radv: add RADV_DEBUG=pso_history

This dumps pipeline hash + shader VA to /tmp/radv_pso_history.log. Can
be very useful when investigating GPU hangs using UMR to get the fossils
back with the PC.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33912>

docs/envvars.rst

@@ -1402,6 +1402,9 @@ RADV driver environment variables
       dump vertex shader prologs
    ``psocachestats``
      dump PSO cache stats (hits/misses) to verify precompilation of shaders
+   ``pso_history``
+     dump PSO history (pipeline hash + shader VA) to /tmp/radv_pso_history.log.
+     Useful for debugging GPU hangs with UMR and Fossilize.
    ``shaders``
       dump shaders
    ``shaderstats``

src/amd/vulkan/radv_debug.h

@@ -70,6 +70,7 @@ enum {
    RADV_DEBUG_DUMP_NIR = 1ull << 55,
    RADV_DEBUG_DUMP_ASM = 1ull << 56,
    RADV_DEBUG_DUMP_BACKEND_IR = 1ull << 57,
+   RADV_DEBUG_PSO_HISTORY = 1ull << 58,
    RADV_DEBUG_DUMP_SHADERS = RADV_DEBUG_DUMP_VS | RADV_DEBUG_DUMP_TCS | RADV_DEBUG_DUMP_TES | RADV_DEBUG_DUMP_GS |
                              RADV_DEBUG_DUMP_PS | RADV_DEBUG_DUMP_TASK | RADV_DEBUG_DUMP_MESH | RADV_DEBUG_DUMP_CS |
                              RADV_DEBUG_DUMP_NIR | RADV_DEBUG_DUMP_ASM | RADV_DEBUG_DUMP_BACKEND_IR,

src/amd/vulkan/radv_instance.c

@@ -85,6 +85,7 @@ static const struct debug_control radv_debug_options[] = {{"nofastclears", RADV_
                                                           {"nir", RADV_DEBUG_DUMP_NIR},
                                                           {"asm", RADV_DEBUG_DUMP_ASM},
                                                           {"ir", RADV_DEBUG_DUMP_BACKEND_IR},
+                                                          {"pso_history", RADV_DEBUG_PSO_HISTORY},
                                                           {NULL, 0}};
 
 const char *
@@ -400,6 +401,14 @@ radv_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, const VkAllocationC
       instance->debug_flags |= shader_stage_flags;
    }
 
+   if (instance->debug_flags & RADV_DEBUG_PSO_HISTORY) {
+      const char *filename = "/tmp/radv_pso_history.log";
+
+      instance->pso_history_logfile = fopen(filename, "w");
+      if (!instance->pso_history_logfile)
+         fprintf(stderr, "radv: Failed to open log file: %s.\n", filename);
+   }
+
    /* When RADV_FORCE_FAMILY is set, the driver creates a null
     * device that allows to test the compiler without having an
     * AMDGPU instance.
@@ -433,6 +442,9 @@ radv_DestroyInstance(VkInstance _instance, const VkAllocationCallbacks *pAllocat
 
    VG(VALGRIND_DESTROY_MEMPOOL(instance));
 
+   if (instance->pso_history_logfile)
+      fclose(instance->pso_history_logfile);
+
    simple_mtx_destroy(&instance->shader_dump_mtx);
 
    driDestroyOptionCache(&instance->drirc.options);

src/amd/vulkan/radv_instance.h

@@ -80,6 +80,8 @@ struct radv_instance {
       int override_vram_size;
       int override_uniform_offset_alignment;
    } drirc;
+
+   FILE *pso_history_logfile;
 };
 
 VK_DEFINE_HANDLE_CASTS(radv_instance, vk.base, VkInstance, VK_OBJECT_TYPE_INSTANCE)

src/amd/vulkan/radv_pipeline.c

@@ -61,8 +61,8 @@ radv_pipeline_capture_shader_stats(const struct radv_device *device, VkPipelineC
 
    /* Capture shader statistics when RGP is enabled to correlate shader hashes with Fossilize. */
    return (flags & VK_PIPELINE_CREATE_2_CAPTURE_STATISTICS_BIT_KHR) ||
-          (instance->debug_flags & RADV_DEBUG_DUMP_SHADER_STATS) || device->keep_shader_info ||
-          (instance->vk.trace_mode & RADV_TRACE_MODE_RGP);
+          (instance->debug_flags & (RADV_DEBUG_DUMP_SHADER_STATS | RADV_DEBUG_PSO_HISTORY)) ||
+          device->keep_shader_info || (instance->vk.trace_mode & RADV_TRACE_MODE_RGP);
 }
 
 bool
@@ -1238,3 +1238,61 @@ radv_pipeline_hash_shader_stage(VkPipelineCreateFlags2 pipeline_flags, const VkP
    _mesa_sha1_update(ctx, shader_sha1, sizeof(shader_sha1));
    _mesa_sha1_update(ctx, stage_key, sizeof(*stage_key));
 }
+
+static void
+radv_print_pso_history(const struct radv_pipeline *pipeline, const struct radv_shader *shader, FILE *output)
+{
+   const uint64_t start_addr = radv_shader_get_va(shader) & ((1ull << 48) - 1);
+   const uint64_t end_addr = start_addr + shader->code_size;
+
+   fprintf(output, "pipeline_hash=%.16llx, VA=%.16llx-%.16llx, stage=%s\n", (long long)pipeline->pipeline_hash,
+           (long long)start_addr, (long long)end_addr, _mesa_shader_stage_to_string(shader->info.stage));
+   fflush(output);
+}
+
+void
+radv_pipeline_report_pso_history(const struct radv_device *device, struct radv_pipeline *pipeline)
+{
+   const struct radv_physical_device *pdev = radv_device_physical(device);
+   const struct radv_instance *instance = radv_physical_device_instance(pdev);
+   FILE *output = instance->pso_history_logfile ? instance->pso_history_logfile : stderr;
+
+   if (!(instance->debug_flags & RADV_DEBUG_PSO_HISTORY))
+      return;
+
+   /* Only report PSO history for application pipelines. */
+   if (pipeline->is_internal)
+      return;
+
+   switch (pipeline->type) {
+   case RADV_PIPELINE_GRAPHICS:
+      for (uint32_t i = 0; i < MESA_VULKAN_SHADER_STAGES; i++) {
+         const struct radv_shader *shader = pipeline->shaders[i];
+
+         if (shader)
+            radv_print_pso_history(pipeline, shader, output);
+      }
+
+      if (pipeline->gs_copy_shader)
+         radv_print_pso_history(pipeline, pipeline->gs_copy_shader, output);
+      break;
+   case RADV_PIPELINE_COMPUTE:
+      radv_print_pso_history(pipeline, pipeline->shaders[MESA_SHADER_COMPUTE], output);
+      break;
+   case RADV_PIPELINE_RAY_TRACING: {
+      struct radv_ray_tracing_pipeline *rt_pipeline = radv_pipeline_to_ray_tracing(pipeline);
+
+      radv_print_pso_history(pipeline, rt_pipeline->prolog, output);
+
+      for (uint32_t i = 0; i < rt_pipeline->stage_count; i++) {
+         const struct radv_shader *shader = rt_pipeline->stages[i].shader;
+
+         if (shader)
+            radv_print_pso_history(pipeline, shader, output);
+      }
+      break;
+   }
+   default:
+      break;
+   }
+}

src/amd/vulkan/radv_pipeline.h

@@ -110,4 +110,6 @@ void radv_pipeline_hash_shader_stage(VkPipelineCreateFlags2 pipeline_flags,
                                      const VkPipelineShaderStageCreateInfo *sinfo,
                                      const struct radv_shader_stage_key *stage_key, struct mesa_sha1 *ctx);
 
+void radv_pipeline_report_pso_history(const struct radv_device *device, struct radv_pipeline *pipeline);
+
 #endif /* RADV_PIPELINE_H */

src/amd/vulkan/radv_pipeline_compute.c

@@ -307,6 +307,8 @@ radv_compute_pipeline_create(VkDevice _device, VkPipelineCache _cache, const VkC
 
    radv_compute_pipeline_init(pipeline, pipeline_layout, pipeline->base.shaders[MESA_SHADER_COMPUTE]);
 
+   radv_pipeline_report_pso_history(device, &pipeline->base);
+
    *pPipeline = radv_pipeline_to_handle(&pipeline->base);
    radv_rmv_log_compute_pipeline_create(device, &pipeline->base, pipeline->base.is_internal);
    return VK_SUCCESS;

src/amd/vulkan/radv_pipeline_graphics.c

@@ -3484,6 +3484,8 @@ radv_graphics_pipeline_create(VkDevice _device, VkPipelineCache _cache, const Vk
       return result;
    }
 
+   radv_pipeline_report_pso_history(device, &pipeline->base);
+
    *pPipeline = radv_pipeline_to_handle(&pipeline->base);
    radv_rmv_log_graphics_pipeline_create(device, &pipeline->base, pipeline->base.is_internal);
    return VK_SUCCESS;
@@ -3592,6 +3594,8 @@ radv_graphics_lib_pipeline_create(VkDevice _device, VkPipelineCache _cache,
       return result;
    }
 
+   radv_pipeline_report_pso_history(device, &pipeline->base.base);
+
    *pPipeline = radv_pipeline_to_handle(&pipeline->base.base);
 
    return VK_SUCCESS;

src/amd/vulkan/radv_pipeline_rt.c

@@ -1157,6 +1157,8 @@ radv_rt_pipeline_create(VkDevice _device, VkPipelineCache _cache, const VkRayTra
       }
    }
 
+   radv_pipeline_report_pso_history(device, &pipeline->base.base);
+
    *pPipeline = radv_pipeline_to_handle(&pipeline->base.base);
    radv_rmv_log_rt_pipeline_create(device, pipeline);