2 * Copyright (C) 2017-2019 Alyssa Rosenzweig
3 * Copyright (C) 2017-2019 Connor Abbott
4 * Copyright (C) 2019 Collabora, Ltd.
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:
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
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
26 #include <panfrost-job.h>
33 #include "util/u_math.h"
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
);
40 #define MEMORY_PROP(obj, p) {\
42 char *a = pointer_as_memory_reference(obj->p); \
43 pandecode_prop("%s = %s", #p, a); \
48 #define DYN_MEMORY_PROP(obj, no, p) { \
50 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
53 /* Semantic logging type.
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
59 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
62 enum pandecode_log_type
{
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__)
72 unsigned pandecode_indent
= 0;
75 pandecode_make_indent(void)
77 for (unsigned i
= 0; i
< pandecode_indent
; ++i
)
82 pandecode_log_typed(enum pandecode_log_type type
, const char *format
, ...)
86 pandecode_make_indent();
88 if (type
== PANDECODE_MESSAGE
)
90 else if (type
== PANDECODE_PROPERTY
)
97 if (type
== PANDECODE_PROPERTY
)
102 pandecode_log_cont(const char *format
, ...)
106 va_start(ap
, format
);
111 struct pandecode_flag_info
{
117 pandecode_log_decoded_flags(const struct pandecode_flag_info
*flag_info
,
120 bool decodable_flags_found
= false;
122 for (int i
= 0; flag_info
[i
].name
; i
++) {
123 if ((flags
& flag_info
[i
].flag
) != flag_info
[i
].flag
)
126 if (!decodable_flags_found
) {
127 decodable_flags_found
= true;
129 pandecode_log_cont(" | ");
132 pandecode_log_cont("%s", flag_info
[i
].name
);
134 flags
&= ~flag_info
[i
].flag
;
137 if (decodable_flags_found
) {
139 pandecode_log_cont(" | 0x%" PRIx64
, flags
);
141 pandecode_log_cont("0x%" PRIx64
, flags
);
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
),
156 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
157 static const struct pandecode_flag_info clear_flag_info
[] = {
160 FLAG_INFO(SLOW_STENCIL
),
165 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
166 static const struct pandecode_flag_info mask_flag_info
[] = {
175 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
176 static const struct pandecode_flag_info u3_flag_info
[] = {
178 FLAG_INFO(CAN_DISCARD
),
179 FLAG_INFO(HAS_BLEND_SHADER
),
180 FLAG_INFO(DEPTH_TEST
),
184 static const struct pandecode_flag_info u4_flag_info
[] = {
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
),
195 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
196 static const struct pandecode_flag_info fb_fmt_flag_info
[] = {
204 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
205 static const struct pandecode_flag_info mfbd_fmt_flag_info
[] = {
212 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
213 static const struct pandecode_flag_info mfbd_extra_flag_info
[] = {
221 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
222 static const struct pandecode_flag_info shader_midgard1_flag_info
[] = {
224 FLAG_INFO(HELPER_INVOCATIONS
),
225 FLAG_INFO(READS_TILEBUFFER
),
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
),
240 extern char *replace_fragment
;
241 extern char *replace_vertex
;
244 pandecode_job_type_name(enum mali_job_type type
)
246 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
250 DEFINE_CASE(SET_VALUE
);
251 DEFINE_CASE(CACHE_FLUSH
);
252 DEFINE_CASE(COMPUTE
);
256 DEFINE_CASE(FRAGMENT
);
258 case JOB_NOT_STARTED
:
259 return "NOT_STARTED";
262 pandecode_log("Warning! Unknown job type %x\n", type
);
270 pandecode_draw_mode_name(enum mali_draw_mode mode
)
272 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
275 DEFINE_CASE(DRAW_NONE
);
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
);
285 DEFINE_CASE(QUAD_STRIP
);
288 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
294 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
296 pandecode_func_name(enum mali_func mode
)
303 DEFINE_CASE(GREATER
);
304 DEFINE_CASE(NOTEQUAL
);
309 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
314 /* Why is this duplicated? Who knows... */
315 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
317 pandecode_alt_func_name(enum mali_alt_func mode
)
324 DEFINE_CASE(GREATER
);
325 DEFINE_CASE(NOTEQUAL
);
330 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
335 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
337 pandecode_stencil_op_name(enum mali_stencil_op op
)
341 DEFINE_CASE(REPLACE
);
344 DEFINE_CASE(INCR_WRAP
);
345 DEFINE_CASE(DECR_WRAP
);
350 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
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
)
362 DEFINE_CASE(POT_DIVIDE
);
364 DEFINE_CASE(NPOT_DIVIDE
);
365 default: return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
371 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
373 pandecode_channel_name(enum mali_channel channel
)
382 DEFINE_CASE(RESERVED_0
);
383 DEFINE_CASE(RESERVED_1
);
386 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
391 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
393 pandecode_wrap_mode_name(enum mali_wrap_mode op
)
397 DEFINE_CASE(CLAMP_TO_EDGE
);
398 DEFINE_CASE(CLAMP_TO_BORDER
);
399 DEFINE_CASE(MIRRORED_REPEAT
);
402 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
407 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
409 pandecode_texture_type(enum mali_texture_type type
)
418 unreachable("Unknown case");
423 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
425 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt
)
429 DEFINE_CASE(UNKNOWN
);
434 unreachable("Invalid case");
440 pandecode_decode_fbd_type(enum mali_fbd_type type
)
442 if (type
== MALI_SFBD
) return "SFBD";
443 else if (type
== MALI_MFBD
) return "MFBD";
444 else return "WATFBD /* XXX */";
448 pandecode_replay_sfbd(uint64_t gpu_va
, int job_no
)
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
);
453 pandecode_log("struct mali_single_framebuffer framebuffer_%d = {\n", job_no
);
456 pandecode_prop("unknown1 = 0x%" PRIx32
, s
->unknown1
);
457 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->unknown2
);
459 pandecode_log(".format = ");
460 pandecode_log_decoded_flags(fb_fmt_flag_info
, s
->format
);
461 pandecode_log_cont(",\n");
463 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", s
->width
+ 1);
464 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", s
->height
+ 1);
466 MEMORY_PROP(s
, framebuffer
);
467 pandecode_prop("stride = %d", s
->stride
);
469 /* Earlier in the actual commandstream -- right before width -- but we
470 * delay to flow nicer */
472 pandecode_log(".clear_flags = ");
473 pandecode_log_decoded_flags(clear_flag_info
, s
->clear_flags
);
474 pandecode_log_cont(",\n");
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
));
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
));
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
);
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
);
500 if (s
->clear_stencil
) {
501 pandecode_prop("clear_stencil = 0x%x", s
->clear_stencil
);
504 MEMORY_PROP(s
, unknown_address_0
);
505 MEMORY_PROP(s
, tiler_polygon_list
);
506 MEMORY_PROP(s
, tiler_polygon_list_body
);
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
);
512 MEMORY_PROP(s
, tiler_heap_free
);
513 MEMORY_PROP(s
, tiler_heap_end
);
516 pandecode_log("};\n");
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
);
523 printf(".zero3 = {");
525 for (int i
= 0; i
< sizeof(s
->zero3
) / sizeof(s
->zero3
[0]); ++i
)
526 printf("%X, ", s
->zero3
[i
]);
530 printf(".zero6 = {");
532 for (int i
= 0; i
< sizeof(s
->zero6
) / sizeof(s
->zero6
[0]); ++i
)
533 printf("%X, ", s
->zero6
[i
]);
539 pandecode_u32_slide(unsigned name
, const u32
*slide
, unsigned count
)
541 pandecode_log(".unknown%d = {", name
);
543 for (int i
= 0; i
< count
; ++i
)
544 printf("%X, ", slide
[i
]);
546 pandecode_log("},\n");
549 #define SHORT_SLIDE(num) \
550 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
553 pandecode_compute_fbd(uint64_t gpu_va
, int job_no
)
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
);
558 pandecode_log("struct mali_compute_fbd framebuffer_%d = {\n", job_no
);
568 pandecode_replay_swizzle(unsigned swizzle
)
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));
578 pandecode_rt_format(struct mali_rt_format format
)
580 pandecode_log(".format = {\n");
583 pandecode_prop("unk1 = 0x%" PRIx32
, format
.unk1
);
584 pandecode_prop("unk2 = 0x%" PRIx32
, format
.unk2
);
585 pandecode_prop("unk3 = 0x%" PRIx32
, format
.unk3
);
587 pandecode_prop("block = %s",
588 pandecode_mfbd_block_format(format
.block
));
590 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
591 MALI_NEGATIVE(format
.nr_channels
));
593 pandecode_log(".flags = ");
594 pandecode_log_decoded_flags(mfbd_fmt_flag_info
, format
.flags
);
595 pandecode_log_cont(",\n");
597 pandecode_replay_swizzle(format
.swizzle
);
599 pandecode_prop("unk4 = 0x%" PRIx32
, format
.unk4
);
602 pandecode_log("},\n");
606 pandecode_render_target(uint64_t gpu_va
, unsigned job_no
, const struct bifrost_framebuffer
*fb
)
608 pandecode_log("struct bifrost_render_target rts_list_%d[] = {\n", job_no
);
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
);
617 pandecode_log("{\n");
620 pandecode_rt_format(rt
->format
);
622 if (rt
->format
.block
== MALI_MFBD_BLOCK_AFBC
) {
623 pandecode_log(".afbc = {\n");
626 char *a
= pointer_as_memory_reference(rt
->afbc
.metadata
);
627 pandecode_prop("metadata = %s", a
);
630 pandecode_prop("stride = %d", rt
->afbc
.stride
);
631 pandecode_prop("unk = 0x%" PRIx32
, rt
->afbc
.unk
);
634 pandecode_log("},\n");
636 pandecode_log(".chunknown = {\n");
639 pandecode_prop("unk = 0x%" PRIx64
, rt
->chunknown
.unk
);
641 char *a
= pointer_as_memory_reference(rt
->chunknown
.pointer
);
642 pandecode_prop("pointer = %s", a
);
646 pandecode_log("},\n");
649 MEMORY_PROP(rt
, framebuffer
);
650 pandecode_prop("framebuffer_stride = %d", rt
->framebuffer_stride
);
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
);
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
);
667 pandecode_log("},\n");
671 pandecode_log("};\n");
675 pandecode_replay_mfbd_bfr(uint64_t gpu_va
, int job_no
, bool with_render_targets
)
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
);
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.
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
692 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(fb
->sample_locations
);
694 const u16
*PANDECODE_PTR_VAR(samples
, smem
, fb
->sample_locations
);
696 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no
);
699 for (int i
= 0; i
< 32 + 16; i
++) {
700 pandecode_log("%d, %d,\n", samples
[2 * i
], samples
[2 * i
+ 1]);
704 pandecode_log("};\n");
707 pandecode_log("struct bifrost_framebuffer framebuffer_%d = {\n", job_no
);
710 pandecode_prop("unk0 = 0x%x", fb
->unk0
);
712 if (fb
->sample_locations
)
713 pandecode_prop("sample_locations = sample_locations_%d", job_no
);
715 /* Assume that unknown1 was emitted in the last job for
717 MEMORY_PROP(fb
, unknown1
);
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
);
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);
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
);
733 pandecode_log(".mfbd_flags = ");
734 pandecode_log_decoded_flags(mfbd_flag_info
, fb
->mfbd_flags
);
735 pandecode_log_cont(",\n");
737 pandecode_prop("clear_stencil = 0x%x", fb
->clear_stencil
);
738 pandecode_prop("clear_depth = %f", fb
->clear_depth
);
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
);
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
);
753 bool nonzero_weights
= false;
755 for (unsigned w
= 0; w
< ARRAY_SIZE(fb
->tiler_weights
); ++w
) {
756 nonzero_weights
|= fb
->tiler_weights
[w
] != 0x0;
759 if (nonzero_weights
) {
760 pandecode_log(".tiler_weights = {");
762 for (unsigned w
= 0; w
< ARRAY_SIZE(fb
->tiler_weights
); ++w
) {
763 pandecode_log("%d, ", fb
->tiler_weights
[w
]);
770 pandecode_log("};\n");
772 gpu_va
+= sizeof(struct bifrost_framebuffer
);
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
);
778 pandecode_log("struct bifrost_fb_extra fb_extra_%d = {\n", job_no
);
781 MEMORY_PROP(fbx
, checksum
);
783 if (fbx
->checksum_stride
)
784 pandecode_prop("checksum_stride = %d", fbx
->checksum_stride
);
786 pandecode_log(".flags = ");
787 pandecode_log_decoded_flags(mfbd_extra_flag_info
, fbx
->flags
);
788 pandecode_log_cont(",\n");
790 if (fbx
->flags
& MALI_EXTRA_AFBC_ZS
) {
791 pandecode_log(".ds_afbc = {\n");
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
);
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
,
803 pandecode_prop("padding = 0x%" PRIx64
,
804 fbx
->ds_afbc
.padding
);
808 pandecode_log("},\n");
810 pandecode_log(".ds_linear = {\n");
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
);
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
);
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
);
840 pandecode_log("},\n");
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
);
850 pandecode_log("};\n");
852 gpu_va
+= sizeof(struct bifrost_fb_extra
);
855 if (with_render_targets
)
856 pandecode_render_target(gpu_va
, job_no
, fb
);
858 /* Passback the render target count */
859 return MALI_NEGATIVE(fb
->rt_count_1
);
863 pandecode_replay_attributes(const struct pandecode_mapped_memory
*mem
,
864 mali_ptr addr
, int job_no
, char *suffix
,
865 int count
, bool varying
)
867 char *prefix
= varying
? "varyings" : "attributes";
869 union mali_attr
*attr
= pandecode_fetch_gpu_mem(mem
, addr
, sizeof(union mali_attr
) * count
);
872 snprintf(base
, sizeof(base
), "%s_data_%d%s", prefix
, job_no
, suffix
);
874 for (int i
= 0; i
< count
; ++i
) {
875 enum mali_attr_mode mode
= attr
[i
].elements
& 7;
877 if (mode
== MALI_ATTR_UNUSED
)
880 mali_ptr raw_elements
= attr
[i
].elements
& ~7;
882 /* TODO: Do we maybe want to dump the attribute values
883 * themselves given the specified format? Or is that too hard?
886 char *a
= pointer_as_memory_reference(raw_elements
);
887 pandecode_log("mali_ptr %s_%d_p = %s;\n", base
, i
, a
);
891 pandecode_log("union mali_attr %s_%d[] = {\n", prefix
, job_no
);
894 for (int i
= 0; i
< count
; ++i
) {
895 pandecode_log("{\n");
898 pandecode_prop("elements = (%s_%d_p) | %s", base
, i
, pandecode_attr_mode_name(attr
[i
].elements
& 7));
899 pandecode_prop("shift = %d", attr
[i
].shift
);
900 pandecode_prop("extra_flags = %d", attr
[i
].extra_flags
);
901 pandecode_prop("stride = 0x%" PRIx32
, attr
[i
].stride
);
902 pandecode_prop("size = 0x%" PRIx32
, attr
[i
].size
);
904 pandecode_log("}, \n");
906 if ((attr
[i
].elements
& 7) == MALI_ATTR_NPOT_DIVIDE
) {
908 pandecode_log("{\n");
910 pandecode_prop("unk = 0x%x", attr
[i
].unk
);
911 pandecode_prop("magic_divisor = 0x%08x", attr
[i
].magic_divisor
);
912 if (attr
[i
].zero
!= 0)
913 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr
[i
].zero
);
914 pandecode_prop("divisor = %d", attr
[i
].divisor
);
916 pandecode_log("}, \n");
922 pandecode_log("};\n");
926 pandecode_replay_shader_address(const char *name
, mali_ptr ptr
)
928 /* TODO: Decode flags */
929 mali_ptr shader_ptr
= ptr
& ~15;
931 char *a
= pointer_as_memory_reference(shader_ptr
);
932 pandecode_prop("%s = (%s) | %d", name
, a
, (int) (ptr
& 15));
939 all_zero(unsigned *buffer
, unsigned count
)
941 for (unsigned i
= 0; i
< count
; ++i
) {
950 pandecode_replay_stencil(const char *name
, const struct mali_stencil_test
*stencil
)
952 if (all_zero((unsigned *) stencil
, sizeof(stencil
) / sizeof(unsigned)))
955 const char *func
= pandecode_func_name(stencil
->func
);
956 const char *sfail
= pandecode_stencil_op_name(stencil
->sfail
);
957 const char *dpfail
= pandecode_stencil_op_name(stencil
->dpfail
);
958 const char *dppass
= pandecode_stencil_op_name(stencil
->dppass
);
961 pandecode_msg("Stencil zero tripped: %X\n", stencil
->zero
);
963 pandecode_log(".stencil_%s = {\n", name
);
965 pandecode_prop("ref = %d", stencil
->ref
);
966 pandecode_prop("mask = 0x%02X", stencil
->mask
);
967 pandecode_prop("func = %s", func
);
968 pandecode_prop("sfail = %s", sfail
);
969 pandecode_prop("dpfail = %s", dpfail
);
970 pandecode_prop("dppass = %s", dppass
);
972 pandecode_log("},\n");
976 pandecode_replay_blend_equation(const struct mali_blend_equation
*blend
)
979 pandecode_msg("Blend zero tripped: %X\n", blend
->zero1
);
981 pandecode_log(".equation = {\n");
984 pandecode_prop("rgb_mode = 0x%X", blend
->rgb_mode
);
985 pandecode_prop("alpha_mode = 0x%X", blend
->alpha_mode
);
987 pandecode_log(".color_mask = ");
988 pandecode_log_decoded_flags(mask_flag_info
, blend
->color_mask
);
989 pandecode_log_cont(",\n");
992 pandecode_log("},\n");
995 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
998 decode_bifrost_constant(u16 constant
)
1000 float lo
= (float) (constant
& 0xFF);
1001 float hi
= (float) (constant
>> 8);
1003 return (hi
/ 255.0) + (lo
/ 65535.0);
1007 pandecode_bifrost_blend(void *descs
, int job_no
, int rt_no
)
1009 struct bifrost_blend_rt
*b
=
1010 ((struct bifrost_blend_rt
*) descs
) + rt_no
;
1012 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1015 pandecode_prop("flags = 0x%" PRIx16
, b
->flags
);
1016 pandecode_prop("constant = 0x%" PRIx8
" /* %f */",
1017 b
->constant
, decode_bifrost_constant(b
->constant
));
1019 /* TODO figure out blend shader enable bit */
1020 pandecode_replay_blend_equation(&b
->equation
);
1021 pandecode_prop("unk2 = 0x%" PRIx16
, b
->unk2
);
1022 pandecode_prop("index = 0x%" PRIx16
, b
->index
);
1023 pandecode_prop("shader = 0x%" PRIx32
, b
->shader
);
1026 pandecode_log("},\n");
1032 pandecode_midgard_blend(union midgard_blend
*blend
, bool is_shader
)
1034 if (all_zero((unsigned *) blend
, sizeof(blend
) / sizeof(unsigned)))
1037 pandecode_log(".blend = {\n");
1041 pandecode_replay_shader_address("shader", blend
->shader
);
1043 pandecode_replay_blend_equation(&blend
->equation
);
1044 pandecode_prop("constant = %f", blend
->constant
);
1048 pandecode_log("},\n");
1050 /* Return blend shader to disassemble if present */
1051 return is_shader
? (blend
->shader
& ~0xF) : 0;
1055 pandecode_midgard_blend_mrt(void *descs
, int job_no
, int rt_no
)
1057 struct midgard_blend_rt
*b
=
1058 ((struct midgard_blend_rt
*) descs
) + rt_no
;
1060 /* Flags determine presence of blend shader */
1061 bool is_shader
= (b
->flags
& 0xF) >= 0x2;
1063 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1066 pandecode_prop("flags = 0x%" PRIx64
, b
->flags
);
1068 mali_ptr shader
= pandecode_midgard_blend(&b
->blend
, is_shader
);
1071 pandecode_log("};\n");
1077 pandecode_replay_attribute_meta(int job_no
, int count
, const struct mali_vertex_tiler_postfix
*v
, bool varying
, char *suffix
)
1080 char *prefix
= varying
? "varying" : "attribute";
1081 unsigned max_index
= 0;
1082 snprintf(base
, sizeof(base
), "%s_meta", prefix
);
1084 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base
, job_no
, suffix
);
1087 struct mali_attr_meta
*attr_meta
;
1088 mali_ptr p
= varying
? (v
->varying_meta
& ~0xF) : v
->attribute_meta
;
1090 struct pandecode_mapped_memory
*attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
);
1092 for (int i
= 0; i
< count
; ++i
, p
+= sizeof(struct mali_attr_meta
)) {
1093 attr_meta
= pandecode_fetch_gpu_mem(attr_mem
, p
,
1096 pandecode_log("{\n");
1098 pandecode_prop("index = %d", attr_meta
->index
);
1100 if (attr_meta
->index
> max_index
)
1101 max_index
= attr_meta
->index
;
1102 pandecode_replay_swizzle(attr_meta
->swizzle
);
1103 pandecode_prop("format = %s", pandecode_format_name(attr_meta
->format
));
1105 pandecode_prop("unknown1 = 0x%" PRIx64
, (u64
) attr_meta
->unknown1
);
1106 pandecode_prop("unknown3 = 0x%" PRIx64
, (u64
) attr_meta
->unknown3
);
1107 pandecode_prop("src_offset = 0x%" PRIx64
, (u64
) attr_meta
->src_offset
);
1109 pandecode_log("},\n");
1114 pandecode_log("};\n");
1120 pandecode_replay_indices(uintptr_t pindices
, uint32_t index_count
, int job_no
)
1122 struct pandecode_mapped_memory
*imem
= pandecode_find_mapped_gpu_mem_containing(pindices
);
1125 /* Indices are literally just a u32 array :) */
1127 uint32_t *PANDECODE_PTR_VAR(indices
, imem
, pindices
);
1129 pandecode_log("uint32_t indices_%d[] = {\n", job_no
);
1132 for (unsigned i
= 0; i
< (index_count
+ 1); i
+= 3)
1133 pandecode_log("%d, %d, %d,\n",
1139 pandecode_log("};\n");
1143 /* return bits [lo, hi) of word */
1145 bits(u32 word
, u32 lo
, u32 hi
)
1148 return word
; // avoid undefined behavior with the shift
1150 return (word
>> lo
) & ((1 << (hi
- lo
)) - 1);
1154 pandecode_replay_vertex_tiler_prefix(struct mali_vertex_tiler_prefix
*p
, int job_no
)
1156 pandecode_log_cont("{\n");
1159 pandecode_prop("invocation_count = 0x%" PRIx32
, p
->invocation_count
);
1160 pandecode_prop("size_y_shift = %d", p
->size_y_shift
);
1161 pandecode_prop("size_z_shift = %d", p
->size_z_shift
);
1162 pandecode_prop("workgroups_x_shift = %d", p
->workgroups_x_shift
);
1163 pandecode_prop("workgroups_y_shift = %d", p
->workgroups_y_shift
);
1164 pandecode_prop("workgroups_z_shift = %d", p
->workgroups_z_shift
);
1165 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32
, p
->workgroups_x_shift_2
);
1167 /* Decode invocation_count. See the comment before the definition of
1168 * invocation_count for an explanation.
1170 pandecode_msg("size: (%d, %d, %d)\n",
1171 bits(p
->invocation_count
, 0, p
->size_y_shift
) + 1,
1172 bits(p
->invocation_count
, p
->size_y_shift
, p
->size_z_shift
) + 1,
1173 bits(p
->invocation_count
, p
->size_z_shift
,
1174 p
->workgroups_x_shift
) + 1);
1175 pandecode_msg("workgroups: (%d, %d, %d)\n",
1176 bits(p
->invocation_count
, p
->workgroups_x_shift
,
1177 p
->workgroups_y_shift
) + 1,
1178 bits(p
->invocation_count
, p
->workgroups_y_shift
,
1179 p
->workgroups_z_shift
) + 1,
1180 bits(p
->invocation_count
, p
->workgroups_z_shift
,
1184 if (p
->unknown_draw
)
1185 pandecode_prop("unknown_draw = 0x%" PRIx32
, p
->unknown_draw
);
1187 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32
, p
->workgroups_x_shift_3
);
1189 pandecode_prop("draw_mode = %s", pandecode_draw_mode_name(p
->draw_mode
));
1191 /* Index count only exists for tiler jobs anyway */
1194 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32
")", p
->index_count
+ 1);
1196 if (p
->negative_start
)
1197 pandecode_prop("negative_start = %d", p
->negative_start
);
1199 DYN_MEMORY_PROP(p
, job_no
, indices
);
1202 pandecode_msg("Zero tripped\n");
1203 pandecode_prop("zero1 = 0x%" PRIx32
, p
->zero1
);
1207 pandecode_log("},\n");
1211 pandecode_replay_uniform_buffers(mali_ptr pubufs
, int ubufs_count
, int job_no
)
1213 struct pandecode_mapped_memory
*umem
= pandecode_find_mapped_gpu_mem_containing(pubufs
);
1215 struct mali_uniform_buffer_meta
*PANDECODE_PTR_VAR(ubufs
, umem
, pubufs
);
1217 for (int i
= 0; i
< ubufs_count
; i
++) {
1218 mali_ptr ptr
= ubufs
[i
].ptr
<< 2;
1219 struct pandecode_mapped_memory
*umem2
= pandecode_find_mapped_gpu_mem_containing(ptr
);
1220 uint32_t *PANDECODE_PTR_VAR(ubuf
, umem2
, ptr
);
1222 snprintf(name
, sizeof(name
), "ubuf_%d", i
);
1223 /* The blob uses ubuf 0 to upload internal stuff and
1224 * uniforms that won't fit/are accessed indirectly, so
1225 * it puts it in the batchbuffer.
1227 pandecode_log("uint32_t %s_%d[] = {\n", name
, job_no
);
1230 for (int j
= 0; j
<= ubufs
[i
].size
; j
++) {
1231 for (int k
= 0; k
< 4; k
++) {
1233 pandecode_log("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1235 pandecode_log_cont("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1239 pandecode_log_cont("\n");
1243 pandecode_log("};\n");
1246 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%d[] = {\n",
1250 for (int i
= 0; i
< ubufs_count
; i
++) {
1251 pandecode_log("{\n");
1253 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs
[i
].size
+ 1);
1254 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i
, job_no
);
1256 pandecode_log("},\n");
1260 pandecode_log("};\n");
1264 pandecode_replay_scratchpad(uintptr_t pscratchpad
, int job_no
, char *suffix
)
1267 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(pscratchpad
);
1269 struct bifrost_scratchpad
*PANDECODE_PTR_VAR(scratchpad
, mem
, pscratchpad
);
1271 if (scratchpad
->zero
)
1272 pandecode_msg("XXX scratchpad zero tripped");
1274 pandecode_log("struct bifrost_scratchpad scratchpad_%d%s = {\n", job_no
, suffix
);
1277 pandecode_prop("flags = 0x%x", scratchpad
->flags
);
1278 MEMORY_PROP(scratchpad
, gpu_scratchpad
);
1281 pandecode_log("};\n");
1285 pandecode_shader_disassemble(mali_ptr shader_ptr
, int shader_no
, int type
,
1288 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(shader_ptr
);
1289 uint8_t *PANDECODE_PTR_VAR(code
, mem
, shader_ptr
);
1291 /* Compute maximum possible size */
1292 size_t sz
= mem
->length
- (shader_ptr
- mem
->gpu_va
);
1294 /* Print some boilerplate to clearly denote the assembly (which doesn't
1295 * obey indentation rules), and actually do the disassembly! */
1300 disassemble_bifrost(code
, sz
, false);
1302 disassemble_midgard(code
, sz
);
1309 pandecode_replay_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix
*p
,
1310 int job_no
, enum mali_job_type job_type
,
1311 char *suffix
, bool is_bifrost
)
1313 mali_ptr shader_meta_ptr
= (u64
) (uintptr_t) (p
->_shader_upper
<< 4);
1314 struct pandecode_mapped_memory
*attr_mem
;
1316 unsigned rt_count
= 1;
1318 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1319 * are the only things actually needed from the FBD, vertex/tiler jobs
1320 * no longer reference the FBD -- instead, this field points to some
1321 * info about the scratchpad.
1324 pandecode_replay_scratchpad(p
->framebuffer
& ~FBD_TYPE
, job_no
, suffix
);
1325 else if (p
->framebuffer
& MALI_MFBD
)
1326 rt_count
= pandecode_replay_mfbd_bfr((u64
) ((uintptr_t) p
->framebuffer
) & FBD_MASK
, job_no
, false);
1327 else if (job_type
== JOB_TYPE_COMPUTE
)
1328 pandecode_compute_fbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1330 pandecode_replay_sfbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1332 int varying_count
= 0, attribute_count
= 0, uniform_count
= 0, uniform_buffer_count
= 0;
1333 int texture_count
= 0, sampler_count
= 0;
1335 if (shader_meta_ptr
) {
1336 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr
);
1337 struct mali_shader_meta
*PANDECODE_PTR_VAR(s
, smem
, shader_meta_ptr
);
1339 pandecode_log("struct mali_shader_meta shader_meta_%d%s = {\n", job_no
, suffix
);
1342 /* Save for dumps */
1343 attribute_count
= s
->attribute_count
;
1344 varying_count
= s
->varying_count
;
1345 texture_count
= s
->texture_count
;
1346 sampler_count
= s
->sampler_count
;
1349 uniform_count
= s
->bifrost2
.uniform_count
;
1350 uniform_buffer_count
= s
->bifrost1
.uniform_buffer_count
;
1352 uniform_count
= s
->midgard1
.uniform_count
;
1353 uniform_buffer_count
= s
->midgard1
.uniform_buffer_count
;
1356 mali_ptr shader_ptr
= pandecode_replay_shader_address("shader", s
->shader
);
1358 pandecode_prop("texture_count = %" PRId16
, s
->texture_count
);
1359 pandecode_prop("sampler_count = %" PRId16
, s
->sampler_count
);
1360 pandecode_prop("attribute_count = %" PRId16
, s
->attribute_count
);
1361 pandecode_prop("varying_count = %" PRId16
, s
->varying_count
);
1364 pandecode_log(".bifrost1 = {\n");
1367 pandecode_prop("uniform_buffer_count = %" PRId32
, s
->bifrost1
.uniform_buffer_count
);
1368 pandecode_prop("unk1 = 0x%" PRIx32
, s
->bifrost1
.unk1
);
1371 pandecode_log("},\n");
1373 pandecode_log(".midgard1 = {\n");
1376 pandecode_prop("uniform_count = %" PRId16
, s
->midgard1
.uniform_count
);
1377 pandecode_prop("uniform_buffer_count = %" PRId16
, s
->midgard1
.uniform_buffer_count
);
1378 pandecode_prop("work_count = %" PRId16
, s
->midgard1
.work_count
);
1380 pandecode_log(".flags = ");
1381 pandecode_log_decoded_flags(shader_midgard1_flag_info
, s
->midgard1
.flags
);
1382 pandecode_log_cont(",\n");
1384 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->midgard1
.unknown2
);
1387 pandecode_log("},\n");
1390 if (s
->depth_units
|| s
->depth_factor
) {
1392 pandecode_prop("depth_units = %f", s
->depth_units
);
1394 pandecode_prop("depth_units = MALI_NEGATIVE(%f)", s
->depth_units
- 1.0f
);
1396 pandecode_prop("depth_factor = %f", s
->depth_factor
);
1399 if (s
->alpha_coverage
) {
1400 bool invert_alpha_coverage
= s
->alpha_coverage
& 0xFFF0;
1401 uint16_t inverted_coverage
= invert_alpha_coverage
? ~s
->alpha_coverage
: s
->alpha_coverage
;
1403 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1404 invert_alpha_coverage
? "~" : "",
1405 MALI_GET_ALPHA_COVERAGE(inverted_coverage
));
1408 if (s
->unknown2_3
|| s
->unknown2_4
) {
1409 pandecode_log(".unknown2_3 = ");
1411 int unknown2_3
= s
->unknown2_3
;
1412 int unknown2_4
= s
->unknown2_4
;
1414 /* We're not quite sure what these flags mean without the depth test, if anything */
1416 if (unknown2_3
& (MALI_DEPTH_TEST
| MALI_DEPTH_FUNC_MASK
)) {
1417 const char *func
= pandecode_func_name(MALI_GET_DEPTH_FUNC(unknown2_3
));
1418 unknown2_3
&= ~MALI_DEPTH_FUNC_MASK
;
1420 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func
);
1423 pandecode_log_decoded_flags(u3_flag_info
, unknown2_3
);
1424 pandecode_log_cont(",\n");
1426 pandecode_log(".unknown2_4 = ");
1427 pandecode_log_decoded_flags(u4_flag_info
, unknown2_4
);
1428 pandecode_log_cont(",\n");
1431 if (s
->stencil_mask_front
|| s
->stencil_mask_back
) {
1432 pandecode_prop("stencil_mask_front = 0x%02X", s
->stencil_mask_front
);
1433 pandecode_prop("stencil_mask_back = 0x%02X", s
->stencil_mask_back
);
1436 pandecode_replay_stencil("front", &s
->stencil_front
);
1437 pandecode_replay_stencil("back", &s
->stencil_back
);
1440 pandecode_log(".bifrost2 = {\n");
1443 pandecode_prop("unk3 = 0x%" PRIx32
, s
->bifrost2
.unk3
);
1444 pandecode_prop("preload_regs = 0x%" PRIx32
, s
->bifrost2
.preload_regs
);
1445 pandecode_prop("uniform_count = %" PRId32
, s
->bifrost2
.uniform_count
);
1446 pandecode_prop("unk4 = 0x%" PRIx32
, s
->bifrost2
.unk4
);
1449 pandecode_log("},\n");
1450 } else if (s
->midgard2
.unknown2_7
) {
1451 pandecode_log(".midgard2 = {\n");
1454 pandecode_prop("unknown2_7 = 0x%" PRIx32
, s
->midgard2
.unknown2_7
);
1456 pandecode_log("},\n");
1460 pandecode_prop("unknown2_8 = 0x%" PRIx32
, s
->unknown2_8
);
1463 /* TODO: Blend shaders routing/disasm */
1465 pandecode_midgard_blend(&s
->blend
, false);
1469 pandecode_log("};\n");
1471 /* MRT blend fields are used whenever MFBD is used, with
1472 * per-RT descriptors */
1474 if (job_type
== JOB_TYPE_TILER
) {
1475 void* blend_base
= (void *) (s
+ 1);
1477 for (unsigned i
= 0; i
< rt_count
; i
++) {
1478 mali_ptr shader
= 0;
1481 shader
= pandecode_bifrost_blend(blend_base
, job_no
, i
);
1483 shader
= pandecode_midgard_blend_mrt(blend_base
, job_no
, i
);
1486 pandecode_shader_disassemble(shader
, job_no
, job_type
, false);
1490 pandecode_shader_disassemble(shader_ptr
, job_no
, job_type
, is_bifrost
);
1492 pandecode_msg("<no shader>\n");
1495 struct pandecode_mapped_memory
*fmem
= pandecode_find_mapped_gpu_mem_containing(p
->viewport
);
1496 struct mali_viewport
*PANDECODE_PTR_VAR(f
, fmem
, p
->viewport
);
1498 pandecode_log("struct mali_viewport viewport_%d%s = {\n", job_no
, suffix
);
1501 pandecode_prop("clip_minx = %f", f
->clip_minx
);
1502 pandecode_prop("clip_miny = %f", f
->clip_miny
);
1503 pandecode_prop("clip_minz = %f", f
->clip_minz
);
1504 pandecode_prop("clip_maxx = %f", f
->clip_maxx
);
1505 pandecode_prop("clip_maxy = %f", f
->clip_maxy
);
1506 pandecode_prop("clip_maxz = %f", f
->clip_maxz
);
1508 /* Only the higher coordinates are MALI_POSITIVE scaled */
1510 pandecode_prop("viewport0 = { %d, %d }",
1511 f
->viewport0
[0], f
->viewport0
[1]);
1513 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1514 f
->viewport1
[0] + 1, f
->viewport1
[1] + 1);
1517 pandecode_log("};\n");
1520 if (p
->attribute_meta
) {
1521 unsigned max_attr_index
= pandecode_replay_attribute_meta(job_no
, attribute_count
, p
, false, suffix
);
1523 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->attributes
);
1524 pandecode_replay_attributes(attr_mem
, p
->attributes
, job_no
, suffix
, max_attr_index
+ 1, false);
1527 /* Varyings are encoded like attributes but not actually sent; we just
1528 * pass a zero buffer with the right stride/size set, (or whatever)
1529 * since the GPU will write to it itself */
1532 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->varyings
);
1534 /* Number of descriptors depends on whether there are
1535 * non-internal varyings */
1537 pandecode_replay_attributes(attr_mem
, p
->varyings
, job_no
, suffix
, varying_count
> 1 ? 4 : 1, true);
1540 if (p
->varying_meta
) {
1541 pandecode_replay_attribute_meta(job_no
, varying_count
, p
, true, suffix
);
1544 bool is_compute
= job_type
== JOB_TYPE_COMPUTE
;
1546 if (p
->uniforms
&& !is_compute
) {
1547 int rows
= uniform_count
, width
= 4;
1548 size_t sz
= rows
* width
* sizeof(float);
1550 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1551 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1552 u32
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1554 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no
, suffix
);
1558 for (int row
= 0; row
< rows
; row
++) {
1559 for (int i
= 0; i
< width
; i
++) {
1560 u32 v
= uniforms
[i
];
1562 memcpy(&f
, &v
, sizeof(v
));
1563 pandecode_log_cont("%X /* %f */, ", v
, f
);
1566 pandecode_log_cont("\n");
1572 pandecode_log("};\n");
1573 } else if (p
->uniforms
) {
1574 int rows
= uniform_count
* 2;
1575 size_t sz
= rows
* sizeof(mali_ptr
);
1577 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1578 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1579 mali_ptr
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1581 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no
, suffix
);
1585 for (int row
= 0; row
< rows
; row
++) {
1586 char *a
= pointer_as_memory_reference(uniforms
[row
]);
1587 pandecode_log("%s,\n", a
);
1592 pandecode_log("};\n");
1596 if (p
->uniform_buffers
) {
1597 pandecode_replay_uniform_buffers(p
->uniform_buffers
, uniform_buffer_count
, job_no
);
1600 if (p
->texture_trampoline
) {
1601 struct pandecode_mapped_memory
*mmem
= pandecode_find_mapped_gpu_mem_containing(p
->texture_trampoline
);
1604 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
);
1606 pandecode_log("uint64_t texture_trampoline_%d[] = {\n", job_no
);
1609 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1610 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1611 char *a
= pointer_as_memory_reference(*u
);
1612 pandecode_log("%s,\n", a
);
1617 pandecode_log("};\n");
1619 /* Now, finally, descend down into the texture descriptor */
1620 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1621 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1622 struct pandecode_mapped_memory
*tmem
= pandecode_find_mapped_gpu_mem_containing(*u
);
1625 struct mali_texture_descriptor
*PANDECODE_PTR_VAR(t
, tmem
, *u
);
1627 pandecode_log("struct mali_texture_descriptor texture_descriptor_%d_%d = {\n", job_no
, tex
);
1630 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", t
->width
+ 1);
1631 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", t
->height
+ 1);
1632 pandecode_prop("depth = MALI_POSITIVE(%" PRId16
")", t
->depth
+ 1);
1633 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16
")", t
->array_size
+ 1);
1634 pandecode_prop("unknown3 = %" PRId16
, t
->unknown3
);
1635 pandecode_prop("unknown3A = %" PRId8
, t
->unknown3A
);
1636 pandecode_prop("nr_mipmap_levels = %" PRId8
, t
->nr_mipmap_levels
);
1638 struct mali_texture_format f
= t
->format
;
1640 pandecode_log(".format = {\n");
1643 pandecode_replay_swizzle(f
.swizzle
);
1644 pandecode_prop("format = %s", pandecode_format_name(f
.format
));
1645 pandecode_prop("type = %s", pandecode_texture_type(f
.type
));
1646 pandecode_prop("srgb = %" PRId32
, f
.srgb
);
1647 pandecode_prop("unknown1 = %" PRId32
, f
.unknown1
);
1648 pandecode_prop("usage2 = 0x%" PRIx32
, f
.usage2
);
1651 pandecode_log("},\n");
1653 pandecode_replay_swizzle(t
->swizzle
);
1655 if (t
->swizzle_zero
) {
1656 /* Shouldn't happen */
1657 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1658 pandecode_prop("swizzle_zero = %d", t
->swizzle_zero
);
1661 pandecode_prop("unknown3 = 0x%" PRIx32
, t
->unknown3
);
1663 pandecode_prop("unknown5 = 0x%" PRIx32
, t
->unknown5
);
1664 pandecode_prop("unknown6 = 0x%" PRIx32
, t
->unknown6
);
1665 pandecode_prop("unknown7 = 0x%" PRIx32
, t
->unknown7
);
1667 pandecode_log(".payload = {\n");
1670 /* A bunch of bitmap pointers follow.
1671 * We work out the correct number,
1672 * based on the mipmap/cubemap
1673 * properties, but dump extra
1674 * possibilities to futureproof */
1676 int bitmap_count
= MALI_NEGATIVE(t
->nr_mipmap_levels
);
1677 bool manual_stride
= f
.usage2
& MALI_TEX_MANUAL_STRIDE
;
1679 /* Miptree for each face */
1680 if (f
.type
== MALI_TEX_CUBE
)
1683 /* Array of textures */
1684 bitmap_count
*= MALI_NEGATIVE(t
->array_size
);
1686 /* Stride for each element */
1690 /* Sanity check the size */
1691 int max_count
= sizeof(t
->payload
) / sizeof(t
->payload
[0]);
1692 assert (bitmap_count
<= max_count
);
1694 /* Dump more to be safe, but not _that_ much more */
1695 int safe_count
= MIN2(bitmap_count
* 2, max_count
);
1697 for (int i
= 0; i
< safe_count
; ++i
) {
1698 char *prefix
= (i
>= bitmap_count
) ? "// " : "";
1700 /* How we dump depends if this is a stride or a pointer */
1702 if ((f
.usage2
& MALI_TEX_MANUAL_STRIDE
) && (i
& 1)) {
1703 /* signed 32-bit snuck in as a 64-bit pointer */
1704 uint64_t stride_set
= t
->payload
[i
];
1705 uint32_t clamped_stride
= stride_set
;
1706 int32_t stride
= clamped_stride
;
1707 assert(stride_set
== clamped_stride
);
1708 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix
, stride
);
1710 char *a
= pointer_as_memory_reference(t
->payload
[i
]);
1711 pandecode_log("%s%s, \n", prefix
, a
);
1717 pandecode_log("},\n");
1720 pandecode_log("};\n");
1726 if (p
->sampler_descriptor
) {
1727 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(p
->sampler_descriptor
);
1730 struct mali_sampler_descriptor
*s
;
1732 mali_ptr d
= p
->sampler_descriptor
;
1734 for (int i
= 0; i
< sampler_count
; ++i
) {
1735 s
= pandecode_fetch_gpu_mem(smem
, d
+ sizeof(*s
) * i
, sizeof(*s
));
1737 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%d_%d = {\n", job_no
, i
);
1740 /* Only the lower two bits are understood right now; the rest we display as hex */
1741 pandecode_log(".filter_mode = MALI_TEX_MIN(%s) | MALI_TEX_MAG(%s) | 0x%" PRIx32
",\n",
1742 MALI_FILTER_NAME(s
->filter_mode
& MALI_TEX_MIN_MASK
),
1743 MALI_FILTER_NAME(s
->filter_mode
& MALI_TEX_MAG_MASK
),
1744 s
->filter_mode
& ~3);
1746 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->min_lod
));
1747 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->max_lod
));
1749 pandecode_prop("wrap_s = %s", pandecode_wrap_mode_name(s
->wrap_s
));
1750 pandecode_prop("wrap_t = %s", pandecode_wrap_mode_name(s
->wrap_t
));
1751 pandecode_prop("wrap_r = %s", pandecode_wrap_mode_name(s
->wrap_r
));
1753 pandecode_prop("compare_func = %s", pandecode_alt_func_name(s
->compare_func
));
1755 if (s
->zero
|| s
->zero2
) {
1756 pandecode_msg("Zero tripped\n");
1757 pandecode_prop("zero = 0x%X, 0x%X\n", s
->zero
, s
->zero2
);
1760 pandecode_prop("seamless_cube_map = %d", s
->seamless_cube_map
);
1762 pandecode_prop("border_color = { %f, %f, %f, %f }",
1766 s
->border_color
[3]);
1769 pandecode_log("};\n");
1776 pandecode_replay_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix
*p
, int job_no
, bool is_bifrost
)
1778 pandecode_log_cont("{\n");
1781 MEMORY_PROP(p
, position_varying
);
1782 DYN_MEMORY_PROP(p
, job_no
, uniform_buffers
);
1783 DYN_MEMORY_PROP(p
, job_no
, texture_trampoline
);
1784 DYN_MEMORY_PROP(p
, job_no
, sampler_descriptor
);
1785 DYN_MEMORY_PROP(p
, job_no
, uniforms
);
1786 DYN_MEMORY_PROP(p
, job_no
, attributes
);
1787 DYN_MEMORY_PROP(p
, job_no
, attribute_meta
);
1788 DYN_MEMORY_PROP(p
, job_no
, varyings
);
1789 DYN_MEMORY_PROP(p
, job_no
, varying_meta
);
1790 DYN_MEMORY_PROP(p
, job_no
, viewport
);
1791 DYN_MEMORY_PROP(p
, job_no
, occlusion_counter
);
1794 pandecode_prop("framebuffer = scratchpad_%d_p", job_no
);
1796 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no
, p
->framebuffer
& MALI_MFBD
? "MALI_MFBD" : "0");
1798 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no
);
1799 pandecode_prop("flags = %d", p
->flags
);
1802 pandecode_log("},\n");
1806 pandecode_replay_vertex_only_bfr(struct bifrost_vertex_only
*v
)
1808 pandecode_log_cont("{\n");
1811 pandecode_prop("unk2 = 0x%x", v
->unk2
);
1813 if (v
->zero0
|| v
->zero1
) {
1814 pandecode_msg("vertex only zero tripped");
1815 pandecode_prop("zero0 = 0x%" PRIx32
, v
->zero0
);
1816 pandecode_prop("zero1 = 0x%" PRIx64
, v
->zero1
);
1820 pandecode_log("}\n");
1824 pandecode_replay_tiler_heap_meta(mali_ptr gpu_va
, int job_no
)
1827 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
1828 const struct bifrost_tiler_heap_meta
*PANDECODE_PTR_VAR(h
, mem
, gpu_va
);
1830 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no
);
1834 pandecode_msg("tiler heap zero tripped\n");
1835 pandecode_prop("zero = 0x%x", h
->zero
);
1838 for (int i
= 0; i
< 12; i
++) {
1839 if (h
->zeros
[i
] != 0) {
1840 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1845 pandecode_prop("heap_size = 0x%x", h
->heap_size
);
1846 MEMORY_PROP(h
, tiler_heap_start
);
1847 MEMORY_PROP(h
, tiler_heap_free
);
1849 /* this might point to the beginning of another buffer, when it's
1850 * really the end of the tiler heap buffer, so we have to be careful
1853 char *a
= pointer_as_memory_reference(h
->tiler_heap_end
- 1);
1854 pandecode_prop("tiler_heap_end = %s + 1", a
);
1858 pandecode_log("};\n");
1862 pandecode_replay_tiler_meta(mali_ptr gpu_va
, int job_no
)
1864 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
1865 const struct bifrost_tiler_meta
*PANDECODE_PTR_VAR(t
, mem
, gpu_va
);
1867 pandecode_replay_tiler_heap_meta(t
->tiler_heap_meta
, job_no
);
1869 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no
);
1872 if (t
->zero0
|| t
->zero1
) {
1873 pandecode_msg("tiler meta zero tripped");
1874 pandecode_prop("zero0 = 0x%" PRIx64
, t
->zero0
);
1875 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
1878 pandecode_prop("hierarchy_mask = 0x%" PRIx16
, t
->hierarchy_mask
);
1879 pandecode_prop("flags = 0x%" PRIx16
, t
->flags
);
1881 pandecode_prop("width = MALI_POSITIVE(%d)", t
->width
+ 1);
1882 pandecode_prop("height = MALI_POSITIVE(%d)", t
->height
+ 1);
1883 DYN_MEMORY_PROP(t
, job_no
, tiler_heap_meta
);
1885 for (int i
= 0; i
< 12; i
++) {
1886 if (t
->zeros
[i
] != 0) {
1887 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64
"\n",
1893 pandecode_log("};\n");
1897 pandecode_replay_gl_enables(uint32_t gl_enables
, int job_type
)
1899 pandecode_log(".gl_enables = ");
1901 pandecode_log_decoded_flags(gl_enable_flag_info
, gl_enables
);
1903 pandecode_log_cont(",\n");
1907 pandecode_replay_primitive_size(union midgard_primitive_size u
, bool constant
)
1909 if (u
.pointer
== 0x0)
1912 pandecode_log(".primitive_size = {\n");
1916 pandecode_prop("constant = %f", u
.constant
);
1918 MEMORY_PROP((&u
), pointer
);
1922 pandecode_log("},\n");
1926 pandecode_replay_tiler_only_bfr(const struct bifrost_tiler_only
*t
, int job_no
)
1928 pandecode_log_cont("{\n");
1931 /* TODO: gl_PointSize on Bifrost */
1932 pandecode_replay_primitive_size(t
->primitive_size
, true);
1934 DYN_MEMORY_PROP(t
, job_no
, tiler_meta
);
1935 pandecode_replay_gl_enables(t
->gl_enables
, JOB_TYPE_TILER
);
1937 if (t
->zero1
|| t
->zero2
|| t
->zero3
|| t
->zero4
|| t
->zero5
1938 || t
->zero6
|| t
->zero7
|| t
->zero8
) {
1939 pandecode_msg("tiler only zero tripped");
1940 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
1941 pandecode_prop("zero2 = 0x%" PRIx64
, t
->zero2
);
1942 pandecode_prop("zero3 = 0x%" PRIx64
, t
->zero3
);
1943 pandecode_prop("zero4 = 0x%" PRIx64
, t
->zero4
);
1944 pandecode_prop("zero5 = 0x%" PRIx64
, t
->zero5
);
1945 pandecode_prop("zero6 = 0x%" PRIx64
, t
->zero6
);
1946 pandecode_prop("zero7 = 0x%" PRIx32
, t
->zero7
);
1947 pandecode_prop("zero8 = 0x%" PRIx64
, t
->zero8
);
1951 pandecode_log("},\n");
1955 pandecode_replay_vertex_job_bfr(const struct mali_job_descriptor_header
*h
,
1956 const struct pandecode_mapped_memory
*mem
,
1957 mali_ptr payload
, int job_no
)
1959 struct bifrost_payload_vertex
*PANDECODE_PTR_VAR(v
, mem
, payload
);
1961 pandecode_replay_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", true);
1963 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no
);
1966 pandecode_log(".prefix = ");
1967 pandecode_replay_vertex_tiler_prefix(&v
->prefix
, job_no
);
1969 pandecode_log(".vertex = ");
1970 pandecode_replay_vertex_only_bfr(&v
->vertex
);
1972 pandecode_log(".postfix = ");
1973 pandecode_replay_vertex_tiler_postfix(&v
->postfix
, job_no
, true);
1976 pandecode_log("};\n");
1982 pandecode_replay_tiler_job_bfr(const struct mali_job_descriptor_header
*h
,
1983 const struct pandecode_mapped_memory
*mem
,
1984 mali_ptr payload
, int job_no
)
1986 struct bifrost_payload_tiler
*PANDECODE_PTR_VAR(t
, mem
, payload
);
1988 pandecode_replay_vertex_tiler_postfix_pre(&t
->postfix
, job_no
, h
->job_type
, "", true);
1990 pandecode_replay_indices(t
->prefix
.indices
, t
->prefix
.index_count
, job_no
);
1991 pandecode_replay_tiler_meta(t
->tiler
.tiler_meta
, job_no
);
1993 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no
);
1996 pandecode_log(".prefix = ");
1997 pandecode_replay_vertex_tiler_prefix(&t
->prefix
, job_no
);
1999 pandecode_log(".tiler = ");
2000 pandecode_replay_tiler_only_bfr(&t
->tiler
, job_no
);
2002 pandecode_log(".postfix = ");
2003 pandecode_replay_vertex_tiler_postfix(&t
->postfix
, job_no
, true);
2006 pandecode_log("};\n");
2012 pandecode_replay_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header
*h
,
2013 const struct pandecode_mapped_memory
*mem
,
2014 mali_ptr payload
, int job_no
)
2016 struct midgard_payload_vertex_tiler
*PANDECODE_PTR_VAR(v
, mem
, payload
);
2018 pandecode_replay_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", false);
2020 pandecode_replay_indices(v
->prefix
.indices
, v
->prefix
.index_count
, job_no
);
2022 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no
);
2025 bool has_primitive_pointer
= v
->prefix
.unknown_draw
& MALI_DRAW_VARYING_SIZE
;
2026 pandecode_replay_primitive_size(v
->primitive_size
, !has_primitive_pointer
);
2028 pandecode_log(".prefix = ");
2029 pandecode_replay_vertex_tiler_prefix(&v
->prefix
, job_no
);
2031 pandecode_replay_gl_enables(v
->gl_enables
, h
->job_type
);
2034 pandecode_prop("draw_start = %d", v
->draw_start
);
2039 pandecode_msg("Zero tripped\n");
2040 pandecode_prop("zero3 = 0x%" PRIx32
, v
->zero3
);
2046 pandecode_msg("Zero tripped\n");
2047 pandecode_prop("zero5 = 0x%" PRIx64
, v
->zero5
);
2050 pandecode_log(".postfix = ");
2051 pandecode_replay_vertex_tiler_postfix(&v
->postfix
, job_no
, false);
2054 pandecode_log("};\n");
2060 pandecode_replay_fragment_job(const struct pandecode_mapped_memory
*mem
,
2061 mali_ptr payload
, int job_no
,
2064 const struct mali_payload_fragment
*PANDECODE_PTR_VAR(s
, mem
, payload
);
2066 bool fbd_dumped
= false;
2068 if (!is_bifrost
&& (s
->framebuffer
& FBD_TYPE
) == MALI_SFBD
) {
2069 /* Only SFBDs are understood, not MFBDs. We're speculating,
2070 * based on the versioning, kernel code, etc, that the
2071 * difference is between Single FrameBuffer Descriptor and
2072 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2073 * up with multi-framebuffer support being added (T7xx onwards,
2074 * including Gxx). In any event, there's some field shuffling
2075 * that we haven't looked into yet. */
2077 pandecode_replay_sfbd(s
->framebuffer
& FBD_MASK
, job_no
);
2079 } else if ((s
->framebuffer
& FBD_TYPE
) == MALI_MFBD
) {
2080 /* We don't know if Bifrost supports SFBD's at all, since the
2081 * driver never uses them. And the format is different from
2082 * Midgard anyways, due to the tiler heap and scratchpad being
2083 * moved out into separate structures, so it's not clear what a
2084 * Bifrost SFBD would even look like without getting an actual
2085 * trace, which appears impossible.
2088 pandecode_replay_mfbd_bfr(s
->framebuffer
& FBD_MASK
, job_no
, true);
2092 uintptr_t p
= (uintptr_t) s
->framebuffer
& FBD_MASK
;
2094 pandecode_log("struct mali_payload_fragment payload_%d = {\n", job_no
);
2097 /* See the comments by the macro definitions for mathematical context
2098 * on why this is so weird */
2100 if (MALI_TILE_COORD_FLAGS(s
->max_tile_coord
) || MALI_TILE_COORD_FLAGS(s
->min_tile_coord
))
2101 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2103 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2104 MALI_TILE_COORD_X(s
->min_tile_coord
) << MALI_TILE_SHIFT
,
2105 MALI_TILE_COORD_Y(s
->min_tile_coord
) << MALI_TILE_SHIFT
);
2107 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2108 (MALI_TILE_COORD_X(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
,
2109 (MALI_TILE_COORD_Y(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
);
2111 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2112 * we have to fallback on offsets. */
2114 const char *fbd_type
= s
->framebuffer
& MALI_MFBD
? "MALI_MFBD" : "MALI_SFBD";
2117 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no
, fbd_type
);
2119 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p
), fbd_type
);
2122 pandecode_log("};\n");
2127 static int job_descriptor_number
= 0;
2130 pandecode_replay_jc(mali_ptr jc_gpu_va
, bool bifrost
)
2132 struct mali_job_descriptor_header
*h
;
2134 int start_number
= 0;
2140 struct pandecode_mapped_memory
*mem
=
2141 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va
);
2145 h
= PANDECODE_PTR(mem
, jc_gpu_va
, struct mali_job_descriptor_header
);
2147 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2148 * high 32-bits of the 64-bit pointer are reused to store
2151 int offset
= h
->job_descriptor_size
== MALI_JOB_32
&&
2152 h
->job_type
!= JOB_TYPE_FRAGMENT
? 4 : 0;
2153 mali_ptr payload_ptr
= jc_gpu_va
+ sizeof(*h
) - offset
;
2155 payload
= pandecode_fetch_gpu_mem(mem
, payload_ptr
,
2158 int job_no
= job_descriptor_number
++;
2161 start_number
= job_no
;
2163 pandecode_log("struct mali_job_descriptor_header job_%d = {\n", job_no
);
2166 pandecode_prop("job_type = %s", pandecode_job_type_name(h
->job_type
));
2168 /* Save for next job fixing */
2169 last_size
= h
->job_descriptor_size
;
2171 if (h
->job_descriptor_size
)
2172 pandecode_prop("job_descriptor_size = %d", h
->job_descriptor_size
);
2174 if (h
->exception_status
!= 0x1)
2175 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2176 h
->exception_status
,
2177 (h
->exception_status
>> 16) & 0xFFFF,
2178 (h
->exception_status
>> 8) & 0x3,
2179 h
->exception_status
& 0xFF);
2181 if (h
->first_incomplete_task
)
2182 pandecode_prop("first_incomplete_task = %d", h
->first_incomplete_task
);
2184 if (h
->fault_pointer
)
2185 pandecode_prop("fault_pointer = 0x%" PRIx64
, h
->fault_pointer
);
2188 pandecode_prop("job_barrier = %d", h
->job_barrier
);
2190 pandecode_prop("job_index = %d", h
->job_index
);
2192 if (h
->unknown_flags
)
2193 pandecode_prop("unknown_flags = %d", h
->unknown_flags
);
2195 if (h
->job_dependency_index_1
)
2196 pandecode_prop("job_dependency_index_1 = %d", h
->job_dependency_index_1
);
2198 if (h
->job_dependency_index_2
)
2199 pandecode_prop("job_dependency_index_2 = %d", h
->job_dependency_index_2
);
2202 pandecode_log("};\n");
2204 /* Do not touch the field yet -- decode the payload first, and
2205 * don't touch that either. This is essential for the uploads
2206 * to occur in sequence and therefore be dynamically allocated
2207 * correctly. Do note the size, however, for that related
2210 switch (h
->job_type
) {
2211 case JOB_TYPE_SET_VALUE
: {
2212 struct mali_payload_set_value
*s
= payload
;
2214 pandecode_log("struct mali_payload_set_value payload_%d = {\n", job_no
);
2216 MEMORY_PROP(s
, out
);
2217 pandecode_prop("unknown = 0x%" PRIX64
, s
->unknown
);
2219 pandecode_log("};\n");
2224 case JOB_TYPE_TILER
:
2225 case JOB_TYPE_VERTEX
:
2226 case JOB_TYPE_COMPUTE
:
2228 if (h
->job_type
== JOB_TYPE_TILER
)
2229 pandecode_replay_tiler_job_bfr(h
, mem
, payload_ptr
, job_no
);
2231 pandecode_replay_vertex_job_bfr(h
, mem
, payload_ptr
, job_no
);
2233 pandecode_replay_vertex_or_tiler_job_mdg(h
, mem
, payload_ptr
, job_no
);
2237 case JOB_TYPE_FRAGMENT
:
2238 pandecode_replay_fragment_job(mem
, payload_ptr
, job_no
, bifrost
);
2245 /* Handle linkage */
2248 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no
- 1);
2251 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no
);
2253 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no
);
2258 } while ((jc_gpu_va
= h
->job_descriptor_size
? h
->next_job_64
: h
->next_job_32
));
2260 return start_number
;