panfrost/midgard: Use unsigned blend patch offset
[mesa.git] / src / gallium / drivers / 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 int pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost);
39
40 #define MEMORY_PROP(obj, p) {\
41 if (obj->p) { \
42 char *a = pointer_as_memory_reference(obj->p); \
43 pandecode_prop("%s = %s", #p, a); \
44 free(a); \
45 } \
46 }
47
48 #define DYN_MEMORY_PROP(obj, no, p) { \
49 if (obj->p) \
50 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
51 }
52
53 /* Semantic logging type.
54 *
55 * Raw: for raw messages to be printed as is.
56 * Message: for helpful information to be commented out in replays.
57 * Property: for properties of a struct
58 *
59 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
60 */
61
62 enum pandecode_log_type {
63 PANDECODE_RAW,
64 PANDECODE_MESSAGE,
65 PANDECODE_PROPERTY
66 };
67
68 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
69 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
70 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
71
72 unsigned pandecode_indent = 0;
73
74 static void
75 pandecode_make_indent(void)
76 {
77 for (unsigned i = 0; i < pandecode_indent; ++i)
78 printf(" ");
79 }
80
81 static void
82 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
83 {
84 va_list ap;
85
86 pandecode_make_indent();
87
88 if (type == PANDECODE_MESSAGE)
89 printf("// ");
90 else if (type == PANDECODE_PROPERTY)
91 printf(".");
92
93 va_start(ap, format);
94 vprintf(format, ap);
95 va_end(ap);
96
97 if (type == PANDECODE_PROPERTY)
98 printf(",\n");
99 }
100
101 static void
102 pandecode_log_cont(const char *format, ...)
103 {
104 va_list ap;
105
106 va_start(ap, format);
107 vprintf(format, ap);
108 va_end(ap);
109 }
110
111 struct pandecode_flag_info {
112 u64 flag;
113 const char *name;
114 };
115
116 static void
117 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
118 u64 flags)
119 {
120 bool decodable_flags_found = false;
121
122 for (int i = 0; flag_info[i].name; i++) {
123 if ((flags & flag_info[i].flag) != flag_info[i].flag)
124 continue;
125
126 if (!decodable_flags_found) {
127 decodable_flags_found = true;
128 } else {
129 pandecode_log_cont(" | ");
130 }
131
132 pandecode_log_cont("%s", flag_info[i].name);
133
134 flags &= ~flag_info[i].flag;
135 }
136
137 if (decodable_flags_found) {
138 if (flags)
139 pandecode_log_cont(" | 0x%" PRIx64, flags);
140 } else {
141 pandecode_log_cont("0x%" PRIx64, flags);
142 }
143 }
144
145 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
146 static const struct pandecode_flag_info gl_enable_flag_info[] = {
147 FLAG_INFO(OCCLUSION_QUERY),
148 FLAG_INFO(OCCLUSION_PRECISE),
149 FLAG_INFO(FRONT_CCW_TOP),
150 FLAG_INFO(CULL_FACE_FRONT),
151 FLAG_INFO(CULL_FACE_BACK),
152 {}
153 };
154 #undef FLAG_INFO
155
156 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
157 static const struct pandecode_flag_info clear_flag_info[] = {
158 FLAG_INFO(FAST),
159 FLAG_INFO(SLOW),
160 FLAG_INFO(SLOW_STENCIL),
161 {}
162 };
163 #undef FLAG_INFO
164
165 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
166 static const struct pandecode_flag_info mask_flag_info[] = {
167 FLAG_INFO(R),
168 FLAG_INFO(G),
169 FLAG_INFO(B),
170 FLAG_INFO(A),
171 {}
172 };
173 #undef FLAG_INFO
174
175 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
176 static const struct pandecode_flag_info u3_flag_info[] = {
177 FLAG_INFO(HAS_MSAA),
178 FLAG_INFO(CAN_DISCARD),
179 FLAG_INFO(HAS_BLEND_SHADER),
180 FLAG_INFO(DEPTH_TEST),
181 {}
182 };
183
184 static const struct pandecode_flag_info u4_flag_info[] = {
185 FLAG_INFO(NO_MSAA),
186 FLAG_INFO(NO_DITHER),
187 FLAG_INFO(DEPTH_RANGE_A),
188 FLAG_INFO(DEPTH_RANGE_B),
189 FLAG_INFO(STENCIL_TEST),
190 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
191 {}
192 };
193 #undef FLAG_INFO
194
195 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
196 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
197 FLAG_INFO(MSAA_A),
198 FLAG_INFO(MSAA_B),
199 FLAG_INFO(MSAA_8),
200 {}
201 };
202 #undef FLAG_INFO
203
204 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
205 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
206 FLAG_INFO(MSAA),
207 FLAG_INFO(SRGB),
208 {}
209 };
210 #undef FLAG_INFO
211
212 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
213 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
214 FLAG_INFO(PRESENT),
215 FLAG_INFO(AFBC),
216 FLAG_INFO(ZS),
217 {}
218 };
219 #undef FLAG_INFO
220
221 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
222 static const struct pandecode_flag_info shader_midgard1_flag_info [] = {
223 FLAG_INFO(EARLY_Z),
224 FLAG_INFO(HELPER_INVOCATIONS),
225 FLAG_INFO(READS_TILEBUFFER),
226 FLAG_INFO(READS_ZS),
227 {}
228 };
229 #undef FLAG_INFO
230
231 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
232 static const struct pandecode_flag_info mfbd_flag_info [] = {
233 FLAG_INFO(DEPTH_WRITE),
234 FLAG_INFO(EXTRA),
235 {}
236 };
237 #undef FLAG_INFO
238
239
240 extern char *replace_fragment;
241 extern char *replace_vertex;
242
243 static char *
244 pandecode_job_type_name(enum mali_job_type type)
245 {
246 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
247
248 switch (type) {
249 DEFINE_CASE(NULL);
250 DEFINE_CASE(SET_VALUE);
251 DEFINE_CASE(CACHE_FLUSH);
252 DEFINE_CASE(COMPUTE);
253 DEFINE_CASE(VERTEX);
254 DEFINE_CASE(TILER);
255 DEFINE_CASE(FUSED);
256 DEFINE_CASE(FRAGMENT);
257
258 case JOB_NOT_STARTED:
259 return "NOT_STARTED";
260
261 default:
262 pandecode_log("Warning! Unknown job type %x\n", type);
263 return "!?!?!?";
264 }
265
266 #undef DEFINE_CASE
267 }
268
269 static char *
270 pandecode_draw_mode_name(enum mali_draw_mode mode)
271 {
272 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
273
274 switch (mode) {
275 DEFINE_CASE(DRAW_NONE);
276 DEFINE_CASE(POINTS);
277 DEFINE_CASE(LINES);
278 DEFINE_CASE(TRIANGLES);
279 DEFINE_CASE(TRIANGLE_STRIP);
280 DEFINE_CASE(TRIANGLE_FAN);
281 DEFINE_CASE(LINE_STRIP);
282 DEFINE_CASE(LINE_LOOP);
283 DEFINE_CASE(POLYGON);
284 DEFINE_CASE(QUADS);
285 DEFINE_CASE(QUAD_STRIP);
286
287 default:
288 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
289 }
290
291 #undef DEFINE_CASE
292 }
293
294 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
295 static char *
296 pandecode_func_name(enum mali_func mode)
297 {
298 switch (mode) {
299 DEFINE_CASE(NEVER);
300 DEFINE_CASE(LESS);
301 DEFINE_CASE(EQUAL);
302 DEFINE_CASE(LEQUAL);
303 DEFINE_CASE(GREATER);
304 DEFINE_CASE(NOTEQUAL);
305 DEFINE_CASE(GEQUAL);
306 DEFINE_CASE(ALWAYS);
307
308 default:
309 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
310 }
311 }
312 #undef DEFINE_CASE
313
314 /* Why is this duplicated? Who knows... */
315 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
316 static char *
317 pandecode_alt_func_name(enum mali_alt_func mode)
318 {
319 switch (mode) {
320 DEFINE_CASE(NEVER);
321 DEFINE_CASE(LESS);
322 DEFINE_CASE(EQUAL);
323 DEFINE_CASE(LEQUAL);
324 DEFINE_CASE(GREATER);
325 DEFINE_CASE(NOTEQUAL);
326 DEFINE_CASE(GEQUAL);
327 DEFINE_CASE(ALWAYS);
328
329 default:
330 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
331 }
332 }
333 #undef DEFINE_CASE
334
335 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
336 static char *
337 pandecode_stencil_op_name(enum mali_stencil_op op)
338 {
339 switch (op) {
340 DEFINE_CASE(KEEP);
341 DEFINE_CASE(REPLACE);
342 DEFINE_CASE(ZERO);
343 DEFINE_CASE(INVERT);
344 DEFINE_CASE(INCR_WRAP);
345 DEFINE_CASE(DECR_WRAP);
346 DEFINE_CASE(INCR);
347 DEFINE_CASE(DECR);
348
349 default:
350 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
351 }
352 }
353
354 #undef DEFINE_CASE
355
356 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
357 static char *pandecode_attr_mode_name(enum mali_attr_mode mode)
358 {
359 switch(mode) {
360 DEFINE_CASE(UNUSED);
361 DEFINE_CASE(LINEAR);
362 DEFINE_CASE(POT_DIVIDE);
363 DEFINE_CASE(MODULO);
364 DEFINE_CASE(NPOT_DIVIDE);
365 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
366 }
367 }
368
369 #undef DEFINE_CASE
370
371 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
372 static char *
373 pandecode_channel_name(enum mali_channel channel)
374 {
375 switch (channel) {
376 DEFINE_CASE(RED);
377 DEFINE_CASE(GREEN);
378 DEFINE_CASE(BLUE);
379 DEFINE_CASE(ALPHA);
380 DEFINE_CASE(ZERO);
381 DEFINE_CASE(ONE);
382 DEFINE_CASE(RESERVED_0);
383 DEFINE_CASE(RESERVED_1);
384
385 default:
386 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
387 }
388 }
389 #undef DEFINE_CASE
390
391 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
392 static char *
393 pandecode_wrap_mode_name(enum mali_wrap_mode op)
394 {
395 switch (op) {
396 DEFINE_CASE(REPEAT);
397 DEFINE_CASE(CLAMP_TO_EDGE);
398 DEFINE_CASE(CLAMP_TO_BORDER);
399 DEFINE_CASE(MIRRORED_REPEAT);
400
401 default:
402 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
403 }
404 }
405 #undef DEFINE_CASE
406
407 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
408 static char *
409 pandecode_texture_type(enum mali_texture_type type)
410 {
411 switch (type) {
412 DEFINE_CASE(1D);
413 DEFINE_CASE(2D);
414 DEFINE_CASE(3D);
415 DEFINE_CASE(CUBE);
416
417 default:
418 unreachable("Unknown case");
419 }
420 }
421 #undef DEFINE_CASE
422
423 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
424 static char *
425 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
426 {
427 switch (fmt) {
428 DEFINE_CASE(TILED);
429 DEFINE_CASE(UNKNOWN);
430 DEFINE_CASE(LINEAR);
431 DEFINE_CASE(AFBC);
432
433 default:
434 unreachable("Invalid case");
435 }
436 }
437 #undef DEFINE_CASE
438
439 static inline char *
440 pandecode_decode_fbd_type(enum mali_fbd_type type)
441 {
442 if (type == MALI_SFBD) return "SFBD";
443 else if (type == MALI_MFBD) return "MFBD";
444 else return "WATFBD /* XXX */";
445 }
446
447 static void
448 pandecode_replay_sfbd(uint64_t gpu_va, int job_no)
449 {
450 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
451 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
452
453 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
454 pandecode_indent++;
455
456 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
457 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
458
459 pandecode_log(".format = ");
460 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
461 pandecode_log_cont(",\n");
462
463 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
464 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
465
466 MEMORY_PROP(s, framebuffer);
467 pandecode_prop("stride = %d", s->stride);
468
469 /* Earlier in the actual commandstream -- right before width -- but we
470 * delay to flow nicer */
471
472 pandecode_log(".clear_flags = ");
473 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
474 pandecode_log_cont(",\n");
475
476 if (s->depth_buffer | s->depth_buffer_enable) {
477 MEMORY_PROP(s, depth_buffer);
478 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
479 }
480
481 if (s->stencil_buffer | s->stencil_buffer_enable) {
482 MEMORY_PROP(s, stencil_buffer);
483 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
484 }
485
486 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
487 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
488 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
489 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
490 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
491 }
492
493 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
494 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
495 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
496 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
497 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
498 }
499
500 if (s->clear_stencil) {
501 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
502 }
503
504 MEMORY_PROP(s, unknown_address_0);
505 MEMORY_PROP(s, tiler_polygon_list);
506 MEMORY_PROP(s, tiler_polygon_list_body);
507
508 pandecode_prop("tiler_resolution_check = 0x%" PRIx32, s->tiler_resolution_check);
509 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, s->tiler_hierarchy_mask);
510 pandecode_prop("tiler_flags = 0x%" PRIx16, s->tiler_flags);
511
512 MEMORY_PROP(s, tiler_heap_free);
513 MEMORY_PROP(s, tiler_heap_end);
514
515 pandecode_indent--;
516 pandecode_log("};\n");
517
518 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
519 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
520 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
521 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
522
523 printf(".zero3 = {");
524
525 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
526 printf("%X, ", s->zero3[i]);
527
528 printf("},\n");
529
530 printf(".zero6 = {");
531
532 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
533 printf("%X, ", s->zero6[i]);
534
535 printf("},\n");
536 }
537
538 static void
539 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
540 {
541 pandecode_log(".unknown%d = {", name);
542
543 for (int i = 0; i < count; ++i)
544 printf("%X, ", slide[i]);
545
546 pandecode_log("},\n");
547 }
548
549 #define SHORT_SLIDE(num) \
550 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
551
552 static void
553 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
554 {
555 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
556 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
557
558 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
559 pandecode_indent++;
560
561 SHORT_SLIDE(1);
562
563 pandecode_indent--;
564 printf("},\n");
565 }
566
567 static void
568 pandecode_replay_swizzle(unsigned swizzle)
569 {
570 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
571 pandecode_channel_name((swizzle >> 0) & 0x7),
572 pandecode_channel_name((swizzle >> 3) & 0x7),
573 pandecode_channel_name((swizzle >> 6) & 0x7),
574 pandecode_channel_name((swizzle >> 9) & 0x7));
575 }
576
577 static void
578 pandecode_rt_format(struct mali_rt_format format)
579 {
580 pandecode_log(".format = {\n");
581 pandecode_indent++;
582
583 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
584 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
585 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
586
587 pandecode_prop("block = %s",
588 pandecode_mfbd_block_format(format.block));
589
590 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
591 MALI_NEGATIVE(format.nr_channels));
592
593 pandecode_log(".flags = ");
594 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
595 pandecode_log_cont(",\n");
596
597 pandecode_replay_swizzle(format.swizzle);
598
599 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
600
601 pandecode_indent--;
602 pandecode_log("},\n");
603 }
604
605 static void
606 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
607 {
608 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
609 pandecode_indent++;
610
611 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
612 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
613 struct pandecode_mapped_memory *mem =
614 pandecode_find_mapped_gpu_mem_containing(rt_va);
615 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
616
617 pandecode_log("{\n");
618 pandecode_indent++;
619
620 pandecode_rt_format(rt->format);
621
622 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
623 pandecode_log(".afbc = {\n");
624 pandecode_indent++;
625
626 char *a = pointer_as_memory_reference(rt->afbc.metadata);
627 pandecode_prop("metadata = %s", a);
628 free(a);
629
630 pandecode_prop("stride = %d", rt->afbc.stride);
631 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
632
633 pandecode_indent--;
634 pandecode_log("},\n");
635 } else {
636 pandecode_log(".chunknown = {\n");
637 pandecode_indent++;
638
639 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
640
641 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
642 pandecode_prop("pointer = %s", a);
643 free(a);
644
645 pandecode_indent--;
646 pandecode_log("},\n");
647 }
648
649 MEMORY_PROP(rt, framebuffer);
650 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
651
652 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
653 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
654 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
655 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
656 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
657 }
658
659 if (rt->zero1 || rt->zero2 || rt->zero3) {
660 pandecode_msg("render target zeros tripped\n");
661 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
662 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
663 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
664 }
665
666 pandecode_indent--;
667 pandecode_log("},\n");
668 }
669
670 pandecode_indent--;
671 pandecode_log("};\n");
672 }
673
674 static unsigned
675 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
676 {
677 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
678 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
679
680 if (fb->sample_locations) {
681 /* The blob stores all possible sample locations in a single buffer
682 * allocated on startup, and just switches the pointer when switching
683 * MSAA state. For now, we just put the data into the cmdstream, but we
684 * should do something like what the blob does with a real driver.
685 *
686 * There seem to be 32 slots for sample locations, followed by another
687 * 16. The second 16 is just the center location followed by 15 zeros
688 * in all the cases I've identified (maybe shader vs. depth/color
689 * samples?).
690 */
691
692 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
693
694 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
695
696 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
697 pandecode_indent++;
698
699 for (int i = 0; i < 32 + 16; i++) {
700 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
701 }
702
703 pandecode_indent--;
704 pandecode_log("};\n");
705 }
706
707 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
708 pandecode_indent++;
709
710 pandecode_prop("unk0 = 0x%x", fb->unk0);
711
712 if (fb->sample_locations)
713 pandecode_prop("sample_locations = sample_locations_%d", job_no);
714
715 /* Assume that unknown1 was emitted in the last job for
716 * now */
717 MEMORY_PROP(fb, unknown1);
718
719 pandecode_prop("tiler_polygon_list_size = 0x%x", fb->tiler_polygon_list_size);
720 pandecode_prop("tiler_hierarchy_mask = 0x%" PRIx16, fb->tiler_hierarchy_mask);
721 pandecode_prop("tiler_flags = 0x%" PRIx16, fb->tiler_flags);
722
723 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
724 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
725 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
726 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
727
728 pandecode_prop("unk1 = 0x%x", fb->unk1);
729 pandecode_prop("unk2 = 0x%x", fb->unk2);
730 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
731 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
732
733 pandecode_log(".mfbd_flags = ");
734 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
735 pandecode_log_cont(",\n");
736
737 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
738 pandecode_prop("clear_depth = %f", fb->clear_depth);
739
740 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
741 MEMORY_PROP(fb, scratchpad);
742 MEMORY_PROP(fb, tiler_polygon_list);
743 MEMORY_PROP(fb, tiler_polygon_list_body);
744 MEMORY_PROP(fb, tiler_heap_start);
745 MEMORY_PROP(fb, tiler_heap_end);
746
747 if (fb->zero3 || fb->zero4) {
748 pandecode_msg("framebuffer zeros tripped\n");
749 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
750 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
751 }
752
753 bool nonzero_weights = false;
754
755 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
756 nonzero_weights |= fb->tiler_weights[w] != 0x0;
757 }
758
759 if (nonzero_weights) {
760 pandecode_log(".tiler_weights = {");
761
762 for (unsigned w = 0; w < ARRAY_SIZE(fb->tiler_weights); ++w) {
763 pandecode_log("%d, ", fb->tiler_weights[w]);
764 }
765
766 pandecode_log("},");
767 }
768
769 pandecode_indent--;
770 pandecode_log("};\n");
771
772 gpu_va += sizeof(struct bifrost_framebuffer);
773
774 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
775 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
776 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
777
778 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
779 pandecode_indent++;
780
781 MEMORY_PROP(fbx, checksum);
782
783 if (fbx->checksum_stride)
784 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
785
786 pandecode_log(".flags = ");
787 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
788 pandecode_log_cont(",\n");
789
790 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
791 pandecode_log(".ds_afbc = {\n");
792 pandecode_indent++;
793
794 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
795 pandecode_prop("depth_stencil_afbc_stride = %d",
796 fbx->ds_afbc.depth_stencil_afbc_stride);
797 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
798
799 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
800 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
801 pandecode_prop("zero1 = 0x%" PRIx32,
802 fbx->ds_afbc.zero1);
803 pandecode_prop("padding = 0x%" PRIx64,
804 fbx->ds_afbc.padding);
805 }
806
807 pandecode_indent--;
808 pandecode_log("},\n");
809 } else {
810 pandecode_log(".ds_linear = {\n");
811 pandecode_indent++;
812
813 if (fbx->ds_linear.depth) {
814 MEMORY_PROP((&fbx->ds_linear), depth);
815 pandecode_prop("depth_stride = %d",
816 fbx->ds_linear.depth_stride);
817 }
818
819 if (fbx->ds_linear.stencil) {
820 MEMORY_PROP((&fbx->ds_linear), stencil);
821 pandecode_prop("stencil_stride = %d",
822 fbx->ds_linear.stencil_stride);
823 }
824
825 if (fbx->ds_linear.depth_stride_zero ||
826 fbx->ds_linear.stencil_stride_zero ||
827 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
828 pandecode_msg("Depth/stencil zeros tripped\n");
829 pandecode_prop("depth_stride_zero = 0x%x",
830 fbx->ds_linear.depth_stride_zero);
831 pandecode_prop("stencil_stride_zero = 0x%x",
832 fbx->ds_linear.stencil_stride_zero);
833 pandecode_prop("zero1 = 0x%" PRIx32,
834 fbx->ds_linear.zero1);
835 pandecode_prop("zero2 = 0x%" PRIx32,
836 fbx->ds_linear.zero2);
837 }
838
839 pandecode_indent--;
840 pandecode_log("},\n");
841 }
842
843 if (fbx->zero3 || fbx->zero4) {
844 pandecode_msg("fb_extra zeros tripped\n");
845 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
846 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
847 }
848
849 pandecode_indent--;
850 pandecode_log("};\n");
851
852 gpu_va += sizeof(struct bifrost_fb_extra);
853 }
854
855 if (with_render_targets)
856 pandecode_render_target(gpu_va, job_no, fb);
857
858 /* Passback the render target count */
859 return MALI_NEGATIVE(fb->rt_count_1);
860 }
861
862 /* Just add a comment decoding the shift/odd fields forming the padded vertices
863 * count */
864
865 static void
866 pandecode_padded_vertices(unsigned shift, unsigned k)
867 {
868 unsigned odd = 2*k + 1;
869 unsigned pot = 1 << shift;
870 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
871 }
872
873 /* Given a magic divisor, recover what we were trying to divide by.
874 *
875 * Let m represent the magic divisor. By definition, m is an element on Z, whre
876 * 0 <= m < 2^N, for N bits in m.
877 *
878 * Let q represent the number we would like to divide by.
879 *
880 * By definition of a magic divisor for N-bit unsigned integers (a number you
881 * multiply by to magically get division), m is a number such that:
882 *
883 * (m * x) & (2^N - 1) = floor(x/q).
884 * for all x on Z where 0 <= x < 2^N
885 *
886 * Ignore the case where any of the above values equals zero; it is irrelevant
887 * for our purposes (instanced arrays).
888 *
889 * Choose x = q. Then:
890 *
891 * (m * x) & (2^N - 1) = floor(x/q).
892 * (m * q) & (2^N - 1) = floor(q/q).
893 *
894 * floor(q/q) = floor(1) = 1, therefore:
895 *
896 * (m * q) & (2^N - 1) = 1
897 *
898 * Recall the identity that the bitwise AND of one less than a power-of-two
899 * equals the modulo with that power of two, i.e. for all x:
900 *
901 * x & (2^N - 1) = x % N
902 *
903 * Therefore:
904 *
905 * mq % (2^N) = 1
906 *
907 * By definition, a modular multiplicative inverse of a number m is the number
908 * q such that with respect to a modulos M:
909 *
910 * mq % M = 1
911 *
912 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
913 *
914 */
915
916 static void
917 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
918 {
919 /* Compute the modular inverse of `magic` with respect to 2^(32 -
920 * shift) the most lame way possible... just repeatedly add.
921 * Asymptoptically slow but nobody cares in practice, unless you have
922 * massive numbers of vertices or high divisors. */
923
924 unsigned inverse = 0;
925
926 /* Magic implicitly has the highest bit set */
927 magic |= (1 << 31);
928
929 /* Depending on rounding direction */
930 if (extra)
931 magic++;
932
933 for (;;) {
934 uint32_t product = magic * inverse;
935
936 if (shift) {
937 product >>= shift;
938 }
939
940 if (product == 1)
941 break;
942
943 ++inverse;
944 }
945
946 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
947
948 /* Recall we're supposed to divide by (gl_level_divisor *
949 * padded_num_vertices) */
950
951 unsigned padded_num_vertices = inverse / orig_divisor;
952
953 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
954 }
955
956 static void
957 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
958 mali_ptr addr, int job_no, char *suffix,
959 int count, bool varying)
960 {
961 char *prefix = varying ? "varyings" : "attributes";
962
963 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
964
965 char base[128];
966 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
967
968 for (int i = 0; i < count; ++i) {
969 enum mali_attr_mode mode = attr[i].elements & 7;
970
971 if (mode == MALI_ATTR_UNUSED)
972 continue;
973
974 mali_ptr raw_elements = attr[i].elements & ~7;
975
976 /* TODO: Do we maybe want to dump the attribute values
977 * themselves given the specified format? Or is that too hard?
978 * */
979
980 char *a = pointer_as_memory_reference(raw_elements);
981 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
982 free(a);
983 }
984
985 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
986 pandecode_indent++;
987
988 for (int i = 0; i < count; ++i) {
989 pandecode_log("{\n");
990 pandecode_indent++;
991
992 unsigned mode = attr[i].elements & 7;
993 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(mode));
994 pandecode_prop("shift = %d", attr[i].shift);
995 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
996 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
997 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
998
999 /* Decode further where possible */
1000
1001 if (mode == MALI_ATTR_MODULO) {
1002 pandecode_padded_vertices(
1003 attr[i].shift,
1004 attr[i].extra_flags);
1005 }
1006
1007 pandecode_indent--;
1008 pandecode_log("}, \n");
1009
1010 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1011 i++;
1012 pandecode_log("{\n");
1013 pandecode_indent++;
1014 pandecode_prop("unk = 0x%x", attr[i].unk);
1015 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1016 if (attr[i].zero != 0)
1017 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1018 pandecode_prop("divisor = %d", attr[i].divisor);
1019 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1020 pandecode_indent--;
1021 pandecode_log("}, \n");
1022 }
1023
1024 }
1025
1026 pandecode_indent--;
1027 pandecode_log("};\n");
1028 }
1029
1030 static mali_ptr
1031 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
1032 {
1033 /* TODO: Decode flags */
1034 mali_ptr shader_ptr = ptr & ~15;
1035
1036 char *a = pointer_as_memory_reference(shader_ptr);
1037 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1038 free(a);
1039
1040 return shader_ptr;
1041 }
1042
1043 static bool
1044 all_zero(unsigned *buffer, unsigned count)
1045 {
1046 for (unsigned i = 0; i < count; ++i) {
1047 if (buffer[i])
1048 return false;
1049 }
1050
1051 return true;
1052 }
1053
1054 static void
1055 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
1056 {
1057 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1058 return;
1059
1060 const char *func = pandecode_func_name(stencil->func);
1061 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
1062 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
1063 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
1064
1065 if (stencil->zero)
1066 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1067
1068 pandecode_log(".stencil_%s = {\n", name);
1069 pandecode_indent++;
1070 pandecode_prop("ref = %d", stencil->ref);
1071 pandecode_prop("mask = 0x%02X", stencil->mask);
1072 pandecode_prop("func = %s", func);
1073 pandecode_prop("sfail = %s", sfail);
1074 pandecode_prop("dpfail = %s", dpfail);
1075 pandecode_prop("dppass = %s", dppass);
1076 pandecode_indent--;
1077 pandecode_log("},\n");
1078 }
1079
1080 static void
1081 pandecode_replay_blend_equation(const struct mali_blend_equation *blend)
1082 {
1083 if (blend->zero1)
1084 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1085
1086 pandecode_log(".equation = {\n");
1087 pandecode_indent++;
1088
1089 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1090 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1091
1092 pandecode_log(".color_mask = ");
1093 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1094 pandecode_log_cont(",\n");
1095
1096 pandecode_indent--;
1097 pandecode_log("},\n");
1098 }
1099
1100 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1101
1102 static unsigned
1103 decode_bifrost_constant(u16 constant)
1104 {
1105 float lo = (float) (constant & 0xFF);
1106 float hi = (float) (constant >> 8);
1107
1108 return (hi / 255.0) + (lo / 65535.0);
1109 }
1110
1111 static mali_ptr
1112 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1113 {
1114 struct bifrost_blend_rt *b =
1115 ((struct bifrost_blend_rt *) descs) + rt_no;
1116
1117 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1118 pandecode_indent++;
1119
1120 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1121 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1122 b->constant, decode_bifrost_constant(b->constant));
1123
1124 /* TODO figure out blend shader enable bit */
1125 pandecode_replay_blend_equation(&b->equation);
1126 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1127 pandecode_prop("index = 0x%" PRIx16, b->index);
1128 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1129
1130 pandecode_indent--;
1131 pandecode_log("},\n");
1132
1133 return 0;
1134 }
1135
1136 static mali_ptr
1137 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1138 {
1139 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1140 return 0;
1141
1142 pandecode_log(".blend = {\n");
1143 pandecode_indent++;
1144
1145 if (is_shader) {
1146 pandecode_replay_shader_address("shader", blend->shader);
1147 } else {
1148 pandecode_replay_blend_equation(&blend->equation);
1149 pandecode_prop("constant = %f", blend->constant);
1150 }
1151
1152 pandecode_indent--;
1153 pandecode_log("},\n");
1154
1155 /* Return blend shader to disassemble if present */
1156 return is_shader ? (blend->shader & ~0xF) : 0;
1157 }
1158
1159 static mali_ptr
1160 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1161 {
1162 struct midgard_blend_rt *b =
1163 ((struct midgard_blend_rt *) descs) + rt_no;
1164
1165 /* Flags determine presence of blend shader */
1166 bool is_shader = (b->flags & 0xF) >= 0x2;
1167
1168 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1169 pandecode_indent++;
1170
1171 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1172
1173 mali_ptr shader = pandecode_midgard_blend(&b->blend, is_shader);
1174
1175 pandecode_indent--;
1176 pandecode_log("};\n");
1177
1178 return shader;
1179 }
1180
1181 static int
1182 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1183 {
1184 char base[128];
1185 char *prefix = varying ? "varying" : "attribute";
1186 unsigned max_index = 0;
1187 snprintf(base, sizeof(base), "%s_meta", prefix);
1188
1189 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1190 pandecode_indent++;
1191
1192 struct mali_attr_meta *attr_meta;
1193 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1194
1195 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1196
1197 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1198 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1199 sizeof(*attr_mem));
1200
1201 pandecode_log("{\n");
1202 pandecode_indent++;
1203 pandecode_prop("index = %d", attr_meta->index);
1204
1205 if (attr_meta->index > max_index)
1206 max_index = attr_meta->index;
1207 pandecode_replay_swizzle(attr_meta->swizzle);
1208 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
1209
1210 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1211 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1212 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1213 pandecode_indent--;
1214 pandecode_log("},\n");
1215
1216 }
1217
1218 pandecode_indent--;
1219 pandecode_log("};\n");
1220
1221 return max_index;
1222 }
1223
1224 static void
1225 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1226 {
1227 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1228
1229 if (imem) {
1230 /* Indices are literally just a u32 array :) */
1231
1232 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1233
1234 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1235 pandecode_indent++;
1236
1237 for (unsigned i = 0; i < (index_count + 1); i += 3)
1238 pandecode_log("%d, %d, %d,\n",
1239 indices[i],
1240 indices[i + 1],
1241 indices[i + 2]);
1242
1243 pandecode_indent--;
1244 pandecode_log("};\n");
1245 }
1246 }
1247
1248 /* return bits [lo, hi) of word */
1249 static u32
1250 bits(u32 word, u32 lo, u32 hi)
1251 {
1252 if (hi - lo >= 32)
1253 return word; // avoid undefined behavior with the shift
1254
1255 return (word >> lo) & ((1 << (hi - lo)) - 1);
1256 }
1257
1258 static void
1259 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1260 {
1261 pandecode_log_cont("{\n");
1262 pandecode_indent++;
1263
1264 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1265 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1266 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1267 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1268 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1269 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1270 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1271
1272 /* Decode invocation_count. See the comment before the definition of
1273 * invocation_count for an explanation.
1274 */
1275 pandecode_msg("size: (%d, %d, %d)\n",
1276 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1277 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1278 bits(p->invocation_count, p->size_z_shift,
1279 p->workgroups_x_shift) + 1);
1280 pandecode_msg("workgroups: (%d, %d, %d)\n",
1281 bits(p->invocation_count, p->workgroups_x_shift,
1282 p->workgroups_y_shift) + 1,
1283 bits(p->invocation_count, p->workgroups_y_shift,
1284 p->workgroups_z_shift) + 1,
1285 bits(p->invocation_count, p->workgroups_z_shift,
1286 32) + 1);
1287
1288 /* TODO: Decode */
1289 if (p->unknown_draw)
1290 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1291
1292 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1293
1294 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
1295
1296 /* Index count only exists for tiler jobs anyway */
1297
1298 if (p->index_count)
1299 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1300
1301 if (p->negative_start)
1302 pandecode_prop("negative_start = %d", p->negative_start);
1303
1304 DYN_MEMORY_PROP(p, job_no, indices);
1305
1306 if (p->zero1) {
1307 pandecode_msg("Zero tripped\n");
1308 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1309 }
1310
1311 pandecode_indent--;
1312 pandecode_log("},\n");
1313 }
1314
1315 static void
1316 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1317 {
1318 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1319
1320 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1321
1322 for (int i = 0; i < ubufs_count; i++) {
1323 mali_ptr ptr = ubufs[i].ptr << 2;
1324 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1325 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1326 char name[50];
1327 snprintf(name, sizeof(name), "ubuf_%d", i);
1328 /* The blob uses ubuf 0 to upload internal stuff and
1329 * uniforms that won't fit/are accessed indirectly, so
1330 * it puts it in the batchbuffer.
1331 */
1332 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1333 pandecode_indent++;
1334
1335 for (int j = 0; j <= ubufs[i].size; j++) {
1336 for (int k = 0; k < 4; k++) {
1337 if (k == 0)
1338 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1339 else
1340 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1341
1342 }
1343
1344 pandecode_log_cont("\n");
1345 }
1346
1347 pandecode_indent--;
1348 pandecode_log("};\n");
1349 }
1350
1351 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1352 job_no);
1353 pandecode_indent++;
1354
1355 for (int i = 0; i < ubufs_count; i++) {
1356 pandecode_log("{\n");
1357 pandecode_indent++;
1358 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1359 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1360 pandecode_indent--;
1361 pandecode_log("},\n");
1362 }
1363
1364 pandecode_indent--;
1365 pandecode_log("};\n");
1366 }
1367
1368 static void
1369 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1370 {
1371
1372 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1373
1374 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1375
1376 if (scratchpad->zero)
1377 pandecode_msg("XXX scratchpad zero tripped");
1378
1379 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1380 pandecode_indent++;
1381
1382 pandecode_prop("flags = 0x%x", scratchpad->flags);
1383 MEMORY_PROP(scratchpad, gpu_scratchpad);
1384
1385 pandecode_indent--;
1386 pandecode_log("};\n");
1387 }
1388
1389 static void
1390 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1391 bool is_bifrost)
1392 {
1393 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1394 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1395
1396 /* Compute maximum possible size */
1397 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1398
1399 /* Print some boilerplate to clearly denote the assembly (which doesn't
1400 * obey indentation rules), and actually do the disassembly! */
1401
1402 printf("\n\n");
1403
1404 if (is_bifrost) {
1405 disassemble_bifrost(code, sz, false);
1406 } else {
1407 disassemble_midgard(code, sz);
1408 }
1409
1410 printf("\n\n");
1411 }
1412
1413 static void
1414 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1415 int job_no, enum mali_job_type job_type,
1416 char *suffix, bool is_bifrost)
1417 {
1418 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1419 struct pandecode_mapped_memory *attr_mem;
1420
1421 unsigned rt_count = 1;
1422
1423 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1424 * are the only things actually needed from the FBD, vertex/tiler jobs
1425 * no longer reference the FBD -- instead, this field points to some
1426 * info about the scratchpad.
1427 */
1428 if (is_bifrost)
1429 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1430 else if (p->framebuffer & MALI_MFBD)
1431 rt_count = pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1432 else if (job_type == JOB_TYPE_COMPUTE)
1433 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1434 else
1435 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1436
1437 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1438 int texture_count = 0, sampler_count = 0;
1439
1440 if (shader_meta_ptr) {
1441 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1442 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1443
1444 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1445 pandecode_indent++;
1446
1447 /* Save for dumps */
1448 attribute_count = s->attribute_count;
1449 varying_count = s->varying_count;
1450 texture_count = s->texture_count;
1451 sampler_count = s->sampler_count;
1452
1453 if (is_bifrost) {
1454 uniform_count = s->bifrost2.uniform_count;
1455 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1456 } else {
1457 uniform_count = s->midgard1.uniform_count;
1458 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1459 }
1460
1461 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1462
1463 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1464 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1465 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1466 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1467
1468 if (is_bifrost) {
1469 pandecode_log(".bifrost1 = {\n");
1470 pandecode_indent++;
1471
1472 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1473 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1474
1475 pandecode_indent--;
1476 pandecode_log("},\n");
1477 } else {
1478 pandecode_log(".midgard1 = {\n");
1479 pandecode_indent++;
1480
1481 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1482 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1483 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1484
1485 pandecode_log(".flags = ");
1486 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1487 pandecode_log_cont(",\n");
1488
1489 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1490
1491 pandecode_indent--;
1492 pandecode_log("},\n");
1493 }
1494
1495 if (s->depth_units || s->depth_factor) {
1496 if (is_bifrost)
1497 pandecode_prop("depth_units = %f", s->depth_units);
1498 else
1499 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s->depth_units - 1.0f);
1500
1501 pandecode_prop("depth_factor = %f", s->depth_factor);
1502 }
1503
1504 if (s->alpha_coverage) {
1505 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1506 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1507
1508 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1509 invert_alpha_coverage ? "~" : "",
1510 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1511 }
1512
1513 if (s->unknown2_3 || s->unknown2_4) {
1514 pandecode_log(".unknown2_3 = ");
1515
1516 int unknown2_3 = s->unknown2_3;
1517 int unknown2_4 = s->unknown2_4;
1518
1519 /* We're not quite sure what these flags mean without the depth test, if anything */
1520
1521 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1522 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1523 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1524
1525 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1526 }
1527
1528 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1529 pandecode_log_cont(",\n");
1530
1531 pandecode_log(".unknown2_4 = ");
1532 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1533 pandecode_log_cont(",\n");
1534 }
1535
1536 if (s->stencil_mask_front || s->stencil_mask_back) {
1537 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1538 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1539 }
1540
1541 pandecode_replay_stencil("front", &s->stencil_front);
1542 pandecode_replay_stencil("back", &s->stencil_back);
1543
1544 if (is_bifrost) {
1545 pandecode_log(".bifrost2 = {\n");
1546 pandecode_indent++;
1547
1548 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1549 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1550 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1551 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1552
1553 pandecode_indent--;
1554 pandecode_log("},\n");
1555 } else if (s->midgard2.unknown2_7) {
1556 pandecode_log(".midgard2 = {\n");
1557 pandecode_indent++;
1558
1559 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1560 pandecode_indent--;
1561 pandecode_log("},\n");
1562 }
1563
1564 if (s->unknown2_8)
1565 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1566
1567 if (!is_bifrost) {
1568 /* TODO: Blend shaders routing/disasm */
1569
1570 pandecode_midgard_blend(&s->blend, false);
1571 }
1572
1573 pandecode_indent--;
1574 pandecode_log("};\n");
1575
1576 /* MRT blend fields are used whenever MFBD is used, with
1577 * per-RT descriptors */
1578
1579 if (job_type == JOB_TYPE_TILER) {
1580 void* blend_base = (void *) (s + 1);
1581
1582 for (unsigned i = 0; i < rt_count; i++) {
1583 mali_ptr shader = 0;
1584
1585 if (is_bifrost)
1586 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1587 else
1588 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1589
1590 if (shader)
1591 pandecode_shader_disassemble(shader, job_no, job_type, false);
1592 }
1593 }
1594
1595 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1596 } else
1597 pandecode_msg("<no shader>\n");
1598
1599 if (p->viewport) {
1600 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1601 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1602
1603 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1604 pandecode_indent++;
1605
1606 pandecode_prop("clip_minx = %f", f->clip_minx);
1607 pandecode_prop("clip_miny = %f", f->clip_miny);
1608 pandecode_prop("clip_minz = %f", f->clip_minz);
1609 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1610 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1611 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1612
1613 /* Only the higher coordinates are MALI_POSITIVE scaled */
1614
1615 pandecode_prop("viewport0 = { %d, %d }",
1616 f->viewport0[0], f->viewport0[1]);
1617
1618 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1619 f->viewport1[0] + 1, f->viewport1[1] + 1);
1620
1621 pandecode_indent--;
1622 pandecode_log("};\n");
1623 }
1624
1625 if (p->attribute_meta) {
1626 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1627
1628 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1629 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1630 }
1631
1632 /* Varyings are encoded like attributes but not actually sent; we just
1633 * pass a zero buffer with the right stride/size set, (or whatever)
1634 * since the GPU will write to it itself */
1635
1636 if (p->varyings) {
1637 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1638
1639 /* Number of descriptors depends on whether there are
1640 * non-internal varyings */
1641
1642 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1643 }
1644
1645 if (p->varying_meta) {
1646 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1647 }
1648
1649 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1650
1651 if (p->uniforms && !is_compute) {
1652 int rows = uniform_count, width = 4;
1653 size_t sz = rows * width * sizeof(float);
1654
1655 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1656 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1657 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1658
1659 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1660
1661 pandecode_indent++;
1662
1663 for (int row = 0; row < rows; row++) {
1664 for (int i = 0; i < width; i++) {
1665 u32 v = uniforms[i];
1666 float f;
1667 memcpy(&f, &v, sizeof(v));
1668 pandecode_log_cont("%X /* %f */, ", v, f);
1669 }
1670
1671 pandecode_log_cont("\n");
1672
1673 uniforms += width;
1674 }
1675
1676 pandecode_indent--;
1677 pandecode_log("};\n");
1678 } else if (p->uniforms) {
1679 int rows = uniform_count * 2;
1680 size_t sz = rows * sizeof(mali_ptr);
1681
1682 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1683 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1684 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1685
1686 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1687
1688 pandecode_indent++;
1689
1690 for (int row = 0; row < rows; row++) {
1691 char *a = pointer_as_memory_reference(uniforms[row]);
1692 pandecode_log("%s,\n", a);
1693 free(a);
1694 }
1695
1696 pandecode_indent--;
1697 pandecode_log("};\n");
1698
1699 }
1700
1701 if (p->uniform_buffers) {
1702 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1703 }
1704
1705 if (p->texture_trampoline) {
1706 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1707
1708 if (mmem) {
1709 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1710
1711 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1712 pandecode_indent++;
1713
1714 for (int tex = 0; tex < texture_count; ++tex) {
1715 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1716 char *a = pointer_as_memory_reference(*u);
1717 pandecode_log("%s,\n", a);
1718 free(a);
1719 }
1720
1721 pandecode_indent--;
1722 pandecode_log("};\n");
1723
1724 /* Now, finally, descend down into the texture descriptor */
1725 for (int tex = 0; tex < texture_count; ++tex) {
1726 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1727 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1728
1729 if (tmem) {
1730 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1731
1732 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1733 pandecode_indent++;
1734
1735 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1736 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1737 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1738 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1739 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1740 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1741 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1742
1743 struct mali_texture_format f = t->format;
1744
1745 pandecode_log(".format = {\n");
1746 pandecode_indent++;
1747
1748 pandecode_replay_swizzle(f.swizzle);
1749 pandecode_prop("format = %s", pandecode_format_name(f.format));
1750 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1751 pandecode_prop("srgb = %" PRId32, f.srgb);
1752 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1753 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1754
1755 pandecode_indent--;
1756 pandecode_log("},\n");
1757
1758 pandecode_replay_swizzle(t->swizzle);
1759
1760 if (t->swizzle_zero) {
1761 /* Shouldn't happen */
1762 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1763 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1764 }
1765
1766 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1767
1768 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1769 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1770 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1771
1772 pandecode_log(".payload = {\n");
1773 pandecode_indent++;
1774
1775 /* A bunch of bitmap pointers follow.
1776 * We work out the correct number,
1777 * based on the mipmap/cubemap
1778 * properties, but dump extra
1779 * possibilities to futureproof */
1780
1781 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1782 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1783
1784 /* Miptree for each face */
1785 if (f.type == MALI_TEX_CUBE)
1786 bitmap_count *= 6;
1787
1788 /* Array of textures */
1789 bitmap_count *= MALI_NEGATIVE(t->array_size);
1790
1791 /* Stride for each element */
1792 if (manual_stride)
1793 bitmap_count *= 2;
1794
1795 /* Sanity check the size */
1796 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1797 assert (bitmap_count <= max_count);
1798
1799 /* Dump more to be safe, but not _that_ much more */
1800 int safe_count = MIN2(bitmap_count * 2, max_count);
1801
1802 for (int i = 0; i < safe_count; ++i) {
1803 char *prefix = (i >= bitmap_count) ? "// " : "";
1804
1805 /* How we dump depends if this is a stride or a pointer */
1806
1807 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1808 /* signed 32-bit snuck in as a 64-bit pointer */
1809 uint64_t stride_set = t->payload[i];
1810 uint32_t clamped_stride = stride_set;
1811 int32_t stride = clamped_stride;
1812 assert(stride_set == clamped_stride);
1813 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1814 } else {
1815 char *a = pointer_as_memory_reference(t->payload[i]);
1816 pandecode_log("%s%s, \n", prefix, a);
1817 free(a);
1818 }
1819 }
1820
1821 pandecode_indent--;
1822 pandecode_log("},\n");
1823
1824 pandecode_indent--;
1825 pandecode_log("};\n");
1826 }
1827 }
1828 }
1829 }
1830
1831 if (p->sampler_descriptor) {
1832 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1833
1834 if (smem) {
1835 struct mali_sampler_descriptor *s;
1836
1837 mali_ptr d = p->sampler_descriptor;
1838
1839 for (int i = 0; i < sampler_count; ++i) {
1840 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1841
1842 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1843 pandecode_indent++;
1844
1845 /* Only the lower two bits are understood right now; the rest we display as hex */
1846 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1847 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1848 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1849 s->filter_mode & ~3);
1850
1851 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1852 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1853
1854 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1855 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1856 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1857
1858 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1859
1860 if (s->zero || s->zero2) {
1861 pandecode_msg("Zero tripped\n");
1862 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1863 }
1864
1865 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1866
1867 pandecode_prop("border_color = { %f, %f, %f, %f }",
1868 s->border_color[0],
1869 s->border_color[1],
1870 s->border_color[2],
1871 s->border_color[3]);
1872
1873 pandecode_indent--;
1874 pandecode_log("};\n");
1875 }
1876 }
1877 }
1878 }
1879
1880 static void
1881 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1882 {
1883 pandecode_log_cont("{\n");
1884 pandecode_indent++;
1885
1886 MEMORY_PROP(p, position_varying);
1887 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1888 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1889 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1890 DYN_MEMORY_PROP(p, job_no, uniforms);
1891 DYN_MEMORY_PROP(p, job_no, attributes);
1892 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1893 DYN_MEMORY_PROP(p, job_no, varyings);
1894 DYN_MEMORY_PROP(p, job_no, varying_meta);
1895 DYN_MEMORY_PROP(p, job_no, viewport);
1896 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1897
1898 if (is_bifrost)
1899 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1900 else
1901 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1902
1903 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1904 pandecode_prop("flags = %d", p->flags);
1905
1906 pandecode_indent--;
1907 pandecode_log("},\n");
1908 }
1909
1910 static void
1911 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1912 {
1913 pandecode_log_cont("{\n");
1914 pandecode_indent++;
1915
1916 pandecode_prop("unk2 = 0x%x", v->unk2);
1917
1918 if (v->zero0 || v->zero1) {
1919 pandecode_msg("vertex only zero tripped");
1920 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1921 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1922 }
1923
1924 pandecode_indent--;
1925 pandecode_log("}\n");
1926 }
1927
1928 static void
1929 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1930 {
1931
1932 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1933 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1934
1935 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1936 pandecode_indent++;
1937
1938 if (h->zero) {
1939 pandecode_msg("tiler heap zero tripped\n");
1940 pandecode_prop("zero = 0x%x", h->zero);
1941 }
1942
1943 for (int i = 0; i < 12; i++) {
1944 if (h->zeros[i] != 0) {
1945 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1946 i, h->zeros[i]);
1947 }
1948 }
1949
1950 pandecode_prop("heap_size = 0x%x", h->heap_size);
1951 MEMORY_PROP(h, tiler_heap_start);
1952 MEMORY_PROP(h, tiler_heap_free);
1953
1954 /* this might point to the beginning of another buffer, when it's
1955 * really the end of the tiler heap buffer, so we have to be careful
1956 * here.
1957 */
1958 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1959 pandecode_prop("tiler_heap_end = %s + 1", a);
1960 free(a);
1961
1962 pandecode_indent--;
1963 pandecode_log("};\n");
1964 }
1965
1966 static void
1967 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1968 {
1969 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1970 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1971
1972 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1973
1974 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
1975 pandecode_indent++;
1976
1977 if (t->zero0 || t->zero1) {
1978 pandecode_msg("tiler meta zero tripped");
1979 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1980 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1981 }
1982
1983 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
1984 pandecode_prop("flags = 0x%" PRIx16, t->flags);
1985
1986 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1987 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
1988 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
1989
1990 for (int i = 0; i < 12; i++) {
1991 if (t->zeros[i] != 0) {
1992 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
1993 i, t->zeros[i]);
1994 }
1995 }
1996
1997 pandecode_indent--;
1998 pandecode_log("};\n");
1999 }
2000
2001 static void
2002 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
2003 {
2004 pandecode_log(".gl_enables = ");
2005
2006 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2007
2008 pandecode_log_cont(",\n");
2009 }
2010
2011 static void
2012 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
2013 {
2014 if (u.pointer == 0x0)
2015 return;
2016
2017 pandecode_log(".primitive_size = {\n");
2018 pandecode_indent++;
2019
2020 if (constant) {
2021 pandecode_prop("constant = %f", u.constant);
2022 } else {
2023 MEMORY_PROP((&u), pointer);
2024 }
2025
2026 pandecode_indent--;
2027 pandecode_log("},\n");
2028 }
2029
2030 static void
2031 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2032 {
2033 pandecode_log_cont("{\n");
2034 pandecode_indent++;
2035
2036 /* TODO: gl_PointSize on Bifrost */
2037 pandecode_replay_primitive_size(t->primitive_size, true);
2038
2039 DYN_MEMORY_PROP(t, job_no, tiler_meta);
2040 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2041
2042 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2043 || t->zero6 || t->zero7 || t->zero8) {
2044 pandecode_msg("tiler only zero tripped");
2045 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2046 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2047 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2048 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2049 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2050 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2051 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2052 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2053 }
2054
2055 pandecode_indent--;
2056 pandecode_log("},\n");
2057 }
2058
2059 static int
2060 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2061 const struct pandecode_mapped_memory *mem,
2062 mali_ptr payload, int job_no)
2063 {
2064 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2065
2066 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2067
2068 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2069 pandecode_indent++;
2070
2071 pandecode_log(".prefix = ");
2072 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
2073
2074 pandecode_log(".vertex = ");
2075 pandecode_replay_vertex_only_bfr(&v->vertex);
2076
2077 pandecode_log(".postfix = ");
2078 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
2079
2080 pandecode_indent--;
2081 pandecode_log("};\n");
2082
2083 return sizeof(*v);
2084 }
2085
2086 static int
2087 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2088 const struct pandecode_mapped_memory *mem,
2089 mali_ptr payload, int job_no)
2090 {
2091 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2092
2093 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2094
2095 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
2096 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
2097
2098 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2099 pandecode_indent++;
2100
2101 pandecode_log(".prefix = ");
2102 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
2103
2104 pandecode_log(".tiler = ");
2105 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
2106
2107 pandecode_log(".postfix = ");
2108 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
2109
2110 pandecode_indent--;
2111 pandecode_log("};\n");
2112
2113 return sizeof(*t);
2114 }
2115
2116 static int
2117 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2118 const struct pandecode_mapped_memory *mem,
2119 mali_ptr payload, int job_no)
2120 {
2121 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2122
2123 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2124
2125 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
2126
2127 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2128 pandecode_indent++;
2129
2130 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2131 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
2132
2133 pandecode_log(".prefix = ");
2134 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
2135
2136 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
2137
2138 if (v->instance_shift || v->instance_odd) {
2139 pandecode_prop("instance_shift = 0x%d /* %d */",
2140 v->instance_shift, 1 << v->instance_shift);
2141 pandecode_prop("instance_odd = 0x%X /* %d */",
2142 v->instance_odd, (2 * v->instance_odd) + 1);
2143
2144 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2145 }
2146
2147 if (v->draw_start)
2148 pandecode_prop("draw_start = %d", v->draw_start);
2149
2150 #ifndef __LP64__
2151
2152 if (v->zero3) {
2153 pandecode_msg("Zero tripped\n");
2154 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
2155 }
2156
2157 #endif
2158
2159 if (v->zero5) {
2160 pandecode_msg("Zero tripped\n");
2161 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2162 }
2163
2164 pandecode_log(".postfix = ");
2165 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
2166
2167 pandecode_indent--;
2168 pandecode_log("};\n");
2169
2170 return sizeof(*v);
2171 }
2172
2173 static int
2174 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
2175 mali_ptr payload, int job_no,
2176 bool is_bifrost)
2177 {
2178 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2179
2180 bool fbd_dumped = false;
2181
2182 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2183 /* Only SFBDs are understood, not MFBDs. We're speculating,
2184 * based on the versioning, kernel code, etc, that the
2185 * difference is between Single FrameBuffer Descriptor and
2186 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2187 * up with multi-framebuffer support being added (T7xx onwards,
2188 * including Gxx). In any event, there's some field shuffling
2189 * that we haven't looked into yet. */
2190
2191 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
2192 fbd_dumped = true;
2193 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2194 /* We don't know if Bifrost supports SFBD's at all, since the
2195 * driver never uses them. And the format is different from
2196 * Midgard anyways, due to the tiler heap and scratchpad being
2197 * moved out into separate structures, so it's not clear what a
2198 * Bifrost SFBD would even look like without getting an actual
2199 * trace, which appears impossible.
2200 */
2201
2202 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2203 fbd_dumped = true;
2204 }
2205
2206 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2207 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2208 pandecode_indent++;
2209
2210 /* See the comments by the macro definitions for mathematical context
2211 * on why this is so weird */
2212
2213 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2214 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2215
2216 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2217 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2218 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2219
2220 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2221 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2222 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2223
2224 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2225 * we have to fallback on offsets. */
2226
2227 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2228
2229 if (fbd_dumped)
2230 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2231 else
2232 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2233
2234 pandecode_indent--;
2235 pandecode_log("};\n");
2236
2237 return sizeof(*s);
2238 }
2239
2240 static int job_descriptor_number = 0;
2241
2242 int
2243 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
2244 {
2245 struct mali_job_descriptor_header *h;
2246
2247 int start_number = 0;
2248
2249 bool first = true;
2250 bool last_size;
2251
2252 do {
2253 struct pandecode_mapped_memory *mem =
2254 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2255
2256 void *payload;
2257
2258 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2259
2260 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2261 * high 32-bits of the 64-bit pointer are reused to store
2262 * something else.
2263 */
2264 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2265 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2266 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2267
2268 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2269 MALI_PAYLOAD_SIZE);
2270
2271 int job_no = job_descriptor_number++;
2272
2273 if (first)
2274 start_number = job_no;
2275
2276 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2277 pandecode_indent++;
2278
2279 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
2280
2281 /* Save for next job fixing */
2282 last_size = h->job_descriptor_size;
2283
2284 if (h->job_descriptor_size)
2285 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2286
2287 if (h->exception_status != 0x1)
2288 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2289 h->exception_status,
2290 (h->exception_status >> 16) & 0xFFFF,
2291 (h->exception_status >> 8) & 0x3,
2292 h->exception_status & 0xFF);
2293
2294 if (h->first_incomplete_task)
2295 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2296
2297 if (h->fault_pointer)
2298 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2299
2300 if (h->job_barrier)
2301 pandecode_prop("job_barrier = %d", h->job_barrier);
2302
2303 pandecode_prop("job_index = %d", h->job_index);
2304
2305 if (h->unknown_flags)
2306 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2307
2308 if (h->job_dependency_index_1)
2309 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2310
2311 if (h->job_dependency_index_2)
2312 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2313
2314 pandecode_indent--;
2315 pandecode_log("};\n");
2316
2317 /* Do not touch the field yet -- decode the payload first, and
2318 * don't touch that either. This is essential for the uploads
2319 * to occur in sequence and therefore be dynamically allocated
2320 * correctly. Do note the size, however, for that related
2321 * reason. */
2322
2323 switch (h->job_type) {
2324 case JOB_TYPE_SET_VALUE: {
2325 struct mali_payload_set_value *s = payload;
2326 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2327 pandecode_indent++;
2328 MEMORY_PROP(s, out);
2329 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2330 pandecode_indent--;
2331 pandecode_log("};\n");
2332
2333 break;
2334 }
2335
2336 case JOB_TYPE_TILER:
2337 case JOB_TYPE_VERTEX:
2338 case JOB_TYPE_COMPUTE:
2339 if (bifrost) {
2340 if (h->job_type == JOB_TYPE_TILER)
2341 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2342 else
2343 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2344 } else
2345 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2346
2347 break;
2348
2349 case JOB_TYPE_FRAGMENT:
2350 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2351 break;
2352
2353 default:
2354 break;
2355 }
2356
2357 /* Handle linkage */
2358
2359 if (!first) {
2360 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2361
2362 if (last_size)
2363 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2364 else
2365 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2366 }
2367
2368 first = false;
2369
2370 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2371
2372 return start_number;
2373 }