#include <stdlib.h>
#include <stdio.h>
+#include <sys/utsname.h>
+#include "util/mesa-sha1.h"
#include "sid.h"
#include "gfx9d.h"
#include "ac_debug.h"
device->trace_bo = ws->buffer_create(ws, TRACE_BO_SIZE, 8,
RADEON_DOMAIN_VRAM,
- RADEON_FLAG_CPU_ACCESS);
+ RADEON_FLAG_CPU_ACCESS|
+ RADEON_FLAG_NO_INTERPROCESS_SHARING,
+ RADV_BO_PRIORITY_UPLOAD_BUFFER);
if (!device->trace_bo)
return false;
}
static void
-radv_dump_trace(struct radv_device *device, struct radeon_winsys_cs *cs)
+radv_dump_trace(struct radv_device *device, struct radeon_cmdbuf *cs)
{
const char *filename = getenv("RADV_TRACE_FILE");
FILE *f = fopen(filename, "w");
radv_dump_mmapped_reg(device, f, R_00803C_GRBM_STATUS_SE3);
radv_dump_mmapped_reg(device, f, R_00D034_SDMA0_STATUS_REG);
radv_dump_mmapped_reg(device, f, R_00D834_SDMA1_STATUS_REG);
- if (info->chip_class <= VI) {
+ if (info->chip_class <= GFX8) {
radv_dump_mmapped_reg(device, f, R_000E50_SRBM_STATUS);
radv_dump_mmapped_reg(device, f, R_000E4C_SRBM_STATUS2);
radv_dump_mmapped_reg(device, f, R_000E54_SRBM_STATUS3);
fprintf(f, "\n");
}
+static const char *
+radv_get_descriptor_name(enum VkDescriptorType type)
+{
+ switch (type) {
+ case VK_DESCRIPTOR_TYPE_SAMPLER:
+ return "SAMPLER";
+ case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
+ return "COMBINED_IMAGE_SAMPLER";
+ case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
+ return "SAMPLED_IMAGE";
+ case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
+ return "STORAGE_IMAGE";
+ case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
+ return "UNIFORM_TEXEL_BUFFER";
+ case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
+ return "STORAGE_TEXEL_BUFFER";
+ case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
+ return "UNIFORM_BUFFER";
+ case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
+ return "STORAGE_BUFFER";
+ case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
+ return "UNIFORM_BUFFER_DYNAMIC";
+ case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
+ return "STORAGE_BUFFER_DYNAMIC";
+ case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
+ return "INPUT_ATTACHMENT";
+ default:
+ return "UNKNOWN";
+ }
+}
+
+static void
+radv_dump_buffer_descriptor(enum chip_class chip_class, const uint32_t *desc,
+ FILE *f)
+{
+ fprintf(f, COLOR_CYAN " Buffer:" COLOR_RESET "\n");
+ for (unsigned j = 0; j < 4; j++)
+ ac_dump_reg(f, chip_class, R_008F00_SQ_BUF_RSRC_WORD0 + j * 4,
+ desc[j], 0xffffffff);
+}
+
+static void
+radv_dump_image_descriptor(enum chip_class chip_class, const uint32_t *desc,
+ FILE *f)
+{
+ fprintf(f, COLOR_CYAN " Image:" COLOR_RESET "\n");
+ for (unsigned j = 0; j < 8; j++)
+ ac_dump_reg(f, chip_class, R_008F10_SQ_IMG_RSRC_WORD0 + j * 4,
+ desc[j], 0xffffffff);
+
+ fprintf(f, COLOR_CYAN " FMASK:" COLOR_RESET "\n");
+ for (unsigned j = 0; j < 8; j++)
+ ac_dump_reg(f, chip_class, R_008F10_SQ_IMG_RSRC_WORD0 + j * 4,
+ desc[8 + j], 0xffffffff);
+}
+
+static void
+radv_dump_sampler_descriptor(enum chip_class chip_class, const uint32_t *desc,
+ FILE *f)
+{
+ fprintf(f, COLOR_CYAN " Sampler state:" COLOR_RESET "\n");
+ for (unsigned j = 0; j < 4; j++) {
+ ac_dump_reg(f, chip_class, R_008F30_SQ_IMG_SAMP_WORD0 + j * 4,
+ desc[j], 0xffffffff);
+ }
+}
+
+static void
+radv_dump_combined_image_sampler_descriptor(enum chip_class chip_class,
+ const uint32_t *desc, FILE *f)
+{
+ radv_dump_image_descriptor(chip_class, desc, f);
+ radv_dump_sampler_descriptor(chip_class, desc + 16, f);
+}
+
+static void
+radv_dump_descriptor_set(enum chip_class chip_class,
+ struct radv_descriptor_set *set, unsigned id, FILE *f)
+{
+ const struct radv_descriptor_set_layout *layout;
+ int i;
+
+ if (!set)
+ return;
+ layout = set->layout;
+
+ fprintf(f, "** descriptor set (%d) **\n", id);
+ fprintf(f, "va: 0x%"PRIx64"\n", set->va);
+ fprintf(f, "size: %d\n", set->size);
+ fprintf(f, "mapped_ptr:\n");
+
+ for (i = 0; i < set->size / 4; i++) {
+ fprintf(f, "\t[0x%x] = 0x%08x\n", i, set->mapped_ptr[i]);
+ }
+ fprintf(f, "\n");
+
+ fprintf(f, "\t*** layout ***\n");
+ fprintf(f, "\tbinding_count: %d\n", layout->binding_count);
+ fprintf(f, "\tsize: %d\n", layout->size);
+ fprintf(f, "\tshader_stages: %x\n", layout->shader_stages);
+ fprintf(f, "\tdynamic_shader_stages: %x\n",
+ layout->dynamic_shader_stages);
+ fprintf(f, "\tbuffer_count: %d\n", layout->buffer_count);
+ fprintf(f, "\tdynamic_offset_count: %d\n",
+ layout->dynamic_offset_count);
+ fprintf(f, "\n");
+
+ for (i = 0; i < set->layout->binding_count; i++) {
+ uint32_t *desc =
+ set->mapped_ptr + layout->binding[i].offset / 4;
+
+ fprintf(f, "\t\t**** binding layout (%d) ****\n", i);
+ fprintf(f, "\t\ttype: %s\n",
+ radv_get_descriptor_name(layout->binding[i].type));
+ fprintf(f, "\t\tarray_size: %d\n",
+ layout->binding[i].array_size);
+ fprintf(f, "\t\toffset: %d\n",
+ layout->binding[i].offset);
+ fprintf(f, "\t\tbuffer_offset: %d\n",
+ layout->binding[i].buffer_offset);
+ fprintf(f, "\t\tdynamic_offset_offset: %d\n",
+ layout->binding[i].dynamic_offset_offset);
+ fprintf(f, "\t\tdynamic_offset_count: %d\n",
+ layout->binding[i].dynamic_offset_count);
+ fprintf(f, "\t\tsize: %d\n",
+ layout->binding[i].size);
+ fprintf(f, "\t\timmutable_samplers_offset: %d\n",
+ layout->binding[i].immutable_samplers_offset);
+ fprintf(f, "\t\timmutable_samplers_equal: %d\n",
+ layout->binding[i].immutable_samplers_equal);
+ fprintf(f, "\n");
+
+ switch (layout->binding[i].type) {
+ case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER:
+ case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER:
+ case VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER:
+ case VK_DESCRIPTOR_TYPE_STORAGE_TEXEL_BUFFER:
+ radv_dump_buffer_descriptor(chip_class, desc, f);
+ break;
+ case VK_DESCRIPTOR_TYPE_STORAGE_IMAGE:
+ case VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE:
+ case VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT:
+ radv_dump_image_descriptor(chip_class, desc, f);
+ break;
+ case VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER:
+ radv_dump_combined_image_sampler_descriptor(chip_class, desc, f);
+ break;
+ case VK_DESCRIPTOR_TYPE_SAMPLER:
+ radv_dump_sampler_descriptor(chip_class, desc, f);
+ break;
+ case VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER_DYNAMIC:
+ case VK_DESCRIPTOR_TYPE_STORAGE_BUFFER_DYNAMIC:
+ /* todo */
+ break;
+ default:
+ assert(!"unknown descriptor type");
+ break;
+ }
+ fprintf(f, "\n");
+ }
+ fprintf(f, "\n\n");
+}
+
+static void
+radv_dump_descriptors(struct radv_pipeline *pipeline, FILE *f)
+{
+ struct radv_device *device = pipeline->device;
+ enum chip_class chip_class = device->physical_device->rad_info.chip_class;
+ uint64_t *ptr = (uint64_t *)device->trace_id_ptr;
+ int i;
+
+ fprintf(f, "List of descriptors:\n");
+ for (i = 0; i < MAX_SETS; i++) {
+ struct radv_descriptor_set *set =
+ (struct radv_descriptor_set *)ptr[i + 3];
+
+ radv_dump_descriptor_set(chip_class, set, i, f);
+ }
+}
+
struct radv_shader_inst {
char text[160]; /* one disasm line */
unsigned offset; /* instruction offset */
}
static void
-radv_dump_annotated_shader(struct radv_pipeline *pipeline,
- struct radv_shader_variant *shader,
- gl_shader_stage stage,
- struct ac_wave_info *waves, unsigned num_waves,
- FILE *f)
+radv_dump_annotated_shader(struct radv_shader_variant *shader,
+ gl_shader_stage stage, struct ac_wave_info *waves,
+ unsigned num_waves, FILE *f)
{
- struct radv_device *device = pipeline->device;
uint64_t start_addr, end_addr;
unsigned i;
if (!shader)
return;
- start_addr = device->ws->buffer_get_va(shader->bo) + shader->bo_offset;
+ start_addr = radv_buffer_get_va(shader->bo) + shader->bo_offset;
end_addr = start_addr + shader->code_size;
/* See if any wave executes the shader. */
static void
radv_dump_annotated_shaders(struct radv_pipeline *pipeline,
- struct radv_shader_variant *compute_shader,
- FILE *f)
+ VkShaderStageFlagBits active_stages, FILE *f)
{
struct ac_wave_info waves[AC_MAX_WAVES_PER_CHIP];
unsigned num_waves = ac_get_wave_info(waves);
- unsigned mask;
fprintf(f, COLOR_CYAN "The number of active waves = %u" COLOR_RESET
"\n\n", num_waves);
/* Dump annotated active graphics shaders. */
- mask = pipeline->active_stages;
- while (mask) {
- int stage = u_bit_scan(&mask);
+ while (active_stages) {
+ int stage = u_bit_scan(&active_stages);
- radv_dump_annotated_shader(pipeline, pipeline->shaders[stage],
+ radv_dump_annotated_shader(pipeline->shaders[stage],
stage, waves, num_waves, f);
}
- radv_dump_annotated_shader(pipeline, compute_shader,
- MESA_SHADER_COMPUTE, waves, num_waves, f);
-
/* Print waves executing shaders that are not currently bound. */
unsigned i;
bool found = false;
if (!shader)
return;
- fprintf(f, "%s:\n%s\n\n", radv_get_shader_name(shader, stage),
- shader->disasm_string);
+ fprintf(f, "%s:\n\n", radv_get_shader_name(shader, stage));
+
+ if (shader->spirv) {
+ unsigned char sha1[21];
+ char sha1buf[41];
+
+ _mesa_sha1_compute(shader->spirv, shader->spirv_size, sha1);
+ _mesa_sha1_format(sha1buf, sha1);
+
+ fprintf(f, "SPIRV (sha1: %s):\n", sha1buf);
+ radv_print_spirv(shader->spirv, shader->spirv_size, f);
+ }
+
+ if (shader->nir) {
+ fprintf(f, "NIR:\n");
+ nir_print_shader(shader->nir, f);
+ }
+
+ fprintf(f, "LLVM IR:\n%s\n", shader->llvm_ir_string);
+ fprintf(f, "DISASM:\n%s\n", shader->disasm_string);
radv_shader_dump_stats(pipeline->device, shader, stage, f);
}
static void
radv_dump_shaders(struct radv_pipeline *pipeline,
- struct radv_shader_variant *compute_shader, FILE *f)
+ VkShaderStageFlagBits active_stages, FILE *f)
{
- unsigned mask;
-
/* Dump active graphics shaders. */
- mask = pipeline->active_stages;
- while (mask) {
- int stage = u_bit_scan(&mask);
+ while (active_stages) {
+ int stage = u_bit_scan(&active_stages);
radv_dump_shader(pipeline, pipeline->shaders[stage], stage, f);
}
+}
- radv_dump_shader(pipeline, compute_shader, MESA_SHADER_COMPUTE, f);
+static void
+radv_dump_pipeline_state(struct radv_pipeline *pipeline,
+ VkShaderStageFlagBits active_stages, FILE *f)
+{
+ radv_dump_shaders(pipeline, active_stages, f);
+ radv_dump_annotated_shaders(pipeline, active_stages, f);
+ radv_dump_descriptors(pipeline, f);
}
static void
radv_dump_graphics_state(struct radv_pipeline *graphics_pipeline,
struct radv_pipeline *compute_pipeline, FILE *f)
{
- struct radv_shader_variant *compute_shader =
- compute_pipeline ? compute_pipeline->shaders[MESA_SHADER_COMPUTE] : NULL;
+ VkShaderStageFlagBits active_stages;
- if (!graphics_pipeline)
- return;
+ if (graphics_pipeline) {
+ active_stages = graphics_pipeline->active_stages;
+ radv_dump_pipeline_state(graphics_pipeline, active_stages, f);
+ }
- radv_dump_shaders(graphics_pipeline, compute_shader, f);
- radv_dump_annotated_shaders(graphics_pipeline, compute_shader, f);
+ if (compute_pipeline) {
+ active_stages = VK_SHADER_STAGE_COMPUTE_BIT;
+ radv_dump_pipeline_state(compute_pipeline, active_stages, f);
+ }
}
static void
radv_dump_compute_state(struct radv_pipeline *compute_pipeline, FILE *f)
{
+ VkShaderStageFlagBits active_stages = VK_SHADER_STAGE_COMPUTE_BIT;
+
if (!compute_pipeline)
return;
- radv_dump_shaders(compute_pipeline,
- compute_pipeline->shaders[MESA_SHADER_COMPUTE], f);
- radv_dump_annotated_shaders(compute_pipeline,
- compute_pipeline->shaders[MESA_SHADER_COMPUTE],
- f);
+ radv_dump_pipeline_state(compute_pipeline, active_stages, f);
}
static struct radv_pipeline *
return (struct radv_pipeline *)ptr[2];
}
+static void
+radv_dump_dmesg(FILE *f)
+{
+ char line[2000];
+ FILE *p;
+
+ p = popen("dmesg | tail -n60", "r");
+ if (!p)
+ return;
+
+ fprintf(f, "\nLast 60 lines of dmesg:\n\n");
+ while (fgets(line, sizeof(line), p))
+ fputs(line, f);
+ fprintf(f, "\n");
+
+ pclose(p);
+}
+
+void
+radv_dump_enabled_options(struct radv_device *device, FILE *f)
+{
+ uint64_t mask;
+
+ if (device->instance->debug_flags) {
+ fprintf(f, "Enabled debug options: ");
+
+ mask = device->instance->debug_flags;
+ while (mask) {
+ int i = u_bit_scan64(&mask);
+ fprintf(f, "%s, ", radv_get_debug_option_name(i));
+ }
+ fprintf(f, "\n");
+ }
+
+ if (device->instance->perftest_flags) {
+ fprintf(f, "Enabled perftest options: ");
+
+ mask = device->instance->perftest_flags;
+ while (mask) {
+ int i = u_bit_scan64(&mask);
+ fprintf(f, "%s, ", radv_get_perftest_option_name(i));
+ }
+ fprintf(f, "\n");
+ }
+}
+
+static void
+radv_dump_device_name(struct radv_device *device, FILE *f)
+{
+ struct radeon_info *info = &device->physical_device->rad_info;
+ char kernel_version[128] = {};
+ struct utsname uname_data;
+ const char *chip_name;
+
+ chip_name = device->ws->get_chip_name(device->ws);
+
+ if (uname(&uname_data) == 0)
+ snprintf(kernel_version, sizeof(kernel_version),
+ " / %s", uname_data.release);
+
+ fprintf(f, "Device name: %s (%s DRM %i.%i.%i%s, LLVM "
+ MESA_LLVM_VERSION_STRING ")\n\n",
+ chip_name, device->physical_device->name,
+ info->drm_major, info->drm_minor, info->drm_patchlevel,
+ kernel_version);
+}
+
static bool
radv_gpu_hang_occured(struct radv_queue *queue, enum ring_type ring)
{
}
void
-radv_check_gpu_hangs(struct radv_queue *queue, struct radeon_winsys_cs *cs)
+radv_check_gpu_hangs(struct radv_queue *queue, struct radeon_cmdbuf *cs)
{
struct radv_pipeline *graphics_pipeline, *compute_pipeline;
struct radv_device *device = queue->device;
graphics_pipeline = radv_get_saved_graphics_pipeline(device);
compute_pipeline = radv_get_saved_compute_pipeline(device);
+ fprintf(stderr, "GPU hang report:\n\n");
+ radv_dump_device_name(device, stderr);
+
+ radv_dump_enabled_options(device, stderr);
+ radv_dump_dmesg(stderr);
+
if (vm_fault_occurred) {
fprintf(stderr, "VM fault report.\n\n");
fprintf(stderr, "Failing VM page: 0x%08"PRIx64"\n\n", addr);
}
void
-radv_print_spirv(struct radv_shader_module *module, FILE *fp)
+radv_print_spirv(uint32_t *data, uint32_t size, FILE *fp)
{
char path[] = "/tmp/fileXXXXXX";
char line[2048], command[128];
if (fd < 0)
return;
- if (write(fd, module->data, module->size) == -1)
+ if (write(fd, data, size) == -1)
goto fail;
sprintf(command, "spirv-dis %s", path);