scl/strato
1
0
Fork 0
mirror of https://github.com/Takiiiiiiii/strato.git synced 2025-07-17 08:46:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use semaphores for presentation engine frame synchronisation

Browse Source 
Avoids waits on the CPU which can be costly and confuse the scheduler, also reduces latency significantly.
This commit is contained in:
Billy Laws
2022-10-16 20:47:17 +01:00
parent 0670e0e0dc
commit 1a0819fb76
4 changed files with 106 additions and 87 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
app/src/main/cpp/skyline/gpu/presentation_engine.cpp
View File

@ -24,7 +24,8 @@ namespace skyline::gpu {
PresentationEngine::PresentationEngine(const DeviceState &state, GPU &gpu)
: state{state},
gpu{gpu},
acquireFence{gpu.vkDevice, vk::FenceCreateInfo{}},
presentSemaphores{util::MakeFilledArray<vk::raii::Semaphore, MaxSwapchainImageCount>(gpu.vkDevice, vk::SemaphoreCreateInfo{})},
acquireSemaphores{util::MakeFilledArray<vk::raii::Semaphore, MaxSwapchainImageCount>(gpu.vkDevice, vk::SemaphoreCreateInfo{})},
presentationTrack{static_cast<u64>(trace::TrackIds::Presentation), perfetto::ProcessTrack::Current()},
vsyncEvent{std::make_shared<kernel::type::KEvent>(state, true)},
choreographerThread{&PresentationEngine::ChoreographerThread, this},
@ -116,35 +117,31 @@ namespace skyline::gpu {
windowScalingMode = frame.scalingMode;
}
if (frame.transform != windowTransform) {
if ((result = window->perform(window, NATIVE_WINDOW_SET_BUFFERS_TRANSFORM, static_cast<i32>(frame.transform))))
throw exception("Setting the buffer transform to '{}' failed with {}", ToString(frame.transform), result);
windowTransform = frame.transform;
}
if ((result = window->perform(window, NATIVE_WINDOW_SET_BUFFERS_TRANSFORM, static_cast<i32>(frame.transform))))
throw exception("Setting the buffer transform to '{}' failed with {}", ToString(frame.transform), result);
windowTransform = frame.transform;
gpu.vkDevice.resetFences(*acquireFence);
auto &acquireSemaphore{acquireSemaphores[acquireSemaphoreIndex]};
acquireSemaphoreIndex = (acquireSemaphoreIndex + 1) % swapchainImageCount;
std::pair<vk::Result, u32> nextImage;
while (nextImage = vkSwapchain->acquireNextImage(std::numeric_limits<u64>::max(), {}, *acquireFence), nextImage.first != vk::Result::eSuccess) [[unlikely]] {
while (nextImage = vkSwapchain->acquireNextImage(std::numeric_limits<u64>::max(), *acquireSemaphore, {}), nextImage.first != vk::Result::eSuccess) [[unlikely]] {
if (nextImage.first == vk::Result::eSuboptimalKHR)
surfaceCondition.wait(lock, [this]() { return vkSurface.has_value(); });
else
throw exception("vkAcquireNextImageKHR returned an unhandled result '{}'", vk::to_string(nextImage.first));
}
auto &nextImageTexture{images.at(nextImage.second)};
std::ignore = gpu.vkDevice.waitForFences(*acquireFence, true, std::numeric_limits<u64>::max());
auto &nextImageTexture{images.at(nextImage.second)};
auto &presentSemaphore{presentSemaphores[nextImage.second]};
texture->SynchronizeHost();
nextImageTexture->CopyFrom(texture, vk::ImageSubresourceRange{
nextImageTexture->CopyFrom(texture, *acquireSemaphore, *presentSemaphore, vk::ImageSubresourceRange{
.aspectMask = vk::ImageAspectFlagBits::eColor,
.levelCount = 1,
.layerCount = 1,
});
// Wait on the copy to the swapchain image to complete before submitting for presentation
nextImageTexture->WaitOnFence();
auto getMonotonicNsNow{[]() -> i64 {
timespec time;
if (clock_gettime(CLOCK_MONOTONIC, &time))
@ -194,6 +191,8 @@ namespace skyline::gpu {
.swapchainCount = 1,
.pSwapchains = &**vkSwapchain,
.pImageIndices = &nextImage.second,
.waitSemaphoreCount = 1,
.pWaitSemaphores = &*presentSemaphore,
}); // We don't care about suboptimal images as they are caused by not respecting the transform hint, we handle transformations externally
}
@ -328,6 +327,7 @@ namespace skyline::gpu {
swapchainFormat = format;
swapchainExtent = extent;
swapchainImageCount = vkImages.size();
}
void PresentationEngine::UpdateSurface(jobject newSurface) {