pan/decode: Fix duplicate heap_end property
[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_end);
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("no_preload = 0x%" PRIx32, format.no_preload);
656
657 if (format.zero)
658 pandecode_prop("zero = 0x%" PRIx32, format.zero);
659
660 pandecode_indent--;
661 pandecode_log("},\n");
662 }
663
664 static void
665 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
666 {
667 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
668 pandecode_indent++;
669
670 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
671 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
672 struct pandecode_mapped_memory *mem =
673 pandecode_find_mapped_gpu_mem_containing(rt_va);
674 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
675
676 pandecode_log("{\n");
677 pandecode_indent++;
678
679 pandecode_rt_format(rt->format);
680
681 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
682 pandecode_log(".afbc = {\n");
683 pandecode_indent++;
684
685 char *a = pointer_as_memory_reference(rt->afbc.metadata);
686 pandecode_prop("metadata = %s", a);
687 free(a);
688
689 pandecode_prop("stride = %d", rt->afbc.stride);
690 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
691
692 pandecode_indent--;
693 pandecode_log("},\n");
694 } else {
695 pandecode_log(".chunknown = {\n");
696 pandecode_indent++;
697
698 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
699
700 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
701 pandecode_prop("pointer = %s", a);
702 free(a);
703
704 pandecode_indent--;
705 pandecode_log("},\n");
706 }
707
708 MEMORY_PROP(rt, framebuffer);
709 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
710
711 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
712 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
713 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
714 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
715 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
716 }
717
718 if (rt->zero1 || rt->zero2 || rt->zero3) {
719 pandecode_msg("render target zeros tripped\n");
720 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
721 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
722 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
723 }
724
725 pandecode_indent--;
726 pandecode_log("},\n");
727 }
728
729 pandecode_indent--;
730 pandecode_log("};\n");
731 }
732
733 static unsigned
734 pandecode_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
735 {
736 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
737 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
738
739 if (fb->sample_locations) {
740 /* The blob stores all possible sample locations in a single buffer
741 * allocated on startup, and just switches the pointer when switching
742 * MSAA state. For now, we just put the data into the cmdstream, but we
743 * should do something like what the blob does with a real driver.
744 *
745 * There seem to be 32 slots for sample locations, followed by another
746 * 16. The second 16 is just the center location followed by 15 zeros
747 * in all the cases I've identified (maybe shader vs. depth/color
748 * samples?).
749 */
750
751 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
752
753 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
754
755 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
756 pandecode_indent++;
757
758 for (int i = 0; i < 32 + 16; i++) {
759 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
760 }
761
762 pandecode_indent--;
763 pandecode_log("};\n");
764 }
765
766 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
767 pandecode_indent++;
768
769 pandecode_prop("unk0 = 0x%x", fb->unk0);
770
771 if (fb->sample_locations)
772 pandecode_prop("sample_locations = sample_locations_%d", job_no);
773
774 /* Assume that unknown1 was emitted in the last job for
775 * now */
776 MEMORY_PROP(fb, unknown1);
777
778 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
779 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
780 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
781 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
782
783 pandecode_prop("unk1 = 0x%x", fb->unk1);
784 pandecode_prop("unk2 = 0x%x", fb->unk2);
785 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
786 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
787
788 pandecode_log(".mfbd_flags = ");
789 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
790 pandecode_log_cont(",\n");
791
792 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
793 pandecode_prop("clear_depth = %f", fb->clear_depth);
794
795 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
796 MEMORY_PROP(fb, scratchpad);
797 const struct midgard_tiler_descriptor t = fb->tiler;
798 pandecode_midgard_tiler_descriptor(&t);
799
800 if (fb->zero3 || fb->zero4) {
801 pandecode_msg("framebuffer zeros tripped\n");
802 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
803 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
804 }
805
806 pandecode_indent--;
807 pandecode_log("};\n");
808
809 gpu_va += sizeof(struct bifrost_framebuffer);
810
811 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
812 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
813 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
814
815 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
816 pandecode_indent++;
817
818 MEMORY_PROP(fbx, checksum);
819
820 if (fbx->checksum_stride)
821 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
822
823 pandecode_log(".flags = ");
824 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
825 pandecode_log_cont(",\n");
826
827 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
828 pandecode_log(".ds_afbc = {\n");
829 pandecode_indent++;
830
831 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil_afbc_metadata);
832 pandecode_prop("depth_stencil_afbc_stride = %d",
833 fbx->ds_afbc.depth_stencil_afbc_stride);
834 MEMORY_PROP_DIR(fbx->ds_afbc, depth_stencil);
835
836 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
837 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
838 pandecode_prop("zero1 = 0x%" PRIx32,
839 fbx->ds_afbc.zero1);
840 pandecode_prop("padding = 0x%" PRIx64,
841 fbx->ds_afbc.padding);
842 }
843
844 pandecode_indent--;
845 pandecode_log("},\n");
846 } else {
847 pandecode_log(".ds_linear = {\n");
848 pandecode_indent++;
849
850 if (fbx->ds_linear.depth) {
851 MEMORY_PROP_DIR(fbx->ds_linear, depth);
852 pandecode_prop("depth_stride = %d",
853 fbx->ds_linear.depth_stride);
854 }
855
856 if (fbx->ds_linear.stencil) {
857 MEMORY_PROP_DIR(fbx->ds_linear, stencil);
858 pandecode_prop("stencil_stride = %d",
859 fbx->ds_linear.stencil_stride);
860 }
861
862 if (fbx->ds_linear.depth_stride_zero ||
863 fbx->ds_linear.stencil_stride_zero ||
864 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
865 pandecode_msg("Depth/stencil zeros tripped\n");
866 pandecode_prop("depth_stride_zero = 0x%x",
867 fbx->ds_linear.depth_stride_zero);
868 pandecode_prop("stencil_stride_zero = 0x%x",
869 fbx->ds_linear.stencil_stride_zero);
870 pandecode_prop("zero1 = 0x%" PRIx32,
871 fbx->ds_linear.zero1);
872 pandecode_prop("zero2 = 0x%" PRIx32,
873 fbx->ds_linear.zero2);
874 }
875
876 pandecode_indent--;
877 pandecode_log("},\n");
878 }
879
880 if (fbx->zero3 || fbx->zero4) {
881 pandecode_msg("fb_extra zeros tripped\n");
882 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
883 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
884 }
885
886 pandecode_indent--;
887 pandecode_log("};\n");
888
889 gpu_va += sizeof(struct bifrost_fb_extra);
890 }
891
892 if (with_render_targets)
893 pandecode_render_target(gpu_va, job_no, fb);
894
895 /* Passback the render target count */
896 return MALI_NEGATIVE(fb->rt_count_1);
897 }
898
899 /* Just add a comment decoding the shift/odd fields forming the padded vertices
900 * count */
901
902 static void
903 pandecode_padded_vertices(unsigned shift, unsigned k)
904 {
905 unsigned odd = 2*k + 1;
906 unsigned pot = 1 << shift;
907 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
908 }
909
910 /* Given a magic divisor, recover what we were trying to divide by.
911 *
912 * Let m represent the magic divisor. By definition, m is an element on Z, whre
913 * 0 <= m < 2^N, for N bits in m.
914 *
915 * Let q represent the number we would like to divide by.
916 *
917 * By definition of a magic divisor for N-bit unsigned integers (a number you
918 * multiply by to magically get division), m is a number such that:
919 *
920 * (m * x) & (2^N - 1) = floor(x/q).
921 * for all x on Z where 0 <= x < 2^N
922 *
923 * Ignore the case where any of the above values equals zero; it is irrelevant
924 * for our purposes (instanced arrays).
925 *
926 * Choose x = q. Then:
927 *
928 * (m * x) & (2^N - 1) = floor(x/q).
929 * (m * q) & (2^N - 1) = floor(q/q).
930 *
931 * floor(q/q) = floor(1) = 1, therefore:
932 *
933 * (m * q) & (2^N - 1) = 1
934 *
935 * Recall the identity that the bitwise AND of one less than a power-of-two
936 * equals the modulo with that power of two, i.e. for all x:
937 *
938 * x & (2^N - 1) = x % N
939 *
940 * Therefore:
941 *
942 * mq % (2^N) = 1
943 *
944 * By definition, a modular multiplicative inverse of a number m is the number
945 * q such that with respect to a modulos M:
946 *
947 * mq % M = 1
948 *
949 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
950 *
951 */
952
953 static void
954 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
955 {
956 #if 0
957 /* Compute the modular inverse of `magic` with respect to 2^(32 -
958 * shift) the most lame way possible... just repeatedly add.
959 * Asymptoptically slow but nobody cares in practice, unless you have
960 * massive numbers of vertices or high divisors. */
961
962 unsigned inverse = 0;
963
964 /* Magic implicitly has the highest bit set */
965 magic |= (1 << 31);
966
967 /* Depending on rounding direction */
968 if (extra)
969 magic++;
970
971 for (;;) {
972 uint32_t product = magic * inverse;
973
974 if (shift) {
975 product >>= shift;
976 }
977
978 if (product == 1)
979 break;
980
981 ++inverse;
982 }
983
984 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
985
986 /* Recall we're supposed to divide by (gl_level_divisor *
987 * padded_num_vertices) */
988
989 unsigned padded_num_vertices = inverse / orig_divisor;
990
991 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
992 #endif
993 }
994
995 static void
996 pandecode_attributes(const struct pandecode_mapped_memory *mem,
997 mali_ptr addr, int job_no, char *suffix,
998 int count, bool varying)
999 {
1000 char *prefix = varying ? "varyings" : "attributes";
1001
1002 if (!addr) {
1003 pandecode_msg("no %s\n", prefix);
1004 return;
1005 }
1006
1007 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
1008
1009 char base[128];
1010 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
1011
1012 for (int i = 0; i < count; ++i) {
1013 enum mali_attr_mode mode = attr[i].elements & 7;
1014
1015 if (mode == MALI_ATTR_UNUSED)
1016 continue;
1017
1018 mali_ptr raw_elements = attr[i].elements & ~7;
1019
1020 /* TODO: Do we maybe want to dump the attribute values
1021 * themselves given the specified format? Or is that too hard?
1022 * */
1023
1024 char *a = pointer_as_memory_reference(raw_elements);
1025 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
1026 free(a);
1027 }
1028
1029 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1030 pandecode_indent++;
1031
1032 for (int i = 0; i < count; ++i) {
1033 pandecode_log("{\n");
1034 pandecode_indent++;
1035
1036 unsigned mode = attr[i].elements & 7;
1037 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode(mode));
1038 pandecode_prop("shift = %d", attr[i].shift);
1039 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1040 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1041 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1042
1043 /* Decode further where possible */
1044
1045 if (mode == MALI_ATTR_MODULO) {
1046 pandecode_padded_vertices(
1047 attr[i].shift,
1048 attr[i].extra_flags);
1049 }
1050
1051 pandecode_indent--;
1052 pandecode_log("}, \n");
1053
1054 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1055 i++;
1056 pandecode_log("{\n");
1057 pandecode_indent++;
1058 pandecode_prop("unk = 0x%x", attr[i].unk);
1059 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1060 if (attr[i].zero != 0)
1061 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1062 pandecode_prop("divisor = %d", attr[i].divisor);
1063 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1064 pandecode_indent--;
1065 pandecode_log("}, \n");
1066 }
1067
1068 }
1069
1070 pandecode_indent--;
1071 pandecode_log("};\n");
1072 }
1073
1074 static mali_ptr
1075 pandecode_shader_address(const char *name, mali_ptr ptr)
1076 {
1077 /* TODO: Decode flags */
1078 mali_ptr shader_ptr = ptr & ~15;
1079
1080 char *a = pointer_as_memory_reference(shader_ptr);
1081 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1082 free(a);
1083
1084 return shader_ptr;
1085 }
1086
1087 static bool
1088 all_zero(unsigned *buffer, unsigned count)
1089 {
1090 for (unsigned i = 0; i < count; ++i) {
1091 if (buffer[i])
1092 return false;
1093 }
1094
1095 return true;
1096 }
1097
1098 static void
1099 pandecode_stencil(const char *name, const struct mali_stencil_test *stencil)
1100 {
1101 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1102 return;
1103
1104 const char *func = pandecode_func(stencil->func);
1105 const char *sfail = pandecode_stencil_op(stencil->sfail);
1106 const char *dpfail = pandecode_stencil_op(stencil->dpfail);
1107 const char *dppass = pandecode_stencil_op(stencil->dppass);
1108
1109 if (stencil->zero)
1110 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1111
1112 pandecode_log(".stencil_%s = {\n", name);
1113 pandecode_indent++;
1114 pandecode_prop("ref = %d", stencil->ref);
1115 pandecode_prop("mask = 0x%02X", stencil->mask);
1116 pandecode_prop("func = %s", func);
1117 pandecode_prop("sfail = %s", sfail);
1118 pandecode_prop("dpfail = %s", dpfail);
1119 pandecode_prop("dppass = %s", dppass);
1120 pandecode_indent--;
1121 pandecode_log("},\n");
1122 }
1123
1124 static void
1125 pandecode_blend_equation(const struct mali_blend_equation *blend)
1126 {
1127 if (blend->zero1)
1128 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1129
1130 pandecode_log(".equation = {\n");
1131 pandecode_indent++;
1132
1133 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1134 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1135
1136 pandecode_log(".color_mask = ");
1137 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1138 pandecode_log_cont(",\n");
1139
1140 pandecode_indent--;
1141 pandecode_log("},\n");
1142 }
1143
1144 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1145
1146 static unsigned
1147 decode_bifrost_constant(u16 constant)
1148 {
1149 float lo = (float) (constant & 0xFF);
1150 float hi = (float) (constant >> 8);
1151
1152 return (hi / 255.0) + (lo / 65535.0);
1153 }
1154
1155 static mali_ptr
1156 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1157 {
1158 struct bifrost_blend_rt *b =
1159 ((struct bifrost_blend_rt *) descs) + rt_no;
1160
1161 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1162 pandecode_indent++;
1163
1164 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1165 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1166 b->constant, decode_bifrost_constant(b->constant));
1167
1168 /* TODO figure out blend shader enable bit */
1169 pandecode_blend_equation(&b->equation);
1170 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1171 pandecode_prop("index = 0x%" PRIx16, b->index);
1172 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1173
1174 pandecode_indent--;
1175 pandecode_log("},\n");
1176
1177 return 0;
1178 }
1179
1180 static mali_ptr
1181 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1182 {
1183 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1184 return 0;
1185
1186 pandecode_log(".blend = {\n");
1187 pandecode_indent++;
1188
1189 if (is_shader) {
1190 pandecode_shader_address("shader", blend->shader);
1191 } else {
1192 pandecode_blend_equation(&blend->equation);
1193 pandecode_prop("constant = %f", blend->constant);
1194 }
1195
1196 pandecode_indent--;
1197 pandecode_log("},\n");
1198
1199 /* Return blend shader to disassemble if present */
1200 return is_shader ? (blend->shader & ~0xF) : 0;
1201 }
1202
1203 static mali_ptr
1204 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1205 {
1206 struct midgard_blend_rt *b =
1207 ((struct midgard_blend_rt *) descs) + rt_no;
1208
1209 /* Flags determine presence of blend shader */
1210 bool is_shader = (b->flags & 0xF) >= 0x2;
1211
1212 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1213 pandecode_indent++;
1214
1215 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1216
1217 union midgard_blend blend = b->blend;
1218 mali_ptr shader = pandecode_midgard_blend(&blend, is_shader);
1219
1220 pandecode_indent--;
1221 pandecode_log("};\n");
1222
1223 return shader;
1224 }
1225
1226 static int
1227 pandecode_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1228 {
1229 char base[128];
1230 char *prefix = varying ? "varying" : "attribute";
1231 unsigned max_index = 0;
1232 snprintf(base, sizeof(base), "%s_meta", prefix);
1233
1234 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1235 pandecode_indent++;
1236
1237 struct mali_attr_meta *attr_meta;
1238 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1239
1240 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1241
1242 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1243 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1244 sizeof(*attr_mem));
1245
1246 pandecode_log("{\n");
1247 pandecode_indent++;
1248 pandecode_prop("index = %d", attr_meta->index);
1249
1250 if (attr_meta->index > max_index)
1251 max_index = attr_meta->index;
1252 pandecode_swizzle(attr_meta->swizzle);
1253 pandecode_prop("format = %s", pandecode_format(attr_meta->format));
1254
1255 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1256 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1257 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1258 pandecode_indent--;
1259 pandecode_log("},\n");
1260
1261 }
1262
1263 pandecode_indent--;
1264 pandecode_log("};\n");
1265
1266 return count ? (max_index + 1) : 0;
1267 }
1268
1269 static void
1270 pandecode_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1271 {
1272 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1273
1274 if (imem) {
1275 /* Indices are literally just a u32 array :) */
1276
1277 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1278
1279 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1280 pandecode_indent++;
1281
1282 for (unsigned i = 0; i < (index_count + 1); i += 3)
1283 pandecode_log("%d, %d, %d,\n",
1284 indices[i],
1285 indices[i + 1],
1286 indices[i + 2]);
1287
1288 pandecode_indent--;
1289 pandecode_log("};\n");
1290 }
1291 }
1292
1293 /* return bits [lo, hi) of word */
1294 static u32
1295 bits(u32 word, u32 lo, u32 hi)
1296 {
1297 if (hi - lo >= 32)
1298 return word; // avoid undefined behavior with the shift
1299
1300 return (word >> lo) & ((1 << (hi - lo)) - 1);
1301 }
1302
1303 static void
1304 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1305 {
1306 pandecode_log_cont("{\n");
1307 pandecode_indent++;
1308
1309 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1310 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1311 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1312 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1313 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1314 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1315 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1316
1317 /* Decode invocation_count. See the comment before the definition of
1318 * invocation_count for an explanation.
1319 */
1320 pandecode_msg("size: (%d, %d, %d)\n",
1321 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1322 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1323 bits(p->invocation_count, p->size_z_shift,
1324 p->workgroups_x_shift) + 1);
1325 pandecode_msg("workgroups: (%d, %d, %d)\n",
1326 bits(p->invocation_count, p->workgroups_x_shift,
1327 p->workgroups_y_shift) + 1,
1328 bits(p->invocation_count, p->workgroups_y_shift,
1329 p->workgroups_z_shift) + 1,
1330 bits(p->invocation_count, p->workgroups_z_shift,
1331 32) + 1);
1332
1333 /* TODO: Decode */
1334 if (p->unknown_draw)
1335 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1336
1337 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1338
1339 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p->draw_mode));
1340
1341 /* Index count only exists for tiler jobs anyway */
1342
1343 if (p->index_count)
1344 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1345
1346 if (p->offset_bias_correction)
1347 pandecode_prop("offset_bias_correction = %d", p->offset_bias_correction);
1348
1349 DYN_MEMORY_PROP(p, job_no, indices);
1350
1351 if (p->zero1) {
1352 pandecode_msg("Zero tripped\n");
1353 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1354 }
1355
1356 pandecode_indent--;
1357 pandecode_log("},\n");
1358 }
1359
1360 static void
1361 pandecode_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1362 {
1363 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1364
1365 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1366
1367 for (int i = 0; i < ubufs_count; i++) {
1368 mali_ptr ptr = ubufs[i].ptr << 2;
1369 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1370 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1371 char name[50];
1372 snprintf(name, sizeof(name), "ubuf_%d", i);
1373 /* The blob uses ubuf 0 to upload internal stuff and
1374 * uniforms that won't fit/are accessed indirectly, so
1375 * it puts it in the batchbuffer.
1376 */
1377 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1378 pandecode_indent++;
1379
1380 for (int j = 0; j <= ubufs[i].size; j++) {
1381 for (int k = 0; k < 4; k++) {
1382 if (k == 0)
1383 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1384 else
1385 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1386
1387 }
1388
1389 pandecode_log_cont("\n");
1390 }
1391
1392 pandecode_indent--;
1393 pandecode_log("};\n");
1394 }
1395
1396 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64"_%d[] = {\n",
1397 pubufs, job_no);
1398 pandecode_indent++;
1399
1400 for (int i = 0; i < ubufs_count; i++) {
1401 pandecode_log("{\n");
1402 pandecode_indent++;
1403 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1404 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1405 pandecode_indent--;
1406 pandecode_log("},\n");
1407 }
1408
1409 pandecode_indent--;
1410 pandecode_log("};\n");
1411 }
1412
1413 static void
1414 pandecode_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1415 {
1416
1417 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1418
1419 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1420
1421 if (scratchpad->zero)
1422 pandecode_msg("XXX scratchpad zero tripped");
1423
1424 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1425 pandecode_indent++;
1426
1427 pandecode_prop("flags = 0x%x", scratchpad->flags);
1428 MEMORY_PROP(scratchpad, gpu_scratchpad);
1429
1430 pandecode_indent--;
1431 pandecode_log("};\n");
1432 }
1433
1434 static unsigned shader_id = 0;
1435
1436 static void
1437 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1438 bool is_bifrost, unsigned nr_regs)
1439 {
1440 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1441 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1442
1443 /* Compute maximum possible size */
1444 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1445
1446 /* Print some boilerplate to clearly denote the assembly (which doesn't
1447 * obey indentation rules), and actually do the disassembly! */
1448
1449 printf("\n\n");
1450
1451 char prefix[512];
1452
1453 snprintf(prefix, sizeof(prefix) - 1, "shader%d - %s shader: ",
1454 shader_id++,
1455 (type == JOB_TYPE_TILER) ? "FRAGMENT" : "VERTEX");
1456
1457 if (is_bifrost) {
1458 disassemble_bifrost(code, sz, false);
1459 } else {
1460 disassemble_midgard(code, sz, true, nr_regs, prefix);
1461 }
1462
1463 printf("\n\n");
1464 }
1465
1466 static void
1467 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1468 int job_no, enum mali_job_type job_type,
1469 char *suffix, bool is_bifrost)
1470 {
1471 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1472 struct pandecode_mapped_memory *attr_mem;
1473
1474 unsigned rt_count = 1;
1475
1476 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1477 * are the only things actually needed from the FBD, vertex/tiler jobs
1478 * no longer reference the FBD -- instead, this field points to some
1479 * info about the scratchpad.
1480 */
1481 if (is_bifrost)
1482 pandecode_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1483 else if (p->framebuffer & MALI_MFBD)
1484 rt_count = pandecode_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1485 else if (job_type == JOB_TYPE_COMPUTE)
1486 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1487 else
1488 pandecode_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1489
1490 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1491 int texture_count = 0, sampler_count = 0;
1492
1493 if (shader_meta_ptr) {
1494 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1495 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1496
1497 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1498 pandecode_indent++;
1499
1500 /* Save for dumps */
1501 attribute_count = s->attribute_count;
1502 varying_count = s->varying_count;
1503 texture_count = s->texture_count;
1504 sampler_count = s->sampler_count;
1505
1506 if (is_bifrost) {
1507 uniform_count = s->bifrost2.uniform_count;
1508 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1509 } else {
1510 uniform_count = s->midgard1.uniform_count;
1511 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1512 }
1513
1514 mali_ptr shader_ptr = pandecode_shader_address("shader", s->shader);
1515
1516 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1517 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1518 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1519 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1520
1521 unsigned nr_registers = 0;
1522
1523 if (is_bifrost) {
1524 pandecode_log(".bifrost1 = {\n");
1525 pandecode_indent++;
1526
1527 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1528 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1529
1530 pandecode_indent--;
1531 pandecode_log("},\n");
1532 } else {
1533 pandecode_log(".midgard1 = {\n");
1534 pandecode_indent++;
1535
1536 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1537 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1538 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1539 nr_registers = s->midgard1.work_count;
1540
1541 pandecode_log(".flags = ");
1542 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1543 pandecode_log_cont(",\n");
1544
1545 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1546
1547 pandecode_indent--;
1548 pandecode_log("},\n");
1549 }
1550
1551 if (s->depth_units || s->depth_factor) {
1552 pandecode_prop("depth_factor = %f", s->depth_factor);
1553 pandecode_prop("depth_units = %f", s->depth_units);
1554 }
1555
1556 if (s->alpha_coverage) {
1557 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1558 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1559
1560 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1561 invert_alpha_coverage ? "~" : "",
1562 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1563 }
1564
1565 if (s->unknown2_3 || s->unknown2_4) {
1566 pandecode_log(".unknown2_3 = ");
1567
1568 int unknown2_3 = s->unknown2_3;
1569 int unknown2_4 = s->unknown2_4;
1570
1571 /* We're not quite sure what these flags mean without the depth test, if anything */
1572
1573 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1574 const char *func = pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3));
1575 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1576
1577 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1578 }
1579
1580 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1581 pandecode_log_cont(",\n");
1582
1583 pandecode_log(".unknown2_4 = ");
1584 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1585 pandecode_log_cont(",\n");
1586 }
1587
1588 if (s->stencil_mask_front || s->stencil_mask_back) {
1589 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1590 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1591 }
1592
1593 pandecode_stencil("front", &s->stencil_front);
1594 pandecode_stencil("back", &s->stencil_back);
1595
1596 if (is_bifrost) {
1597 pandecode_log(".bifrost2 = {\n");
1598 pandecode_indent++;
1599
1600 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1601 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1602 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1603 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1604
1605 pandecode_indent--;
1606 pandecode_log("},\n");
1607 } else if (s->midgard2.unknown2_7) {
1608 pandecode_log(".midgard2 = {\n");
1609 pandecode_indent++;
1610
1611 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1612 pandecode_indent--;
1613 pandecode_log("},\n");
1614 }
1615
1616 if (s->unknown2_8)
1617 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1618
1619 if (!is_bifrost) {
1620 /* TODO: Blend shaders routing/disasm */
1621
1622 union midgard_blend blend = s->blend;
1623 pandecode_midgard_blend(&blend, false);
1624 }
1625
1626 pandecode_indent--;
1627 pandecode_log("};\n");
1628
1629 /* MRT blend fields are used whenever MFBD is used, with
1630 * per-RT descriptors */
1631
1632 if (job_type == JOB_TYPE_TILER) {
1633 void* blend_base = (void *) (s + 1);
1634
1635 for (unsigned i = 0; i < rt_count; i++) {
1636 mali_ptr shader = 0;
1637
1638 if (is_bifrost)
1639 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1640 else
1641 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1642
1643 if (shader & ~0xF)
1644 pandecode_shader_disassemble(shader, job_no, job_type, false, 0);
1645 }
1646 }
1647
1648 if (shader_ptr & ~0xF)
1649 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost, nr_registers);
1650 } else
1651 pandecode_msg("<no shader>\n");
1652
1653 if (p->viewport) {
1654 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1655 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1656
1657 pandecode_log("struct mali_viewport viewport_%"PRIx64"_%d%s = {\n", p->viewport, job_no, suffix);
1658 pandecode_indent++;
1659
1660 pandecode_prop("clip_minx = %f", f->clip_minx);
1661 pandecode_prop("clip_miny = %f", f->clip_miny);
1662 pandecode_prop("clip_minz = %f", f->clip_minz);
1663 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1664 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1665 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1666
1667 /* Only the higher coordinates are MALI_POSITIVE scaled */
1668
1669 pandecode_prop("viewport0 = { %d, %d }",
1670 f->viewport0[0], f->viewport0[1]);
1671
1672 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1673 f->viewport1[0] + 1, f->viewport1[1] + 1);
1674
1675 pandecode_indent--;
1676 pandecode_log("};\n");
1677 }
1678
1679 if (p->attribute_meta) {
1680 unsigned max_attr_index = pandecode_attribute_meta(job_no, attribute_count, p, false, suffix);
1681
1682 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1683 pandecode_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1684 }
1685
1686 /* Varyings are encoded like attributes but not actually sent; we just
1687 * pass a zero buffer with the right stride/size set, (or whatever)
1688 * since the GPU will write to it itself */
1689
1690 if (p->varying_meta) {
1691 varying_count = pandecode_attribute_meta(job_no, varying_count, p, true, suffix);
1692 }
1693
1694 if (p->varyings) {
1695 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1696
1697 /* Number of descriptors depends on whether there are
1698 * non-internal varyings */
1699
1700 pandecode_attributes(attr_mem, p->varyings, job_no, suffix, varying_count, true);
1701 }
1702
1703 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1704
1705 if (p->uniforms && !is_compute) {
1706 int rows = uniform_count, width = 4;
1707 size_t sz = rows * width * sizeof(float);
1708
1709 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1710 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1711 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1712
1713 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1714
1715 pandecode_indent++;
1716
1717 for (int row = 0; row < rows; row++) {
1718 for (int i = 0; i < width; i++) {
1719 u32 v = uniforms[i];
1720 float f;
1721 memcpy(&f, &v, sizeof(v));
1722 pandecode_log_cont("%X /* %f */, ", v, f);
1723 }
1724
1725 pandecode_log_cont("\n");
1726
1727 uniforms += width;
1728 }
1729
1730 pandecode_indent--;
1731 pandecode_log("};\n");
1732 } else if (p->uniforms) {
1733 int rows = uniform_count * 2;
1734 size_t sz = rows * sizeof(mali_ptr);
1735
1736 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1737 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1738 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1739
1740 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1741
1742 pandecode_indent++;
1743
1744 for (int row = 0; row < rows; row++) {
1745 char *a = pointer_as_memory_reference(uniforms[row]);
1746 pandecode_log("%s,\n", a);
1747 free(a);
1748 }
1749
1750 pandecode_indent--;
1751 pandecode_log("};\n");
1752
1753 }
1754
1755 if (p->uniform_buffers) {
1756 pandecode_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1757 }
1758
1759 if (p->texture_trampoline) {
1760 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1761
1762 if (mmem) {
1763 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1764
1765 pandecode_log("uint64_t texture_trampoline_%"PRIx64"_%d[] = {\n", p->texture_trampoline, job_no);
1766 pandecode_indent++;
1767
1768 for (int tex = 0; tex < texture_count; ++tex) {
1769 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1770 char *a = pointer_as_memory_reference(*u);
1771 pandecode_log("%s,\n", a);
1772 free(a);
1773 }
1774
1775 pandecode_indent--;
1776 pandecode_log("};\n");
1777
1778 /* Now, finally, descend down into the texture descriptor */
1779 for (int tex = 0; tex < texture_count; ++tex) {
1780 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1781 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1782
1783 if (tmem) {
1784 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1785
1786 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1787 pandecode_indent++;
1788
1789 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1790 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1791 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1792 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1793 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1794 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1795 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1796
1797 struct mali_texture_format f = t->format;
1798
1799 pandecode_log(".format = {\n");
1800 pandecode_indent++;
1801
1802 pandecode_swizzle(f.swizzle);
1803 pandecode_prop("format = %s", pandecode_format(f.format));
1804 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1805 pandecode_prop("srgb = %" PRId32, f.srgb);
1806 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1807 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1808
1809 pandecode_indent--;
1810 pandecode_log("},\n");
1811
1812 pandecode_swizzle(t->swizzle);
1813
1814 if (t->swizzle_zero) {
1815 /* Shouldn't happen */
1816 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1817 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1818 }
1819
1820 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1821
1822 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1823 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1824 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1825
1826 pandecode_log(".payload = {\n");
1827 pandecode_indent++;
1828
1829 /* A bunch of bitmap pointers follow.
1830 * We work out the correct number,
1831 * based on the mipmap/cubemap
1832 * properties, but dump extra
1833 * possibilities to futureproof */
1834
1835 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1836 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1837
1838 /* Miptree for each face */
1839 if (f.type == MALI_TEX_CUBE)
1840 bitmap_count *= 6;
1841
1842 /* Array of textures */
1843 bitmap_count *= MALI_NEGATIVE(t->array_size);
1844
1845 /* Stride for each element */
1846 if (manual_stride)
1847 bitmap_count *= 2;
1848
1849 /* Sanity check the size */
1850 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1851 assert (bitmap_count <= max_count);
1852
1853 /* Dump more to be safe, but not _that_ much more */
1854 int safe_count = MIN2(bitmap_count * 2, max_count);
1855
1856 for (int i = 0; i < safe_count; ++i) {
1857 char *prefix = (i >= bitmap_count) ? "// " : "";
1858
1859 /* How we dump depends if this is a stride or a pointer */
1860
1861 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1862 /* signed 32-bit snuck in as a 64-bit pointer */
1863 uint64_t stride_set = t->payload[i];
1864 uint32_t clamped_stride = stride_set;
1865 int32_t stride = clamped_stride;
1866 assert(stride_set == clamped_stride);
1867 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1868 } else {
1869 char *a = pointer_as_memory_reference(t->payload[i]);
1870 pandecode_log("%s%s, \n", prefix, a);
1871 free(a);
1872 }
1873 }
1874
1875 pandecode_indent--;
1876 pandecode_log("},\n");
1877
1878 pandecode_indent--;
1879 pandecode_log("};\n");
1880 }
1881 }
1882 }
1883 }
1884
1885 if (p->sampler_descriptor) {
1886 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1887
1888 if (smem) {
1889 struct mali_sampler_descriptor *s;
1890
1891 mali_ptr d = p->sampler_descriptor;
1892
1893 for (int i = 0; i < sampler_count; ++i) {
1894 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1895
1896 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64"_%d_%d = {\n", d + sizeof(*s) * i, job_no, i);
1897 pandecode_indent++;
1898
1899 pandecode_log(".filter_mode = ");
1900 pandecode_log_decoded_flags(sampler_flag_info, s->filter_mode);
1901 pandecode_log_cont(",\n");
1902
1903 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1904 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1905
1906 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s->wrap_s));
1907 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s->wrap_t));
1908 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s->wrap_r));
1909
1910 pandecode_prop("compare_func = %s", pandecode_alt_func(s->compare_func));
1911
1912 if (s->zero || s->zero2) {
1913 pandecode_msg("Zero tripped\n");
1914 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1915 }
1916
1917 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1918
1919 pandecode_prop("border_color = { %f, %f, %f, %f }",
1920 s->border_color[0],
1921 s->border_color[1],
1922 s->border_color[2],
1923 s->border_color[3]);
1924
1925 pandecode_indent--;
1926 pandecode_log("};\n");
1927 }
1928 }
1929 }
1930 }
1931
1932 static void
1933 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1934 {
1935 pandecode_log_cont("{\n");
1936 pandecode_indent++;
1937
1938 MEMORY_PROP(p, position_varying);
1939 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1940 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1941 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1942 DYN_MEMORY_PROP(p, job_no, uniforms);
1943 DYN_MEMORY_PROP(p, job_no, attributes);
1944 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1945 DYN_MEMORY_PROP(p, job_no, varyings);
1946 DYN_MEMORY_PROP(p, job_no, varying_meta);
1947 DYN_MEMORY_PROP(p, job_no, viewport);
1948 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1949
1950 if (is_bifrost)
1951 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1952 else
1953 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1954
1955 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1956 pandecode_prop("flags = %d", p->flags);
1957
1958 pandecode_indent--;
1959 pandecode_log("},\n");
1960 }
1961
1962 static void
1963 pandecode_vertex_only_bfr(struct bifrost_vertex_only *v)
1964 {
1965 pandecode_log_cont("{\n");
1966 pandecode_indent++;
1967
1968 pandecode_prop("unk2 = 0x%x", v->unk2);
1969
1970 if (v->zero0 || v->zero1) {
1971 pandecode_msg("vertex only zero tripped");
1972 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1973 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1974 }
1975
1976 pandecode_indent--;
1977 pandecode_log("}\n");
1978 }
1979
1980 static void
1981 pandecode_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1982 {
1983
1984 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1985 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1986
1987 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1988 pandecode_indent++;
1989
1990 if (h->zero) {
1991 pandecode_msg("tiler heap zero tripped\n");
1992 pandecode_prop("zero = 0x%x", h->zero);
1993 }
1994
1995 for (int i = 0; i < 12; i++) {
1996 if (h->zeros[i] != 0) {
1997 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1998 i, h->zeros[i]);
1999 }
2000 }
2001
2002 pandecode_prop("heap_size = 0x%x", h->heap_size);
2003 MEMORY_PROP(h, tiler_heap_start);
2004 MEMORY_PROP(h, tiler_heap_free);
2005
2006 /* this might point to the beginning of another buffer, when it's
2007 * really the end of the tiler heap buffer, so we have to be careful
2008 * here. but for zero length, we need the same pointer.
2009 */
2010
2011 if (h->tiler_heap_end == h->tiler_heap_start) {
2012 MEMORY_PROP(h, tiler_heap_start);
2013 } else {
2014 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
2015 pandecode_prop("tiler_heap_end = %s + 1", a);
2016 free(a);
2017 }
2018
2019 pandecode_indent--;
2020 pandecode_log("};\n");
2021 }
2022
2023 static void
2024 pandecode_tiler_meta(mali_ptr gpu_va, int job_no)
2025 {
2026 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
2027 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
2028
2029 pandecode_tiler_heap_meta(t->tiler_heap_meta, job_no);
2030
2031 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
2032 pandecode_indent++;
2033
2034 if (t->zero0 || t->zero1) {
2035 pandecode_msg("tiler meta zero tripped");
2036 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
2037 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2038 }
2039
2040 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
2041 pandecode_prop("flags = 0x%" PRIx16, t->flags);
2042
2043 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
2044 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2045 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
2046
2047 for (int i = 0; i < 12; i++) {
2048 if (t->zeros[i] != 0) {
2049 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2050 i, t->zeros[i]);
2051 }
2052 }
2053
2054 pandecode_indent--;
2055 pandecode_log("};\n");
2056 }
2057
2058 static void
2059 pandecode_gl_enables(uint32_t gl_enables, int job_type)
2060 {
2061 pandecode_log(".gl_enables = ");
2062
2063 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2064
2065 pandecode_log_cont(",\n");
2066 }
2067
2068 static void
2069 pandecode_primitive_size(union midgard_primitive_size u, bool constant)
2070 {
2071 if (u.pointer == 0x0)
2072 return;
2073
2074 pandecode_log(".primitive_size = {\n");
2075 pandecode_indent++;
2076
2077 if (constant) {
2078 pandecode_prop("constant = %f", u.constant);
2079 } else {
2080 MEMORY_PROP((&u), pointer);
2081 }
2082
2083 pandecode_indent--;
2084 pandecode_log("},\n");
2085 }
2086
2087 static void
2088 pandecode_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2089 {
2090 pandecode_log_cont("{\n");
2091 pandecode_indent++;
2092
2093 /* TODO: gl_PointSize on Bifrost */
2094 pandecode_primitive_size(t->primitive_size, true);
2095
2096 DYN_MEMORY_PROP(t, job_no, tiler_meta);
2097 pandecode_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2098
2099 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2100 || t->zero6 || t->zero7 || t->zero8) {
2101 pandecode_msg("tiler only zero tripped");
2102 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2103 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2104 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2105 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2106 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2107 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2108 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2109 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2110 }
2111
2112 pandecode_indent--;
2113 pandecode_log("},\n");
2114 }
2115
2116 static int
2117 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2118 const struct pandecode_mapped_memory *mem,
2119 mali_ptr payload, int job_no)
2120 {
2121 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2122
2123 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2124
2125 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2126 pandecode_indent++;
2127
2128 pandecode_log(".prefix = ");
2129 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2130
2131 pandecode_log(".vertex = ");
2132 pandecode_vertex_only_bfr(&v->vertex);
2133
2134 pandecode_log(".postfix = ");
2135 pandecode_vertex_tiler_postfix(&v->postfix, job_no, true);
2136
2137 pandecode_indent--;
2138 pandecode_log("};\n");
2139
2140 return sizeof(*v);
2141 }
2142
2143 static int
2144 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2145 const struct pandecode_mapped_memory *mem,
2146 mali_ptr payload, int job_no)
2147 {
2148 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2149
2150 pandecode_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2151
2152 pandecode_indices(t->prefix.indices, t->prefix.index_count, job_no);
2153 pandecode_tiler_meta(t->tiler.tiler_meta, job_no);
2154
2155 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2156 pandecode_indent++;
2157
2158 pandecode_log(".prefix = ");
2159 pandecode_vertex_tiler_prefix(&t->prefix, job_no);
2160
2161 pandecode_log(".tiler = ");
2162 pandecode_tiler_only_bfr(&t->tiler, job_no);
2163
2164 pandecode_log(".postfix = ");
2165 pandecode_vertex_tiler_postfix(&t->postfix, job_no, true);
2166
2167 pandecode_indent--;
2168 pandecode_log("};\n");
2169
2170 return sizeof(*t);
2171 }
2172
2173 static int
2174 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2175 const struct pandecode_mapped_memory *mem,
2176 mali_ptr payload, int job_no)
2177 {
2178 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2179
2180 pandecode_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2181
2182 pandecode_indices(v->prefix.indices, v->prefix.index_count, job_no);
2183
2184 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2185 pandecode_indent++;
2186
2187 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2188 pandecode_primitive_size(v->primitive_size, !has_primitive_pointer);
2189
2190 pandecode_log(".prefix = ");
2191 pandecode_vertex_tiler_prefix(&v->prefix, job_no);
2192
2193 pandecode_gl_enables(v->gl_enables, h->job_type);
2194
2195 if (v->instance_shift || v->instance_odd) {
2196 pandecode_prop("instance_shift = 0x%d /* %d */",
2197 v->instance_shift, 1 << v->instance_shift);
2198 pandecode_prop("instance_odd = 0x%X /* %d */",
2199 v->instance_odd, (2 * v->instance_odd) + 1);
2200
2201 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2202 }
2203
2204 if (v->offset_start)
2205 pandecode_prop("offset_start = %d", v->offset_start);
2206
2207 if (v->zero5) {
2208 pandecode_msg("Zero tripped\n");
2209 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2210 }
2211
2212 pandecode_log(".postfix = ");
2213 pandecode_vertex_tiler_postfix(&v->postfix, job_no, false);
2214
2215 pandecode_indent--;
2216 pandecode_log("};\n");
2217
2218 return sizeof(*v);
2219 }
2220
2221 static int
2222 pandecode_fragment_job(const struct pandecode_mapped_memory *mem,
2223 mali_ptr payload, int job_no,
2224 bool is_bifrost)
2225 {
2226 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2227
2228 bool fbd_dumped = false;
2229
2230 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2231 /* Only SFBDs are understood, not MFBDs. We're speculating,
2232 * based on the versioning, kernel code, etc, that the
2233 * difference is between Single FrameBuffer Descriptor and
2234 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2235 * up with multi-framebuffer support being added (T7xx onwards,
2236 * including Gxx). In any event, there's some field shuffling
2237 * that we haven't looked into yet. */
2238
2239 pandecode_sfbd(s->framebuffer & FBD_MASK, job_no);
2240 fbd_dumped = true;
2241 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2242 /* We don't know if Bifrost supports SFBD's at all, since the
2243 * driver never uses them. And the format is different from
2244 * Midgard anyways, due to the tiler heap and scratchpad being
2245 * moved out into separate structures, so it's not clear what a
2246 * Bifrost SFBD would even look like without getting an actual
2247 * trace, which appears impossible.
2248 */
2249
2250 pandecode_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2251 fbd_dumped = true;
2252 }
2253
2254 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2255 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2256 pandecode_indent++;
2257
2258 /* See the comments by the macro definitions for mathematical context
2259 * on why this is so weird */
2260
2261 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2262 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2263
2264 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2265 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2266 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2267
2268 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2269 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2270 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2271
2272 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2273 * we have to fallback on offsets. */
2274
2275 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2276
2277 if (fbd_dumped)
2278 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2279 else
2280 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2281
2282 pandecode_indent--;
2283 pandecode_log("};\n");
2284
2285 return sizeof(*s);
2286 }
2287
2288 static int job_descriptor_number = 0;
2289
2290 int
2291 pandecode_jc(mali_ptr jc_gpu_va, bool bifrost)
2292 {
2293 struct mali_job_descriptor_header *h;
2294
2295 int start_number = 0;
2296
2297 bool first = true;
2298 bool last_size;
2299
2300 do {
2301 struct pandecode_mapped_memory *mem =
2302 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2303
2304 void *payload;
2305
2306 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2307
2308 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2309 * high 32-bits of the 64-bit pointer are reused to store
2310 * something else.
2311 */
2312 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2313 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2314 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2315
2316 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2317 MALI_PAYLOAD_SIZE);
2318
2319 int job_no = job_descriptor_number++;
2320
2321 if (first)
2322 start_number = job_no;
2323
2324 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2325 pandecode_indent++;
2326
2327 pandecode_prop("job_type = %s", pandecode_job_type(h->job_type));
2328
2329 /* Save for next job fixing */
2330 last_size = h->job_descriptor_size;
2331
2332 if (h->job_descriptor_size)
2333 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2334
2335 if (h->exception_status != 0x1)
2336 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2337 h->exception_status,
2338 (h->exception_status >> 16) & 0xFFFF,
2339 (h->exception_status >> 8) & 0x3,
2340 h->exception_status & 0xFF);
2341
2342 if (h->first_incomplete_task)
2343 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2344
2345 if (h->fault_pointer)
2346 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2347
2348 if (h->job_barrier)
2349 pandecode_prop("job_barrier = %d", h->job_barrier);
2350
2351 pandecode_prop("job_index = %d", h->job_index);
2352
2353 if (h->unknown_flags)
2354 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2355
2356 if (h->job_dependency_index_1)
2357 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2358
2359 if (h->job_dependency_index_2)
2360 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2361
2362 pandecode_indent--;
2363 pandecode_log("};\n");
2364
2365 /* Do not touch the field yet -- decode the payload first, and
2366 * don't touch that either. This is essential for the uploads
2367 * to occur in sequence and therefore be dynamically allocated
2368 * correctly. Do note the size, however, for that related
2369 * reason. */
2370
2371 switch (h->job_type) {
2372 case JOB_TYPE_SET_VALUE: {
2373 struct mali_payload_set_value *s = payload;
2374 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2375 pandecode_indent++;
2376 MEMORY_PROP(s, out);
2377 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2378 pandecode_indent--;
2379 pandecode_log("};\n");
2380
2381 break;
2382 }
2383
2384 case JOB_TYPE_TILER:
2385 case JOB_TYPE_VERTEX:
2386 case JOB_TYPE_COMPUTE:
2387 if (bifrost) {
2388 if (h->job_type == JOB_TYPE_TILER)
2389 pandecode_tiler_job_bfr(h, mem, payload_ptr, job_no);
2390 else
2391 pandecode_vertex_job_bfr(h, mem, payload_ptr, job_no);
2392 } else
2393 pandecode_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2394
2395 break;
2396
2397 case JOB_TYPE_FRAGMENT:
2398 pandecode_fragment_job(mem, payload_ptr, job_no, bifrost);
2399 break;
2400
2401 default:
2402 break;
2403 }
2404
2405 /* Handle linkage */
2406
2407 if (!first) {
2408 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2409
2410 if (last_size)
2411 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2412 else
2413 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2414 }
2415
2416 first = false;
2417
2418 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2419
2420 return start_number;
2421 }