*/
#include "si_pipe.h"
-#include "si_shader.h"
#include "sid.h"
#include "sid_tables.h"
+#include "radeon/radeon_elf_util.h"
#include "ddebug/dd_util.h"
+#include "util/u_memory.h"
+#include "ac_debug.h"
+DEBUG_GET_ONCE_OPTION(replace_shaders, "RADEON_REPLACE_SHADERS", NULL)
-static void si_dump_shader(struct si_shader_ctx_state *state, const char *name,
- FILE *f)
+static void si_dump_shader(struct si_screen *sscreen,
+ struct si_shader_ctx_state *state, FILE *f)
{
- if (!state->cso || !state->current)
+ struct si_shader *current = state->current;
+
+ if (!state->cso || !current)
return;
- fprintf(f, "%s shader disassembly:\n", name);
- si_dump_shader_key(state->cso->type, &state->current->key, f);
- fprintf(f, "%s\n\n", state->current->binary.disasm_string);
+ if (current->shader_log)
+ fwrite(current->shader_log, current->shader_log_size, 1, f);
+ else
+ si_shader_dump(sscreen, state->current, NULL,
+ state->cso->info.processor, f, false);
}
-/* Parsed IBs are difficult to read without colors. Use "less -R file" to
- * read them, or use "aha -b -f file" to convert them to html.
+/**
+ * Shader compiles can be overridden with arbitrary ELF objects by setting
+ * the environment variable RADEON_REPLACE_SHADERS=num1:filename1[;num2:filename2]
*/
-#define COLOR_RESET "\033[0m"
-#define COLOR_RED "\033[31m"
-#define COLOR_GREEN "\033[1;32m"
-#define COLOR_YELLOW "\033[1;33m"
-#define COLOR_CYAN "\033[1;36m"
-
-#define INDENT_PKT 8
-
-static void print_spaces(FILE *f, unsigned num)
+bool si_replace_shader(unsigned num, struct radeon_shader_binary *binary)
{
- fprintf(f, "%*s", num, "");
-}
-
-static void print_value(FILE *file, uint32_t value, int bits)
-{
- /* Guess if it's int or float */
- if (value <= (1 << 15))
- fprintf(file, "%u\n", value);
- else {
- float f = uif(value);
-
- if (fabs(f) < 100000 && f*10 == floor(f*10))
- fprintf(file, "%.1ff\n", f);
- else
- /* Don't print more leading zeros than there are bits. */
- fprintf(file, "0x%0*x\n", bits / 4, value);
- }
-}
-
-static void print_named_value(FILE *file, const char *name, uint32_t value,
- int bits)
-{
- print_spaces(file, INDENT_PKT);
- fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ", name);
- print_value(file, value, bits);
-}
-
-static void si_dump_reg(FILE *file, unsigned offset, uint32_t value,
- uint32_t field_mask)
-{
- int r, f;
-
- for (r = 0; r < ARRAY_SIZE(reg_table); r++) {
- const struct si_reg *reg = ®_table[r];
-
- if (reg->offset == offset) {
- bool first_field = true;
-
- print_spaces(file, INDENT_PKT);
- fprintf(file, COLOR_YELLOW "%s" COLOR_RESET " <- ",
- reg->name);
-
- if (!reg->num_fields) {
- print_value(file, value, 32);
- return;
- }
+ const char *p = debug_get_option_replace_shaders();
+ const char *semicolon;
+ char *copy = NULL;
+ FILE *f;
+ long filesize, nread;
+ char *buf = NULL;
+ bool replaced = false;
- for (f = 0; f < reg->num_fields; f++) {
- const struct si_field *field = ®->fields[f];
- uint32_t val = (value & field->mask) >>
- (ffs(field->mask) - 1);
+ if (!p)
+ return false;
- if (!(field->mask & field_mask))
- continue;
+ while (*p) {
+ unsigned long i;
+ char *endp;
+ i = strtoul(p, &endp, 0);
- /* Indent the field. */
- if (!first_field)
- print_spaces(file,
- INDENT_PKT + strlen(reg->name) + 4);
+ p = endp;
+ if (*p != ':') {
+ fprintf(stderr, "RADEON_REPLACE_SHADERS formatted badly.\n");
+ exit(1);
+ }
+ ++p;
- /* Print the field. */
- fprintf(file, "%s = ", field->name);
+ if (i == num)
+ break;
- if (val < field->num_values && field->values[val])
- fprintf(file, "%s\n", field->values[val]);
- else
- print_value(file, val,
- util_bitcount(field->mask));
+ p = strchr(p, ';');
+ if (!p)
+ return false;
+ ++p;
+ }
+ if (!*p)
+ return false;
- first_field = false;
- }
- return;
+ semicolon = strchr(p, ';');
+ if (semicolon) {
+ p = copy = strndup(p, semicolon - p);
+ if (!copy) {
+ fprintf(stderr, "out of memory\n");
+ return false;
}
}
- fprintf(file, COLOR_YELLOW "0x%05x" COLOR_RESET " = 0x%08x", offset, value);
-}
-
-static void si_parse_set_reg_packet(FILE *f, uint32_t *ib, unsigned count,
- unsigned reg_offset)
-{
- unsigned reg = (ib[1] << 2) + reg_offset;
- int i;
+ fprintf(stderr, "radeonsi: replace shader %u by %s\n", num, p);
- for (i = 0; i < count; i++)
- si_dump_reg(f, reg + i*4, ib[2+i], ~0);
-}
+ f = fopen(p, "r");
+ if (!f) {
+ perror("radeonsi: failed to open file");
+ goto out_free;
+ }
-static uint32_t *si_parse_packet3(FILE *f, uint32_t *ib, int *num_dw,
- int trace_id)
-{
- unsigned count = PKT_COUNT_G(ib[0]);
- unsigned op = PKT3_IT_OPCODE_G(ib[0]);
- const char *predicate = PKT3_PREDICATE(ib[0]) ? "(predicate)" : "";
- int i;
+ if (fseek(f, 0, SEEK_END) != 0)
+ goto file_error;
- /* Print the name first. */
- for (i = 0; i < ARRAY_SIZE(packet3_table); i++)
- if (packet3_table[i].op == op)
- break;
+ filesize = ftell(f);
+ if (filesize < 0)
+ goto file_error;
- if (i < ARRAY_SIZE(packet3_table))
- if (op == PKT3_SET_CONTEXT_REG ||
- op == PKT3_SET_CONFIG_REG ||
- op == PKT3_SET_UCONFIG_REG ||
- op == PKT3_SET_SH_REG)
- fprintf(f, COLOR_CYAN "%s%s" COLOR_CYAN ":\n",
- packet3_table[i].name, predicate);
- else
- fprintf(f, COLOR_GREEN "%s%s" COLOR_RESET ":\n",
- packet3_table[i].name, predicate);
- else
- fprintf(f, COLOR_RED "PKT3_UNKNOWN 0x%x%s" COLOR_RESET ":\n",
- op, predicate);
+ if (fseek(f, 0, SEEK_SET) != 0)
+ goto file_error;
- /* Print the contents. */
- switch (op) {
- case PKT3_SET_CONTEXT_REG:
- si_parse_set_reg_packet(f, ib, count, SI_CONTEXT_REG_OFFSET);
- break;
- case PKT3_SET_CONFIG_REG:
- si_parse_set_reg_packet(f, ib, count, SI_CONFIG_REG_OFFSET);
- break;
- case PKT3_SET_UCONFIG_REG:
- si_parse_set_reg_packet(f, ib, count, CIK_UCONFIG_REG_OFFSET);
- break;
- case PKT3_SET_SH_REG:
- si_parse_set_reg_packet(f, ib, count, SI_SH_REG_OFFSET);
- break;
- case PKT3_DRAW_PREAMBLE:
- si_dump_reg(f, R_030908_VGT_PRIMITIVE_TYPE, ib[1], ~0);
- si_dump_reg(f, R_028AA8_IA_MULTI_VGT_PARAM, ib[2], ~0);
- si_dump_reg(f, R_028B58_VGT_LS_HS_CONFIG, ib[3], ~0);
- break;
- case PKT3_ACQUIRE_MEM:
- si_dump_reg(f, R_0301F0_CP_COHER_CNTL, ib[1], ~0);
- si_dump_reg(f, R_0301F4_CP_COHER_SIZE, ib[2], ~0);
- si_dump_reg(f, R_030230_CP_COHER_SIZE_HI, ib[3], ~0);
- si_dump_reg(f, R_0301F8_CP_COHER_BASE, ib[4], ~0);
- si_dump_reg(f, R_0301E4_CP_COHER_BASE_HI, ib[5], ~0);
- print_named_value(f, "POLL_INTERVAL", ib[6], 16);
- break;
- case PKT3_SURFACE_SYNC:
- si_dump_reg(f, R_0085F0_CP_COHER_CNTL, ib[1], ~0);
- si_dump_reg(f, R_0085F4_CP_COHER_SIZE, ib[2], ~0);
- si_dump_reg(f, R_0085F8_CP_COHER_BASE, ib[3], ~0);
- print_named_value(f, "POLL_INTERVAL", ib[4], 16);
- break;
- case PKT3_EVENT_WRITE:
- si_dump_reg(f, R_028A90_VGT_EVENT_INITIATOR, ib[1],
- S_028A90_EVENT_TYPE(~0));
- print_named_value(f, "EVENT_INDEX", (ib[1] >> 8) & 0xf, 4);
- print_named_value(f, "INV_L2", (ib[1] >> 20) & 0x1, 1);
- if (count > 0) {
- print_named_value(f, "ADDRESS_LO", ib[2], 32);
- print_named_value(f, "ADDRESS_HI", ib[3], 16);
- }
- break;
- case PKT3_DRAW_INDEX_AUTO:
- si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[1], ~0);
- si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[2], ~0);
- break;
- case PKT3_DRAW_INDEX_2:
- si_dump_reg(f, R_028A78_VGT_DMA_MAX_SIZE, ib[1], ~0);
- si_dump_reg(f, R_0287E8_VGT_DMA_BASE, ib[2], ~0);
- si_dump_reg(f, R_0287E4_VGT_DMA_BASE_HI, ib[3], ~0);
- si_dump_reg(f, R_030930_VGT_NUM_INDICES, ib[4], ~0);
- si_dump_reg(f, R_0287F0_VGT_DRAW_INITIATOR, ib[5], ~0);
- break;
- case PKT3_INDEX_TYPE:
- si_dump_reg(f, R_028A7C_VGT_DMA_INDEX_TYPE, ib[1], ~0);
- break;
- case PKT3_NUM_INSTANCES:
- si_dump_reg(f, R_030934_VGT_NUM_INSTANCES, ib[1], ~0);
- break;
- case PKT3_WRITE_DATA:
- si_dump_reg(f, R_370_CONTROL, ib[1], ~0);
- si_dump_reg(f, R_371_DST_ADDR_LO, ib[2], ~0);
- si_dump_reg(f, R_372_DST_ADDR_HI, ib[3], ~0);
- for (i = 2; i < count; i++) {
- print_spaces(f, INDENT_PKT);
- fprintf(f, "0x%08x\n", ib[2+i]);
- }
- break;
- case PKT3_CP_DMA:
- si_dump_reg(f, R_410_CP_DMA_WORD0, ib[1], ~0);
- si_dump_reg(f, R_411_CP_DMA_WORD1, ib[2], ~0);
- si_dump_reg(f, R_412_CP_DMA_WORD2, ib[3], ~0);
- si_dump_reg(f, R_413_CP_DMA_WORD3, ib[4], ~0);
- si_dump_reg(f, R_414_COMMAND, ib[5], ~0);
- break;
- case PKT3_DMA_DATA:
- si_dump_reg(f, R_500_DMA_DATA_WORD0, ib[1], ~0);
- si_dump_reg(f, R_501_SRC_ADDR_LO, ib[2], ~0);
- si_dump_reg(f, R_502_SRC_ADDR_HI, ib[3], ~0);
- si_dump_reg(f, R_503_DST_ADDR_LO, ib[4], ~0);
- si_dump_reg(f, R_504_DST_ADDR_HI, ib[5], ~0);
- si_dump_reg(f, R_414_COMMAND, ib[6], ~0);
- break;
- case PKT3_NOP:
- if (ib[0] == 0xffff1000) {
- count = -1; /* One dword NOP. */
- break;
- } else if (count == 0 && SI_IS_TRACE_POINT(ib[1])) {
- unsigned packet_id = SI_GET_TRACE_POINT_ID(ib[1]);
-
- print_spaces(f, INDENT_PKT);
- fprintf(f, COLOR_RED "Trace point ID: %u\n", packet_id);
-
- if (trace_id == -1)
- break; /* tracing was disabled */
-
- print_spaces(f, INDENT_PKT);
- if (packet_id < trace_id)
- fprintf(f, COLOR_RED
- "This trace point was reached by the CP."
- COLOR_RESET "\n");
- else if (packet_id == trace_id)
- fprintf(f, COLOR_RED
- "!!!!! This is the last trace point that "
- "was reached by the CP !!!!!"
- COLOR_RESET "\n");
- else if (packet_id+1 == trace_id)
- fprintf(f, COLOR_RED
- "!!!!! This is the first trace point that "
- "was NOT been reached by the CP !!!!!"
- COLOR_RESET "\n");
- else
- fprintf(f, COLOR_RED
- "!!!!! This trace point was NOT reached "
- "by the CP !!!!!"
- COLOR_RESET "\n");
- break;
- }
- /* fall through, print all dwords */
- default:
- for (i = 0; i < count+1; i++) {
- print_spaces(f, INDENT_PKT);
- fprintf(f, "0x%08x\n", ib[1+i]);
- }
+ buf = MALLOC(filesize);
+ if (!buf) {
+ fprintf(stderr, "out of memory\n");
+ goto out_close;
}
- ib += count + 2;
- *num_dw -= count + 2;
- return ib;
-}
-
-/**
- * Parse and print an IB into a file.
- *
- * \param f file
- * \param ib IB
- * \param num_dw size of the IB
- * \param chip_class chip class
- * \param trace_id the last trace ID that is known to have been reached
- * and executed by the CP, typically read from a buffer
- */
-static void si_parse_ib(FILE *f, uint32_t *ib, int num_dw, int trace_id,
- const char *name)
-{
- fprintf(f, "------------------ %s begin ------------------\n", name);
+ nread = fread(buf, 1, filesize, f);
+ if (nread != filesize)
+ goto file_error;
- while (num_dw > 0) {
- unsigned type = PKT_TYPE_G(ib[0]);
+ radeon_elf_read(buf, filesize, binary);
+ replaced = true;
- switch (type) {
- case 3:
- ib = si_parse_packet3(f, ib, &num_dw, trace_id);
- break;
- case 2:
- /* type-2 nop */
- if (ib[0] == 0x80000000) {
- fprintf(f, COLOR_GREEN "NOP (type 2)" COLOR_RESET "\n");
- ib++;
- break;
- }
- /* fall through */
- default:
- fprintf(f, "Unknown packet type %i\n", type);
- return;
- }
- }
-
- fprintf(f, "------------------- %s end -------------------\n", name);
- if (num_dw < 0) {
- printf("Packet ends after the end of IB.\n");
- exit(0);
- }
- fprintf(f, "\n");
+out_close:
+ fclose(f);
+out_free:
+ FREE(buf);
+ free(copy);
+ return replaced;
+
+file_error:
+ perror("radeonsi: reading shader");
+ goto out_close;
}
+/* Parsed IBs are difficult to read without colors. Use "less -R file" to
+ * read them, or use "aha -b -f file" to convert them to html.
+ */
+#define COLOR_RESET "\033[0m"
+#define COLOR_RED "\033[31m"
+#define COLOR_GREEN "\033[1;32m"
+#define COLOR_YELLOW "\033[1;33m"
+#define COLOR_CYAN "\033[1;36m"
+
static void si_dump_mmapped_reg(struct si_context *sctx, FILE *f,
unsigned offset)
{
uint32_t value;
if (ws->read_registers(ws, offset, 1, &value))
- si_dump_reg(f, offset, value, ~0);
+ ac_dump_reg(f, offset, value, ~0);
}
static void si_dump_debug_registers(struct si_context *sctx, FILE *f)
{
int last_trace_id = -1;
- if (!sctx->last_ib)
+ if (!sctx->last_gfx.ib)
return;
if (sctx->last_trace_buf) {
* waited for the context, so this buffer should be idle.
* If the GPU is hung, there is no point in waiting for it.
*/
- uint32_t *map = sctx->b.ws->buffer_map(sctx->last_trace_buf->cs_buf,
+ uint32_t *map = sctx->b.ws->buffer_map(sctx->last_trace_buf->buf,
NULL,
PIPE_TRANSFER_UNSYNCHRONIZED |
PIPE_TRANSFER_READ);
}
if (sctx->init_config)
- si_parse_ib(f, sctx->init_config->pm4, sctx->init_config->ndw,
- -1, "IB2: Init config");
+ ac_parse_ib(f, sctx->init_config->pm4, sctx->init_config->ndw,
+ -1, "IB2: Init config", sctx->b.chip_class);
if (sctx->init_config_gs_rings)
- si_parse_ib(f, sctx->init_config_gs_rings->pm4,
+ ac_parse_ib(f, sctx->init_config_gs_rings->pm4,
sctx->init_config_gs_rings->ndw,
- -1, "IB2: Init GS rings");
+ -1, "IB2: Init GS rings", sctx->b.chip_class);
- si_parse_ib(f, sctx->last_ib, sctx->last_ib_dw_size,
- last_trace_id, "IB");
- free(sctx->last_ib); /* dump only once */
- sctx->last_ib = NULL;
- r600_resource_reference(&sctx->last_trace_buf, NULL);
+ ac_parse_ib(f, sctx->last_gfx.ib, sctx->last_gfx.num_dw,
+ last_trace_id, "IB", sctx->b.chip_class);
}
static const char *priority_to_string(enum radeon_bo_priority priority)
ITEM(IB2),
ITEM(DRAW_INDIRECT),
ITEM(INDEX_BUFFER),
- ITEM(CP_DMA),
ITEM(VCE),
ITEM(UVD),
ITEM(SDMA_BUFFER),
ITEM(SDMA_TEXTURE),
- ITEM(USER_SHADER),
- ITEM(INTERNAL_SHADER),
+ ITEM(CP_DMA),
ITEM(CONST_BUFFER),
ITEM(DESCRIPTORS),
ITEM(BORDER_COLORS),
ITEM(SAMPLER_BUFFER),
ITEM(VERTEX_BUFFER),
ITEM(SHADER_RW_BUFFER),
- ITEM(RINGS_STREAMOUT),
- ITEM(SCRATCH_BUFFER),
ITEM(COMPUTE_GLOBAL),
ITEM(SAMPLER_TEXTURE),
ITEM(SHADER_RW_IMAGE),
ITEM(CMASK),
ITEM(DCC),
ITEM(HTILE),
+ ITEM(SHADER_BINARY),
+ ITEM(SHADER_RINGS),
+ ITEM(SCRATCH_BUFFER),
};
#undef ITEM
a->vm_address > b->vm_address ? 1 : 0;
}
-static void si_dump_last_bo_list(struct si_context *sctx, FILE *f)
+static void si_dump_bo_list(struct si_context *sctx,
+ const struct radeon_saved_cs *saved, FILE *f)
{
unsigned i,j;
- if (!sctx->last_bo_list)
+ if (!saved->bo_list)
return;
/* Sort the list according to VM adddresses first. */
- qsort(sctx->last_bo_list, sctx->last_bo_count,
- sizeof(sctx->last_bo_list[0]), (void*)bo_list_compare_va);
+ qsort(saved->bo_list, saved->bo_count,
+ sizeof(saved->bo_list[0]), (void*)bo_list_compare_va);
fprintf(f, "Buffer list (in units of pages = 4kB):\n"
COLOR_YELLOW " Size VM start page "
"VM end page Usage" COLOR_RESET "\n");
- for (i = 0; i < sctx->last_bo_count; i++) {
+ for (i = 0; i < saved->bo_count; i++) {
/* Note: Buffer sizes are expected to be aligned to 4k by the winsys. */
- const unsigned page_size = 4096;
- uint64_t va = sctx->last_bo_list[i].vm_address;
- uint64_t size = sctx->last_bo_list[i].buf->size;
+ const unsigned page_size = sctx->b.screen->info.gart_page_size;
+ uint64_t va = saved->bo_list[i].vm_address;
+ uint64_t size = saved->bo_list[i].bo_size;
bool hit = false;
/* If there's unused virtual memory between 2 buffers, print it. */
if (i) {
- uint64_t previous_va_end = sctx->last_bo_list[i-1].vm_address +
- sctx->last_bo_list[i-1].buf->size;
+ uint64_t previous_va_end = saved->bo_list[i-1].vm_address +
+ saved->bo_list[i-1].bo_size;
if (va > previous_va_end) {
fprintf(f, " %10"PRIu64" -- hole --\n",
}
/* Print the buffer. */
- fprintf(f, " %10"PRIu64" 0x%013"PRIx64" 0x%013"PRIx64" ",
+ fprintf(f, " %10"PRIu64" 0x%013"PRIX64" 0x%013"PRIX64" ",
size / page_size, va / page_size, (va + size) / page_size);
/* Print the usage. */
for (j = 0; j < 64; j++) {
- if (!(sctx->last_bo_list[i].priority_usage & (1llu << j)))
+ if (!(saved->bo_list[i].priority_usage & (1llu << j)))
continue;
fprintf(f, "%s%s", !hit ? "" : ", ", priority_to_string(j));
}
fprintf(f, "\nNote: The holes represent memory not used by the IB.\n"
" Other buffers can still be allocated there.\n\n");
-
- for (i = 0; i < sctx->last_bo_count; i++)
- pb_reference(&sctx->last_bo_list[i].buf, NULL);
- free(sctx->last_bo_list);
- sctx->last_bo_list = NULL;
}
static void si_dump_framebuffer(struct si_context *sctx, FILE *f)
}
}
+static void si_dump_descriptor_list(struct si_descriptors *desc,
+ const char *shader_name,
+ const char *elem_name,
+ unsigned num_elements,
+ FILE *f)
+{
+ unsigned i, j;
+ uint32_t *cpu_list = desc->list;
+ uint32_t *gpu_list = desc->gpu_list;
+ const char *list_note = "GPU list";
+
+ if (!gpu_list) {
+ gpu_list = cpu_list;
+ list_note = "CPU list";
+ }
+
+ for (i = 0; i < num_elements; i++) {
+ fprintf(f, COLOR_GREEN "%s%s slot %u (%s):" COLOR_RESET "\n",
+ shader_name, elem_name, i, list_note);
+
+ switch (desc->element_dw_size) {
+ case 4:
+ for (j = 0; j < 4; j++)
+ ac_dump_reg(f, R_008F00_SQ_BUF_RSRC_WORD0 + j*4,
+ gpu_list[j], 0xffffffff);
+ break;
+ case 8:
+ for (j = 0; j < 8; j++)
+ ac_dump_reg(f, R_008F10_SQ_IMG_RSRC_WORD0 + j*4,
+ gpu_list[j], 0xffffffff);
+
+ fprintf(f, COLOR_CYAN " Buffer:" COLOR_RESET "\n");
+ for (j = 0; j < 4; j++)
+ ac_dump_reg(f, R_008F00_SQ_BUF_RSRC_WORD0 + j*4,
+ gpu_list[4+j], 0xffffffff);
+ break;
+ case 16:
+ for (j = 0; j < 8; j++)
+ ac_dump_reg(f, R_008F10_SQ_IMG_RSRC_WORD0 + j*4,
+ gpu_list[j], 0xffffffff);
+
+ fprintf(f, COLOR_CYAN " Buffer:" COLOR_RESET "\n");
+ for (j = 0; j < 4; j++)
+ ac_dump_reg(f, R_008F00_SQ_BUF_RSRC_WORD0 + j*4,
+ gpu_list[4+j], 0xffffffff);
+
+ fprintf(f, COLOR_CYAN " FMASK:" COLOR_RESET "\n");
+ for (j = 0; j < 8; j++)
+ ac_dump_reg(f, R_008F10_SQ_IMG_RSRC_WORD0 + j*4,
+ gpu_list[8+j], 0xffffffff);
+
+ fprintf(f, COLOR_CYAN " Sampler state:" COLOR_RESET "\n");
+ for (j = 0; j < 4; j++)
+ ac_dump_reg(f, R_008F30_SQ_IMG_SAMP_WORD0 + j*4,
+ gpu_list[12+j], 0xffffffff);
+ break;
+ }
+
+ if (memcmp(gpu_list, cpu_list, desc->element_dw_size * 4) != 0) {
+ fprintf(f, COLOR_RED "!!!!! This slot was corrupted in GPU memory !!!!!"
+ COLOR_RESET "\n");
+ }
+
+ fprintf(f, "\n");
+ gpu_list += desc->element_dw_size;
+ cpu_list += desc->element_dw_size;
+ }
+}
+
+static void si_dump_descriptors(struct si_context *sctx,
+ struct si_shader_ctx_state *state,
+ FILE *f)
+{
+ if (!state->cso || !state->current)
+ return;
+
+ unsigned type = state->cso->type;
+ const struct tgsi_shader_info *info = &state->cso->info;
+ struct si_descriptors *descs =
+ &sctx->descriptors[SI_DESCS_FIRST_SHADER +
+ type * SI_NUM_SHADER_DESCS];
+ static const char *shader_name[] = {"VS", "PS", "GS", "TCS", "TES", "CS"};
+
+ static const char *elem_name[] = {
+ " - Constant buffer",
+ " - Shader buffer",
+ " - Sampler",
+ " - Image",
+ };
+ unsigned num_elements[] = {
+ util_last_bit(info->const_buffers_declared),
+ util_last_bit(info->shader_buffers_declared),
+ util_last_bit(info->samplers_declared),
+ util_last_bit(info->images_declared),
+ };
+
+ if (type == PIPE_SHADER_VERTEX) {
+ si_dump_descriptor_list(&sctx->vertex_buffers, shader_name[type],
+ " - Vertex buffer", info->num_inputs, f);
+ }
+
+ for (unsigned i = 0; i < SI_NUM_SHADER_DESCS; ++i, ++descs)
+ si_dump_descriptor_list(descs, shader_name[type], elem_name[i],
+ num_elements[i], f);
+}
+
static void si_dump_debug_state(struct pipe_context *ctx, FILE *f,
unsigned flags)
{
struct si_context *sctx = (struct si_context*)ctx;
- if (flags & PIPE_DEBUG_DEVICE_IS_HUNG)
+ if (flags & PIPE_DUMP_DEVICE_STATUS_REGISTERS)
si_dump_debug_registers(sctx, f);
- si_dump_framebuffer(sctx, f);
- si_dump_shader(&sctx->vs_shader, "Vertex", f);
- si_dump_shader(&sctx->tcs_shader, "Tessellation control", f);
- si_dump_shader(&sctx->tes_shader, "Tessellation evaluation", f);
- si_dump_shader(&sctx->gs_shader, "Geometry", f);
- si_dump_shader(&sctx->ps_shader, "Fragment", f);
+ if (flags & PIPE_DUMP_CURRENT_STATES)
+ si_dump_framebuffer(sctx, f);
+
+ if (flags & PIPE_DUMP_CURRENT_SHADERS) {
+ si_dump_shader(sctx->screen, &sctx->vs_shader, f);
+ si_dump_shader(sctx->screen, &sctx->tcs_shader, f);
+ si_dump_shader(sctx->screen, &sctx->tes_shader, f);
+ si_dump_shader(sctx->screen, &sctx->gs_shader, f);
+ si_dump_shader(sctx->screen, &sctx->ps_shader, f);
+
+ si_dump_descriptor_list(&sctx->descriptors[SI_DESCS_RW_BUFFERS],
+ "", "RW buffers", SI_NUM_RW_BUFFERS, f);
+ si_dump_descriptors(sctx, &sctx->vs_shader, f);
+ si_dump_descriptors(sctx, &sctx->tcs_shader, f);
+ si_dump_descriptors(sctx, &sctx->tes_shader, f);
+ si_dump_descriptors(sctx, &sctx->gs_shader, f);
+ si_dump_descriptors(sctx, &sctx->ps_shader, f);
+ }
+
+ if (flags & PIPE_DUMP_LAST_COMMAND_BUFFER) {
+ si_dump_bo_list(sctx, &sctx->last_gfx, f);
+ si_dump_last_ib(sctx, f);
+
+ fprintf(f, "Done.\n");
+
+ /* dump only once */
+ radeon_clear_saved_cs(&sctx->last_gfx);
+ r600_resource_reference(&sctx->last_trace_buf, NULL);
+ }
+}
+
+static void si_dump_dma(struct si_context *sctx,
+ struct radeon_saved_cs *saved, FILE *f)
+{
+ static const char ib_name[] = "sDMA IB";
+ unsigned i;
+
+ si_dump_bo_list(sctx, saved, f);
+
+ fprintf(f, "------------------ %s begin ------------------\n", ib_name);
- si_dump_last_bo_list(sctx, f);
- si_dump_last_ib(sctx, f);
+ for (i = 0; i < saved->num_dw; ++i) {
+ fprintf(f, " %08x\n", saved->ib[i]);
+ }
+
+ fprintf(f, "------------------- %s end -------------------\n", ib_name);
+ fprintf(f, "\n");
- fprintf(f, "Done.\n");
+ fprintf(f, "SDMA Dump Done.\n");
}
static bool si_vm_fault_occured(struct si_context *sctx, uint32_t *out_addr)
while (fgets(line, sizeof(line), p)) {
char *msg, len;
+ if (!line[0] || line[0] == '\n')
+ continue;
+
/* Get the timestamp. */
if (sscanf(line, "[%u.%u]", &sec, &usec) != 2) {
- assert(0);
+ static bool hit = false;
+ if (!hit) {
+ fprintf(stderr, "%s: failed to parse line '%s'\n",
+ __func__, line);
+ hit = true;
+ }
continue;
}
timestamp = sec * 1000000llu + usec;
return fault;
}
-void si_check_vm_faults(struct si_context *sctx)
+void si_check_vm_faults(struct r600_common_context *ctx,
+ struct radeon_saved_cs *saved, enum ring_type ring)
{
+ struct si_context *sctx = (struct si_context *)ctx;
struct pipe_screen *screen = sctx->b.b.screen;
FILE *f;
uint32_t addr;
-
- /* Use conservative timeout 800ms, after which we won't wait any
- * longer and assume the GPU is hung.
- */
- sctx->b.ws->fence_wait(sctx->b.ws, sctx->last_gfx_fence, 800*1000*1000);
+ char cmd_line[4096];
if (!si_vm_fault_occured(sctx, &addr))
return;
- f = dd_get_debug_file();
+ f = dd_get_debug_file(false);
if (!f)
return;
fprintf(f, "VM fault report.\n\n");
+ if (os_get_command_line(cmd_line, sizeof(cmd_line)))
+ fprintf(f, "Command: %s\n", cmd_line);
fprintf(f, "Driver vendor: %s\n", screen->get_vendor(screen));
fprintf(f, "Device vendor: %s\n", screen->get_device_vendor(screen));
fprintf(f, "Device name: %s\n\n", screen->get_name(screen));
fprintf(f, "Failing VM page: 0x%08x\n\n", addr);
- si_dump_last_bo_list(sctx, f);
- si_dump_last_ib(sctx, f);
+ if (sctx->apitrace_call_number)
+ fprintf(f, "Last apitrace call: %u\n\n",
+ sctx->apitrace_call_number);
+
+ switch (ring) {
+ case RING_GFX:
+ si_dump_debug_state(&sctx->b.b, f,
+ PIPE_DUMP_CURRENT_STATES |
+ PIPE_DUMP_CURRENT_SHADERS |
+ PIPE_DUMP_LAST_COMMAND_BUFFER);
+ break;
+
+ case RING_DMA:
+ si_dump_dma(sctx, saved, f);
+ break;
+
+ default:
+ break;
+ }
+
fclose(f);
fprintf(stderr, "Detected a VM fault, exiting...\n");
void si_init_debug_functions(struct si_context *sctx)
{
sctx->b.b.dump_debug_state = si_dump_debug_state;
+ sctx->b.check_vm_faults = si_check_vm_faults;
/* Set the initial dmesg timestamp for this context, so that
* only new messages will be checked for VM faults.
projects / mesa.git / blobdiff