liamwhite
GitHub
commit
020dbcdbc7
@ -0,0 +1,29 @@
|
||||
// SPDX-FileCopyrightText: Copyright 2022 yuzu Emulator Project
|
||||
// SPDX-License-Identifier: GPL-2.0-or-later
|
||||
|
||||
#version 460 core
|
||||
|
||||
layout (local_size_x = 16, local_size_y = 8, local_size_z = 1) in;
|
||||
|
||||
layout (binding = 0) buffer ThreadData {
|
||||
uint data[];
|
||||
};
|
||||
|
||||
uint xorshift32(uint x) {
|
||||
x ^= x << 13;
|
||||
x ^= x >> 17;
|
||||
x ^= x << 5;
|
||||
return x;
|
||||
}
|
||||
|
||||
uint getGlobalIndex() {
|
||||
return gl_GlobalInvocationID.x + gl_GlobalInvocationID.y * gl_WorkGroupSize.y * gl_NumWorkGroups.y;
|
||||
}
|
||||
|
||||
void main() {
|
||||
uint myIndex = xorshift32(getGlobalIndex());
|
||||
uint otherIndex = xorshift32(myIndex);
|
||||
|
||||
uint otherValue = atomicAdd(data[otherIndex % data.length()], 0) + 1;
|
||||
atomicAdd(data[myIndex % data.length()], otherValue);
|
||||
}
|
||||
@ -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
|
||||
@ -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
|
||||
Loading…
Reference in New Issue