From a05bdb172d198904a76b4b166b43fb31a94a0dd3 Mon Sep 17 00:00:00 2001 From: goeiecool9999 <7033575+goeiecool9999@users.noreply.github.com> Date: Sun, 15 Sep 2024 20:23:11 +0200 Subject: [PATCH] Vulkan: Add explicit synchronization on frame boundaries (#1290) --- .../Latte/Renderer/Vulkan/SwapchainInfoVk.cpp | 34 ++++++++- .../Latte/Renderer/Vulkan/SwapchainInfoVk.h | 8 ++ src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h | 3 + .../Latte/Renderer/Vulkan/VulkanRenderer.cpp | 73 ++++++++++++++++++- .../HW/Latte/Renderer/Vulkan/VulkanRenderer.h | 4 +- 5 files changed, 119 insertions(+), 3 deletions(-) diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.cpp b/src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.cpp index 75ff02ba..56a3ab12 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.cpp +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/SwapchainInfoVk.cpp @@ -146,8 +146,17 @@ void SwapchainInfoVk::Create() UnrecoverableError("Failed to create semaphore for swapchain acquire"); } + VkFenceCreateInfo fenceInfo = {}; + fenceInfo.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; + fenceInfo.flags = VK_FENCE_CREATE_SIGNALED_BIT; + result = vkCreateFence(m_logicalDevice, &fenceInfo, nullptr, &m_imageAvailableFence); + if (result != VK_SUCCESS) + UnrecoverableError("Failed to create fence for swapchain"); + m_acquireIndex = 0; hasDefinedSwapchainImage = false; + + m_queueDepth = 0; } void SwapchainInfoVk::Cleanup() @@ -177,6 +186,12 @@ void SwapchainInfoVk::Cleanup() m_swapchainFramebuffers.clear(); + if (m_imageAvailableFence) + { + WaitAvailableFence(); + vkDestroyFence(m_logicalDevice, m_imageAvailableFence, nullptr); + m_imageAvailableFence = nullptr; + } if (m_swapchain) { vkDestroySwapchainKHR(m_logicalDevice, m_swapchain, nullptr); @@ -189,6 +204,18 @@ bool SwapchainInfoVk::IsValid() const return m_swapchain && !m_acquireSemaphores.empty(); } +void SwapchainInfoVk::WaitAvailableFence() +{ + if(m_awaitableFence != VK_NULL_HANDLE) + vkWaitForFences(m_logicalDevice, 1, &m_awaitableFence, VK_TRUE, UINT64_MAX); + m_awaitableFence = VK_NULL_HANDLE; +} + +void SwapchainInfoVk::ResetAvailableFence() const +{ + vkResetFences(m_logicalDevice, 1, &m_imageAvailableFence); +} + VkSemaphore SwapchainInfoVk::ConsumeAcquireSemaphore() { VkSemaphore ret = m_currentSemaphore; @@ -198,8 +225,10 @@ VkSemaphore SwapchainInfoVk::ConsumeAcquireSemaphore() bool SwapchainInfoVk::AcquireImage() { + ResetAvailableFence(); + VkSemaphore acquireSemaphore = m_acquireSemaphores[m_acquireIndex]; - VkResult result = vkAcquireNextImageKHR(m_logicalDevice, m_swapchain, 1'000'000'000, acquireSemaphore, nullptr, &swapchainImageIndex); + VkResult result = vkAcquireNextImageKHR(m_logicalDevice, m_swapchain, 1'000'000'000, acquireSemaphore, m_imageAvailableFence, &swapchainImageIndex); if (result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR) m_shouldRecreate = true; if (result == VK_TIMEOUT) @@ -216,6 +245,7 @@ bool SwapchainInfoVk::AcquireImage() return false; } m_currentSemaphore = acquireSemaphore; + m_awaitableFence = m_imageAvailableFence; m_acquireIndex = (m_acquireIndex + 1) % m_swapchainImages.size(); return true; @@ -319,6 +349,7 @@ VkExtent2D SwapchainInfoVk::ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& cap VkPresentModeKHR SwapchainInfoVk::ChoosePresentMode(const std::vector& modes) { + m_maxQueued = 0; const auto vsyncState = (VSync)GetConfig().vsync.GetValue(); if (vsyncState == VSync::MAILBOX) { @@ -345,6 +376,7 @@ VkPresentModeKHR SwapchainInfoVk::ChoosePresentMode(const std::vector m_acquireSemaphores; // indexed by m_acquireIndex + VkFence m_imageAvailableFence{}; + VkFence m_awaitableFence = VK_NULL_HANDLE; VkSemaphore m_currentSemaphore = VK_NULL_HANDLE; std::array m_swapchainQueueFamilyIndices; diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h index 0489bb4e..6bde2a0b 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanAPI.h @@ -188,6 +188,9 @@ VKFUNC_DEVICE(vkCmdPipelineBarrier2KHR); VKFUNC_DEVICE(vkCmdBeginRenderingKHR); VKFUNC_DEVICE(vkCmdEndRenderingKHR); +// khr_present_wait +VKFUNC_DEVICE(vkWaitForPresentKHR); + // transform feedback extension VKFUNC_DEVICE(vkCmdBindTransformFeedbackBuffersEXT); VKFUNC_DEVICE(vkCmdBeginTransformFeedbackEXT); diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp index f464c7a3..12d1d975 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.cpp @@ -47,7 +47,9 @@ const std::vector kOptionalDeviceExtensions = VK_EXT_FILTER_CUBIC_EXTENSION_NAME, // not supported by any device yet VK_EXT_EXTERNAL_MEMORY_HOST_EXTENSION_NAME, VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME, - VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME + VK_KHR_SHADER_FLOAT_CONTROLS_EXTENSION_NAME, + VK_KHR_PRESENT_WAIT_EXTENSION_NAME, + VK_KHR_PRESENT_ID_EXTENSION_NAME }; const std::vector kRequiredDeviceExtensions = @@ -252,12 +254,24 @@ void VulkanRenderer::GetDeviceFeatures() pcc.pNext = prevStruct; prevStruct = &pcc; + VkPhysicalDevicePresentIdFeaturesKHR pidf{}; + pidf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_ID_FEATURES_KHR; + pidf.pNext = prevStruct; + prevStruct = &pidf; + + VkPhysicalDevicePresentWaitFeaturesKHR pwf{}; + pwf.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_WAIT_FEATURES_KHR; + pwf.pNext = prevStruct; + prevStruct = &pwf; + VkPhysicalDeviceFeatures2 physicalDeviceFeatures2{}; physicalDeviceFeatures2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2; physicalDeviceFeatures2.pNext = prevStruct; vkGetPhysicalDeviceFeatures2(m_physicalDevice, &physicalDeviceFeatures2); + cemuLog_log(LogType::Force, "Vulkan: present_wait extension: {}", (pwf.presentWait && pidf.presentId) ? "supported" : "unsupported"); + /* Get Vulkan device properties and limits */ VkPhysicalDeviceFloatControlsPropertiesKHR pfcp{}; prevStruct = nullptr; @@ -490,6 +504,24 @@ VulkanRenderer::VulkanRenderer() customBorderColorFeature.customBorderColors = VK_TRUE; customBorderColorFeature.customBorderColorWithoutFormat = VK_TRUE; } + // enable VK_KHR_present_id + VkPhysicalDevicePresentIdFeaturesKHR presentIdFeature{}; + if(m_featureControl.deviceExtensions.present_wait) + { + presentIdFeature.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_ID_FEATURES_KHR; + presentIdFeature.pNext = deviceExtensionFeatures; + deviceExtensionFeatures = &presentIdFeature; + presentIdFeature.presentId = VK_TRUE; + } + // enable VK_KHR_present_wait + VkPhysicalDevicePresentWaitFeaturesKHR presentWaitFeature{}; + if(m_featureControl.deviceExtensions.present_wait) + { + presentWaitFeature.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRESENT_WAIT_FEATURES_KHR; + presentWaitFeature.pNext = deviceExtensionFeatures; + deviceExtensionFeatures = &presentWaitFeature; + presentWaitFeature.presentWait = VK_TRUE; + } std::vector used_extensions; VkDeviceCreateInfo createInfo = CreateDeviceCreateInfo(queueCreateInfos, deviceFeatures, deviceExtensionFeatures, used_extensions); @@ -1047,6 +1079,10 @@ VkDeviceCreateInfo VulkanRenderer::CreateDeviceCreateInfo(const std::vector 0); + // wait for the previous frame to finish rendering + WaitCommandBufferFinished(m_commandBufferIDOfPrevFrame); + m_commandBufferIDOfPrevFrame = currentFrameCmdBufferID; + + chainInfo.WaitAvailableFence(); + + VkPresentIdKHR presentId = {}; + VkPresentInfoKHR presentInfo = {}; presentInfo.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR; presentInfo.swapchainCount = 1; @@ -2709,6 +2756,24 @@ void VulkanRenderer::SwapBuffer(bool mainWindow) presentInfo.waitSemaphoreCount = 1; presentInfo.pWaitSemaphores = &presentSemaphore; + // if present_wait is available and enabled, add frame markers to present requests + // and limit the number of queued present operations + if (m_featureControl.deviceExtensions.present_wait && chainInfo.m_maxQueued > 0) + { + presentId.sType = VK_STRUCTURE_TYPE_PRESENT_ID_KHR; + presentId.swapchainCount = 1; + presentId.pPresentIds = &chainInfo.m_presentId; + + presentInfo.pNext = &presentId; + + if(chainInfo.m_queueDepth >= chainInfo.m_maxQueued) + { + uint64 waitFrameId = chainInfo.m_presentId - chainInfo.m_queueDepth; + vkWaitForPresentKHR(m_logicalDevice, chainInfo.m_swapchain, waitFrameId, 40'000'000); + chainInfo.m_queueDepth--; + } + } + VkResult result = vkQueuePresentKHR(m_presentQueue, &presentInfo); if (result < 0 && result != VK_ERROR_OUT_OF_DATE_KHR) { @@ -2717,6 +2782,12 @@ void VulkanRenderer::SwapBuffer(bool mainWindow) if(result == VK_ERROR_OUT_OF_DATE_KHR || result == VK_SUBOPTIMAL_KHR) chainInfo.m_shouldRecreate = true; + if(result >= 0) + { + chainInfo.m_queueDepth++; + chainInfo.m_presentId++; + } + chainInfo.hasDefinedSwapchainImage = false; chainInfo.swapchainImageIndex = -1; diff --git a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h index 6df53da4..ce97b5e9 100644 --- a/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h +++ b/src/Cafe/HW/Latte/Renderer/Vulkan/VulkanRenderer.h @@ -450,6 +450,7 @@ private: bool synchronization2 = false; // VK_KHR_synchronization2 bool dynamic_rendering = false; // VK_KHR_dynamic_rendering bool shader_float_controls = false; // VK_KHR_shader_float_controls + bool present_wait = false; // VK_KHR_present_wait }deviceExtensions; struct @@ -457,7 +458,7 @@ private: bool shaderRoundingModeRTEFloat32{ false }; }shaderFloatControls; // from VK_KHR_shader_float_controls - struct + struct { bool debug_utils = false; // VK_EXT_DEBUG_UTILS }instanceExtensions; @@ -635,6 +636,7 @@ private: size_t m_commandBufferIndex = 0; // current buffer being filled size_t m_commandBufferSyncIndex = 0; // latest buffer that finished execution (updated on submit) + size_t m_commandBufferIDOfPrevFrame = 0; std::array m_cmd_buffer_fences; std::array m_commandBuffers; std::array m_commandBufferSemaphores;