Merge pull request #3653 from ReinUsesLisp/nsight-aftermath

renderer_vulkan: Integrate Nvidia Nsight Aftermath on Windows
This commit is contained in:
Fernando Sahmkow 2020-04-22 11:39:01 -04:00 committed by GitHub
commit afae40a99e
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
12 changed files with 389 additions and 77 deletions

View file

@ -160,6 +160,8 @@ if (ENABLE_VULKAN)
renderer_vulkan/fixed_pipeline_state.h renderer_vulkan/fixed_pipeline_state.h
renderer_vulkan/maxwell_to_vk.cpp renderer_vulkan/maxwell_to_vk.cpp
renderer_vulkan/maxwell_to_vk.h renderer_vulkan/maxwell_to_vk.h
renderer_vulkan/nsight_aftermath_tracker.cpp
renderer_vulkan/nsight_aftermath_tracker.h
renderer_vulkan/renderer_vulkan.h renderer_vulkan/renderer_vulkan.h
renderer_vulkan/renderer_vulkan.cpp renderer_vulkan/renderer_vulkan.cpp
renderer_vulkan/vk_blit_screen.cpp renderer_vulkan/vk_blit_screen.cpp
@ -213,19 +215,30 @@ if (ENABLE_VULKAN)
renderer_vulkan/wrapper.cpp renderer_vulkan/wrapper.cpp
renderer_vulkan/wrapper.h renderer_vulkan/wrapper.h
) )
target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
target_compile_definitions(video_core PRIVATE HAS_VULKAN)
endif() endif()
create_target_directory_groups(video_core) create_target_directory_groups(video_core)
target_link_libraries(video_core PUBLIC common core) target_link_libraries(video_core PUBLIC common core)
target_link_libraries(video_core PRIVATE glad) target_link_libraries(video_core PRIVATE glad)
if (ENABLE_VULKAN) if (ENABLE_VULKAN)
target_include_directories(video_core PRIVATE sirit ../../externals/Vulkan-Headers/include)
target_compile_definitions(video_core PRIVATE HAS_VULKAN)
target_link_libraries(video_core PRIVATE sirit) target_link_libraries(video_core PRIVATE sirit)
endif() endif()
if (ENABLE_NSIGHT_AFTERMATH)
if (NOT DEFINED ENV{NSIGHT_AFTERMATH_SDK})
message(ERROR "Environment variable NSIGHT_AFTERMATH_SDK has to be provided")
endif()
if (NOT WIN32)
message(ERROR "Nsight Aftermath doesn't support non-Windows platforms")
endif()
target_compile_definitions(video_core PRIVATE HAS_NSIGHT_AFTERMATH)
target_include_directories(video_core PRIVATE "$ENV{NSIGHT_AFTERMATH_SDK}/include")
endif()
if (MSVC) if (MSVC)
target_compile_options(video_core PRIVATE /we4267) target_compile_options(video_core PRIVATE /we4267)
else() else()

View file

@ -0,0 +1,220 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#ifdef HAS_NSIGHT_AFTERMATH
#include <mutex>
#include <string>
#include <string_view>
#include <utility>
#include <vector>
#include <fmt/format.h>
#define VK_NO_PROTOTYPES
#include <vulkan/vulkan.h>
#include <GFSDK_Aftermath.h>
#include <GFSDK_Aftermath_Defines.h>
#include <GFSDK_Aftermath_GpuCrashDump.h>
#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
#include "common/common_paths.h"
#include "common/common_types.h"
#include "common/file_util.h"
#include "common/logging/log.h"
#include "common/scope_exit.h"
#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
namespace Vulkan {
static constexpr char AFTERMATH_LIB_NAME[] = "GFSDK_Aftermath_Lib.x64.dll";
NsightAftermathTracker::NsightAftermathTracker() = default;
NsightAftermathTracker::~NsightAftermathTracker() {
if (initialized) {
(void)GFSDK_Aftermath_DisableGpuCrashDumps();
}
}
bool NsightAftermathTracker::Initialize() {
if (!dl.Open(AFTERMATH_LIB_NAME)) {
LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath DLL");
return false;
}
if (!dl.GetSymbol("GFSDK_Aftermath_DisableGpuCrashDumps",
&GFSDK_Aftermath_DisableGpuCrashDumps) ||
!dl.GetSymbol("GFSDK_Aftermath_EnableGpuCrashDumps",
&GFSDK_Aftermath_EnableGpuCrashDumps) ||
!dl.GetSymbol("GFSDK_Aftermath_GetShaderDebugInfoIdentifier",
&GFSDK_Aftermath_GetShaderDebugInfoIdentifier) ||
!dl.GetSymbol("GFSDK_Aftermath_GetShaderHashSpirv", &GFSDK_Aftermath_GetShaderHashSpirv) ||
!dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_CreateDecoder",
&GFSDK_Aftermath_GpuCrashDump_CreateDecoder) ||
!dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_DestroyDecoder",
&GFSDK_Aftermath_GpuCrashDump_DestroyDecoder) ||
!dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GenerateJSON",
&GFSDK_Aftermath_GpuCrashDump_GenerateJSON) ||
!dl.GetSymbol("GFSDK_Aftermath_GpuCrashDump_GetJSON",
&GFSDK_Aftermath_GpuCrashDump_GetJSON)) {
LOG_ERROR(Render_Vulkan, "Failed to load Nsight Aftermath function pointers");
return false;
}
dump_dir = FileUtil::GetUserPath(FileUtil::UserPath::LogDir) + "gpucrash";
(void)FileUtil::DeleteDirRecursively(dump_dir);
if (!FileUtil::CreateDir(dump_dir)) {
LOG_ERROR(Render_Vulkan, "Failed to create Nsight Aftermath dump directory");
return false;
}
if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_EnableGpuCrashDumps(
GFSDK_Aftermath_Version_API, GFSDK_Aftermath_GpuCrashDumpWatchedApiFlags_Vulkan,
GFSDK_Aftermath_GpuCrashDumpFeatureFlags_Default, GpuCrashDumpCallback,
ShaderDebugInfoCallback, CrashDumpDescriptionCallback, this))) {
LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_EnableGpuCrashDumps failed");
return false;
}
LOG_INFO(Render_Vulkan, "Nsight Aftermath dump directory is \"{}\"", dump_dir);
initialized = true;
return true;
}
void NsightAftermathTracker::SaveShader(const std::vector<u32>& spirv) const {
if (!initialized) {
return;
}
std::vector<u32> spirv_copy = spirv;
GFSDK_Aftermath_SpirvCode shader;
shader.pData = spirv_copy.data();
shader.size = static_cast<u32>(spirv_copy.size() * 4);
std::scoped_lock lock{mutex};
GFSDK_Aftermath_ShaderHash hash;
if (!GFSDK_Aftermath_SUCCEED(
GFSDK_Aftermath_GetShaderHashSpirv(GFSDK_Aftermath_Version_API, &shader, &hash))) {
LOG_ERROR(Render_Vulkan, "Failed to hash SPIR-V module");
return;
}
FileUtil::IOFile file(fmt::format("{}/source_{:016x}.spv", dump_dir, hash.hash), "wb");
if (!file.IsOpen()) {
LOG_ERROR(Render_Vulkan, "Failed to dump SPIR-V module with hash={:016x}", hash.hash);
return;
}
if (file.WriteArray(spirv.data(), spirv.size()) != spirv.size()) {
LOG_ERROR(Render_Vulkan, "Failed to write SPIR-V module with hash={:016x}", hash.hash);
return;
}
}
void NsightAftermathTracker::OnGpuCrashDumpCallback(const void* gpu_crash_dump,
u32 gpu_crash_dump_size) {
std::scoped_lock lock{mutex};
LOG_CRITICAL(Render_Vulkan, "called");
GFSDK_Aftermath_GpuCrashDump_Decoder decoder;
if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_CreateDecoder(
GFSDK_Aftermath_Version_API, gpu_crash_dump, gpu_crash_dump_size, &decoder))) {
LOG_ERROR(Render_Vulkan, "Failed to create decoder");
return;
}
SCOPE_EXIT({ GFSDK_Aftermath_GpuCrashDump_DestroyDecoder(decoder); });
u32 json_size = 0;
if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GpuCrashDump_GenerateJSON(
decoder, GFSDK_Aftermath_GpuCrashDumpDecoderFlags_ALL_INFO,
GFSDK_Aftermath_GpuCrashDumpFormatterFlags_NONE, nullptr, nullptr, nullptr, nullptr,
this, &json_size))) {
LOG_ERROR(Render_Vulkan, "Failed to generate JSON");
return;
}
std::vector<char> json(json_size);
if (!GFSDK_Aftermath_SUCCEED(
GFSDK_Aftermath_GpuCrashDump_GetJSON(decoder, json_size, json.data()))) {
LOG_ERROR(Render_Vulkan, "Failed to query JSON");
return;
}
const std::string base_name = [this] {
const int id = dump_id++;
if (id == 0) {
return fmt::format("{}/crash.nv-gpudmp", dump_dir);
} else {
return fmt::format("{}/crash_{}.nv-gpudmp", dump_dir, id);
}
}();
std::string_view dump_view(static_cast<const char*>(gpu_crash_dump), gpu_crash_dump_size);
if (FileUtil::WriteStringToFile(false, base_name, dump_view) != gpu_crash_dump_size) {
LOG_ERROR(Render_Vulkan, "Failed to write dump file");
return;
}
const std::string_view json_view(json.data(), json.size());
if (FileUtil::WriteStringToFile(true, base_name + ".json", json_view) != json.size()) {
LOG_ERROR(Render_Vulkan, "Failed to write JSON");
return;
}
}
void NsightAftermathTracker::OnShaderDebugInfoCallback(const void* shader_debug_info,
u32 shader_debug_info_size) {
std::scoped_lock lock{mutex};
GFSDK_Aftermath_ShaderDebugInfoIdentifier identifier;
if (!GFSDK_Aftermath_SUCCEED(GFSDK_Aftermath_GetShaderDebugInfoIdentifier(
GFSDK_Aftermath_Version_API, shader_debug_info, shader_debug_info_size, &identifier))) {
LOG_ERROR(Render_Vulkan, "GFSDK_Aftermath_GetShaderDebugInfoIdentifier failed");
return;
}
const std::string path =
fmt::format("{}/shader_{:016x}{:016x}.nvdbg", dump_dir, identifier.id[0], identifier.id[1]);
FileUtil::IOFile file(path, "wb");
if (!file.IsOpen()) {
LOG_ERROR(Render_Vulkan, "Failed to create file {}", path);
return;
}
if (file.WriteBytes(static_cast<const u8*>(shader_debug_info), shader_debug_info_size) !=
shader_debug_info_size) {
LOG_ERROR(Render_Vulkan, "Failed to write file {}", path);
return;
}
}
void NsightAftermathTracker::OnCrashDumpDescriptionCallback(
PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description) {
add_description(GFSDK_Aftermath_GpuCrashDumpDescriptionKey_ApplicationName, "yuzu");
}
void NsightAftermathTracker::GpuCrashDumpCallback(const void* gpu_crash_dump,
u32 gpu_crash_dump_size, void* user_data) {
static_cast<NsightAftermathTracker*>(user_data)->OnGpuCrashDumpCallback(gpu_crash_dump,
gpu_crash_dump_size);
}
void NsightAftermathTracker::ShaderDebugInfoCallback(const void* shader_debug_info,
u32 shader_debug_info_size, void* user_data) {
static_cast<NsightAftermathTracker*>(user_data)->OnShaderDebugInfoCallback(
shader_debug_info, shader_debug_info_size);
}
void NsightAftermathTracker::CrashDumpDescriptionCallback(
PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data) {
static_cast<NsightAftermathTracker*>(user_data)->OnCrashDumpDescriptionCallback(
add_description);
}
} // namespace Vulkan
#endif // HAS_NSIGHT_AFTERMATH

View file

@ -0,0 +1,87 @@
// Copyright 2020 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include <mutex>
#include <string>
#include <vector>
#define VK_NO_PROTOTYPES
#include <vulkan/vulkan.h>
#ifdef HAS_NSIGHT_AFTERMATH
#include <GFSDK_Aftermath_Defines.h>
#include <GFSDK_Aftermath_GpuCrashDump.h>
#include <GFSDK_Aftermath_GpuCrashDumpDecoding.h>
#endif
#include "common/common_types.h"
#include "common/dynamic_library.h"
namespace Vulkan {
class NsightAftermathTracker {
public:
NsightAftermathTracker();
~NsightAftermathTracker();
NsightAftermathTracker(const NsightAftermathTracker&) = delete;
NsightAftermathTracker& operator=(const NsightAftermathTracker&) = delete;
// Delete move semantics because Aftermath initialization uses a pointer to this.
NsightAftermathTracker(NsightAftermathTracker&&) = delete;
NsightAftermathTracker& operator=(NsightAftermathTracker&&) = delete;
bool Initialize();
void SaveShader(const std::vector<u32>& spirv) const;
private:
#ifdef HAS_NSIGHT_AFTERMATH
static void GpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size,
void* user_data);
static void ShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size,
void* user_data);
static void CrashDumpDescriptionCallback(
PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description, void* user_data);
void OnGpuCrashDumpCallback(const void* gpu_crash_dump, u32 gpu_crash_dump_size);
void OnShaderDebugInfoCallback(const void* shader_debug_info, u32 shader_debug_info_size);
void OnCrashDumpDescriptionCallback(
PFN_GFSDK_Aftermath_AddGpuCrashDumpDescription add_description);
mutable std::mutex mutex;
std::string dump_dir;
int dump_id = 0;
bool initialized = false;
Common::DynamicLibrary dl;
PFN_GFSDK_Aftermath_DisableGpuCrashDumps GFSDK_Aftermath_DisableGpuCrashDumps;
PFN_GFSDK_Aftermath_EnableGpuCrashDumps GFSDK_Aftermath_EnableGpuCrashDumps;
PFN_GFSDK_Aftermath_GetShaderDebugInfoIdentifier GFSDK_Aftermath_GetShaderDebugInfoIdentifier;
PFN_GFSDK_Aftermath_GetShaderHashSpirv GFSDK_Aftermath_GetShaderHashSpirv;
PFN_GFSDK_Aftermath_GpuCrashDump_CreateDecoder GFSDK_Aftermath_GpuCrashDump_CreateDecoder;
PFN_GFSDK_Aftermath_GpuCrashDump_DestroyDecoder GFSDK_Aftermath_GpuCrashDump_DestroyDecoder;
PFN_GFSDK_Aftermath_GpuCrashDump_GenerateJSON GFSDK_Aftermath_GpuCrashDump_GenerateJSON;
PFN_GFSDK_Aftermath_GpuCrashDump_GetJSON GFSDK_Aftermath_GpuCrashDump_GetJSON;
#endif
};
#ifndef HAS_NSIGHT_AFTERMATH
inline NsightAftermathTracker::NsightAftermathTracker() = default;
inline NsightAftermathTracker::~NsightAftermathTracker() = default;
inline bool NsightAftermathTracker::Initialize() {
return false;
}
inline void NsightAftermathTracker::SaveShader(const std::vector<u32>&) const {}
#endif
} // namespace Vulkan

View file

@ -105,6 +105,8 @@ vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplat
} }
vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const { vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
device.SaveShader(code);
VkShaderModuleCreateInfo ci; VkShaderModuleCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO; ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
ci.pNext = nullptr; ci.pNext = nullptr;

View file

@ -9,6 +9,7 @@
#include <string_view> #include <string_view>
#include <thread> #include <thread>
#include <unordered_set> #include <unordered_set>
#include <utility>
#include <vector> #include <vector>
#include "common/assert.h" #include "common/assert.h"
@ -167,6 +168,7 @@ bool VKDevice::Create() {
VkPhysicalDeviceFeatures2 features2; VkPhysicalDeviceFeatures2 features2;
features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
features2.pNext = nullptr; features2.pNext = nullptr;
const void* first_next = &features2;
void** next = &features2.pNext; void** next = &features2.pNext;
auto& features = features2.features; auto& features = features2.features;
@ -296,7 +298,19 @@ bool VKDevice::Create() {
LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted"); LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
} }
logical = vk::Device::Create(physical, queue_cis, extensions, features2, dld); VkDeviceDiagnosticsConfigCreateInfoNV diagnostics_nv;
if (nv_device_diagnostics_config) {
nsight_aftermath_tracker.Initialize();
diagnostics_nv.sType = VK_STRUCTURE_TYPE_DEVICE_DIAGNOSTICS_CONFIG_CREATE_INFO_NV;
diagnostics_nv.pNext = &features2;
diagnostics_nv.flags = VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_SHADER_DEBUG_INFO_BIT_NV |
VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_RESOURCE_TRACKING_BIT_NV |
VK_DEVICE_DIAGNOSTICS_CONFIG_ENABLE_AUTOMATIC_CHECKPOINTS_BIT_NV;
first_next = &diagnostics_nv;
}
logical = vk::Device::Create(physical, queue_cis, extensions, first_next, dld);
if (!logical) { if (!logical) {
LOG_ERROR(Render_Vulkan, "Failed to create logical device"); LOG_ERROR(Render_Vulkan, "Failed to create logical device");
return false; return false;
@ -344,17 +358,12 @@ VkFormat VKDevice::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFla
void VKDevice::ReportLoss() const { void VKDevice::ReportLoss() const {
LOG_CRITICAL(Render_Vulkan, "Device loss occured!"); LOG_CRITICAL(Render_Vulkan, "Device loss occured!");
// Wait some time to let the log flush // Wait for the log to flush and for Nsight Aftermath to dump the results
std::this_thread::sleep_for(std::chrono::seconds{1}); std::this_thread::sleep_for(std::chrono::seconds{3});
if (!nv_device_diagnostic_checkpoints) {
return;
} }
[[maybe_unused]] const std::vector data = graphics_queue.GetCheckpointDataNV(dld); void VKDevice::SaveShader(const std::vector<u32>& spirv) const {
// Catch here in debug builds (or with optimizations disabled) the last graphics pipeline to be nsight_aftermath_tracker.SaveShader(spirv);
// executed. It can be done on a debugger by evaluating the expression:
// *(VKGraphicsPipeline*)data[0]
} }
bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const { bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const {
@ -527,8 +536,8 @@ std::vector<const char*> VKDevice::LoadExtensions() {
Test(extension, has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME, Test(extension, has_ext_transform_feedback, VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME,
false); false);
if (Settings::values.renderer_debug) { if (Settings::values.renderer_debug) {
Test(extension, nv_device_diagnostic_checkpoints, Test(extension, nv_device_diagnostics_config,
VK_NV_DEVICE_DIAGNOSTIC_CHECKPOINTS_EXTENSION_NAME, true); VK_NV_DEVICE_DIAGNOSTICS_CONFIG_EXTENSION_NAME, true);
} }
} }

View file

@ -10,6 +10,7 @@
#include <vector> #include <vector>
#include "common/common_types.h" #include "common/common_types.h"
#include "video_core/renderer_vulkan/nsight_aftermath_tracker.h"
#include "video_core/renderer_vulkan/wrapper.h" #include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan { namespace Vulkan {
@ -43,6 +44,9 @@ public:
/// Reports a device loss. /// Reports a device loss.
void ReportLoss() const; void ReportLoss() const;
/// Reports a shader to Nsight Aftermath.
void SaveShader(const std::vector<u32>& spirv) const;
/// Returns the dispatch loader with direct function pointers of the device. /// Returns the dispatch loader with direct function pointers of the device.
const vk::DeviceDispatch& GetDispatchLoader() const { const vk::DeviceDispatch& GetDispatchLoader() const {
return dld; return dld;
@ -173,11 +177,6 @@ public:
return ext_transform_feedback; return ext_transform_feedback;
} }
/// Returns true if the device supports VK_NV_device_diagnostic_checkpoints.
bool IsNvDeviceDiagnosticCheckpoints() const {
return nv_device_diagnostic_checkpoints;
}
/// Returns the vendor name reported from Vulkan. /// Returns the vendor name reported from Vulkan.
std::string_view GetVendorName() const { std::string_view GetVendorName() const {
return vendor_name; return vendor_name;
@ -233,7 +232,7 @@ private:
bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted. bool ext_depth_range_unrestricted{}; ///< Support for VK_EXT_depth_range_unrestricted.
bool ext_shader_viewport_index_layer{}; ///< Support for VK_EXT_shader_viewport_index_layer. bool ext_shader_viewport_index_layer{}; ///< Support for VK_EXT_shader_viewport_index_layer.
bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback. bool ext_transform_feedback{}; ///< Support for VK_EXT_transform_feedback.
bool nv_device_diagnostic_checkpoints{}; ///< Support for VK_NV_device_diagnostic_checkpoints. bool nv_device_diagnostics_config{}; ///< Support for VK_NV_device_diagnostics_config.
// Telemetry parameters // Telemetry parameters
std::string vendor_name; ///< Device's driver name. std::string vendor_name; ///< Device's driver name.
@ -241,6 +240,9 @@ private:
/// Format properties dictionary. /// Format properties dictionary.
std::unordered_map<VkFormat, VkFormatProperties> format_properties; std::unordered_map<VkFormat, VkFormatProperties> format_properties;
/// Nsight Aftermath GPU crash tracker
NsightAftermathTracker nsight_aftermath_tracker;
}; };
} // namespace Vulkan } // namespace Vulkan

View file

@ -148,6 +148,8 @@ std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
continue; continue;
} }
device.SaveShader(stage->code);
ci.codeSize = stage->code.size() * sizeof(u32); ci.codeSize = stage->code.size() * sizeof(u32);
ci.pCode = stage->code.data(); ci.pCode = stage->code.data();
modules.push_back(device.GetLogical().CreateShaderModule(ci)); modules.push_back(device.GetLogical().CreateShaderModule(ci));

View file

@ -113,8 +113,19 @@ u64 HostCounter::BlockingQuery() const {
if (ticks >= cache.Scheduler().Ticks()) { if (ticks >= cache.Scheduler().Ticks()) {
cache.Scheduler().Flush(); cache.Scheduler().Flush();
} }
return cache.Device().GetLogical().GetQueryResult<u64>( u64 data;
query.first, query.second, VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT); const VkResult result = cache.Device().GetLogical().GetQueryResults(
query.first, query.second, 1, sizeof(data), &data, sizeof(data),
VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
switch (result) {
case VK_SUCCESS:
return data;
case VK_ERROR_DEVICE_LOST:
cache.Device().ReportLoss();
[[fallthrough]];
default:
throw vk::Exception(result);
}
} }
} // namespace Vulkan } // namespace Vulkan

View file

@ -347,11 +347,6 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
buffer_bindings.Bind(scheduler); buffer_bindings.Bind(scheduler);
if (device.IsNvDeviceDiagnosticCheckpoints()) {
scheduler.Record(
[&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(&pipeline); });
}
BeginTransformFeedback(); BeginTransformFeedback();
const auto pipeline_layout = pipeline.GetLayout(); const auto pipeline_layout = pipeline.GetLayout();
@ -478,11 +473,6 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
TransitionImages(image_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, TransitionImages(image_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT); VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
if (device.IsNvDeviceDiagnosticCheckpoints()) {
scheduler.Record(
[&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(nullptr); });
}
scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y, scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
grid_z = launch_desc.grid_dim_z, pipeline_handle = pipeline.GetHandle(), grid_z = launch_desc.grid_dim_z, pipeline_handle = pipeline.GetHandle(),
layout = pipeline.GetLayout(), layout = pipeline.GetLayout(),

View file

@ -166,7 +166,15 @@ void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
submit_info.pCommandBuffers = current_cmdbuf.address(); submit_info.pCommandBuffers = current_cmdbuf.address();
submit_info.signalSemaphoreCount = semaphore ? 1 : 0; submit_info.signalSemaphoreCount = semaphore ? 1 : 0;
submit_info.pSignalSemaphores = &semaphore; submit_info.pSignalSemaphores = &semaphore;
device.GetGraphicsQueue().Submit(submit_info, *current_fence); switch (const VkResult result = device.GetGraphicsQueue().Submit(submit_info, *current_fence)) {
case VK_SUCCESS:
break;
case VK_ERROR_DEVICE_LOST:
device.ReportLoss();
[[fallthrough]];
default:
vk::Check(result);
}
} }
void VKScheduler::AllocateNewContext() { void VKScheduler::AllocateNewContext() {

View file

@ -61,7 +61,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkCmdPipelineBarrier); X(vkCmdPipelineBarrier);
X(vkCmdPushConstants); X(vkCmdPushConstants);
X(vkCmdSetBlendConstants); X(vkCmdSetBlendConstants);
X(vkCmdSetCheckpointNV);
X(vkCmdSetDepthBias); X(vkCmdSetDepthBias);
X(vkCmdSetDepthBounds); X(vkCmdSetDepthBounds);
X(vkCmdSetScissor); X(vkCmdSetScissor);
@ -116,7 +115,6 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkGetFenceStatus); X(vkGetFenceStatus);
X(vkGetImageMemoryRequirements); X(vkGetImageMemoryRequirements);
X(vkGetQueryPoolResults); X(vkGetQueryPoolResults);
X(vkGetQueueCheckpointDataNV);
X(vkMapMemory); X(vkMapMemory);
X(vkQueueSubmit); X(vkQueueSubmit);
X(vkResetFences); X(vkResetFences);
@ -409,17 +407,6 @@ DebugCallback Instance::TryCreateDebugCallback(
return DebugCallback(messenger, handle, *dld); return DebugCallback(messenger, handle, *dld);
} }
std::vector<VkCheckpointDataNV> Queue::GetCheckpointDataNV(const DeviceDispatch& dld) const {
if (!dld.vkGetQueueCheckpointDataNV) {
return {};
}
u32 num;
dld.vkGetQueueCheckpointDataNV(queue, &num, nullptr);
std::vector<VkCheckpointDataNV> checkpoints(num);
dld.vkGetQueueCheckpointDataNV(queue, &num, checkpoints.data());
return checkpoints;
}
void Buffer::BindMemory(VkDeviceMemory memory, VkDeviceSize offset) const { void Buffer::BindMemory(VkDeviceMemory memory, VkDeviceSize offset) const {
Check(dld->vkBindBufferMemory(owner, handle, memory, offset)); Check(dld->vkBindBufferMemory(owner, handle, memory, offset));
} }
@ -469,12 +456,11 @@ std::vector<VkImage> SwapchainKHR::GetImages() const {
} }
Device Device::Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci, Device Device::Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
Span<const char*> enabled_extensions, Span<const char*> enabled_extensions, const void* next,
const VkPhysicalDeviceFeatures2& enabled_features,
DeviceDispatch& dld) noexcept { DeviceDispatch& dld) noexcept {
VkDeviceCreateInfo ci; VkDeviceCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO; ci.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
ci.pNext = &enabled_features; ci.pNext = next;
ci.flags = 0; ci.flags = 0;
ci.queueCreateInfoCount = queues_ci.size(); ci.queueCreateInfoCount = queues_ci.size();
ci.pQueueCreateInfos = queues_ci.data(); ci.pQueueCreateInfos = queues_ci.data();

View file

@ -197,7 +197,6 @@ struct DeviceDispatch : public InstanceDispatch {
PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier; PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier;
PFN_vkCmdPushConstants vkCmdPushConstants; PFN_vkCmdPushConstants vkCmdPushConstants;
PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants; PFN_vkCmdSetBlendConstants vkCmdSetBlendConstants;
PFN_vkCmdSetCheckpointNV vkCmdSetCheckpointNV;
PFN_vkCmdSetDepthBias vkCmdSetDepthBias; PFN_vkCmdSetDepthBias vkCmdSetDepthBias;
PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds; PFN_vkCmdSetDepthBounds vkCmdSetDepthBounds;
PFN_vkCmdSetScissor vkCmdSetScissor; PFN_vkCmdSetScissor vkCmdSetScissor;
@ -252,7 +251,6 @@ struct DeviceDispatch : public InstanceDispatch {
PFN_vkGetFenceStatus vkGetFenceStatus; PFN_vkGetFenceStatus vkGetFenceStatus;
PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements; PFN_vkGetImageMemoryRequirements vkGetImageMemoryRequirements;
PFN_vkGetQueryPoolResults vkGetQueryPoolResults; PFN_vkGetQueryPoolResults vkGetQueryPoolResults;
PFN_vkGetQueueCheckpointDataNV vkGetQueueCheckpointDataNV;
PFN_vkMapMemory vkMapMemory; PFN_vkMapMemory vkMapMemory;
PFN_vkQueueSubmit vkQueueSubmit; PFN_vkQueueSubmit vkQueueSubmit;
PFN_vkResetFences vkResetFences; PFN_vkResetFences vkResetFences;
@ -567,12 +565,8 @@ public:
/// Construct a queue handle. /// Construct a queue handle.
constexpr Queue(VkQueue queue, const DeviceDispatch& dld) noexcept : queue{queue}, dld{&dld} {} constexpr Queue(VkQueue queue, const DeviceDispatch& dld) noexcept : queue{queue}, dld{&dld} {}
/// Returns the checkpoint data. VkResult Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const noexcept {
/// @note Returns an empty vector when the function pointer is not present. return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence);
std::vector<VkCheckpointDataNV> GetCheckpointDataNV(const DeviceDispatch& dld) const;
void Submit(Span<VkSubmitInfo> submit_infos, VkFence fence) const {
Check(dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence));
} }
VkResult Present(const VkPresentInfoKHR& present_info) const noexcept { VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
@ -659,8 +653,7 @@ class Device : public Handle<VkDevice, NoOwner, DeviceDispatch> {
public: public:
static Device Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci, static Device Create(VkPhysicalDevice physical_device, Span<VkDeviceQueueCreateInfo> queues_ci,
Span<const char*> enabled_extensions, Span<const char*> enabled_extensions, const void* next,
const VkPhysicalDeviceFeatures2& enabled_features,
DeviceDispatch& dld) noexcept; DeviceDispatch& dld) noexcept;
Queue GetQueue(u32 family_index) const noexcept; Queue GetQueue(u32 family_index) const noexcept;
@ -734,18 +727,11 @@ public:
dld->vkResetQueryPoolEXT(handle, query_pool, first, count); dld->vkResetQueryPoolEXT(handle, query_pool, first, count);
} }
void GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size, VkResult GetQueryResults(VkQueryPool query_pool, u32 first, u32 count, std::size_t data_size,
void* data, VkDeviceSize stride, VkQueryResultFlags flags) const { void* data, VkDeviceSize stride, VkQueryResultFlags flags) const
Check(dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride, noexcept {
flags)); return dld->vkGetQueryPoolResults(handle, query_pool, first, count, data_size, data, stride,
} flags);
template <typename T>
T GetQueryResult(VkQueryPool query_pool, u32 first, VkQueryResultFlags flags) const {
static_assert(std::is_trivially_copyable_v<T>);
T value;
GetQueryResults(query_pool, first, 1, sizeof(T), &value, sizeof(T), flags);
return value;
} }
}; };
@ -920,10 +906,6 @@ public:
dld->vkCmdPushConstants(handle, layout, flags, offset, size, values); dld->vkCmdPushConstants(handle, layout, flags, offset, size, values);
} }
void SetCheckpointNV(const void* checkpoint_marker) const noexcept {
dld->vkCmdSetCheckpointNV(handle, checkpoint_marker);
}
void SetViewport(u32 first, Span<VkViewport> viewports) const noexcept { void SetViewport(u32 first, Span<VkViewport> viewports) const noexcept {
dld->vkCmdSetViewport(handle, first, viewports.size(), viewports.data()); dld->vkCmdSetViewport(handle, first, viewports.size(), viewports.data());
} }