Emulator-Archive/ppsspp
1
0
Fork 0
mirror of https://github.com/hrydgard/ppsspp.git synced 2024-11-26 23:10:38 +00:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

Vulkan profiler: Show CPU command buffer recording timing even if GPU timing is unavailable

Browse Source
This commit is contained in:
Henrik Rydgård 2023-06-14 00:19:58 +02:00
parent fbd10e4722
commit ae29fd2951
6 changed files with 20 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
Common/GPU/Vulkan/VulkanFrameData.cpp
View File

@ -121,7 +121,7 @@ VkCommandBuffer FrameData::GetInitCmd(VulkanContext *vulkan) {
}
// Good spot to reset the query pool.
if (profilingEnabled_) {
if (profile.enabled) {
vkCmdResetQueryPool(initCmd, profile.queryPool, 0, MAX_TIMESTAMP_QUERIES);
vkCmdWriteTimestamp(initCmd, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, profile.queryPool, 0);
}
@ -138,7 +138,7 @@ void FrameData::SubmitPending(VulkanContext *vulkan, FrameSubmitType type, Frame
VkFence fenceToTrigger = VK_NULL_HANDLE;
if (hasInitCommands) {
if (profilingEnabled_) {
if (profile.enabled) {
// Pre-allocated query ID 1 - end of init cmdbuf.
vkCmdWriteTimestamp(initCmd, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, profile.queryPool, 1);
}

6
Common/GPU/Vulkan/VulkanFrameData.h
View File

@ -19,11 +19,14 @@ enum class VKRRunType {
};
struct QueueProfileContext {
bool enabled = false;
bool timestampsEnabled = false;
VkQueryPool queryPool;
std::vector<std::string> timestampDescriptions;
std::string profileSummary;
double cpuStartTime;
double cpuEndTime;
double descWriteTime;
};
class VKRFramebuffer;
@ -92,8 +95,7 @@ struct FrameData {
uint32_t curSwapchainImage = -1;
// Profiling.
QueueProfileContext profile;
bool profilingEnabled_ = false;
QueueProfileContext profile{};
// Async readback cache.
DenseHashMap<ReadbackKey, CachedReadback*, nullptr> readbacks_;

4
Common/GPU/Vulkan/VulkanQueueRunner.cpp
View File

@ -339,7 +339,7 @@ void VulkanQueueRunner::PreprocessSteps(std::vector<VKRStep *> &steps) {
}
void VulkanQueueRunner::RunSteps(std::vector<VKRStep *> &steps, FrameData &frameData, FrameDataShared &frameDataShared, bool keepSteps) {
QueueProfileContext *profile = frameData.profilingEnabled_ ? &frameData.profile : nullptr;
QueueProfileContext *profile = frameData.profile.enabled ? &frameData.profile : nullptr;
if (profile)
profile->cpuStartTime = time_now_d();
@ -412,7 +412,7 @@ void VulkanQueueRunner::RunSteps(std::vector<VKRStep *> &steps, FrameData &frame
break;
}
if (profile && profile->timestampDescriptions.size() + 1 < MAX_TIMESTAMP_QUERIES) {
if (profile && profile->timestampsEnabled && profile->timestampDescriptions.size() + 1 < MAX_TIMESTAMP_QUERIES) {
vkCmdWriteTimestamp(cmd, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, profile->queryPool, (uint32_t)profile->timestampDescriptions.size());
profile->timestampDescriptions.push_back(StepToString(step));
}

16
Common/GPU/Vulkan/VulkanRenderManager.cpp
View File

@ -552,13 +552,14 @@ void VulkanRenderManager::BeginFrame(bool enableProfiling, bool enableLogProfile
int validBits = vulkan_->GetQueueFamilyProperties(vulkan_->GetGraphicsQueueFamilyIndex()).timestampValidBits;
// Can't set this until after the fence.
frameData.profilingEnabled_ = enableProfiling && validBits > 0;
frameData.profile.enabled = enableProfiling;
frameData.profile.timestampsEnabled = enableProfiling && validBits > 0;
uint64_t queryResults[MAX_TIMESTAMP_QUERIES];
if (frameData.profilingEnabled_) {
if (enableProfiling) {
// Pull the profiling results from last time and produce a summary!
if (!frameData.profile.timestampDescriptions.empty()) {
if (!frameData.profile.timestampDescriptions.empty() && frameData.profile.timestampsEnabled) {
int numQueries = (int)frameData.profile.timestampDescriptions.size();
VkResult res = vkGetQueryPoolResults(
vulkan_->GetDevice(),
@ -596,7 +597,12 @@ void VulkanRenderManager::BeginFrame(bool enableProfiling, bool enableLogProfile
frameData.profile.profileSummary = "(error getting GPU profile - not ready?)";
}
} else {
frameData.profile.profileSummary = "(no GPU profile data collected)";
std::stringstream str;
char line[256];
renderCPUTimeMs_.Update((frameData.profile.cpuEndTime - frameData.profile.cpuStartTime) * 1000.0);
renderCPUTimeMs_.Format(line, sizeof(line));
str << line;
frameData.profile.profileSummary = str.str();
}
}
@ -607,7 +613,7 @@ void VulkanRenderManager::BeginFrame(bool enableProfiling, bool enableLogProfile
vulkan_->BeginFrame(enableLogProfiler ? GetInitCmd() : VK_NULL_HANDLE);
frameData.profile.timestampDescriptions.clear();
if (frameData.profilingEnabled_) {
if (frameData.profile.timestampsEnabled) {
// For various reasons, we need to always use an init cmd buffer in this case to perform the vkCmdResetQueryPool,
// unless we want to limit ourselves to only measure the main cmd buffer.
// Later versions of Vulkan have support for clearing queries on the CPU timeline, but we don't want to rely on that.

2
Common/GPU/Vulkan/VulkanRenderManager.h
View File

@ -453,8 +453,6 @@ public:
return outOfDateFrames_ > VulkanContext::MAX_INFLIGHT_FRAMES;
}
void Invalidate(InvalidationFlags flags);
void ResetStats();
void DrainCompileQueue();

1
Common/Math/Statistics.h
View File

@ -32,6 +32,7 @@ struct SimpleStat {
void Format(char *buffer, size_t sz);
private:
SimpleStat() {}
const char *name_;
// These are initialized in Reset().