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/macros.h"
34 #include "util/u_math.h"
36 #include "pan_pretty_print.h"
37 #include "midgard/disassemble.h"
38 #include "bifrost/disassemble.h"
40 int pandecode_jc(mali_ptr jc_gpu_va
, bool bifrost
);
42 #define MEMORY_PROP(obj, p) {\
44 char *a = pointer_as_memory_reference(obj->p); \
45 pandecode_prop("%s = %s", #p, a); \
50 #define MEMORY_PROP_DIR(obj, p) {\
52 char *a = pointer_as_memory_reference(obj.p); \
53 pandecode_prop("%s = %s", #p, a); \
58 /* Semantic logging type.
60 * Raw: for raw messages to be printed as is.
61 * Message: for helpful information to be commented out in replays.
62 * Property: for properties of a struct
64 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
67 enum pandecode_log_type
{
73 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
74 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
75 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
77 unsigned pandecode_indent
= 0;
80 pandecode_make_indent(void)
82 for (unsigned i
= 0; i
< pandecode_indent
; ++i
)
87 pandecode_log_typed(enum pandecode_log_type type
, const char *format
, ...)
91 pandecode_make_indent();
93 if (type
== PANDECODE_MESSAGE
)
95 else if (type
== PANDECODE_PROPERTY
)
102 if (type
== PANDECODE_PROPERTY
)
107 pandecode_log_cont(const char *format
, ...)
111 va_start(ap
, format
);
116 struct pandecode_flag_info
{
122 pandecode_log_decoded_flags(const struct pandecode_flag_info
*flag_info
,
125 bool decodable_flags_found
= false;
127 for (int i
= 0; flag_info
[i
].name
; i
++) {
128 if ((flags
& flag_info
[i
].flag
) != flag_info
[i
].flag
)
131 if (!decodable_flags_found
) {
132 decodable_flags_found
= true;
134 pandecode_log_cont(" | ");
137 pandecode_log_cont("%s", flag_info
[i
].name
);
139 flags
&= ~flag_info
[i
].flag
;
142 if (decodable_flags_found
) {
144 pandecode_log_cont(" | 0x%" PRIx64
, flags
);
146 pandecode_log_cont("0x%" PRIx64
, flags
);
150 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
151 static const struct pandecode_flag_info gl_enable_flag_info
[] = {
152 FLAG_INFO(OCCLUSION_QUERY
),
153 FLAG_INFO(OCCLUSION_PRECISE
),
154 FLAG_INFO(FRONT_CCW_TOP
),
155 FLAG_INFO(CULL_FACE_FRONT
),
156 FLAG_INFO(CULL_FACE_BACK
),
161 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
162 static const struct pandecode_flag_info clear_flag_info
[] = {
165 FLAG_INFO(SLOW_STENCIL
),
170 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
171 static const struct pandecode_flag_info mask_flag_info
[] = {
180 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
181 static const struct pandecode_flag_info u3_flag_info
[] = {
183 FLAG_INFO(CAN_DISCARD
),
184 FLAG_INFO(HAS_BLEND_SHADER
),
185 FLAG_INFO(DEPTH_TEST
),
189 static const struct pandecode_flag_info u4_flag_info
[] = {
191 FLAG_INFO(NO_DITHER
),
192 FLAG_INFO(DEPTH_RANGE_A
),
193 FLAG_INFO(DEPTH_RANGE_B
),
194 FLAG_INFO(STENCIL_TEST
),
195 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER
),
200 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
201 static const struct pandecode_flag_info fb_fmt_flag_info
[] = {
209 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
210 static const struct pandecode_flag_info mfbd_fmt_flag_info
[] = {
217 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
218 static const struct pandecode_flag_info mfbd_extra_flag_info
[] = {
226 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
227 static const struct pandecode_flag_info shader_midgard1_flag_info
[] = {
229 FLAG_INFO(HELPER_INVOCATIONS
),
230 FLAG_INFO(READS_TILEBUFFER
),
236 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
237 static const struct pandecode_flag_info mfbd_flag_info
[] = {
238 FLAG_INFO(DEPTH_WRITE
),
244 #define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
245 static const struct pandecode_flag_info sampler_flag_info
[] = {
246 FLAG_INFO(MAG_NEAREST
),
247 FLAG_INFO(MIN_NEAREST
),
248 FLAG_INFO(MIP_LINEAR_1
),
249 FLAG_INFO(MIP_LINEAR_2
),
250 FLAG_INFO(NORM_COORDS
),
255 extern char *replace_fragment
;
256 extern char *replace_vertex
;
259 pandecode_job_type(enum mali_job_type type
)
261 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
265 DEFINE_CASE(SET_VALUE
);
266 DEFINE_CASE(CACHE_FLUSH
);
267 DEFINE_CASE(COMPUTE
);
271 DEFINE_CASE(FRAGMENT
);
273 case JOB_NOT_STARTED
:
274 return "NOT_STARTED";
277 pandecode_log("Warning! Unknown job type %x\n", type
);
285 pandecode_draw_mode(enum mali_draw_mode mode
)
287 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
290 DEFINE_CASE(DRAW_NONE
);
293 DEFINE_CASE(TRIANGLES
);
294 DEFINE_CASE(TRIANGLE_STRIP
);
295 DEFINE_CASE(TRIANGLE_FAN
);
296 DEFINE_CASE(LINE_STRIP
);
297 DEFINE_CASE(LINE_LOOP
);
298 DEFINE_CASE(POLYGON
);
300 DEFINE_CASE(QUAD_STRIP
);
303 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
309 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
311 pandecode_func(enum mali_func mode
)
318 DEFINE_CASE(GREATER
);
319 DEFINE_CASE(NOTEQUAL
);
324 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
329 /* Why is this duplicated? Who knows... */
330 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
332 pandecode_alt_func(enum mali_alt_func mode
)
339 DEFINE_CASE(GREATER
);
340 DEFINE_CASE(NOTEQUAL
);
345 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
350 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
352 pandecode_stencil_op(enum mali_stencil_op op
)
356 DEFINE_CASE(REPLACE
);
359 DEFINE_CASE(INCR_WRAP
);
360 DEFINE_CASE(DECR_WRAP
);
365 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
371 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
372 static char *pandecode_attr_mode(enum mali_attr_mode mode
)
377 DEFINE_CASE(POT_DIVIDE
);
379 DEFINE_CASE(NPOT_DIVIDE
);
381 DEFINE_CASE(INTERNAL
);
383 return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
389 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
391 pandecode_channel(enum mali_channel channel
)
400 DEFINE_CASE(RESERVED_0
);
401 DEFINE_CASE(RESERVED_1
);
404 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
409 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
411 pandecode_wrap_mode(enum mali_wrap_mode op
)
415 DEFINE_CASE(CLAMP_TO_EDGE
);
416 DEFINE_CASE(CLAMP_TO_BORDER
);
417 DEFINE_CASE(MIRRORED_REPEAT
);
420 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
425 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
427 pandecode_texture_type(enum mali_texture_type type
)
436 unreachable("Unknown case");
441 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
443 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt
)
447 DEFINE_CASE(UNKNOWN
);
452 unreachable("Invalid case");
457 #define DEFINE_CASE(name) case MALI_EXCEPTION_ACCESS_## name: return ""#name
459 pandecode_exception_access(enum mali_exception_access fmt
)
463 DEFINE_CASE(EXECUTE
);
468 unreachable("Invalid case");
473 /* Midgard's tiler descriptor is embedded within the
477 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor
*t
)
479 pandecode_log(".tiler = {\n");
482 pandecode_prop("hierarchy_mask = 0x%" PRIx16
, t
->hierarchy_mask
);
483 pandecode_prop("flags = 0x%" PRIx16
, t
->flags
);
484 pandecode_prop("polygon_list_size = 0x%x", t
->polygon_list_size
);
486 MEMORY_PROP(t
, polygon_list
);
487 MEMORY_PROP(t
, polygon_list_body
);
489 MEMORY_PROP(t
, heap_start
);
491 if (t
->heap_start
== t
->heap_end
) {
492 /* Print identically to show symmetry for empty tiler heaps */
493 MEMORY_PROP(t
, heap_end
);
495 /* Points to the end of a buffer */
496 char *a
= pointer_as_memory_reference(t
->heap_end
- 1);
497 pandecode_prop("heap_end = %s + 1", a
);
501 bool nonzero_weights
= false;
503 for (unsigned w
= 0; w
< ARRAY_SIZE(t
->weights
); ++w
) {
504 nonzero_weights
|= t
->weights
[w
] != 0x0;
507 if (nonzero_weights
) {
508 pandecode_log(".weights = {");
510 for (unsigned w
= 0; w
< ARRAY_SIZE(t
->weights
); ++w
) {
511 pandecode_log("%d, ", t
->weights
[w
]);
518 pandecode_log("}\n");
522 pandecode_sfbd(uint64_t gpu_va
, int job_no
)
524 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
525 const struct mali_single_framebuffer
*PANDECODE_PTR_VAR(s
, mem
, (mali_ptr
) gpu_va
);
527 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
530 pandecode_prop("unknown1 = 0x%" PRIx32
, s
->unknown1
);
531 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->unknown2
);
533 pandecode_log(".format = ");
534 pandecode_log_decoded_flags(fb_fmt_flag_info
, s
->format
);
535 pandecode_log_cont(",\n");
537 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", s
->width
+ 1);
538 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", s
->height
+ 1);
540 MEMORY_PROP(s
, framebuffer
);
541 pandecode_prop("stride = %d", s
->stride
);
543 /* Earlier in the actual commandstream -- right before width -- but we
544 * delay to flow nicer */
546 pandecode_log(".clear_flags = ");
547 pandecode_log_decoded_flags(clear_flag_info
, s
->clear_flags
);
548 pandecode_log_cont(",\n");
550 if (s
->depth_buffer
| s
->depth_buffer_enable
) {
551 MEMORY_PROP(s
, depth_buffer
);
552 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s
->depth_buffer_enable
));
555 if (s
->stencil_buffer
| s
->stencil_buffer_enable
) {
556 MEMORY_PROP(s
, stencil_buffer
);
557 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s
->stencil_buffer_enable
));
560 if (s
->clear_color_1
| s
->clear_color_2
| s
->clear_color_3
| s
->clear_color_4
) {
561 pandecode_prop("clear_color_1 = 0x%" PRIx32
, s
->clear_color_1
);
562 pandecode_prop("clear_color_2 = 0x%" PRIx32
, s
->clear_color_2
);
563 pandecode_prop("clear_color_3 = 0x%" PRIx32
, s
->clear_color_3
);
564 pandecode_prop("clear_color_4 = 0x%" PRIx32
, s
->clear_color_4
);
567 if (s
->clear_depth_1
!= 0 || s
->clear_depth_2
!= 0 || s
->clear_depth_3
!= 0 || s
->clear_depth_4
!= 0) {
568 pandecode_prop("clear_depth_1 = %f", s
->clear_depth_1
);
569 pandecode_prop("clear_depth_2 = %f", s
->clear_depth_2
);
570 pandecode_prop("clear_depth_3 = %f", s
->clear_depth_3
);
571 pandecode_prop("clear_depth_4 = %f", s
->clear_depth_4
);
574 if (s
->clear_stencil
) {
575 pandecode_prop("clear_stencil = 0x%x", s
->clear_stencil
);
578 MEMORY_PROP(s
, unknown_address_0
);
579 const struct midgard_tiler_descriptor t
= s
->tiler
;
580 pandecode_midgard_tiler_descriptor(&t
);
583 pandecode_log("};\n");
585 pandecode_prop("zero0 = 0x%" PRIx64
, s
->zero0
);
586 pandecode_prop("zero1 = 0x%" PRIx64
, s
->zero1
);
587 pandecode_prop("zero2 = 0x%" PRIx32
, s
->zero2
);
588 pandecode_prop("zero4 = 0x%" PRIx32
, s
->zero4
);
590 printf(".zero3 = {");
592 for (int i
= 0; i
< sizeof(s
->zero3
) / sizeof(s
->zero3
[0]); ++i
)
593 printf("%X, ", s
->zero3
[i
]);
597 printf(".zero6 = {");
599 for (int i
= 0; i
< sizeof(s
->zero6
) / sizeof(s
->zero6
[0]); ++i
)
600 printf("%X, ", s
->zero6
[i
]);
606 pandecode_u32_slide(unsigned name
, const u32
*slide
, unsigned count
)
608 pandecode_log(".unknown%d = {", name
);
610 for (int i
= 0; i
< count
; ++i
)
611 printf("%X, ", slide
[i
]);
613 pandecode_log("},\n");
616 #define SHORT_SLIDE(num) \
617 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
620 pandecode_compute_fbd(uint64_t gpu_va
, int job_no
)
622 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
623 const struct mali_compute_fbd
*PANDECODE_PTR_VAR(s
, mem
, (mali_ptr
) gpu_va
);
625 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
635 pandecode_swizzle(unsigned swizzle
)
637 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
638 pandecode_channel((swizzle
>> 0) & 0x7),
639 pandecode_channel((swizzle
>> 3) & 0x7),
640 pandecode_channel((swizzle
>> 6) & 0x7),
641 pandecode_channel((swizzle
>> 9) & 0x7));
645 pandecode_rt_format(struct mali_rt_format format
)
647 pandecode_log(".format = {\n");
650 pandecode_prop("unk1 = 0x%" PRIx32
, format
.unk1
);
651 pandecode_prop("unk2 = 0x%" PRIx32
, format
.unk2
);
652 pandecode_prop("unk3 = 0x%" PRIx32
, format
.unk3
);
654 pandecode_prop("block = %s",
655 pandecode_mfbd_block_format(format
.block
));
657 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
658 MALI_NEGATIVE(format
.nr_channels
));
660 pandecode_log(".flags = ");
661 pandecode_log_decoded_flags(mfbd_fmt_flag_info
, format
.flags
);
662 pandecode_log_cont(",\n");
664 pandecode_swizzle(format
.swizzle
);
666 pandecode_prop("no_preload = 0x%" PRIx32
, format
.no_preload
);
669 pandecode_prop("zero = 0x%" PRIx32
, format
.zero
);
672 pandecode_log("},\n");
676 pandecode_render_target(uint64_t gpu_va
, unsigned job_no
, const struct bifrost_framebuffer
*fb
)
678 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64
"_%d[] = {\n", gpu_va
, job_no
);
681 for (int i
= 0; i
< MALI_NEGATIVE(fb
->rt_count_1
); i
++) {
682 mali_ptr rt_va
= gpu_va
+ i
* sizeof(struct bifrost_render_target
);
683 struct pandecode_mapped_memory
*mem
=
684 pandecode_find_mapped_gpu_mem_containing(rt_va
);
685 const struct bifrost_render_target
*PANDECODE_PTR_VAR(rt
, mem
, (mali_ptr
) rt_va
);
687 pandecode_log("{\n");
690 pandecode_rt_format(rt
->format
);
692 if (rt
->format
.block
== MALI_MFBD_BLOCK_AFBC
) {
693 pandecode_log(".afbc = {\n");
696 char *a
= pointer_as_memory_reference(rt
->afbc
.metadata
);
697 pandecode_prop("metadata = %s", a
);
700 pandecode_prop("stride = %d", rt
->afbc
.stride
);
701 pandecode_prop("unk = 0x%" PRIx32
, rt
->afbc
.unk
);
704 pandecode_log("},\n");
706 pandecode_log(".chunknown = {\n");
709 pandecode_prop("unk = 0x%" PRIx64
, rt
->chunknown
.unk
);
711 char *a
= pointer_as_memory_reference(rt
->chunknown
.pointer
);
712 pandecode_prop("pointer = %s", a
);
716 pandecode_log("},\n");
719 MEMORY_PROP(rt
, framebuffer
);
720 pandecode_prop("framebuffer_stride = %d", rt
->framebuffer_stride
);
722 if (rt
->clear_color_1
| rt
->clear_color_2
| rt
->clear_color_3
| rt
->clear_color_4
) {
723 pandecode_prop("clear_color_1 = 0x%" PRIx32
, rt
->clear_color_1
);
724 pandecode_prop("clear_color_2 = 0x%" PRIx32
, rt
->clear_color_2
);
725 pandecode_prop("clear_color_3 = 0x%" PRIx32
, rt
->clear_color_3
);
726 pandecode_prop("clear_color_4 = 0x%" PRIx32
, rt
->clear_color_4
);
729 if (rt
->zero1
|| rt
->zero2
|| rt
->zero3
) {
730 pandecode_msg("render target zeros tripped\n");
731 pandecode_prop("zero1 = 0x%" PRIx64
, rt
->zero1
);
732 pandecode_prop("zero2 = 0x%" PRIx32
, rt
->zero2
);
733 pandecode_prop("zero3 = 0x%" PRIx32
, rt
->zero3
);
737 pandecode_log("},\n");
741 pandecode_log("};\n");
745 pandecode_mfbd_bfr(uint64_t gpu_va
, int job_no
, bool with_render_targets
)
747 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
748 const struct bifrost_framebuffer
*PANDECODE_PTR_VAR(fb
, mem
, (mali_ptr
) gpu_va
);
750 if (fb
->sample_locations
) {
751 /* The blob stores all possible sample locations in a single buffer
752 * allocated on startup, and just switches the pointer when switching
753 * MSAA state. For now, we just put the data into the cmdstream, but we
754 * should do something like what the blob does with a real driver.
756 * There seem to be 32 slots for sample locations, followed by another
757 * 16. The second 16 is just the center location followed by 15 zeros
758 * in all the cases I've identified (maybe shader vs. depth/color
762 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(fb
->sample_locations
);
764 const u16
*PANDECODE_PTR_VAR(samples
, smem
, fb
->sample_locations
);
766 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no
);
769 for (int i
= 0; i
< 32 + 16; i
++) {
770 pandecode_log("%d, %d,\n", samples
[2 * i
], samples
[2 * i
+ 1]);
774 pandecode_log("};\n");
777 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
780 pandecode_prop("unk0 = 0x%x", fb
->unk0
);
782 if (fb
->sample_locations
)
783 pandecode_prop("sample_locations = sample_locations_%d", job_no
);
785 /* Assume that unknown1 was emitted in the last job for
787 MEMORY_PROP(fb
, unknown1
);
789 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb
->width1
+ 1);
790 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb
->height1
+ 1);
791 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb
->width2
+ 1);
792 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb
->height2
+ 1);
794 pandecode_prop("unk1 = 0x%x", fb
->unk1
);
795 pandecode_prop("unk2 = 0x%x", fb
->unk2
);
796 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb
->rt_count_1
+ 1);
797 pandecode_prop("rt_count_2 = %d", fb
->rt_count_2
);
799 pandecode_log(".mfbd_flags = ");
800 pandecode_log_decoded_flags(mfbd_flag_info
, fb
->mfbd_flags
);
801 pandecode_log_cont(",\n");
803 pandecode_prop("clear_stencil = 0x%x", fb
->clear_stencil
);
804 pandecode_prop("clear_depth = %f", fb
->clear_depth
);
806 pandecode_prop("unknown2 = 0x%x", fb
->unknown2
);
807 MEMORY_PROP(fb
, scratchpad
);
808 const struct midgard_tiler_descriptor t
= fb
->tiler
;
809 pandecode_midgard_tiler_descriptor(&t
);
811 if (fb
->zero3
|| fb
->zero4
) {
812 pandecode_msg("framebuffer zeros tripped\n");
813 pandecode_prop("zero3 = 0x%" PRIx32
, fb
->zero3
);
814 pandecode_prop("zero4 = 0x%" PRIx32
, fb
->zero4
);
818 pandecode_log("};\n");
820 gpu_va
+= sizeof(struct bifrost_framebuffer
);
822 if ((fb
->mfbd_flags
& MALI_MFBD_EXTRA
) && with_render_targets
) {
823 mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
824 const struct bifrost_fb_extra
*PANDECODE_PTR_VAR(fbx
, mem
, (mali_ptr
) gpu_va
);
826 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
829 MEMORY_PROP(fbx
, checksum
);
831 if (fbx
->checksum_stride
)
832 pandecode_prop("checksum_stride = %d", fbx
->checksum_stride
);
834 pandecode_log(".flags = ");
835 pandecode_log_decoded_flags(mfbd_extra_flag_info
, fbx
->flags
);
836 pandecode_log_cont(",\n");
838 if (fbx
->flags
& MALI_EXTRA_AFBC_ZS
) {
839 pandecode_log(".ds_afbc = {\n");
842 MEMORY_PROP_DIR(fbx
->ds_afbc
, depth_stencil_afbc_metadata
);
843 pandecode_prop("depth_stencil_afbc_stride = %d",
844 fbx
->ds_afbc
.depth_stencil_afbc_stride
);
845 MEMORY_PROP_DIR(fbx
->ds_afbc
, depth_stencil
);
847 if (fbx
->ds_afbc
.zero1
|| fbx
->ds_afbc
.padding
) {
848 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
849 pandecode_prop("zero1 = 0x%" PRIx32
,
851 pandecode_prop("padding = 0x%" PRIx64
,
852 fbx
->ds_afbc
.padding
);
856 pandecode_log("},\n");
858 pandecode_log(".ds_linear = {\n");
861 if (fbx
->ds_linear
.depth
) {
862 MEMORY_PROP_DIR(fbx
->ds_linear
, depth
);
863 pandecode_prop("depth_stride = %d",
864 fbx
->ds_linear
.depth_stride
);
867 if (fbx
->ds_linear
.stencil
) {
868 MEMORY_PROP_DIR(fbx
->ds_linear
, stencil
);
869 pandecode_prop("stencil_stride = %d",
870 fbx
->ds_linear
.stencil_stride
);
873 if (fbx
->ds_linear
.depth_stride_zero
||
874 fbx
->ds_linear
.stencil_stride_zero
||
875 fbx
->ds_linear
.zero1
|| fbx
->ds_linear
.zero2
) {
876 pandecode_msg("Depth/stencil zeros tripped\n");
877 pandecode_prop("depth_stride_zero = 0x%x",
878 fbx
->ds_linear
.depth_stride_zero
);
879 pandecode_prop("stencil_stride_zero = 0x%x",
880 fbx
->ds_linear
.stencil_stride_zero
);
881 pandecode_prop("zero1 = 0x%" PRIx32
,
882 fbx
->ds_linear
.zero1
);
883 pandecode_prop("zero2 = 0x%" PRIx32
,
884 fbx
->ds_linear
.zero2
);
888 pandecode_log("},\n");
891 if (fbx
->zero3
|| fbx
->zero4
) {
892 pandecode_msg("fb_extra zeros tripped\n");
893 pandecode_prop("zero3 = 0x%" PRIx64
, fbx
->zero3
);
894 pandecode_prop("zero4 = 0x%" PRIx64
, fbx
->zero4
);
898 pandecode_log("};\n");
900 gpu_va
+= sizeof(struct bifrost_fb_extra
);
903 if (with_render_targets
)
904 pandecode_render_target(gpu_va
, job_no
, fb
);
906 /* Passback the render target count */
907 return MALI_NEGATIVE(fb
->rt_count_1
);
910 /* Just add a comment decoding the shift/odd fields forming the padded vertices
914 pandecode_padded_vertices(unsigned shift
, unsigned k
)
916 unsigned odd
= 2*k
+ 1;
917 unsigned pot
= 1 << shift
;
918 pandecode_msg("padded_num_vertices = %d\n", odd
* pot
);
921 /* Given a magic divisor, recover what we were trying to divide by.
923 * Let m represent the magic divisor. By definition, m is an element on Z, whre
924 * 0 <= m < 2^N, for N bits in m.
926 * Let q represent the number we would like to divide by.
928 * By definition of a magic divisor for N-bit unsigned integers (a number you
929 * multiply by to magically get division), m is a number such that:
931 * (m * x) & (2^N - 1) = floor(x/q).
932 * for all x on Z where 0 <= x < 2^N
934 * Ignore the case where any of the above values equals zero; it is irrelevant
935 * for our purposes (instanced arrays).
937 * Choose x = q. Then:
939 * (m * x) & (2^N - 1) = floor(x/q).
940 * (m * q) & (2^N - 1) = floor(q/q).
942 * floor(q/q) = floor(1) = 1, therefore:
944 * (m * q) & (2^N - 1) = 1
946 * Recall the identity that the bitwise AND of one less than a power-of-two
947 * equals the modulo with that power of two, i.e. for all x:
949 * x & (2^N - 1) = x % N
955 * By definition, a modular multiplicative inverse of a number m is the number
956 * q such that with respect to a modulos M:
960 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
965 pandecode_magic_divisor(uint32_t magic
, unsigned shift
, unsigned orig_divisor
, unsigned extra
)
968 /* Compute the modular inverse of `magic` with respect to 2^(32 -
969 * shift) the most lame way possible... just repeatedly add.
970 * Asymptoptically slow but nobody cares in practice, unless you have
971 * massive numbers of vertices or high divisors. */
973 unsigned inverse
= 0;
975 /* Magic implicitly has the highest bit set */
978 /* Depending on rounding direction */
983 uint32_t product
= magic
* inverse
;
995 pandecode_msg("dividing by %d (maybe off by two)\n", inverse
);
997 /* Recall we're supposed to divide by (gl_level_divisor *
998 * padded_num_vertices) */
1000 unsigned padded_num_vertices
= inverse
/ orig_divisor
;
1002 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices
);
1007 pandecode_attributes(const struct pandecode_mapped_memory
*mem
,
1008 mali_ptr addr
, int job_no
, char *suffix
,
1009 int count
, bool varying
)
1011 char *prefix
= varying
? "varyings" : "attributes";
1014 pandecode_msg("no %s\n", prefix
);
1018 union mali_attr
*attr
= pandecode_fetch_gpu_mem(mem
, addr
, sizeof(union mali_attr
) * count
);
1021 snprintf(base
, sizeof(base
), "%s_data_%d%s", prefix
, job_no
, suffix
);
1023 for (int i
= 0; i
< count
; ++i
) {
1024 enum mali_attr_mode mode
= attr
[i
].elements
& 7;
1026 if (mode
== MALI_ATTR_UNUSED
)
1029 mali_ptr raw_elements
= attr
[i
].elements
& ~7;
1031 /* TODO: Do we maybe want to dump the attribute values
1032 * themselves given the specified format? Or is that too hard?
1035 char *a
= pointer_as_memory_reference(raw_elements
);
1036 pandecode_log("mali_ptr %s_%d_p = %s;\n", base
, i
, a
);
1040 pandecode_log("union mali_attr %s_%d[] = {\n", prefix
, job_no
);
1043 for (int i
= 0; i
< count
; ++i
) {
1044 pandecode_log("{\n");
1047 unsigned mode
= attr
[i
].elements
& 7;
1048 pandecode_prop("elements = (%s_%d_p) | %s", base
, i
, pandecode_attr_mode(mode
));
1049 pandecode_prop("shift = %d", attr
[i
].shift
);
1050 pandecode_prop("extra_flags = %d", attr
[i
].extra_flags
);
1051 pandecode_prop("stride = 0x%" PRIx32
, attr
[i
].stride
);
1052 pandecode_prop("size = 0x%" PRIx32
, attr
[i
].size
);
1054 /* Decode further where possible */
1056 if (mode
== MALI_ATTR_MODULO
) {
1057 pandecode_padded_vertices(
1059 attr
[i
].extra_flags
);
1063 pandecode_log("}, \n");
1065 if (mode
== MALI_ATTR_NPOT_DIVIDE
) {
1067 pandecode_log("{\n");
1069 pandecode_prop("unk = 0x%x", attr
[i
].unk
);
1070 pandecode_prop("magic_divisor = 0x%08x", attr
[i
].magic_divisor
);
1071 if (attr
[i
].zero
!= 0)
1072 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr
[i
].zero
);
1073 pandecode_prop("divisor = %d", attr
[i
].divisor
);
1074 pandecode_magic_divisor(attr
[i
].magic_divisor
, attr
[i
- 1].shift
, attr
[i
].divisor
, attr
[i
- 1].extra_flags
);
1076 pandecode_log("}, \n");
1082 pandecode_log("};\n");
1086 pandecode_shader_address(const char *name
, mali_ptr ptr
)
1088 /* TODO: Decode flags */
1089 mali_ptr shader_ptr
= ptr
& ~15;
1091 char *a
= pointer_as_memory_reference(shader_ptr
);
1092 pandecode_prop("%s = (%s) | %d", name
, a
, (int) (ptr
& 15));
1099 all_zero(unsigned *buffer
, unsigned count
)
1101 for (unsigned i
= 0; i
< count
; ++i
) {
1110 pandecode_stencil(const char *name
, const struct mali_stencil_test
*stencil
)
1112 if (all_zero((unsigned *) stencil
, sizeof(stencil
) / sizeof(unsigned)))
1115 const char *func
= pandecode_func(stencil
->func
);
1116 const char *sfail
= pandecode_stencil_op(stencil
->sfail
);
1117 const char *dpfail
= pandecode_stencil_op(stencil
->dpfail
);
1118 const char *dppass
= pandecode_stencil_op(stencil
->dppass
);
1121 pandecode_msg("Stencil zero tripped: %X\n", stencil
->zero
);
1123 pandecode_log(".stencil_%s = {\n", name
);
1125 pandecode_prop("ref = %d", stencil
->ref
);
1126 pandecode_prop("mask = 0x%02X", stencil
->mask
);
1127 pandecode_prop("func = %s", func
);
1128 pandecode_prop("sfail = %s", sfail
);
1129 pandecode_prop("dpfail = %s", dpfail
);
1130 pandecode_prop("dppass = %s", dppass
);
1132 pandecode_log("},\n");
1136 pandecode_blend_equation(const struct mali_blend_equation
*blend
)
1139 pandecode_msg("Blend zero tripped: %X\n", blend
->zero1
);
1141 pandecode_log(".equation = {\n");
1144 pandecode_prop("rgb_mode = 0x%X", blend
->rgb_mode
);
1145 pandecode_prop("alpha_mode = 0x%X", blend
->alpha_mode
);
1147 pandecode_log(".color_mask = ");
1148 pandecode_log_decoded_flags(mask_flag_info
, blend
->color_mask
);
1149 pandecode_log_cont(",\n");
1152 pandecode_log("},\n");
1155 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1158 decode_bifrost_constant(u16 constant
)
1160 float lo
= (float) (constant
& 0xFF);
1161 float hi
= (float) (constant
>> 8);
1163 return (hi
/ 255.0) + (lo
/ 65535.0);
1167 pandecode_bifrost_blend(void *descs
, int job_no
, int rt_no
)
1169 struct bifrost_blend_rt
*b
=
1170 ((struct bifrost_blend_rt
*) descs
) + rt_no
;
1172 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1175 pandecode_prop("flags = 0x%" PRIx16
, b
->flags
);
1176 pandecode_prop("constant = 0x%" PRIx8
" /* %f */",
1177 b
->constant
, decode_bifrost_constant(b
->constant
));
1179 /* TODO figure out blend shader enable bit */
1180 pandecode_blend_equation(&b
->equation
);
1181 pandecode_prop("unk2 = 0x%" PRIx16
, b
->unk2
);
1182 pandecode_prop("index = 0x%" PRIx16
, b
->index
);
1183 pandecode_prop("shader = 0x%" PRIx32
, b
->shader
);
1186 pandecode_log("},\n");
1192 pandecode_midgard_blend(union midgard_blend
*blend
, bool is_shader
)
1194 if (all_zero((unsigned *) blend
, sizeof(blend
) / sizeof(unsigned)))
1197 pandecode_log(".blend = {\n");
1201 pandecode_shader_address("shader", blend
->shader
);
1203 pandecode_blend_equation(&blend
->equation
);
1204 pandecode_prop("constant = %f", blend
->constant
);
1208 pandecode_log("},\n");
1210 /* Return blend shader to disassemble if present */
1211 return is_shader
? (blend
->shader
& ~0xF) : 0;
1215 pandecode_midgard_blend_mrt(void *descs
, int job_no
, int rt_no
)
1217 struct midgard_blend_rt
*b
=
1218 ((struct midgard_blend_rt
*) descs
) + rt_no
;
1220 /* Flags determine presence of blend shader */
1221 bool is_shader
= (b
->flags
& 0xF) >= 0x2;
1223 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1226 pandecode_prop("flags = 0x%" PRIx64
, b
->flags
);
1228 union midgard_blend blend
= b
->blend
;
1229 mali_ptr shader
= pandecode_midgard_blend(&blend
, is_shader
);
1232 pandecode_log("};\n");
1238 pandecode_attribute_meta(int job_no
, int count
, const struct mali_vertex_tiler_postfix
*v
, bool varying
, char *suffix
)
1241 char *prefix
= varying
? "varying" : "attribute";
1242 unsigned max_index
= 0;
1243 snprintf(base
, sizeof(base
), "%s_meta", prefix
);
1245 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base
, job_no
, suffix
);
1248 struct mali_attr_meta
*attr_meta
;
1249 mali_ptr p
= varying
? (v
->varying_meta
& ~0xF) : v
->attribute_meta
;
1251 struct pandecode_mapped_memory
*attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
);
1253 for (int i
= 0; i
< count
; ++i
, p
+= sizeof(struct mali_attr_meta
)) {
1254 attr_meta
= pandecode_fetch_gpu_mem(attr_mem
, p
,
1257 pandecode_log("{\n");
1259 pandecode_prop("index = %d", attr_meta
->index
);
1261 if (attr_meta
->index
> max_index
)
1262 max_index
= attr_meta
->index
;
1263 pandecode_swizzle(attr_meta
->swizzle
);
1264 pandecode_prop("format = %s", pandecode_format(attr_meta
->format
));
1266 pandecode_prop("unknown1 = 0x%" PRIx64
, (u64
) attr_meta
->unknown1
);
1267 pandecode_prop("unknown3 = 0x%" PRIx64
, (u64
) attr_meta
->unknown3
);
1268 pandecode_prop("src_offset = %d", attr_meta
->src_offset
);
1270 pandecode_log("},\n");
1275 pandecode_log("};\n");
1277 return count
? (max_index
+ 1) : 0;
1281 pandecode_indices(uintptr_t pindices
, uint32_t index_count
, int job_no
)
1283 struct pandecode_mapped_memory
*imem
= pandecode_find_mapped_gpu_mem_containing(pindices
);
1286 /* Indices are literally just a u32 array :) */
1288 uint32_t *PANDECODE_PTR_VAR(indices
, imem
, pindices
);
1290 pandecode_log("uint32_t indices_%d[] = {\n", job_no
);
1293 for (unsigned i
= 0; i
< (index_count
+ 1); i
+= 3)
1294 pandecode_log("%d, %d, %d,\n",
1300 pandecode_log("};\n");
1304 /* return bits [lo, hi) of word */
1306 bits(u32 word
, u32 lo
, u32 hi
)
1309 return word
; // avoid undefined behavior with the shift
1311 return (word
>> lo
) & ((1 << (hi
- lo
)) - 1);
1315 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix
*p
, int job_no
)
1317 pandecode_log_cont("{\n");
1320 pandecode_prop("invocation_count = 0x%" PRIx32
, p
->invocation_count
);
1321 pandecode_prop("size_y_shift = %d", p
->size_y_shift
);
1322 pandecode_prop("size_z_shift = %d", p
->size_z_shift
);
1323 pandecode_prop("workgroups_x_shift = %d", p
->workgroups_x_shift
);
1324 pandecode_prop("workgroups_y_shift = %d", p
->workgroups_y_shift
);
1325 pandecode_prop("workgroups_z_shift = %d", p
->workgroups_z_shift
);
1326 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32
, p
->workgroups_x_shift_2
);
1328 /* Decode invocation_count. See the comment before the definition of
1329 * invocation_count for an explanation.
1331 pandecode_msg("size: (%d, %d, %d)\n",
1332 bits(p
->invocation_count
, 0, p
->size_y_shift
) + 1,
1333 bits(p
->invocation_count
, p
->size_y_shift
, p
->size_z_shift
) + 1,
1334 bits(p
->invocation_count
, p
->size_z_shift
,
1335 p
->workgroups_x_shift
) + 1);
1336 pandecode_msg("workgroups: (%d, %d, %d)\n",
1337 bits(p
->invocation_count
, p
->workgroups_x_shift
,
1338 p
->workgroups_y_shift
) + 1,
1339 bits(p
->invocation_count
, p
->workgroups_y_shift
,
1340 p
->workgroups_z_shift
) + 1,
1341 bits(p
->invocation_count
, p
->workgroups_z_shift
,
1345 if (p
->unknown_draw
)
1346 pandecode_prop("unknown_draw = 0x%" PRIx32
, p
->unknown_draw
);
1348 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32
, p
->workgroups_x_shift_3
);
1350 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p
->draw_mode
));
1352 /* Index count only exists for tiler jobs anyway */
1355 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32
")", p
->index_count
+ 1);
1357 if (p
->offset_bias_correction
)
1358 pandecode_prop("offset_bias_correction = %d", p
->offset_bias_correction
);
1361 pandecode_msg("Zero tripped\n");
1362 pandecode_prop("zero1 = 0x%" PRIx32
, p
->zero1
);
1366 pandecode_log("},\n");
1370 pandecode_uniform_buffers(mali_ptr pubufs
, int ubufs_count
, int job_no
)
1372 struct pandecode_mapped_memory
*umem
= pandecode_find_mapped_gpu_mem_containing(pubufs
);
1374 struct mali_uniform_buffer_meta
*PANDECODE_PTR_VAR(ubufs
, umem
, pubufs
);
1376 for (int i
= 0; i
< ubufs_count
; i
++) {
1377 mali_ptr ptr
= ubufs
[i
].ptr
<< 2;
1378 struct pandecode_mapped_memory
*umem2
= pandecode_find_mapped_gpu_mem_containing(ptr
);
1379 uint32_t *PANDECODE_PTR_VAR(ubuf
, umem2
, ptr
);
1381 snprintf(name
, sizeof(name
), "ubuf_%d", i
);
1382 /* The blob uses ubuf 0 to upload internal stuff and
1383 * uniforms that won't fit/are accessed indirectly, so
1384 * it puts it in the batchbuffer.
1386 pandecode_log("uint32_t %s_%d[] = {\n", name
, job_no
);
1389 for (int j
= 0; j
<= ubufs
[i
].size
; j
++) {
1390 for (int k
= 0; k
< 4; k
++) {
1392 pandecode_log("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1394 pandecode_log_cont("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1398 pandecode_log_cont("\n");
1402 pandecode_log("};\n");
1405 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64
"_%d[] = {\n",
1409 for (int i
= 0; i
< ubufs_count
; i
++) {
1410 pandecode_log("{\n");
1412 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs
[i
].size
+ 1);
1413 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i
, job_no
);
1415 pandecode_log("},\n");
1419 pandecode_log("};\n");
1423 pandecode_scratchpad(uintptr_t pscratchpad
, int job_no
, char *suffix
)
1426 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(pscratchpad
);
1428 struct bifrost_scratchpad
*PANDECODE_PTR_VAR(scratchpad
, mem
, pscratchpad
);
1430 if (scratchpad
->zero
)
1431 pandecode_msg("XXX scratchpad zero tripped");
1433 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64
"_%d%s = {\n", pscratchpad
, job_no
, suffix
);
1436 pandecode_prop("flags = 0x%x", scratchpad
->flags
);
1437 MEMORY_PROP(scratchpad
, gpu_scratchpad
);
1440 pandecode_log("};\n");
1443 static unsigned shader_id
= 0;
1446 pandecode_shader_disassemble(mali_ptr shader_ptr
, int shader_no
, int type
,
1447 bool is_bifrost
, unsigned nr_regs
)
1449 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(shader_ptr
);
1450 uint8_t *PANDECODE_PTR_VAR(code
, mem
, shader_ptr
);
1452 /* Compute maximum possible size */
1453 size_t sz
= mem
->length
- (shader_ptr
- mem
->gpu_va
);
1455 /* Print some boilerplate to clearly denote the assembly (which doesn't
1456 * obey indentation rules), and actually do the disassembly! */
1462 snprintf(prefix
, sizeof(prefix
) - 1, "shader%d - %s shader: ",
1464 (type
== JOB_TYPE_TILER
) ? "FRAGMENT" : "VERTEX");
1467 disassemble_bifrost(code
, sz
, false);
1469 disassemble_midgard(code
, sz
, true, nr_regs
, prefix
);
1476 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix
*p
,
1477 int job_no
, enum mali_job_type job_type
,
1478 char *suffix
, bool is_bifrost
)
1480 mali_ptr shader_meta_ptr
= (u64
) (uintptr_t) (p
->_shader_upper
<< 4);
1481 struct pandecode_mapped_memory
*attr_mem
;
1483 unsigned rt_count
= 1;
1485 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1486 * are the only things actually needed from the FBD, vertex/tiler jobs
1487 * no longer reference the FBD -- instead, this field points to some
1488 * info about the scratchpad.
1491 pandecode_scratchpad(p
->framebuffer
& ~FBD_TYPE
, job_no
, suffix
);
1492 else if (p
->framebuffer
& MALI_MFBD
)
1493 rt_count
= pandecode_mfbd_bfr((u64
) ((uintptr_t) p
->framebuffer
) & FBD_MASK
, job_no
, false);
1494 else if (job_type
== JOB_TYPE_COMPUTE
)
1495 pandecode_compute_fbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1497 pandecode_sfbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1499 int varying_count
= 0, attribute_count
= 0, uniform_count
= 0, uniform_buffer_count
= 0;
1500 int texture_count
= 0, sampler_count
= 0;
1502 if (shader_meta_ptr
) {
1503 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr
);
1504 struct mali_shader_meta
*PANDECODE_PTR_VAR(s
, smem
, shader_meta_ptr
);
1506 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64
"_%d%s = {\n", shader_meta_ptr
, job_no
, suffix
);
1509 /* Save for dumps */
1510 attribute_count
= s
->attribute_count
;
1511 varying_count
= s
->varying_count
;
1512 texture_count
= s
->texture_count
;
1513 sampler_count
= s
->sampler_count
;
1516 uniform_count
= s
->bifrost2
.uniform_count
;
1517 uniform_buffer_count
= s
->bifrost1
.uniform_buffer_count
;
1519 uniform_count
= s
->midgard1
.uniform_count
;
1520 uniform_buffer_count
= s
->midgard1
.uniform_buffer_count
;
1523 mali_ptr shader_ptr
= pandecode_shader_address("shader", s
->shader
);
1525 pandecode_prop("texture_count = %" PRId16
, s
->texture_count
);
1526 pandecode_prop("sampler_count = %" PRId16
, s
->sampler_count
);
1527 pandecode_prop("attribute_count = %" PRId16
, s
->attribute_count
);
1528 pandecode_prop("varying_count = %" PRId16
, s
->varying_count
);
1530 unsigned nr_registers
= 0;
1533 pandecode_log(".bifrost1 = {\n");
1536 pandecode_prop("uniform_buffer_count = %" PRId32
, s
->bifrost1
.uniform_buffer_count
);
1537 pandecode_prop("unk1 = 0x%" PRIx32
, s
->bifrost1
.unk1
);
1540 pandecode_log("},\n");
1542 pandecode_log(".midgard1 = {\n");
1545 pandecode_prop("uniform_count = %" PRId16
, s
->midgard1
.uniform_count
);
1546 pandecode_prop("uniform_buffer_count = %" PRId16
, s
->midgard1
.uniform_buffer_count
);
1547 pandecode_prop("work_count = %" PRId16
, s
->midgard1
.work_count
);
1548 nr_registers
= s
->midgard1
.work_count
;
1550 pandecode_log(".flags = ");
1551 pandecode_log_decoded_flags(shader_midgard1_flag_info
, s
->midgard1
.flags
);
1552 pandecode_log_cont(",\n");
1554 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->midgard1
.unknown2
);
1557 pandecode_log("},\n");
1560 if (s
->depth_units
|| s
->depth_factor
) {
1561 pandecode_prop("depth_factor = %f", s
->depth_factor
);
1562 pandecode_prop("depth_units = %f", s
->depth_units
);
1565 if (s
->alpha_coverage
) {
1566 bool invert_alpha_coverage
= s
->alpha_coverage
& 0xFFF0;
1567 uint16_t inverted_coverage
= invert_alpha_coverage
? ~s
->alpha_coverage
: s
->alpha_coverage
;
1569 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1570 invert_alpha_coverage
? "~" : "",
1571 MALI_GET_ALPHA_COVERAGE(inverted_coverage
));
1574 if (s
->unknown2_3
|| s
->unknown2_4
) {
1575 pandecode_log(".unknown2_3 = ");
1577 int unknown2_3
= s
->unknown2_3
;
1578 int unknown2_4
= s
->unknown2_4
;
1580 /* We're not quite sure what these flags mean without the depth test, if anything */
1582 if (unknown2_3
& (MALI_DEPTH_TEST
| MALI_DEPTH_FUNC_MASK
)) {
1583 const char *func
= pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3
));
1584 unknown2_3
&= ~MALI_DEPTH_FUNC_MASK
;
1586 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func
);
1589 pandecode_log_decoded_flags(u3_flag_info
, unknown2_3
);
1590 pandecode_log_cont(",\n");
1592 pandecode_log(".unknown2_4 = ");
1593 pandecode_log_decoded_flags(u4_flag_info
, unknown2_4
);
1594 pandecode_log_cont(",\n");
1597 if (s
->stencil_mask_front
|| s
->stencil_mask_back
) {
1598 pandecode_prop("stencil_mask_front = 0x%02X", s
->stencil_mask_front
);
1599 pandecode_prop("stencil_mask_back = 0x%02X", s
->stencil_mask_back
);
1602 pandecode_stencil("front", &s
->stencil_front
);
1603 pandecode_stencil("back", &s
->stencil_back
);
1606 pandecode_log(".bifrost2 = {\n");
1609 pandecode_prop("unk3 = 0x%" PRIx32
, s
->bifrost2
.unk3
);
1610 pandecode_prop("preload_regs = 0x%" PRIx32
, s
->bifrost2
.preload_regs
);
1611 pandecode_prop("uniform_count = %" PRId32
, s
->bifrost2
.uniform_count
);
1612 pandecode_prop("unk4 = 0x%" PRIx32
, s
->bifrost2
.unk4
);
1615 pandecode_log("},\n");
1616 } else if (s
->midgard2
.unknown2_7
) {
1617 pandecode_log(".midgard2 = {\n");
1620 pandecode_prop("unknown2_7 = 0x%" PRIx32
, s
->midgard2
.unknown2_7
);
1622 pandecode_log("},\n");
1626 pandecode_prop("unknown2_8 = 0x%" PRIx32
, s
->unknown2_8
);
1629 /* TODO: Blend shaders routing/disasm */
1631 union midgard_blend blend
= s
->blend
;
1632 pandecode_midgard_blend(&blend
, false);
1636 pandecode_log("};\n");
1638 /* MRT blend fields are used whenever MFBD is used, with
1639 * per-RT descriptors */
1641 if (job_type
== JOB_TYPE_TILER
) {
1642 void* blend_base
= (void *) (s
+ 1);
1644 for (unsigned i
= 0; i
< rt_count
; i
++) {
1645 mali_ptr shader
= 0;
1648 shader
= pandecode_bifrost_blend(blend_base
, job_no
, i
);
1650 shader
= pandecode_midgard_blend_mrt(blend_base
, job_no
, i
);
1653 pandecode_shader_disassemble(shader
, job_no
, job_type
, false, 0);
1657 if (shader_ptr
& ~0xF)
1658 pandecode_shader_disassemble(shader_ptr
, job_no
, job_type
, is_bifrost
, nr_registers
);
1660 pandecode_msg("<no shader>\n");
1663 struct pandecode_mapped_memory
*fmem
= pandecode_find_mapped_gpu_mem_containing(p
->viewport
);
1664 struct mali_viewport
*PANDECODE_PTR_VAR(f
, fmem
, p
->viewport
);
1666 pandecode_log("struct mali_viewport viewport_%"PRIx64
"_%d%s = {\n", p
->viewport
, job_no
, suffix
);
1669 pandecode_prop("clip_minx = %f", f
->clip_minx
);
1670 pandecode_prop("clip_miny = %f", f
->clip_miny
);
1671 pandecode_prop("clip_minz = %f", f
->clip_minz
);
1672 pandecode_prop("clip_maxx = %f", f
->clip_maxx
);
1673 pandecode_prop("clip_maxy = %f", f
->clip_maxy
);
1674 pandecode_prop("clip_maxz = %f", f
->clip_maxz
);
1676 /* Only the higher coordinates are MALI_POSITIVE scaled */
1678 pandecode_prop("viewport0 = { %d, %d }",
1679 f
->viewport0
[0], f
->viewport0
[1]);
1681 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1682 f
->viewport1
[0] + 1, f
->viewport1
[1] + 1);
1685 pandecode_log("};\n");
1688 if (p
->attribute_meta
) {
1689 unsigned max_attr_index
= pandecode_attribute_meta(job_no
, attribute_count
, p
, false, suffix
);
1691 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->attributes
);
1692 pandecode_attributes(attr_mem
, p
->attributes
, job_no
, suffix
, max_attr_index
+ 1, false);
1695 /* Varyings are encoded like attributes but not actually sent; we just
1696 * pass a zero buffer with the right stride/size set, (or whatever)
1697 * since the GPU will write to it itself */
1699 if (p
->varying_meta
) {
1700 varying_count
= pandecode_attribute_meta(job_no
, varying_count
, p
, true, suffix
);
1704 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->varyings
);
1706 /* Number of descriptors depends on whether there are
1707 * non-internal varyings */
1709 pandecode_attributes(attr_mem
, p
->varyings
, job_no
, suffix
, varying_count
, true);
1712 bool is_compute
= job_type
== JOB_TYPE_COMPUTE
;
1714 if (p
->uniforms
&& !is_compute
) {
1715 int rows
= uniform_count
, width
= 4;
1716 size_t sz
= rows
* width
* sizeof(float);
1718 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1719 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1720 u32
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1722 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no
, suffix
);
1726 for (int row
= 0; row
< rows
; row
++) {
1727 for (int i
= 0; i
< width
; i
++) {
1728 u32 v
= uniforms
[i
];
1730 memcpy(&f
, &v
, sizeof(v
));
1731 pandecode_log_cont("%X /* %f */, ", v
, f
);
1734 pandecode_log_cont("\n");
1740 pandecode_log("};\n");
1741 } else if (p
->uniforms
) {
1742 int rows
= uniform_count
* 2;
1743 size_t sz
= rows
* sizeof(mali_ptr
);
1745 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1746 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1747 mali_ptr
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1749 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no
, suffix
);
1753 for (int row
= 0; row
< rows
; row
++) {
1754 char *a
= pointer_as_memory_reference(uniforms
[row
]);
1755 pandecode_log("%s,\n", a
);
1760 pandecode_log("};\n");
1764 if (p
->uniform_buffers
) {
1765 pandecode_uniform_buffers(p
->uniform_buffers
, uniform_buffer_count
, job_no
);
1768 if (p
->texture_trampoline
) {
1769 struct pandecode_mapped_memory
*mmem
= pandecode_find_mapped_gpu_mem_containing(p
->texture_trampoline
);
1772 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
);
1774 pandecode_log("uint64_t texture_trampoline_%"PRIx64
"_%d[] = {\n", p
->texture_trampoline
, job_no
);
1777 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1778 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1779 char *a
= pointer_as_memory_reference(*u
);
1780 pandecode_log("%s,\n", a
);
1785 pandecode_log("};\n");
1787 /* Now, finally, descend down into the texture descriptor */
1788 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1789 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1790 struct pandecode_mapped_memory
*tmem
= pandecode_find_mapped_gpu_mem_containing(*u
);
1793 struct mali_texture_descriptor
*PANDECODE_PTR_VAR(t
, tmem
, *u
);
1795 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64
"_%d_%d = {\n", *u
, job_no
, tex
);
1798 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", t
->width
+ 1);
1799 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", t
->height
+ 1);
1800 pandecode_prop("depth = MALI_POSITIVE(%" PRId16
")", t
->depth
+ 1);
1801 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16
")", t
->array_size
+ 1);
1802 pandecode_prop("unknown3 = %" PRId16
, t
->unknown3
);
1803 pandecode_prop("unknown3A = %" PRId8
, t
->unknown3A
);
1804 pandecode_prop("nr_mipmap_levels = %" PRId8
, t
->nr_mipmap_levels
);
1806 struct mali_texture_format f
= t
->format
;
1808 pandecode_log(".format = {\n");
1811 pandecode_swizzle(f
.swizzle
);
1812 pandecode_prop("format = %s", pandecode_format(f
.format
));
1813 pandecode_prop("type = %s", pandecode_texture_type(f
.type
));
1814 pandecode_prop("srgb = %" PRId32
, f
.srgb
);
1815 pandecode_prop("unknown1 = %" PRId32
, f
.unknown1
);
1816 pandecode_prop("usage2 = 0x%" PRIx32
, f
.usage2
);
1819 pandecode_log("},\n");
1821 pandecode_swizzle(t
->swizzle
);
1823 if (t
->swizzle_zero
) {
1824 /* Shouldn't happen */
1825 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1826 pandecode_prop("swizzle_zero = %d", t
->swizzle_zero
);
1829 pandecode_prop("unknown3 = 0x%" PRIx32
, t
->unknown3
);
1831 pandecode_prop("unknown5 = 0x%" PRIx32
, t
->unknown5
);
1832 pandecode_prop("unknown6 = 0x%" PRIx32
, t
->unknown6
);
1833 pandecode_prop("unknown7 = 0x%" PRIx32
, t
->unknown7
);
1835 pandecode_log(".payload = {\n");
1838 /* A bunch of bitmap pointers follow.
1839 * We work out the correct number,
1840 * based on the mipmap/cubemap
1841 * properties, but dump extra
1842 * possibilities to futureproof */
1844 int bitmap_count
= MALI_NEGATIVE(t
->nr_mipmap_levels
);
1845 bool manual_stride
= f
.usage2
& MALI_TEX_MANUAL_STRIDE
;
1847 /* Miptree for each face */
1848 if (f
.type
== MALI_TEX_CUBE
)
1851 /* Array of textures */
1852 bitmap_count
*= MALI_NEGATIVE(t
->array_size
);
1854 /* Stride for each element */
1858 /* Sanity check the size */
1859 int max_count
= sizeof(t
->payload
) / sizeof(t
->payload
[0]);
1860 assert (bitmap_count
<= max_count
);
1862 /* Dump more to be safe, but not _that_ much more */
1863 int safe_count
= MIN2(bitmap_count
* 2, max_count
);
1865 for (int i
= 0; i
< safe_count
; ++i
) {
1866 char *prefix
= (i
>= bitmap_count
) ? "// " : "";
1868 /* How we dump depends if this is a stride or a pointer */
1870 if ((f
.usage2
& MALI_TEX_MANUAL_STRIDE
) && (i
& 1)) {
1871 /* signed 32-bit snuck in as a 64-bit pointer */
1872 uint64_t stride_set
= t
->payload
[i
];
1873 uint32_t clamped_stride
= stride_set
;
1874 int32_t stride
= clamped_stride
;
1875 assert(stride_set
== clamped_stride
);
1876 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix
, stride
);
1878 char *a
= pointer_as_memory_reference(t
->payload
[i
]);
1879 pandecode_log("%s%s, \n", prefix
, a
);
1885 pandecode_log("},\n");
1888 pandecode_log("};\n");
1894 if (p
->sampler_descriptor
) {
1895 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(p
->sampler_descriptor
);
1898 struct mali_sampler_descriptor
*s
;
1900 mali_ptr d
= p
->sampler_descriptor
;
1902 for (int i
= 0; i
< sampler_count
; ++i
) {
1903 s
= pandecode_fetch_gpu_mem(smem
, d
+ sizeof(*s
) * i
, sizeof(*s
));
1905 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64
"_%d_%d = {\n", d
+ sizeof(*s
) * i
, job_no
, i
);
1908 pandecode_log(".filter_mode = ");
1909 pandecode_log_decoded_flags(sampler_flag_info
, s
->filter_mode
);
1910 pandecode_log_cont(",\n");
1912 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->min_lod
));
1913 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->max_lod
));
1915 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s
->wrap_s
));
1916 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s
->wrap_t
));
1917 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s
->wrap_r
));
1919 pandecode_prop("compare_func = %s", pandecode_alt_func(s
->compare_func
));
1921 if (s
->zero
|| s
->zero2
) {
1922 pandecode_msg("Zero tripped\n");
1923 pandecode_prop("zero = 0x%X, 0x%X\n", s
->zero
, s
->zero2
);
1926 pandecode_prop("seamless_cube_map = %d", s
->seamless_cube_map
);
1928 pandecode_prop("border_color = { %f, %f, %f, %f }",
1932 s
->border_color
[3]);
1935 pandecode_log("};\n");
1942 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix
*p
, int job_no
, bool is_bifrost
)
1944 if (!(p
->position_varying
|| p
->occlusion_counter
|| p
->flags
))
1947 pandecode_log(".postfix = {\n");
1950 MEMORY_PROP(p
, position_varying
);
1951 MEMORY_PROP(p
, occlusion_counter
);
1954 pandecode_prop("flags = %d", p
->flags
);
1957 pandecode_log("},\n");
1961 pandecode_vertex_only_bfr(struct bifrost_vertex_only
*v
)
1963 pandecode_log_cont("{\n");
1966 pandecode_prop("unk2 = 0x%x", v
->unk2
);
1968 if (v
->zero0
|| v
->zero1
) {
1969 pandecode_msg("vertex only zero tripped");
1970 pandecode_prop("zero0 = 0x%" PRIx32
, v
->zero0
);
1971 pandecode_prop("zero1 = 0x%" PRIx64
, v
->zero1
);
1975 pandecode_log("}\n");
1979 pandecode_tiler_heap_meta(mali_ptr gpu_va
, int job_no
)
1982 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
1983 const struct bifrost_tiler_heap_meta
*PANDECODE_PTR_VAR(h
, mem
, gpu_va
);
1985 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no
);
1989 pandecode_msg("tiler heap zero tripped\n");
1990 pandecode_prop("zero = 0x%x", h
->zero
);
1993 for (int i
= 0; i
< 12; i
++) {
1994 if (h
->zeros
[i
] != 0) {
1995 pandecode_msg("tiler heap zero %d tripped, value %x\n",
2000 pandecode_prop("heap_size = 0x%x", h
->heap_size
);
2001 MEMORY_PROP(h
, tiler_heap_start
);
2002 MEMORY_PROP(h
, tiler_heap_free
);
2004 /* this might point to the beginning of another buffer, when it's
2005 * really the end of the tiler heap buffer, so we have to be careful
2006 * here. but for zero length, we need the same pointer.
2009 if (h
->tiler_heap_end
== h
->tiler_heap_start
) {
2010 MEMORY_PROP(h
, tiler_heap_start
);
2012 char *a
= pointer_as_memory_reference(h
->tiler_heap_end
- 1);
2013 pandecode_prop("tiler_heap_end = %s + 1", a
);
2018 pandecode_log("};\n");
2022 pandecode_tiler_meta(mali_ptr gpu_va
, int job_no
)
2024 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
2025 const struct bifrost_tiler_meta
*PANDECODE_PTR_VAR(t
, mem
, gpu_va
);
2027 pandecode_tiler_heap_meta(t
->tiler_heap_meta
, job_no
);
2029 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no
);
2032 if (t
->zero0
|| t
->zero1
) {
2033 pandecode_msg("tiler meta zero tripped");
2034 pandecode_prop("zero0 = 0x%" PRIx64
, t
->zero0
);
2035 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
2038 pandecode_prop("hierarchy_mask = 0x%" PRIx16
, t
->hierarchy_mask
);
2039 pandecode_prop("flags = 0x%" PRIx16
, t
->flags
);
2041 pandecode_prop("width = MALI_POSITIVE(%d)", t
->width
+ 1);
2042 pandecode_prop("height = MALI_POSITIVE(%d)", t
->height
+ 1);
2044 for (int i
= 0; i
< 12; i
++) {
2045 if (t
->zeros
[i
] != 0) {
2046 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64
"\n",
2052 pandecode_log("};\n");
2056 pandecode_gl_enables(uint32_t gl_enables
, int job_type
)
2058 pandecode_log(".gl_enables = ");
2060 pandecode_log_decoded_flags(gl_enable_flag_info
, gl_enables
);
2062 pandecode_log_cont(",\n");
2066 pandecode_primitive_size(union midgard_primitive_size u
, bool constant
)
2068 if (u
.pointer
== 0x0)
2071 pandecode_log(".primitive_size = {\n");
2075 pandecode_prop("constant = %f", u
.constant
);
2077 MEMORY_PROP((&u
), pointer
);
2081 pandecode_log("},\n");
2085 pandecode_tiler_only_bfr(const struct bifrost_tiler_only
*t
, int job_no
)
2087 pandecode_log_cont("{\n");
2090 /* TODO: gl_PointSize on Bifrost */
2091 pandecode_primitive_size(t
->primitive_size
, true);
2093 pandecode_gl_enables(t
->gl_enables
, JOB_TYPE_TILER
);
2095 if (t
->zero1
|| t
->zero2
|| t
->zero3
|| t
->zero4
|| t
->zero5
2096 || t
->zero6
|| t
->zero7
|| t
->zero8
) {
2097 pandecode_msg("tiler only zero tripped");
2098 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
2099 pandecode_prop("zero2 = 0x%" PRIx64
, t
->zero2
);
2100 pandecode_prop("zero3 = 0x%" PRIx64
, t
->zero3
);
2101 pandecode_prop("zero4 = 0x%" PRIx64
, t
->zero4
);
2102 pandecode_prop("zero5 = 0x%" PRIx64
, t
->zero5
);
2103 pandecode_prop("zero6 = 0x%" PRIx64
, t
->zero6
);
2104 pandecode_prop("zero7 = 0x%" PRIx32
, t
->zero7
);
2105 pandecode_prop("zero8 = 0x%" PRIx64
, t
->zero8
);
2109 pandecode_log("},\n");
2113 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header
*h
,
2114 const struct pandecode_mapped_memory
*mem
,
2115 mali_ptr payload
, int job_no
)
2117 struct bifrost_payload_vertex
*PANDECODE_PTR_VAR(v
, mem
, payload
);
2119 pandecode_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", true);
2121 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no
);
2124 pandecode_log(".prefix = ");
2125 pandecode_vertex_tiler_prefix(&v
->prefix
, job_no
);
2127 pandecode_log(".vertex = ");
2128 pandecode_vertex_only_bfr(&v
->vertex
);
2130 pandecode_vertex_tiler_postfix(&v
->postfix
, job_no
, true);
2133 pandecode_log("};\n");
2139 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header
*h
,
2140 const struct pandecode_mapped_memory
*mem
,
2141 mali_ptr payload
, int job_no
)
2143 struct bifrost_payload_tiler
*PANDECODE_PTR_VAR(t
, mem
, payload
);
2145 pandecode_vertex_tiler_postfix_pre(&t
->postfix
, job_no
, h
->job_type
, "", true);
2147 pandecode_indices(t
->prefix
.indices
, t
->prefix
.index_count
, job_no
);
2148 pandecode_tiler_meta(t
->tiler
.tiler_meta
, job_no
);
2150 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no
);
2153 pandecode_log(".prefix = ");
2154 pandecode_vertex_tiler_prefix(&t
->prefix
, job_no
);
2156 pandecode_log(".tiler = ");
2157 pandecode_tiler_only_bfr(&t
->tiler
, job_no
);
2159 pandecode_vertex_tiler_postfix(&t
->postfix
, job_no
, true);
2162 pandecode_log("};\n");
2168 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header
*h
,
2169 const struct pandecode_mapped_memory
*mem
,
2170 mali_ptr payload
, int job_no
)
2172 struct midgard_payload_vertex_tiler
*PANDECODE_PTR_VAR(v
, mem
, payload
);
2174 pandecode_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", false);
2176 pandecode_indices(v
->prefix
.indices
, v
->prefix
.index_count
, job_no
);
2178 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no
);
2181 bool has_primitive_pointer
= v
->prefix
.unknown_draw
& MALI_DRAW_VARYING_SIZE
;
2182 pandecode_primitive_size(v
->primitive_size
, !has_primitive_pointer
);
2184 pandecode_log(".prefix = ");
2185 pandecode_vertex_tiler_prefix(&v
->prefix
, job_no
);
2187 pandecode_gl_enables(v
->gl_enables
, h
->job_type
);
2189 if (v
->instance_shift
|| v
->instance_odd
) {
2190 pandecode_prop("instance_shift = 0x%d /* %d */",
2191 v
->instance_shift
, 1 << v
->instance_shift
);
2192 pandecode_prop("instance_odd = 0x%X /* %d */",
2193 v
->instance_odd
, (2 * v
->instance_odd
) + 1);
2195 pandecode_padded_vertices(v
->instance_shift
, v
->instance_odd
);
2198 if (v
->offset_start
)
2199 pandecode_prop("offset_start = %d", v
->offset_start
);
2202 pandecode_msg("Zero tripped\n");
2203 pandecode_prop("zero5 = 0x%" PRIx64
, v
->zero5
);
2206 pandecode_vertex_tiler_postfix(&v
->postfix
, job_no
, false);
2209 pandecode_log("};\n");
2215 pandecode_fragment_job(const struct pandecode_mapped_memory
*mem
,
2216 mali_ptr payload
, int job_no
,
2219 const struct mali_payload_fragment
*PANDECODE_PTR_VAR(s
, mem
, payload
);
2221 bool fbd_dumped
= false;
2223 if (!is_bifrost
&& (s
->framebuffer
& FBD_TYPE
) == MALI_SFBD
) {
2224 /* Only SFBDs are understood, not MFBDs. We're speculating,
2225 * based on the versioning, kernel code, etc, that the
2226 * difference is between Single FrameBuffer Descriptor and
2227 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2228 * up with multi-framebuffer support being added (T7xx onwards,
2229 * including Gxx). In any event, there's some field shuffling
2230 * that we haven't looked into yet. */
2232 pandecode_sfbd(s
->framebuffer
& FBD_MASK
, job_no
);
2234 } else if ((s
->framebuffer
& FBD_TYPE
) == MALI_MFBD
) {
2235 /* We don't know if Bifrost supports SFBD's at all, since the
2236 * driver never uses them. And the format is different from
2237 * Midgard anyways, due to the tiler heap and scratchpad being
2238 * moved out into separate structures, so it's not clear what a
2239 * Bifrost SFBD would even look like without getting an actual
2240 * trace, which appears impossible.
2243 pandecode_mfbd_bfr(s
->framebuffer
& FBD_MASK
, job_no
, true);
2247 uintptr_t p
= (uintptr_t) s
->framebuffer
& FBD_MASK
;
2248 pandecode_log("struct mali_payload_fragment payload_%"PRIx64
"_%d = {\n", payload
, job_no
);
2251 /* See the comments by the macro definitions for mathematical context
2252 * on why this is so weird */
2254 if (MALI_TILE_COORD_FLAGS(s
->max_tile_coord
) || MALI_TILE_COORD_FLAGS(s
->min_tile_coord
))
2255 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2257 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2258 MALI_TILE_COORD_X(s
->min_tile_coord
) << MALI_TILE_SHIFT
,
2259 MALI_TILE_COORD_Y(s
->min_tile_coord
) << MALI_TILE_SHIFT
);
2261 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2262 (MALI_TILE_COORD_X(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
,
2263 (MALI_TILE_COORD_Y(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
);
2265 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2266 * we have to fallback on offsets. */
2268 const char *fbd_type
= s
->framebuffer
& MALI_MFBD
? "MALI_MFBD" : "MALI_SFBD";
2271 unsigned extra_flags
= (s
->framebuffer
& ~FBD_MASK
) & ~MALI_MFBD
;
2274 pandecode_prop("framebuffer = framebuffer_%d_p | %s | 0x%X", job_no
,
2275 fbd_type
, extra_flags
);
2277 pandecode_prop("framebuffer = %s | %s | 0x%X", pointer_as_memory_reference(p
),
2278 fbd_type
, extra_flags
);
2281 pandecode_log("};\n");
2286 static int job_descriptor_number
= 0;
2289 pandecode_jc(mali_ptr jc_gpu_va
, bool bifrost
)
2291 struct mali_job_descriptor_header
*h
;
2293 int start_number
= 0;
2299 struct pandecode_mapped_memory
*mem
=
2300 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va
);
2304 h
= PANDECODE_PTR(mem
, jc_gpu_va
, struct mali_job_descriptor_header
);
2306 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2307 * high 32-bits of the 64-bit pointer are reused to store
2310 int offset
= h
->job_descriptor_size
== MALI_JOB_32
&&
2311 h
->job_type
!= JOB_TYPE_FRAGMENT
? 4 : 0;
2312 mali_ptr payload_ptr
= jc_gpu_va
+ sizeof(*h
) - offset
;
2314 payload
= pandecode_fetch_gpu_mem(mem
, payload_ptr
,
2317 int job_no
= job_descriptor_number
++;
2320 start_number
= job_no
;
2322 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64
"_%d = {\n", jc_gpu_va
, job_no
);
2325 pandecode_prop("job_type = %s", pandecode_job_type(h
->job_type
));
2327 /* Save for next job fixing */
2328 last_size
= h
->job_descriptor_size
;
2330 if (h
->job_descriptor_size
)
2331 pandecode_prop("job_descriptor_size = %d", h
->job_descriptor_size
);
2333 if (h
->exception_status
!= 0x1)
2334 pandecode_prop("exception_status = %x (source ID: 0x%x access: %s exception: 0x%x)",
2335 h
->exception_status
,
2336 (h
->exception_status
>> 16) & 0xFFFF,
2337 pandecode_exception_access((h
->exception_status
>> 8) & 0x3),
2338 h
->exception_status
& 0xFF);
2340 if (h
->first_incomplete_task
)
2341 pandecode_prop("first_incomplete_task = %d", h
->first_incomplete_task
);
2343 if (h
->fault_pointer
)
2344 pandecode_prop("fault_pointer = 0x%" PRIx64
, h
->fault_pointer
);
2347 pandecode_prop("job_barrier = %d", h
->job_barrier
);
2349 pandecode_prop("job_index = %d", h
->job_index
);
2351 if (h
->unknown_flags
)
2352 pandecode_prop("unknown_flags = %d", h
->unknown_flags
);
2354 if (h
->job_dependency_index_1
)
2355 pandecode_prop("job_dependency_index_1 = %d", h
->job_dependency_index_1
);
2357 if (h
->job_dependency_index_2
)
2358 pandecode_prop("job_dependency_index_2 = %d", h
->job_dependency_index_2
);
2361 pandecode_log("};\n");
2363 /* Do not touch the field yet -- decode the payload first, and
2364 * don't touch that either. This is essential for the uploads
2365 * to occur in sequence and therefore be dynamically allocated
2366 * correctly. Do note the size, however, for that related
2369 switch (h
->job_type
) {
2370 case JOB_TYPE_SET_VALUE
: {
2371 struct mali_payload_set_value
*s
= payload
;
2372 pandecode_log("struct mali_payload_set_value payload_%"PRIx64
"_%d = {\n", payload_ptr
, job_no
);
2374 MEMORY_PROP(s
, out
);
2375 pandecode_prop("unknown = 0x%" PRIX64
, s
->unknown
);
2377 pandecode_log("};\n");
2382 case JOB_TYPE_TILER
:
2383 case JOB_TYPE_VERTEX
:
2384 case JOB_TYPE_COMPUTE
:
2386 if (h
->job_type
== JOB_TYPE_TILER
)
2387 pandecode_tiler_job_bfr(h
, mem
, payload_ptr
, job_no
);
2389 pandecode_vertex_job_bfr(h
, mem
, payload_ptr
, job_no
);
2391 pandecode_vertex_or_tiler_job_mdg(h
, mem
, payload_ptr
, job_no
);
2395 case JOB_TYPE_FRAGMENT
:
2396 pandecode_fragment_job(mem
, payload_ptr
, job_no
, bifrost
);
2403 /* Handle linkage */
2406 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no
- 1);
2409 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no
);
2411 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no
);
2416 } while ((jc_gpu_va
= h
->job_descriptor_size
? h
->next_job_64
: h
->next_job_32
));
2418 return start_number
;