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