->workgroups_x_shift) + 1);
1322 pandecode_msg("workgroups: (%d, %d, %d)\n",
1323 bits(p->invocation_count, p->workgroups_x_shift,
1324 p->workgroups_y_shift) + 1,
1325 bits(p->invocation_count, p->workgroups_y_shift,
1326 p->workgroups_z_shift) + 1,
1327 bits(p->invocation_count, p->workgroups_z_shift,
1328 32) + 1);
1329
1330 /* TODO: Decode */
1331 if (p->unknown_draw)
1332 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1333
1334 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1335
1336 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
1337
1338 /* Index count only exists for tiler jobs anyway */
1339
1340 if (p->index_count)
1341 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1342
1343 if (p->offset_bias_correction)
1344 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1345
1346 DYN_MEMORY_PROP(p, job_no, indices);
1347
1348 if (p->zero1) {
1349 pandecode_msg("Zero tripped\n");
1350 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1351 }
1352
1353 pandecode_indent--;
1354 pandecode_log("},\n");
1355 }
1356
1357 static void
1358 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1359 {
1360 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1361
1362 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1363
1364 for (int i = 0; i < ubufs_count; i++) {
1365 mali_ptr ptr = ubufs[i].ptr << 2;
1366 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1367 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1368 char name[50];
1369 snprintf(name, sizeof(name), "ubuf_%d", i);
1370 /* The blob uses ubuf 0 to upload internal stuff and
1371 * uniforms that won't fit/are accessed indirectly, so
1372 * it puts it in the batchbuffer.
1373 */
1374 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1375 pandecode_indent++;
1376
1377 for (int j = 0; j <= ubufs[i].size; j++) {
1378 for (int k = 0; k < 4; k++) {
1379 if (k == 0)
1380 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1381 else
1382 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1383
1384 }
1385
1386 pandecode_log_cont("\n");
1387 }
1388
1389 pandecode_indent--;
1390 pandecode_log("};\n");
1391 }
1392
1393 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64"_%d[] = {\n",
1394 pubufs, job_no);
1395 pandecode_indent++;
1396
1397 for (int i = 0; i < ubufs_count; i++) {
1398 pandecode_log("{\n");
1399 pandecode_indent++;
1400 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1401 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1402 pandecode_indent--;
1403 pandecode_log("},\n");
1404 }
1405
1406 pandecode_indent--;
1407 pandecode_log("};\n");
1408 }
1409
1410 static void
1411 pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1412 {
1413
1414 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1415
1416 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1417
1418 if (scratchpad->zero)
1419 pandecode_msg("XXX scratchpad zero tripped");
1420
1421 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1422 pandecode_indent++;
1423
1424 pandecode_prop("flags = 0x%x", scratchpad->flags);
1425 MEMORY_PROP(scratchpad, gpu_scratchpad);
1426
1427 pandecode_indent--;
1428 pandecode_log("};\n");
1429 }
1430
1431 static unsigned shader_id = 0;
1432
1433 static void
1434 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1435 bool is_bifrost, unsigned nr_regs)
1436 {
1437 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1438 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1439
1440 /* Compute maximum possible size */
1441 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1442
1443 /* Print some boilerplate to clearly denote the assembly (which doesn't
1444 * obey indentation rules), and actually do the disassembly! */
1445
1446 printf("\n\n");
1447
1448 char prefix[512];
1449
1450 snprintf(prefix, sizeof(prefix) - 1, "shader%d - %s shader: ",
1451 shader_id++,
1452 (type == JOB_TYPE_TILER) ? "FRAGMENT" : "VERTEX");
1453
1454 if (is_bifrost) {
1455 disassemble_bifrost(code, sz, false);
1456 } else {
1457 disassemble_midgard(code, sz, true, nr_regs, prefix);
1458 }
1459
1460 printf("\n\n");
1461 }
1462
1463 static void
1464 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1465 int job_no, enum mali_job_type job_type,
1466 char *suffix, bool is_bifrost)
1467 {
1468 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1469 struct pandecode_mapped_memory *attr_mem;
1470
1471 unsigned rt_count = 1;
1472
1473 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1474 * are the only things actually needed from the FBD, vertex/tiler jobs
1475 * no longer reference the FBD -- instead, this field points to some
1476 * info about the scratchpad.
1477 */
1478 if (is_bifrost)
1479 pandecode_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1480 else if (p->framebuffer & MALI_MFBD)
1481 rt_count = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1482 else if (job_type == JOB_TYPE_COMPUTE)
1483 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1484 else
1485 pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1486
1487 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1488 int texture_count = 0, sampler_count = 0;
1489
1490 if (shader_meta_ptr) {
1491 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1492 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1493
1494 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1495 pandecode_indent++;
1496
1497 /* Save for dumps */
1498 attribute_count = s->attribute_count;
1499 varying_count = s->varying_count;
1500 texture_count = s->texture_count;
1501 sampler_count = s->sampler_count;
1502
1503 if (is_bifrost) {
1504 uniform_count = s->bifrost2.uniform_count;
1505 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1506 } else {
1507 uniform_count = s->midgard1.uniform_count;
1508 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1509 }
1510
1511 mali_ptr shader_ptr = pandecode_shader_address("shader", s->shader);
1512
1513 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1514 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1515 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1516 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1517
1518 unsigned nr_registers = 0;
1519
1520 if (is_bifrost) {
1521 pandecode_log(".bifrost1 = {\n");
1522 pandecode_indent++;
1523
1524 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1525 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1526
1527 pandecode_indent--;
1528 pandecode_log("},\n");
1529 } else {
1530 pandecode_log(".midgard1 = {\n");
1531 pandecode_indent++;
1532
1533 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1534 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1535 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1536 nr_registers = s->midgard1.work_count;
1537
1538 pandecode_log(".flags = ");
1539 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1540 pandecode_log_cont(",\n");
1541
1542 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1543
1544 pandecode_indent--;
1545 pandecode_log("},\n");
1546 }
1547
1548 if (s->depth_units || s->depth_factor) {
1549 pandecode_prop("depth_factor = %f", s->depth_factor);
1550 pandecode_prop("depth_units = %f", s->depth_units);
1551 }
1552
1553 if (s->alpha_coverage) {
1554 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1555 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1556
1557 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1558 invert_alpha_coverage ? "~" : "",
1559 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1560 }
1561
1562 if (s->unknown2_3 || s->unknown2_4) {
1563 pandecode_log(".unknown2_3 = ");
1564
1565 int unknown2_3 = s->unknown2_3;
1566 int unknown2_4 = s->unknown2_4;
1567
1568 /* We're not quite sure what these flags mean without the depth test, if anything */
1569
1570 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1571 const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
1572 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1573
1574 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1575 }
1576
1577 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1578 pandecode_log_cont(",\n");
1579
1580 pandecode_log(".unknown2_4 = ");
1581 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1582 pandecode_log_cont(",\n");
1583 }
1584
1585 if (s->stencil_mask_front || s->stencil_mask_back) {
1586 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1587 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1588 }
1589
1590 pandecode_stencil("front", &s->stencil_front);
1591 pandecode_stencil("back", &s->stencil_back);
1592
1593 if (is_bifrost) {
1594 pandecode_log(".bifrost2 = {\n");
1595 pandecode_indent++;
1596
1597 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1598 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1599 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1600 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1601
1602 pandecode_indent--;
1603 pandecode_log("},\n");
1604 } else if (s->midgard2.unknown2_7) {
1605 pandecode_log(".midgard2 = {\n");
1606 pandecode_indent++;
1607
1608 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1609 pandecode_indent--;
1610 pandecode_log("},\n");
1611 }
1612
1613 if (s->unknown2_8)
1614 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1615
1616 if (!is_bifrost) {
1617 /* TODO: Blend shaders routing/disasm */
1618
1619 union midgard_blend blend = s->blend;
1620 pandecode_midgard_blend(&blend, false);
1621 }
1622
1623 pandecode_indent--;
1624 pandecode_log("};\n");
1625
1626 /* MRT blend fields are used whenever MFBD is used, with
1627 * per-RT descriptors */
1628
1629 if (job_type == JOB_TYPE_TILER) {
1630 void* blend_base = (void *) (s + 1);
1631
1632 for (unsigned i = 0; i < rt_count; i++) {
1633 mali_ptr shader = 0;
1634
1635 if (is_bifrost)
1636 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1637 else
1638 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1639
1640 if (shader & ~0xF)
1641 pandecode_shader_disassemble(shader, job_no, job_type, false, 0);
1642 }
1643 }
1644
1645 if (shader_ptr & ~0xF)
1646 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost, nr_registers);
1647 } else
1648 pandecode_msg("<no shader>\n");
1649
1650 if (p->viewport) {
1651 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1652 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1653
1654 pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
1655 pandecode_indent++;
1656
1657 pandecode_prop("clip_minx = %f", f->clip_minx);
1658 pandecode_prop("clip_miny = %f", f->clip_miny);
1659 pandecode_prop("clip_minz = %f", f->clip_minz);
1660 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1661 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1662 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1663
1664 /* Only the higher coordinates are MALI_POSITIVE scaled */
1665
1666 pandecode_prop("viewport0 = { %d, %d }",
1667 f->viewport0[0], f->viewport0[1]);
1668
1669 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1670 f->viewport1[0] + 1, f->viewport1[1] + 1);
1671
1672 pandecode_indent--;
1673 pandecode_log("};\n");
1674 }
1675
1676 if (p->attribute_meta) {
1677 unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1678
1679 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1680 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1681 }
1682
1683 /* Varyings are encoded like attributes but not actually sent; we just
1684 * pass a zero buffer with the right stride/size set, (or whatever)
1685 * since the GPU will write to it itself */
1686
1687 if (p->varying_meta) {
1688 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1689 }
1690
1691 if (p->varyings) {
1692 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1693
1694 /* Number of descriptors depends on whether there are
1695 * non-internal varyings */
1696
1697 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
1698 }
1699
1700 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1701
1702 if (p->uniforms && !is_compute) {
1703 int rows = uniform_count, width = 4;
1704 size_t sz = rows * width * sizeof(float);
1705
1706 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1707 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1708 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1709
1710 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1711
1712 pandecode_indent++;
1713
1714 for (int row = 0; row < rows; row++) {
1715 for (int i = 0; i < width; i++) {
1716 u32 v = uniforms[i];
1717 float f;
1718 memcpy(&f, &v, sizeof(v));
1719 pandecode_log_cont("%X /* %f */, ", v, f);
1720 }
1721
1722 pandecode_log_cont("\n");
1723
1724 uniforms += width;
1725 }
1726
1727 pandecode_indent--;
1728 pandecode_log("};\n");
1729 } else if (p->uniforms) {
1730 int rows = uniform_count * 2;
1731 size_t sz = rows * sizeof(mali_ptr);
1732
1733 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1734 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1735 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1736
1737 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1738
1739 pandecode_indent++;
1740
1741 for (int row = 0; row < rows; row++) {
1742 char *a = pointer_as_memory_reference(uniforms[row]);
1743 pandecode_log("%s,\n", a);
1744 free(a);
1745 }
1746
1747 pandecode_indent--;
1748 pandecode_log("};\n");
1749
1750 }
1751
1752 if (p->uniform_buffers) {
1753 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1754 }
1755
1756 if (p->texture_trampoline) {
1757 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1758
1759 if (mmem) {
1760 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1761
1762 pandecode_log("uint64_t texture_trampoline_%"PRIx64"_%d[] = {\n", p->texture_trampoline, job_no);
1763 pandecode_indent++;
1764
1765 for (int tex = 0; tex < texture_count; ++tex) {
1766 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1767 char *a = pointer_as_memory_reference(*u);
1768 pandecode_log("%s,\n", a);
1769 free(a);
1770 }
1771
1772 pandecode_indent--;
1773 pandecode_log("};\n");
1774
1775 /* Now, finally, descend down into the texture descriptor */
1776 for (int tex = 0; tex < texture_count; ++tex) {
1777 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1778 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1779
1780 if (tmem) {
1781 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1782
1783 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1784 pandecode_indent++;
1785
1786 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1787 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1788 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1789 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1790 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1791 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1792 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1793
1794 struct mali_texture_format f = t->format;
1795
1796 pandecode_log(".format = {\n");
1797 pandecode_indent++;
1798
1799 pandecode_swizzle(f.swizzle);
1800 pandecode_prop("format = %s", pandecode_format(f.format));
1801 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1802 pandecode_prop("srgb = %" PRId32, f.srgb);
1803 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1804 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1805
1806 pandecode_indent--;
1807 pandecode_log("},\n");
1808
1809 pandecode_swizzle(t->swizzle);
1810
1811 if (t->swizzle_zero) {
1812 /* Shouldn't happen */
1813 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1814 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1815 }
1816
1817 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1818
1819 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1820 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1821 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1822
1823 pandecode_log(".payload = {\n");
1824 pandecode_indent++;
1825
1826 /* A bunch of bitmap pointers follow.
1827 * We work out the correct number,
1828 * based on the mipmap/cubemap
1829 * properties, but dump extra
1830 * possibilities to futureproof */
1831
1832 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1833 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1834
1835 /* Miptree for each face */
1836 if (f.type == MALI_TEX_CUBE)
1837 bitmap_count *= 6;
1838
1839 /* Array of textures */
1840 bitmap_count *= MALI_NEGATIVE(t->array_size);
1841
1842 /* Stride for each element */
1843 if (manual_stride)
1844 bitmap_count *= 2;
1845
1846 /* Sanity check the size */
1847 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1848 assert (bitmap_count <= max_count);
1849
1850 /* Dump more to be safe, but not _that_ much more */
1851 int safe_count = MIN2(bitmap_count * 2, max_count);
1852
1853 for (int i = 0; i < safe_count; ++i) {
1854 char *prefix = (i >= bitmap_count) ? "// " : "";
1855
1856 /* How we dump depends if this is a stride or a pointer */
1857
1858 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1859 /* signed 32-bit snuck in as a 64-bit pointer */
1860 uint64_t stride_set = t->payload[i];
1861 uint32_t clamped_stride = stride_set;
1862 int32_t stride = clamped_stride;
1863 assert(stride_set == clamped_stride);
1864 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1865 } else {
1866 char *a = pointer_as_memory_reference(t->payload[i]);
1867 pandecode_log("%s%s, \n", prefix, a);
1868 free(a);
1869 }
1870 }
1871
1872 pandecode_indent--;
1873 pandecode_log("},\n");
1874
1875 pandecode_indent--;
1876 pandecode_log("};\n");
1877 }
1878 }
1879 }
1880 }
1881
1882 if (p->sampler_descriptor) {
1883 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1884
1885 if (smem) {
1886 struct mali_sampler_descriptor *s;
1887
1888 mali_ptr d = p->sampler_descriptor;
1889
1890 for (int i = 0; i < sampler_count; ++i) {
1891 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1892
1893 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", d + sizeof(*s) * i, job_no, i);
1894 pandecode_indent++;
1895
1896 pandecode_log(".filter_mode = ");
1897 pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
1898 pandecode_log_cont(",\n");
1899
1900 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1901 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1902
1903 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
1904 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
1905 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
1906
1907 pandecode_prop("compare_func = %s", pandecode_alt_func(s->compare_func));
1908
1909 if (s->zero || s->zero2) {
1910 pandecode_msg("Zero tripped\n");
1911 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1912 }
1913
1914 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1915
1916 pandecode_prop("border_color = { %f, %f, %f, %f }",
1917 s->border_color[0],
1918 s->border_color[1],
1919 s->border_color[2],
1920 s->border_color[3]);
1921
1922 pandecode_indent--;
1923 pandecode_log("};\n");
1924 }
1925 }
1926 }
1927 }
1928
1929 static void
1930 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1931 {
1932 pandecode_log_cont("{\n");
1933 pandecode_indent++;
1934
1935 MEMORY_PROP(p, position_varying);
1936 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1937 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1938 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1939 DYN_MEMORY_PROP(p, job_no, uniforms);
1940 DYN_MEMORY_PROP(p, job_no, attributes);
1941 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1942 DYN_MEMORY_PROP(p, job_no, varyings);
1943 DYN_MEMORY_PROP(p, job_no, varying_meta);
1944 DYN_MEMORY_PROP(p, job_no, viewport);
1945 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1946
1947 if (is_bifrost)
1948 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1949 else
1950 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1951
1952 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1953 pandecode_prop("flags = %d", p->flags);
1954
1955 pandecode_indent--;
1956 pandecode_log("},\n");
1957 }
1958
1959 static void
1960 pandecode_vertex_only_bfr(struct bifrost_vertex_only *v)
1961 {
1962 pandecode_log_cont("{\n");
1963 pandecode_indent++;
1964
1965 pandecode_prop("unk2 = 0x%x", v->unk2);
1966
1967 if (v->zero0 || v->zero1) {
1968 pandecode_msg("vertex only zero tripped");
1969 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1970 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1971 }
1972
1973 pandecode_indent--;
1974 pandecode_log("}\n");
1975 }
1976
1977 static void
1978 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1979 {
1980
1981 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1982 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1983
1984 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1985 pandecode_indent++;
1986
1987 if (h->zero) {
1988 pandecode_msg("tiler heap zero tripped\n");
1989 pandecode_prop("zero = 0x%x", h->zero);
1990 }
1991
1992 for (int i = 0; i < 12; i++) {
1993 if (h->zeros[i] != 0) {
1994 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1995 i, h->zeros[i]);
1996 }
1997 }
1998
1999 pandecode_prop("heap_size = 0x%x", h->heap_size);
2000 MEMORY_PROP(h, tiler_heap_start);
2001 MEMORY_PROP(h, tiler_heap_free);
2002
2003 /* this might point to the beginning of another buffer, when it's
2004 * really the end of the tiler heap buffer, so we have to be careful
2005 * here. but for zero length, we need the same pointer.
2006 */
2007
2008 if (h->tiler_heap_end == h->tiler_heap_start) {
2009 MEMORY_PROP(h, tiler_heap_start);
2010 } else {
2011 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2012 pandecode_prop("tiler_heap_end = %s + 1", a);
2013 free(a);
2014 }
2015
2016 pandecode_indent--;
2017 pandecode_log("};\n");
2018 }
2019
2020 static void
2021 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2022 {
2023 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2024 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2025
2026 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2027
2028 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
2029 pandecode_indent++;
2030
2031 if (t->zero0 || t->zero1) {
2032 pandecode_msg("tiler meta zero tripped");
2033 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2034 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2035 }
2036
2037 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2038 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2039
2040 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2041 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2042 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
2043
2044 for (int i = 0; i < 12; i++) {
2045 if (t->zeros[i] != 0) {
2046 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2047 i, t->zeros[i]);
2048 }
2049 }
2050
2051 pandecode_indent--;
2052 pandecode_log("};\n");
2053 }
2054
2055 static void
2056 pandecode_gl_enables(uint32_t gl_enables, int job_type)
2057 {
2058 pandecode_log(".gl_enables = ");
2059
2060 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2061
2062 pandecode_log_cont(",\n");
2063 }
2064
2065 static void
2066 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2067 {
2068 if (u.pointer == 0x0)
2069 return;
2070
2071 pandecode_log(".primitive_size = {\n");
2072 pandecode_indent++;
2073
2074 if (constant) {
2075 pandecode_prop("constant = %f", u.constant);
2076 } else {
2077 MEMORY_PROP((&u), pointer);
2078 }
2079
2080 pandecode_indent--;
2081 pandecode_log("},\n");
2082 }
2083
2084 static void
2085 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2086 {
2087 pandecode_log_cont("{\n");
2088 pandecode_indent++;
2089
2090 /* TODO: gl_PointSize on Bifrost */
2091 pandecode_primitive_size(t->primitive_size, true);
2092
2093 DYN_MEMORY_PROP(t, job_no, tiler_meta);
2094 pandecode_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2095
2096 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2097 || t->zero6 || t->zero7 || t->zero8) {
2098 pandecode_msg("tiler only zero tripped");
2099 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2100 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2101 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2102 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2103 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2104 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2105 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2106 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2107 }
2108
2109 pandecode_indent--;
2110 pandecode_log("},\n");
2111 }
2112
2113 static int
2114 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2115 const struct pandecode_mapped_memory *mem,
2116 mali_ptr payload, int job_no)
2117 {
2118 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2119
2120 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2121
2122 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2123 pandecode_indent++;
2124
2125 pandecode_log(".prefix = ");
2126 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2127
2128 pandecode_log(".vertex = ");
2129 pandecode_vertex_only_bfr(&v->vertex);
2130
2131 pandecode_log(".postfix = ");
2132 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2133
2134 pandecode_indent--;
2135 pandecode_log("};\n");
2136
2137 return sizeof(*v);
2138 }
2139
2140 static int
2141 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2142 const struct pandecode_mapped_memory *mem,
2143 mali_ptr payload, int job_no)
2144 {
2145 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2146
2147 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2148
2149 pandecode_indices(t->prefix.indices, t->prefix.index_count, job_no);
2150 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2151
2152 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2153 pandecode_indent++;
2154
2155 pandecode_log(".prefix = ");
2156 pandecode_vertex_tiler_prefix(&t->prefix, job_no);
2157
2158 pandecode_log(".tiler = ");
2159 pandecode_tiler_only_bfr(&t->tiler, job_no);
2160
2161 pandecode_log(".postfix = ");
2162 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2163
2164 pandecode_indent--;
2165 pandecode_log("};\n");
2166
2167 return sizeof(*t);
2168 }
2169
2170 static int
2171 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2172 const struct pandecode_mapped_memory *mem,
2173 mali_ptr payload, int job_no)
2174 {
2175 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2176
2177 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2178
2179 pandecode_indices(v->prefix.indices, v->prefix.index_count, job_no);
2180
2181 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2182 pandecode_indent++;
2183
2184 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2185 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2186
2187 pandecode_log(".prefix = ");
2188 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2189
2190 pandecode_gl_enables(v->gl_enables, h->job_type);
2191
2192 if (v->instance_shift || v->instance_odd) {
2193 pandecode_prop("instance_shift = 0x%d /* %d */",
2194 v->instance_shift, 1 << v->instance_shift);
2195 pandecode_prop("instance_odd = 0x%X /* %d */",
2196 v->instance_odd, (2 * v->instance_odd) + 1);
2197
2198 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2199 }
2200
2201 if (v->offset_start)
2202 pandecode_prop("offset_start = %d", v->offset_start);
2203
2204 if (v->zero5) {
2205 pandecode_msg("Zero tripped\n");
2206 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2207 }
2208
2209 pandecode_log(".postfix = ");
2210 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2211
2212 pandecode_indent--;
2213 pandecode_log("};\n");
2214
2215 return sizeof(*v);
2216 }
2217
2218 static int
2219 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2220 mali_ptr payload, int job_no,
2221 bool is_bifrost)
2222 {
2223 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2224
2225 bool fbd_dumped = false;
2226
2227 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2228 /* Only SFBDs are understood, not MFBDs. We're speculating,
2229 * based on the versioning, kernel code, etc, that the
2230 * difference is between Single FrameBuffer Descriptor and
2231 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2232 * up with multi-framebuffer support being added (T7xx onwards,
2233 * including Gxx). In any event, there's some field shuffling
2234 * that we haven't looked into yet. */
2235
2236 pandecode_sfbd(s->framebuffer & FBD_MASK, job_no);
2237 fbd_dumped = true;
2238 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2239 /* We don't know if Bifrost supports SFBD's at all, since the
2240 * driver never uses them. And the format is different from
2241 * Midgard anyways, due to the tiler heap and scratchpad being
2242 * moved out into separate structures, so it's not clear what a
2243 * Bifrost SFBD would even look like without getting an actual
2244 * trace, which appears impossible.
2245 */
2246
2247 pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2248 fbd_dumped = true;
2249 }
2250
2251 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2252 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2253 pandecode_indent++;
2254
2255 /* See the comments by the macro definitions for mathematical context
2256 * on why this is so weird */
2257
2258 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2259 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2260
2261 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2262 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2263 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2264
2265 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2266 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2267 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2268
2269 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2270 * we have to fallback on offsets. */
2271
2272 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2273
2274 if (fbd_dumped)
2275 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2276 else
2277 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2278
2279 pandecode_indent--;
2280 pandecode_log("};\n");
2281
2282 return sizeof(*s);
2283 }
2284
2285 static int job_descriptor_number = 0;
2286
2287 int
2288 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
2289 {
2290 struct mali_job_descriptor_header *h;
2291
2292 int start_number = 0;
2293
2294 bool first = true;
2295 bool last_size;
2296
2297 do {
2298 struct pandecode_mapped_memory *mem =
2299 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2300
2301 void *payload;
2302
2303 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2304
2305 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2306 * high 32-bits of the 64-bit pointer are reused to store
2307 * something else.
2308 */
2309 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2310 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2311 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2312
2313 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2314 MALI_PAYLOAD_SIZE);
2315
2316 int job_no = job_descriptor_number++;
2317
2318 if (first)
2319 start_number = job_no;
2320
2321 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2322 pandecode_indent++;
2323
2324 pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
2325
2326 /* Save for next job fixing */
2327 last_size = h->job_descriptor_size;
2328
2329 if (h->job_descriptor_size)
2330 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2331
2332 if (h->exception_status != 0x1)
2333 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2334 h->exception_status,
2335 (h->exception_status >> 16) & 0xFFFF,
2336 (h->exception_status >> 8) & 0x3,
2337 h->exception_status & 0xFF);
2338
2339 if (h->first_incomplete_task)
2340 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2341
2342 if (h->fault_pointer)
2343 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2344
2345 if (h->job_barrier)
2346 pandecode_prop("job_barrier = %d", h->job_barrier);
2347
2348 pandecode_prop("job_index = %d", h->job_index);
2349
2350 if (h->unknown_flags)
2351 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2352
2353 if (h->job_dependency_index_1)
2354 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2355
2356 if (h->job_dependency_index_2)
2357 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2358
2359 pandecode_indent--;
2360 pandecode_log("};\n");
2361
2362 /* Do not touch the field yet -- decode the payload first, and
2363 * don't touch that either. This is essential for the uploads
2364 * to occur in sequence and therefore be dynamically allocated
2365 * correctly. Do note the size, however, for that related
2366 * reason. */
2367
2368 switch (h->job_type) {
2369 case JOB_TYPE_SET_VALUE: {
2370 struct mali_payload_set_value *s = payload;
2371 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2372 pandecode_indent++;
2373 MEMORY_PROP(s, out);
2374 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2375 pandecode_indent--;
2376 pandecode_log("};\n");
2377
2378 break;
2379 }
2380
2381 case JOB_TYPE_TILER:
2382 case JOB_TYPE_VERTEX:
2383 case JOB_TYPE_COMPUTE:
2384 if (bifrost) {
2385 if (h->job_type == JOB_TYPE_TILER)
2386 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
2387 else
2388 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
2389 } else
2390 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2391
2392 break;
2393
2394 case JOB_TYPE_FRAGMENT:
2395 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost);
2396 break;
2397
2398 default:
2399 break;
2400 }
2401
2402 /* Handle linkage */
2403
2404 if (!first) {
2405 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2406
2407 if (last_size)
2408 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2409 else
2410 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2411 }
2412
2413 first = false;
2414
2415 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2416
2417 return start_number;
2418 }