debuggerx/eden
1
0
Fork 0
mirror of https://git.eden-emu.dev/eden-emu/eden.git synced 2026-08-22 15:48:56 +00:00
Code Issues Packages Projects Releases Wiki Activity

[vulkan] Initial implementation Synchronization2

Browse Source
This commit is contained in:
CamilleLaVey
2026-07-03 00:58:40 -04:00
parent 4c65780f11
commit 5af01bef9f
5 changed files with 225 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/video_core/renderer_vulkan/vk_master_semaphore.cpp
+124
View File
@@ -130,6 +130,70 @@ VkResult MasterSemaphore::SubmitQueueTimeline(vk::CommandBuffer& cmdbuf,
VkSemaphore wait_semaphore, u64 host_tick) {
const VkSemaphore timeline_semaphore = *semaphore;
if (device.HasSynchronization2()) {
const std::array<VkCommandBufferSubmitInfo, 2> cmdbuffer_infos{{
{
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO,
.pNext = nullptr,
.commandBuffer = *upload_cmdbuf,
.deviceMask = 0,
},
{
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO,
.pNext = nullptr,
.commandBuffer = *cmdbuf,
.deviceMask = 0,
},
}};
std::array<VkSemaphoreSubmitInfo, 2> signal_infos{{
{
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO,
.pNext = nullptr,
.semaphore = timeline_semaphore,
.value = host_tick,
.stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.deviceIndex = 0,
},
{},
}};
u32 num_signal_semaphores = 1;
if (signal_semaphore) {
signal_infos[1] = VkSemaphoreSubmitInfo{
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO,
.pNext = nullptr,
.semaphore = signal_semaphore,
.value = 0,
.stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.deviceIndex = 0,
};
num_signal_semaphores = 2;
}
const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
const VkSemaphoreSubmitInfo wait_info{
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO,
.pNext = nullptr,
.semaphore = wait_semaphore,
.value = 0,
.stageMask = static_cast<VkPipelineStageFlags2>(wait_stage_mask),
.deviceIndex = 0,
};
const VkSubmitInfo2 submit_info2{
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2,
.pNext = nullptr,
.flags = 0,
.waitSemaphoreInfoCount = num_wait_semaphores,
.pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr,
.commandBufferInfoCount = static_cast<u32>(cmdbuffer_infos.size()),
.pCommandBufferInfos = cmdbuffer_infos.data(),
.signalSemaphoreInfoCount = num_signal_semaphores,
.pSignalSemaphoreInfos = signal_infos.data(),
};
return device.GetGraphicsQueue().Submit2(submit_info2);
}
const u32 num_signal_semaphores = signal_semaphore ? 2 : 1;
const std::array signal_values{host_tick, u64(0)};
const std::array signal_semaphores{timeline_semaphore, signal_semaphore};
@@ -172,6 +236,66 @@ VkResult MasterSemaphore::SubmitQueueFence(vk::CommandBuffer& cmdbuf,
vk::CommandBuffer& upload_cmdbuf,
VkSemaphore signal_semaphore, VkSemaphore wait_semaphore,
u64 host_tick) {
if (device.HasSynchronization2()) {
const std::array<VkCommandBufferSubmitInfo, 2> cmdbuffer_infos{{
{
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO,
.pNext = nullptr,
.commandBuffer = *upload_cmdbuf,
.deviceMask = 0,
},
{
.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_SUBMIT_INFO,
.pNext = nullptr,
.commandBuffer = *cmdbuf,
.deviceMask = 0,
},
}};
const u32 num_signal_semaphores = signal_semaphore ? 1 : 0;
const VkSemaphoreSubmitInfo signal_info{
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO,
.pNext = nullptr,
.semaphore = signal_semaphore,
.value = 0,
.stageMask = VK_PIPELINE_STAGE_2_ALL_COMMANDS_BIT,
.deviceIndex = 0,
};
const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
const VkSemaphoreSubmitInfo wait_info{
.sType = VK_STRUCTURE_TYPE_SEMAPHORE_SUBMIT_INFO,
.pNext = nullptr,
.semaphore = wait_semaphore,
.value = 0,
.stageMask = static_cast<VkPipelineStageFlags2>(wait_stage_mask),
.deviceIndex = 0,
};
const VkSubmitInfo2 submit_info2{
.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO_2,
.pNext = nullptr,
.flags = 0,
.waitSemaphoreInfoCount = num_wait_semaphores,
.pWaitSemaphoreInfos = num_wait_semaphores ? &wait_info : nullptr,
.commandBufferInfoCount = static_cast<u32>(cmdbuffer_infos.size()),
.pCommandBufferInfos = cmdbuffer_infos.data(),
.signalSemaphoreInfoCount = num_signal_semaphores,
.pSignalSemaphoreInfos = num_signal_semaphores ? &signal_info : nullptr,
};
auto fence = GetFreeFence();
auto result = device.GetGraphicsQueue().Submit2(submit_info2, *fence);
if (result == VK_SUCCESS) {
std::scoped_lock lock{wait_mutex};
wait_queue.emplace(host_tick, std::move(fence));
wait_cv.notify_one();
}
return result;
}
const u32 num_signal_semaphores = signal_semaphore ? 1 : 0;
const u32 num_wait_semaphores = wait_semaphore ? 1 : 0;
src/video_core/vulkan_common/vulkan_device.cpp
+5
View File
@@ -1375,6 +1375,11 @@ void Device::RemoveUnsuitableExtensions() {
// VK_KHR_maintenance8
extensions.maintenance8 = loaded_extensions.contains(VK_KHR_MAINTENANCE_8_EXTENSION_NAME);
RemoveExtensionIfUnsuitable(extensions.maintenance8, VK_KHR_MAINTENANCE_8_EXTENSION_NAME);
// VK_KHR_synchronization2
extensions.synchronization2 = features.synchronization2.synchronization2;
RemoveExtensionFeatureIfUnsuitable(extensions.synchronization2, features.synchronization2,
VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME);
}
void Device::SetupFamilies(VkSurfaceKHR surface) {
src/video_core/vulkan_common/vulkan_device.h
+8 -1
View File
@@ -43,7 +43,8 @@ VK_DEFINE_HANDLE(VmaAllocator)
FEATURE(EXT, ShaderDemoteToHelperInvocation, SHADER_DEMOTE_TO_HELPER_INVOCATION, \
shader_demote_to_helper_invocation) \
FEATURE(EXT, SubgroupSizeControl, SUBGROUP_SIZE_CONTROL, subgroup_size_control) \
FEATURE(KHR, Maintenance4, MAINTENANCE_4, maintenance4)
FEATURE(KHR, Maintenance4, MAINTENANCE_4, maintenance4) \
FEATURE(KHR, Synchronization2, SYNCHRONIZATION_2, synchronization2)
#define FOR_EACH_VK_FEATURE_1_4(FEATURE)
@@ -180,6 +181,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
FEATURE_NAME(robustness2, nullDescriptor) \
FEATURE_NAME(shader_float16_int8, shaderFloat16) \
FEATURE_NAME(shader_float16_int8, shaderInt8) \
FEATURE_NAME(synchronization2, synchronization2) \
FEATURE_NAME(timeline_semaphore, timelineSemaphore) \
FEATURE_NAME(transform_feedback, transformFeedback) \
FEATURE_NAME(uniform_buffer_standard_layout, uniformBufferStandardLayout) \
@@ -722,6 +724,11 @@ FN_MAX_LIMIT_LIST
bool HasTimelineSemaphore() const;
/// Returns true if the device supports VK_KHR_synchronization2.
bool HasSynchronization2() const {
return extensions.synchronization2;
}
/// Returns the minimum supported version of SPIR-V.
u32 SupportedSpirvVersion() const {
if (instance_version >= VK_API_VERSION_1_3) {
src/video_core/vulkan_common/vulkan_wrapper.cpp
+10
View File
@@ -123,6 +123,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkCmdEndDebugUtilsLabelEXT);
X(vkCmdFillBuffer);
X(vkCmdPipelineBarrier);
X(vkCmdPipelineBarrier2);
X(vkCmdPushConstants);
X(vkCmdPushDescriptorSetWithTemplateKHR);
X(vkCmdSetBlendConstants);
@@ -226,6 +227,7 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
X(vkGetSemaphoreCounterValue);
X(vkMapMemory);
X(vkQueueSubmit);
X(vkQueueSubmit2);
X(vkResetFences);
X(vkResetQueryPool);
X(vkSetDebugUtilsObjectNameEXT);
@@ -252,6 +254,14 @@ void Load(VkDevice device, DeviceDispatch& dld) noexcept {
Proc(dld.vkCmdDrawIndirectCount, dld, "vkCmdDrawIndirectCountKHR", device);
Proc(dld.vkCmdDrawIndexedIndirectCount, dld, "vkCmdDrawIndexedIndirectCountKHR", device);
}
// Synchronization2 is core in Vulkan 1.3, otherwise requires VK_KHR_synchronization2
if (!dld.vkCmdPipelineBarrier2) {
Proc(dld.vkCmdPipelineBarrier2, dld, "vkCmdPipelineBarrier2KHR", device);
}
if (!dld.vkQueueSubmit2) {
Proc(dld.vkQueueSubmit2, dld, "vkQueueSubmit2KHR", device);
}
#undef X
}
src/video_core/vulkan_common/vulkan_wrapper.h
+78
View File
@@ -6,6 +6,7 @@
#pragma once
#include <array>
#include <exception>
#include <limits>
#include <memory>
@@ -237,6 +238,7 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkCmdEndTransformFeedbackEXT vkCmdEndTransformFeedbackEXT{};
PFN_vkCmdFillBuffer vkCmdFillBuffer{};
PFN_vkCmdPipelineBarrier vkCmdPipelineBarrier{};
PFN_vkCmdPipelineBarrier2 vkCmdPipelineBarrier2{};
PFN_vkCmdPushConstants vkCmdPushConstants{};
PFN_vkCmdPushDescriptorSetWithTemplateKHR vkCmdPushDescriptorSetWithTemplateKHR{};
PFN_vkCmdResolveImage vkCmdResolveImage{};
@@ -340,6 +342,7 @@ struct DeviceDispatch : InstanceDispatch {
PFN_vkGetSemaphoreCounterValue vkGetSemaphoreCounterValue{};
PFN_vkMapMemory vkMapMemory{};
PFN_vkQueueSubmit vkQueueSubmit{};
PFN_vkQueueSubmit2 vkQueueSubmit2{};
PFN_vkResetFences vkResetFences{};
PFN_vkResetQueryPool vkResetQueryPool{};
PFN_vkSetDebugUtilsObjectNameEXT vkSetDebugUtilsObjectNameEXT{};
@@ -819,6 +822,13 @@ public:
return dld->vkQueueSubmit(queue, submit_infos.size(), submit_infos.data(), fence);
}
/// Submits using VK_KHR_synchronization2 / Vulkan 1.3 vkQueueSubmit2.
/// Only valid to call when the device dispatch table has vkQueueSubmit2 loaded.
VkResult Submit2(Span<VkSubmitInfo2> submit_infos,
VkFence fence = VK_NULL_HANDLE) const noexcept {
return dld->vkQueueSubmit2(queue, submit_infos.size(), submit_infos.data(), fence);
}
VkResult Present(const VkPresentInfoKHR& present_info) const noexcept {
return dld->vkQueuePresentKHR(queue, &present_info);
}
@@ -1287,6 +1297,74 @@ public:
VkDependencyFlags dependency_flags, Span<VkMemoryBarrier> memory_barriers,
Span<VkBufferMemoryBarrier> buffer_barriers,
Span<VkImageMemoryBarrier> image_barriers) const noexcept {
// Legacy VkPipelineStageFlagBits/VkAccessFlagBits are bit-compatible with their
// Synchronization2 *2 counterparts, so barriers can be widened without a lookup table.
static constexpr u32 MaxBarriers = 16;
if (dld->vkCmdPipelineBarrier2 && memory_barriers.size() <= MaxBarriers &&
buffer_barriers.size() <= MaxBarriers && image_barriers.size() <= MaxBarriers) {
const auto src_stage_mask2 = static_cast<VkPipelineStageFlags2>(src_stage_mask);
const auto dst_stage_mask2 = static_cast<VkPipelineStageFlags2>(dst_stage_mask);
std::array<VkMemoryBarrier2, MaxBarriers> memory_barriers2;
for (u32 i = 0; i < memory_barriers.size(); ++i) {
memory_barriers2[i] = VkMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(memory_barriers[i].srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(memory_barriers[i].dstAccessMask),
};
}
std::array<VkBufferMemoryBarrier2, MaxBarriers> buffer_barriers2;
for (u32 i = 0; i < buffer_barriers.size(); ++i) {
const auto& barrier = buffer_barriers[i];
buffer_barriers2[i] = VkBufferMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(barrier.srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(barrier.dstAccessMask),
.srcQueueFamilyIndex = barrier.srcQueueFamilyIndex,
.dstQueueFamilyIndex = barrier.dstQueueFamilyIndex,
.buffer = barrier.buffer,
.offset = barrier.offset,
.size = barrier.size,
};
}
std::array<VkImageMemoryBarrier2, MaxBarriers> image_barriers2;
for (u32 i = 0; i < image_barriers.size(); ++i) {
const auto& barrier = image_barriers[i];
image_barriers2[i] = VkImageMemoryBarrier2{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER_2,
.pNext = nullptr,
.srcStageMask = src_stage_mask2,
.srcAccessMask = static_cast<VkAccessFlags2>(barrier.srcAccessMask),
.dstStageMask = dst_stage_mask2,
.dstAccessMask = static_cast<VkAccessFlags2>(barrier.dstAccessMask),
.oldLayout = barrier.oldLayout,
.newLayout = barrier.newLayout,
.srcQueueFamilyIndex = barrier.srcQueueFamilyIndex,
.dstQueueFamilyIndex = barrier.dstQueueFamilyIndex,
.image = barrier.image,
.subresourceRange = barrier.subresourceRange,
};
}
const VkDependencyInfo dependency_info{
.sType = VK_STRUCTURE_TYPE_DEPENDENCY_INFO,
.pNext = nullptr,
.dependencyFlags = dependency_flags,
.memoryBarrierCount = memory_barriers.size(),
.pMemoryBarriers = memory_barriers2.data(),
.bufferMemoryBarrierCount = buffer_barriers.size(),
.pBufferMemoryBarriers = buffer_barriers2.data(),
.imageMemoryBarrierCount = image_barriers.size(),
.pImageMemoryBarriers = image_barriers2.data(),
};
dld->vkCmdPipelineBarrier2(handle, &dependency_info);
return;
}
dld->vkCmdPipelineBarrier(handle, src_stage_mask, dst_stage_mask, dependency_flags,
memory_barriers.size(), memory_barriers.data(),
buffer_barriers.size(), buffer_barriers.data(),