pan/decode: Depth units/factor are identical to GL
[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_replay_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 DYN_MEMORY_PROP(obj, no, p) { \
50 if (obj->p) \
51 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
52 }
53
54 /* Semantic logging type.
55 *
56 * Raw: for raw messages to be printed as is.
57 * Message: for helpful information to be commented out in replays.
58 * Property: for properties of a struct
59 *
60 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
61 */
62
63 enum pandecode_log_type {
64 PANDECODE_RAW,
65 PANDECODE_MESSAGE,
66 PANDECODE_PROPERTY
67 };
68
69 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
70 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
71 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
72
73 unsigned pandecode_indent = 0;
74
75 static void
76 pandecode_make_indent(void)
77 {
78 for (unsigned i = 0; i < pandecode_indent; ++i)
79 printf(" ");
80 }
81
82 static void
83 pandecode_log_typed(enum pandecode_log_type type, const char *format, ...)
84 {
85 va_list ap;
86
87 pandecode_make_indent();
88
89 if (type == PANDECODE_MESSAGE)
90 printf("// ");
91 else if (type == PANDECODE_PROPERTY)
92 printf(".");
93
94 va_start(ap, format);
95 vprintf(format, ap);
96 va_end(ap);
97
98 if (type == PANDECODE_PROPERTY)
99 printf(",\n");
100 }
101
102 static void
103 pandecode_log_cont(const char *format, ...)
104 {
105 va_list ap;
106
107 va_start(ap, format);
108 vprintf(format, ap);
109 va_end(ap);
110 }
111
112 struct pandecode_flag_info {
113 u64 flag;
114 const char *name;
115 };
116
117 static void
118 pandecode_log_decoded_flags(const struct pandecode_flag_info *flag_info,
119 u64 flags)
120 {
121 bool decodable_flags_found = false;
122
123 for (int i = 0; flag_info[i].name; i++) {
124 if ((flags & flag_info[i].flag) != flag_info[i].flag)
125 continue;
126
127 if (!decodable_flags_found) {
128 decodable_flags_found = true;
129 } else {
130 pandecode_log_cont(" | ");
131 }
132
133 pandecode_log_cont("%s", flag_info[i].name);
134
135 flags &= ~flag_info[i].flag;
136 }
137
138 if (decodable_flags_found) {
139 if (flags)
140 pandecode_log_cont(" | 0x%" PRIx64, flags);
141 } else {
142 pandecode_log_cont("0x%" PRIx64, flags);
143 }
144 }
145
146 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
147 static const struct pandecode_flag_info gl_enable_flag_info[] = {
148 FLAG_INFO(OCCLUSION_QUERY),
149 FLAG_INFO(OCCLUSION_PRECISE),
150 FLAG_INFO(FRONT_CCW_TOP),
151 FLAG_INFO(CULL_FACE_FRONT),
152 FLAG_INFO(CULL_FACE_BACK),
153 {}
154 };
155 #undef FLAG_INFO
156
157 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
158 static const struct pandecode_flag_info clear_flag_info[] = {
159 FLAG_INFO(FAST),
160 FLAG_INFO(SLOW),
161 FLAG_INFO(SLOW_STENCIL),
162 {}
163 };
164 #undef FLAG_INFO
165
166 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
167 static const struct pandecode_flag_info mask_flag_info[] = {
168 FLAG_INFO(R),
169 FLAG_INFO(G),
170 FLAG_INFO(B),
171 FLAG_INFO(A),
172 {}
173 };
174 #undef FLAG_INFO
175
176 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
177 static const struct pandecode_flag_info u3_flag_info[] = {
178 FLAG_INFO(HAS_MSAA),
179 FLAG_INFO(CAN_DISCARD),
180 FLAG_INFO(HAS_BLEND_SHADER),
181 FLAG_INFO(DEPTH_TEST),
182 {}
183 };
184
185 static const struct pandecode_flag_info u4_flag_info[] = {
186 FLAG_INFO(NO_MSAA),
187 FLAG_INFO(NO_DITHER),
188 FLAG_INFO(DEPTH_RANGE_A),
189 FLAG_INFO(DEPTH_RANGE_B),
190 FLAG_INFO(STENCIL_TEST),
191 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER),
192 {}
193 };
194 #undef FLAG_INFO
195
196 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
197 static const struct pandecode_flag_info fb_fmt_flag_info[] = {
198 FLAG_INFO(MSAA_A),
199 FLAG_INFO(MSAA_B),
200 FLAG_INFO(MSAA_8),
201 {}
202 };
203 #undef FLAG_INFO
204
205 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
206 static const struct pandecode_flag_info mfbd_fmt_flag_info[] = {
207 FLAG_INFO(MSAA),
208 FLAG_INFO(SRGB),
209 {}
210 };
211 #undef FLAG_INFO
212
213 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
214 static const struct pandecode_flag_info mfbd_extra_flag_info[] = {
215 FLAG_INFO(PRESENT),
216 FLAG_INFO(AFBC),
217 FLAG_INFO(ZS),
218 {}
219 };
220 #undef FLAG_INFO
221
222 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
223 static const struct pandecode_flag_info shader_midgard1_flag_info [] = {
224 FLAG_INFO(EARLY_Z),
225 FLAG_INFO(HELPER_INVOCATIONS),
226 FLAG_INFO(READS_TILEBUFFER),
227 FLAG_INFO(READS_ZS),
228 {}
229 };
230 #undef FLAG_INFO
231
232 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
233 static const struct pandecode_flag_info mfbd_flag_info [] = {
234 FLAG_INFO(DEPTH_WRITE),
235 FLAG_INFO(EXTRA),
236 {}
237 };
238 #undef FLAG_INFO
239
240
241 extern char *replace_fragment;
242 extern char *replace_vertex;
243
244 static char *
245 pandecode_job_type_name(enum mali_job_type type)
246 {
247 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
248
249 switch (type) {
250 DEFINE_CASE(NULL);
251 DEFINE_CASE(SET_VALUE);
252 DEFINE_CASE(CACHE_FLUSH);
253 DEFINE_CASE(COMPUTE);
254 DEFINE_CASE(VERTEX);
255 DEFINE_CASE(TILER);
256 DEFINE_CASE(FUSED);
257 DEFINE_CASE(FRAGMENT);
258
259 case JOB_NOT_STARTED:
260 return "NOT_STARTED";
261
262 default:
263 pandecode_log("Warning! Unknown job type %x\n", type);
264 return "!?!?!?";
265 }
266
267 #undef DEFINE_CASE
268 }
269
270 static char *
271 pandecode_draw_mode_name(enum mali_draw_mode mode)
272 {
273 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
274
275 switch (mode) {
276 DEFINE_CASE(DRAW_NONE);
277 DEFINE_CASE(POINTS);
278 DEFINE_CASE(LINES);
279 DEFINE_CASE(TRIANGLES);
280 DEFINE_CASE(TRIANGLE_STRIP);
281 DEFINE_CASE(TRIANGLE_FAN);
282 DEFINE_CASE(LINE_STRIP);
283 DEFINE_CASE(LINE_LOOP);
284 DEFINE_CASE(POLYGON);
285 DEFINE_CASE(QUADS);
286 DEFINE_CASE(QUAD_STRIP);
287
288 default:
289 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
290 }
291
292 #undef DEFINE_CASE
293 }
294
295 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
296 static char *
297 pandecode_func_name(enum mali_func mode)
298 {
299 switch (mode) {
300 DEFINE_CASE(NEVER);
301 DEFINE_CASE(LESS);
302 DEFINE_CASE(EQUAL);
303 DEFINE_CASE(LEQUAL);
304 DEFINE_CASE(GREATER);
305 DEFINE_CASE(NOTEQUAL);
306 DEFINE_CASE(GEQUAL);
307 DEFINE_CASE(ALWAYS);
308
309 default:
310 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
311 }
312 }
313 #undef DEFINE_CASE
314
315 /* Why is this duplicated? Who knows... */
316 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
317 static char *
318 pandecode_alt_func_name(enum mali_alt_func mode)
319 {
320 switch (mode) {
321 DEFINE_CASE(NEVER);
322 DEFINE_CASE(LESS);
323 DEFINE_CASE(EQUAL);
324 DEFINE_CASE(LEQUAL);
325 DEFINE_CASE(GREATER);
326 DEFINE_CASE(NOTEQUAL);
327 DEFINE_CASE(GEQUAL);
328 DEFINE_CASE(ALWAYS);
329
330 default:
331 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
332 }
333 }
334 #undef DEFINE_CASE
335
336 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
337 static char *
338 pandecode_stencil_op_name(enum mali_stencil_op op)
339 {
340 switch (op) {
341 DEFINE_CASE(KEEP);
342 DEFINE_CASE(REPLACE);
343 DEFINE_CASE(ZERO);
344 DEFINE_CASE(INVERT);
345 DEFINE_CASE(INCR_WRAP);
346 DEFINE_CASE(DECR_WRAP);
347 DEFINE_CASE(INCR);
348 DEFINE_CASE(DECR);
349
350 default:
351 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
352 }
353 }
354
355 #undef DEFINE_CASE
356
357 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
358 static char *pandecode_attr_mode_name(enum mali_attr_mode mode)
359 {
360 switch(mode) {
361 DEFINE_CASE(UNUSED);
362 DEFINE_CASE(LINEAR);
363 DEFINE_CASE(POT_DIVIDE);
364 DEFINE_CASE(MODULO);
365 DEFINE_CASE(NPOT_DIVIDE);
366 default:
367 return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
368 }
369 }
370
371 #undef DEFINE_CASE
372
373 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
374 static char *
375 pandecode_channel_name(enum mali_channel channel)
376 {
377 switch (channel) {
378 DEFINE_CASE(RED);
379 DEFINE_CASE(GREEN);
380 DEFINE_CASE(BLUE);
381 DEFINE_CASE(ALPHA);
382 DEFINE_CASE(ZERO);
383 DEFINE_CASE(ONE);
384 DEFINE_CASE(RESERVED_0);
385 DEFINE_CASE(RESERVED_1);
386
387 default:
388 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
389 }
390 }
391 #undef DEFINE_CASE
392
393 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
394 static char *
395 pandecode_wrap_mode_name(enum mali_wrap_mode op)
396 {
397 switch (op) {
398 DEFINE_CASE(REPEAT);
399 DEFINE_CASE(CLAMP_TO_EDGE);
400 DEFINE_CASE(CLAMP_TO_BORDER);
401 DEFINE_CASE(MIRRORED_REPEAT);
402
403 default:
404 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
405 }
406 }
407 #undef DEFINE_CASE
408
409 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
410 static char *
411 pandecode_texture_type(enum mali_texture_type type)
412 {
413 switch (type) {
414 DEFINE_CASE(1D);
415 DEFINE_CASE(2D);
416 DEFINE_CASE(3D);
417 DEFINE_CASE(CUBE);
418
419 default:
420 unreachable("Unknown case");
421 }
422 }
423 #undef DEFINE_CASE
424
425 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
426 static char *
427 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt)
428 {
429 switch (fmt) {
430 DEFINE_CASE(TILED);
431 DEFINE_CASE(UNKNOWN);
432 DEFINE_CASE(LINEAR);
433 DEFINE_CASE(AFBC);
434
435 default:
436 unreachable("Invalid case");
437 }
438 }
439 #undef DEFINE_CASE
440
441 static inline char *
442 pandecode_decode_fbd_type(enum mali_fbd_type type)
443 {
444 if (type == MALI_SFBD) return "SFBD";
445 else if (type == MALI_MFBD) return "MFBD";
446 else return "WATFBD /* XXX */";
447 }
448
449 /* Midgard's tiler descriptor is embedded within the
450 * larger FBD */
451
452 static void
453 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor *t)
454 {
455 pandecode_log(".tiler = {\n");
456 pandecode_indent++;
457
458 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
459 pandecode_prop("flags = 0x%" PRIx16, t->flags);
460 pandecode_prop("polygon_list_size = 0x%x", t->polygon_list_size);
461
462 MEMORY_PROP(t, polygon_list);
463 MEMORY_PROP(t, polygon_list_body);
464
465 MEMORY_PROP(t, heap_start);
466
467 {
468 /* Points to the end of a buffer */
469 char *a = pointer_as_memory_reference(t->heap_end - 1);
470 pandecode_prop("heap_end = %s + 1", a);
471 free(a);
472 }
473
474 bool nonzero_weights = false;
475
476 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
477 nonzero_weights |= t->weights[w] != 0x0;
478 }
479
480 if (nonzero_weights) {
481 pandecode_log(".weights = {");
482
483 for (unsigned w = 0; w < ARRAY_SIZE(t->weights); ++w) {
484 pandecode_log("%d, ", t->weights[w]);
485 }
486
487 pandecode_log("},");
488 }
489
490 pandecode_indent--;
491 pandecode_log("}\n");
492 }
493
494 static void
495 pandecode_replay_sfbd(uint64_t gpu_va, int job_no)
496 {
497 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
498 const struct mali_single_framebuffer *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
499
500 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
501 pandecode_indent++;
502
503 pandecode_prop("unknown1 = 0x%" PRIx32, s->unknown1);
504 pandecode_prop("unknown2 = 0x%" PRIx32, s->unknown2);
505
506 pandecode_log(".format = ");
507 pandecode_log_decoded_flags(fb_fmt_flag_info, s->format);
508 pandecode_log_cont(",\n");
509
510 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", s->width + 1);
511 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", s->height + 1);
512
513 MEMORY_PROP(s, framebuffer);
514 pandecode_prop("stride = %d", s->stride);
515
516 /* Earlier in the actual commandstream -- right before width -- but we
517 * delay to flow nicer */
518
519 pandecode_log(".clear_flags = ");
520 pandecode_log_decoded_flags(clear_flag_info, s->clear_flags);
521 pandecode_log_cont(",\n");
522
523 if (s->depth_buffer | s->depth_buffer_enable) {
524 MEMORY_PROP(s, depth_buffer);
525 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s->depth_buffer_enable));
526 }
527
528 if (s->stencil_buffer | s->stencil_buffer_enable) {
529 MEMORY_PROP(s, stencil_buffer);
530 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s->stencil_buffer_enable));
531 }
532
533 if (s->clear_color_1 | s->clear_color_2 | s->clear_color_3 | s->clear_color_4) {
534 pandecode_prop("clear_color_1 = 0x%" PRIx32, s->clear_color_1);
535 pandecode_prop("clear_color_2 = 0x%" PRIx32, s->clear_color_2);
536 pandecode_prop("clear_color_3 = 0x%" PRIx32, s->clear_color_3);
537 pandecode_prop("clear_color_4 = 0x%" PRIx32, s->clear_color_4);
538 }
539
540 if (s->clear_depth_1 != 0 || s->clear_depth_2 != 0 || s->clear_depth_3 != 0 || s->clear_depth_4 != 0) {
541 pandecode_prop("clear_depth_1 = %f", s->clear_depth_1);
542 pandecode_prop("clear_depth_2 = %f", s->clear_depth_2);
543 pandecode_prop("clear_depth_3 = %f", s->clear_depth_3);
544 pandecode_prop("clear_depth_4 = %f", s->clear_depth_4);
545 }
546
547 if (s->clear_stencil) {
548 pandecode_prop("clear_stencil = 0x%x", s->clear_stencil);
549 }
550
551 MEMORY_PROP(s, unknown_address_0);
552 pandecode_midgard_tiler_descriptor(&s->tiler);
553
554 pandecode_indent--;
555 pandecode_log("};\n");
556
557 pandecode_prop("zero0 = 0x%" PRIx64, s->zero0);
558 pandecode_prop("zero1 = 0x%" PRIx64, s->zero1);
559 pandecode_prop("zero2 = 0x%" PRIx32, s->zero2);
560 pandecode_prop("zero4 = 0x%" PRIx32, s->zero4);
561
562 printf(".zero3 = {");
563
564 for (int i = 0; i < sizeof(s->zero3) / sizeof(s->zero3[0]); ++i)
565 printf("%X, ", s->zero3[i]);
566
567 printf("},\n");
568
569 printf(".zero6 = {");
570
571 for (int i = 0; i < sizeof(s->zero6) / sizeof(s->zero6[0]); ++i)
572 printf("%X, ", s->zero6[i]);
573
574 printf("},\n");
575 }
576
577 static void
578 pandecode_u32_slide(unsigned name, const u32 *slide, unsigned count)
579 {
580 pandecode_log(".unknown%d = {", name);
581
582 for (int i = 0; i < count; ++i)
583 printf("%X, ", slide[i]);
584
585 pandecode_log("},\n");
586 }
587
588 #define SHORT_SLIDE(num) \
589 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
590
591 static void
592 pandecode_compute_fbd(uint64_t gpu_va, int job_no)
593 {
594 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
595 const struct mali_compute_fbd *PANDECODE_PTR_VAR(s, mem, (mali_ptr) gpu_va);
596
597 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
598 pandecode_indent++;
599
600 SHORT_SLIDE(1);
601
602 pandecode_indent--;
603 printf("},\n");
604 }
605
606 static void
607 pandecode_replay_swizzle(unsigned swizzle)
608 {
609 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
610 pandecode_channel_name((swizzle >> 0) & 0x7),
611 pandecode_channel_name((swizzle >> 3) & 0x7),
612 pandecode_channel_name((swizzle >> 6) & 0x7),
613 pandecode_channel_name((swizzle >> 9) & 0x7));
614 }
615
616 static void
617 pandecode_rt_format(struct mali_rt_format format)
618 {
619 pandecode_log(".format = {\n");
620 pandecode_indent++;
621
622 pandecode_prop("unk1 = 0x%" PRIx32, format.unk1);
623 pandecode_prop("unk2 = 0x%" PRIx32, format.unk2);
624 pandecode_prop("unk3 = 0x%" PRIx32, format.unk3);
625
626 pandecode_prop("block = %s",
627 pandecode_mfbd_block_format(format.block));
628
629 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
630 MALI_NEGATIVE(format.nr_channels));
631
632 pandecode_log(".flags = ");
633 pandecode_log_decoded_flags(mfbd_fmt_flag_info, format.flags);
634 pandecode_log_cont(",\n");
635
636 pandecode_replay_swizzle(format.swizzle);
637
638 pandecode_prop("unk4 = 0x%" PRIx32, format.unk4);
639
640 pandecode_indent--;
641 pandecode_log("},\n");
642 }
643
644 static void
645 pandecode_render_target(uint64_t gpu_va, unsigned job_no, const struct bifrost_framebuffer *fb)
646 {
647 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64"_%d[] = {\n", gpu_va, job_no);
648 pandecode_indent++;
649
650 for (int i = 0; i < MALI_NEGATIVE(fb->rt_count_1); i++) {
651 mali_ptr rt_va = gpu_va + i * sizeof(struct bifrost_render_target);
652 struct pandecode_mapped_memory *mem =
653 pandecode_find_mapped_gpu_mem_containing(rt_va);
654 const struct bifrost_render_target *PANDECODE_PTR_VAR(rt, mem, (mali_ptr) rt_va);
655
656 pandecode_log("{\n");
657 pandecode_indent++;
658
659 pandecode_rt_format(rt->format);
660
661 if (rt->format.block == MALI_MFBD_BLOCK_AFBC) {
662 pandecode_log(".afbc = {\n");
663 pandecode_indent++;
664
665 char *a = pointer_as_memory_reference(rt->afbc.metadata);
666 pandecode_prop("metadata = %s", a);
667 free(a);
668
669 pandecode_prop("stride = %d", rt->afbc.stride);
670 pandecode_prop("unk = 0x%" PRIx32, rt->afbc.unk);
671
672 pandecode_indent--;
673 pandecode_log("},\n");
674 } else {
675 pandecode_log(".chunknown = {\n");
676 pandecode_indent++;
677
678 pandecode_prop("unk = 0x%" PRIx64, rt->chunknown.unk);
679
680 char *a = pointer_as_memory_reference(rt->chunknown.pointer);
681 pandecode_prop("pointer = %s", a);
682 free(a);
683
684 pandecode_indent--;
685 pandecode_log("},\n");
686 }
687
688 MEMORY_PROP(rt, framebuffer);
689 pandecode_prop("framebuffer_stride = %d", rt->framebuffer_stride);
690
691 if (rt->clear_color_1 | rt->clear_color_2 | rt->clear_color_3 | rt->clear_color_4) {
692 pandecode_prop("clear_color_1 = 0x%" PRIx32, rt->clear_color_1);
693 pandecode_prop("clear_color_2 = 0x%" PRIx32, rt->clear_color_2);
694 pandecode_prop("clear_color_3 = 0x%" PRIx32, rt->clear_color_3);
695 pandecode_prop("clear_color_4 = 0x%" PRIx32, rt->clear_color_4);
696 }
697
698 if (rt->zero1 || rt->zero2 || rt->zero3) {
699 pandecode_msg("render target zeros tripped\n");
700 pandecode_prop("zero1 = 0x%" PRIx64, rt->zero1);
701 pandecode_prop("zero2 = 0x%" PRIx32, rt->zero2);
702 pandecode_prop("zero3 = 0x%" PRIx32, rt->zero3);
703 }
704
705 pandecode_indent--;
706 pandecode_log("},\n");
707 }
708
709 pandecode_indent--;
710 pandecode_log("};\n");
711 }
712
713 static unsigned
714 pandecode_replay_mfbd_bfr(uint64_t gpu_va, int job_no, bool with_render_targets)
715 {
716 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
717 const struct bifrost_framebuffer *PANDECODE_PTR_VAR(fb, mem, (mali_ptr) gpu_va);
718
719 if (fb->sample_locations) {
720 /* The blob stores all possible sample locations in a single buffer
721 * allocated on startup, and just switches the pointer when switching
722 * MSAA state. For now, we just put the data into the cmdstream, but we
723 * should do something like what the blob does with a real driver.
724 *
725 * There seem to be 32 slots for sample locations, followed by another
726 * 16. The second 16 is just the center location followed by 15 zeros
727 * in all the cases I've identified (maybe shader vs. depth/color
728 * samples?).
729 */
730
731 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(fb->sample_locations);
732
733 const u16 *PANDECODE_PTR_VAR(samples, smem, fb->sample_locations);
734
735 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no);
736 pandecode_indent++;
737
738 for (int i = 0; i < 32 + 16; i++) {
739 pandecode_log("%d, %d,\n", samples[2 * i], samples[2 * i + 1]);
740 }
741
742 pandecode_indent--;
743 pandecode_log("};\n");
744 }
745
746 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64"_%d = {\n", gpu_va, job_no);
747 pandecode_indent++;
748
749 pandecode_prop("unk0 = 0x%x", fb->unk0);
750
751 if (fb->sample_locations)
752 pandecode_prop("sample_locations = sample_locations_%d", job_no);
753
754 /* Assume that unknown1 was emitted in the last job for
755 * now */
756 MEMORY_PROP(fb, unknown1);
757
758 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb->width1 + 1);
759 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb->height1 + 1);
760 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb->width2 + 1);
761 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb->height2 + 1);
762
763 pandecode_prop("unk1 = 0x%x", fb->unk1);
764 pandecode_prop("unk2 = 0x%x", fb->unk2);
765 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb->rt_count_1 + 1);
766 pandecode_prop("rt_count_2 = %d", fb->rt_count_2);
767
768 pandecode_log(".mfbd_flags = ");
769 pandecode_log_decoded_flags(mfbd_flag_info, fb->mfbd_flags);
770 pandecode_log_cont(",\n");
771
772 pandecode_prop("clear_stencil = 0x%x", fb->clear_stencil);
773 pandecode_prop("clear_depth = %f", fb->clear_depth);
774
775 pandecode_prop("unknown2 = 0x%x", fb->unknown2);
776 MEMORY_PROP(fb, scratchpad);
777 pandecode_midgard_tiler_descriptor(&fb->tiler);
778
779 if (fb->zero3 || fb->zero4) {
780 pandecode_msg("framebuffer zeros tripped\n");
781 pandecode_prop("zero3 = 0x%" PRIx32, fb->zero3);
782 pandecode_prop("zero4 = 0x%" PRIx32, fb->zero4);
783 }
784
785 pandecode_indent--;
786 pandecode_log("};\n");
787
788 gpu_va += sizeof(struct bifrost_framebuffer);
789
790 if ((fb->mfbd_flags & MALI_MFBD_EXTRA) && with_render_targets) {
791 mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
792 const struct bifrost_fb_extra *PANDECODE_PTR_VAR(fbx, mem, (mali_ptr) gpu_va);
793
794 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64"_%d = {\n", gpu_va, job_no);
795 pandecode_indent++;
796
797 MEMORY_PROP(fbx, checksum);
798
799 if (fbx->checksum_stride)
800 pandecode_prop("checksum_stride = %d", fbx->checksum_stride);
801
802 pandecode_log(".flags = ");
803 pandecode_log_decoded_flags(mfbd_extra_flag_info, fbx->flags);
804 pandecode_log_cont(",\n");
805
806 if (fbx->flags & MALI_EXTRA_AFBC_ZS) {
807 pandecode_log(".ds_afbc = {\n");
808 pandecode_indent++;
809
810 MEMORY_PROP((&fbx->ds_afbc), depth_stencil_afbc_metadata);
811 pandecode_prop("depth_stencil_afbc_stride = %d",
812 fbx->ds_afbc.depth_stencil_afbc_stride);
813 MEMORY_PROP((&fbx->ds_afbc), depth_stencil);
814
815 if (fbx->ds_afbc.zero1 || fbx->ds_afbc.padding) {
816 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
817 pandecode_prop("zero1 = 0x%" PRIx32,
818 fbx->ds_afbc.zero1);
819 pandecode_prop("padding = 0x%" PRIx64,
820 fbx->ds_afbc.padding);
821 }
822
823 pandecode_indent--;
824 pandecode_log("},\n");
825 } else {
826 pandecode_log(".ds_linear = {\n");
827 pandecode_indent++;
828
829 if (fbx->ds_linear.depth) {
830 MEMORY_PROP((&fbx->ds_linear), depth);
831 pandecode_prop("depth_stride = %d",
832 fbx->ds_linear.depth_stride);
833 }
834
835 if (fbx->ds_linear.stencil) {
836 MEMORY_PROP((&fbx->ds_linear), stencil);
837 pandecode_prop("stencil_stride = %d",
838 fbx->ds_linear.stencil_stride);
839 }
840
841 if (fbx->ds_linear.depth_stride_zero ||
842 fbx->ds_linear.stencil_stride_zero ||
843 fbx->ds_linear.zero1 || fbx->ds_linear.zero2) {
844 pandecode_msg("Depth/stencil zeros tripped\n");
845 pandecode_prop("depth_stride_zero = 0x%x",
846 fbx->ds_linear.depth_stride_zero);
847 pandecode_prop("stencil_stride_zero = 0x%x",
848 fbx->ds_linear.stencil_stride_zero);
849 pandecode_prop("zero1 = 0x%" PRIx32,
850 fbx->ds_linear.zero1);
851 pandecode_prop("zero2 = 0x%" PRIx32,
852 fbx->ds_linear.zero2);
853 }
854
855 pandecode_indent--;
856 pandecode_log("},\n");
857 }
858
859 if (fbx->zero3 || fbx->zero4) {
860 pandecode_msg("fb_extra zeros tripped\n");
861 pandecode_prop("zero3 = 0x%" PRIx64, fbx->zero3);
862 pandecode_prop("zero4 = 0x%" PRIx64, fbx->zero4);
863 }
864
865 pandecode_indent--;
866 pandecode_log("};\n");
867
868 gpu_va += sizeof(struct bifrost_fb_extra);
869 }
870
871 if (with_render_targets)
872 pandecode_render_target(gpu_va, job_no, fb);
873
874 /* Passback the render target count */
875 return MALI_NEGATIVE(fb->rt_count_1);
876 }
877
878 /* Just add a comment decoding the shift/odd fields forming the padded vertices
879 * count */
880
881 static void
882 pandecode_padded_vertices(unsigned shift, unsigned k)
883 {
884 unsigned odd = 2*k + 1;
885 unsigned pot = 1 << shift;
886 pandecode_msg("padded_num_vertices = %d\n", odd * pot);
887 }
888
889 /* Given a magic divisor, recover what we were trying to divide by.
890 *
891 * Let m represent the magic divisor. By definition, m is an element on Z, whre
892 * 0 <= m < 2^N, for N bits in m.
893 *
894 * Let q represent the number we would like to divide by.
895 *
896 * By definition of a magic divisor for N-bit unsigned integers (a number you
897 * multiply by to magically get division), m is a number such that:
898 *
899 * (m * x) & (2^N - 1) = floor(x/q).
900 * for all x on Z where 0 <= x < 2^N
901 *
902 * Ignore the case where any of the above values equals zero; it is irrelevant
903 * for our purposes (instanced arrays).
904 *
905 * Choose x = q. Then:
906 *
907 * (m * x) & (2^N - 1) = floor(x/q).
908 * (m * q) & (2^N - 1) = floor(q/q).
909 *
910 * floor(q/q) = floor(1) = 1, therefore:
911 *
912 * (m * q) & (2^N - 1) = 1
913 *
914 * Recall the identity that the bitwise AND of one less than a power-of-two
915 * equals the modulo with that power of two, i.e. for all x:
916 *
917 * x & (2^N - 1) = x % N
918 *
919 * Therefore:
920 *
921 * mq % (2^N) = 1
922 *
923 * By definition, a modular multiplicative inverse of a number m is the number
924 * q such that with respect to a modulos M:
925 *
926 * mq % M = 1
927 *
928 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
929 *
930 */
931
932 static void
933 pandecode_magic_divisor(uint32_t magic, unsigned shift, unsigned orig_divisor, unsigned extra)
934 {
935 /* Compute the modular inverse of `magic` with respect to 2^(32 -
936 * shift) the most lame way possible... just repeatedly add.
937 * Asymptoptically slow but nobody cares in practice, unless you have
938 * massive numbers of vertices or high divisors. */
939
940 unsigned inverse = 0;
941
942 /* Magic implicitly has the highest bit set */
943 magic |= (1 << 31);
944
945 /* Depending on rounding direction */
946 if (extra)
947 magic++;
948
949 for (;;) {
950 uint32_t product = magic * inverse;
951
952 if (shift) {
953 product >>= shift;
954 }
955
956 if (product == 1)
957 break;
958
959 ++inverse;
960 }
961
962 pandecode_msg("dividing by %d (maybe off by two)\n", inverse);
963
964 /* Recall we're supposed to divide by (gl_level_divisor *
965 * padded_num_vertices) */
966
967 unsigned padded_num_vertices = inverse / orig_divisor;
968
969 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices);
970 }
971
972 static void
973 pandecode_replay_attributes(const struct pandecode_mapped_memory *mem,
974 mali_ptr addr, int job_no, char *suffix,
975 int count, bool varying)
976 {
977 char *prefix = varying ? "varyings" : "attributes";
978
979 union mali_attr *attr = pandecode_fetch_gpu_mem(mem, addr, sizeof(union mali_attr) * count);
980
981 char base[128];
982 snprintf(base, sizeof(base), "%s_data_%d%s", prefix, job_no, suffix);
983
984 for (int i = 0; i < count; ++i) {
985 enum mali_attr_mode mode = attr[i].elements & 7;
986
987 if (mode == MALI_ATTR_UNUSED)
988 continue;
989
990 mali_ptr raw_elements = attr[i].elements & ~7;
991
992 /* TODO: Do we maybe want to dump the attribute values
993 * themselves given the specified format? Or is that too hard?
994 * */
995
996 char *a = pointer_as_memory_reference(raw_elements);
997 pandecode_log("mali_ptr %s_%d_p = %s;\n", base, i, a);
998 free(a);
999 }
1000
1001 pandecode_log("union mali_attr %s_%d[] = {\n", prefix, job_no);
1002 pandecode_indent++;
1003
1004 for (int i = 0; i < count; ++i) {
1005 pandecode_log("{\n");
1006 pandecode_indent++;
1007
1008 unsigned mode = attr[i].elements & 7;
1009 pandecode_prop("elements = (%s_%d_p) | %s", base, i, pandecode_attr_mode_name(mode));
1010 pandecode_prop("shift = %d", attr[i].shift);
1011 pandecode_prop("extra_flags = %d", attr[i].extra_flags);
1012 pandecode_prop("stride = 0x%" PRIx32, attr[i].stride);
1013 pandecode_prop("size = 0x%" PRIx32, attr[i].size);
1014
1015 /* Decode further where possible */
1016
1017 if (mode == MALI_ATTR_MODULO) {
1018 pandecode_padded_vertices(
1019 attr[i].shift,
1020 attr[i].extra_flags);
1021 }
1022
1023 pandecode_indent--;
1024 pandecode_log("}, \n");
1025
1026 if (mode == MALI_ATTR_NPOT_DIVIDE) {
1027 i++;
1028 pandecode_log("{\n");
1029 pandecode_indent++;
1030 pandecode_prop("unk = 0x%x", attr[i].unk);
1031 pandecode_prop("magic_divisor = 0x%08x", attr[i].magic_divisor);
1032 if (attr[i].zero != 0)
1033 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr[i].zero);
1034 pandecode_prop("divisor = %d", attr[i].divisor);
1035 pandecode_magic_divisor(attr[i].magic_divisor, attr[i - 1].shift, attr[i].divisor, attr[i - 1].extra_flags);
1036 pandecode_indent--;
1037 pandecode_log("}, \n");
1038 }
1039
1040 }
1041
1042 pandecode_indent--;
1043 pandecode_log("};\n");
1044 }
1045
1046 static mali_ptr
1047 pandecode_replay_shader_address(const char *name, mali_ptr ptr)
1048 {
1049 /* TODO: Decode flags */
1050 mali_ptr shader_ptr = ptr & ~15;
1051
1052 char *a = pointer_as_memory_reference(shader_ptr);
1053 pandecode_prop("%s = (%s) | %d", name, a, (int) (ptr & 15));
1054 free(a);
1055
1056 return shader_ptr;
1057 }
1058
1059 static bool
1060 all_zero(unsigned *buffer, unsigned count)
1061 {
1062 for (unsigned i = 0; i < count; ++i) {
1063 if (buffer[i])
1064 return false;
1065 }
1066
1067 return true;
1068 }
1069
1070 static void
1071 pandecode_replay_stencil(const char *name, const struct mali_stencil_test *stencil)
1072 {
1073 if (all_zero((unsigned *) stencil, sizeof(stencil) / sizeof(unsigned)))
1074 return;
1075
1076 const char *func = pandecode_func_name(stencil->func);
1077 const char *sfail = pandecode_stencil_op_name(stencil->sfail);
1078 const char *dpfail = pandecode_stencil_op_name(stencil->dpfail);
1079 const char *dppass = pandecode_stencil_op_name(stencil->dppass);
1080
1081 if (stencil->zero)
1082 pandecode_msg("Stencil zero tripped: %X\n", stencil->zero);
1083
1084 pandecode_log(".stencil_%s = {\n", name);
1085 pandecode_indent++;
1086 pandecode_prop("ref = %d", stencil->ref);
1087 pandecode_prop("mask = 0x%02X", stencil->mask);
1088 pandecode_prop("func = %s", func);
1089 pandecode_prop("sfail = %s", sfail);
1090 pandecode_prop("dpfail = %s", dpfail);
1091 pandecode_prop("dppass = %s", dppass);
1092 pandecode_indent--;
1093 pandecode_log("},\n");
1094 }
1095
1096 static void
1097 pandecode_replay_blend_equation(const struct mali_blend_equation *blend)
1098 {
1099 if (blend->zero1)
1100 pandecode_msg("Blend zero tripped: %X\n", blend->zero1);
1101
1102 pandecode_log(".equation = {\n");
1103 pandecode_indent++;
1104
1105 pandecode_prop("rgb_mode = 0x%X", blend->rgb_mode);
1106 pandecode_prop("alpha_mode = 0x%X", blend->alpha_mode);
1107
1108 pandecode_log(".color_mask = ");
1109 pandecode_log_decoded_flags(mask_flag_info, blend->color_mask);
1110 pandecode_log_cont(",\n");
1111
1112 pandecode_indent--;
1113 pandecode_log("},\n");
1114 }
1115
1116 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1117
1118 static unsigned
1119 decode_bifrost_constant(u16 constant)
1120 {
1121 float lo = (float) (constant & 0xFF);
1122 float hi = (float) (constant >> 8);
1123
1124 return (hi / 255.0) + (lo / 65535.0);
1125 }
1126
1127 static mali_ptr
1128 pandecode_bifrost_blend(void *descs, int job_no, int rt_no)
1129 {
1130 struct bifrost_blend_rt *b =
1131 ((struct bifrost_blend_rt *) descs) + rt_no;
1132
1133 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1134 pandecode_indent++;
1135
1136 pandecode_prop("flags = 0x%" PRIx16, b->flags);
1137 pandecode_prop("constant = 0x%" PRIx8 " /* %f */",
1138 b->constant, decode_bifrost_constant(b->constant));
1139
1140 /* TODO figure out blend shader enable bit */
1141 pandecode_replay_blend_equation(&b->equation);
1142 pandecode_prop("unk2 = 0x%" PRIx16, b->unk2);
1143 pandecode_prop("index = 0x%" PRIx16, b->index);
1144 pandecode_prop("shader = 0x%" PRIx32, b->shader);
1145
1146 pandecode_indent--;
1147 pandecode_log("},\n");
1148
1149 return 0;
1150 }
1151
1152 static mali_ptr
1153 pandecode_midgard_blend(union midgard_blend *blend, bool is_shader)
1154 {
1155 if (all_zero((unsigned *) blend, sizeof(blend) / sizeof(unsigned)))
1156 return 0;
1157
1158 pandecode_log(".blend = {\n");
1159 pandecode_indent++;
1160
1161 if (is_shader) {
1162 pandecode_replay_shader_address("shader", blend->shader);
1163 } else {
1164 pandecode_replay_blend_equation(&blend->equation);
1165 pandecode_prop("constant = %f", blend->constant);
1166 }
1167
1168 pandecode_indent--;
1169 pandecode_log("},\n");
1170
1171 /* Return blend shader to disassemble if present */
1172 return is_shader ? (blend->shader & ~0xF) : 0;
1173 }
1174
1175 static mali_ptr
1176 pandecode_midgard_blend_mrt(void *descs, int job_no, int rt_no)
1177 {
1178 struct midgard_blend_rt *b =
1179 ((struct midgard_blend_rt *) descs) + rt_no;
1180
1181 /* Flags determine presence of blend shader */
1182 bool is_shader = (b->flags & 0xF) >= 0x2;
1183
1184 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no, rt_no);
1185 pandecode_indent++;
1186
1187 pandecode_prop("flags = 0x%" PRIx64, b->flags);
1188
1189 mali_ptr shader = pandecode_midgard_blend(&b->blend, is_shader);
1190
1191 pandecode_indent--;
1192 pandecode_log("};\n");
1193
1194 return shader;
1195 }
1196
1197 static int
1198 pandecode_replay_attribute_meta(int job_no, int count, const struct mali_vertex_tiler_postfix *v, bool varying, char *suffix)
1199 {
1200 char base[128];
1201 char *prefix = varying ? "varying" : "attribute";
1202 unsigned max_index = 0;
1203 snprintf(base, sizeof(base), "%s_meta", prefix);
1204
1205 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base, job_no, suffix);
1206 pandecode_indent++;
1207
1208 struct mali_attr_meta *attr_meta;
1209 mali_ptr p = varying ? (v->varying_meta & ~0xF) : v->attribute_meta;
1210
1211 struct pandecode_mapped_memory *attr_mem = pandecode_find_mapped_gpu_mem_containing(p);
1212
1213 for (int i = 0; i < count; ++i, p += sizeof(struct mali_attr_meta)) {
1214 attr_meta = pandecode_fetch_gpu_mem(attr_mem, p,
1215 sizeof(*attr_mem));
1216
1217 pandecode_log("{\n");
1218 pandecode_indent++;
1219 pandecode_prop("index = %d", attr_meta->index);
1220
1221 if (attr_meta->index > max_index)
1222 max_index = attr_meta->index;
1223 pandecode_replay_swizzle(attr_meta->swizzle);
1224 pandecode_prop("format = %s", pandecode_format_name(attr_meta->format));
1225
1226 pandecode_prop("unknown1 = 0x%" PRIx64, (u64) attr_meta->unknown1);
1227 pandecode_prop("unknown3 = 0x%" PRIx64, (u64) attr_meta->unknown3);
1228 pandecode_prop("src_offset = %d", attr_meta->src_offset);
1229 pandecode_indent--;
1230 pandecode_log("},\n");
1231
1232 }
1233
1234 pandecode_indent--;
1235 pandecode_log("};\n");
1236
1237 return max_index;
1238 }
1239
1240 static void
1241 pandecode_replay_indices(uintptr_t pindices, uint32_t index_count, int job_no)
1242 {
1243 struct pandecode_mapped_memory *imem = pandecode_find_mapped_gpu_mem_containing(pindices);
1244
1245 if (imem) {
1246 /* Indices are literally just a u32 array :) */
1247
1248 uint32_t *PANDECODE_PTR_VAR(indices, imem, pindices);
1249
1250 pandecode_log("uint32_t indices_%d[] = {\n", job_no);
1251 pandecode_indent++;
1252
1253 for (unsigned i = 0; i < (index_count + 1); i += 3)
1254 pandecode_log("%d, %d, %d,\n",
1255 indices[i],
1256 indices[i + 1],
1257 indices[i + 2]);
1258
1259 pandecode_indent--;
1260 pandecode_log("};\n");
1261 }
1262 }
1263
1264 /* return bits [lo, hi) of word */
1265 static u32
1266 bits(u32 word, u32 lo, u32 hi)
1267 {
1268 if (hi - lo >= 32)
1269 return word; // avoid undefined behavior with the shift
1270
1271 return (word >> lo) & ((1 << (hi - lo)) - 1);
1272 }
1273
1274 static void
1275 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix *p, int job_no)
1276 {
1277 pandecode_log_cont("{\n");
1278 pandecode_indent++;
1279
1280 pandecode_prop("invocation_count = 0x%" PRIx32, p->invocation_count);
1281 pandecode_prop("size_y_shift = %d", p->size_y_shift);
1282 pandecode_prop("size_z_shift = %d", p->size_z_shift);
1283 pandecode_prop("workgroups_x_shift = %d", p->workgroups_x_shift);
1284 pandecode_prop("workgroups_y_shift = %d", p->workgroups_y_shift);
1285 pandecode_prop("workgroups_z_shift = %d", p->workgroups_z_shift);
1286 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32, p->workgroups_x_shift_2);
1287
1288 /* Decode invocation_count. See the comment before the definition of
1289 * invocation_count for an explanation.
1290 */
1291 pandecode_msg("size: (%d, %d, %d)\n",
1292 bits(p->invocation_count, 0, p->size_y_shift) + 1,
1293 bits(p->invocation_count, p->size_y_shift, p->size_z_shift) + 1,
1294 bits(p->invocation_count, p->size_z_shift,
1295 p->workgroups_x_shift) + 1);
1296 pandecode_msg("workgroups: (%d, %d, %d)\n",
1297 bits(p->invocation_count, p->workgroups_x_shift,
1298 p->workgroups_y_shift) + 1,
1299 bits(p->invocation_count, p->workgroups_y_shift,
1300 p->workgroups_z_shift) + 1,
1301 bits(p->invocation_count, p->workgroups_z_shift,
1302 32) + 1);
1303
1304 /* TODO: Decode */
1305 if (p->unknown_draw)
1306 pandecode_prop("unknown_draw = 0x%" PRIx32, p->unknown_draw);
1307
1308 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32, p->workgroups_x_shift_3);
1309
1310 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p->draw_mode));
1311
1312 /* Index count only exists for tiler jobs anyway */
1313
1314 if (p->index_count)
1315 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32 ")", p->index_count + 1);
1316
1317 if (p->negative_start)
1318 pandecode_prop("negative_start = %d", p->negative_start);
1319
1320 DYN_MEMORY_PROP(p, job_no, indices);
1321
1322 if (p->zero1) {
1323 pandecode_msg("Zero tripped\n");
1324 pandecode_prop("zero1 = 0x%" PRIx32, p->zero1);
1325 }
1326
1327 pandecode_indent--;
1328 pandecode_log("},\n");
1329 }
1330
1331 static void
1332 pandecode_replay_uniform_buffers(mali_ptr pubufs, int ubufs_count, int job_no)
1333 {
1334 struct pandecode_mapped_memory *umem = pandecode_find_mapped_gpu_mem_containing(pubufs);
1335
1336 struct mali_uniform_buffer_meta *PANDECODE_PTR_VAR(ubufs, umem, pubufs);
1337
1338 for (int i = 0; i < ubufs_count; i++) {
1339 mali_ptr ptr = ubufs[i].ptr << 2;
1340 struct pandecode_mapped_memory *umem2 = pandecode_find_mapped_gpu_mem_containing(ptr);
1341 uint32_t *PANDECODE_PTR_VAR(ubuf, umem2, ptr);
1342 char name[50];
1343 snprintf(name, sizeof(name), "ubuf_%d", i);
1344 /* The blob uses ubuf 0 to upload internal stuff and
1345 * uniforms that won't fit/are accessed indirectly, so
1346 * it puts it in the batchbuffer.
1347 */
1348 pandecode_log("uint32_t %s_%d[] = {\n", name, job_no);
1349 pandecode_indent++;
1350
1351 for (int j = 0; j <= ubufs[i].size; j++) {
1352 for (int k = 0; k < 4; k++) {
1353 if (k == 0)
1354 pandecode_log("0x%"PRIx32", ", ubuf[4 * j + k]);
1355 else
1356 pandecode_log_cont("0x%"PRIx32", ", ubuf[4 * j + k]);
1357
1358 }
1359
1360 pandecode_log_cont("\n");
1361 }
1362
1363 pandecode_indent--;
1364 pandecode_log("};\n");
1365 }
1366
1367 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1368 job_no);
1369 pandecode_indent++;
1370
1371 for (int i = 0; i < ubufs_count; i++) {
1372 pandecode_log("{\n");
1373 pandecode_indent++;
1374 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs[i].size + 1);
1375 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i, job_no);
1376 pandecode_indent--;
1377 pandecode_log("},\n");
1378 }
1379
1380 pandecode_indent--;
1381 pandecode_log("};\n");
1382 }
1383
1384 static void
1385 pandecode_replay_scratchpad(uintptr_t pscratchpad, int job_no, char *suffix)
1386 {
1387
1388 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(pscratchpad);
1389
1390 struct bifrost_scratchpad *PANDECODE_PTR_VAR(scratchpad, mem, pscratchpad);
1391
1392 if (scratchpad->zero)
1393 pandecode_msg("XXX scratchpad zero tripped");
1394
1395 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64"_%d%s = {\n", pscratchpad, job_no, suffix);
1396 pandecode_indent++;
1397
1398 pandecode_prop("flags = 0x%x", scratchpad->flags);
1399 MEMORY_PROP(scratchpad, gpu_scratchpad);
1400
1401 pandecode_indent--;
1402 pandecode_log("};\n");
1403 }
1404
1405 static void
1406 pandecode_shader_disassemble(mali_ptr shader_ptr, int shader_no, int type,
1407 bool is_bifrost)
1408 {
1409 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(shader_ptr);
1410 uint8_t *PANDECODE_PTR_VAR(code, mem, shader_ptr);
1411
1412 /* Compute maximum possible size */
1413 size_t sz = mem->length - (shader_ptr - mem->gpu_va);
1414
1415 /* Print some boilerplate to clearly denote the assembly (which doesn't
1416 * obey indentation rules), and actually do the disassembly! */
1417
1418 printf("\n\n");
1419
1420 if (is_bifrost) {
1421 disassemble_bifrost(code, sz, false);
1422 } else {
1423 disassemble_midgard(code, sz);
1424 }
1425
1426 printf("\n\n");
1427 }
1428
1429 static void
1430 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix *p,
1431 int job_no, enum mali_job_type job_type,
1432 char *suffix, bool is_bifrost)
1433 {
1434 mali_ptr shader_meta_ptr = (u64) (uintptr_t) (p->_shader_upper << 4);
1435 struct pandecode_mapped_memory *attr_mem;
1436
1437 unsigned rt_count = 1;
1438
1439 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1440 * are the only things actually needed from the FBD, vertex/tiler jobs
1441 * no longer reference the FBD -- instead, this field points to some
1442 * info about the scratchpad.
1443 */
1444 if (is_bifrost)
1445 pandecode_replay_scratchpad(p->framebuffer & ~FBD_TYPE, job_no, suffix);
1446 else if (p->framebuffer & MALI_MFBD)
1447 rt_count = pandecode_replay_mfbd_bfr((u64) ((uintptr_t) p->framebuffer) & FBD_MASK, job_no, false);
1448 else if (job_type == JOB_TYPE_COMPUTE)
1449 pandecode_compute_fbd((u64) (uintptr_t) p->framebuffer, job_no);
1450 else
1451 pandecode_replay_sfbd((u64) (uintptr_t) p->framebuffer, job_no);
1452
1453 int varying_count = 0, attribute_count = 0, uniform_count = 0, uniform_buffer_count = 0;
1454 int texture_count = 0, sampler_count = 0;
1455
1456 if (shader_meta_ptr) {
1457 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr);
1458 struct mali_shader_meta *PANDECODE_PTR_VAR(s, smem, shader_meta_ptr);
1459
1460 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64"_%d%s = {\n", shader_meta_ptr, job_no, suffix);
1461 pandecode_indent++;
1462
1463 /* Save for dumps */
1464 attribute_count = s->attribute_count;
1465 varying_count = s->varying_count;
1466 texture_count = s->texture_count;
1467 sampler_count = s->sampler_count;
1468
1469 if (is_bifrost) {
1470 uniform_count = s->bifrost2.uniform_count;
1471 uniform_buffer_count = s->bifrost1.uniform_buffer_count;
1472 } else {
1473 uniform_count = s->midgard1.uniform_count;
1474 uniform_buffer_count = s->midgard1.uniform_buffer_count;
1475 }
1476
1477 mali_ptr shader_ptr = pandecode_replay_shader_address("shader", s->shader);
1478
1479 pandecode_prop("texture_count = %" PRId16, s->texture_count);
1480 pandecode_prop("sampler_count = %" PRId16, s->sampler_count);
1481 pandecode_prop("attribute_count = %" PRId16, s->attribute_count);
1482 pandecode_prop("varying_count = %" PRId16, s->varying_count);
1483
1484 if (is_bifrost) {
1485 pandecode_log(".bifrost1 = {\n");
1486 pandecode_indent++;
1487
1488 pandecode_prop("uniform_buffer_count = %" PRId32, s->bifrost1.uniform_buffer_count);
1489 pandecode_prop("unk1 = 0x%" PRIx32, s->bifrost1.unk1);
1490
1491 pandecode_indent--;
1492 pandecode_log("},\n");
1493 } else {
1494 pandecode_log(".midgard1 = {\n");
1495 pandecode_indent++;
1496
1497 pandecode_prop("uniform_count = %" PRId16, s->midgard1.uniform_count);
1498 pandecode_prop("uniform_buffer_count = %" PRId16, s->midgard1.uniform_buffer_count);
1499 pandecode_prop("work_count = %" PRId16, s->midgard1.work_count);
1500
1501 pandecode_log(".flags = ");
1502 pandecode_log_decoded_flags(shader_midgard1_flag_info, s->midgard1.flags);
1503 pandecode_log_cont(",\n");
1504
1505 pandecode_prop("unknown2 = 0x%" PRIx32, s->midgard1.unknown2);
1506
1507 pandecode_indent--;
1508 pandecode_log("},\n");
1509 }
1510
1511 if (s->depth_units || s->depth_factor) {
1512 pandecode_prop("depth_factor = %f", s->depth_factor);
1513 pandecode_prop("depth_units = %f", s->depth_units);
1514 }
1515
1516 if (s->alpha_coverage) {
1517 bool invert_alpha_coverage = s->alpha_coverage & 0xFFF0;
1518 uint16_t inverted_coverage = invert_alpha_coverage ? ~s->alpha_coverage : s->alpha_coverage;
1519
1520 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1521 invert_alpha_coverage ? "~" : "",
1522 MALI_GET_ALPHA_COVERAGE(inverted_coverage));
1523 }
1524
1525 if (s->unknown2_3 || s->unknown2_4) {
1526 pandecode_log(".unknown2_3 = ");
1527
1528 int unknown2_3 = s->unknown2_3;
1529 int unknown2_4 = s->unknown2_4;
1530
1531 /* We're not quite sure what these flags mean without the depth test, if anything */
1532
1533 if (unknown2_3 & (MALI_DEPTH_TEST | MALI_DEPTH_FUNC_MASK)) {
1534 const char *func = pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3));
1535 unknown2_3 &= ~MALI_DEPTH_FUNC_MASK;
1536
1537 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func);
1538 }
1539
1540 pandecode_log_decoded_flags(u3_flag_info, unknown2_3);
1541 pandecode_log_cont(",\n");
1542
1543 pandecode_log(".unknown2_4 = ");
1544 pandecode_log_decoded_flags(u4_flag_info, unknown2_4);
1545 pandecode_log_cont(",\n");
1546 }
1547
1548 if (s->stencil_mask_front || s->stencil_mask_back) {
1549 pandecode_prop("stencil_mask_front = 0x%02X", s->stencil_mask_front);
1550 pandecode_prop("stencil_mask_back = 0x%02X", s->stencil_mask_back);
1551 }
1552
1553 pandecode_replay_stencil("front", &s->stencil_front);
1554 pandecode_replay_stencil("back", &s->stencil_back);
1555
1556 if (is_bifrost) {
1557 pandecode_log(".bifrost2 = {\n");
1558 pandecode_indent++;
1559
1560 pandecode_prop("unk3 = 0x%" PRIx32, s->bifrost2.unk3);
1561 pandecode_prop("preload_regs = 0x%" PRIx32, s->bifrost2.preload_regs);
1562 pandecode_prop("uniform_count = %" PRId32, s->bifrost2.uniform_count);
1563 pandecode_prop("unk4 = 0x%" PRIx32, s->bifrost2.unk4);
1564
1565 pandecode_indent--;
1566 pandecode_log("},\n");
1567 } else if (s->midgard2.unknown2_7) {
1568 pandecode_log(".midgard2 = {\n");
1569 pandecode_indent++;
1570
1571 pandecode_prop("unknown2_7 = 0x%" PRIx32, s->midgard2.unknown2_7);
1572 pandecode_indent--;
1573 pandecode_log("},\n");
1574 }
1575
1576 if (s->unknown2_8)
1577 pandecode_prop("unknown2_8 = 0x%" PRIx32, s->unknown2_8);
1578
1579 if (!is_bifrost) {
1580 /* TODO: Blend shaders routing/disasm */
1581
1582 pandecode_midgard_blend(&s->blend, false);
1583 }
1584
1585 pandecode_indent--;
1586 pandecode_log("};\n");
1587
1588 /* MRT blend fields are used whenever MFBD is used, with
1589 * per-RT descriptors */
1590
1591 if (job_type == JOB_TYPE_TILER) {
1592 void* blend_base = (void *) (s + 1);
1593
1594 for (unsigned i = 0; i < rt_count; i++) {
1595 mali_ptr shader = 0;
1596
1597 if (is_bifrost)
1598 shader = pandecode_bifrost_blend(blend_base, job_no, i);
1599 else
1600 shader = pandecode_midgard_blend_mrt(blend_base, job_no, i);
1601
1602 if (shader)
1603 pandecode_shader_disassemble(shader, job_no, job_type, false);
1604 }
1605 }
1606
1607 pandecode_shader_disassemble(shader_ptr, job_no, job_type, is_bifrost);
1608 } else
1609 pandecode_msg("<no shader>\n");
1610
1611 if (p->viewport) {
1612 struct pandecode_mapped_memory *fmem = pandecode_find_mapped_gpu_mem_containing(p->viewport);
1613 struct mali_viewport *PANDECODE_PTR_VAR(f, fmem, p->viewport);
1614
1615 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no, suffix);
1616 pandecode_indent++;
1617
1618 pandecode_prop("clip_minx = %f", f->clip_minx);
1619 pandecode_prop("clip_miny = %f", f->clip_miny);
1620 pandecode_prop("clip_minz = %f", f->clip_minz);
1621 pandecode_prop("clip_maxx = %f", f->clip_maxx);
1622 pandecode_prop("clip_maxy = %f", f->clip_maxy);
1623 pandecode_prop("clip_maxz = %f", f->clip_maxz);
1624
1625 /* Only the higher coordinates are MALI_POSITIVE scaled */
1626
1627 pandecode_prop("viewport0 = { %d, %d }",
1628 f->viewport0[0], f->viewport0[1]);
1629
1630 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1631 f->viewport1[0] + 1, f->viewport1[1] + 1);
1632
1633 pandecode_indent--;
1634 pandecode_log("};\n");
1635 }
1636
1637 if (p->attribute_meta) {
1638 unsigned max_attr_index = pandecode_replay_attribute_meta(job_no, attribute_count, p, false, suffix);
1639
1640 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->attributes);
1641 pandecode_replay_attributes(attr_mem, p->attributes, job_no, suffix, max_attr_index + 1, false);
1642 }
1643
1644 /* Varyings are encoded like attributes but not actually sent; we just
1645 * pass a zero buffer with the right stride/size set, (or whatever)
1646 * since the GPU will write to it itself */
1647
1648 if (p->varyings) {
1649 attr_mem = pandecode_find_mapped_gpu_mem_containing(p->varyings);
1650
1651 /* Number of descriptors depends on whether there are
1652 * non-internal varyings */
1653
1654 pandecode_replay_attributes(attr_mem, p->varyings, job_no, suffix, varying_count > 1 ? 4 : 1, true);
1655 }
1656
1657 if (p->varying_meta) {
1658 pandecode_replay_attribute_meta(job_no, varying_count, p, true, suffix);
1659 }
1660
1661 bool is_compute = job_type == JOB_TYPE_COMPUTE;
1662
1663 if (p->uniforms && !is_compute) {
1664 int rows = uniform_count, width = 4;
1665 size_t sz = rows * width * sizeof(float);
1666
1667 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1668 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1669 u32 *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1670
1671 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no, suffix);
1672
1673 pandecode_indent++;
1674
1675 for (int row = 0; row < rows; row++) {
1676 for (int i = 0; i < width; i++) {
1677 u32 v = uniforms[i];
1678 float f;
1679 memcpy(&f, &v, sizeof(v));
1680 pandecode_log_cont("%X /* %f */, ", v, f);
1681 }
1682
1683 pandecode_log_cont("\n");
1684
1685 uniforms += width;
1686 }
1687
1688 pandecode_indent--;
1689 pandecode_log("};\n");
1690 } else if (p->uniforms) {
1691 int rows = uniform_count * 2;
1692 size_t sz = rows * sizeof(mali_ptr);
1693
1694 struct pandecode_mapped_memory *uniform_mem = pandecode_find_mapped_gpu_mem_containing(p->uniforms);
1695 pandecode_fetch_gpu_mem(uniform_mem, p->uniforms, sz);
1696 mali_ptr *PANDECODE_PTR_VAR(uniforms, uniform_mem, p->uniforms);
1697
1698 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no, suffix);
1699
1700 pandecode_indent++;
1701
1702 for (int row = 0; row < rows; row++) {
1703 char *a = pointer_as_memory_reference(uniforms[row]);
1704 pandecode_log("%s,\n", a);
1705 free(a);
1706 }
1707
1708 pandecode_indent--;
1709 pandecode_log("};\n");
1710
1711 }
1712
1713 if (p->uniform_buffers) {
1714 pandecode_replay_uniform_buffers(p->uniform_buffers, uniform_buffer_count, job_no);
1715 }
1716
1717 if (p->texture_trampoline) {
1718 struct pandecode_mapped_memory *mmem = pandecode_find_mapped_gpu_mem_containing(p->texture_trampoline);
1719
1720 if (mmem) {
1721 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline);
1722
1723 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no);
1724 pandecode_indent++;
1725
1726 for (int tex = 0; tex < texture_count; ++tex) {
1727 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1728 char *a = pointer_as_memory_reference(*u);
1729 pandecode_log("%s,\n", a);
1730 free(a);
1731 }
1732
1733 pandecode_indent--;
1734 pandecode_log("};\n");
1735
1736 /* Now, finally, descend down into the texture descriptor */
1737 for (int tex = 0; tex < texture_count; ++tex) {
1738 mali_ptr *PANDECODE_PTR_VAR(u, mmem, p->texture_trampoline + tex * sizeof(mali_ptr));
1739 struct pandecode_mapped_memory *tmem = pandecode_find_mapped_gpu_mem_containing(*u);
1740
1741 if (tmem) {
1742 struct mali_texture_descriptor *PANDECODE_PTR_VAR(t, tmem, *u);
1743
1744 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64"_%d_%d = {\n", *u, job_no, tex);
1745 pandecode_indent++;
1746
1747 pandecode_prop("width = MALI_POSITIVE(%" PRId16 ")", t->width + 1);
1748 pandecode_prop("height = MALI_POSITIVE(%" PRId16 ")", t->height + 1);
1749 pandecode_prop("depth = MALI_POSITIVE(%" PRId16 ")", t->depth + 1);
1750 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16 ")", t->array_size + 1);
1751 pandecode_prop("unknown3 = %" PRId16, t->unknown3);
1752 pandecode_prop("unknown3A = %" PRId8, t->unknown3A);
1753 pandecode_prop("nr_mipmap_levels = %" PRId8, t->nr_mipmap_levels);
1754
1755 struct mali_texture_format f = t->format;
1756
1757 pandecode_log(".format = {\n");
1758 pandecode_indent++;
1759
1760 pandecode_replay_swizzle(f.swizzle);
1761 pandecode_prop("format = %s", pandecode_format_name(f.format));
1762 pandecode_prop("type = %s", pandecode_texture_type(f.type));
1763 pandecode_prop("srgb = %" PRId32, f.srgb);
1764 pandecode_prop("unknown1 = %" PRId32, f.unknown1);
1765 pandecode_prop("usage2 = 0x%" PRIx32, f.usage2);
1766
1767 pandecode_indent--;
1768 pandecode_log("},\n");
1769
1770 pandecode_replay_swizzle(t->swizzle);
1771
1772 if (t->swizzle_zero) {
1773 /* Shouldn't happen */
1774 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1775 pandecode_prop("swizzle_zero = %d", t->swizzle_zero);
1776 }
1777
1778 pandecode_prop("unknown3 = 0x%" PRIx32, t->unknown3);
1779
1780 pandecode_prop("unknown5 = 0x%" PRIx32, t->unknown5);
1781 pandecode_prop("unknown6 = 0x%" PRIx32, t->unknown6);
1782 pandecode_prop("unknown7 = 0x%" PRIx32, t->unknown7);
1783
1784 pandecode_log(".payload = {\n");
1785 pandecode_indent++;
1786
1787 /* A bunch of bitmap pointers follow.
1788 * We work out the correct number,
1789 * based on the mipmap/cubemap
1790 * properties, but dump extra
1791 * possibilities to futureproof */
1792
1793 int bitmap_count = MALI_NEGATIVE(t->nr_mipmap_levels);
1794 bool manual_stride = f.usage2 & MALI_TEX_MANUAL_STRIDE;
1795
1796 /* Miptree for each face */
1797 if (f.type == MALI_TEX_CUBE)
1798 bitmap_count *= 6;
1799
1800 /* Array of textures */
1801 bitmap_count *= MALI_NEGATIVE(t->array_size);
1802
1803 /* Stride for each element */
1804 if (manual_stride)
1805 bitmap_count *= 2;
1806
1807 /* Sanity check the size */
1808 int max_count = sizeof(t->payload) / sizeof(t->payload[0]);
1809 assert (bitmap_count <= max_count);
1810
1811 /* Dump more to be safe, but not _that_ much more */
1812 int safe_count = MIN2(bitmap_count * 2, max_count);
1813
1814 for (int i = 0; i < safe_count; ++i) {
1815 char *prefix = (i >= bitmap_count) ? "// " : "";
1816
1817 /* How we dump depends if this is a stride or a pointer */
1818
1819 if ((f.usage2 & MALI_TEX_MANUAL_STRIDE) && (i & 1)) {
1820 /* signed 32-bit snuck in as a 64-bit pointer */
1821 uint64_t stride_set = t->payload[i];
1822 uint32_t clamped_stride = stride_set;
1823 int32_t stride = clamped_stride;
1824 assert(stride_set == clamped_stride);
1825 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix, stride);
1826 } else {
1827 char *a = pointer_as_memory_reference(t->payload[i]);
1828 pandecode_log("%s%s, \n", prefix, a);
1829 free(a);
1830 }
1831 }
1832
1833 pandecode_indent--;
1834 pandecode_log("},\n");
1835
1836 pandecode_indent--;
1837 pandecode_log("};\n");
1838 }
1839 }
1840 }
1841 }
1842
1843 if (p->sampler_descriptor) {
1844 struct pandecode_mapped_memory *smem = pandecode_find_mapped_gpu_mem_containing(p->sampler_descriptor);
1845
1846 if (smem) {
1847 struct mali_sampler_descriptor *s;
1848
1849 mali_ptr d = p->sampler_descriptor;
1850
1851 for (int i = 0; i < sampler_count; ++i) {
1852 s = pandecode_fetch_gpu_mem(smem, d + sizeof(*s) * i, sizeof(*s));
1853
1854 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no, i);
1855 pandecode_indent++;
1856
1857 /* Only the lower two bits are understood right now; the rest we display as hex */
1858 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32",\n",
1859 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MIN_MASK),
1860 MALI_FILTER_NAME(s->filter_mode & MALI_TEX_MAG_MASK),
1861 s->filter_mode & ~3);
1862
1863 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s->min_lod));
1864 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s->max_lod));
1865
1866 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s->wrap_s));
1867 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s->wrap_t));
1868 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s->wrap_r));
1869
1870 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s->compare_func));
1871
1872 if (s->zero || s->zero2) {
1873 pandecode_msg("Zero tripped\n");
1874 pandecode_prop("zero = 0x%X, 0x%X\n", s->zero, s->zero2);
1875 }
1876
1877 pandecode_prop("seamless_cube_map = %d", s->seamless_cube_map);
1878
1879 pandecode_prop("border_color = { %f, %f, %f, %f }",
1880 s->border_color[0],
1881 s->border_color[1],
1882 s->border_color[2],
1883 s->border_color[3]);
1884
1885 pandecode_indent--;
1886 pandecode_log("};\n");
1887 }
1888 }
1889 }
1890 }
1891
1892 static void
1893 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix *p, int job_no, bool is_bifrost)
1894 {
1895 pandecode_log_cont("{\n");
1896 pandecode_indent++;
1897
1898 MEMORY_PROP(p, position_varying);
1899 DYN_MEMORY_PROP(p, job_no, uniform_buffers);
1900 DYN_MEMORY_PROP(p, job_no, texture_trampoline);
1901 DYN_MEMORY_PROP(p, job_no, sampler_descriptor);
1902 DYN_MEMORY_PROP(p, job_no, uniforms);
1903 DYN_MEMORY_PROP(p, job_no, attributes);
1904 DYN_MEMORY_PROP(p, job_no, attribute_meta);
1905 DYN_MEMORY_PROP(p, job_no, varyings);
1906 DYN_MEMORY_PROP(p, job_no, varying_meta);
1907 DYN_MEMORY_PROP(p, job_no, viewport);
1908 DYN_MEMORY_PROP(p, job_no, occlusion_counter);
1909
1910 if (is_bifrost)
1911 pandecode_prop("framebuffer = scratchpad_%d_p", job_no);
1912 else
1913 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, p->framebuffer & MALI_MFBD ? "MALI_MFBD" : "0");
1914
1915 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no);
1916 pandecode_prop("flags = %d", p->flags);
1917
1918 pandecode_indent--;
1919 pandecode_log("},\n");
1920 }
1921
1922 static void
1923 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only *v)
1924 {
1925 pandecode_log_cont("{\n");
1926 pandecode_indent++;
1927
1928 pandecode_prop("unk2 = 0x%x", v->unk2);
1929
1930 if (v->zero0 || v->zero1) {
1931 pandecode_msg("vertex only zero tripped");
1932 pandecode_prop("zero0 = 0x%" PRIx32, v->zero0);
1933 pandecode_prop("zero1 = 0x%" PRIx64, v->zero1);
1934 }
1935
1936 pandecode_indent--;
1937 pandecode_log("}\n");
1938 }
1939
1940 static void
1941 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va, int job_no)
1942 {
1943
1944 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1945 const struct bifrost_tiler_heap_meta *PANDECODE_PTR_VAR(h, mem, gpu_va);
1946
1947 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no);
1948 pandecode_indent++;
1949
1950 if (h->zero) {
1951 pandecode_msg("tiler heap zero tripped\n");
1952 pandecode_prop("zero = 0x%x", h->zero);
1953 }
1954
1955 for (int i = 0; i < 12; i++) {
1956 if (h->zeros[i] != 0) {
1957 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1958 i, h->zeros[i]);
1959 }
1960 }
1961
1962 pandecode_prop("heap_size = 0x%x", h->heap_size);
1963 MEMORY_PROP(h, tiler_heap_start);
1964 MEMORY_PROP(h, tiler_heap_free);
1965
1966 /* this might point to the beginning of another buffer, when it's
1967 * really the end of the tiler heap buffer, so we have to be careful
1968 * here.
1969 */
1970 char *a = pointer_as_memory_reference(h->tiler_heap_end - 1);
1971 pandecode_prop("tiler_heap_end = %s + 1", a);
1972 free(a);
1973
1974 pandecode_indent--;
1975 pandecode_log("};\n");
1976 }
1977
1978 static void
1979 pandecode_replay_tiler_meta(mali_ptr gpu_va, int job_no)
1980 {
1981 struct pandecode_mapped_memory *mem = pandecode_find_mapped_gpu_mem_containing(gpu_va);
1982 const struct bifrost_tiler_meta *PANDECODE_PTR_VAR(t, mem, gpu_va);
1983
1984 pandecode_replay_tiler_heap_meta(t->tiler_heap_meta, job_no);
1985
1986 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no);
1987 pandecode_indent++;
1988
1989 if (t->zero0 || t->zero1) {
1990 pandecode_msg("tiler meta zero tripped");
1991 pandecode_prop("zero0 = 0x%" PRIx64, t->zero0);
1992 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
1993 }
1994
1995 pandecode_prop("hierarchy_mask = 0x%" PRIx16, t->hierarchy_mask);
1996 pandecode_prop("flags = 0x%" PRIx16, t->flags);
1997
1998 pandecode_prop("width = MALI_POSITIVE(%d)", t->width + 1);
1999 pandecode_prop("height = MALI_POSITIVE(%d)", t->height + 1);
2000 DYN_MEMORY_PROP(t, job_no, tiler_heap_meta);
2001
2002 for (int i = 0; i < 12; i++) {
2003 if (t->zeros[i] != 0) {
2004 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64 "\n",
2005 i, t->zeros[i]);
2006 }
2007 }
2008
2009 pandecode_indent--;
2010 pandecode_log("};\n");
2011 }
2012
2013 static void
2014 pandecode_replay_gl_enables(uint32_t gl_enables, int job_type)
2015 {
2016 pandecode_log(".gl_enables = ");
2017
2018 pandecode_log_decoded_flags(gl_enable_flag_info, gl_enables);
2019
2020 pandecode_log_cont(",\n");
2021 }
2022
2023 static void
2024 pandecode_replay_primitive_size(union midgard_primitive_size u, bool constant)
2025 {
2026 if (u.pointer == 0x0)
2027 return;
2028
2029 pandecode_log(".primitive_size = {\n");
2030 pandecode_indent++;
2031
2032 if (constant) {
2033 pandecode_prop("constant = %f", u.constant);
2034 } else {
2035 MEMORY_PROP((&u), pointer);
2036 }
2037
2038 pandecode_indent--;
2039 pandecode_log("},\n");
2040 }
2041
2042 static void
2043 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only *t, int job_no)
2044 {
2045 pandecode_log_cont("{\n");
2046 pandecode_indent++;
2047
2048 /* TODO: gl_PointSize on Bifrost */
2049 pandecode_replay_primitive_size(t->primitive_size, true);
2050
2051 DYN_MEMORY_PROP(t, job_no, tiler_meta);
2052 pandecode_replay_gl_enables(t->gl_enables, JOB_TYPE_TILER);
2053
2054 if (t->zero1 || t->zero2 || t->zero3 || t->zero4 || t->zero5
2055 || t->zero6 || t->zero7 || t->zero8) {
2056 pandecode_msg("tiler only zero tripped");
2057 pandecode_prop("zero1 = 0x%" PRIx64, t->zero1);
2058 pandecode_prop("zero2 = 0x%" PRIx64, t->zero2);
2059 pandecode_prop("zero3 = 0x%" PRIx64, t->zero3);
2060 pandecode_prop("zero4 = 0x%" PRIx64, t->zero4);
2061 pandecode_prop("zero5 = 0x%" PRIx64, t->zero5);
2062 pandecode_prop("zero6 = 0x%" PRIx64, t->zero6);
2063 pandecode_prop("zero7 = 0x%" PRIx32, t->zero7);
2064 pandecode_prop("zero8 = 0x%" PRIx64, t->zero8);
2065 }
2066
2067 pandecode_indent--;
2068 pandecode_log("},\n");
2069 }
2070
2071 static int
2072 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header *h,
2073 const struct pandecode_mapped_memory *mem,
2074 mali_ptr payload, int job_no)
2075 {
2076 struct bifrost_payload_vertex *PANDECODE_PTR_VAR(v, mem, payload);
2077
2078 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", true);
2079
2080 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no);
2081 pandecode_indent++;
2082
2083 pandecode_log(".prefix = ");
2084 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
2085
2086 pandecode_log(".vertex = ");
2087 pandecode_replay_vertex_only_bfr(&v->vertex);
2088
2089 pandecode_log(".postfix = ");
2090 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, true);
2091
2092 pandecode_indent--;
2093 pandecode_log("};\n");
2094
2095 return sizeof(*v);
2096 }
2097
2098 static int
2099 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header *h,
2100 const struct pandecode_mapped_memory *mem,
2101 mali_ptr payload, int job_no)
2102 {
2103 struct bifrost_payload_tiler *PANDECODE_PTR_VAR(t, mem, payload);
2104
2105 pandecode_replay_vertex_tiler_postfix_pre(&t->postfix, job_no, h->job_type, "", true);
2106
2107 pandecode_replay_indices(t->prefix.indices, t->prefix.index_count, job_no);
2108 pandecode_replay_tiler_meta(t->tiler.tiler_meta, job_no);
2109
2110 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no);
2111 pandecode_indent++;
2112
2113 pandecode_log(".prefix = ");
2114 pandecode_replay_vertex_tiler_prefix(&t->prefix, job_no);
2115
2116 pandecode_log(".tiler = ");
2117 pandecode_replay_tiler_only_bfr(&t->tiler, job_no);
2118
2119 pandecode_log(".postfix = ");
2120 pandecode_replay_vertex_tiler_postfix(&t->postfix, job_no, true);
2121
2122 pandecode_indent--;
2123 pandecode_log("};\n");
2124
2125 return sizeof(*t);
2126 }
2127
2128 static int
2129 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header *h,
2130 const struct pandecode_mapped_memory *mem,
2131 mali_ptr payload, int job_no)
2132 {
2133 struct midgard_payload_vertex_tiler *PANDECODE_PTR_VAR(v, mem, payload);
2134
2135 pandecode_replay_vertex_tiler_postfix_pre(&v->postfix, job_no, h->job_type, "", false);
2136
2137 pandecode_replay_indices(v->prefix.indices, v->prefix.index_count, job_no);
2138
2139 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no);
2140 pandecode_indent++;
2141
2142 bool has_primitive_pointer = v->prefix.unknown_draw & MALI_DRAW_VARYING_SIZE;
2143 pandecode_replay_primitive_size(v->primitive_size, !has_primitive_pointer);
2144
2145 pandecode_log(".prefix = ");
2146 pandecode_replay_vertex_tiler_prefix(&v->prefix, job_no);
2147
2148 pandecode_replay_gl_enables(v->gl_enables, h->job_type);
2149
2150 if (v->instance_shift || v->instance_odd) {
2151 pandecode_prop("instance_shift = 0x%d /* %d */",
2152 v->instance_shift, 1 << v->instance_shift);
2153 pandecode_prop("instance_odd = 0x%X /* %d */",
2154 v->instance_odd, (2 * v->instance_odd) + 1);
2155
2156 pandecode_padded_vertices(v->instance_shift, v->instance_odd);
2157 }
2158
2159 if (v->draw_start)
2160 pandecode_prop("draw_start = %d", v->draw_start);
2161
2162 #ifndef __LP64__
2163
2164 if (v->zero3) {
2165 pandecode_msg("Zero tripped\n");
2166 pandecode_prop("zero3 = 0x%" PRIx32, v->zero3);
2167 }
2168
2169 #endif
2170
2171 if (v->zero5) {
2172 pandecode_msg("Zero tripped\n");
2173 pandecode_prop("zero5 = 0x%" PRIx64, v->zero5);
2174 }
2175
2176 pandecode_log(".postfix = ");
2177 pandecode_replay_vertex_tiler_postfix(&v->postfix, job_no, false);
2178
2179 pandecode_indent--;
2180 pandecode_log("};\n");
2181
2182 return sizeof(*v);
2183 }
2184
2185 static int
2186 pandecode_replay_fragment_job(const struct pandecode_mapped_memory *mem,
2187 mali_ptr payload, int job_no,
2188 bool is_bifrost)
2189 {
2190 const struct mali_payload_fragment *PANDECODE_PTR_VAR(s, mem, payload);
2191
2192 bool fbd_dumped = false;
2193
2194 if (!is_bifrost && (s->framebuffer & FBD_TYPE) == MALI_SFBD) {
2195 /* Only SFBDs are understood, not MFBDs. We're speculating,
2196 * based on the versioning, kernel code, etc, that the
2197 * difference is between Single FrameBuffer Descriptor and
2198 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2199 * up with multi-framebuffer support being added (T7xx onwards,
2200 * including Gxx). In any event, there's some field shuffling
2201 * that we haven't looked into yet. */
2202
2203 pandecode_replay_sfbd(s->framebuffer & FBD_MASK, job_no);
2204 fbd_dumped = true;
2205 } else if ((s->framebuffer & FBD_TYPE) == MALI_MFBD) {
2206 /* We don't know if Bifrost supports SFBD's at all, since the
2207 * driver never uses them. And the format is different from
2208 * Midgard anyways, due to the tiler heap and scratchpad being
2209 * moved out into separate structures, so it's not clear what a
2210 * Bifrost SFBD would even look like without getting an actual
2211 * trace, which appears impossible.
2212 */
2213
2214 pandecode_replay_mfbd_bfr(s->framebuffer & FBD_MASK, job_no, true);
2215 fbd_dumped = true;
2216 }
2217
2218 uintptr_t p = (uintptr_t) s->framebuffer & FBD_MASK;
2219 pandecode_log("struct mali_payload_fragment payload_%"PRIx64"_%d = {\n", payload, job_no);
2220 pandecode_indent++;
2221
2222 /* See the comments by the macro definitions for mathematical context
2223 * on why this is so weird */
2224
2225 if (MALI_TILE_COORD_FLAGS(s->max_tile_coord) || MALI_TILE_COORD_FLAGS(s->min_tile_coord))
2226 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2227
2228 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2229 MALI_TILE_COORD_X(s->min_tile_coord) << MALI_TILE_SHIFT,
2230 MALI_TILE_COORD_Y(s->min_tile_coord) << MALI_TILE_SHIFT);
2231
2232 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2233 (MALI_TILE_COORD_X(s->max_tile_coord) + 1) << MALI_TILE_SHIFT,
2234 (MALI_TILE_COORD_Y(s->max_tile_coord) + 1) << MALI_TILE_SHIFT);
2235
2236 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2237 * we have to fallback on offsets. */
2238
2239 const char *fbd_type = s->framebuffer & MALI_MFBD ? "MALI_MFBD" : "MALI_SFBD";
2240
2241 if (fbd_dumped)
2242 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no, fbd_type);
2243 else
2244 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p), fbd_type);
2245
2246 pandecode_indent--;
2247 pandecode_log("};\n");
2248
2249 return sizeof(*s);
2250 }
2251
2252 static int job_descriptor_number = 0;
2253
2254 int
2255 pandecode_replay_jc(mali_ptr jc_gpu_va, bool bifrost)
2256 {
2257 struct mali_job_descriptor_header *h;
2258
2259 int start_number = 0;
2260
2261 bool first = true;
2262 bool last_size;
2263
2264 do {
2265 struct pandecode_mapped_memory *mem =
2266 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va);
2267
2268 void *payload;
2269
2270 h = PANDECODE_PTR(mem, jc_gpu_va, struct mali_job_descriptor_header);
2271
2272 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2273 * high 32-bits of the 64-bit pointer are reused to store
2274 * something else.
2275 */
2276 int offset = h->job_descriptor_size == MALI_JOB_32 &&
2277 h->job_type != JOB_TYPE_FRAGMENT ? 4 : 0;
2278 mali_ptr payload_ptr = jc_gpu_va + sizeof(*h) - offset;
2279
2280 payload = pandecode_fetch_gpu_mem(mem, payload_ptr,
2281 MALI_PAYLOAD_SIZE);
2282
2283 int job_no = job_descriptor_number++;
2284
2285 if (first)
2286 start_number = job_no;
2287
2288 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64"_%d = {\n", jc_gpu_va, job_no);
2289 pandecode_indent++;
2290
2291 pandecode_prop("job_type = %s", pandecode_job_type_name(h->job_type));
2292
2293 /* Save for next job fixing */
2294 last_size = h->job_descriptor_size;
2295
2296 if (h->job_descriptor_size)
2297 pandecode_prop("job_descriptor_size = %d", h->job_descriptor_size);
2298
2299 if (h->exception_status != 0x1)
2300 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2301 h->exception_status,
2302 (h->exception_status >> 16) & 0xFFFF,
2303 (h->exception_status >> 8) & 0x3,
2304 h->exception_status & 0xFF);
2305
2306 if (h->first_incomplete_task)
2307 pandecode_prop("first_incomplete_task = %d", h->first_incomplete_task);
2308
2309 if (h->fault_pointer)
2310 pandecode_prop("fault_pointer = 0x%" PRIx64, h->fault_pointer);
2311
2312 if (h->job_barrier)
2313 pandecode_prop("job_barrier = %d", h->job_barrier);
2314
2315 pandecode_prop("job_index = %d", h->job_index);
2316
2317 if (h->unknown_flags)
2318 pandecode_prop("unknown_flags = %d", h->unknown_flags);
2319
2320 if (h->job_dependency_index_1)
2321 pandecode_prop("job_dependency_index_1 = %d", h->job_dependency_index_1);
2322
2323 if (h->job_dependency_index_2)
2324 pandecode_prop("job_dependency_index_2 = %d", h->job_dependency_index_2);
2325
2326 pandecode_indent--;
2327 pandecode_log("};\n");
2328
2329 /* Do not touch the field yet -- decode the payload first, and
2330 * don't touch that either. This is essential for the uploads
2331 * to occur in sequence and therefore be dynamically allocated
2332 * correctly. Do note the size, however, for that related
2333 * reason. */
2334
2335 switch (h->job_type) {
2336 case JOB_TYPE_SET_VALUE: {
2337 struct mali_payload_set_value *s = payload;
2338 pandecode_log("struct mali_payload_set_value payload_%"PRIx64"_%d = {\n", payload_ptr, job_no);
2339 pandecode_indent++;
2340 MEMORY_PROP(s, out);
2341 pandecode_prop("unknown = 0x%" PRIX64, s->unknown);
2342 pandecode_indent--;
2343 pandecode_log("};\n");
2344
2345 break;
2346 }
2347
2348 case JOB_TYPE_TILER:
2349 case JOB_TYPE_VERTEX:
2350 case JOB_TYPE_COMPUTE:
2351 if (bifrost) {
2352 if (h->job_type == JOB_TYPE_TILER)
2353 pandecode_replay_tiler_job_bfr(h, mem, payload_ptr, job_no);
2354 else
2355 pandecode_replay_vertex_job_bfr(h, mem, payload_ptr, job_no);
2356 } else
2357 pandecode_replay_vertex_or_tiler_job_mdg(h, mem, payload_ptr, job_no);
2358
2359 break;
2360
2361 case JOB_TYPE_FRAGMENT:
2362 pandecode_replay_fragment_job(mem, payload_ptr, job_no, bifrost);
2363 break;
2364
2365 default:
2366 break;
2367 }
2368
2369 /* Handle linkage */
2370
2371 if (!first) {
2372 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no - 1);
2373
2374 if (last_size)
2375 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no);
2376 else
2377 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no);
2378 }
2379
2380 first = false;
2381
2382 } while ((jc_gpu_va = h->job_descriptor_size ? h->next_job_64 : h->next_job_32));
2383
2384 return start_number;
2385 }