Skip to content

A
angle
Commit 6165602f by Shahbaz Youssefi Committed by Commit Bot
Options

Vulkan: Throttle CPU to avoid excessive frame queuing

Unthrottled, the CPU can generate and queue an increasingly large number of frames with the GPU lagging behind, especially with vsync enabled. Assuming N swapchain images, this commit adds a wait on fence for the Nth previous frame before submitting new work, that is the CPU is always at most N frames ahead of the GPU. Bug: angleproject:2908 Change-Id: Ieb2bf20168bfe9bc9d8e2219f682b01347c21dec Reviewed-on: https://chromium-review.googlesource.com/c/1296953 Commit-Queue: Shahbaz Youssefi <syoussefi@chromium.org> Reviewed-by: 's avatarYuly Novikov <ynovikov@chromium.org>
parent 79927344
Showing with 91 additions and 39 deletions
src/libANGLE/renderer/vulkan/RendererVk.cpp
......@@ -28,13 +28,15 @@
#include "libANGLE/renderer/vulkan/vk_format_utils.h"
#include "platform/Platform.h"
#include "third_party/trace_event/trace_event.h"
// Consts
namespace
{
const uint32_t kMockVendorID = 0xba5eba11;
const uint32_t kMockDeviceID = 0xf005ba11;
constexpr char kMockDeviceName[] = "Vulkan Mock Device";
constexpr size_t kInFlightCommandsLimit = 50000u;
constexpr size_t kInFlightCommandsLimit = 100u;
} // anonymous namespace
namespace rx
......@@ -838,6 +840,8 @@ angle::Result RendererVk::finish(vk::Context *context)
{
if (!mCommandGraph.empty())
{
TRACE_EVENT0("gpu.angle", "RendererVk::finish");
vk::Scoped<vk::CommandBuffer> commandBatch(mDevice);
ANGLE_TRY(flushCommandGraph(context, &commandBatch.get()));
......@@ -881,6 +885,8 @@ void RendererVk::freeAllInFlightResources()
garbage.destroy(mDevice);
}
mGarbage.clear();
mLastCompletedQueueSerial = mLastSubmittedQueueSerial;
}
angle::Result RendererVk::checkCompletedCommands(vk::Context *context)
......@@ -940,15 +946,14 @@ angle::Result RendererVk::submitFrame(vk::Context *context,
mInFlightCommands.emplace_back(scopedBatch.release());
// Check that mInFlightCommands isn't growing too fast
// If it is, wait for the queue to complete work it has alread been assigned
if (mInFlightCommands.size() > kInFlightCommandsLimit)
{
vkQueueWaitIdle(mQueue);
}
// CPU should be throttled to avoid mInFlightCommands from growing too fast. That is done on
// swap() though, and there could be multiple submissions in between (through glFlush() calls),
// so the limit is larger than the expected number of images.
ASSERT(mInFlightCommands.size() <= kInFlightCommandsLimit);
// Increment the queue serial. If this fails, we should restart ANGLE.
// TODO(jmadill): Overflow check.
mLastSubmittedQueueSerial = mCurrentQueueSerial;
mCurrentQueueSerial = mQueueSerialFactory.generate();
ANGLE_TRY(checkCompletedCommands(context));
......@@ -1042,6 +1047,8 @@ angle::Result RendererVk::flush(vk::Context *context)
return angle::Result::Continue();
}
TRACE_EVENT0("gpu.angle", "RendererVk::flush");
vk::Scoped<vk::CommandBuffer> commandBatch(mDevice);
ANGLE_TRY(flushCommandGraph(context, &commandBatch.get()));
......
src/libANGLE/renderer/vulkan/RendererVk.h
......@@ -84,6 +84,7 @@ class RendererVk : angle::NonCopyable
uint32_t getMaxActiveTextures();
Serial getCurrentQueueSerial() const { return mCurrentQueueSerial; }
Serial getLastSubmittedQueueSerial() const { return mLastSubmittedQueueSerial; }
Serial getLastCompletedQueueSerial() const { return mLastCompletedQueueSerial; }
bool isSerialInUse(Serial serial) const;
......@@ -214,6 +215,7 @@ class RendererVk : angle::NonCopyable
SerialFactory mQueueSerialFactory;
SerialFactory mShaderSerialFactory;
Serial mLastCompletedQueueSerial;
Serial mLastSubmittedQueueSerial;
Serial mCurrentQueueSerial;
bool mDeviceLost;
......
src/libANGLE/renderer/vulkan/SurfaceVk.cpp
......@@ -18,6 +18,7 @@
#include "libANGLE/renderer/vulkan/FramebufferVk.h"
#include "libANGLE/renderer/vulkan/RendererVk.h"
#include "libANGLE/renderer/vulkan/vk_format_utils.h"
#include "third_party/trace_event/trace_event.h"
namespace rx
{
......@@ -31,22 +32,37 @@ VkPresentModeKHR GetDesiredPresentMode(const std::vector<VkPresentModeKHR> &pres
{
ASSERT(!presentModes.empty());
// Use FIFO mode for v-sync, since it throttles you to the display rate. Mailbox is more
// similar to triple-buffering. For now we hard-code Mailbox for perf testing.
// Use FIFO mode for v-sync, since it throttles you to the display rate.
//
// However, for performance testing (for now), we want to issue draws as fast as possible so we
// use either of the following, if available, in order specified here:
//
// - Mailbox is similar to triple-buffering.
// - Immediate is similar to single-buffering.
//
// TODO(jmadill): Properly select present mode and re-create display if changed.
VkPresentModeKHR bestChoice = VK_PRESENT_MODE_MAILBOX_KHR;
bool mailboxAvailable = false;
bool immediateAvailable = false;
for (VkPresentModeKHR presentMode : presentModes)
{
if (presentMode == bestChoice)
switch (presentMode)
{
return bestChoice;
case VK_PRESENT_MODE_MAILBOX_KHR:
mailboxAvailable = true;
break;
case VK_PRESENT_MODE_IMMEDIATE_KHR:
immediateAvailable = true;
break;
default:
break;
}
}
WARN() << "Present mode " << bestChoice << " not available. Falling back to "
<< presentModes[0];
return presentModes[0];
// Note that VK_PRESENT_MODE_FIFO_KHR is guaranteed to be available.
return mailboxAvailable
? VK_PRESENT_MODE_MAILBOX_KHR
: immediateAvailable ? VK_PRESENT_MODE_IMMEDIATE_KHR : VK_PRESENT_MODE_FIFO_KHR;
}
constexpr VkImageUsageFlags kSurfaceVKImageUsageFlags =
......@@ -261,9 +277,11 @@ WindowSurfaceVk::WindowSurfaceVk(const egl::SurfaceState &surfaceState,
mSurface(VK_NULL_HANDLE),
mInstance(VK_NULL_HANDLE),
mSwapchain(VK_NULL_HANDLE),
mSwapchainPresentMode(VK_PRESENT_MODE_FIFO_KHR),
mColorRenderTarget(nullptr, nullptr, 0),
mDepthStencilRenderTarget(&mDepthStencilImage, &mDepthStencilImageView, 0),
mCurrentSwapchainImageIndex(0)
mCurrentSwapchainImageIndex(0),
mCurrentSwapSerialIndex(0)
{
}
......@@ -376,8 +394,7 @@ angle::Result WindowSurfaceVk::initializeImpl(DisplayVk *displayVk)
ASSERT(minSwapInterval == 0 || minSwapInterval == 1);
ASSERT(maxSwapInterval == 0 || maxSwapInterval == 1);
VkPresentModeKHR swapchainPresentMode =
GetDesiredPresentMode(presentModes, minSwapInterval, maxSwapInterval);
mSwapchainPresentMode = GetDesiredPresentMode(presentModes, minSwapInterval, maxSwapInterval);
// Determine number of swapchain images. Aim for one more than the minimum.
uint32_t minImageCount = surfaceCaps.minImageCount + 1;
......@@ -436,24 +453,24 @@ angle::Result WindowSurfaceVk::initializeImpl(DisplayVk *displayVk)
VkImageUsageFlags imageUsageFlags = kSurfaceVKColorImageUsageFlags;
VkSwapchainCreateInfoKHR swapchainInfo = {};
swapchainInfo.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR;
swapchainInfo.flags = 0;
swapchainInfo.surface = mSurface;
swapchainInfo.minImageCount = minImageCount;
swapchainInfo.imageFormat = nativeFormat;
swapchainInfo.imageColorSpace = VK_COLORSPACE_SRGB_NONLINEAR_KHR;
swapchainInfo.imageExtent.width = width;
swapchainInfo.imageExtent.height = height;
swapchainInfo.imageArrayLayers = 1;
swapchainInfo.imageUsage = imageUsageFlags;
swapchainInfo.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
swapchainInfo.queueFamilyIndexCount = 0;
swapchainInfo.pQueueFamilyIndices = nullptr;
swapchainInfo.preTransform = preTransform;
swapchainInfo.compositeAlpha = compositeAlpha;
swapchainInfo.presentMode = swapchainPresentMode;
swapchainInfo.clipped = VK_TRUE;
swapchainInfo.oldSwapchain = VK_NULL_HANDLE;
swapchainInfo.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR;
swapchainInfo.flags = 0;
swapchainInfo.surface = mSurface;
swapchainInfo.minImageCount = minImageCount;
swapchainInfo.imageFormat = nativeFormat;
swapchainInfo.imageColorSpace = VK_COLORSPACE_SRGB_NONLINEAR_KHR;
swapchainInfo.imageExtent.width = width;
swapchainInfo.imageExtent.height = height;
swapchainInfo.imageArrayLayers = 1;
swapchainInfo.imageUsage = imageUsageFlags;
swapchainInfo.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
swapchainInfo.queueFamilyIndexCount = 0;
swapchainInfo.pQueueFamilyIndices = nullptr;
swapchainInfo.preTransform = preTransform;
swapchainInfo.compositeAlpha = compositeAlpha;
swapchainInfo.presentMode = mSwapchainPresentMode;
swapchainInfo.clipped = VK_TRUE;
swapchainInfo.oldSwapchain = VK_NULL_HANDLE;
VkDevice device = renderer->getDevice();
ANGLE_VK_TRY(displayVk, vkCreateSwapchainKHR(device, &swapchainInfo, nullptr, &mSwapchain));
......@@ -473,6 +490,7 @@ angle::Result WindowSurfaceVk::initializeImpl(DisplayVk *displayVk)
transparentBlack.float32[3] = 0.0f;
mSwapchainImages.resize(imageCount);
mSwapSerials.resize(imageCount);
for (uint32_t imageIndex = 0; imageIndex < imageCount; ++imageIndex)
{
......@@ -549,6 +567,14 @@ angle::Result WindowSurfaceVk::swapImpl(DisplayVk *displayVk)
{
RendererVk *renderer = displayVk->getRenderer();
// If the swapchain is not in mailbox mode, throttle the submissions. NOTE(syoussefi): this can
// be done in mailbox mode too, just currently unnecessary.
if (mSwapchainPresentMode != VK_PRESENT_MODE_MAILBOX_KHR)
{
TRACE_EVENT0("gpu.angle", "WindowSurfaceVk::swapImpl: Throttle CPU");
ANGLE_TRY(renderer->finishToSerial(displayVk, mSwapSerials[mCurrentSwapSerialIndex]));
}
vk::CommandBuffer *swapCommands = nullptr;
ANGLE_TRY(mSwapchainImages[mCurrentSwapchainImageIndex].image.recordCommands(displayVk,
&swapCommands));
......@@ -561,7 +587,12 @@ angle::Result WindowSurfaceVk::swapImpl(DisplayVk *displayVk)
ANGLE_TRY(renderer->flush(displayVk));
// Ask the renderer what semaphore it signalled in the last flush.
// Remember the serial of the last submission.
mSwapSerials[mCurrentSwapSerialIndex++] = renderer->getLastSubmittedQueueSerial();
mCurrentSwapSerialIndex =
mCurrentSwapSerialIndex == mSwapSerials.size() ? 0 : mCurrentSwapSerialIndex;
// Ask the renderer what semaphore it signaled in the last flush.
const vk::Semaphore *commandsCompleteSemaphore =
renderer->getSubmitLastSignaledSemaphore(displayVk);
......@@ -577,8 +608,13 @@ angle::Result WindowSurfaceVk::swapImpl(DisplayVk *displayVk)
ANGLE_VK_TRY(displayVk, vkQueuePresentKHR(renderer->getQueue(), &presentInfo));
// Get the next available swapchain image.
ANGLE_TRY(nextSwapchainImage(displayVk));
{
// Note: TRACE_EVENT0 is put here instead of inside the function to workaround this issue:
// http://anglebug.com/2927
TRACE_EVENT0("gpu.angle", "nextSwapchainImage");
// Get the next available swapchain image.
ANGLE_TRY(nextSwapchainImage(displayVk));
}
ANGLE_TRY(renderer->syncPipelineCacheVk(displayVk));
......
src/libANGLE/renderer/vulkan/SurfaceVk.h
......@@ -146,6 +146,7 @@ class WindowSurfaceVk : public SurfaceImpl
angle::Result swapImpl(DisplayVk *displayVk);
VkSwapchainKHR mSwapchain;
VkPresentModeKHR mSwapchainPresentMode;
RenderTargetVk mColorRenderTarget;
RenderTargetVk mDepthStencilRenderTarget;
......@@ -165,6 +166,12 @@ class WindowSurfaceVk : public SurfaceImpl
std::vector<SwapchainImage> mSwapchainImages;
// A circular buffer, with the same size as mSwapchainImages (N), that stores the serial of the
// renderer on every swap. In FIFO present modes, the CPU is throttled by waiting for the
// Nth previous serial to finish.
std::vector<Serial> mSwapSerials;
size_t mCurrentSwapSerialIndex;
vk::ImageHelper mDepthStencilImage;
vk::ImageView mDepthStencilImageView;
};
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment