summaryrefslogtreecommitdiffstats
path: root/src/video_core/renderer_vulkan
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/renderer_vulkan')
-rw-r--r--src/video_core/renderer_vulkan/renderer_vulkan.cpp8
-rw-r--r--src/video_core/renderer_vulkan/renderer_vulkan.h5
-rw-r--r--src/video_core/renderer_vulkan/vk_buffer_cache.cpp11
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.cpp23
-rw-r--r--src/video_core/renderer_vulkan/vk_rasterizer.h1
-rw-r--r--src/video_core/renderer_vulkan/vk_turbo_mode.cpp205
-rw-r--r--src/video_core/renderer_vulkan/vk_turbo_mode.h26
7 files changed, 277 insertions, 2 deletions
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.cpp b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
index f502a7d09..1578cb206 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.cpp
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.cpp
@@ -78,6 +78,8 @@ std::string BuildCommaSeparatedExtensions(std::vector<std::string> available_ext
return separated_extensions;
}
+} // Anonymous namespace
+
Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dld,
VkSurfaceKHR surface) {
const std::vector<VkPhysicalDevice> devices = instance.EnumeratePhysicalDevices();
@@ -89,7 +91,6 @@ Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dl
const vk::PhysicalDevice physical_device(devices[device_index], dld);
return Device(*instance, physical_device, surface, dld);
}
-} // Anonymous namespace
RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
Core::Frontend::EmuWindow& emu_window,
@@ -98,7 +99,7 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
: RendererBase(emu_window, std::move(context_)), telemetry_session(telemetry_session_),
cpu_memory(cpu_memory_), gpu(gpu_), library(OpenLibrary()),
instance(CreateInstance(library, dld, VK_API_VERSION_1_1, render_window.GetWindowInfo().type,
- true, Settings::values.renderer_debug.GetValue())),
+ Settings::values.renderer_debug.GetValue())),
debug_callback(Settings::values.renderer_debug ? CreateDebugCallback(instance) : nullptr),
surface(CreateSurface(instance, render_window)),
device(CreateDevice(instance, dld, *surface)), memory_allocator(device, false),
@@ -109,6 +110,9 @@ RendererVulkan::RendererVulkan(Core::TelemetrySession& telemetry_session_,
screen_info),
rasterizer(render_window, gpu, cpu_memory, screen_info, device, memory_allocator,
state_tracker, scheduler) {
+ if (Settings::values.renderer_force_max_clock.GetValue()) {
+ turbo_mode.emplace(instance, dld);
+ }
Report();
} catch (const vk::Exception& exception) {
LOG_ERROR(Render_Vulkan, "Vulkan initialization failed with error: {}", exception.what());
diff --git a/src/video_core/renderer_vulkan/renderer_vulkan.h b/src/video_core/renderer_vulkan/renderer_vulkan.h
index e7bfecb20..009e75e0d 100644
--- a/src/video_core/renderer_vulkan/renderer_vulkan.h
+++ b/src/video_core/renderer_vulkan/renderer_vulkan.h
@@ -13,6 +13,7 @@
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_state_tracker.h"
#include "video_core/renderer_vulkan/vk_swapchain.h"
+#include "video_core/renderer_vulkan/vk_turbo_mode.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
@@ -31,6 +32,9 @@ class GPU;
namespace Vulkan {
+Device CreateDevice(const vk::Instance& instance, const vk::InstanceDispatch& dld,
+ VkSurfaceKHR surface);
+
class RendererVulkan final : public VideoCore::RendererBase {
public:
explicit RendererVulkan(Core::TelemetrySession& telemtry_session,
@@ -74,6 +78,7 @@ private:
Swapchain swapchain;
BlitScreen blit_screen;
RasterizerVulkan rasterizer;
+ std::optional<TurboMode> turbo_mode;
};
} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
index bc78287ce..1cfb4c2ff 100644
--- a/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_buffer_cache.cpp
@@ -340,6 +340,9 @@ void BufferCacheRuntime::Finish() {
void BufferCacheRuntime::CopyBuffer(VkBuffer dst_buffer, VkBuffer src_buffer,
std::span<const VideoCommon::BufferCopy> copies, bool barrier) {
+ if (dst_buffer == VK_NULL_HANDLE || src_buffer == VK_NULL_HANDLE) {
+ return;
+ }
static constexpr VkMemoryBarrier READ_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
.pNext = nullptr,
@@ -398,6 +401,9 @@ void BufferCacheRuntime::PostCopyBarrier() {
}
void BufferCacheRuntime::ClearBuffer(VkBuffer dest_buffer, u32 offset, size_t size, u32 value) {
+ if (dest_buffer == VK_NULL_HANDLE) {
+ return;
+ }
static constexpr VkMemoryBarrier READ_BARRIER{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
.pNext = nullptr,
@@ -477,6 +483,11 @@ void BufferCacheRuntime::BindVertexBuffer(u32 index, VkBuffer buffer, u32 offset
cmdbuf.BindVertexBuffers2EXT(index, 1, &buffer, &vk_offset, &vk_size, &vk_stride);
});
} else {
+ if (!device.HasNullDescriptor() && buffer == VK_NULL_HANDLE) {
+ ReserveNullBuffer();
+ buffer = *null_buffer;
+ offset = 0;
+ }
scheduler.Record([index, buffer, offset](vk::CommandBuffer cmdbuf) {
cmdbuf.BindVertexBuffer(index, buffer, offset);
});
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 242bf9602..ed4a72166 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -186,6 +186,7 @@ void RasterizerVulkan::PrepareDraw(bool is_indexed, Func&& draw_func) {
SCOPE_EXIT({ gpu.TickWork(); });
FlushWork();
+ gpu_memory->FlushCaching();
query_cache.UpdateCounters();
@@ -393,6 +394,7 @@ void RasterizerVulkan::Clear(u32 layer_count) {
void RasterizerVulkan::DispatchCompute() {
FlushWork();
+ gpu_memory->FlushCaching();
ComputePipeline* const pipeline{pipeline_cache.CurrentComputePipeline()};
if (!pipeline) {
@@ -481,6 +483,27 @@ void RasterizerVulkan::InvalidateRegion(VAddr addr, u64 size, VideoCommon::Cache
}
}
+void RasterizerVulkan::InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) {
+ {
+ std::scoped_lock lock{texture_cache.mutex};
+ for (const auto& [addr, size] : sequences) {
+ texture_cache.WriteMemory(addr, size);
+ }
+ }
+ {
+ std::scoped_lock lock{buffer_cache.mutex};
+ for (const auto& [addr, size] : sequences) {
+ buffer_cache.WriteMemory(addr, size);
+ }
+ }
+ {
+ for (const auto& [addr, size] : sequences) {
+ query_cache.InvalidateRegion(addr, size);
+ pipeline_cache.InvalidateRegion(addr, size);
+ }
+ }
+}
+
void RasterizerVulkan::OnCPUWrite(VAddr addr, u64 size) {
if (addr == 0 || size == 0) {
return;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.h b/src/video_core/renderer_vulkan/vk_rasterizer.h
index c661e5b19..472cc64d9 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.h
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.h
@@ -79,6 +79,7 @@ public:
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
void InvalidateRegion(VAddr addr, u64 size,
VideoCommon::CacheType which = VideoCommon::CacheType::All) override;
+ void InnerInvalidation(std::span<const std::pair<VAddr, std::size_t>> sequences) override;
void OnCPUWrite(VAddr addr, u64 size) override;
void InvalidateGPUCache() override;
void UnmapMemory(VAddr addr, u64 size) override;
diff --git a/src/video_core/renderer_vulkan/vk_turbo_mode.cpp b/src/video_core/renderer_vulkan/vk_turbo_mode.cpp
new file mode 100644
index 000000000..852b86f84
--- /dev/null
+++ b/src/video_core/renderer_vulkan/vk_turbo_mode.cpp
@@ -0,0 +1,205 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "common/literals.h"
+#include "video_core/host_shaders/vulkan_turbo_mode_comp_spv.h"
+#include "video_core/renderer_vulkan/renderer_vulkan.h"
+#include "video_core/renderer_vulkan/vk_shader_util.h"
+#include "video_core/renderer_vulkan/vk_turbo_mode.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+
+namespace Vulkan {
+
+using namespace Common::Literals;
+
+TurboMode::TurboMode(const vk::Instance& instance, const vk::InstanceDispatch& dld)
+ : m_device{CreateDevice(instance, dld, VK_NULL_HANDLE)}, m_allocator{m_device, false} {
+ m_thread = std::jthread([&](auto stop_token) { Run(stop_token); });
+}
+
+TurboMode::~TurboMode() = default;
+
+void TurboMode::Run(std::stop_token stop_token) {
+ auto& dld = m_device.GetLogical();
+
+ // Allocate buffer. 2MiB should be sufficient.
+ auto buffer = dld.CreateBuffer(VkBufferCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .size = 2_MiB,
+ .usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
+ .sharingMode = VK_SHARING_MODE_EXCLUSIVE,
+ .queueFamilyIndexCount = 0,
+ .pQueueFamilyIndices = nullptr,
+ });
+
+ // Commit some device local memory for the buffer.
+ auto commit = m_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
+
+ // Create the descriptor pool to contain our descriptor.
+ constexpr VkDescriptorPoolSize pool_size{
+ .type = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ };
+
+ auto descriptor_pool = dld.CreateDescriptorPool(VkDescriptorPoolCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT,
+ .maxSets = 1,
+ .poolSizeCount = 1,
+ .pPoolSizes = &pool_size,
+ });
+
+ // Create the descriptor set layout from the pool.
+ constexpr VkDescriptorSetLayoutBinding layout_binding{
+ .binding = 0,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .descriptorCount = 1,
+ .stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
+ .pImmutableSamplers = nullptr,
+ };
+
+ auto descriptor_set_layout = dld.CreateDescriptorSetLayout(VkDescriptorSetLayoutCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .bindingCount = 1,
+ .pBindings = &layout_binding,
+ });
+
+ // Actually create the descriptor set.
+ auto descriptor_set = descriptor_pool.Allocate(VkDescriptorSetAllocateInfo{
+ .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
+ .pNext = nullptr,
+ .descriptorPool = *descriptor_pool,
+ .descriptorSetCount = 1,
+ .pSetLayouts = descriptor_set_layout.address(),
+ });
+
+ // Create the shader.
+ auto shader = BuildShader(m_device, VULKAN_TURBO_MODE_COMP_SPV);
+
+ // Create the pipeline layout.
+ auto pipeline_layout = dld.CreatePipelineLayout(VkPipelineLayoutCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .setLayoutCount = 1,
+ .pSetLayouts = descriptor_set_layout.address(),
+ .pushConstantRangeCount = 0,
+ .pPushConstantRanges = nullptr,
+ });
+
+ // Actually create the pipeline.
+ const VkPipelineShaderStageCreateInfo shader_stage{
+ .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .stage = VK_SHADER_STAGE_COMPUTE_BIT,
+ .module = *shader,
+ .pName = "main",
+ .pSpecializationInfo = nullptr,
+ };
+
+ auto pipeline = dld.CreateComputePipeline(VkComputePipelineCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ .stage = shader_stage,
+ .layout = *pipeline_layout,
+ .basePipelineHandle = VK_NULL_HANDLE,
+ .basePipelineIndex = 0,
+ });
+
+ // Create a fence to wait on.
+ auto fence = dld.CreateFence(VkFenceCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO,
+ .pNext = nullptr,
+ .flags = 0,
+ });
+
+ // Create a command pool to allocate a command buffer from.
+ auto command_pool = dld.CreateCommandPool(VkCommandPoolCreateInfo{
+ .sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO,
+ .pNext = nullptr,
+ .flags =
+ VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT,
+ .queueFamilyIndex = m_device.GetGraphicsFamily(),
+ });
+
+ // Create a single command buffer.
+ auto cmdbufs = command_pool.Allocate(1, VK_COMMAND_BUFFER_LEVEL_PRIMARY);
+ auto cmdbuf = vk::CommandBuffer{cmdbufs[0], m_device.GetDispatchLoader()};
+
+ while (!stop_token.stop_requested()) {
+ // Reset the fence.
+ fence.Reset();
+
+ // Update descriptor set.
+ const VkDescriptorBufferInfo buffer_info{
+ .buffer = *buffer,
+ .offset = 0,
+ .range = VK_WHOLE_SIZE,
+ };
+
+ const VkWriteDescriptorSet buffer_write{
+ .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
+ .pNext = nullptr,
+ .dstSet = descriptor_set[0],
+ .dstBinding = 0,
+ .dstArrayElement = 0,
+ .descriptorCount = 1,
+ .descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
+ .pImageInfo = nullptr,
+ .pBufferInfo = &buffer_info,
+ .pTexelBufferView = nullptr,
+ };
+
+ dld.UpdateDescriptorSets(std::array{buffer_write}, {});
+
+ // Set up the command buffer.
+ cmdbuf.Begin(VkCommandBufferBeginInfo{
+ .sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO,
+ .pNext = nullptr,
+ .flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT,
+ .pInheritanceInfo = nullptr,
+ });
+
+ // Clear the buffer.
+ cmdbuf.FillBuffer(*buffer, 0, VK_WHOLE_SIZE, 0);
+
+ // Bind descriptor set.
+ cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0,
+ descriptor_set, {});
+
+ // Bind the pipeline.
+ cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline);
+
+ // Dispatch.
+ cmdbuf.Dispatch(64, 64, 1);
+
+ // Finish.
+ cmdbuf.End();
+
+ const VkSubmitInfo submit_info{
+ .sType = VK_STRUCTURE_TYPE_SUBMIT_INFO,
+ .pNext = nullptr,
+ .waitSemaphoreCount = 0,
+ .pWaitSemaphores = nullptr,
+ .pWaitDstStageMask = nullptr,
+ .commandBufferCount = 1,
+ .pCommandBuffers = cmdbuf.address(),
+ .signalSemaphoreCount = 0,
+ .pSignalSemaphores = nullptr,
+ };
+
+ m_device.GetGraphicsQueue().Submit(std::array{submit_info}, *fence);
+
+ // Wait for completion.
+ fence.Wait();
+ }
+}
+
+} // namespace Vulkan
diff --git a/src/video_core/renderer_vulkan/vk_turbo_mode.h b/src/video_core/renderer_vulkan/vk_turbo_mode.h
new file mode 100644
index 000000000..2060e2395
--- /dev/null
+++ b/src/video_core/renderer_vulkan/vk_turbo_mode.h
@@ -0,0 +1,26 @@
+// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#pragma once
+
+#include "common/polyfill_thread.h"
+#include "video_core/vulkan_common/vulkan_device.h"
+#include "video_core/vulkan_common/vulkan_memory_allocator.h"
+#include "video_core/vulkan_common/vulkan_wrapper.h"
+
+namespace Vulkan {
+
+class TurboMode {
+public:
+ explicit TurboMode(const vk::Instance& instance, const vk::InstanceDispatch& dld);
+ ~TurboMode();
+
+private:
+ void Run(std::stop_token stop_token);
+
+ Device m_device;
+ MemoryAllocator m_allocator;
+ std::jthread m_thread;
+};
+
+} // namespace Vulkan
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-29 08:16:18 +0200