2 * Copyright © 2016 Intel Corporation
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
35 #include <sys/types.h>
40 #include "util/macros.h"
42 #include "common/gen_decoder.h"
43 #include "intel_aub.h"
44 #include "gen_disasm.h"
46 /* Below is the only command missing from intel_aub.h in libdrm
47 * So, reuse intel_aub.h from libdrm and #define the
48 * AUB_MI_BATCH_BUFFER_END as below
50 #define AUB_MI_BATCH_BUFFER_END (0x0500 << 16)
53 #define BLUE_HEADER CSI "0;44m"
54 #define GREEN_HEADER CSI "1;42m"
55 #define NORMAL CSI "0m"
59 static bool option_full_decode
= true;
60 static bool option_print_offsets
= true;
61 static enum { COLOR_AUTO
, COLOR_ALWAYS
, COLOR_NEVER
} option_color
;
66 char *input_file
= NULL
, *xml_path
= NULL
;
67 struct gen_spec
*spec
;
68 struct gen_disasm
*disasm
;
70 uint64_t gtt_size
, gtt_end
;
72 uint64_t general_state_base
;
73 uint64_t surface_state_base
;
74 uint64_t dynamic_state_base
;
75 uint64_t instruction_base
;
76 uint64_t instruction_bound
;
80 static inline uint32_t
81 field(uint32_t value
, int start
, int end
)
85 mask
= ~0U >> (31 - end
+ start
);
87 return (value
>> start
) & mask
;
90 struct brw_instruction
;
93 valid_offset(uint32_t offset
)
95 return offset
< gtt_end
;
99 decode_group(struct gen_group
*strct
, const uint32_t *p
, int starting_dword
)
101 uint64_t offset
= option_print_offsets
? (void *) p
- gtt
: 0;
102 gen_print_group(outfile
, strct
, offset
, p
, starting_dword
,
103 option_color
== COLOR_ALWAYS
);
107 dump_binding_table(struct gen_spec
*spec
, uint32_t offset
)
109 uint32_t *pointers
, i
;
111 struct gen_group
*surface_state
;
113 surface_state
= gen_spec_find_struct(spec
, "RENDER_SURFACE_STATE");
114 if (surface_state
== NULL
) {
115 fprintf(outfile
, "did not find RENDER_SURFACE_STATE info\n");
119 start
= surface_state_base
+ offset
;
120 pointers
= gtt
+ start
;
121 for (i
= 0; i
< 16; i
++) {
122 if (pointers
[i
] == 0)
124 start
= pointers
[i
] + surface_state_base
;
125 if (!valid_offset(start
)) {
126 fprintf(outfile
, "pointer %u: %08x <not valid>\n",
130 fprintf(outfile
, "pointer %u: %08x\n", i
, pointers
[i
]);
133 decode_group(surface_state
, gtt
+ start
, 0);
138 handle_3dstate_index_buffer(struct gen_spec
*spec
, uint32_t *p
)
141 uint32_t length
, i
, type
, size
;
144 type
= (p
[1] >> 8) & 3;
146 length
= p
[4] / size
;
150 fprintf(outfile
, "\t");
152 for (i
= 0; i
< length
; i
++) {
155 fprintf(outfile
, "%3d ", ((uint8_t *)start
)[i
]);
158 fprintf(outfile
, "%3d ", ((uint16_t *)start
)[i
]);
161 fprintf(outfile
, "%3d ", ((uint32_t *)start
)[i
]);
165 if (length
< p
[4] / size
)
166 fprintf(outfile
, "...\n");
168 fprintf(outfile
, "\n");
171 static inline uint64_t
172 get_address(struct gen_spec
*spec
, uint32_t *p
)
174 /* Addresses are always guaranteed to be page-aligned and sometimes
175 * hardware packets have extra stuff stuffed in the bottom 12 bits.
177 uint64_t addr
= p
[0] & ~0xfffu
;
179 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0)) {
180 /* On Broadwell and above, we have 48-bit addresses which consume two
181 * dwords. Some packets require that these get stored in a "canonical
182 * form" which means that bit 47 is sign-extended through the upper
183 * bits. In order to correctly handle those aub dumps, we need to mask
184 * off the top 16 bits.
186 addr
|= ((uint64_t)p
[1] & 0xffff) << 32;
192 static inline uint64_t
193 get_offset(uint32_t *p
, uint32_t start
, uint32_t end
)
195 assert(start
<= end
);
198 uint64_t mask
= (~0ull >> (64 - (end
- start
+ 1))) << start
;
200 uint64_t offset
= p
[0];
202 offset
|= (uint64_t) p
[1] << 32;
204 return offset
& mask
;
208 handle_state_base_address(struct gen_spec
*spec
, uint32_t *p
)
210 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0)) {
212 general_state_base
= get_address(spec
, &p
[1]);
214 surface_state_base
= get_address(spec
, &p
[4]);
216 dynamic_state_base
= get_address(spec
, &p
[6]);
218 instruction_base
= get_address(spec
, &p
[10]);
220 instruction_bound
= p
[15] & 0xfff;
223 surface_state_base
= get_address(spec
, &p
[2]);
225 dynamic_state_base
= get_address(spec
, &p
[3]);
227 instruction_base
= get_address(spec
, &p
[5]);
229 instruction_bound
= get_address(spec
, &p
[9]);
234 dump_samplers(struct gen_spec
*spec
, uint32_t offset
)
238 struct gen_group
*sampler_state
;
240 sampler_state
= gen_spec_find_struct(spec
, "SAMPLER_STATE");
242 start
= dynamic_state_base
+ offset
;
243 for (i
= 0; i
< 4; i
++) {
244 fprintf(outfile
, "sampler state %d\n", i
);
245 decode_group(sampler_state
, gtt
+ start
+ i
* 16, 0);
250 handle_media_interface_descriptor_load(struct gen_spec
*spec
, uint32_t *p
)
252 int i
, length
= p
[2] / 32;
253 struct gen_group
*descriptor_structure
;
254 uint32_t *descriptors
;
256 struct brw_instruction
*insns
;
258 descriptor_structure
=
259 gen_spec_find_struct(spec
, "INTERFACE_DESCRIPTOR_DATA");
260 if (descriptor_structure
== NULL
) {
261 fprintf(outfile
, "did not find INTERFACE_DESCRIPTOR_DATA info\n");
265 start
= dynamic_state_base
+ p
[3];
266 descriptors
= gtt
+ start
;
267 for (i
= 0; i
< length
; i
++, descriptors
+= 8) {
268 fprintf(outfile
, "descriptor %u: %08x\n", i
, *descriptors
);
269 decode_group(descriptor_structure
, descriptors
, 0);
271 start
= instruction_base
+ descriptors
[0];
272 if (!valid_offset(start
)) {
273 fprintf(outfile
, "kernel: %08"PRIx64
" <not valid>\n", start
);
276 fprintf(outfile
, "kernel: %08"PRIx64
"\n", start
);
279 insns
= (struct brw_instruction
*) (gtt
+ start
);
280 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
282 dump_samplers(spec
, descriptors
[3] & ~0x1f);
283 dump_binding_table(spec
, descriptors
[4] & ~0x1f);
287 /* Heuristic to determine whether a uint32_t is probably actually a float
288 * (http://stackoverflow.com/a/2953466)
292 probably_float(uint32_t bits
)
294 int exp
= ((bits
& 0x7f800000U
) >> 23) - 127;
295 uint32_t mant
= bits
& 0x007fffff;
298 if (exp
== -127 && mant
== 0)
301 /* +- 1 billionth to 1 billion */
302 if (-30 <= exp
&& exp
<= 30)
305 /* some value with only a few binary digits */
306 if ((mant
& 0x0000ffff) == 0)
313 handle_3dstate_vertex_buffers(struct gen_spec
*spec
, uint32_t *p
)
315 uint32_t *end
, *s
, *dw
, *dwend
;
317 int n
, i
, count
, stride
;
319 end
= (p
[0] & 0xff) + p
+ 2;
320 for (s
= &p
[1], n
= 0; s
< end
; s
+= 4, n
++) {
321 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
322 offset
= *(uint64_t *) &s
[1];
323 dwend
= gtt
+ offset
+ s
[3];
326 dwend
= gtt
+ s
[2] + 1;
329 stride
= field(s
[0], 0, 11);
331 fprintf(outfile
, "vertex buffer %d, size %d\n", n
, s
[3]);
332 for (dw
= gtt
+ offset
, i
= 0; dw
< dwend
&& i
< 256; dw
++) {
333 if (count
== 0 && count
% (8 * 4) == 0)
334 fprintf(outfile
, " ");
336 if (probably_float(*dw
))
337 fprintf(outfile
, " %8.2f", *(float *) dw
);
339 fprintf(outfile
, " 0x%08x", *dw
);
344 if (count
== stride
) {
345 fprintf(outfile
, "\n");
347 } else if (count
% (8 * 4) == 0) {
348 fprintf(outfile
, "\n");
350 fprintf(outfile
, " ");
353 if (count
> 0 && count
% (8 * 4) != 0)
354 fprintf(outfile
, "\n");
359 handle_3dstate_vs(struct gen_spec
*spec
, uint32_t *p
)
362 struct brw_instruction
*insns
;
365 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
366 start
= get_offset(&p
[1], 6, 63);
367 vs_enable
= p
[7] & 1;
369 start
= get_offset(&p
[1], 6, 31);
370 vs_enable
= p
[5] & 1;
374 fprintf(outfile
, "instruction_base %08"PRIx64
", start %08"PRIx64
"\n",
375 instruction_base
, start
);
377 insns
= (struct brw_instruction
*) (gtt
+ instruction_base
+ start
);
378 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
383 handle_3dstate_hs(struct gen_spec
*spec
, uint32_t *p
)
386 struct brw_instruction
*insns
;
389 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
390 start
= get_offset(&p
[3], 6, 63);
392 start
= get_offset(&p
[3], 6, 31);
395 hs_enable
= p
[2] & 0x80000000;
398 fprintf(outfile
, "instruction_base %08"PRIx64
", start %08"PRIx64
"\n",
399 instruction_base
, start
);
401 insns
= (struct brw_instruction
*) (gtt
+ instruction_base
+ start
);
402 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
407 handle_3dstate_constant(struct gen_spec
*spec
, uint32_t *p
)
413 for (i
= 0; i
< 4; i
++) {
414 length
= (p
[1 + i
/ 2] >> (i
& 1) * 16) & 0xffff;
415 f
= (float *) (gtt
+ p
[3 + i
* 2] + dynamic_state_base
);
417 for (j
= 0; j
< length
* 8; j
++) {
418 if (probably_float(dw
[j
]))
419 fprintf(outfile
, " %04.3f", f
[j
]);
421 fprintf(outfile
, " 0x%08x", dw
[j
]);
424 fprintf(outfile
, "\n");
430 handle_3dstate_ps(struct gen_spec
*spec
, uint32_t *p
)
432 uint32_t mask
= ~((1 << 6) - 1);
434 struct brw_instruction
*insns
;
435 static const char unused
[] = "unused";
436 static const char *pixel_type
[3] = {"8 pixel", "16 pixel", "32 pixel"};
437 const char *k0
, *k1
, *k2
;
438 uint32_t k_mask
, k1_offset
, k2_offset
;
440 if (gen_spec_get_gen(spec
) >= gen_make_gen(8, 0)) {
451 #define DISPATCH_16 2
452 #define DISPATCH_32 4
465 case DISPATCH_8
| DISPATCH_16
:
475 case DISPATCH_16
| DISPATCH_32
:
480 case DISPATCH_8
| DISPATCH_16
| DISPATCH_32
:
492 start
= instruction_base
+ (p
[1] & mask
);
493 fprintf(outfile
, " Kernel[0] %s\n", k0
);
495 insns
= (struct brw_instruction
*) (gtt
+ start
);
496 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
499 start
= instruction_base
+ (p
[k1_offset
] & mask
);
500 fprintf(outfile
, " Kernel[1] %s\n", k1
);
502 insns
= (struct brw_instruction
*) (gtt
+ start
);
503 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
506 start
= instruction_base
+ (p
[k2_offset
] & mask
);
507 fprintf(outfile
, " Kernel[2] %s\n", k2
);
509 insns
= (struct brw_instruction
*) (gtt
+ start
);
510 gen_disasm_disassemble(disasm
, insns
, 0, stdout
);
515 handle_3dstate_binding_table_pointers(struct gen_spec
*spec
, uint32_t *p
)
517 dump_binding_table(spec
, p
[1]);
521 handle_3dstate_sampler_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
523 dump_samplers(spec
, p
[1]);
527 handle_3dstate_viewport_state_pointers_cc(struct gen_spec
*spec
, uint32_t *p
)
530 struct gen_group
*cc_viewport
;
532 cc_viewport
= gen_spec_find_struct(spec
, "CC_VIEWPORT");
534 start
= dynamic_state_base
+ (p
[1] & ~0x1fu
);
535 for (uint32_t i
= 0; i
< 4; i
++) {
536 fprintf(outfile
, "viewport %d\n", i
);
537 decode_group(cc_viewport
, gtt
+ start
+ i
* 8, 0);
542 handle_3dstate_viewport_state_pointers_sf_clip(struct gen_spec
*spec
,
546 struct gen_group
*sf_clip_viewport
;
548 sf_clip_viewport
= gen_spec_find_struct(spec
, "SF_CLIP_VIEWPORT");
550 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
551 for (uint32_t i
= 0; i
< 4; i
++) {
552 fprintf(outfile
, "viewport %d\n", i
);
553 decode_group(sf_clip_viewport
, gtt
+ start
+ i
* 64, 0);
558 handle_3dstate_blend_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
561 struct gen_group
*blend_state
;
563 blend_state
= gen_spec_find_struct(spec
, "BLEND_STATE");
565 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
566 decode_group(blend_state
, gtt
+ start
, 0);
570 handle_3dstate_cc_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
573 struct gen_group
*cc_state
;
575 cc_state
= gen_spec_find_struct(spec
, "COLOR_CALC_STATE");
577 start
= dynamic_state_base
+ (p
[1] & ~0x3fu
);
578 decode_group(cc_state
, gtt
+ start
, 0);
582 handle_3dstate_scissor_state_pointers(struct gen_spec
*spec
, uint32_t *p
)
585 struct gen_group
*scissor_rect
;
587 scissor_rect
= gen_spec_find_struct(spec
, "SCISSOR_RECT");
589 start
= dynamic_state_base
+ (p
[1] & ~0x1fu
);
590 decode_group(scissor_rect
, gtt
+ start
, 0);
594 handle_load_register_imm(struct gen_spec
*spec
, uint32_t *p
)
596 struct gen_group
*reg
= gen_spec_find_register(spec
, p
[1]);
599 fprintf(outfile
, "register %s (0x%x): 0x%x\n",
600 reg
->name
, reg
->register_offset
, p
[2]);
601 decode_group(reg
, &p
[2], 0);
605 #define ARRAY_LENGTH(a) (sizeof (a) / sizeof (a)[0])
607 #define STATE_BASE_ADDRESS 0x61010000
609 #define MEDIA_INTERFACE_DESCRIPTOR_LOAD 0x70020000
611 #define _3DSTATE_INDEX_BUFFER 0x780a0000
612 #define _3DSTATE_VERTEX_BUFFERS 0x78080000
614 #define _3DSTATE_VS 0x78100000
615 #define _3DSTATE_GS 0x78110000
616 #define _3DSTATE_HS 0x781b0000
617 #define _3DSTATE_DS 0x781d0000
619 #define _3DSTATE_CONSTANT_VS 0x78150000
620 #define _3DSTATE_CONSTANT_GS 0x78160000
621 #define _3DSTATE_CONSTANT_PS 0x78170000
622 #define _3DSTATE_CONSTANT_HS 0x78190000
623 #define _3DSTATE_CONSTANT_DS 0x781A0000
625 #define _3DSTATE_PS 0x78200000
627 #define _3DSTATE_BINDING_TABLE_POINTERS_VS 0x78260000
628 #define _3DSTATE_BINDING_TABLE_POINTERS_HS 0x78270000
629 #define _3DSTATE_BINDING_TABLE_POINTERS_DS 0x78280000
630 #define _3DSTATE_BINDING_TABLE_POINTERS_GS 0x78290000
631 #define _3DSTATE_BINDING_TABLE_POINTERS_PS 0x782a0000
633 #define _3DSTATE_SAMPLER_STATE_POINTERS_VS 0x782b0000
634 #define _3DSTATE_SAMPLER_STATE_POINTERS_GS 0x782e0000
635 #define _3DSTATE_SAMPLER_STATE_POINTERS_PS 0x782f0000
637 #define _3DSTATE_VIEWPORT_STATE_POINTERS_CC 0x78230000
638 #define _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP 0x78210000
639 #define _3DSTATE_BLEND_STATE_POINTERS 0x78240000
640 #define _3DSTATE_CC_STATE_POINTERS 0x780e0000
641 #define _3DSTATE_SCISSOR_STATE_POINTERS 0x780f0000
643 #define _MI_LOAD_REGISTER_IMM 0x11000000
645 struct custom_handler
{
647 void (*handle
)(struct gen_spec
*spec
, uint32_t *p
);
648 } custom_handlers
[] = {
649 { STATE_BASE_ADDRESS
, handle_state_base_address
},
650 { MEDIA_INTERFACE_DESCRIPTOR_LOAD
, handle_media_interface_descriptor_load
},
651 { _3DSTATE_VERTEX_BUFFERS
, handle_3dstate_vertex_buffers
},
652 { _3DSTATE_INDEX_BUFFER
, handle_3dstate_index_buffer
},
653 { _3DSTATE_VS
, handle_3dstate_vs
},
654 { _3DSTATE_GS
, handle_3dstate_vs
},
655 { _3DSTATE_DS
, handle_3dstate_vs
},
656 { _3DSTATE_HS
, handle_3dstate_hs
},
657 { _3DSTATE_CONSTANT_VS
, handle_3dstate_constant
},
658 { _3DSTATE_CONSTANT_GS
, handle_3dstate_constant
},
659 { _3DSTATE_CONSTANT_PS
, handle_3dstate_constant
},
660 { _3DSTATE_CONSTANT_HS
, handle_3dstate_constant
},
661 { _3DSTATE_CONSTANT_DS
, handle_3dstate_constant
},
662 { _3DSTATE_PS
, handle_3dstate_ps
},
664 { _3DSTATE_BINDING_TABLE_POINTERS_VS
, handle_3dstate_binding_table_pointers
},
665 { _3DSTATE_BINDING_TABLE_POINTERS_HS
, handle_3dstate_binding_table_pointers
},
666 { _3DSTATE_BINDING_TABLE_POINTERS_DS
, handle_3dstate_binding_table_pointers
},
667 { _3DSTATE_BINDING_TABLE_POINTERS_GS
, handle_3dstate_binding_table_pointers
},
668 { _3DSTATE_BINDING_TABLE_POINTERS_PS
, handle_3dstate_binding_table_pointers
},
670 { _3DSTATE_SAMPLER_STATE_POINTERS_VS
, handle_3dstate_sampler_state_pointers
},
671 { _3DSTATE_SAMPLER_STATE_POINTERS_GS
, handle_3dstate_sampler_state_pointers
},
672 { _3DSTATE_SAMPLER_STATE_POINTERS_PS
, handle_3dstate_sampler_state_pointers
},
674 { _3DSTATE_VIEWPORT_STATE_POINTERS_CC
, handle_3dstate_viewport_state_pointers_cc
},
675 { _3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP
, handle_3dstate_viewport_state_pointers_sf_clip
},
676 { _3DSTATE_BLEND_STATE_POINTERS
, handle_3dstate_blend_state_pointers
},
677 { _3DSTATE_CC_STATE_POINTERS
, handle_3dstate_cc_state_pointers
},
678 { _3DSTATE_SCISSOR_STATE_POINTERS
, handle_3dstate_scissor_state_pointers
},
679 { _MI_LOAD_REGISTER_IMM
, handle_load_register_imm
}
683 parse_commands(struct gen_spec
*spec
, uint32_t *cmds
, int size
, int engine
)
685 uint32_t *p
, *end
= cmds
+ size
/ 4;
686 unsigned int length
, i
;
687 struct gen_group
*inst
;
689 for (p
= cmds
; p
< end
; p
+= length
) {
690 inst
= gen_spec_find_instruction(spec
, p
);
692 fprintf(outfile
, "unknown instruction %08x\n", p
[0]);
693 length
= (p
[0] & 0xff) + 2;
696 length
= gen_group_get_length(inst
, p
);
698 const char *color
, *reset_color
= NORMAL
;
701 if (option_full_decode
) {
702 if ((p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_START
||
703 (p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END
)
704 color
= GREEN_HEADER
;
710 if (option_color
== COLOR_NEVER
) {
715 if (option_print_offsets
)
716 offset
= (void *) p
- gtt
;
720 fprintf(outfile
, "%s0x%08"PRIx64
": 0x%08x: %-80s%s\n",
722 gen_group_get_name(inst
), reset_color
);
724 if (option_full_decode
) {
725 decode_group(inst
, p
, 1);
727 for (i
= 0; i
< ARRAY_LENGTH(custom_handlers
); i
++) {
728 if (gen_group_get_opcode(inst
) == custom_handlers
[i
].opcode
)
729 custom_handlers
[i
].handle(spec
, p
);
733 if ((p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_START
) {
734 uint64_t start
= get_address(spec
, &p
[1]);
736 if (p
[0] & (1 << 22)) {
737 /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" set acts
738 * like a subroutine call. Commands that come afterwards get
739 * processed once the 2nd level batch buffer returns with
740 * MI_BATCH_BUFFER_END.
742 parse_commands(spec
, gtt
+ start
, gtt_end
- start
, engine
);
744 /* MI_BATCH_BUFFER_START with "2nd Level Batch Buffer" unset acts
745 * like a goto. Nothing after it will ever get processed. In
746 * order to prevent the recursion from growing, we just reset the
750 /* We don't know where secondaries end so use the GTT end */
755 } else if ((p
[0] & 0xffff0000) == AUB_MI_BATCH_BUFFER_END
) {
761 #define GEN_ENGINE_RENDER 1
762 #define GEN_ENGINE_BLITTER 2
765 handle_trace_block(uint32_t *p
)
767 int operation
= p
[1] & AUB_TRACE_OPERATION_MASK
;
768 int type
= p
[1] & AUB_TRACE_TYPE_MASK
;
769 int address_space
= p
[1] & AUB_TRACE_ADDRESS_SPACE_MASK
;
770 uint64_t offset
= p
[3];
771 uint32_t size
= p
[4];
772 int header_length
= p
[0] & 0xffff;
773 uint32_t *data
= p
+ header_length
+ 2;
774 int engine
= GEN_ENGINE_RENDER
;
776 if (gen_spec_get_gen(spec
) >= gen_make_gen(8,0))
777 offset
+= (uint64_t) p
[5] << 32;
780 case AUB_TRACE_OP_DATA_WRITE
:
781 if (address_space
!= AUB_TRACE_MEMTYPE_GTT
)
783 if (gtt_size
< offset
+ size
) {
784 fprintf(stderr
, "overflow gtt space: %s\n", strerror(errno
));
787 memcpy((char *) gtt
+ offset
, data
, size
);
788 if (gtt_end
< offset
+ size
)
789 gtt_end
= offset
+ size
;
791 case AUB_TRACE_OP_COMMAND_WRITE
:
793 case AUB_TRACE_TYPE_RING_PRB0
:
794 engine
= GEN_ENGINE_RENDER
;
796 case AUB_TRACE_TYPE_RING_PRB2
:
797 engine
= GEN_ENGINE_BLITTER
;
800 fprintf(outfile
, "command write to unknown ring %d\n", type
);
804 parse_commands(spec
, data
, size
, engine
);
811 handle_trace_header(uint32_t *p
)
813 /* The intel_aubdump tool from IGT is kind enough to put a PCI-ID= tag in
814 * the AUB header comment. If the user hasn't specified a hardware
815 * generation, try to use the one from the AUB file.
817 uint32_t *end
= p
+ (p
[0] & 0xffff) + 2;
819 if (end
> &p
[12] && p
[12] > 0)
820 sscanf((char *)&p
[13], "PCI-ID=%i", &aub_pci_id
);
825 struct gen_device_info devinfo
;
826 if (!gen_get_device_info(pci_id
, &devinfo
)) {
827 fprintf(stderr
, "can't find device information: pci_id=0x%x\n", pci_id
);
831 if (xml_path
== NULL
)
832 spec
= gen_spec_load(&devinfo
);
834 spec
= gen_spec_load_from_path(&devinfo
, xml_path
);
835 disasm
= gen_disasm_create(pci_id
);
837 if (spec
== NULL
|| disasm
== NULL
)
840 fprintf(outfile
, "%sAubinator: Intel AUB file decoder.%-80s%s\n",
841 GREEN_HEADER
, "", NORMAL
);
844 fprintf(outfile
, "File name: %s\n", input_file
);
847 fprintf(outfile
, "PCI ID: 0x%x\n", aub_pci_id
);
850 strncpy(app_name
, (char *)&p
[2], 32);
852 fprintf(outfile
, "Application name: %s\n", app_name
);
854 fprintf(outfile
, "Decoding as: %s\n", gen_get_device_name(pci_id
));
856 /* Throw in a new line before the first batch */
857 fprintf(outfile
, "\n");
863 uint32_t *map
, *end
, *cursor
;
867 static struct aub_file
*
868 aub_file_open(const char *filename
)
870 struct aub_file
*file
;
874 file
= calloc(1, sizeof *file
);
875 fd
= open(filename
, O_RDONLY
);
877 fprintf(stderr
, "open %s failed: %s\n", filename
, strerror(errno
));
881 if (fstat(fd
, &sb
) == -1) {
882 fprintf(stderr
, "stat failed: %s\n", strerror(errno
));
886 file
->map
= mmap(NULL
, sb
.st_size
,
887 PROT_READ
, MAP_SHARED
, fd
, 0);
888 if (file
->map
== MAP_FAILED
) {
889 fprintf(stderr
, "mmap failed: %s\n", strerror(errno
));
893 file
->cursor
= file
->map
;
894 file
->end
= file
->map
+ sb
.st_size
/ 4;
899 static struct aub_file
*
902 struct aub_file
*file
;
904 file
= calloc(1, sizeof *file
);
905 file
->stream
= stdin
;
910 #define TYPE(dw) (((dw) >> 29) & 7)
911 #define OPCODE(dw) (((dw) >> 23) & 0x3f)
912 #define SUBOPCODE(dw) (((dw) >> 16) & 0x7f)
914 #define MAKE_HEADER(type, opcode, subopcode) \
915 (((type) << 29) | ((opcode) << 23) | ((subopcode) << 16))
919 /* Classic AUB opcodes */
920 #define OPCODE_AUB 0x01
921 #define SUBOPCODE_HEADER 0x05
922 #define SUBOPCODE_BLOCK 0x41
923 #define SUBOPCODE_BMP 0x1e
925 /* Newer version AUB opcode */
926 #define OPCODE_NEW_AUB 0x2e
927 #define SUBOPCODE_VERSION 0x00
928 #define SUBOPCODE_REG_WRITE 0x03
929 #define SUBOPCODE_MEM_POLL 0x05
930 #define SUBOPCODE_MEM_WRITE 0x06
932 #define MAKE_GEN(major, minor) ( ((major) << 8) | (minor) )
938 { "bwr", MAKE_GEN(4, 0) },
939 { "cln", MAKE_GEN(4, 0) },
940 { "blc", MAKE_GEN(4, 0) },
941 { "ctg", MAKE_GEN(4, 0) },
942 { "el", MAKE_GEN(4, 0) },
943 { "il", MAKE_GEN(4, 0) },
944 { "sbr", MAKE_GEN(6, 0) },
945 { "ivb", MAKE_GEN(7, 0) },
946 { "lrb2", MAKE_GEN(0, 0) },
947 { "hsw", MAKE_GEN(7, 5) },
948 { "vlv", MAKE_GEN(7, 0) },
949 { "bdw", MAKE_GEN(8, 0) },
950 { "skl", MAKE_GEN(9, 0) },
951 { "chv", MAKE_GEN(8, 0) },
952 { "bxt", MAKE_GEN(9, 0) }
957 AUB_ITEM_DECODE_FAILED
,
958 AUB_ITEM_DECODE_NEED_MORE_DATA
,
962 aub_file_decode_batch(struct aub_file
*file
)
964 uint32_t *p
, h
, device
, data_type
, *new_cursor
;
965 int header_length
, bias
;
967 if (file
->end
- file
->cursor
< 1)
968 return AUB_ITEM_DECODE_NEED_MORE_DATA
;
972 header_length
= h
& 0xffff;
982 fprintf(outfile
, "unknown opcode %d at %td/%td\n",
983 OPCODE(h
), file
->cursor
- file
->map
,
984 file
->end
- file
->map
);
985 return AUB_ITEM_DECODE_FAILED
;
988 new_cursor
= p
+ header_length
+ bias
;
989 if ((h
& 0xffff0000) == MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_BLOCK
)) {
990 if (file
->end
- file
->cursor
< 4)
991 return AUB_ITEM_DECODE_NEED_MORE_DATA
;
992 new_cursor
+= p
[4] / 4;
995 if (new_cursor
> file
->end
)
996 return AUB_ITEM_DECODE_NEED_MORE_DATA
;
998 switch (h
& 0xffff0000) {
999 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_HEADER
):
1000 handle_trace_header(p
);
1002 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_BLOCK
):
1003 handle_trace_block(p
);
1005 case MAKE_HEADER(TYPE_AUB
, OPCODE_AUB
, SUBOPCODE_BMP
):
1007 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_VERSION
):
1008 fprintf(outfile
, "version block: dw1 %08x\n", p
[1]);
1009 device
= (p
[1] >> 8) & 0xff;
1010 fprintf(outfile
, " device %s\n", device_map
[device
].name
);
1012 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_REG_WRITE
):
1013 fprintf(outfile
, "register write block: (dwords %d)\n", h
& 0xffff);
1014 fprintf(outfile
, " reg 0x%x, data 0x%x\n", p
[1], p
[5]);
1016 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_MEM_WRITE
):
1017 fprintf(outfile
, "memory write block (dwords %d):\n", h
& 0xffff);
1018 fprintf(outfile
, " address 0x%"PRIx64
"\n", *(uint64_t *) &p
[1]);
1019 data_type
= (p
[3] >> 20) & 0xff;
1021 fprintf(outfile
, " data type 0x%x\n", data_type
);
1022 fprintf(outfile
, " address space 0x%x\n", (p
[3] >> 28) & 0xf);
1024 case MAKE_HEADER(TYPE_AUB
, OPCODE_NEW_AUB
, SUBOPCODE_MEM_POLL
):
1025 fprintf(outfile
, "memory poll block (dwords %d):\n", h
& 0xffff);
1028 fprintf(outfile
, "unknown block type=0x%x, opcode=0x%x, "
1029 "subopcode=0x%x (%08x)\n", TYPE(h
), OPCODE(h
), SUBOPCODE(h
), h
);
1032 file
->cursor
= new_cursor
;
1034 return AUB_ITEM_DECODE_OK
;
1038 aub_file_more_stuff(struct aub_file
*file
)
1040 return file
->cursor
< file
->end
|| (file
->stream
&& !feof(file
->stream
));
1043 #define AUB_READ_BUFFER_SIZE (4096)
1044 #define MAX(a, b) ((a) < (b) ? (b) : (a))
1047 aub_file_data_grow(struct aub_file
*file
)
1049 size_t old_size
= (file
->mem_end
- file
->map
) * 4;
1050 size_t new_size
= MAX(old_size
* 2, AUB_READ_BUFFER_SIZE
);
1051 uint32_t *new_start
= realloc(file
->map
, new_size
);
1053 file
->cursor
= new_start
+ (file
->cursor
- file
->map
);
1054 file
->end
= new_start
+ (file
->end
- file
->map
);
1055 file
->map
= new_start
;
1056 file
->mem_end
= file
->map
+ (new_size
/ 4);
1060 aub_file_data_load(struct aub_file
*file
)
1064 if (file
->stream
== NULL
)
1067 /* First remove any consumed data */
1068 if (file
->cursor
> file
->map
) {
1069 memmove(file
->map
, file
->cursor
,
1070 (file
->end
- file
->cursor
) * 4);
1071 file
->end
-= file
->cursor
- file
->map
;
1072 file
->cursor
= file
->map
;
1075 /* Then load some new data in */
1076 if ((file
->mem_end
- file
->end
) < (AUB_READ_BUFFER_SIZE
/ 4))
1077 aub_file_data_grow(file
);
1079 r
= fread(file
->end
, 1, (file
->mem_end
- file
->end
) * 4, file
->stream
);
1094 if (pipe(fds
) == -1)
1104 execlp("less", "less", "-FRSi", NULL
);
1113 print_help(const char *progname
, FILE *file
)
1116 "Usage: %s [OPTION]... [FILE]\n"
1117 "Decode aub file contents from either FILE or the standard input.\n\n"
1118 "A valid --gen option must be provided.\n\n"
1119 " --help display this help and exit\n"
1120 " --gen=platform decode for given platform (ivb, byt, hsw, bdw, chv, skl, kbl or bxt)\n"
1121 " --headers decode only command headers\n"
1122 " --color[=WHEN] colorize the output; WHEN can be 'auto' (default\n"
1123 " if omitted), 'always', or 'never'\n"
1124 " --no-pager don't launch pager\n"
1125 " --no-offsets don't print instruction offsets\n"
1126 " --xml=DIR load hardware xml description from directory DIR\n",
1130 int main(int argc
, char *argv
[])
1132 struct aub_file
*file
;
1134 bool help
= false, pager
= true;
1139 { "ivb", 0x0166 }, /* Intel(R) Ivybridge Mobile GT2 */
1140 { "hsw", 0x0416 }, /* Intel(R) Haswell Mobile GT2 */
1141 { "byt", 0x0155 }, /* Intel(R) Bay Trail */
1142 { "bdw", 0x1616 }, /* Intel(R) HD Graphics 5500 (Broadwell GT2) */
1143 { "chv", 0x22B3 }, /* Intel(R) HD Graphics (Cherryview) */
1144 { "skl", 0x1912 }, /* Intel(R) HD Graphics 530 (Skylake GT2) */
1145 { "kbl", 0x591D }, /* Intel(R) Kabylake GT2 */
1146 { "bxt", 0x0A84 } /* Intel(R) HD Graphics (Broxton) */
1148 const struct option aubinator_opts
[] = {
1149 { "help", no_argument
, (int *) &help
, true },
1150 { "no-pager", no_argument
, (int *) &pager
, false },
1151 { "no-offsets", no_argument
, (int *) &option_print_offsets
, false },
1152 { "gen", required_argument
, NULL
, 'g' },
1153 { "headers", no_argument
, (int *) &option_full_decode
, false },
1154 { "color", required_argument
, NULL
, 'c' },
1155 { "xml", required_argument
, NULL
, 'x' },
1156 { NULL
, 0, NULL
, 0 }
1162 while ((c
= getopt_long(argc
, argv
, "", aubinator_opts
, &i
)) != -1) {
1165 for (i
= 0; i
< ARRAY_SIZE(gens
); i
++) {
1166 if (!strcmp(optarg
, gens
[i
].name
)) {
1167 pci_id
= gens
[i
].pci_id
;
1171 if (i
== ARRAY_SIZE(gens
)) {
1172 fprintf(stderr
, "can't parse gen: '%s', expected ivb, byt, hsw, "
1173 "bdw, chv, skl, kbl or bxt\n", optarg
);
1178 if (optarg
== NULL
|| strcmp(optarg
, "always") == 0)
1179 option_color
= COLOR_ALWAYS
;
1180 else if (strcmp(optarg
, "never") == 0)
1181 option_color
= COLOR_NEVER
;
1182 else if (strcmp(optarg
, "auto") == 0)
1183 option_color
= COLOR_AUTO
;
1185 fprintf(stderr
, "invalid value for --color: %s", optarg
);
1190 xml_path
= strdup(optarg
);
1197 if (help
|| argc
== 1) {
1198 print_help(argv
[0], stderr
);
1203 input_file
= argv
[optind
];
1205 /* Do this before we redirect stdout to pager. */
1206 if (option_color
== COLOR_AUTO
)
1207 option_color
= isatty(1) ? COLOR_ALWAYS
: COLOR_NEVER
;
1209 if (isatty(1) && pager
)
1212 if (input_file
== NULL
)
1213 file
= aub_file_stdin();
1215 file
= aub_file_open(input_file
);
1217 /* mmap a terabyte for our gtt space. */
1218 gtt_size
= 1ull << 40;
1219 gtt
= mmap(NULL
, gtt_size
, PROT_READ
| PROT_WRITE
,
1220 MAP_PRIVATE
| MAP_ANONYMOUS
| MAP_NORESERVE
, -1, 0);
1221 if (gtt
== MAP_FAILED
) {
1222 fprintf(stderr
, "failed to alloc gtt space: %s\n", strerror(errno
));
1226 while (aub_file_more_stuff(file
)) {
1227 switch (aub_file_decode_batch(file
)) {
1228 case AUB_ITEM_DECODE_OK
:
1230 case AUB_ITEM_DECODE_NEED_MORE_DATA
:
1231 if (!file
->stream
) {
1232 file
->cursor
= file
->end
;
1235 if (aub_file_more_stuff(file
) && !aub_file_data_load(file
)) {
1236 fprintf(stderr
, "failed to load data from stdin\n");
1241 fprintf(stderr
, "failed to parse aubdump data\n");
1248 /* close the stdout which is opened to write the output */
1254 return EXIT_SUCCESS
;