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"
39 int pandecode_jc(mali_ptr jc_gpu_va
, bool bifrost
);
41 #define MEMORY_PROP(obj, p) {\
43 char *a = pointer_as_memory_reference(obj->p); \
44 pandecode_prop("%s = %s", #p, a); \
49 #define MEMORY_PROP_DIR(obj, p) {\
51 char *a = pointer_as_memory_reference(obj.p); \
52 pandecode_prop("%s = %s", #p, a); \
57 #define DYN_MEMORY_PROP(obj, no, p) { \
59 pandecode_prop("%s = %s_%d_p", #p, #p, no); \
62 /* Semantic logging type.
64 * Raw: for raw messages to be printed as is.
65 * Message: for helpful information to be commented out in replays.
66 * Property: for properties of a struct
68 * Use one of pandecode_log, pandecode_msg, or pandecode_prop as syntax sugar.
71 enum pandecode_log_type
{
77 #define pandecode_log(...) pandecode_log_typed(PANDECODE_RAW, __VA_ARGS__)
78 #define pandecode_msg(...) pandecode_log_typed(PANDECODE_MESSAGE, __VA_ARGS__)
79 #define pandecode_prop(...) pandecode_log_typed(PANDECODE_PROPERTY, __VA_ARGS__)
81 unsigned pandecode_indent
= 0;
84 pandecode_make_indent(void)
86 for (unsigned i
= 0; i
< pandecode_indent
; ++i
)
91 pandecode_log_typed(enum pandecode_log_type type
, const char *format
, ...)
95 pandecode_make_indent();
97 if (type
== PANDECODE_MESSAGE
)
99 else if (type
== PANDECODE_PROPERTY
)
102 va_start(ap
, format
);
106 if (type
== PANDECODE_PROPERTY
)
111 pandecode_log_cont(const char *format
, ...)
115 va_start(ap
, format
);
120 struct pandecode_flag_info
{
126 pandecode_log_decoded_flags(const struct pandecode_flag_info
*flag_info
,
129 bool decodable_flags_found
= false;
131 for (int i
= 0; flag_info
[i
].name
; i
++) {
132 if ((flags
& flag_info
[i
].flag
) != flag_info
[i
].flag
)
135 if (!decodable_flags_found
) {
136 decodable_flags_found
= true;
138 pandecode_log_cont(" | ");
141 pandecode_log_cont("%s", flag_info
[i
].name
);
143 flags
&= ~flag_info
[i
].flag
;
146 if (decodable_flags_found
) {
148 pandecode_log_cont(" | 0x%" PRIx64
, flags
);
150 pandecode_log_cont("0x%" PRIx64
, flags
);
154 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
155 static const struct pandecode_flag_info gl_enable_flag_info
[] = {
156 FLAG_INFO(OCCLUSION_QUERY
),
157 FLAG_INFO(OCCLUSION_PRECISE
),
158 FLAG_INFO(FRONT_CCW_TOP
),
159 FLAG_INFO(CULL_FACE_FRONT
),
160 FLAG_INFO(CULL_FACE_BACK
),
165 #define FLAG_INFO(flag) { MALI_CLEAR_##flag, "MALI_CLEAR_" #flag }
166 static const struct pandecode_flag_info clear_flag_info
[] = {
169 FLAG_INFO(SLOW_STENCIL
),
174 #define FLAG_INFO(flag) { MALI_MASK_##flag, "MALI_MASK_" #flag }
175 static const struct pandecode_flag_info mask_flag_info
[] = {
184 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
185 static const struct pandecode_flag_info u3_flag_info
[] = {
187 FLAG_INFO(CAN_DISCARD
),
188 FLAG_INFO(HAS_BLEND_SHADER
),
189 FLAG_INFO(DEPTH_TEST
),
193 static const struct pandecode_flag_info u4_flag_info
[] = {
195 FLAG_INFO(NO_DITHER
),
196 FLAG_INFO(DEPTH_RANGE_A
),
197 FLAG_INFO(DEPTH_RANGE_B
),
198 FLAG_INFO(STENCIL_TEST
),
199 FLAG_INFO(SAMPLE_ALPHA_TO_COVERAGE_NO_BLEND_SHADER
),
204 #define FLAG_INFO(flag) { MALI_FRAMEBUFFER_##flag, "MALI_FRAMEBUFFER_" #flag }
205 static const struct pandecode_flag_info fb_fmt_flag_info
[] = {
213 #define FLAG_INFO(flag) { MALI_MFBD_FORMAT_##flag, "MALI_MFBD_FORMAT_" #flag }
214 static const struct pandecode_flag_info mfbd_fmt_flag_info
[] = {
221 #define FLAG_INFO(flag) { MALI_EXTRA_##flag, "MALI_EXTRA_" #flag }
222 static const struct pandecode_flag_info mfbd_extra_flag_info
[] = {
230 #define FLAG_INFO(flag) { MALI_##flag, "MALI_" #flag }
231 static const struct pandecode_flag_info shader_midgard1_flag_info
[] = {
233 FLAG_INFO(HELPER_INVOCATIONS
),
234 FLAG_INFO(READS_TILEBUFFER
),
240 #define FLAG_INFO(flag) { MALI_MFBD_##flag, "MALI_MFBD_" #flag }
241 static const struct pandecode_flag_info mfbd_flag_info
[] = {
242 FLAG_INFO(DEPTH_WRITE
),
248 #define FLAG_INFO(flag) { MALI_SAMP_##flag, "MALI_SAMP_" #flag }
249 static const struct pandecode_flag_info sampler_flag_info
[] = {
250 FLAG_INFO(MAG_NEAREST
),
251 FLAG_INFO(MIN_NEAREST
),
252 FLAG_INFO(MIP_LINEAR_1
),
253 FLAG_INFO(MIP_LINEAR_2
),
254 FLAG_INFO(NORM_COORDS
),
259 extern char *replace_fragment
;
260 extern char *replace_vertex
;
263 pandecode_job_type(enum mali_job_type type
)
265 #define DEFINE_CASE(name) case JOB_TYPE_ ## name: return "JOB_TYPE_" #name
269 DEFINE_CASE(SET_VALUE
);
270 DEFINE_CASE(CACHE_FLUSH
);
271 DEFINE_CASE(COMPUTE
);
275 DEFINE_CASE(FRAGMENT
);
277 case JOB_NOT_STARTED
:
278 return "NOT_STARTED";
281 pandecode_log("Warning! Unknown job type %x\n", type
);
289 pandecode_draw_mode(enum mali_draw_mode mode
)
291 #define DEFINE_CASE(name) case MALI_ ## name: return "MALI_" #name
294 DEFINE_CASE(DRAW_NONE
);
297 DEFINE_CASE(TRIANGLES
);
298 DEFINE_CASE(TRIANGLE_STRIP
);
299 DEFINE_CASE(TRIANGLE_FAN
);
300 DEFINE_CASE(LINE_STRIP
);
301 DEFINE_CASE(LINE_LOOP
);
302 DEFINE_CASE(POLYGON
);
304 DEFINE_CASE(QUAD_STRIP
);
307 return "MALI_TRIANGLES /* XXX: Unknown GL mode, check dump */";
313 #define DEFINE_CASE(name) case MALI_FUNC_ ## name: return "MALI_FUNC_" #name
315 pandecode_func(enum mali_func mode
)
322 DEFINE_CASE(GREATER
);
323 DEFINE_CASE(NOTEQUAL
);
328 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
333 /* Why is this duplicated? Who knows... */
334 #define DEFINE_CASE(name) case MALI_ALT_FUNC_ ## name: return "MALI_ALT_FUNC_" #name
336 pandecode_alt_func(enum mali_alt_func mode
)
343 DEFINE_CASE(GREATER
);
344 DEFINE_CASE(NOTEQUAL
);
349 return "MALI_FUNC_NEVER /* XXX: Unknown function, check dump */";
354 #define DEFINE_CASE(name) case MALI_STENCIL_ ## name: return "MALI_STENCIL_" #name
356 pandecode_stencil_op(enum mali_stencil_op op
)
360 DEFINE_CASE(REPLACE
);
363 DEFINE_CASE(INCR_WRAP
);
364 DEFINE_CASE(DECR_WRAP
);
369 return "MALI_STENCIL_KEEP /* XXX: Unknown stencil op, check dump */";
375 #define DEFINE_CASE(name) case MALI_ATTR_ ## name: return "MALI_ATTR_" #name
376 static char *pandecode_attr_mode(enum mali_attr_mode mode
)
381 DEFINE_CASE(POT_DIVIDE
);
383 DEFINE_CASE(NPOT_DIVIDE
);
385 DEFINE_CASE(INTERNAL
);
387 return "MALI_ATTR_UNUSED /* XXX: Unknown stencil op, check dump */";
393 #define DEFINE_CASE(name) case MALI_CHANNEL_## name: return "MALI_CHANNEL_" #name
395 pandecode_channel(enum mali_channel channel
)
404 DEFINE_CASE(RESERVED_0
);
405 DEFINE_CASE(RESERVED_1
);
408 return "MALI_CHANNEL_ZERO /* XXX: Unknown channel, check dump */";
413 #define DEFINE_CASE(name) case MALI_WRAP_## name: return "MALI_WRAP_" #name
415 pandecode_wrap_mode(enum mali_wrap_mode op
)
419 DEFINE_CASE(CLAMP_TO_EDGE
);
420 DEFINE_CASE(CLAMP_TO_BORDER
);
421 DEFINE_CASE(MIRRORED_REPEAT
);
424 return "MALI_WRAP_REPEAT /* XXX: Unknown wrap mode, check dump */";
429 #define DEFINE_CASE(name) case MALI_TEX_## name: return "MALI_TEX_" #name
431 pandecode_texture_type(enum mali_texture_type type
)
440 unreachable("Unknown case");
445 #define DEFINE_CASE(name) case MALI_MFBD_BLOCK_## name: return "MALI_MFBD_BLOCK_" #name
447 pandecode_mfbd_block_format(enum mali_mfbd_block_format fmt
)
451 DEFINE_CASE(UNKNOWN
);
456 unreachable("Invalid case");
461 /* Midgard's tiler descriptor is embedded within the
465 pandecode_midgard_tiler_descriptor(const struct midgard_tiler_descriptor
*t
)
467 pandecode_log(".tiler = {\n");
470 pandecode_prop("hierarchy_mask = 0x%" PRIx16
, t
->hierarchy_mask
);
471 pandecode_prop("flags = 0x%" PRIx16
, t
->flags
);
472 pandecode_prop("polygon_list_size = 0x%x", t
->polygon_list_size
);
474 MEMORY_PROP(t
, polygon_list
);
475 MEMORY_PROP(t
, polygon_list_body
);
477 MEMORY_PROP(t
, heap_start
);
479 if (t
->heap_start
== t
->heap_end
) {
480 /* Print identically to show symmetry for empty tiler heaps */
481 MEMORY_PROP(t
, heap_start
);
483 /* Points to the end of a buffer */
484 char *a
= pointer_as_memory_reference(t
->heap_end
- 1);
485 pandecode_prop("heap_end = %s + 1", a
);
489 bool nonzero_weights
= false;
491 for (unsigned w
= 0; w
< ARRAY_SIZE(t
->weights
); ++w
) {
492 nonzero_weights
|= t
->weights
[w
] != 0x0;
495 if (nonzero_weights
) {
496 pandecode_log(".weights = {");
498 for (unsigned w
= 0; w
< ARRAY_SIZE(t
->weights
); ++w
) {
499 pandecode_log("%d, ", t
->weights
[w
]);
506 pandecode_log("}\n");
510 pandecode_sfbd(uint64_t gpu_va
, int job_no
)
512 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
513 const struct mali_single_framebuffer
*PANDECODE_PTR_VAR(s
, mem
, (mali_ptr
) gpu_va
);
515 pandecode_log("struct mali_single_framebuffer framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
518 pandecode_prop("unknown1 = 0x%" PRIx32
, s
->unknown1
);
519 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->unknown2
);
521 pandecode_log(".format = ");
522 pandecode_log_decoded_flags(fb_fmt_flag_info
, s
->format
);
523 pandecode_log_cont(",\n");
525 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", s
->width
+ 1);
526 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", s
->height
+ 1);
528 MEMORY_PROP(s
, framebuffer
);
529 pandecode_prop("stride = %d", s
->stride
);
531 /* Earlier in the actual commandstream -- right before width -- but we
532 * delay to flow nicer */
534 pandecode_log(".clear_flags = ");
535 pandecode_log_decoded_flags(clear_flag_info
, s
->clear_flags
);
536 pandecode_log_cont(",\n");
538 if (s
->depth_buffer
| s
->depth_buffer_enable
) {
539 MEMORY_PROP(s
, depth_buffer
);
540 pandecode_prop("depth_buffer_enable = %s", DS_ENABLE(s
->depth_buffer_enable
));
543 if (s
->stencil_buffer
| s
->stencil_buffer_enable
) {
544 MEMORY_PROP(s
, stencil_buffer
);
545 pandecode_prop("stencil_buffer_enable = %s", DS_ENABLE(s
->stencil_buffer_enable
));
548 if (s
->clear_color_1
| s
->clear_color_2
| s
->clear_color_3
| s
->clear_color_4
) {
549 pandecode_prop("clear_color_1 = 0x%" PRIx32
, s
->clear_color_1
);
550 pandecode_prop("clear_color_2 = 0x%" PRIx32
, s
->clear_color_2
);
551 pandecode_prop("clear_color_3 = 0x%" PRIx32
, s
->clear_color_3
);
552 pandecode_prop("clear_color_4 = 0x%" PRIx32
, s
->clear_color_4
);
555 if (s
->clear_depth_1
!= 0 || s
->clear_depth_2
!= 0 || s
->clear_depth_3
!= 0 || s
->clear_depth_4
!= 0) {
556 pandecode_prop("clear_depth_1 = %f", s
->clear_depth_1
);
557 pandecode_prop("clear_depth_2 = %f", s
->clear_depth_2
);
558 pandecode_prop("clear_depth_3 = %f", s
->clear_depth_3
);
559 pandecode_prop("clear_depth_4 = %f", s
->clear_depth_4
);
562 if (s
->clear_stencil
) {
563 pandecode_prop("clear_stencil = 0x%x", s
->clear_stencil
);
566 MEMORY_PROP(s
, unknown_address_0
);
567 const struct midgard_tiler_descriptor t
= s
->tiler
;
568 pandecode_midgard_tiler_descriptor(&t
);
571 pandecode_log("};\n");
573 pandecode_prop("zero0 = 0x%" PRIx64
, s
->zero0
);
574 pandecode_prop("zero1 = 0x%" PRIx64
, s
->zero1
);
575 pandecode_prop("zero2 = 0x%" PRIx32
, s
->zero2
);
576 pandecode_prop("zero4 = 0x%" PRIx32
, s
->zero4
);
578 printf(".zero3 = {");
580 for (int i
= 0; i
< sizeof(s
->zero3
) / sizeof(s
->zero3
[0]); ++i
)
581 printf("%X, ", s
->zero3
[i
]);
585 printf(".zero6 = {");
587 for (int i
= 0; i
< sizeof(s
->zero6
) / sizeof(s
->zero6
[0]); ++i
)
588 printf("%X, ", s
->zero6
[i
]);
594 pandecode_u32_slide(unsigned name
, const u32
*slide
, unsigned count
)
596 pandecode_log(".unknown%d = {", name
);
598 for (int i
= 0; i
< count
; ++i
)
599 printf("%X, ", slide
[i
]);
601 pandecode_log("},\n");
604 #define SHORT_SLIDE(num) \
605 pandecode_u32_slide(num, s->unknown ## num, ARRAY_SIZE(s->unknown ## num))
608 pandecode_compute_fbd(uint64_t gpu_va
, int job_no
)
610 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
611 const struct mali_compute_fbd
*PANDECODE_PTR_VAR(s
, mem
, (mali_ptr
) gpu_va
);
613 pandecode_log("struct mali_compute_fbd framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
623 pandecode_swizzle(unsigned swizzle
)
625 pandecode_prop("swizzle = %s | (%s << 3) | (%s << 6) | (%s << 9)",
626 pandecode_channel((swizzle
>> 0) & 0x7),
627 pandecode_channel((swizzle
>> 3) & 0x7),
628 pandecode_channel((swizzle
>> 6) & 0x7),
629 pandecode_channel((swizzle
>> 9) & 0x7));
633 pandecode_rt_format(struct mali_rt_format format
)
635 pandecode_log(".format = {\n");
638 pandecode_prop("unk1 = 0x%" PRIx32
, format
.unk1
);
639 pandecode_prop("unk2 = 0x%" PRIx32
, format
.unk2
);
640 pandecode_prop("unk3 = 0x%" PRIx32
, format
.unk3
);
642 pandecode_prop("block = %s",
643 pandecode_mfbd_block_format(format
.block
));
645 pandecode_prop("nr_channels = MALI_POSITIVE(%d)",
646 MALI_NEGATIVE(format
.nr_channels
));
648 pandecode_log(".flags = ");
649 pandecode_log_decoded_flags(mfbd_fmt_flag_info
, format
.flags
);
650 pandecode_log_cont(",\n");
652 pandecode_swizzle(format
.swizzle
);
654 pandecode_prop("unk4 = 0x%" PRIx32
, format
.unk4
);
657 pandecode_log("},\n");
661 pandecode_render_target(uint64_t gpu_va
, unsigned job_no
, const struct bifrost_framebuffer
*fb
)
663 pandecode_log("struct bifrost_render_target rts_list_%"PRIx64
"_%d[] = {\n", gpu_va
, job_no
);
666 for (int i
= 0; i
< MALI_NEGATIVE(fb
->rt_count_1
); i
++) {
667 mali_ptr rt_va
= gpu_va
+ i
* sizeof(struct bifrost_render_target
);
668 struct pandecode_mapped_memory
*mem
=
669 pandecode_find_mapped_gpu_mem_containing(rt_va
);
670 const struct bifrost_render_target
*PANDECODE_PTR_VAR(rt
, mem
, (mali_ptr
) rt_va
);
672 pandecode_log("{\n");
675 pandecode_rt_format(rt
->format
);
677 if (rt
->format
.block
== MALI_MFBD_BLOCK_AFBC
) {
678 pandecode_log(".afbc = {\n");
681 char *a
= pointer_as_memory_reference(rt
->afbc
.metadata
);
682 pandecode_prop("metadata = %s", a
);
685 pandecode_prop("stride = %d", rt
->afbc
.stride
);
686 pandecode_prop("unk = 0x%" PRIx32
, rt
->afbc
.unk
);
689 pandecode_log("},\n");
691 pandecode_log(".chunknown = {\n");
694 pandecode_prop("unk = 0x%" PRIx64
, rt
->chunknown
.unk
);
696 char *a
= pointer_as_memory_reference(rt
->chunknown
.pointer
);
697 pandecode_prop("pointer = %s", a
);
701 pandecode_log("},\n");
704 MEMORY_PROP(rt
, framebuffer
);
705 pandecode_prop("framebuffer_stride = %d", rt
->framebuffer_stride
);
707 if (rt
->clear_color_1
| rt
->clear_color_2
| rt
->clear_color_3
| rt
->clear_color_4
) {
708 pandecode_prop("clear_color_1 = 0x%" PRIx32
, rt
->clear_color_1
);
709 pandecode_prop("clear_color_2 = 0x%" PRIx32
, rt
->clear_color_2
);
710 pandecode_prop("clear_color_3 = 0x%" PRIx32
, rt
->clear_color_3
);
711 pandecode_prop("clear_color_4 = 0x%" PRIx32
, rt
->clear_color_4
);
714 if (rt
->zero1
|| rt
->zero2
|| rt
->zero3
) {
715 pandecode_msg("render target zeros tripped\n");
716 pandecode_prop("zero1 = 0x%" PRIx64
, rt
->zero1
);
717 pandecode_prop("zero2 = 0x%" PRIx32
, rt
->zero2
);
718 pandecode_prop("zero3 = 0x%" PRIx32
, rt
->zero3
);
722 pandecode_log("},\n");
726 pandecode_log("};\n");
730 pandecode_mfbd_bfr(uint64_t gpu_va
, int job_no
, bool with_render_targets
)
732 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
733 const struct bifrost_framebuffer
*PANDECODE_PTR_VAR(fb
, mem
, (mali_ptr
) gpu_va
);
735 if (fb
->sample_locations
) {
736 /* The blob stores all possible sample locations in a single buffer
737 * allocated on startup, and just switches the pointer when switching
738 * MSAA state. For now, we just put the data into the cmdstream, but we
739 * should do something like what the blob does with a real driver.
741 * There seem to be 32 slots for sample locations, followed by another
742 * 16. The second 16 is just the center location followed by 15 zeros
743 * in all the cases I've identified (maybe shader vs. depth/color
747 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(fb
->sample_locations
);
749 const u16
*PANDECODE_PTR_VAR(samples
, smem
, fb
->sample_locations
);
751 pandecode_log("uint16_t sample_locations_%d[] = {\n", job_no
);
754 for (int i
= 0; i
< 32 + 16; i
++) {
755 pandecode_log("%d, %d,\n", samples
[2 * i
], samples
[2 * i
+ 1]);
759 pandecode_log("};\n");
762 pandecode_log("struct bifrost_framebuffer framebuffer_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
765 pandecode_prop("unk0 = 0x%x", fb
->unk0
);
767 if (fb
->sample_locations
)
768 pandecode_prop("sample_locations = sample_locations_%d", job_no
);
770 /* Assume that unknown1 was emitted in the last job for
772 MEMORY_PROP(fb
, unknown1
);
774 pandecode_prop("width1 = MALI_POSITIVE(%d)", fb
->width1
+ 1);
775 pandecode_prop("height1 = MALI_POSITIVE(%d)", fb
->height1
+ 1);
776 pandecode_prop("width2 = MALI_POSITIVE(%d)", fb
->width2
+ 1);
777 pandecode_prop("height2 = MALI_POSITIVE(%d)", fb
->height2
+ 1);
779 pandecode_prop("unk1 = 0x%x", fb
->unk1
);
780 pandecode_prop("unk2 = 0x%x", fb
->unk2
);
781 pandecode_prop("rt_count_1 = MALI_POSITIVE(%d)", fb
->rt_count_1
+ 1);
782 pandecode_prop("rt_count_2 = %d", fb
->rt_count_2
);
784 pandecode_log(".mfbd_flags = ");
785 pandecode_log_decoded_flags(mfbd_flag_info
, fb
->mfbd_flags
);
786 pandecode_log_cont(",\n");
788 pandecode_prop("clear_stencil = 0x%x", fb
->clear_stencil
);
789 pandecode_prop("clear_depth = %f", fb
->clear_depth
);
791 pandecode_prop("unknown2 = 0x%x", fb
->unknown2
);
792 MEMORY_PROP(fb
, scratchpad
);
793 const struct midgard_tiler_descriptor t
= fb
->tiler
;
794 pandecode_midgard_tiler_descriptor(&t
);
796 if (fb
->zero3
|| fb
->zero4
) {
797 pandecode_msg("framebuffer zeros tripped\n");
798 pandecode_prop("zero3 = 0x%" PRIx32
, fb
->zero3
);
799 pandecode_prop("zero4 = 0x%" PRIx32
, fb
->zero4
);
803 pandecode_log("};\n");
805 gpu_va
+= sizeof(struct bifrost_framebuffer
);
807 if ((fb
->mfbd_flags
& MALI_MFBD_EXTRA
) && with_render_targets
) {
808 mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
809 const struct bifrost_fb_extra
*PANDECODE_PTR_VAR(fbx
, mem
, (mali_ptr
) gpu_va
);
811 pandecode_log("struct bifrost_fb_extra fb_extra_%"PRIx64
"_%d = {\n", gpu_va
, job_no
);
814 MEMORY_PROP(fbx
, checksum
);
816 if (fbx
->checksum_stride
)
817 pandecode_prop("checksum_stride = %d", fbx
->checksum_stride
);
819 pandecode_log(".flags = ");
820 pandecode_log_decoded_flags(mfbd_extra_flag_info
, fbx
->flags
);
821 pandecode_log_cont(",\n");
823 if (fbx
->flags
& MALI_EXTRA_AFBC_ZS
) {
824 pandecode_log(".ds_afbc = {\n");
827 MEMORY_PROP_DIR(fbx
->ds_afbc
, depth_stencil_afbc_metadata
);
828 pandecode_prop("depth_stencil_afbc_stride = %d",
829 fbx
->ds_afbc
.depth_stencil_afbc_stride
);
830 MEMORY_PROP_DIR(fbx
->ds_afbc
, depth_stencil
);
832 if (fbx
->ds_afbc
.zero1
|| fbx
->ds_afbc
.padding
) {
833 pandecode_msg("Depth/stencil AFBC zeros tripped\n");
834 pandecode_prop("zero1 = 0x%" PRIx32
,
836 pandecode_prop("padding = 0x%" PRIx64
,
837 fbx
->ds_afbc
.padding
);
841 pandecode_log("},\n");
843 pandecode_log(".ds_linear = {\n");
846 if (fbx
->ds_linear
.depth
) {
847 MEMORY_PROP_DIR(fbx
->ds_linear
, depth
);
848 pandecode_prop("depth_stride = %d",
849 fbx
->ds_linear
.depth_stride
);
852 if (fbx
->ds_linear
.stencil
) {
853 MEMORY_PROP_DIR(fbx
->ds_linear
, stencil
);
854 pandecode_prop("stencil_stride = %d",
855 fbx
->ds_linear
.stencil_stride
);
858 if (fbx
->ds_linear
.depth_stride_zero
||
859 fbx
->ds_linear
.stencil_stride_zero
||
860 fbx
->ds_linear
.zero1
|| fbx
->ds_linear
.zero2
) {
861 pandecode_msg("Depth/stencil zeros tripped\n");
862 pandecode_prop("depth_stride_zero = 0x%x",
863 fbx
->ds_linear
.depth_stride_zero
);
864 pandecode_prop("stencil_stride_zero = 0x%x",
865 fbx
->ds_linear
.stencil_stride_zero
);
866 pandecode_prop("zero1 = 0x%" PRIx32
,
867 fbx
->ds_linear
.zero1
);
868 pandecode_prop("zero2 = 0x%" PRIx32
,
869 fbx
->ds_linear
.zero2
);
873 pandecode_log("},\n");
876 if (fbx
->zero3
|| fbx
->zero4
) {
877 pandecode_msg("fb_extra zeros tripped\n");
878 pandecode_prop("zero3 = 0x%" PRIx64
, fbx
->zero3
);
879 pandecode_prop("zero4 = 0x%" PRIx64
, fbx
->zero4
);
883 pandecode_log("};\n");
885 gpu_va
+= sizeof(struct bifrost_fb_extra
);
888 if (with_render_targets
)
889 pandecode_render_target(gpu_va
, job_no
, fb
);
891 /* Passback the render target count */
892 return MALI_NEGATIVE(fb
->rt_count_1
);
895 /* Just add a comment decoding the shift/odd fields forming the padded vertices
899 pandecode_padded_vertices(unsigned shift
, unsigned k
)
901 unsigned odd
= 2*k
+ 1;
902 unsigned pot
= 1 << shift
;
903 pandecode_msg("padded_num_vertices = %d\n", odd
* pot
);
906 /* Given a magic divisor, recover what we were trying to divide by.
908 * Let m represent the magic divisor. By definition, m is an element on Z, whre
909 * 0 <= m < 2^N, for N bits in m.
911 * Let q represent the number we would like to divide by.
913 * By definition of a magic divisor for N-bit unsigned integers (a number you
914 * multiply by to magically get division), m is a number such that:
916 * (m * x) & (2^N - 1) = floor(x/q).
917 * for all x on Z where 0 <= x < 2^N
919 * Ignore the case where any of the above values equals zero; it is irrelevant
920 * for our purposes (instanced arrays).
922 * Choose x = q. Then:
924 * (m * x) & (2^N - 1) = floor(x/q).
925 * (m * q) & (2^N - 1) = floor(q/q).
927 * floor(q/q) = floor(1) = 1, therefore:
929 * (m * q) & (2^N - 1) = 1
931 * Recall the identity that the bitwise AND of one less than a power-of-two
932 * equals the modulo with that power of two, i.e. for all x:
934 * x & (2^N - 1) = x % N
940 * By definition, a modular multiplicative inverse of a number m is the number
941 * q such that with respect to a modulos M:
945 * Therefore, q is the modular multiplicative inverse of m with modulus 2^N.
950 pandecode_magic_divisor(uint32_t magic
, unsigned shift
, unsigned orig_divisor
, unsigned extra
)
953 /* Compute the modular inverse of `magic` with respect to 2^(32 -
954 * shift) the most lame way possible... just repeatedly add.
955 * Asymptoptically slow but nobody cares in practice, unless you have
956 * massive numbers of vertices or high divisors. */
958 unsigned inverse
= 0;
960 /* Magic implicitly has the highest bit set */
963 /* Depending on rounding direction */
968 uint32_t product
= magic
* inverse
;
980 pandecode_msg("dividing by %d (maybe off by two)\n", inverse
);
982 /* Recall we're supposed to divide by (gl_level_divisor *
983 * padded_num_vertices) */
985 unsigned padded_num_vertices
= inverse
/ orig_divisor
;
987 pandecode_msg("padded_num_vertices = %d\n", padded_num_vertices
);
992 pandecode_attributes(const struct pandecode_mapped_memory
*mem
,
993 mali_ptr addr
, int job_no
, char *suffix
,
994 int count
, bool varying
)
996 char *prefix
= varying
? "varyings" : "attributes";
998 union mali_attr
*attr
= pandecode_fetch_gpu_mem(mem
, addr
, sizeof(union mali_attr
) * count
);
1001 snprintf(base
, sizeof(base
), "%s_data_%d%s", prefix
, job_no
, suffix
);
1003 for (int i
= 0; i
< count
; ++i
) {
1004 enum mali_attr_mode mode
= attr
[i
].elements
& 7;
1006 if (mode
== MALI_ATTR_UNUSED
)
1009 mali_ptr raw_elements
= attr
[i
].elements
& ~7;
1011 /* TODO: Do we maybe want to dump the attribute values
1012 * themselves given the specified format? Or is that too hard?
1015 char *a
= pointer_as_memory_reference(raw_elements
);
1016 pandecode_log("mali_ptr %s_%d_p = %s;\n", base
, i
, a
);
1020 pandecode_log("union mali_attr %s_%d[] = {\n", prefix
, job_no
);
1023 for (int i
= 0; i
< count
; ++i
) {
1024 pandecode_log("{\n");
1027 unsigned mode
= attr
[i
].elements
& 7;
1028 pandecode_prop("elements = (%s_%d_p) | %s", base
, i
, pandecode_attr_mode(mode
));
1029 pandecode_prop("shift = %d", attr
[i
].shift
);
1030 pandecode_prop("extra_flags = %d", attr
[i
].extra_flags
);
1031 pandecode_prop("stride = 0x%" PRIx32
, attr
[i
].stride
);
1032 pandecode_prop("size = 0x%" PRIx32
, attr
[i
].size
);
1034 /* Decode further where possible */
1036 if (mode
== MALI_ATTR_MODULO
) {
1037 pandecode_padded_vertices(
1039 attr
[i
].extra_flags
);
1043 pandecode_log("}, \n");
1045 if (mode
== MALI_ATTR_NPOT_DIVIDE
) {
1047 pandecode_log("{\n");
1049 pandecode_prop("unk = 0x%x", attr
[i
].unk
);
1050 pandecode_prop("magic_divisor = 0x%08x", attr
[i
].magic_divisor
);
1051 if (attr
[i
].zero
!= 0)
1052 pandecode_prop("zero = 0x%x /* XXX zero tripped */", attr
[i
].zero
);
1053 pandecode_prop("divisor = %d", attr
[i
].divisor
);
1054 pandecode_magic_divisor(attr
[i
].magic_divisor
, attr
[i
- 1].shift
, attr
[i
].divisor
, attr
[i
- 1].extra_flags
);
1056 pandecode_log("}, \n");
1062 pandecode_log("};\n");
1066 pandecode_shader_address(const char *name
, mali_ptr ptr
)
1068 /* TODO: Decode flags */
1069 mali_ptr shader_ptr
= ptr
& ~15;
1071 char *a
= pointer_as_memory_reference(shader_ptr
);
1072 pandecode_prop("%s = (%s) | %d", name
, a
, (int) (ptr
& 15));
1079 all_zero(unsigned *buffer
, unsigned count
)
1081 for (unsigned i
= 0; i
< count
; ++i
) {
1090 pandecode_stencil(const char *name
, const struct mali_stencil_test
*stencil
)
1092 if (all_zero((unsigned *) stencil
, sizeof(stencil
) / sizeof(unsigned)))
1095 const char *func
= pandecode_func(stencil
->func
);
1096 const char *sfail
= pandecode_stencil_op(stencil
->sfail
);
1097 const char *dpfail
= pandecode_stencil_op(stencil
->dpfail
);
1098 const char *dppass
= pandecode_stencil_op(stencil
->dppass
);
1101 pandecode_msg("Stencil zero tripped: %X\n", stencil
->zero
);
1103 pandecode_log(".stencil_%s = {\n", name
);
1105 pandecode_prop("ref = %d", stencil
->ref
);
1106 pandecode_prop("mask = 0x%02X", stencil
->mask
);
1107 pandecode_prop("func = %s", func
);
1108 pandecode_prop("sfail = %s", sfail
);
1109 pandecode_prop("dpfail = %s", dpfail
);
1110 pandecode_prop("dppass = %s", dppass
);
1112 pandecode_log("},\n");
1116 pandecode_blend_equation(const struct mali_blend_equation
*blend
)
1119 pandecode_msg("Blend zero tripped: %X\n", blend
->zero1
);
1121 pandecode_log(".equation = {\n");
1124 pandecode_prop("rgb_mode = 0x%X", blend
->rgb_mode
);
1125 pandecode_prop("alpha_mode = 0x%X", blend
->alpha_mode
);
1127 pandecode_log(".color_mask = ");
1128 pandecode_log_decoded_flags(mask_flag_info
, blend
->color_mask
);
1129 pandecode_log_cont(",\n");
1132 pandecode_log("},\n");
1135 /* Decodes a Bifrost blend constant. See the notes in bifrost_blend_rt */
1138 decode_bifrost_constant(u16 constant
)
1140 float lo
= (float) (constant
& 0xFF);
1141 float hi
= (float) (constant
>> 8);
1143 return (hi
/ 255.0) + (lo
/ 65535.0);
1147 pandecode_bifrost_blend(void *descs
, int job_no
, int rt_no
)
1149 struct bifrost_blend_rt
*b
=
1150 ((struct bifrost_blend_rt
*) descs
) + rt_no
;
1152 pandecode_log("struct bifrost_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1155 pandecode_prop("flags = 0x%" PRIx16
, b
->flags
);
1156 pandecode_prop("constant = 0x%" PRIx8
" /* %f */",
1157 b
->constant
, decode_bifrost_constant(b
->constant
));
1159 /* TODO figure out blend shader enable bit */
1160 pandecode_blend_equation(&b
->equation
);
1161 pandecode_prop("unk2 = 0x%" PRIx16
, b
->unk2
);
1162 pandecode_prop("index = 0x%" PRIx16
, b
->index
);
1163 pandecode_prop("shader = 0x%" PRIx32
, b
->shader
);
1166 pandecode_log("},\n");
1172 pandecode_midgard_blend(union midgard_blend
*blend
, bool is_shader
)
1174 if (all_zero((unsigned *) blend
, sizeof(blend
) / sizeof(unsigned)))
1177 pandecode_log(".blend = {\n");
1181 pandecode_shader_address("shader", blend
->shader
);
1183 pandecode_blend_equation(&blend
->equation
);
1184 pandecode_prop("constant = %f", blend
->constant
);
1188 pandecode_log("},\n");
1190 /* Return blend shader to disassemble if present */
1191 return is_shader
? (blend
->shader
& ~0xF) : 0;
1195 pandecode_midgard_blend_mrt(void *descs
, int job_no
, int rt_no
)
1197 struct midgard_blend_rt
*b
=
1198 ((struct midgard_blend_rt
*) descs
) + rt_no
;
1200 /* Flags determine presence of blend shader */
1201 bool is_shader
= (b
->flags
& 0xF) >= 0x2;
1203 pandecode_log("struct midgard_blend_rt blend_rt_%d_%d = {\n", job_no
, rt_no
);
1206 pandecode_prop("flags = 0x%" PRIx64
, b
->flags
);
1208 union midgard_blend blend
= b
->blend
;
1209 mali_ptr shader
= pandecode_midgard_blend(&blend
, is_shader
);
1212 pandecode_log("};\n");
1218 pandecode_attribute_meta(int job_no
, int count
, const struct mali_vertex_tiler_postfix
*v
, bool varying
, char *suffix
)
1221 char *prefix
= varying
? "varying" : "attribute";
1222 unsigned max_index
= 0;
1223 snprintf(base
, sizeof(base
), "%s_meta", prefix
);
1225 pandecode_log("struct mali_attr_meta %s_%d%s[] = {\n", base
, job_no
, suffix
);
1228 struct mali_attr_meta
*attr_meta
;
1229 mali_ptr p
= varying
? (v
->varying_meta
& ~0xF) : v
->attribute_meta
;
1231 struct pandecode_mapped_memory
*attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
);
1233 for (int i
= 0; i
< count
; ++i
, p
+= sizeof(struct mali_attr_meta
)) {
1234 attr_meta
= pandecode_fetch_gpu_mem(attr_mem
, p
,
1237 pandecode_log("{\n");
1239 pandecode_prop("index = %d", attr_meta
->index
);
1241 if (attr_meta
->index
> max_index
)
1242 max_index
= attr_meta
->index
;
1243 pandecode_swizzle(attr_meta
->swizzle
);
1244 pandecode_prop("format = %s", pandecode_format(attr_meta
->format
));
1246 pandecode_prop("unknown1 = 0x%" PRIx64
, (u64
) attr_meta
->unknown1
);
1247 pandecode_prop("unknown3 = 0x%" PRIx64
, (u64
) attr_meta
->unknown3
);
1248 pandecode_prop("src_offset = %d", attr_meta
->src_offset
);
1250 pandecode_log("},\n");
1255 pandecode_log("};\n");
1257 return count
? (max_index
+ 1) : 0;
1261 pandecode_indices(uintptr_t pindices
, uint32_t index_count
, int job_no
)
1263 struct pandecode_mapped_memory
*imem
= pandecode_find_mapped_gpu_mem_containing(pindices
);
1266 /* Indices are literally just a u32 array :) */
1268 uint32_t *PANDECODE_PTR_VAR(indices
, imem
, pindices
);
1270 pandecode_log("uint32_t indices_%d[] = {\n", job_no
);
1273 for (unsigned i
= 0; i
< (index_count
+ 1); i
+= 3)
1274 pandecode_log("%d, %d, %d,\n",
1280 pandecode_log("};\n");
1284 /* return bits [lo, hi) of word */
1286 bits(u32 word
, u32 lo
, u32 hi
)
1289 return word
; // avoid undefined behavior with the shift
1291 return (word
>> lo
) & ((1 << (hi
- lo
)) - 1);
1295 pandecode_vertex_tiler_prefix(struct mali_vertex_tiler_prefix
*p
, int job_no
)
1297 pandecode_log_cont("{\n");
1300 pandecode_prop("invocation_count = 0x%" PRIx32
, p
->invocation_count
);
1301 pandecode_prop("size_y_shift = %d", p
->size_y_shift
);
1302 pandecode_prop("size_z_shift = %d", p
->size_z_shift
);
1303 pandecode_prop("workgroups_x_shift = %d", p
->workgroups_x_shift
);
1304 pandecode_prop("workgroups_y_shift = %d", p
->workgroups_y_shift
);
1305 pandecode_prop("workgroups_z_shift = %d", p
->workgroups_z_shift
);
1306 pandecode_prop("workgroups_x_shift_2 = 0x%" PRIx32
, p
->workgroups_x_shift_2
);
1308 /* Decode invocation_count. See the comment before the definition of
1309 * invocation_count for an explanation.
1311 pandecode_msg("size: (%d, %d, %d)\n",
1312 bits(p
->invocation_count
, 0, p
->size_y_shift
) + 1,
1313 bits(p
->invocation_count
, p
->size_y_shift
, p
->size_z_shift
) + 1,
1314 bits(p
->invocation_count
, p
->size_z_shift
,
1315 p
->workgroups_x_shift
) + 1);
1316 pandecode_msg("workgroups: (%d, %d, %d)\n",
1317 bits(p
->invocation_count
, p
->workgroups_x_shift
,
1318 p
->workgroups_y_shift
) + 1,
1319 bits(p
->invocation_count
, p
->workgroups_y_shift
,
1320 p
->workgroups_z_shift
) + 1,
1321 bits(p
->invocation_count
, p
->workgroups_z_shift
,
1325 if (p
->unknown_draw
)
1326 pandecode_prop("unknown_draw = 0x%" PRIx32
, p
->unknown_draw
);
1328 pandecode_prop("workgroups_x_shift_3 = 0x%" PRIx32
, p
->workgroups_x_shift_3
);
1330 pandecode_prop("draw_mode = %s", pandecode_draw_mode(p
->draw_mode
));
1332 /* Index count only exists for tiler jobs anyway */
1335 pandecode_prop("index_count = MALI_POSITIVE(%" PRId32
")", p
->index_count
+ 1);
1337 if (p
->negative_start
)
1338 pandecode_prop("negative_start = %d", p
->negative_start
);
1340 DYN_MEMORY_PROP(p
, job_no
, indices
);
1343 pandecode_msg("Zero tripped\n");
1344 pandecode_prop("zero1 = 0x%" PRIx32
, p
->zero1
);
1348 pandecode_log("},\n");
1352 pandecode_uniform_buffers(mali_ptr pubufs
, int ubufs_count
, int job_no
)
1354 struct pandecode_mapped_memory
*umem
= pandecode_find_mapped_gpu_mem_containing(pubufs
);
1356 struct mali_uniform_buffer_meta
*PANDECODE_PTR_VAR(ubufs
, umem
, pubufs
);
1358 for (int i
= 0; i
< ubufs_count
; i
++) {
1359 mali_ptr ptr
= ubufs
[i
].ptr
<< 2;
1360 struct pandecode_mapped_memory
*umem2
= pandecode_find_mapped_gpu_mem_containing(ptr
);
1361 uint32_t *PANDECODE_PTR_VAR(ubuf
, umem2
, ptr
);
1363 snprintf(name
, sizeof(name
), "ubuf_%d", i
);
1364 /* The blob uses ubuf 0 to upload internal stuff and
1365 * uniforms that won't fit/are accessed indirectly, so
1366 * it puts it in the batchbuffer.
1368 pandecode_log("uint32_t %s_%d[] = {\n", name
, job_no
);
1371 for (int j
= 0; j
<= ubufs
[i
].size
; j
++) {
1372 for (int k
= 0; k
< 4; k
++) {
1374 pandecode_log("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1376 pandecode_log_cont("0x%"PRIx32
", ", ubuf
[4 * j
+ k
]);
1380 pandecode_log_cont("\n");
1384 pandecode_log("};\n");
1387 pandecode_log("struct mali_uniform_buffer_meta uniform_buffers_%"PRIx64
"_%d[] = {\n",
1391 for (int i
= 0; i
< ubufs_count
; i
++) {
1392 pandecode_log("{\n");
1394 pandecode_prop("size = MALI_POSITIVE(%d)", ubufs
[i
].size
+ 1);
1395 pandecode_prop("ptr = ubuf_%d_%d_p >> 2", i
, job_no
);
1397 pandecode_log("},\n");
1401 pandecode_log("};\n");
1405 pandecode_scratchpad(uintptr_t pscratchpad
, int job_no
, char *suffix
)
1408 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(pscratchpad
);
1410 struct bifrost_scratchpad
*PANDECODE_PTR_VAR(scratchpad
, mem
, pscratchpad
);
1412 if (scratchpad
->zero
)
1413 pandecode_msg("XXX scratchpad zero tripped");
1415 pandecode_log("struct bifrost_scratchpad scratchpad_%"PRIx64
"_%d%s = {\n", pscratchpad
, job_no
, suffix
);
1418 pandecode_prop("flags = 0x%x", scratchpad
->flags
);
1419 MEMORY_PROP(scratchpad
, gpu_scratchpad
);
1422 pandecode_log("};\n");
1426 pandecode_shader_disassemble(mali_ptr shader_ptr
, int shader_no
, int type
,
1429 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(shader_ptr
);
1430 uint8_t *PANDECODE_PTR_VAR(code
, mem
, shader_ptr
);
1432 /* Compute maximum possible size */
1433 size_t sz
= mem
->length
- (shader_ptr
- mem
->gpu_va
);
1435 /* Print some boilerplate to clearly denote the assembly (which doesn't
1436 * obey indentation rules), and actually do the disassembly! */
1441 disassemble_bifrost(code
, sz
, false);
1443 disassemble_midgard(code
, sz
);
1450 pandecode_vertex_tiler_postfix_pre(const struct mali_vertex_tiler_postfix
*p
,
1451 int job_no
, enum mali_job_type job_type
,
1452 char *suffix
, bool is_bifrost
)
1454 mali_ptr shader_meta_ptr
= (u64
) (uintptr_t) (p
->_shader_upper
<< 4);
1455 struct pandecode_mapped_memory
*attr_mem
;
1457 unsigned rt_count
= 1;
1459 /* On Bifrost, since the tiler heap (for tiler jobs) and the scratchpad
1460 * are the only things actually needed from the FBD, vertex/tiler jobs
1461 * no longer reference the FBD -- instead, this field points to some
1462 * info about the scratchpad.
1465 pandecode_scratchpad(p
->framebuffer
& ~FBD_TYPE
, job_no
, suffix
);
1466 else if (p
->framebuffer
& MALI_MFBD
)
1467 rt_count
= pandecode_mfbd_bfr((u64
) ((uintptr_t) p
->framebuffer
) & FBD_MASK
, job_no
, false);
1468 else if (job_type
== JOB_TYPE_COMPUTE
)
1469 pandecode_compute_fbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1471 pandecode_sfbd((u64
) (uintptr_t) p
->framebuffer
, job_no
);
1473 int varying_count
= 0, attribute_count
= 0, uniform_count
= 0, uniform_buffer_count
= 0;
1474 int texture_count
= 0, sampler_count
= 0;
1476 if (shader_meta_ptr
) {
1477 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(shader_meta_ptr
);
1478 struct mali_shader_meta
*PANDECODE_PTR_VAR(s
, smem
, shader_meta_ptr
);
1480 pandecode_log("struct mali_shader_meta shader_meta_%"PRIx64
"_%d%s = {\n", shader_meta_ptr
, job_no
, suffix
);
1483 /* Save for dumps */
1484 attribute_count
= s
->attribute_count
;
1485 varying_count
= s
->varying_count
;
1486 texture_count
= s
->texture_count
;
1487 sampler_count
= s
->sampler_count
;
1490 uniform_count
= s
->bifrost2
.uniform_count
;
1491 uniform_buffer_count
= s
->bifrost1
.uniform_buffer_count
;
1493 uniform_count
= s
->midgard1
.uniform_count
;
1494 uniform_buffer_count
= s
->midgard1
.uniform_buffer_count
;
1497 mali_ptr shader_ptr
= pandecode_shader_address("shader", s
->shader
);
1499 pandecode_prop("texture_count = %" PRId16
, s
->texture_count
);
1500 pandecode_prop("sampler_count = %" PRId16
, s
->sampler_count
);
1501 pandecode_prop("attribute_count = %" PRId16
, s
->attribute_count
);
1502 pandecode_prop("varying_count = %" PRId16
, s
->varying_count
);
1505 pandecode_log(".bifrost1 = {\n");
1508 pandecode_prop("uniform_buffer_count = %" PRId32
, s
->bifrost1
.uniform_buffer_count
);
1509 pandecode_prop("unk1 = 0x%" PRIx32
, s
->bifrost1
.unk1
);
1512 pandecode_log("},\n");
1514 pandecode_log(".midgard1 = {\n");
1517 pandecode_prop("uniform_count = %" PRId16
, s
->midgard1
.uniform_count
);
1518 pandecode_prop("uniform_buffer_count = %" PRId16
, s
->midgard1
.uniform_buffer_count
);
1519 pandecode_prop("work_count = %" PRId16
, s
->midgard1
.work_count
);
1521 pandecode_log(".flags = ");
1522 pandecode_log_decoded_flags(shader_midgard1_flag_info
, s
->midgard1
.flags
);
1523 pandecode_log_cont(",\n");
1525 pandecode_prop("unknown2 = 0x%" PRIx32
, s
->midgard1
.unknown2
);
1528 pandecode_log("},\n");
1531 if (s
->depth_units
|| s
->depth_factor
) {
1532 pandecode_prop("depth_factor = %f", s
->depth_factor
);
1533 pandecode_prop("depth_units = %f", s
->depth_units
);
1536 if (s
->alpha_coverage
) {
1537 bool invert_alpha_coverage
= s
->alpha_coverage
& 0xFFF0;
1538 uint16_t inverted_coverage
= invert_alpha_coverage
? ~s
->alpha_coverage
: s
->alpha_coverage
;
1540 pandecode_prop("alpha_coverage = %sMALI_ALPHA_COVERAGE(%f)",
1541 invert_alpha_coverage
? "~" : "",
1542 MALI_GET_ALPHA_COVERAGE(inverted_coverage
));
1545 if (s
->unknown2_3
|| s
->unknown2_4
) {
1546 pandecode_log(".unknown2_3 = ");
1548 int unknown2_3
= s
->unknown2_3
;
1549 int unknown2_4
= s
->unknown2_4
;
1551 /* We're not quite sure what these flags mean without the depth test, if anything */
1553 if (unknown2_3
& (MALI_DEPTH_TEST
| MALI_DEPTH_FUNC_MASK
)) {
1554 const char *func
= pandecode_func(MALI_GET_DEPTH_FUNC(unknown2_3
));
1555 unknown2_3
&= ~MALI_DEPTH_FUNC_MASK
;
1557 pandecode_log_cont("MALI_DEPTH_FUNC(%s) | ", func
);
1560 pandecode_log_decoded_flags(u3_flag_info
, unknown2_3
);
1561 pandecode_log_cont(",\n");
1563 pandecode_log(".unknown2_4 = ");
1564 pandecode_log_decoded_flags(u4_flag_info
, unknown2_4
);
1565 pandecode_log_cont(",\n");
1568 if (s
->stencil_mask_front
|| s
->stencil_mask_back
) {
1569 pandecode_prop("stencil_mask_front = 0x%02X", s
->stencil_mask_front
);
1570 pandecode_prop("stencil_mask_back = 0x%02X", s
->stencil_mask_back
);
1573 pandecode_stencil("front", &s
->stencil_front
);
1574 pandecode_stencil("back", &s
->stencil_back
);
1577 pandecode_log(".bifrost2 = {\n");
1580 pandecode_prop("unk3 = 0x%" PRIx32
, s
->bifrost2
.unk3
);
1581 pandecode_prop("preload_regs = 0x%" PRIx32
, s
->bifrost2
.preload_regs
);
1582 pandecode_prop("uniform_count = %" PRId32
, s
->bifrost2
.uniform_count
);
1583 pandecode_prop("unk4 = 0x%" PRIx32
, s
->bifrost2
.unk4
);
1586 pandecode_log("},\n");
1587 } else if (s
->midgard2
.unknown2_7
) {
1588 pandecode_log(".midgard2 = {\n");
1591 pandecode_prop("unknown2_7 = 0x%" PRIx32
, s
->midgard2
.unknown2_7
);
1593 pandecode_log("},\n");
1597 pandecode_prop("unknown2_8 = 0x%" PRIx32
, s
->unknown2_8
);
1600 /* TODO: Blend shaders routing/disasm */
1602 union midgard_blend blend
= s
->blend
;
1603 pandecode_midgard_blend(&blend
, false);
1607 pandecode_log("};\n");
1609 /* MRT blend fields are used whenever MFBD is used, with
1610 * per-RT descriptors */
1612 if (job_type
== JOB_TYPE_TILER
) {
1613 void* blend_base
= (void *) (s
+ 1);
1615 for (unsigned i
= 0; i
< rt_count
; i
++) {
1616 mali_ptr shader
= 0;
1619 shader
= pandecode_bifrost_blend(blend_base
, job_no
, i
);
1621 shader
= pandecode_midgard_blend_mrt(blend_base
, job_no
, i
);
1624 pandecode_shader_disassemble(shader
, job_no
, job_type
, false);
1628 if (shader_ptr
& ~0xF)
1629 pandecode_shader_disassemble(shader_ptr
, job_no
, job_type
, is_bifrost
);
1631 pandecode_msg("<no shader>\n");
1634 struct pandecode_mapped_memory
*fmem
= pandecode_find_mapped_gpu_mem_containing(p
->viewport
);
1635 struct mali_viewport
*PANDECODE_PTR_VAR(f
, fmem
, p
->viewport
);
1637 pandecode_log("struct mali_viewport viewport_%"PRIx64
"_%d%s = {\n", p
->viewport
, job_no
, suffix
);
1640 pandecode_prop("clip_minx = %f", f
->clip_minx
);
1641 pandecode_prop("clip_miny = %f", f
->clip_miny
);
1642 pandecode_prop("clip_minz = %f", f
->clip_minz
);
1643 pandecode_prop("clip_maxx = %f", f
->clip_maxx
);
1644 pandecode_prop("clip_maxy = %f", f
->clip_maxy
);
1645 pandecode_prop("clip_maxz = %f", f
->clip_maxz
);
1647 /* Only the higher coordinates are MALI_POSITIVE scaled */
1649 pandecode_prop("viewport0 = { %d, %d }",
1650 f
->viewport0
[0], f
->viewport0
[1]);
1652 pandecode_prop("viewport1 = { MALI_POSITIVE(%d), MALI_POSITIVE(%d) }",
1653 f
->viewport1
[0] + 1, f
->viewport1
[1] + 1);
1656 pandecode_log("};\n");
1659 if (p
->attribute_meta
) {
1660 unsigned max_attr_index
= pandecode_attribute_meta(job_no
, attribute_count
, p
, false, suffix
);
1662 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->attributes
);
1663 pandecode_attributes(attr_mem
, p
->attributes
, job_no
, suffix
, max_attr_index
+ 1, false);
1666 /* Varyings are encoded like attributes but not actually sent; we just
1667 * pass a zero buffer with the right stride/size set, (or whatever)
1668 * since the GPU will write to it itself */
1670 if (p
->varying_meta
) {
1671 varying_count
= pandecode_attribute_meta(job_no
, varying_count
, p
, true, suffix
);
1675 attr_mem
= pandecode_find_mapped_gpu_mem_containing(p
->varyings
);
1677 /* Number of descriptors depends on whether there are
1678 * non-internal varyings */
1680 pandecode_attributes(attr_mem
, p
->varyings
, job_no
, suffix
, varying_count
, true);
1683 bool is_compute
= job_type
== JOB_TYPE_COMPUTE
;
1685 if (p
->uniforms
&& !is_compute
) {
1686 int rows
= uniform_count
, width
= 4;
1687 size_t sz
= rows
* width
* sizeof(float);
1689 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1690 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1691 u32
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1693 pandecode_log("u32 uniforms_%d%s[] = {\n", job_no
, suffix
);
1697 for (int row
= 0; row
< rows
; row
++) {
1698 for (int i
= 0; i
< width
; i
++) {
1699 u32 v
= uniforms
[i
];
1701 memcpy(&f
, &v
, sizeof(v
));
1702 pandecode_log_cont("%X /* %f */, ", v
, f
);
1705 pandecode_log_cont("\n");
1711 pandecode_log("};\n");
1712 } else if (p
->uniforms
) {
1713 int rows
= uniform_count
* 2;
1714 size_t sz
= rows
* sizeof(mali_ptr
);
1716 struct pandecode_mapped_memory
*uniform_mem
= pandecode_find_mapped_gpu_mem_containing(p
->uniforms
);
1717 pandecode_fetch_gpu_mem(uniform_mem
, p
->uniforms
, sz
);
1718 mali_ptr
*PANDECODE_PTR_VAR(uniforms
, uniform_mem
, p
->uniforms
);
1720 pandecode_log("mali_ptr uniforms_%d%s[] = {\n", job_no
, suffix
);
1724 for (int row
= 0; row
< rows
; row
++) {
1725 char *a
= pointer_as_memory_reference(uniforms
[row
]);
1726 pandecode_log("%s,\n", a
);
1731 pandecode_log("};\n");
1735 if (p
->uniform_buffers
) {
1736 pandecode_uniform_buffers(p
->uniform_buffers
, uniform_buffer_count
, job_no
);
1739 if (p
->texture_trampoline
) {
1740 struct pandecode_mapped_memory
*mmem
= pandecode_find_mapped_gpu_mem_containing(p
->texture_trampoline
);
1743 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
);
1745 pandecode_log("uint64_t texture_trampoline_%"PRIx64
"_%d[] = {\n", p
->texture_trampoline
, job_no
);
1748 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1749 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1750 char *a
= pointer_as_memory_reference(*u
);
1751 pandecode_log("%s,\n", a
);
1756 pandecode_log("};\n");
1758 /* Now, finally, descend down into the texture descriptor */
1759 for (int tex
= 0; tex
< texture_count
; ++tex
) {
1760 mali_ptr
*PANDECODE_PTR_VAR(u
, mmem
, p
->texture_trampoline
+ tex
* sizeof(mali_ptr
));
1761 struct pandecode_mapped_memory
*tmem
= pandecode_find_mapped_gpu_mem_containing(*u
);
1764 struct mali_texture_descriptor
*PANDECODE_PTR_VAR(t
, tmem
, *u
);
1766 pandecode_log("struct mali_texture_descriptor texture_descriptor_%"PRIx64
"_%d_%d = {\n", *u
, job_no
, tex
);
1769 pandecode_prop("width = MALI_POSITIVE(%" PRId16
")", t
->width
+ 1);
1770 pandecode_prop("height = MALI_POSITIVE(%" PRId16
")", t
->height
+ 1);
1771 pandecode_prop("depth = MALI_POSITIVE(%" PRId16
")", t
->depth
+ 1);
1772 pandecode_prop("array_size = MALI_POSITIVE(%" PRId16
")", t
->array_size
+ 1);
1773 pandecode_prop("unknown3 = %" PRId16
, t
->unknown3
);
1774 pandecode_prop("unknown3A = %" PRId8
, t
->unknown3A
);
1775 pandecode_prop("nr_mipmap_levels = %" PRId8
, t
->nr_mipmap_levels
);
1777 struct mali_texture_format f
= t
->format
;
1779 pandecode_log(".format = {\n");
1782 pandecode_swizzle(f
.swizzle
);
1783 pandecode_prop("format = %s", pandecode_format(f
.format
));
1784 pandecode_prop("type = %s", pandecode_texture_type(f
.type
));
1785 pandecode_prop("srgb = %" PRId32
, f
.srgb
);
1786 pandecode_prop("unknown1 = %" PRId32
, f
.unknown1
);
1787 pandecode_prop("usage2 = 0x%" PRIx32
, f
.usage2
);
1790 pandecode_log("},\n");
1792 pandecode_swizzle(t
->swizzle
);
1794 if (t
->swizzle_zero
) {
1795 /* Shouldn't happen */
1796 pandecode_msg("Swizzle zero tripped but replay will be fine anyway");
1797 pandecode_prop("swizzle_zero = %d", t
->swizzle_zero
);
1800 pandecode_prop("unknown3 = 0x%" PRIx32
, t
->unknown3
);
1802 pandecode_prop("unknown5 = 0x%" PRIx32
, t
->unknown5
);
1803 pandecode_prop("unknown6 = 0x%" PRIx32
, t
->unknown6
);
1804 pandecode_prop("unknown7 = 0x%" PRIx32
, t
->unknown7
);
1806 pandecode_log(".payload = {\n");
1809 /* A bunch of bitmap pointers follow.
1810 * We work out the correct number,
1811 * based on the mipmap/cubemap
1812 * properties, but dump extra
1813 * possibilities to futureproof */
1815 int bitmap_count
= MALI_NEGATIVE(t
->nr_mipmap_levels
);
1816 bool manual_stride
= f
.usage2
& MALI_TEX_MANUAL_STRIDE
;
1818 /* Miptree for each face */
1819 if (f
.type
== MALI_TEX_CUBE
)
1822 /* Array of textures */
1823 bitmap_count
*= MALI_NEGATIVE(t
->array_size
);
1825 /* Stride for each element */
1829 /* Sanity check the size */
1830 int max_count
= sizeof(t
->payload
) / sizeof(t
->payload
[0]);
1831 assert (bitmap_count
<= max_count
);
1833 /* Dump more to be safe, but not _that_ much more */
1834 int safe_count
= MIN2(bitmap_count
* 2, max_count
);
1836 for (int i
= 0; i
< safe_count
; ++i
) {
1837 char *prefix
= (i
>= bitmap_count
) ? "// " : "";
1839 /* How we dump depends if this is a stride or a pointer */
1841 if ((f
.usage2
& MALI_TEX_MANUAL_STRIDE
) && (i
& 1)) {
1842 /* signed 32-bit snuck in as a 64-bit pointer */
1843 uint64_t stride_set
= t
->payload
[i
];
1844 uint32_t clamped_stride
= stride_set
;
1845 int32_t stride
= clamped_stride
;
1846 assert(stride_set
== clamped_stride
);
1847 pandecode_log("%s(mali_ptr) %d /* stride */, \n", prefix
, stride
);
1849 char *a
= pointer_as_memory_reference(t
->payload
[i
]);
1850 pandecode_log("%s%s, \n", prefix
, a
);
1856 pandecode_log("},\n");
1859 pandecode_log("};\n");
1865 if (p
->sampler_descriptor
) {
1866 struct pandecode_mapped_memory
*smem
= pandecode_find_mapped_gpu_mem_containing(p
->sampler_descriptor
);
1869 struct mali_sampler_descriptor
*s
;
1871 mali_ptr d
= p
->sampler_descriptor
;
1873 for (int i
= 0; i
< sampler_count
; ++i
) {
1874 s
= pandecode_fetch_gpu_mem(smem
, d
+ sizeof(*s
) * i
, sizeof(*s
));
1876 pandecode_log("struct mali_sampler_descriptor sampler_descriptor_%"PRIx64
"_%d_%d = {\n", d
+ sizeof(*s
) * i
, job_no
, i
);
1879 pandecode_log(".filter_mode = ");
1880 pandecode_log_decoded_flags(sampler_flag_info
, s
->filter_mode
);
1881 pandecode_log_cont(",\n");
1883 pandecode_prop("min_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->min_lod
));
1884 pandecode_prop("max_lod = FIXED_16(%f)", DECODE_FIXED_16(s
->max_lod
));
1886 pandecode_prop("wrap_s = %s", pandecode_wrap_mode(s
->wrap_s
));
1887 pandecode_prop("wrap_t = %s", pandecode_wrap_mode(s
->wrap_t
));
1888 pandecode_prop("wrap_r = %s", pandecode_wrap_mode(s
->wrap_r
));
1890 pandecode_prop("compare_func = %s", pandecode_alt_func(s
->compare_func
));
1892 if (s
->zero
|| s
->zero2
) {
1893 pandecode_msg("Zero tripped\n");
1894 pandecode_prop("zero = 0x%X, 0x%X\n", s
->zero
, s
->zero2
);
1897 pandecode_prop("seamless_cube_map = %d", s
->seamless_cube_map
);
1899 pandecode_prop("border_color = { %f, %f, %f, %f }",
1903 s
->border_color
[3]);
1906 pandecode_log("};\n");
1913 pandecode_vertex_tiler_postfix(const struct mali_vertex_tiler_postfix
*p
, int job_no
, bool is_bifrost
)
1915 pandecode_log_cont("{\n");
1918 MEMORY_PROP(p
, position_varying
);
1919 DYN_MEMORY_PROP(p
, job_no
, uniform_buffers
);
1920 DYN_MEMORY_PROP(p
, job_no
, texture_trampoline
);
1921 DYN_MEMORY_PROP(p
, job_no
, sampler_descriptor
);
1922 DYN_MEMORY_PROP(p
, job_no
, uniforms
);
1923 DYN_MEMORY_PROP(p
, job_no
, attributes
);
1924 DYN_MEMORY_PROP(p
, job_no
, attribute_meta
);
1925 DYN_MEMORY_PROP(p
, job_no
, varyings
);
1926 DYN_MEMORY_PROP(p
, job_no
, varying_meta
);
1927 DYN_MEMORY_PROP(p
, job_no
, viewport
);
1928 DYN_MEMORY_PROP(p
, job_no
, occlusion_counter
);
1931 pandecode_prop("framebuffer = scratchpad_%d_p", job_no
);
1933 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no
, p
->framebuffer
& MALI_MFBD
? "MALI_MFBD" : "0");
1935 pandecode_prop("_shader_upper = (shader_meta_%d_p) >> 4", job_no
);
1936 pandecode_prop("flags = %d", p
->flags
);
1939 pandecode_log("},\n");
1943 pandecode_vertex_only_bfr(struct bifrost_vertex_only
*v
)
1945 pandecode_log_cont("{\n");
1948 pandecode_prop("unk2 = 0x%x", v
->unk2
);
1950 if (v
->zero0
|| v
->zero1
) {
1951 pandecode_msg("vertex only zero tripped");
1952 pandecode_prop("zero0 = 0x%" PRIx32
, v
->zero0
);
1953 pandecode_prop("zero1 = 0x%" PRIx64
, v
->zero1
);
1957 pandecode_log("}\n");
1961 pandecode_tiler_heap_meta(mali_ptr gpu_va
, int job_no
)
1964 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
1965 const struct bifrost_tiler_heap_meta
*PANDECODE_PTR_VAR(h
, mem
, gpu_va
);
1967 pandecode_log("struct mali_tiler_heap_meta tiler_heap_meta_%d = {\n", job_no
);
1971 pandecode_msg("tiler heap zero tripped\n");
1972 pandecode_prop("zero = 0x%x", h
->zero
);
1975 for (int i
= 0; i
< 12; i
++) {
1976 if (h
->zeros
[i
] != 0) {
1977 pandecode_msg("tiler heap zero %d tripped, value %x\n",
1982 pandecode_prop("heap_size = 0x%x", h
->heap_size
);
1983 MEMORY_PROP(h
, tiler_heap_start
);
1984 MEMORY_PROP(h
, tiler_heap_free
);
1986 /* this might point to the beginning of another buffer, when it's
1987 * really the end of the tiler heap buffer, so we have to be careful
1988 * here. but for zero length, we need the same pointer.
1991 if (h
->tiler_heap_end
== h
->tiler_heap_start
) {
1992 MEMORY_PROP(h
, tiler_heap_start
);
1994 char *a
= pointer_as_memory_reference(h
->tiler_heap_end
- 1);
1995 pandecode_prop("tiler_heap_end = %s + 1", a
);
2000 pandecode_log("};\n");
2004 pandecode_tiler_meta(mali_ptr gpu_va
, int job_no
)
2006 struct pandecode_mapped_memory
*mem
= pandecode_find_mapped_gpu_mem_containing(gpu_va
);
2007 const struct bifrost_tiler_meta
*PANDECODE_PTR_VAR(t
, mem
, gpu_va
);
2009 pandecode_tiler_heap_meta(t
->tiler_heap_meta
, job_no
);
2011 pandecode_log("struct bifrost_tiler_meta tiler_meta_%d = {\n", job_no
);
2014 if (t
->zero0
|| t
->zero1
) {
2015 pandecode_msg("tiler meta zero tripped");
2016 pandecode_prop("zero0 = 0x%" PRIx64
, t
->zero0
);
2017 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
2020 pandecode_prop("hierarchy_mask = 0x%" PRIx16
, t
->hierarchy_mask
);
2021 pandecode_prop("flags = 0x%" PRIx16
, t
->flags
);
2023 pandecode_prop("width = MALI_POSITIVE(%d)", t
->width
+ 1);
2024 pandecode_prop("height = MALI_POSITIVE(%d)", t
->height
+ 1);
2025 DYN_MEMORY_PROP(t
, job_no
, tiler_heap_meta
);
2027 for (int i
= 0; i
< 12; i
++) {
2028 if (t
->zeros
[i
] != 0) {
2029 pandecode_msg("tiler heap zero %d tripped, value %" PRIx64
"\n",
2035 pandecode_log("};\n");
2039 pandecode_gl_enables(uint32_t gl_enables
, int job_type
)
2041 pandecode_log(".gl_enables = ");
2043 pandecode_log_decoded_flags(gl_enable_flag_info
, gl_enables
);
2045 pandecode_log_cont(",\n");
2049 pandecode_primitive_size(union midgard_primitive_size u
, bool constant
)
2051 if (u
.pointer
== 0x0)
2054 pandecode_log(".primitive_size = {\n");
2058 pandecode_prop("constant = %f", u
.constant
);
2060 MEMORY_PROP((&u
), pointer
);
2064 pandecode_log("},\n");
2068 pandecode_tiler_only_bfr(const struct bifrost_tiler_only
*t
, int job_no
)
2070 pandecode_log_cont("{\n");
2073 /* TODO: gl_PointSize on Bifrost */
2074 pandecode_primitive_size(t
->primitive_size
, true);
2076 DYN_MEMORY_PROP(t
, job_no
, tiler_meta
);
2077 pandecode_gl_enables(t
->gl_enables
, JOB_TYPE_TILER
);
2079 if (t
->zero1
|| t
->zero2
|| t
->zero3
|| t
->zero4
|| t
->zero5
2080 || t
->zero6
|| t
->zero7
|| t
->zero8
) {
2081 pandecode_msg("tiler only zero tripped");
2082 pandecode_prop("zero1 = 0x%" PRIx64
, t
->zero1
);
2083 pandecode_prop("zero2 = 0x%" PRIx64
, t
->zero2
);
2084 pandecode_prop("zero3 = 0x%" PRIx64
, t
->zero3
);
2085 pandecode_prop("zero4 = 0x%" PRIx64
, t
->zero4
);
2086 pandecode_prop("zero5 = 0x%" PRIx64
, t
->zero5
);
2087 pandecode_prop("zero6 = 0x%" PRIx64
, t
->zero6
);
2088 pandecode_prop("zero7 = 0x%" PRIx32
, t
->zero7
);
2089 pandecode_prop("zero8 = 0x%" PRIx64
, t
->zero8
);
2093 pandecode_log("},\n");
2097 pandecode_vertex_job_bfr(const struct mali_job_descriptor_header
*h
,
2098 const struct pandecode_mapped_memory
*mem
,
2099 mali_ptr payload
, int job_no
)
2101 struct bifrost_payload_vertex
*PANDECODE_PTR_VAR(v
, mem
, payload
);
2103 pandecode_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", true);
2105 pandecode_log("struct bifrost_payload_vertex payload_%d = {\n", job_no
);
2108 pandecode_log(".prefix = ");
2109 pandecode_vertex_tiler_prefix(&v
->prefix
, job_no
);
2111 pandecode_log(".vertex = ");
2112 pandecode_vertex_only_bfr(&v
->vertex
);
2114 pandecode_log(".postfix = ");
2115 pandecode_vertex_tiler_postfix(&v
->postfix
, job_no
, true);
2118 pandecode_log("};\n");
2124 pandecode_tiler_job_bfr(const struct mali_job_descriptor_header
*h
,
2125 const struct pandecode_mapped_memory
*mem
,
2126 mali_ptr payload
, int job_no
)
2128 struct bifrost_payload_tiler
*PANDECODE_PTR_VAR(t
, mem
, payload
);
2130 pandecode_vertex_tiler_postfix_pre(&t
->postfix
, job_no
, h
->job_type
, "", true);
2132 pandecode_indices(t
->prefix
.indices
, t
->prefix
.index_count
, job_no
);
2133 pandecode_tiler_meta(t
->tiler
.tiler_meta
, job_no
);
2135 pandecode_log("struct bifrost_payload_tiler payload_%d = {\n", job_no
);
2138 pandecode_log(".prefix = ");
2139 pandecode_vertex_tiler_prefix(&t
->prefix
, job_no
);
2141 pandecode_log(".tiler = ");
2142 pandecode_tiler_only_bfr(&t
->tiler
, job_no
);
2144 pandecode_log(".postfix = ");
2145 pandecode_vertex_tiler_postfix(&t
->postfix
, job_no
, true);
2148 pandecode_log("};\n");
2154 pandecode_vertex_or_tiler_job_mdg(const struct mali_job_descriptor_header
*h
,
2155 const struct pandecode_mapped_memory
*mem
,
2156 mali_ptr payload
, int job_no
)
2158 struct midgard_payload_vertex_tiler
*PANDECODE_PTR_VAR(v
, mem
, payload
);
2160 pandecode_vertex_tiler_postfix_pre(&v
->postfix
, job_no
, h
->job_type
, "", false);
2162 pandecode_indices(v
->prefix
.indices
, v
->prefix
.index_count
, job_no
);
2164 pandecode_log("struct midgard_payload_vertex_tiler payload_%d = {\n", job_no
);
2167 bool has_primitive_pointer
= v
->prefix
.unknown_draw
& MALI_DRAW_VARYING_SIZE
;
2168 pandecode_primitive_size(v
->primitive_size
, !has_primitive_pointer
);
2170 pandecode_log(".prefix = ");
2171 pandecode_vertex_tiler_prefix(&v
->prefix
, job_no
);
2173 pandecode_gl_enables(v
->gl_enables
, h
->job_type
);
2175 if (v
->instance_shift
|| v
->instance_odd
) {
2176 pandecode_prop("instance_shift = 0x%d /* %d */",
2177 v
->instance_shift
, 1 << v
->instance_shift
);
2178 pandecode_prop("instance_odd = 0x%X /* %d */",
2179 v
->instance_odd
, (2 * v
->instance_odd
) + 1);
2181 pandecode_padded_vertices(v
->instance_shift
, v
->instance_odd
);
2185 pandecode_prop("draw_start = %d", v
->draw_start
);
2188 pandecode_msg("Zero tripped\n");
2189 pandecode_prop("zero5 = 0x%" PRIx64
, v
->zero5
);
2192 pandecode_log(".postfix = ");
2193 pandecode_vertex_tiler_postfix(&v
->postfix
, job_no
, false);
2196 pandecode_log("};\n");
2202 pandecode_fragment_job(const struct pandecode_mapped_memory
*mem
,
2203 mali_ptr payload
, int job_no
,
2206 const struct mali_payload_fragment
*PANDECODE_PTR_VAR(s
, mem
, payload
);
2208 bool fbd_dumped
= false;
2210 if (!is_bifrost
&& (s
->framebuffer
& FBD_TYPE
) == MALI_SFBD
) {
2211 /* Only SFBDs are understood, not MFBDs. We're speculating,
2212 * based on the versioning, kernel code, etc, that the
2213 * difference is between Single FrameBuffer Descriptor and
2214 * Multiple FrmaeBuffer Descriptor; the change apparently lines
2215 * up with multi-framebuffer support being added (T7xx onwards,
2216 * including Gxx). In any event, there's some field shuffling
2217 * that we haven't looked into yet. */
2219 pandecode_sfbd(s
->framebuffer
& FBD_MASK
, job_no
);
2221 } else if ((s
->framebuffer
& FBD_TYPE
) == MALI_MFBD
) {
2222 /* We don't know if Bifrost supports SFBD's at all, since the
2223 * driver never uses them. And the format is different from
2224 * Midgard anyways, due to the tiler heap and scratchpad being
2225 * moved out into separate structures, so it's not clear what a
2226 * Bifrost SFBD would even look like without getting an actual
2227 * trace, which appears impossible.
2230 pandecode_mfbd_bfr(s
->framebuffer
& FBD_MASK
, job_no
, true);
2234 uintptr_t p
= (uintptr_t) s
->framebuffer
& FBD_MASK
;
2235 pandecode_log("struct mali_payload_fragment payload_%"PRIx64
"_%d = {\n", payload
, job_no
);
2238 /* See the comments by the macro definitions for mathematical context
2239 * on why this is so weird */
2241 if (MALI_TILE_COORD_FLAGS(s
->max_tile_coord
) || MALI_TILE_COORD_FLAGS(s
->min_tile_coord
))
2242 pandecode_msg("Tile coordinate flag missed, replay wrong\n");
2244 pandecode_prop("min_tile_coord = MALI_COORDINATE_TO_TILE_MIN(%d, %d)",
2245 MALI_TILE_COORD_X(s
->min_tile_coord
) << MALI_TILE_SHIFT
,
2246 MALI_TILE_COORD_Y(s
->min_tile_coord
) << MALI_TILE_SHIFT
);
2248 pandecode_prop("max_tile_coord = MALI_COORDINATE_TO_TILE_MAX(%d, %d)",
2249 (MALI_TILE_COORD_X(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
,
2250 (MALI_TILE_COORD_Y(s
->max_tile_coord
) + 1) << MALI_TILE_SHIFT
);
2252 /* If the FBD was just decoded, we can refer to it by pointer. If not,
2253 * we have to fallback on offsets. */
2255 const char *fbd_type
= s
->framebuffer
& MALI_MFBD
? "MALI_MFBD" : "MALI_SFBD";
2258 pandecode_prop("framebuffer = framebuffer_%d_p | %s", job_no
, fbd_type
);
2260 pandecode_prop("framebuffer = %s | %s", pointer_as_memory_reference(p
), fbd_type
);
2263 pandecode_log("};\n");
2268 static int job_descriptor_number
= 0;
2271 pandecode_jc(mali_ptr jc_gpu_va
, bool bifrost
)
2273 struct mali_job_descriptor_header
*h
;
2275 int start_number
= 0;
2281 struct pandecode_mapped_memory
*mem
=
2282 pandecode_find_mapped_gpu_mem_containing(jc_gpu_va
);
2286 h
= PANDECODE_PTR(mem
, jc_gpu_va
, struct mali_job_descriptor_header
);
2288 /* On Midgard, for 32-bit jobs except for fragment jobs, the
2289 * high 32-bits of the 64-bit pointer are reused to store
2292 int offset
= h
->job_descriptor_size
== MALI_JOB_32
&&
2293 h
->job_type
!= JOB_TYPE_FRAGMENT
? 4 : 0;
2294 mali_ptr payload_ptr
= jc_gpu_va
+ sizeof(*h
) - offset
;
2296 payload
= pandecode_fetch_gpu_mem(mem
, payload_ptr
,
2299 int job_no
= job_descriptor_number
++;
2302 start_number
= job_no
;
2304 pandecode_log("struct mali_job_descriptor_header job_%"PRIx64
"_%d = {\n", jc_gpu_va
, job_no
);
2307 pandecode_prop("job_type = %s", pandecode_job_type(h
->job_type
));
2309 /* Save for next job fixing */
2310 last_size
= h
->job_descriptor_size
;
2312 if (h
->job_descriptor_size
)
2313 pandecode_prop("job_descriptor_size = %d", h
->job_descriptor_size
);
2315 if (h
->exception_status
!= 0x1)
2316 pandecode_prop("exception_status = %x (source ID: 0x%x access: 0x%x exception: 0x%x)",
2317 h
->exception_status
,
2318 (h
->exception_status
>> 16) & 0xFFFF,
2319 (h
->exception_status
>> 8) & 0x3,
2320 h
->exception_status
& 0xFF);
2322 if (h
->first_incomplete_task
)
2323 pandecode_prop("first_incomplete_task = %d", h
->first_incomplete_task
);
2325 if (h
->fault_pointer
)
2326 pandecode_prop("fault_pointer = 0x%" PRIx64
, h
->fault_pointer
);
2329 pandecode_prop("job_barrier = %d", h
->job_barrier
);
2331 pandecode_prop("job_index = %d", h
->job_index
);
2333 if (h
->unknown_flags
)
2334 pandecode_prop("unknown_flags = %d", h
->unknown_flags
);
2336 if (h
->job_dependency_index_1
)
2337 pandecode_prop("job_dependency_index_1 = %d", h
->job_dependency_index_1
);
2339 if (h
->job_dependency_index_2
)
2340 pandecode_prop("job_dependency_index_2 = %d", h
->job_dependency_index_2
);
2343 pandecode_log("};\n");
2345 /* Do not touch the field yet -- decode the payload first, and
2346 * don't touch that either. This is essential for the uploads
2347 * to occur in sequence and therefore be dynamically allocated
2348 * correctly. Do note the size, however, for that related
2351 switch (h
->job_type
) {
2352 case JOB_TYPE_SET_VALUE
: {
2353 struct mali_payload_set_value
*s
= payload
;
2354 pandecode_log("struct mali_payload_set_value payload_%"PRIx64
"_%d = {\n", payload_ptr
, job_no
);
2356 MEMORY_PROP(s
, out
);
2357 pandecode_prop("unknown = 0x%" PRIX64
, s
->unknown
);
2359 pandecode_log("};\n");
2364 case JOB_TYPE_TILER
:
2365 case JOB_TYPE_VERTEX
:
2366 case JOB_TYPE_COMPUTE
:
2368 if (h
->job_type
== JOB_TYPE_TILER
)
2369 pandecode_tiler_job_bfr(h
, mem
, payload_ptr
, job_no
);
2371 pandecode_vertex_job_bfr(h
, mem
, payload_ptr
, job_no
);
2373 pandecode_vertex_or_tiler_job_mdg(h
, mem
, payload_ptr
, job_no
);
2377 case JOB_TYPE_FRAGMENT
:
2378 pandecode_fragment_job(mem
, payload_ptr
, job_no
, bifrost
);
2385 /* Handle linkage */
2388 pandecode_log("((struct mali_job_descriptor_header *) (uintptr_t) job_%d_p)->", job_no
- 1);
2391 pandecode_log_cont("next_job_64 = job_%d_p;\n\n", job_no
);
2393 pandecode_log_cont("next_job_32 = (u32) (uintptr_t) job_%d_p;\n\n", job_no
);
2398 } while ((jc_gpu_va
= h
->job_descriptor_size
? h
->next_job_64
: h
->next_job_32
));
2400 return start_number
;