2 * Copyright 2015 Advanced Micro Devices, Inc.
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
24 * Marek Olšák <maraeo@gmail.com>
28 #include "si_shader.h"
30 #include "sid_tables.h"
33 static void si_dump_shader(struct si_shader_selector
*sel
, const char *name
,
36 if (!sel
|| !sel
->current
)
39 fprintf(f
, "%s shader disassembly:\n", name
);
40 si_dump_shader_key(sel
->type
, &sel
->current
->key
, f
);
41 fprintf(f
, "%s\n\n", sel
->current
->binary
.disasm_string
);
44 /* Parsed IBs are difficult to read without colors. Use "less -R file" to
45 * read them, or use "aha -b -f file" to convert them to html.
47 #define COLOR_RESET "\033[0m"
48 #define COLOR_RED "\033[31m"
49 #define COLOR_GREEN "\033[1;32m"
50 #define COLOR_YELLOW "\033[1;33m"
51 #define COLOR_CYAN "\033[1;36m"
55 static void print_spaces(FILE *f
, unsigned num
)
57 fprintf(f
, "%*s", num
, "");
60 static void print_value(FILE *file
, uint32_t value
, int bits
)
62 /* Guess if it's int or float */
63 if (value
<= (1 << 15))
64 fprintf(file
, "%u\n", value
);
68 if (fabs(f
) < 100000 && f
*10 == floor(f
*10))
69 fprintf(file
, "%.1ff\n", f
);
71 /* Don't print more leading zeros than there are bits. */
72 fprintf(file
, "0x%0*x\n", bits
/ 4, value
);
76 static void print_named_value(FILE *file
, const char *name
, uint32_t value
,
79 print_spaces(file
, INDENT_PKT
);
80 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ", name
);
81 print_value(file
, value
, bits
);
84 static void si_dump_reg(FILE *file
, unsigned offset
, uint32_t value
,
89 for (r
= 0; r
< ARRAY_SIZE(reg_table
); r
++) {
90 const struct si_reg
*reg
= ®_table
[r
];
92 if (reg
->offset
== offset
) {
93 bool first_field
= true;
95 print_spaces(file
, INDENT_PKT
);
96 fprintf(file
, COLOR_YELLOW
"%s" COLOR_RESET
" <- ",
99 if (!reg
->num_fields
) {
100 print_value(file
, value
, 32);
104 for (f
= 0; f
< reg
->num_fields
; f
++) {
105 const struct si_field
*field
= ®
->fields
[f
];
106 uint32_t val
= (value
& field
->mask
) >>
107 (ffs(field
->mask
) - 1);
109 if (!(field
->mask
& field_mask
))
112 /* Indent the field. */
115 INDENT_PKT
+ strlen(reg
->name
) + 4);
117 /* Print the field. */
118 fprintf(file
, "%s = ", field
->name
);
120 if (val
< field
->num_values
&& field
->values
[val
])
121 fprintf(file
, "%s\n", field
->values
[val
]);
123 print_value(file
, val
,
124 util_bitcount(field
->mask
));
132 fprintf(file
, COLOR_YELLOW
"0x%05x" COLOR_RESET
" = 0x%08x", offset
, value
);
135 static void si_parse_set_reg_packet(FILE *f
, uint32_t *ib
, unsigned count
,
138 unsigned reg
= (ib
[1] << 2) + reg_offset
;
141 for (i
= 0; i
< count
; i
++)
142 si_dump_reg(f
, reg
+ i
*4, ib
[2+i
], ~0);
145 static uint32_t *si_parse_packet3(FILE *f
, uint32_t *ib
, int *num_dw
,
148 unsigned count
= PKT_COUNT_G(ib
[0]);
149 unsigned op
= PKT3_IT_OPCODE_G(ib
[0]);
150 const char *predicate
= PKT3_PREDICATE(ib
[0]) ? "(predicate)" : "";
153 /* Print the name first. */
154 for (i
= 0; i
< ARRAY_SIZE(packet3_table
); i
++)
155 if (packet3_table
[i
].op
== op
)
158 if (i
< ARRAY_SIZE(packet3_table
))
159 if (op
== PKT3_SET_CONTEXT_REG
||
160 op
== PKT3_SET_CONFIG_REG
||
161 op
== PKT3_SET_UCONFIG_REG
||
162 op
== PKT3_SET_SH_REG
)
163 fprintf(f
, COLOR_CYAN
"%s%s" COLOR_CYAN
":\n",
164 packet3_table
[i
].name
, predicate
);
166 fprintf(f
, COLOR_GREEN
"%s%s" COLOR_RESET
":\n",
167 packet3_table
[i
].name
, predicate
);
169 fprintf(f
, COLOR_RED
"PKT3_UNKNOWN 0x%x%s" COLOR_RESET
":\n",
172 /* Print the contents. */
174 case PKT3_SET_CONTEXT_REG
:
175 si_parse_set_reg_packet(f
, ib
, count
, SI_CONTEXT_REG_OFFSET
);
177 case PKT3_SET_CONFIG_REG
:
178 si_parse_set_reg_packet(f
, ib
, count
, SI_CONFIG_REG_OFFSET
);
180 case PKT3_SET_UCONFIG_REG
:
181 si_parse_set_reg_packet(f
, ib
, count
, CIK_UCONFIG_REG_OFFSET
);
183 case PKT3_SET_SH_REG
:
184 si_parse_set_reg_packet(f
, ib
, count
, SI_SH_REG_OFFSET
);
186 case PKT3_DRAW_PREAMBLE
:
187 si_dump_reg(f
, R_030908_VGT_PRIMITIVE_TYPE
, ib
[1], ~0);
188 si_dump_reg(f
, R_028AA8_IA_MULTI_VGT_PARAM
, ib
[2], ~0);
189 si_dump_reg(f
, R_028B58_VGT_LS_HS_CONFIG
, ib
[3], ~0);
191 case PKT3_ACQUIRE_MEM
:
192 si_dump_reg(f
, R_0301F0_CP_COHER_CNTL
, ib
[1], ~0);
193 si_dump_reg(f
, R_0301F4_CP_COHER_SIZE
, ib
[2], ~0);
194 si_dump_reg(f
, R_030230_CP_COHER_SIZE_HI
, ib
[3], ~0);
195 si_dump_reg(f
, R_0301F8_CP_COHER_BASE
, ib
[4], ~0);
196 si_dump_reg(f
, R_0301E4_CP_COHER_BASE_HI
, ib
[5], ~0);
197 print_named_value(f
, "POLL_INTERVAL", ib
[6], 16);
199 case PKT3_SURFACE_SYNC
:
200 si_dump_reg(f
, R_0085F0_CP_COHER_CNTL
, ib
[1], ~0);
201 si_dump_reg(f
, R_0085F4_CP_COHER_SIZE
, ib
[2], ~0);
202 si_dump_reg(f
, R_0085F8_CP_COHER_BASE
, ib
[3], ~0);
203 print_named_value(f
, "POLL_INTERVAL", ib
[4], 16);
205 case PKT3_EVENT_WRITE
:
206 si_dump_reg(f
, R_028A90_VGT_EVENT_INITIATOR
, ib
[1],
207 S_028A90_EVENT_TYPE(~0));
208 print_named_value(f
, "EVENT_INDEX", (ib
[1] >> 8) & 0xf, 4);
209 print_named_value(f
, "INV_L2", (ib
[1] >> 20) & 0x1, 1);
211 print_named_value(f
, "ADDRESS_LO", ib
[2], 32);
212 print_named_value(f
, "ADDRESS_HI", ib
[3], 16);
215 case PKT3_DRAW_INDEX_AUTO
:
216 si_dump_reg(f
, R_030930_VGT_NUM_INDICES
, ib
[1], ~0);
217 si_dump_reg(f
, R_0287F0_VGT_DRAW_INITIATOR
, ib
[2], ~0);
219 case PKT3_DRAW_INDEX_2
:
220 si_dump_reg(f
, R_028A78_VGT_DMA_MAX_SIZE
, ib
[1], ~0);
221 si_dump_reg(f
, R_0287E8_VGT_DMA_BASE
, ib
[2], ~0);
222 si_dump_reg(f
, R_0287E4_VGT_DMA_BASE_HI
, ib
[3], ~0);
223 si_dump_reg(f
, R_030930_VGT_NUM_INDICES
, ib
[4], ~0);
224 si_dump_reg(f
, R_0287F0_VGT_DRAW_INITIATOR
, ib
[5], ~0);
226 case PKT3_INDEX_TYPE
:
227 si_dump_reg(f
, R_028A7C_VGT_DMA_INDEX_TYPE
, ib
[1], ~0);
229 case PKT3_NUM_INSTANCES
:
230 si_dump_reg(f
, R_030934_VGT_NUM_INSTANCES
, ib
[1], ~0);
232 case PKT3_WRITE_DATA
:
233 si_dump_reg(f
, R_370_CONTROL
, ib
[1], ~0);
234 si_dump_reg(f
, R_371_DST_ADDR_LO
, ib
[2], ~0);
235 si_dump_reg(f
, R_372_DST_ADDR_HI
, ib
[3], ~0);
236 for (i
= 2; i
< count
; i
++) {
237 print_spaces(f
, INDENT_PKT
);
238 fprintf(f
, "0x%08x\n", ib
[2+i
]);
242 si_dump_reg(f
, R_410_CP_DMA_WORD0
, ib
[1], ~0);
243 si_dump_reg(f
, R_411_CP_DMA_WORD1
, ib
[2], ~0);
244 si_dump_reg(f
, R_412_CP_DMA_WORD2
, ib
[3], ~0);
245 si_dump_reg(f
, R_413_CP_DMA_WORD3
, ib
[4], ~0);
246 si_dump_reg(f
, R_414_COMMAND
, ib
[5], ~0);
249 si_dump_reg(f
, R_500_DMA_DATA_WORD0
, ib
[1], ~0);
250 si_dump_reg(f
, R_501_SRC_ADDR_LO
, ib
[2], ~0);
251 si_dump_reg(f
, R_502_SRC_ADDR_HI
, ib
[3], ~0);
252 si_dump_reg(f
, R_503_DST_ADDR_LO
, ib
[4], ~0);
253 si_dump_reg(f
, R_504_DST_ADDR_HI
, ib
[5], ~0);
254 si_dump_reg(f
, R_414_COMMAND
, ib
[6], ~0);
257 if (ib
[0] == 0xffff1000) {
258 count
= -1; /* One dword NOP. */
260 } else if (count
== 0 && SI_IS_TRACE_POINT(ib
[1])) {
261 unsigned packet_id
= SI_GET_TRACE_POINT_ID(ib
[1]);
263 print_spaces(f
, INDENT_PKT
);
264 fprintf(f
, COLOR_RED
"Trace point ID: %u\n", packet_id
);
267 break; /* tracing was disabled */
269 print_spaces(f
, INDENT_PKT
);
270 if (packet_id
< trace_id
)
272 "This trace point was reached by the CP."
274 else if (packet_id
== trace_id
)
276 "!!!!! This is the last trace point that "
277 "was reached by the CP !!!!!"
279 else if (packet_id
+1 == trace_id
)
281 "!!!!! This is the first trace point that "
282 "was NOT been reached by the CP !!!!!"
286 "!!!!! This trace point was NOT reached "
291 /* fall through, print all dwords */
293 for (i
= 0; i
< count
+1; i
++) {
294 print_spaces(f
, INDENT_PKT
);
295 fprintf(f
, "0x%08x\n", ib
[1+i
]);
300 *num_dw
-= count
+ 2;
305 * Parse and print an IB into a file.
309 * \param num_dw size of the IB
310 * \param chip_class chip class
311 * \param trace_id the last trace ID that is known to have been reached
312 * and executed by the CP, typically read from a buffer
314 static void si_parse_ib(FILE *f
, uint32_t *ib
, int num_dw
, int trace_id
)
316 fprintf(f
, "------------------ IB begin ------------------\n");
319 unsigned type
= PKT_TYPE_G(ib
[0]);
323 ib
= si_parse_packet3(f
, ib
, &num_dw
, trace_id
);
327 if (ib
[0] == 0x80000000) {
328 fprintf(f
, COLOR_GREEN
"NOP (type 2)" COLOR_RESET
"\n");
334 fprintf(f
, "Unknown packet type %i\n", type
);
339 fprintf(f
, "------------------- IB end -------------------\n");
341 printf("Packet ends after the end of IB.\n");
346 static void si_dump_mmapped_reg(struct si_context
*sctx
, FILE *f
,
349 struct radeon_winsys
*ws
= sctx
->b
.ws
;
352 if (ws
->read_registers(ws
, offset
, 1, &value
))
353 si_dump_reg(f
, offset
, value
, ~0);
356 static void si_dump_debug_registers(struct si_context
*sctx
, FILE *f
)
358 if (sctx
->screen
->b
.info
.drm_major
== 2 &&
359 sctx
->screen
->b
.info
.drm_minor
< 42)
360 return; /* no radeon support */
362 fprintf(f
, "Memory-mapped registers:\n");
363 si_dump_mmapped_reg(sctx
, f
, R_008010_GRBM_STATUS
);
365 /* No other registers can be read on DRM < 3.1.0. */
366 if (sctx
->screen
->b
.info
.drm_major
< 3 ||
367 sctx
->screen
->b
.info
.drm_minor
< 1) {
372 si_dump_mmapped_reg(sctx
, f
, R_008008_GRBM_STATUS2
);
373 si_dump_mmapped_reg(sctx
, f
, R_008014_GRBM_STATUS_SE0
);
374 si_dump_mmapped_reg(sctx
, f
, R_008018_GRBM_STATUS_SE1
);
375 si_dump_mmapped_reg(sctx
, f
, R_008038_GRBM_STATUS_SE2
);
376 si_dump_mmapped_reg(sctx
, f
, R_00803C_GRBM_STATUS_SE3
);
377 si_dump_mmapped_reg(sctx
, f
, R_00D034_SDMA0_STATUS_REG
);
378 si_dump_mmapped_reg(sctx
, f
, R_00D834_SDMA1_STATUS_REG
);
379 si_dump_mmapped_reg(sctx
, f
, R_000E50_SRBM_STATUS
);
380 si_dump_mmapped_reg(sctx
, f
, R_000E4C_SRBM_STATUS2
);
381 si_dump_mmapped_reg(sctx
, f
, R_000E54_SRBM_STATUS3
);
382 si_dump_mmapped_reg(sctx
, f
, R_008680_CP_STAT
);
383 si_dump_mmapped_reg(sctx
, f
, R_008674_CP_STALLED_STAT1
);
384 si_dump_mmapped_reg(sctx
, f
, R_008678_CP_STALLED_STAT2
);
385 si_dump_mmapped_reg(sctx
, f
, R_008670_CP_STALLED_STAT3
);
386 si_dump_mmapped_reg(sctx
, f
, R_008210_CP_CPC_STATUS
);
387 si_dump_mmapped_reg(sctx
, f
, R_008214_CP_CPC_BUSY_STAT
);
388 si_dump_mmapped_reg(sctx
, f
, R_008218_CP_CPC_STALLED_STAT1
);
389 si_dump_mmapped_reg(sctx
, f
, R_00821C_CP_CPF_STATUS
);
390 si_dump_mmapped_reg(sctx
, f
, R_008220_CP_CPF_BUSY_STAT
);
391 si_dump_mmapped_reg(sctx
, f
, R_008224_CP_CPF_STALLED_STAT1
);
395 static void si_dump_debug_state(struct pipe_context
*ctx
, FILE *f
,
398 struct si_context
*sctx
= (struct si_context
*)ctx
;
400 if (flags
& PIPE_DEBUG_DEVICE_IS_HUNG
)
401 si_dump_debug_registers(sctx
, f
);
403 si_dump_shader(sctx
->vs_shader
, "Vertex", f
);
404 si_dump_shader(sctx
->tcs_shader
, "Tessellation control", f
);
405 si_dump_shader(sctx
->tes_shader
, "Tessellation evaluation", f
);
406 si_dump_shader(sctx
->gs_shader
, "Geometry", f
);
407 si_dump_shader(sctx
->ps_shader
, "Fragment", f
);
410 int last_trace_id
= -1;
412 if (sctx
->last_trace_buf
) {
413 /* We are expecting that the ddebug pipe has already
414 * waited for the context, so this buffer should be idle.
415 * If the GPU is hung, there is no point in waiting for it.
418 sctx
->b
.ws
->buffer_map(sctx
->last_trace_buf
->cs_buf
,
420 PIPE_TRANSFER_UNSYNCHRONIZED
|
423 last_trace_id
= *map
;
426 si_parse_ib(f
, sctx
->last_ib
, sctx
->last_ib_dw_size
,
428 free(sctx
->last_ib
); /* dump only once */
429 sctx
->last_ib
= NULL
;
430 r600_resource_reference(&sctx
->last_trace_buf
, NULL
);
433 fprintf(f
, "Done.\n");
436 void si_init_debug_functions(struct si_context
*sctx
)
438 sctx
->b
.b
.dump_debug_state
= si_dump_debug_state
;