AMA25/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Merge pull request #3636 from ReinUsesLisp/drop-vk-hpp

Browse source 
renderer_vulkan: Drop Vulkan-Hpp
This commit is contained in:
Rodrigo Locatti 2020-04-13 17:08:04 -03:00 committed by GitHub
commit 7e4a132a77
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
52 changed files with 2830 additions and 2221 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
externals/Vulkan-Headers vendored

@ -1 +1 @@
Subproject commit d42d0747ee1b7a6726fb8948444b4993f9dcd2e5
Subproject commit 0e78ffd1dcfc3e9f14a966b9660dbc59bd967c5c

1
src/video_core/CMakeLists.txt
View file

@ -156,7 +156,6 @@ add_library(video_core STATIC
if (ENABLE_VULKAN)
target_sources(video_core PRIVATE
renderer_vulkan/declarations.h
renderer_vulkan/fixed_pipeline_state.cpp
renderer_vulkan/fixed_pipeline_state.h
renderer_vulkan/maxwell_to_vk.cpp

60
src/video_core/renderer_vulkan/declarations.h
View file

@ -1,60 +0,0 @@
// Copyright 2019 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
namespace vk {
class DispatchLoaderDynamic;
}
namespace Vulkan {
constexpr vk::DispatchLoaderDynamic* dont_use_me_dld = nullptr;
}
#define VULKAN_HPP_DEFAULT_DISPATCHER (*::Vulkan::dont_use_me_dld)
#define VULKAN_HPP_ENABLE_DYNAMIC_LOADER_TOOL 0
#define VULKAN_HPP_DISPATCH_LOADER_DYNAMIC 1
#include <vulkan/vulkan.hpp>
namespace Vulkan {
// vulkan.hpp unique handlers use DispatchLoaderStatic
template <typename T>
using UniqueHandle = vk::UniqueHandle<T, vk::DispatchLoaderDynamic>;
using UniqueAccelerationStructureNV = UniqueHandle<vk::AccelerationStructureNV>;
using UniqueBuffer = UniqueHandle<vk::Buffer>;
using UniqueBufferView = UniqueHandle<vk::BufferView>;
using UniqueCommandBuffer = UniqueHandle<vk::CommandBuffer>;
using UniqueCommandPool = UniqueHandle<vk::CommandPool>;
using UniqueDescriptorPool = UniqueHandle<vk::DescriptorPool>;
using UniqueDescriptorSet = UniqueHandle<vk::DescriptorSet>;
using UniqueDescriptorSetLayout = UniqueHandle<vk::DescriptorSetLayout>;
using UniqueDescriptorUpdateTemplate = UniqueHandle<vk::DescriptorUpdateTemplate>;
using UniqueDevice = UniqueHandle<vk::Device>;
using UniqueDeviceMemory = UniqueHandle<vk::DeviceMemory>;
using UniqueEvent = UniqueHandle<vk::Event>;
using UniqueFence = UniqueHandle<vk::Fence>;
using UniqueFramebuffer = UniqueHandle<vk::Framebuffer>;
using UniqueImage = UniqueHandle<vk::Image>;
using UniqueImageView = UniqueHandle<vk::ImageView>;
using UniqueInstance = UniqueHandle<vk::Instance>;
using UniqueIndirectCommandsLayoutNVX = UniqueHandle<vk::IndirectCommandsLayoutNVX>;
using UniqueObjectTableNVX = UniqueHandle<vk::ObjectTableNVX>;
using UniquePipeline = UniqueHandle<vk::Pipeline>;
using UniquePipelineCache = UniqueHandle<vk::PipelineCache>;
using UniquePipelineLayout = UniqueHandle<vk::PipelineLayout>;
using UniqueQueryPool = UniqueHandle<vk::QueryPool>;
using UniqueRenderPass = UniqueHandle<vk::RenderPass>;
using UniqueSampler = UniqueHandle<vk::Sampler>;
using UniqueSamplerYcbcrConversion = UniqueHandle<vk::SamplerYcbcrConversion>;
using UniqueSemaphore = UniqueHandle<vk::Semaphore>;
using UniqueShaderModule = UniqueHandle<vk::ShaderModule>;
using UniqueSurfaceKHR = UniqueHandle<vk::SurfaceKHR>;
using UniqueSwapchainKHR = UniqueHandle<vk::SwapchainKHR>;
using UniqueValidationCacheEXT = UniqueHandle<vk::ValidationCacheEXT>;
using UniqueDebugReportCallbackEXT = UniqueHandle<vk::DebugReportCallbackEXT>;
using UniqueDebugUtilsMessengerEXT = UniqueHandle<vk::DebugUtilsMessengerEXT>;
} // namespace Vulkan

523
src/video_core/renderer_vulkan/maxwell_to_vk.cpp
View file

@ -2,13 +2,15 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <iterator>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/surface.h"
namespace Vulkan::MaxwellToVK {
@ -17,88 +19,89 @@ using Maxwell = Tegra::Engines::Maxwell3D::Regs;
namespace Sampler {
vk::Filter Filter(Tegra::Texture::TextureFilter filter) {
VkFilter Filter(Tegra::Texture::TextureFilter filter) {
switch (filter) {
case Tegra::Texture::TextureFilter::Linear:
return vk::Filter::eLinear;
return VK_FILTER_LINEAR;
case Tegra::Texture::TextureFilter::Nearest:
return vk::Filter::eNearest;
return VK_FILTER_NEAREST;
}
UNIMPLEMENTED_MSG("Unimplemented sampler filter={}", static_cast<u32>(filter));
return {};
}
vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter) {
VkSamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter) {
switch (mipmap_filter) {
case Tegra::Texture::TextureMipmapFilter::None:
// TODO(Rodrigo): None seems to be mapped to OpenGL's mag and min filters without mipmapping
// (e.g. GL_NEAREST and GL_LINEAR). Vulkan doesn't have such a thing, find out if we have to
// use an image view with a single mipmap level to emulate this.
return vk::SamplerMipmapMode::eLinear;
return VK_SAMPLER_MIPMAP_MODE_LINEAR;
;
case Tegra::Texture::TextureMipmapFilter::Linear:
return vk::SamplerMipmapMode::eLinear;
return VK_SAMPLER_MIPMAP_MODE_LINEAR;
case Tegra::Texture::TextureMipmapFilter::Nearest:
return vk::SamplerMipmapMode::eNearest;
return VK_SAMPLER_MIPMAP_MODE_NEAREST;
}
UNIMPLEMENTED_MSG("Unimplemented sampler mipmap mode={}", static_cast<u32>(mipmap_filter));
return {};
}
vk::SamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
Tegra::Texture::TextureFilter filter) {
VkSamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
Tegra::Texture::TextureFilter filter) {
switch (wrap_mode) {
case Tegra::Texture::WrapMode::Wrap:
return vk::SamplerAddressMode::eRepeat;
return VK_SAMPLER_ADDRESS_MODE_REPEAT;
case Tegra::Texture::WrapMode::Mirror:
return vk::SamplerAddressMode::eMirroredRepeat;
return VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT;
case Tegra::Texture::WrapMode::ClampToEdge:
return vk::SamplerAddressMode::eClampToEdge;
return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
case Tegra::Texture::WrapMode::Border:
return vk::SamplerAddressMode::eClampToBorder;
return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
case Tegra::Texture::WrapMode::Clamp:
if (device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) {
if (device.GetDriverID() == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR) {
// Nvidia's Vulkan driver defaults to GL_CLAMP on invalid enumerations, we can hack this
// by sending an invalid enumeration.
return static_cast<vk::SamplerAddressMode>(0xcafe);
return static_cast<VkSamplerAddressMode>(0xcafe);
}
// TODO(Rodrigo): Emulate GL_CLAMP properly on other vendors
switch (filter) {
case Tegra::Texture::TextureFilter::Nearest:
return vk::SamplerAddressMode::eClampToEdge;
return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
case Tegra::Texture::TextureFilter::Linear:
return vk::SamplerAddressMode::eClampToBorder;
return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
}
UNREACHABLE();
return vk::SamplerAddressMode::eClampToEdge;
return VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
case Tegra::Texture::WrapMode::MirrorOnceClampToEdge:
return vk::SamplerAddressMode::eMirrorClampToEdge;
return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE;
case Tegra::Texture::WrapMode::MirrorOnceBorder:
UNIMPLEMENTED();
return vk::SamplerAddressMode::eMirrorClampToEdge;
return VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE;
default:
UNIMPLEMENTED_MSG("Unimplemented wrap mode={}", static_cast<u32>(wrap_mode));
return {};
}
}
vk::CompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func) {
VkCompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func) {
switch (depth_compare_func) {
case Tegra::Texture::DepthCompareFunc::Never:
return vk::CompareOp::eNever;
return VK_COMPARE_OP_NEVER;
case Tegra::Texture::DepthCompareFunc::Less:
return vk::CompareOp::eLess;
return VK_COMPARE_OP_LESS;
case Tegra::Texture::DepthCompareFunc::LessEqual:
return vk::CompareOp::eLessOrEqual;
return VK_COMPARE_OP_LESS_OR_EQUAL;
case Tegra::Texture::DepthCompareFunc::Equal:
return vk::CompareOp::eEqual;
return VK_COMPARE_OP_EQUAL;
case Tegra::Texture::DepthCompareFunc::NotEqual:
return vk::CompareOp::eNotEqual;
return VK_COMPARE_OP_NOT_EQUAL;
case Tegra::Texture::DepthCompareFunc::Greater:
return vk::CompareOp::eGreater;
return VK_COMPARE_OP_GREATER;
case Tegra::Texture::DepthCompareFunc::GreaterEqual:
return vk::CompareOp::eGreaterOrEqual;
return VK_COMPARE_OP_GREATER_OR_EQUAL;
case Tegra::Texture::DepthCompareFunc::Always:
return vk::CompareOp::eAlways;
return VK_COMPARE_OP_ALWAYS;
}
UNIMPLEMENTED_MSG("Unimplemented sampler depth compare function={}",
static_cast<u32>(depth_compare_func));
@ -112,92 +115,92 @@ namespace {
enum : u32 { Attachable = 1, Storage = 2 };
struct FormatTuple {
vk::Format format; ///< Vulkan format
int usage; ///< Describes image format usage
VkFormat format; ///< Vulkan format
int usage = 0; ///< Describes image format usage
} constexpr tex_format_tuples[] = {
{vk::Format::eA8B8G8R8UnormPack32, Attachable | Storage}, // ABGR8U
{vk::Format::eA8B8G8R8SnormPack32, Attachable | Storage}, // ABGR8S
{vk::Format::eA8B8G8R8UintPack32, Attachable | Storage}, // ABGR8UI
{vk::Format::eB5G6R5UnormPack16, {}}, // B5G6R5U
{vk::Format::eA2B10G10R10UnormPack32, Attachable | Storage}, // A2B10G10R10U
{vk::Format::eA1R5G5B5UnormPack16, Attachable}, // A1B5G5R5U (flipped with swizzle)
{vk::Format::eR8Unorm, Attachable | Storage}, // R8U
{vk::Format::eR8Uint, Attachable | Storage}, // R8UI
{vk::Format::eR16G16B16A16Sfloat, Attachable | Storage}, // RGBA16F
{vk::Format::eR16G16B16A16Unorm, Attachable | Storage}, // RGBA16U
{vk::Format::eR16G16B16A16Snorm, Attachable | Storage}, // RGBA16S
{vk::Format::eR16G16B16A16Uint, Attachable | Storage}, // RGBA16UI
{vk::Format::eB10G11R11UfloatPack32, Attachable | Storage}, // R11FG11FB10F
{vk::Format::eR32G32B32A32Uint, Attachable | Storage}, // RGBA32UI
{vk::Format::eBc1RgbaUnormBlock, {}}, // DXT1
{vk::Format::eBc2UnormBlock, {}}, // DXT23
{vk::Format::eBc3UnormBlock, {}}, // DXT45
{vk::Format::eBc4UnormBlock, {}}, // DXN1
{vk::Format::eBc5UnormBlock, {}}, // DXN2UNORM
{vk::Format::eBc5SnormBlock, {}}, // DXN2SNORM
{vk::Format::eBc7UnormBlock, {}}, // BC7U
{vk::Format::eBc6HUfloatBlock, {}}, // BC6H_UF16
{vk::Format::eBc6HSfloatBlock, {}}, // BC6H_SF16
{vk::Format::eAstc4x4UnormBlock, {}}, // ASTC_2D_4X4
{vk::Format::eB8G8R8A8Unorm, {}}, // BGRA8
{vk::Format::eR32G32B32A32Sfloat, Attachable | Storage}, // RGBA32F
{vk::Format::eR32G32Sfloat, Attachable | Storage}, // RG32F
{vk::Format::eR32Sfloat, Attachable | Storage}, // R32F
{vk::Format::eR16Sfloat, Attachable | Storage}, // R16F
{vk::Format::eR16Unorm, Attachable | Storage}, // R16U
{vk::Format::eUndefined, {}}, // R16S
{vk::Format::eUndefined, {}}, // R16UI
{vk::Format::eUndefined, {}}, // R16I
{vk::Format::eR16G16Unorm, Attachable | Storage}, // RG16
{vk::Format::eR16G16Sfloat, Attachable | Storage}, // RG16F
{vk::Format::eUndefined, {}}, // RG16UI
{vk::Format::eUndefined, {}}, // RG16I
{vk::Format::eR16G16Snorm, Attachable | Storage}, // RG16S
{vk::Format::eUndefined, {}}, // RGB32F
{vk::Format::eR8G8B8A8Srgb, Attachable}, // RGBA8_SRGB
{vk::Format::eR8G8Unorm, Attachable | Storage}, // RG8U
{vk::Format::eR8G8Snorm, Attachable | Storage}, // RG8S
{vk::Format::eR32G32Uint, Attachable | Storage}, // RG32UI
{vk::Format::eUndefined, {}}, // RGBX16F
{vk::Format::eR32Uint, Attachable | Storage}, // R32UI
{vk::Format::eR32Sint, Attachable | Storage}, // R32I
{vk::Format::eAstc8x8UnormBlock, {}}, // ASTC_2D_8X8
{vk::Format::eUndefined, {}}, // ASTC_2D_8X5
{vk::Format::eUndefined, {}}, // ASTC_2D_5X4
{vk::Format::eUndefined, {}}, // BGRA8_SRGB
{vk::Format::eBc1RgbaSrgbBlock, {}}, // DXT1_SRGB
{vk::Format::eBc2SrgbBlock, {}}, // DXT23_SRGB
{vk::Format::eBc3SrgbBlock, {}}, // DXT45_SRGB
{vk::Format::eBc7SrgbBlock, {}}, // BC7U_SRGB
{vk::Format::eR4G4B4A4UnormPack16, Attachable}, // R4G4B4A4U
{vk::Format::eAstc4x4SrgbBlock, {}}, // ASTC_2D_4X4_SRGB
{vk::Format::eAstc8x8SrgbBlock, {}}, // ASTC_2D_8X8_SRGB
{vk::Format::eAstc8x5SrgbBlock, {}}, // ASTC_2D_8X5_SRGB
{vk::Format::eAstc5x4SrgbBlock, {}}, // ASTC_2D_5X4_SRGB
{vk::Format::eAstc5x5UnormBlock, {}}, // ASTC_2D_5X5
{vk::Format::eAstc5x5SrgbBlock, {}}, // ASTC_2D_5X5_SRGB
{vk::Format::eAstc10x8UnormBlock, {}}, // ASTC_2D_10X8
{vk::Format::eAstc10x8SrgbBlock, {}}, // ASTC_2D_10X8_SRGB
{vk::Format::eAstc6x6UnormBlock, {}}, // ASTC_2D_6X6
{vk::Format::eAstc6x6SrgbBlock, {}}, // ASTC_2D_6X6_SRGB
{vk::Format::eAstc10x10UnormBlock, {}}, // ASTC_2D_10X10
{vk::Format::eAstc10x10SrgbBlock, {}}, // ASTC_2D_10X10_SRGB
{vk::Format::eAstc12x12UnormBlock, {}}, // ASTC_2D_12X12
{vk::Format::eAstc12x12SrgbBlock, {}}, // ASTC_2D_12X12_SRGB
{vk::Format::eAstc8x6UnormBlock, {}}, // ASTC_2D_8X6
{vk::Format::eAstc8x6SrgbBlock, {}}, // ASTC_2D_8X6_SRGB
{vk::Format::eAstc6x5UnormBlock, {}}, // ASTC_2D_6X5
{vk::Format::eAstc6x5SrgbBlock, {}}, // ASTC_2D_6X5_SRGB
{vk::Format::eE5B9G9R9UfloatPack32, {}}, // E5B9G9R9F
{VK_FORMAT_A8B8G8R8_UNORM_PACK32, Attachable | Storage}, // ABGR8U
{VK_FORMAT_A8B8G8R8_SNORM_PACK32, Attachable | Storage}, // ABGR8S
{VK_FORMAT_A8B8G8R8_UINT_PACK32, Attachable | Storage}, // ABGR8UI
{VK_FORMAT_B5G6R5_UNORM_PACK16}, // B5G6R5U
{VK_FORMAT_A2B10G10R10_UNORM_PACK32, Attachable | Storage}, // A2B10G10R10U
{VK_FORMAT_A1R5G5B5_UNORM_PACK16, Attachable}, // A1B5G5R5U (flipped with swizzle)
{VK_FORMAT_R8_UNORM, Attachable | Storage}, // R8U
{VK_FORMAT_R8_UINT, Attachable | Storage}, // R8UI
{VK_FORMAT_R16G16B16A16_SFLOAT, Attachable | Storage}, // RGBA16F
{VK_FORMAT_R16G16B16A16_UNORM, Attachable | Storage}, // RGBA16U
{VK_FORMAT_R16G16B16A16_SNORM, Attachable | Storage}, // RGBA16S
{VK_FORMAT_R16G16B16A16_UINT, Attachable | Storage}, // RGBA16UI
{VK_FORMAT_B10G11R11_UFLOAT_PACK32, Attachable | Storage}, // R11FG11FB10F
{VK_FORMAT_R32G32B32A32_UINT, Attachable | Storage}, // RGBA32UI
{VK_FORMAT_BC1_RGBA_UNORM_BLOCK}, // DXT1
{VK_FORMAT_BC2_UNORM_BLOCK}, // DXT23
{VK_FORMAT_BC3_UNORM_BLOCK}, // DXT45
{VK_FORMAT_BC4_UNORM_BLOCK}, // DXN1
{VK_FORMAT_BC5_UNORM_BLOCK}, // DXN2UNORM
{VK_FORMAT_BC5_SNORM_BLOCK}, // DXN2SNORM
{VK_FORMAT_BC7_UNORM_BLOCK}, // BC7U
{VK_FORMAT_BC6H_UFLOAT_BLOCK}, // BC6H_UF16
{VK_FORMAT_BC6H_SFLOAT_BLOCK}, // BC6H_SF16
{VK_FORMAT_ASTC_4x4_UNORM_BLOCK}, // ASTC_2D_4X4
{VK_FORMAT_B8G8R8A8_UNORM}, // BGRA8
{VK_FORMAT_R32G32B32A32_SFLOAT, Attachable | Storage}, // RGBA32F
{VK_FORMAT_R32G32_SFLOAT, Attachable | Storage}, // RG32F
{VK_FORMAT_R32_SFLOAT, Attachable | Storage}, // R32F
{VK_FORMAT_R16_SFLOAT, Attachable | Storage}, // R16F
{VK_FORMAT_R16_UNORM, Attachable | Storage}, // R16U
{VK_FORMAT_UNDEFINED}, // R16S
{VK_FORMAT_UNDEFINED}, // R16UI
{VK_FORMAT_UNDEFINED}, // R16I
{VK_FORMAT_R16G16_UNORM, Attachable | Storage}, // RG16
{VK_FORMAT_R16G16_SFLOAT, Attachable | Storage}, // RG16F
{VK_FORMAT_UNDEFINED}, // RG16UI
{VK_FORMAT_UNDEFINED}, // RG16I
{VK_FORMAT_R16G16_SNORM, Attachable | Storage}, // RG16S
{VK_FORMAT_UNDEFINED}, // RGB32F
{VK_FORMAT_R8G8B8A8_SRGB, Attachable}, // RGBA8_SRGB
{VK_FORMAT_R8G8_UNORM, Attachable | Storage}, // RG8U
{VK_FORMAT_R8G8_SNORM, Attachable | Storage}, // RG8S
{VK_FORMAT_R32G32_UINT, Attachable | Storage}, // RG32UI
{VK_FORMAT_UNDEFINED}, // RGBX16F
{VK_FORMAT_R32_UINT, Attachable | Storage}, // R32UI
{VK_FORMAT_R32_SINT, Attachable | Storage}, // R32I
{VK_FORMAT_ASTC_8x8_UNORM_BLOCK}, // ASTC_2D_8X8
{VK_FORMAT_UNDEFINED}, // ASTC_2D_8X5
{VK_FORMAT_UNDEFINED}, // ASTC_2D_5X4
{VK_FORMAT_UNDEFINED}, // BGRA8_SRGB
{VK_FORMAT_BC1_RGBA_SRGB_BLOCK}, // DXT1_SRGB
{VK_FORMAT_BC2_SRGB_BLOCK}, // DXT23_SRGB
{VK_FORMAT_BC3_SRGB_BLOCK}, // DXT45_SRGB
{VK_FORMAT_BC7_SRGB_BLOCK}, // BC7U_SRGB
{VK_FORMAT_R4G4B4A4_UNORM_PACK16, Attachable}, // R4G4B4A4U
{VK_FORMAT_ASTC_4x4_SRGB_BLOCK}, // ASTC_2D_4X4_SRGB
{VK_FORMAT_ASTC_8x8_SRGB_BLOCK}, // ASTC_2D_8X8_SRGB
{VK_FORMAT_ASTC_8x5_SRGB_BLOCK}, // ASTC_2D_8X5_SRGB
{VK_FORMAT_ASTC_5x4_SRGB_BLOCK}, // ASTC_2D_5X4_SRGB
{VK_FORMAT_ASTC_5x5_UNORM_BLOCK}, // ASTC_2D_5X5
{VK_FORMAT_ASTC_5x5_SRGB_BLOCK}, // ASTC_2D_5X5_SRGB
{VK_FORMAT_ASTC_10x8_UNORM_BLOCK}, // ASTC_2D_10X8
{VK_FORMAT_ASTC_10x8_SRGB_BLOCK}, // ASTC_2D_10X8_SRGB
{VK_FORMAT_ASTC_6x6_UNORM_BLOCK}, // ASTC_2D_6X6
{VK_FORMAT_ASTC_6x6_SRGB_BLOCK}, // ASTC_2D_6X6_SRGB
{VK_FORMAT_ASTC_10x10_UNORM_BLOCK}, // ASTC_2D_10X10
{VK_FORMAT_ASTC_10x10_SRGB_BLOCK}, // ASTC_2D_10X10_SRGB
{VK_FORMAT_ASTC_12x12_UNORM_BLOCK}, // ASTC_2D_12X12
{VK_FORMAT_ASTC_12x12_SRGB_BLOCK}, // ASTC_2D_12X12_SRGB
{VK_FORMAT_ASTC_8x6_UNORM_BLOCK}, // ASTC_2D_8X6
{VK_FORMAT_ASTC_8x6_SRGB_BLOCK}, // ASTC_2D_8X6_SRGB
{VK_FORMAT_ASTC_6x5_UNORM_BLOCK}, // ASTC_2D_6X5
{VK_FORMAT_ASTC_6x5_SRGB_BLOCK}, // ASTC_2D_6X5_SRGB
{VK_FORMAT_E5B9G9R9_UFLOAT_PACK32}, // E5B9G9R9F
// Depth formats
{vk::Format::eD32Sfloat, Attachable}, // Z32F
{vk::Format::eD16Unorm, Attachable}, // Z16
{VK_FORMAT_D32_SFLOAT, Attachable}, // Z32F
{VK_FORMAT_D16_UNORM, Attachable}, // Z16
// DepthStencil formats
{vk::Format::eD24UnormS8Uint, Attachable}, // Z24S8
{vk::Format::eD24UnormS8Uint, Attachable}, // S8Z24 (emulated)
{vk::Format::eD32SfloatS8Uint, Attachable}, // Z32FS8
{VK_FORMAT_D24_UNORM_S8_UINT, Attachable}, // Z24S8
{VK_FORMAT_D24_UNORM_S8_UINT, Attachable}, // S8Z24 (emulated)
{VK_FORMAT_D32_SFLOAT_S8_UINT, Attachable}, // Z32FS8
};
static_assert(std::size(tex_format_tuples) == VideoCore::Surface::MaxPixelFormat);
@ -212,106 +215,106 @@ FormatInfo SurfaceFormat(const VKDevice& device, FormatType format_type, PixelFo
ASSERT(static_cast<std::size_t>(pixel_format) < std::size(tex_format_tuples));
auto tuple = tex_format_tuples[static_cast<std::size_t>(pixel_format)];
if (tuple.format == vk::Format::eUndefined) {
if (tuple.format == VK_FORMAT_UNDEFINED) {
UNIMPLEMENTED_MSG("Unimplemented texture format with pixel format={}",
static_cast<u32>(pixel_format));
return {vk::Format::eA8B8G8R8UnormPack32, true, true};
return {VK_FORMAT_A8B8G8R8_UNORM_PACK32, true, true};
}
// Use ABGR8 on hardware that doesn't support ASTC natively
if (!device.IsOptimalAstcSupported() && VideoCore::Surface::IsPixelFormatASTC(pixel_format)) {
tuple.format = VideoCore::Surface::IsPixelFormatSRGB(pixel_format)
? vk::Format::eA8B8G8R8SrgbPack32
: vk::Format::eA8B8G8R8UnormPack32;
? VK_FORMAT_A8B8G8R8_SRGB_PACK32
: VK_FORMAT_A8B8G8R8_UNORM_PACK32;
}
const bool attachable = tuple.usage & Attachable;
const bool storage = tuple.usage & Storage;
vk::FormatFeatureFlags usage;
VkFormatFeatureFlags usage;
if (format_type == FormatType::Buffer) {
usage = vk::FormatFeatureFlagBits::eStorageTexelBuffer |
vk::FormatFeatureFlagBits::eUniformTexelBuffer;
usage =
VK_FORMAT_FEATURE_STORAGE_TEXEL_BUFFER_BIT | VK_FORMAT_FEATURE_UNIFORM_TEXEL_BUFFER_BIT;
} else {
usage = vk::FormatFeatureFlagBits::eSampledImage | vk::FormatFeatureFlagBits::eTransferDst |
vk::FormatFeatureFlagBits::eTransferSrc;
usage = VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | VK_FORMAT_FEATURE_TRANSFER_DST_BIT |
VK_FORMAT_FEATURE_TRANSFER_SRC_BIT;
if (attachable) {
usage |= IsZetaFormat(pixel_format) ? vk::FormatFeatureFlagBits::eDepthStencilAttachment
: vk::FormatFeatureFlagBits::eColorAttachment;
usage |= IsZetaFormat(pixel_format) ? VK_FORMAT_FEATURE_DEPTH_STENCIL_ATTACHMENT_BIT
: VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT;
}
if (storage) {
usage |= vk::FormatFeatureFlagBits::eStorageImage;
usage |= VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT;
}
}
return {device.GetSupportedFormat(tuple.format, usage, format_type), attachable, storage};
}
vk::ShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage) {
VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage) {
switch (stage) {
case Tegra::Engines::ShaderType::Vertex:
return vk::ShaderStageFlagBits::eVertex;
return VK_SHADER_STAGE_VERTEX_BIT;
case Tegra::Engines::ShaderType::TesselationControl:
return vk::ShaderStageFlagBits::eTessellationControl;
return VK_SHADER_STAGE_TESSELLATION_CONTROL_BIT;
case Tegra::Engines::ShaderType::TesselationEval:
return vk::ShaderStageFlagBits::eTessellationEvaluation;
return VK_SHADER_STAGE_TESSELLATION_EVALUATION_BIT;
case Tegra::Engines::ShaderType::Geometry:
return vk::ShaderStageFlagBits::eGeometry;
return VK_SHADER_STAGE_GEOMETRY_BIT;
case Tegra::Engines::ShaderType::Fragment:
return vk::ShaderStageFlagBits::eFragment;
return VK_SHADER_STAGE_FRAGMENT_BIT;
case Tegra::Engines::ShaderType::Compute:
return vk::ShaderStageFlagBits::eCompute;
return VK_SHADER_STAGE_COMPUTE_BIT;
}
UNIMPLEMENTED_MSG("Unimplemented shader stage={}", static_cast<u32>(stage));
return {};
}
vk::PrimitiveTopology PrimitiveTopology([[maybe_unused]] const VKDevice& device,
Maxwell::PrimitiveTopology topology) {
VkPrimitiveTopology PrimitiveTopology([[maybe_unused]] const VKDevice& device,
Maxwell::PrimitiveTopology topology) {
switch (topology) {
case Maxwell::PrimitiveTopology::Points:
return vk::PrimitiveTopology::ePointList;
return VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
case Maxwell::PrimitiveTopology::Lines:
return vk::PrimitiveTopology::eLineList;
return VK_PRIMITIVE_TOPOLOGY_LINE_LIST;
case Maxwell::PrimitiveTopology::LineStrip:
return vk::PrimitiveTopology::eLineStrip;
return VK_PRIMITIVE_TOPOLOGY_LINE_STRIP;
case Maxwell::PrimitiveTopology::Triangles:
return vk::PrimitiveTopology::eTriangleList;
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
case Maxwell::PrimitiveTopology::TriangleStrip:
return vk::PrimitiveTopology::eTriangleStrip;
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP;
case Maxwell::PrimitiveTopology::TriangleFan:
return vk::PrimitiveTopology::eTriangleFan;
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN;
case Maxwell::PrimitiveTopology::Quads:
// TODO(Rodrigo): Use VK_PRIMITIVE_TOPOLOGY_QUAD_LIST_EXT whenever it releases
return vk::PrimitiveTopology::eTriangleList;
return VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST;
case Maxwell::PrimitiveTopology::Patches:
return vk::PrimitiveTopology::ePatchList;
return VK_PRIMITIVE_TOPOLOGY_PATCH_LIST;
default:
UNIMPLEMENTED_MSG("Unimplemented topology={}", static_cast<u32>(topology));
return {};
}
}
vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) {
VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size) {
switch (type) {
case Maxwell::VertexAttribute::Type::SignedNorm:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Snorm;
return VK_FORMAT_R8_SNORM;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Snorm;
return VK_FORMAT_R8G8_SNORM;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Snorm;
return VK_FORMAT_R8G8B8_SNORM;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Snorm;
return VK_FORMAT_R8G8B8A8_SNORM;
case Maxwell::VertexAttribute::Size::Size_16:
return vk::Format::eR16Snorm;
return VK_FORMAT_R16_SNORM;
case Maxwell::VertexAttribute::Size::Size_16_16:
return vk::Format::eR16G16Snorm;
return VK_FORMAT_R16G16_SNORM;
case Maxwell::VertexAttribute::Size::Size_16_16_16:
return vk::Format::eR16G16B16Snorm;
return VK_FORMAT_R16G16B16_SNORM;
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Snorm;
return VK_FORMAT_R16G16B16A16_SNORM;
case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
return vk::Format::eA2B10G10R10SnormPack32;
return VK_FORMAT_A2B10G10R10_SNORM_PACK32;
default:
break;
}
@ -319,23 +322,23 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
case Maxwell::VertexAttribute::Type::UnsignedNorm:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Unorm;
return VK_FORMAT_R8_UNORM;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Unorm;
return VK_FORMAT_R8G8_UNORM;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Unorm;
return VK_FORMAT_R8G8B8_UNORM;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Unorm;
return VK_FORMAT_R8G8B8A8_UNORM;
case Maxwell::VertexAttribute::Size::Size_16:
return vk::Format::eR16Unorm;
return VK_FORMAT_R16_UNORM;
case Maxwell::VertexAttribute::Size::Size_16_16:
return vk::Format::eR16G16Unorm;
return VK_FORMAT_R16G16_UNORM;
case Maxwell::VertexAttribute::Size::Size_16_16_16:
return vk::Format::eR16G16B16Unorm;
return VK_FORMAT_R16G16B16_UNORM;
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Unorm;
return VK_FORMAT_R16G16B16A16_UNORM;
case Maxwell::VertexAttribute::Size::Size_10_10_10_2:
return vk::Format::eA2B10G10R10UnormPack32;
return VK_FORMAT_A2B10G10R10_UNORM_PACK32;
default:
break;
}
@ -343,59 +346,59 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
case Maxwell::VertexAttribute::Type::SignedInt:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Sint;
return VK_FORMAT_R16G16B16A16_SINT;
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Sint;
return VK_FORMAT_R8_SINT;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Sint;
return VK_FORMAT_R8G8_SINT;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Sint;
return VK_FORMAT_R8G8B8_SINT;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Sint;
return VK_FORMAT_R8G8B8A8_SINT;
case Maxwell::VertexAttribute::Size::Size_32:
return vk::Format::eR32Sint;
return VK_FORMAT_R32_SINT;
default:
break;
}
case Maxwell::VertexAttribute::Type::UnsignedInt:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Uint;
return VK_FORMAT_R8_UINT;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Uint;
return VK_FORMAT_R8G8_UINT;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Uint;
return VK_FORMAT_R8G8B8_UINT;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Uint;
return VK_FORMAT_R8G8B8A8_UINT;
case Maxwell::VertexAttribute::Size::Size_32:
return vk::Format::eR32Uint;
return VK_FORMAT_R32_UINT;
case Maxwell::VertexAttribute::Size::Size_32_32:
return vk::Format::eR32G32Uint;
return VK_FORMAT_R32G32_UINT;
case Maxwell::VertexAttribute::Size::Size_32_32_32:
return vk::Format::eR32G32B32Uint;
return VK_FORMAT_R32G32B32_UINT;
case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
return vk::Format::eR32G32B32A32Uint;
return VK_FORMAT_R32G32B32A32_UINT;
default:
break;
}
case Maxwell::VertexAttribute::Type::UnsignedScaled:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Uscaled;
return VK_FORMAT_R8_USCALED;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Uscaled;
return VK_FORMAT_R8G8_USCALED;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Uscaled;
return VK_FORMAT_R8G8B8_USCALED;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Uscaled;
return VK_FORMAT_R8G8B8A8_USCALED;
case Maxwell::VertexAttribute::Size::Size_16:
return vk::Format::eR16Uscaled;
return VK_FORMAT_R16_USCALED;
case Maxwell::VertexAttribute::Size::Size_16_16:
return vk::Format::eR16G16Uscaled;
return VK_FORMAT_R16G16_USCALED;
case Maxwell::VertexAttribute::Size::Size_16_16_16:
return vk::Format::eR16G16B16Uscaled;
return VK_FORMAT_R16G16B16_USCALED;
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Uscaled;
return VK_FORMAT_R16G16B16A16_USCALED;
default:
break;
}
@ -403,21 +406,21 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
case Maxwell::VertexAttribute::Type::SignedScaled:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_8:
return vk::Format::eR8Sscaled;
return VK_FORMAT_R8_SSCALED;
case Maxwell::VertexAttribute::Size::Size_8_8:
return vk::Format::eR8G8Sscaled;
return VK_FORMAT_R8G8_SSCALED;
case Maxwell::VertexAttribute::Size::Size_8_8_8:
return vk::Format::eR8G8B8Sscaled;
return VK_FORMAT_R8G8B8_SSCALED;
case Maxwell::VertexAttribute::Size::Size_8_8_8_8:
return vk::Format::eR8G8B8A8Sscaled;
return VK_FORMAT_R8G8B8A8_SSCALED;
case Maxwell::VertexAttribute::Size::Size_16:
return vk::Format::eR16Sscaled;
return VK_FORMAT_R16_SSCALED;
case Maxwell::VertexAttribute::Size::Size_16_16:
return vk::Format::eR16G16Sscaled;
return VK_FORMAT_R16G16_SSCALED;
case Maxwell::VertexAttribute::Size::Size_16_16_16:
return vk::Format::eR16G16B16Sscaled;
return VK_FORMAT_R16G16B16_SSCALED;
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Sscaled;
return VK_FORMAT_R16G16B16A16_SSCALED;
default:
break;
}
@ -425,21 +428,21 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
case Maxwell::VertexAttribute::Type::Float:
switch (size) {
case Maxwell::VertexAttribute::Size::Size_32:
return vk::Format::eR32Sfloat;
return VK_FORMAT_R32_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_32_32:
return vk::Format::eR32G32Sfloat;
return VK_FORMAT_R32G32_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_32_32_32:
return vk::Format::eR32G32B32Sfloat;
return VK_FORMAT_R32G32B32_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_32_32_32_32:
return vk::Format::eR32G32B32A32Sfloat;
return VK_FORMAT_R32G32B32A32_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_16:
return vk::Format::eR16Sfloat;
return VK_FORMAT_R16_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_16_16:
return vk::Format::eR16G16Sfloat;
return VK_FORMAT_R16G16_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_16_16_16:
return vk::Format::eR16G16B16Sfloat;
return VK_FORMAT_R16G16B16_SFLOAT;
case Maxwell::VertexAttribute::Size::Size_16_16_16_16:
return vk::Format::eR16G16B16A16Sfloat;
return VK_FORMAT_R16G16B16A16_SFLOAT;
default:
break;
}
@ -450,210 +453,210 @@ vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttr
return {};
}
vk::CompareOp ComparisonOp(Maxwell::ComparisonOp comparison) {
VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison) {
switch (comparison) {
case Maxwell::ComparisonOp::Never:
case Maxwell::ComparisonOp::NeverOld:
return vk::CompareOp::eNever;
return VK_COMPARE_OP_NEVER;
case Maxwell::ComparisonOp::Less:
case Maxwell::ComparisonOp::LessOld:
return vk::CompareOp::eLess;
return VK_COMPARE_OP_LESS;
case Maxwell::ComparisonOp::Equal:
case Maxwell::ComparisonOp::EqualOld:
return vk::CompareOp::eEqual;
return VK_COMPARE_OP_EQUAL;
case Maxwell::ComparisonOp::LessEqual:
case Maxwell::ComparisonOp::LessEqualOld:
return vk::CompareOp::eLessOrEqual;
return VK_COMPARE_OP_LESS_OR_EQUAL;
case Maxwell::ComparisonOp::Greater:
case Maxwell::ComparisonOp::GreaterOld:
return vk::CompareOp::eGreater;
return VK_COMPARE_OP_GREATER;
case Maxwell::ComparisonOp::NotEqual:
case Maxwell::ComparisonOp::NotEqualOld:
return vk::CompareOp::eNotEqual;
return VK_COMPARE_OP_NOT_EQUAL;
case Maxwell::ComparisonOp::GreaterEqual:
case Maxwell::ComparisonOp::GreaterEqualOld:
return vk::CompareOp::eGreaterOrEqual;
return VK_COMPARE_OP_GREATER_OR_EQUAL;
case Maxwell::ComparisonOp::Always:
case Maxwell::ComparisonOp::AlwaysOld:
return vk::CompareOp::eAlways;
return VK_COMPARE_OP_ALWAYS;
}
UNIMPLEMENTED_MSG("Unimplemented comparison op={}", static_cast<u32>(comparison));
return {};
}
vk::IndexType IndexFormat(const VKDevice& device, Maxwell::IndexFormat index_format) {
VkIndexType IndexFormat(const VKDevice& device, Maxwell::IndexFormat index_format) {
switch (index_format) {
case Maxwell::IndexFormat::UnsignedByte:
if (!device.IsExtIndexTypeUint8Supported()) {
UNIMPLEMENTED_MSG("Native uint8 indices are not supported on this device");
return vk::IndexType::eUint16;
return VK_INDEX_TYPE_UINT16;
}
return vk::IndexType::eUint8EXT;
return VK_INDEX_TYPE_UINT8_EXT;
case Maxwell::IndexFormat::UnsignedShort:
return vk::IndexType::eUint16;
return VK_INDEX_TYPE_UINT16;
case Maxwell::IndexFormat::UnsignedInt:
return vk::IndexType::eUint32;
return VK_INDEX_TYPE_UINT32;
}
UNIMPLEMENTED_MSG("Unimplemented index_format={}", static_cast<u32>(index_format));
return {};
}
vk::StencilOp StencilOp(Maxwell::StencilOp stencil_op) {
VkStencilOp StencilOp(Maxwell::StencilOp stencil_op) {
switch (stencil_op) {
case Maxwell::StencilOp::Keep:
case Maxwell::StencilOp::KeepOGL:
return vk::StencilOp::eKeep;
return VK_STENCIL_OP_KEEP;
case Maxwell::StencilOp::Zero:
case Maxwell::StencilOp::ZeroOGL:
return vk::StencilOp::eZero;
return VK_STENCIL_OP_ZERO;
case Maxwell::StencilOp::Replace:
case Maxwell::StencilOp::ReplaceOGL:
return vk::StencilOp::eReplace;
return VK_STENCIL_OP_REPLACE;
case Maxwell::StencilOp::Incr:
case Maxwell::StencilOp::IncrOGL:
return vk::StencilOp::eIncrementAndClamp;
return VK_STENCIL_OP_INCREMENT_AND_CLAMP;
case Maxwell::StencilOp::Decr:
case Maxwell::StencilOp::DecrOGL:
return vk::StencilOp::eDecrementAndClamp;
return VK_STENCIL_OP_DECREMENT_AND_CLAMP;
case Maxwell::StencilOp::Invert:
case Maxwell::StencilOp::InvertOGL:
return vk::StencilOp::eInvert;
return VK_STENCIL_OP_INVERT;
case Maxwell::StencilOp::IncrWrap:
case Maxwell::StencilOp::IncrWrapOGL:
return vk::StencilOp::eIncrementAndWrap;
return VK_STENCIL_OP_INCREMENT_AND_WRAP;
case Maxwell::StencilOp::DecrWrap:
case Maxwell::StencilOp::DecrWrapOGL:
return vk::StencilOp::eDecrementAndWrap;
return VK_STENCIL_OP_DECREMENT_AND_WRAP;
}
UNIMPLEMENTED_MSG("Unimplemented stencil op={}", static_cast<u32>(stencil_op));
return {};
}
vk::BlendOp BlendEquation(Maxwell::Blend::Equation equation) {
VkBlendOp BlendEquation(Maxwell::Blend::Equation equation) {
switch (equation) {
case Maxwell::Blend::Equation::Add:
case Maxwell::Blend::Equation::AddGL:
return vk::BlendOp::eAdd;
return VK_BLEND_OP_ADD;
case Maxwell::Blend::Equation::Subtract:
case Maxwell::Blend::Equation::SubtractGL:
return vk::BlendOp::eSubtract;
return VK_BLEND_OP_SUBTRACT;
case Maxwell::Blend::Equation::ReverseSubtract:
case Maxwell::Blend::Equation::ReverseSubtractGL:
return vk::BlendOp::eReverseSubtract;
return VK_BLEND_OP_REVERSE_SUBTRACT;
case Maxwell::Blend::Equation::Min:
case Maxwell::Blend::Equation::MinGL:
return vk::BlendOp::eMin;
return VK_BLEND_OP_MIN;
case Maxwell::Blend::Equation::Max:
case Maxwell::Blend::Equation::MaxGL:
return vk::BlendOp::eMax;
return VK_BLEND_OP_MAX;
}
UNIMPLEMENTED_MSG("Unimplemented blend equation={}", static_cast<u32>(equation));
return {};
}
vk::BlendFactor BlendFactor(Maxwell::Blend::Factor factor) {
VkBlendFactor BlendFactor(Maxwell::Blend::Factor factor) {
switch (factor) {
case Maxwell::Blend::Factor::Zero:
case Maxwell::Blend::Factor::ZeroGL:
return vk::BlendFactor::eZero;
return VK_BLEND_FACTOR_ZERO;
case Maxwell::Blend::Factor::One:
case Maxwell::Blend::Factor::OneGL:
return vk::BlendFactor::eOne;
return VK_BLEND_FACTOR_ONE;
case Maxwell::Blend::Factor::SourceColor:
case Maxwell::Blend::Factor::SourceColorGL:
return vk::BlendFactor::eSrcColor;
return VK_BLEND_FACTOR_SRC_COLOR;
case Maxwell::Blend::Factor::OneMinusSourceColor:
case Maxwell::Blend::Factor::OneMinusSourceColorGL:
return vk::BlendFactor::eOneMinusSrcColor;
return VK_BLEND_FACTOR_ONE_MINUS_SRC_COLOR;
case Maxwell::Blend::Factor::SourceAlpha:
case Maxwell::Blend::Factor::SourceAlphaGL:
return vk::BlendFactor::eSrcAlpha;
return VK_BLEND_FACTOR_SRC_ALPHA;
case Maxwell::Blend::Factor::OneMinusSourceAlpha:
case Maxwell::Blend::Factor::OneMinusSourceAlphaGL:
return vk::BlendFactor::eOneMinusSrcAlpha;
return VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA;
case Maxwell::Blend::Factor::DestAlpha:
case Maxwell::Blend::Factor::DestAlphaGL:
return vk::BlendFactor::eDstAlpha;
return VK_BLEND_FACTOR_DST_ALPHA;
case Maxwell::Blend::Factor::OneMinusDestAlpha:
case Maxwell::Blend::Factor::OneMinusDestAlphaGL:
return vk::BlendFactor::eOneMinusDstAlpha;
return VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA;
case Maxwell::Blend::Factor::DestColor:
case Maxwell::Blend::Factor::DestColorGL:
return vk::BlendFactor::eDstColor;
return VK_BLEND_FACTOR_DST_COLOR;
case Maxwell::Blend::Factor::OneMinusDestColor:
case Maxwell::Blend::Factor::OneMinusDestColorGL:
return vk::BlendFactor::eOneMinusDstColor;
return VK_BLEND_FACTOR_ONE_MINUS_DST_COLOR;
case Maxwell::Blend::Factor::SourceAlphaSaturate:
case Maxwell::Blend::Factor::SourceAlphaSaturateGL:
return vk::BlendFactor::eSrcAlphaSaturate;
return VK_BLEND_FACTOR_SRC_ALPHA_SATURATE;
case Maxwell::Blend::Factor::Source1Color:
case Maxwell::Blend::Factor::Source1ColorGL:
return vk::BlendFactor::eSrc1Color;
return VK_BLEND_FACTOR_SRC1_COLOR;
case Maxwell::Blend::Factor::OneMinusSource1Color:
case Maxwell::Blend::Factor::OneMinusSource1ColorGL:
return vk::BlendFactor::eOneMinusSrc1Color;
return VK_BLEND_FACTOR_ONE_MINUS_SRC1_COLOR;
case Maxwell::Blend::Factor::Source1Alpha:
case Maxwell::Blend::Factor::Source1AlphaGL:
return vk::BlendFactor::eSrc1Alpha;
return VK_BLEND_FACTOR_SRC1_ALPHA;
case Maxwell::Blend::Factor::OneMinusSource1Alpha:
case Maxwell::Blend::Factor::OneMinusSource1AlphaGL:
return vk::BlendFactor::eOneMinusSrc1Alpha;
return VK_BLEND_FACTOR_ONE_MINUS_SRC1_ALPHA;
case Maxwell::Blend::Factor::ConstantColor:
case Maxwell::Blend::Factor::ConstantColorGL:
return vk::BlendFactor::eConstantColor;
return VK_BLEND_FACTOR_CONSTANT_COLOR;
case Maxwell::Blend::Factor::OneMinusConstantColor:
case Maxwell::Blend::Factor::OneMinusConstantColorGL:
return vk::BlendFactor::eOneMinusConstantColor;
return VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR;
case Maxwell::Blend::Factor::ConstantAlpha:
case Maxwell::Blend::Factor::ConstantAlphaGL:
return vk::BlendFactor::eConstantAlpha;
return VK_BLEND_FACTOR_CONSTANT_ALPHA;
case Maxwell::Blend::Factor::OneMinusConstantAlpha:
case Maxwell::Blend::Factor::OneMinusConstantAlphaGL:
return vk::BlendFactor::eOneMinusConstantAlpha;
return VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA;
}
UNIMPLEMENTED_MSG("Unimplemented blend factor={}", static_cast<u32>(factor));
return {};
}
vk::FrontFace FrontFace(Maxwell::FrontFace front_face) {
VkFrontFace FrontFace(Maxwell::FrontFace front_face) {
switch (front_face) {
case Maxwell::FrontFace::ClockWise:
return vk::FrontFace::eClockwise;
return VK_FRONT_FACE_CLOCKWISE;
case Maxwell::FrontFace::CounterClockWise:
return vk::FrontFace::eCounterClockwise;
return VK_FRONT_FACE_COUNTER_CLOCKWISE;
}
UNIMPLEMENTED_MSG("Unimplemented front face={}", static_cast<u32>(front_face));
return {};
}
vk::CullModeFlags CullFace(Maxwell::CullFace cull_face) {
VkCullModeFlags CullFace(Maxwell::CullFace cull_face) {
switch (cull_face) {
case Maxwell::CullFace::Front:
return vk::CullModeFlagBits::eFront;
return VK_CULL_MODE_FRONT_BIT;
case Maxwell::CullFace::Back:
return vk::CullModeFlagBits::eBack;
return VK_CULL_MODE_BACK_BIT;
case Maxwell::CullFace::FrontAndBack:
return vk::CullModeFlagBits::eFrontAndBack;
return VK_CULL_MODE_FRONT_AND_BACK;
}
UNIMPLEMENTED_MSG("Unimplemented cull face={}", static_cast<u32>(cull_face));
return {};
}
vk::ComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle) {
VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle) {
switch (swizzle) {
case Tegra::Texture::SwizzleSource::Zero:
return vk::ComponentSwizzle::eZero;
return VK_COMPONENT_SWIZZLE_ZERO;
case Tegra::Texture::SwizzleSource::R:
return vk::ComponentSwizzle::eR;
return VK_COMPONENT_SWIZZLE_R;
case Tegra::Texture::SwizzleSource::G:
return vk::ComponentSwizzle::eG;
return VK_COMPONENT_SWIZZLE_G;
case Tegra::Texture::SwizzleSource::B:
return vk::ComponentSwizzle::eB;
return VK_COMPONENT_SWIZZLE_B;
case Tegra::Texture::SwizzleSource::A:
return vk::ComponentSwizzle::eA;
return VK_COMPONENT_SWIZZLE_A;
case Tegra::Texture::SwizzleSource::OneInt:
case Tegra::Texture::SwizzleSource::OneFloat:
return vk::ComponentSwizzle::eOne;
return VK_COMPONENT_SWIZZLE_ONE;
}
UNIMPLEMENTED_MSG("Unimplemented swizzle source={}", static_cast<u32>(swizzle));
return {};

37
src/video_core/renderer_vulkan/maxwell_to_vk.h
View file

@ -6,8 +6,8 @@
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/surface.h"
#include "video_core/textures/texture.h"
@ -18,46 +18,45 @@ using PixelFormat = VideoCore::Surface::PixelFormat;
namespace Sampler {
vk::Filter Filter(Tegra::Texture::TextureFilter filter);
VkFilter Filter(Tegra::Texture::TextureFilter filter);
vk::SamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter);
VkSamplerMipmapMode MipmapMode(Tegra::Texture::TextureMipmapFilter mipmap_filter);
vk::SamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
Tegra::Texture::TextureFilter filter);
VkSamplerAddressMode WrapMode(const VKDevice& device, Tegra::Texture::WrapMode wrap_mode,
Tegra::Texture::TextureFilter filter);
vk::CompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func);
VkCompareOp DepthCompareFunction(Tegra::Texture::DepthCompareFunc depth_compare_func);
} // namespace Sampler
struct FormatInfo {
vk::Format format;
VkFormat format;
bool attachable;
bool storage;
};
FormatInfo SurfaceFormat(const VKDevice& device, FormatType format_type, PixelFormat pixel_format);
vk::ShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage);
VkShaderStageFlagBits ShaderStage(Tegra::Engines::ShaderType stage);
vk::PrimitiveTopology PrimitiveTopology(const VKDevice& device,
Maxwell::PrimitiveTopology topology);
VkPrimitiveTopology PrimitiveTopology(const VKDevice& device, Maxwell::PrimitiveTopology topology);
vk::Format VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size);
VkFormat VertexFormat(Maxwell::VertexAttribute::Type type, Maxwell::VertexAttribute::Size size);
vk::CompareOp ComparisonOp(Maxwell::ComparisonOp comparison);
VkCompareOp ComparisonOp(Maxwell::ComparisonOp comparison);
vk::IndexType IndexFormat(const VKDevice& device, Maxwell::IndexFormat index_format);
VkIndexType IndexFormat(const VKDevice& device, Maxwell::IndexFormat index_format);
vk::StencilOp StencilOp(Maxwell::StencilOp stencil_op);
VkStencilOp StencilOp(Maxwell::StencilOp stencil_op);
vk::BlendOp BlendEquation(Maxwell::Blend::Equation equation);
VkBlendOp BlendEquation(Maxwell::Blend::Equation equation);
vk::BlendFactor BlendFactor(Maxwell::Blend::Factor factor);
VkBlendFactor BlendFactor(Maxwell::Blend::Factor factor);
vk::FrontFace FrontFace(Maxwell::FrontFace front_face);
VkFrontFace FrontFace(Maxwell::FrontFace front_face);
vk::CullModeFlags CullFace(Maxwell::CullFace cull_face);
VkCullModeFlags CullFace(Maxwell::CullFace cull_face);
vk::ComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle);
VkComponentSwizzle SwizzleSource(Tegra::Texture::SwizzleSource swizzle);
} // namespace Vulkan::MaxwellToVK

159
src/video_core/renderer_vulkan/renderer_vulkan.cpp
View file

@ -24,7 +24,6 @@
#include "core/settings.h"
#include "core/telemetry_session.h"
#include "video_core/gpu.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/renderer_vulkan.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_device.h"
@ -34,8 +33,9 @@
#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/wrapper.h"
// Include these late to avoid changing Vulkan-Hpp's dynamic dispatcher size
// Include these late to avoid polluting previous headers
#ifdef _WIN32
#include <windows.h>
// ensure include order
@ -54,20 +54,19 @@ namespace {
using Core::Frontend::WindowSystemType;
VkBool32 DebugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity_,
VkBool32 DebugCallback(VkDebugUtilsMessageSeverityFlagBitsEXT severity,
VkDebugUtilsMessageTypeFlagsEXT type,
const VkDebugUtilsMessengerCallbackDataEXT* data,
[[maybe_unused]] void* user_data) {
const auto severity{static_cast<vk::DebugUtilsMessageSeverityFlagBitsEXT>(severity_)};
const char* message{data->pMessage};
if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eError) {
if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_ERROR_BIT_EXT) {
LOG_CRITICAL(Render_Vulkan, "{}", message);
} else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning) {
} else if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_WARNING_BIT_EXT) {
LOG_WARNING(Render_Vulkan, "{}", message);
} else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo) {
} else if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_INFO_BIT_EXT) {
LOG_INFO(Render_Vulkan, "{}", message);
} else if (severity & vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose) {
} else if (severity & VK_DEBUG_UTILS_MESSAGE_SEVERITY_VERBOSE_BIT_EXT) {
LOG_DEBUG(Render_Vulkan, "{}", message);
}
return VK_FALSE;
@ -94,22 +93,24 @@ Common::DynamicLibrary OpenVulkanLibrary() {
return library;
}
UniqueInstance CreateInstance(Common::DynamicLibrary& library, vk::DispatchLoaderDynamic& dld,
WindowSystemType window_type = WindowSystemType::Headless,
bool enable_layers = false) {
vk::Instance CreateInstance(Common::DynamicLibrary& library, vk::InstanceDispatch& dld,
WindowSystemType window_type = WindowSystemType::Headless,
bool enable_layers = false) {
if (!library.IsOpen()) {
LOG_ERROR(Render_Vulkan, "Vulkan library not available");
return UniqueInstance{};
return {};
}
PFN_vkGetInstanceProcAddr vkGetInstanceProcAddr;
if (!library.GetSymbol("vkGetInstanceProcAddr", &vkGetInstanceProcAddr)) {
if (!library.GetSymbol("vkGetInstanceProcAddr", &dld.vkGetInstanceProcAddr)) {
LOG_ERROR(Render_Vulkan, "vkGetInstanceProcAddr not present in Vulkan");
return UniqueInstance{};
return {};
}
if (!vk::Load(dld)) {
LOG_ERROR(Render_Vulkan, "Failed to load Vulkan function pointers");
return {};
}
dld.init(vkGetInstanceProcAddr);
std::vector<const char*> extensions;
extensions.reserve(4);
extensions.reserve(6);
switch (window_type) {
case Core::Frontend::WindowSystemType::Headless:
break;
@ -136,45 +137,39 @@ UniqueInstance CreateInstance(Common::DynamicLibrary& library, vk::DispatchLoade
if (enable_layers) {
extensions.push_back(VK_EXT_DEBUG_UTILS_EXTENSION_NAME);
}
extensions.push_back(VK_KHR_GET_PHYSICAL_DEVICE_PROPERTIES_2_EXTENSION_NAME);
u32 num_properties;
if (vk::enumerateInstanceExtensionProperties(nullptr, &num_properties, nullptr, dld) !=
vk::Result::eSuccess) {
LOG_ERROR(Render_Vulkan, "Failed to query number of extension properties");
return UniqueInstance{};
}
std::vector<vk::ExtensionProperties> properties(num_properties);
if (vk::enumerateInstanceExtensionProperties(nullptr, &num_properties, properties.data(),
dld) != vk::Result::eSuccess) {
const std::optional properties = vk::EnumerateInstanceExtensionProperties(dld);
if (!properties) {
LOG_ERROR(Render_Vulkan, "Failed to query extension properties");
return UniqueInstance{};
return {};
}
for (const char* extension : extensions) {
const auto it =
std::find_if(properties.begin(), properties.end(), [extension](const auto& prop) {
std::find_if(properties->begin(), properties->end(), [extension](const auto& prop) {
return !std::strcmp(extension, prop.extensionName);
});
if (it == properties.end()) {
if (it == properties->end()) {
LOG_ERROR(Render_Vulkan, "Required instance extension {} is not available", extension);
return UniqueInstance{};
return {};
}
}
const vk::ApplicationInfo application_info("yuzu Emulator", VK_MAKE_VERSION(0, 1, 0),
"yuzu Emulator", VK_MAKE_VERSION(0, 1, 0),
VK_API_VERSION_1_1);
const std::array layers = {"VK_LAYER_LUNARG_standard_validation"};
const vk::InstanceCreateInfo instance_ci(
{}, &application_info, enable_layers ? static_cast<u32>(layers.size()) : 0, layers.data(),
static_cast<u32>(extensions.size()), extensions.data());
vk::Instance unsafe_instance;
if (vk::createInstance(&instance_ci, nullptr, &unsafe_instance, dld) != vk::Result::eSuccess) {
LOG_ERROR(Render_Vulkan, "Failed to create Vulkan instance");
return UniqueInstance{};
static constexpr std::array layers_data{"VK_LAYER_LUNARG_standard_validation"};
vk::Span<const char*> layers = layers_data;
if (!enable_layers) {
layers = {};
}
dld.init(unsafe_instance);
return UniqueInstance(unsafe_instance, {nullptr, dld});
vk::Instance instance = vk::Instance::Create(layers, extensions, dld);
if (!instance) {
LOG_ERROR(Render_Vulkan, "Failed to create Vulkan instance");
return {};
}
if (!vk::Load(*instance, dld)) {
LOG_ERROR(Render_Vulkan, "Failed to load Vulkan instance function pointers");
}
return instance;
}
std::string GetReadableVersion(u32 version) {
@ -187,14 +182,14 @@ std::string GetDriverVersion(const VKDevice& device) {
// https://github.com/SaschaWillems/vulkan.gpuinfo.org/blob/5dddea46ea1120b0df14eef8f15ff8e318e35462/functions.php#L308-L314
const u32 version = device.GetDriverVersion();
if (device.GetDriverID() == vk::DriverIdKHR::eNvidiaProprietary) {
if (device.GetDriverID() == VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR) {
const u32 major = (version >> 22) & 0x3ff;
const u32 minor = (version >> 14) & 0x0ff;
const u32 secondary = (version >> 6) & 0x0ff;
const u32 tertiary = version & 0x003f;
return fmt::format("{}.{}.{}.{}", major, minor, secondary, tertiary);
}
if (device.GetDriverID() == vk::DriverIdKHR::eIntelProprietaryWindows) {
if (device.GetDriverID() == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR) {
const u32 major = version >> 14;
const u32 minor = version & 0x3fff;
return fmt::format("{}.{}", major, minor);
@ -307,10 +302,8 @@ void RendererVulkan::ShutDown() {
if (!device) {
return;
}
const auto dev = device->GetLogical();
const auto& dld = device->GetDispatchLoader();
if (dev && dld.vkDeviceWaitIdle) {
dev.waitIdle(dld);
if (const auto& dev = device->GetLogical()) {
dev.WaitIdle();
}
rasterizer.reset();
@ -326,23 +319,11 @@ bool RendererVulkan::CreateDebugCallback() {
if (!Settings::values.renderer_debug) {
return true;
}
const vk::DebugUtilsMessengerCreateInfoEXT callback_ci(
{},
vk::DebugUtilsMessageSeverityFlagBitsEXT::eError |
vk::DebugUtilsMessageSeverityFlagBitsEXT::eWarning |
vk::DebugUtilsMessageSeverityFlagBitsEXT::eInfo |
vk::DebugUtilsMessageSeverityFlagBitsEXT::eVerbose,
vk::DebugUtilsMessageTypeFlagBitsEXT::eGeneral |
vk::DebugUtilsMessageTypeFlagBitsEXT::eValidation |
vk::DebugUtilsMessageTypeFlagBitsEXT::ePerformance,
&DebugCallback, nullptr);
vk::DebugUtilsMessengerEXT unsafe_callback;
if (instance->createDebugUtilsMessengerEXT(&callback_ci, nullptr, &unsafe_callback, dld) !=
vk::Result::eSuccess) {
debug_callback = instance.TryCreateDebugCallback(DebugCallback);
if (!debug_callback) {
LOG_ERROR(Render_Vulkan, "Failed to create debug callback");
return false;
}
debug_callback = UniqueDebugUtilsMessengerEXT(unsafe_callback, {*instance, nullptr, dld});
return true;
}
@ -357,8 +338,8 @@ bool RendererVulkan::CreateSurface() {
nullptr, 0, nullptr, hWnd};
const auto vkCreateWin32SurfaceKHR = reinterpret_cast<PFN_vkCreateWin32SurfaceKHR>(
dld.vkGetInstanceProcAddr(*instance, "vkCreateWin32SurfaceKHR"));
if (!vkCreateWin32SurfaceKHR || vkCreateWin32SurfaceKHR(instance.get(), &win32_ci, nullptr,
&unsafe_surface) != VK_SUCCESS) {
if (!vkCreateWin32SurfaceKHR ||
vkCreateWin32SurfaceKHR(*instance, &win32_ci, nullptr, &unsafe_surface) != VK_SUCCESS) {
LOG_ERROR(Render_Vulkan, "Failed to initialize Win32 surface");
return false;
}
@ -372,8 +353,8 @@ bool RendererVulkan::CreateSurface() {
reinterpret_cast<Window>(window_info.render_surface)};
const auto vkCreateXlibSurfaceKHR = reinterpret_cast<PFN_vkCreateXlibSurfaceKHR>(
dld.vkGetInstanceProcAddr(*instance, "vkCreateXlibSurfaceKHR"));
if (!vkCreateXlibSurfaceKHR || vkCreateXlibSurfaceKHR(instance.get(), &xlib_ci, nullptr,
&unsafe_surface) != VK_SUCCESS) {
if (!vkCreateXlibSurfaceKHR ||
vkCreateXlibSurfaceKHR(*instance, &xlib_ci, nullptr, &unsafe_surface) != VK_SUCCESS) {
LOG_ERROR(Render_Vulkan, "Failed to initialize Xlib surface");
return false;
}
@ -386,7 +367,7 @@ bool RendererVulkan::CreateSurface() {
const auto vkCreateWaylandSurfaceKHR = reinterpret_cast<PFN_vkCreateWaylandSurfaceKHR>(
dld.vkGetInstanceProcAddr(*instance, "vkCreateWaylandSurfaceKHR"));
if (!vkCreateWaylandSurfaceKHR ||
vkCreateWaylandSurfaceKHR(instance.get(), &wayland_ci, nullptr, &unsafe_surface) !=
vkCreateWaylandSurfaceKHR(*instance, &wayland_ci, nullptr, &unsafe_surface) !=
VK_SUCCESS) {
LOG_ERROR(Render_Vulkan, "Failed to initialize Wayland surface");
return false;
@ -398,26 +379,30 @@ bool RendererVulkan::CreateSurface() {
return false;
}
surface = UniqueSurfaceKHR(unsafe_surface, {*instance, nullptr, dld});
surface = vk::SurfaceKHR(unsafe_surface, *instance, dld);
return true;
}
bool RendererVulkan::PickDevices() {
const auto devices = instance->enumeratePhysicalDevices(dld);
const auto devices = instance.EnumeratePhysicalDevices();
if (!devices) {
LOG_ERROR(Render_Vulkan, "Failed to enumerate physical devices");
return false;
}
const s32 device_index = Settings::values.vulkan_device;
if (device_index < 0 || device_index >= static_cast<s32>(devices.size())) {
if (device_index < 0 || device_index >= static_cast<s32>(devices->size())) {
LOG_ERROR(Render_Vulkan, "Invalid device index {}!", device_index);
return false;
}
const vk::PhysicalDevice physical_device = devices[static_cast<std::size_t>(device_index)];
if (!VKDevice::IsSuitable(physical_device, *surface, dld)) {
const vk::PhysicalDevice physical_device((*devices)[static_cast<std::size_t>(device_index)],
dld);
if (!VKDevice::IsSuitable(physical_device, *surface)) {
return false;
}
device = std::make_unique<VKDevice>(dld, physical_device, *surface);
return device->Create(*instance);
device = std::make_unique<VKDevice>(*instance, physical_device, *surface, dld);
return device->Create();
}
void RendererVulkan::Report() const {
@ -444,30 +429,22 @@ void RendererVulkan::Report() const {
}
std::vector<std::string> RendererVulkan::EnumerateDevices() {
// Avoid putting DispatchLoaderDynamic, it's too large
auto dld_memory = std::make_unique<vk::DispatchLoaderDynamic>();
auto& dld = *dld_memory;
vk::InstanceDispatch dld;
Common::DynamicLibrary library = OpenVulkanLibrary();
UniqueInstance instance = CreateInstance(library, dld);
vk::Instance instance = CreateInstance(library, dld);
if (!instance) {
return {};
}
u32 num_devices;
if (instance->enumeratePhysicalDevices(&num_devices, nullptr, dld) != vk::Result::eSuccess) {
return {};
}
std::vector<vk::PhysicalDevice> devices(num_devices);
if (instance->enumeratePhysicalDevices(&num_devices, devices.data(), dld) !=
vk::Result::eSuccess) {
const std::optional physical_devices = instance.EnumeratePhysicalDevices();
if (!physical_devices) {
return {};
}
std::vector<std::string> names;
names.reserve(num_devices);
for (auto& device : devices) {
names.push_back(device.getProperties(dld).deviceName);
names.reserve(physical_devices->size());
for (const auto& device : *physical_devices) {
names.push_back(vk::PhysicalDevice(device, dld).GetProperties().deviceName);
}
return names;
}

10
src/video_core/renderer_vulkan/renderer_vulkan.h
View file

@ -12,7 +12,7 @@
#include "common/dynamic_library.h"
#include "video_core/renderer_base.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Core {
class System;
@ -61,14 +61,14 @@ private:
Core::System& system;
Common::DynamicLibrary library;
vk::DispatchLoaderDynamic dld;
vk::InstanceDispatch dld;
UniqueInstance instance;
UniqueSurfaceKHR surface;
vk::Instance instance;
vk::SurfaceKHR surface;
VKScreenInfo screen_info;
UniqueDebugUtilsMessengerEXT debug_callback;
vk::DebugCallback debug_callback;
std::unique_ptr<VKDevice> device;
std::unique_ptr<VKSwapchain> swapchain;
std::unique_ptr<VKMemoryManager> memory_manager;

577
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View file

@ -20,7 +20,6 @@
#include "video_core/gpu.h"
#include "video_core/morton.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/renderer_vulkan.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_device.h"
@ -30,6 +29,7 @@
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/vk_swapchain.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/surface.h"
namespace Vulkan {
@ -140,16 +140,25 @@ struct ScreenRectVertex {
std::array<f32, 2> position;
std::array<f32, 2> tex_coord;
static vk::VertexInputBindingDescription GetDescription() {
return vk::VertexInputBindingDescription(0, sizeof(ScreenRectVertex),
vk::VertexInputRate::eVertex);
static VkVertexInputBindingDescription GetDescription() {
VkVertexInputBindingDescription description;
description.binding = 0;
description.stride = sizeof(ScreenRectVertex);
description.inputRate = VK_VERTEX_INPUT_RATE_VERTEX;
return description;
}
static std::array<vk::VertexInputAttributeDescription, 2> GetAttributes() {
return {vk::VertexInputAttributeDescription(0, 0, vk::Format::eR32G32Sfloat,
offsetof(ScreenRectVertex, position)),
vk::VertexInputAttributeDescription(1, 0, vk::Format::eR32G32Sfloat,
offsetof(ScreenRectVertex, tex_coord))};
static std::array<VkVertexInputAttributeDescription, 2> GetAttributes() {
std::array<VkVertexInputAttributeDescription, 2> attributes;
attributes[0].location = 0;
attributes[0].binding = 0;
attributes[0].format = VK_FORMAT_R32G32_SFLOAT;
attributes[0].offset = offsetof(ScreenRectVertex, position);
attributes[1].location = 1;
attributes[1].binding = 0;
attributes[1].format = VK_FORMAT_R32G32_SFLOAT;
attributes[1].offset = offsetof(ScreenRectVertex, tex_coord);
return attributes;
}
};
@ -172,16 +181,16 @@ std::size_t GetSizeInBytes(const Tegra::FramebufferConfig& framebuffer) {
static_cast<std::size_t>(framebuffer.height) * GetBytesPerPixel(framebuffer);
}
vk::Format GetFormat(const Tegra::FramebufferConfig& framebuffer) {
VkFormat GetFormat(const Tegra::FramebufferConfig& framebuffer) {
switch (framebuffer.pixel_format) {
case Tegra::FramebufferConfig::PixelFormat::ABGR8:
return vk::Format::eA8B8G8R8UnormPack32;
return VK_FORMAT_A8B8G8R8_UNORM_PACK32;
case Tegra::FramebufferConfig::PixelFormat::RGB565:
return vk::Format::eR5G6B5UnormPack16;
return VK_FORMAT_R5G6B5_UNORM_PACK16;
default:
UNIMPLEMENTED_MSG("Unknown framebuffer pixel format: {}",
static_cast<u32>(framebuffer.pixel_format));
return vk::Format::eA8B8G8R8UnormPack32;
return VK_FORMAT_A8B8G8R8_UNORM_PACK32;
}
}
@ -219,8 +228,8 @@ void VKBlitScreen::Recreate() {
CreateDynamicResources();
}
std::tuple<VKFence&, vk::Semaphore> VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
bool use_accelerated) {
std::tuple<VKFence&, VkSemaphore> VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
bool use_accelerated) {
RefreshResources(framebuffer);
// Finish any pending renderpass
@ -255,46 +264,76 @@ std::tuple<VKFence&, vk::Semaphore> VKBlitScreen::Draw(const Tegra::FramebufferC
framebuffer.stride, block_height_log2, framebuffer.height, 0, 1, 1,
map.GetAddress() + image_offset, host_ptr);
blit_image->Transition(0, 1, 0, 1, vk::PipelineStageFlagBits::eTransfer,
vk::AccessFlagBits::eTransferWrite,
vk::ImageLayout::eTransferDstOptimal);
blit_image->Transition(0, 1, 0, 1, VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
const vk::BufferImageCopy copy(image_offset, 0, 0,
{vk::ImageAspectFlagBits::eColor, 0, 0, 1}, {0, 0, 0},
{framebuffer.width, framebuffer.height, 1});
scheduler.Record([buffer_handle = *buffer, image = blit_image->GetHandle(),
copy](auto cmdbuf, auto& dld) {
cmdbuf.copyBufferToImage(buffer_handle, image, vk::ImageLayout::eTransferDstOptimal,
{copy}, dld);
});
VkBufferImageCopy copy;
copy.bufferOffset = image_offset;
copy.bufferRowLength = 0;
copy.bufferImageHeight = 0;
copy.imageSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
copy.imageSubresource.mipLevel = 0;
copy.imageSubresource.baseArrayLayer = 0;
copy.imageSubresource.layerCount = 1;
copy.imageOffset.x = 0;
copy.imageOffset.y = 0;
copy.imageOffset.z = 0;
copy.imageExtent.width = framebuffer.width;
copy.imageExtent.height = framebuffer.height;
copy.imageExtent.depth = 1;
scheduler.Record(
[buffer = *buffer, image = *blit_image->GetHandle(), copy](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
});
}
map.Release();
blit_image->Transition(0, 1, 0, 1, vk::PipelineStageFlagBits::eFragmentShader,
vk::AccessFlagBits::eShaderRead,
vk::ImageLayout::eShaderReadOnlyOptimal);
blit_image->Transition(0, 1, 0, 1, VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
scheduler.Record([renderpass = *renderpass, framebuffer = *framebuffers[image_index],
descriptor_set = descriptor_sets[image_index], buffer = *buffer,
size = swapchain.GetSize(), pipeline = *pipeline,
layout = *pipeline_layout](auto cmdbuf, auto& dld) {
const vk::ClearValue clear_color{std::array{0.0f, 0.0f, 0.0f, 1.0f}};
const vk::RenderPassBeginInfo renderpass_bi(renderpass, framebuffer, {{0, 0}, size}, 1,
&clear_color);
layout = *pipeline_layout](vk::CommandBuffer cmdbuf) {
VkClearValue clear_color;
clear_color.color.float32[0] = 0.0f;
clear_color.color.float32[1] = 0.0f;
clear_color.color.float32[2] = 0.0f;
clear_color.color.float32[3] = 0.0f;
cmdbuf.beginRenderPass(renderpass_bi, vk::SubpassContents::eInline, dld);
cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline, dld);
cmdbuf.setViewport(
0,
{{0.0f, 0.0f, static_cast<f32>(size.width), static_cast<f32>(size.height), 0.0f, 1.0f}},
dld);
cmdbuf.setScissor(0, {{{0, 0}, size}}, dld);
VkRenderPassBeginInfo renderpass_bi;
renderpass_bi.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
renderpass_bi.pNext = nullptr;
renderpass_bi.renderPass = renderpass;
renderpass_bi.framebuffer = framebuffer;
renderpass_bi.renderArea.offset.x = 0;
renderpass_bi.renderArea.offset.y = 0;
renderpass_bi.renderArea.extent = size;
renderpass_bi.clearValueCount = 1;
renderpass_bi.pClearValues = &clear_color;
cmdbuf.bindVertexBuffers(0, {buffer}, {offsetof(BufferData, vertices)}, dld);
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, layout, 0, {descriptor_set}, {},
dld);
cmdbuf.draw(4, 1, 0, 0, dld);
cmdbuf.endRenderPass(dld);
VkViewport viewport;
viewport.x = 0.0f;
viewport.y = 0.0f;
viewport.width = static_cast<float>(size.width);
viewport.height = static_cast<float>(size.height);
viewport.minDepth = 0.0f;
viewport.maxDepth = 1.0f;
VkRect2D scissor;
scissor.offset.x = 0;
scissor.offset.y = 0;
scissor.extent = size;
cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.SetViewport(0, viewport);
cmdbuf.SetScissor(0, scissor);
cmdbuf.BindVertexBuffer(0, buffer, offsetof(BufferData, vertices));
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set, {});
cmdbuf.Draw(4, 1, 0, 0);
cmdbuf.EndRenderPass();
});
return {scheduler.GetFence(), *semaphores[image_index]};
@ -334,165 +373,295 @@ void VKBlitScreen::CreateShaders() {
}
void VKBlitScreen::CreateSemaphores() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
semaphores.resize(image_count);
for (std::size_t i = 0; i < image_count; ++i) {
semaphores[i] = dev.createSemaphoreUnique({}, nullptr, dld);
}
std::generate(semaphores.begin(), semaphores.end(),
[this] { return device.GetLogical().CreateSemaphore(); });
}
void VKBlitScreen::CreateDescriptorPool() {
const std::array<vk::DescriptorPoolSize, 2> pool_sizes{
vk::DescriptorPoolSize{vk::DescriptorType::eUniformBuffer, static_cast<u32>(image_count)},
vk::DescriptorPoolSize{vk::DescriptorType::eCombinedImageSampler,
static_cast<u32>(image_count)}};
const vk::DescriptorPoolCreateInfo pool_ci(
{}, static_cast<u32>(image_count), static_cast<u32>(pool_sizes.size()), pool_sizes.data());
const auto dev = device.GetLogical();
descriptor_pool = dev.createDescriptorPoolUnique(pool_ci, nullptr, device.GetDispatchLoader());
std::array<VkDescriptorPoolSize, 2> pool_sizes;
pool_sizes[0].type = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
pool_sizes[0].descriptorCount = static_cast<u32>(image_count);
pool_sizes[1].type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
pool_sizes[1].descriptorCount = static_cast<u32>(image_count);
VkDescriptorPoolCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT;
ci.maxSets = static_cast<u32>(image_count);
ci.poolSizeCount = static_cast<u32>(pool_sizes.size());
ci.pPoolSizes = pool_sizes.data();
descriptor_pool = device.GetLogical().CreateDescriptorPool(ci);
}
void VKBlitScreen::CreateRenderPass() {
const vk::AttachmentDescription color_attachment(
{}, swapchain.GetImageFormat(), vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eClear,
vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eDontCare,
vk::AttachmentStoreOp::eDontCare, vk::ImageLayout::eUndefined,
vk::ImageLayout::ePresentSrcKHR);
VkAttachmentDescription color_attachment;
color_attachment.flags = 0;
color_attachment.format = swapchain.GetImageFormat();
color_attachment.samples = VK_SAMPLE_COUNT_1_BIT;
color_attachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
color_attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
color_attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
color_attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
color_attachment.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
color_attachment.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR;
const vk::AttachmentReference color_attachment_ref(0, vk::ImageLayout::eColorAttachmentOptimal);
VkAttachmentReference color_attachment_ref;
color_attachment_ref.attachment = 0;
color_attachment_ref.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
const vk::SubpassDescription subpass_description({}, vk::PipelineBindPoint::eGraphics, 0,
nullptr, 1, &color_attachment_ref, nullptr,
nullptr, 0, nullptr);
VkSubpassDescription subpass_description;
subpass_description.flags = 0;
subpass_description.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass_description.inputAttachmentCount = 0;
subpass_description.pInputAttachments = nullptr;
subpass_description.colorAttachmentCount = 1;
subpass_description.pColorAttachments = &color_attachment_ref;
subpass_description.pResolveAttachments = nullptr;
subpass_description.pDepthStencilAttachment = nullptr;
subpass_description.preserveAttachmentCount = 0;
subpass_description.pPreserveAttachments = nullptr;
const vk::SubpassDependency dependency(
VK_SUBPASS_EXTERNAL, 0, vk::PipelineStageFlagBits::eColorAttachmentOutput,
vk::PipelineStageFlagBits::eColorAttachmentOutput, {},
vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite, {});
VkSubpassDependency dependency;
dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
dependency.dstSubpass = 0;
dependency.srcStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
dependency.dstStageMask = VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
dependency.srcAccessMask = 0;
dependency.dstAccessMask =
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
dependency.dependencyFlags = 0;
const vk::RenderPassCreateInfo renderpass_ci({}, 1, &color_attachment, 1, &subpass_description,
1, &dependency);
VkRenderPassCreateInfo renderpass_ci;
renderpass_ci.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
renderpass_ci.pNext = nullptr;
renderpass_ci.flags = 0;
renderpass_ci.attachmentCount = 1;
renderpass_ci.pAttachments = &color_attachment;
renderpass_ci.subpassCount = 1;
renderpass_ci.pSubpasses = &subpass_description;
renderpass_ci.dependencyCount = 1;
renderpass_ci.pDependencies = &dependency;
const auto dev = device.GetLogical();
renderpass = dev.createRenderPassUnique(renderpass_ci, nullptr, device.GetDispatchLoader());
renderpass = device.GetLogical().CreateRenderPass(renderpass_ci);
}
void VKBlitScreen::CreateDescriptorSetLayout() {
const std::array<vk::DescriptorSetLayoutBinding, 2> layout_bindings{
vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eUniformBuffer, 1,
vk::ShaderStageFlagBits::eVertex, nullptr),
vk::DescriptorSetLayoutBinding(1, vk::DescriptorType::eCombinedImageSampler, 1,
vk::ShaderStageFlagBits::eFragment, nullptr)};
const vk::DescriptorSetLayoutCreateInfo descriptor_layout_ci(
{}, static_cast<u32>(layout_bindings.size()), layout_bindings.data());
std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings;
layout_bindings[0].binding = 0;
layout_bindings[0].descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
layout_bindings[0].descriptorCount = 1;
layout_bindings[0].stageFlags = VK_SHADER_STAGE_VERTEX_BIT;
layout_bindings[0].pImmutableSamplers = nullptr;
layout_bindings[1].binding = 1;
layout_bindings[1].descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
layout_bindings[1].descriptorCount = 1;
layout_bindings[1].stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
layout_bindings[1].pImmutableSamplers = nullptr;
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
descriptor_set_layout = dev.createDescriptorSetLayoutUnique(descriptor_layout_ci, nullptr, dld);
VkDescriptorSetLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.bindingCount = static_cast<u32>(layout_bindings.size());
ci.pBindings = layout_bindings.data();
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci);
}
void VKBlitScreen::CreateDescriptorSets() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const std::vector layouts(image_count, *descriptor_set_layout);
descriptor_sets.resize(image_count);
for (std::size_t i = 0; i < image_count; ++i) {
const vk::DescriptorSetLayout layout = *descriptor_set_layout;
const vk::DescriptorSetAllocateInfo descriptor_set_ai(*descriptor_pool, 1, &layout);
const vk::Result result =
dev.allocateDescriptorSets(&descriptor_set_ai, &descriptor_sets[i], dld);
ASSERT(result == vk::Result::eSuccess);
}
VkDescriptorSetAllocateInfo ai;
ai.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO;
ai.pNext = nullptr;
ai.descriptorPool = *descriptor_pool;
ai.descriptorSetCount = static_cast<u32>(image_count);
ai.pSetLayouts = layouts.data();
descriptor_sets = descriptor_pool.Allocate(ai);
}
void VKBlitScreen::CreatePipelineLayout() {
const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &descriptor_set_layout.get(), 0,
nullptr);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
pipeline_layout = dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
VkPipelineLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.setLayoutCount = 1;
ci.pSetLayouts = descriptor_set_layout.address();
ci.pushConstantRangeCount = 0;
ci.pPushConstantRanges = nullptr;
pipeline_layout = device.GetLogical().CreatePipelineLayout(ci);
}
void VKBlitScreen::CreateGraphicsPipeline() {
const std::array shader_stages = {
vk::PipelineShaderStageCreateInfo({}, vk::ShaderStageFlagBits::eVertex, *vertex_shader,
"main", nullptr),
vk::PipelineShaderStageCreateInfo({}, vk::ShaderStageFlagBits::eFragment, *fragment_shader,
"main", nullptr)};
std::array<VkPipelineShaderStageCreateInfo, 2> shader_stages;
shader_stages[0].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
shader_stages[0].pNext = nullptr;
shader_stages[0].flags = 0;
shader_stages[0].stage = VK_SHADER_STAGE_VERTEX_BIT;
shader_stages[0].module = *vertex_shader;
shader_stages[0].pName = "main";
shader_stages[0].pSpecializationInfo = nullptr;
shader_stages[1].sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
shader_stages[1].pNext = nullptr;
shader_stages[1].flags = 0;
shader_stages[1].stage = VK_SHADER_STAGE_FRAGMENT_BIT;
shader_stages[1].module = *fragment_shader;
shader_stages[1].pName = "main";
shader_stages[1].pSpecializationInfo = nullptr;
const auto vertex_binding_description = ScreenRectVertex::GetDescription();
const auto vertex_attrs_description = ScreenRectVertex::GetAttributes();
const vk::PipelineVertexInputStateCreateInfo vertex_input(
{}, 1, &vertex_binding_description, static_cast<u32>(vertex_attrs_description.size()),
vertex_attrs_description.data());
const vk::PipelineInputAssemblyStateCreateInfo input_assembly(
{}, vk::PrimitiveTopology::eTriangleStrip, false);
VkPipelineVertexInputStateCreateInfo vertex_input_ci;
vertex_input_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
vertex_input_ci.pNext = nullptr;
vertex_input_ci.flags = 0;
vertex_input_ci.vertexBindingDescriptionCount = 1;
vertex_input_ci.pVertexBindingDescriptions = &vertex_binding_description;
vertex_input_ci.vertexAttributeDescriptionCount = u32{vertex_attrs_description.size()};
vertex_input_ci.pVertexAttributeDescriptions = vertex_attrs_description.data();
// Set a dummy viewport, it's going to be replaced by dynamic states.
const vk::Viewport viewport(0.0f, 0.0f, 1.0f, 1.0f, 0.0f, 1.0f);
const vk::Rect2D scissor({0, 0}, {1, 1});
VkPipelineInputAssemblyStateCreateInfo input_assembly_ci;
input_assembly_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
input_assembly_ci.pNext = nullptr;
input_assembly_ci.flags = 0;
input_assembly_ci.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP;
input_assembly_ci.primitiveRestartEnable = VK_FALSE;
const vk::PipelineViewportStateCreateInfo viewport_state({}, 1, &viewport, 1, &scissor);
VkPipelineViewportStateCreateInfo viewport_state_ci;
viewport_state_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO;
viewport_state_ci.pNext = nullptr;
viewport_state_ci.flags = 0;
viewport_state_ci.viewportCount = 1;
viewport_state_ci.scissorCount = 1;
const vk::PipelineRasterizationStateCreateInfo rasterizer(
{}, false, false, vk::PolygonMode::eFill, vk::CullModeFlagBits::eNone,
vk::FrontFace::eClockwise, false, 0.0f, 0.0f, 0.0f, 1.0f);
VkPipelineRasterizationStateCreateInfo rasterization_ci;
rasterization_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
rasterization_ci.pNext = nullptr;
rasterization_ci.flags = 0;
rasterization_ci.depthClampEnable = VK_FALSE;
rasterization_ci.rasterizerDiscardEnable = VK_FALSE;
rasterization_ci.polygonMode = VK_POLYGON_MODE_FILL;
rasterization_ci.cullMode = VK_CULL_MODE_NONE;
rasterization_ci.frontFace = VK_FRONT_FACE_CLOCKWISE;
rasterization_ci.depthBiasEnable = VK_FALSE;
rasterization_ci.depthBiasConstantFactor = 0.0f;
rasterization_ci.depthBiasClamp = 0.0f;
rasterization_ci.depthBiasSlopeFactor = 0.0f;
rasterization_ci.lineWidth = 1.0f;
const vk::PipelineMultisampleStateCreateInfo multisampling({}, vk::SampleCountFlagBits::e1,
false, 0.0f, nullptr, false, false);
VkPipelineMultisampleStateCreateInfo multisampling_ci;
multisampling_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
multisampling_ci.pNext = nullptr;
multisampling_ci.flags = 0;
multisampling_ci.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
multisampling_ci.sampleShadingEnable = VK_FALSE;
multisampling_ci.minSampleShading = 0.0f;
multisampling_ci.pSampleMask = nullptr;
multisampling_ci.alphaToCoverageEnable = VK_FALSE;
multisampling_ci.alphaToOneEnable = VK_FALSE;
const vk::PipelineColorBlendAttachmentState color_blend_attachment(
false, vk::BlendFactor::eZero, vk::BlendFactor::eZero, vk::BlendOp::eAdd,
vk::BlendFactor::eZero, vk::BlendFactor::eZero, vk::BlendOp::eAdd,
vk::ColorComponentFlagBits::eR | vk::ColorComponentFlagBits::eG |
vk::ColorComponentFlagBits::eB | vk::ColorComponentFlagBits::eA);
VkPipelineColorBlendAttachmentState color_blend_attachment;
color_blend_attachment.blendEnable = VK_FALSE;
color_blend_attachment.srcColorBlendFactor = VK_BLEND_FACTOR_ZERO;
color_blend_attachment.dstColorBlendFactor = VK_BLEND_FACTOR_ZERO;
color_blend_attachment.colorBlendOp = VK_BLEND_OP_ADD;
color_blend_attachment.srcAlphaBlendFactor = VK_BLEND_FACTOR_ZERO;
color_blend_attachment.dstAlphaBlendFactor = VK_BLEND_FACTOR_ZERO;
color_blend_attachment.alphaBlendOp = VK_BLEND_OP_ADD;
color_blend_attachment.colorWriteMask = VK_COLOR_COMPONENT_R_BIT | VK_COLOR_COMPONENT_G_BIT |
VK_COLOR_COMPONENT_B_BIT | VK_COLOR_COMPONENT_A_BIT;
const vk::PipelineColorBlendStateCreateInfo color_blending(
{}, false, vk::LogicOp::eCopy, 1, &color_blend_attachment, {0.0f, 0.0f, 0.0f, 0.0f});
VkPipelineColorBlendStateCreateInfo color_blend_ci;
color_blend_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO;
color_blend_ci.flags = 0;
color_blend_ci.pNext = nullptr;
color_blend_ci.logicOpEnable = VK_FALSE;
color_blend_ci.logicOp = VK_LOGIC_OP_COPY;
color_blend_ci.attachmentCount = 1;
color_blend_ci.pAttachments = &color_blend_attachment;
color_blend_ci.blendConstants[0] = 0.0f;
color_blend_ci.blendConstants[1] = 0.0f;
color_blend_ci.blendConstants[2] = 0.0f;
color_blend_ci.blendConstants[3] = 0.0f;
const std::array<vk::DynamicState, 2> dynamic_states = {vk::DynamicState::eViewport,
vk::DynamicState::eScissor};
static constexpr std::array dynamic_states = {VK_DYNAMIC_STATE_VIEWPORT,
VK_DYNAMIC_STATE_SCISSOR};
VkPipelineDynamicStateCreateInfo dynamic_state_ci;
dynamic_state_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO;
dynamic_state_ci.pNext = nullptr;
dynamic_state_ci.flags = 0;
dynamic_state_ci.dynamicStateCount = static_cast<u32>(dynamic_states.size());
dynamic_state_ci.pDynamicStates = dynamic_states.data();
const vk::PipelineDynamicStateCreateInfo dynamic_state(
{}, static_cast<u32>(dynamic_states.size()), dynamic_states.data());
VkGraphicsPipelineCreateInfo pipeline_ci;
pipeline_ci.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;
pipeline_ci.pNext = nullptr;
pipeline_ci.flags = 0;
pipeline_ci.stageCount = static_cast<u32>(shader_stages.size());
pipeline_ci.pStages = shader_stages.data();
pipeline_ci.pVertexInputState = &vertex_input_ci;
pipeline_ci.pInputAssemblyState = &input_assembly_ci;
pipeline_ci.pTessellationState = nullptr;
pipeline_ci.pViewportState = &viewport_state_ci;
pipeline_ci.pRasterizationState = &rasterization_ci;
pipeline_ci.pMultisampleState = &multisampling_ci;
pipeline_ci.pDepthStencilState = nullptr;
pipeline_ci.pColorBlendState = &color_blend_ci;
pipeline_ci.pDynamicState = &dynamic_state_ci;
pipeline_ci.layout = *pipeline_layout;
pipeline_ci.renderPass = *renderpass;
pipeline_ci.subpass = 0;
pipeline_ci.basePipelineHandle = 0;
pipeline_ci.basePipelineIndex = 0;
const vk::GraphicsPipelineCreateInfo pipeline_ci(
{}, static_cast<u32>(shader_stages.size()), shader_stages.data(), &vertex_input,
&input_assembly, nullptr, &viewport_state, &rasterizer, &multisampling, nullptr,
&color_blending, &dynamic_state, *pipeline_layout, *renderpass, 0, nullptr, 0);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
pipeline = dev.createGraphicsPipelineUnique({}, pipeline_ci, nullptr, dld);
pipeline = device.GetLogical().CreateGraphicsPipeline(pipeline_ci);
}
void VKBlitScreen::CreateSampler() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const vk::SamplerCreateInfo sampler_ci(
{}, vk::Filter::eLinear, vk::Filter::eLinear, vk::SamplerMipmapMode::eLinear,
vk::SamplerAddressMode::eClampToBorder, vk::SamplerAddressMode::eClampToBorder,
vk::SamplerAddressMode::eClampToBorder, 0.0f, false, 0.0f, false, vk::CompareOp::eNever,
0.0f, 0.0f, vk::BorderColor::eFloatOpaqueBlack, false);
sampler = dev.createSamplerUnique(sampler_ci, nullptr, dld);
VkSamplerCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.magFilter = VK_FILTER_LINEAR;
ci.minFilter = VK_FILTER_NEAREST;
ci.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR;
ci.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
ci.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
ci.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
ci.mipLodBias = 0.0f;
ci.anisotropyEnable = VK_FALSE;
ci.maxAnisotropy = 0.0f;
ci.compareEnable = VK_FALSE;
ci.compareOp = VK_COMPARE_OP_NEVER;
ci.minLod = 0.0f;
ci.maxLod = 0.0f;
ci.borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
ci.unnormalizedCoordinates = VK_FALSE;
sampler = device.GetLogical().CreateSampler(ci);
}
void VKBlitScreen::CreateFramebuffers() {
const vk::Extent2D size{swapchain.GetSize()};
framebuffers.clear();
const VkExtent2D size{swapchain.GetSize()};
framebuffers.resize(image_count);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
VkFramebufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.renderPass = *renderpass;
ci.attachmentCount = 1;
ci.width = size.width;
ci.height = size.height;
ci.layers = 1;
for (std::size_t i = 0; i < image_count; ++i) {
const vk::ImageView image_view{swapchain.GetImageViewIndex(i)};
const vk::FramebufferCreateInfo framebuffer_ci({}, *renderpass, 1, &image_view, size.width,
size.height, 1);
framebuffers[i] = dev.createFramebufferUnique(framebuffer_ci, nullptr, dld);
const VkImageView image_view{swapchain.GetImageViewIndex(i)};
ci.pAttachments = &image_view;
framebuffers[i] = device.GetLogical().CreateFramebuffer(ci);
}
}
@ -507,54 +676,86 @@ void VKBlitScreen::ReleaseRawImages() {
}
void VKBlitScreen::CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
VkBufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.size = CalculateBufferSize(framebuffer);
ci.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
const vk::BufferCreateInfo buffer_ci({}, CalculateBufferSize(framebuffer),
vk::BufferUsageFlagBits::eTransferSrc |
vk::BufferUsageFlagBits::eVertexBuffer |
vk::BufferUsageFlagBits::eUniformBuffer,
vk::SharingMode::eExclusive, 0, nullptr);
buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
buffer_commit = memory_manager.Commit(*buffer, true);
buffer = device.GetLogical().CreateBuffer(ci);
buffer_commit = memory_manager.Commit(buffer, true);
}
void VKBlitScreen::CreateRawImages(const Tegra::FramebufferConfig& framebuffer) {
raw_images.resize(image_count);
raw_buffer_commits.resize(image_count);
const auto format = GetFormat(framebuffer);
for (std::size_t i = 0; i < image_count; ++i) {
const vk::ImageCreateInfo image_ci(
{}, vk::ImageType::e2D, format, {framebuffer.width, framebuffer.height, 1}, 1, 1,
vk::SampleCountFlagBits::e1, vk::ImageTiling::eOptimal,
vk::ImageUsageFlagBits::eTransferDst | vk::ImageUsageFlagBits::eSampled,
vk::SharingMode::eExclusive, 0, nullptr, vk::ImageLayout::eUndefined);
VkImageCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.imageType = VK_IMAGE_TYPE_2D;
ci.format = GetFormat(framebuffer);
ci.extent.width = framebuffer.width;
ci.extent.height = framebuffer.height;
ci.extent.depth = 1;
ci.mipLevels = 1;
ci.arrayLayers = 1;
ci.samples = VK_SAMPLE_COUNT_1_BIT;
ci.tiling = VK_IMAGE_TILING_LINEAR;
ci.usage = VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
ci.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
raw_images[i] =
std::make_unique<VKImage>(device, scheduler, image_ci, vk::ImageAspectFlagBits::eColor);
for (std::size_t i = 0; i < image_count; ++i) {
raw_images[i] = std::make_unique<VKImage>(device, scheduler, ci, VK_IMAGE_ASPECT_COLOR_BIT);
raw_buffer_commits[i] = memory_manager.Commit(raw_images[i]->GetHandle(), false);
}
}
void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, vk::ImageView image_view) const {
const vk::DescriptorSet descriptor_set = descriptor_sets[image_index];
void VKBlitScreen::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const {
VkDescriptorBufferInfo buffer_info;
buffer_info.buffer = *buffer;
buffer_info.offset = offsetof(BufferData, uniform);
buffer_info.range = sizeof(BufferData::uniform);
const vk::DescriptorBufferInfo buffer_info(*buffer, offsetof(BufferData, uniform),
sizeof(BufferData::uniform));
const vk::WriteDescriptorSet ubo_write(descriptor_set, 0, 0, 1,
vk::DescriptorType::eUniformBuffer, nullptr,
&buffer_info, nullptr);
VkWriteDescriptorSet ubo_write;
ubo_write.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
ubo_write.pNext = nullptr;
ubo_write.dstSet = descriptor_sets[image_index];
ubo_write.dstBinding = 0;
ubo_write.dstArrayElement = 0;
ubo_write.descriptorCount = 1;
ubo_write.descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
ubo_write.pImageInfo = nullptr;
ubo_write.pBufferInfo = &buffer_info;
ubo_write.pTexelBufferView = nullptr;
const vk::DescriptorImageInfo image_info(*sampler, image_view,
vk::ImageLayout::eShaderReadOnlyOptimal);
const vk::WriteDescriptorSet sampler_write(descriptor_set, 1, 0, 1,
vk::DescriptorType::eCombinedImageSampler,
&image_info, nullptr, nullptr);
VkDescriptorImageInfo image_info;
image_info.sampler = *sampler;
image_info.imageView = image_view;
image_info.imageLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
dev.updateDescriptorSets({ubo_write, sampler_write}, {}, dld);
VkWriteDescriptorSet sampler_write;
sampler_write.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET;
sampler_write.pNext = nullptr;
sampler_write.dstSet = descriptor_sets[image_index];
sampler_write.dstBinding = 1;
sampler_write.dstArrayElement = 0;
sampler_write.descriptorCount = 1;
sampler_write.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
sampler_write.pImageInfo = &image_info;
sampler_write.pBufferInfo = nullptr;
sampler_write.pTexelBufferView = nullptr;
device.GetLogical().UpdateDescriptorSets(std::array{ubo_write, sampler_write}, {});
}
void VKBlitScreen::SetUniformData(BufferData& data,

32
src/video_core/renderer_vulkan/vk_blit_screen.h
View file

@ -8,9 +8,9 @@
#include <memory>
#include <tuple>
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Core {
class System;
@ -49,8 +49,8 @@ public:
void Recreate();
std::tuple<VKFence&, vk::Semaphore> Draw(const Tegra::FramebufferConfig& framebuffer,
bool use_accelerated);
std::tuple<VKFence&, VkSemaphore> Draw(const Tegra::FramebufferConfig& framebuffer,
bool use_accelerated);
private:
struct BufferData;
@ -74,7 +74,7 @@ private:
void CreateStagingBuffer(const Tegra::FramebufferConfig& framebuffer);
void CreateRawImages(const Tegra::FramebufferConfig& framebuffer);
void UpdateDescriptorSet(std::size_t image_index, vk::ImageView image_view) const;
void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const;
void SetUniformData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
void SetVertexData(BufferData& data, const Tegra::FramebufferConfig& framebuffer) const;
@ -93,23 +93,23 @@ private:
const std::size_t image_count;
const VKScreenInfo& screen_info;
UniqueShaderModule vertex_shader;
UniqueShaderModule fragment_shader;
UniqueDescriptorPool descriptor_pool;
UniqueDescriptorSetLayout descriptor_set_layout;
UniquePipelineLayout pipeline_layout;
UniquePipeline pipeline;
UniqueRenderPass renderpass;
std::vector<UniqueFramebuffer> framebuffers;
std::vector<vk::DescriptorSet> descriptor_sets;
UniqueSampler sampler;
vk::ShaderModule vertex_shader;
vk::ShaderModule fragment_shader;
vk::DescriptorPool descriptor_pool;
vk::DescriptorSetLayout descriptor_set_layout;
vk::PipelineLayout pipeline_layout;
vk::Pipeline pipeline;
vk::RenderPass renderpass;
std::vector<vk::Framebuffer> framebuffers;
vk::DescriptorSets descriptor_sets;
vk::Sampler sampler;
UniqueBuffer buffer;
vk::Buffer buffer;
VKMemoryCommit buffer_commit;
std::vector<std::unique_ptr<VKFenceWatch>> watches;
std::vector<UniqueSemaphore> semaphores;
std::vector<vk::Semaphore> semaphores;
std::vector<std::unique_ptr<VKImage>> raw_images;
std::vector<VKMemoryCommit> raw_buffer_commits;
u32 raw_width = 0;

147
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View file

@ -11,32 +11,31 @@
#include "common/assert.h"
#include "common/bit_util.h"
#include "core/core.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_stream_buffer.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
namespace {
const auto BufferUsage =
vk::BufferUsageFlagBits::eVertexBuffer | vk::BufferUsageFlagBits::eIndexBuffer |
vk::BufferUsageFlagBits::eUniformBuffer | vk::BufferUsageFlagBits::eStorageBuffer;
constexpr VkBufferUsageFlags BUFFER_USAGE =
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT | VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
const auto UploadPipelineStage =
vk::PipelineStageFlagBits::eTransfer | vk::PipelineStageFlagBits::eVertexInput |
vk::PipelineStageFlagBits::eVertexShader | vk::PipelineStageFlagBits::eFragmentShader |
vk::PipelineStageFlagBits::eComputeShader;
constexpr VkPipelineStageFlags UPLOAD_PIPELINE_STAGE =
VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
const auto UploadAccessBarriers =
vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eShaderRead |
vk::AccessFlagBits::eUniformRead | vk::AccessFlagBits::eVertexAttributeRead |
vk::AccessFlagBits::eIndexRead;
constexpr VkAccessFlags UPLOAD_ACCESS_BARRIERS =
VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_UNIFORM_READ_BIT |
VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDEX_READ_BIT;
auto CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
return std::make_unique<VKStreamBuffer>(device, scheduler, BufferUsage);
std::unique_ptr<VKStreamBuffer> CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
return std::make_unique<VKStreamBuffer>(device, scheduler, BUFFER_USAGE);
}
} // Anonymous namespace
@ -44,15 +43,18 @@ auto CreateStreamBuffer(const VKDevice& device, VKScheduler& scheduler) {
CachedBufferBlock::CachedBufferBlock(const VKDevice& device, VKMemoryManager& memory_manager,
VAddr cpu_addr, std::size_t size)
: VideoCommon::BufferBlock{cpu_addr, size} {
const vk::BufferCreateInfo buffer_ci({}, static_cast<vk::DeviceSize>(size),
BufferUsage | vk::BufferUsageFlagBits::eTransferSrc |
vk::BufferUsageFlagBits::eTransferDst,
vk::SharingMode::eExclusive, 0, nullptr);
VkBufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.size = static_cast<VkDeviceSize>(size);
ci.usage = BUFFER_USAGE | VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
const auto& dld{device.GetDispatchLoader()};
const auto dev{device.GetLogical()};
buffer.handle = dev.createBufferUnique(buffer_ci, nullptr, dld);
buffer.commit = memory_manager.Commit(*buffer.handle, false);
buffer.handle = device.GetLogical().CreateBuffer(ci);
buffer.commit = memory_manager.Commit(buffer.handle, false);
}
CachedBufferBlock::~CachedBufferBlock() = default;
@ -60,9 +62,9 @@ CachedBufferBlock::~CachedBufferBlock() = default;
VKBufferCache::VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
const VKDevice& device, VKMemoryManager& memory_manager,
VKScheduler& scheduler, VKStagingBufferPool& staging_pool)
: VideoCommon::BufferCache<Buffer, vk::Buffer, VKStreamBuffer>{rasterizer, system,
CreateStreamBuffer(device,
scheduler)},
: VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer>{rasterizer, system,
CreateStreamBuffer(device,
scheduler)},
device{device}, memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{
staging_pool} {}
@ -72,18 +74,18 @@ Buffer VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
return std::make_shared<CachedBufferBlock>(device, memory_manager, cpu_addr, size);
}
const vk::Buffer* VKBufferCache::ToHandle(const Buffer& buffer) {
const VkBuffer* VKBufferCache::ToHandle(const Buffer& buffer) {
return buffer->GetHandle();
}
const vk::Buffer* VKBufferCache::GetEmptyBuffer(std::size_t size) {
const VkBuffer* VKBufferCache::GetEmptyBuffer(std::size_t size) {
size = std::max(size, std::size_t(4));
const auto& empty = staging_pool.GetUnusedBuffer(size, false);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf, auto& dld) {
cmdbuf.fillBuffer(buffer, 0, size, 0, dld);
scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
cmdbuf.FillBuffer(buffer, 0, size, 0);
});
return &*empty.handle;
return empty.handle.address();
}
void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
@ -93,14 +95,21 @@ void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, st
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
size](auto cmdbuf, auto& dld) {
cmdbuf.copyBuffer(staging, buffer, {{0, offset, size}}, dld);
cmdbuf.pipelineBarrier(
vk::PipelineStageFlagBits::eTransfer, UploadPipelineStage, {}, {},
{vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite, UploadAccessBarriers,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer,
offset, size)},
{}, dld);
size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(staging, buffer, VkBufferCopy{0, offset, size});
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barrier.dstAccessMask = UPLOAD_ACCESS_BARRIERS;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = offset;
barrier.size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
barrier, {});
});
}
@ -109,16 +118,23 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
const auto& staging = staging_pool.GetUnusedBuffer(size, true);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
size](auto cmdbuf, auto& dld) {
cmdbuf.pipelineBarrier(
vk::PipelineStageFlagBits::eVertexShader | vk::PipelineStageFlagBits::eFragmentShader |
vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer, {}, {},
{vk::BufferMemoryBarrier(vk::AccessFlagBits::eShaderWrite,
vk::AccessFlagBits::eTransferRead, VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED, buffer, offset, size)},
{}, dld);
cmdbuf.copyBuffer(buffer, staging, {{offset, 0, size}}, dld);
size](vk::CommandBuffer cmdbuf) {
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = offset;
barrier.size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_VERTEX_SHADER_BIT |
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, {}, barrier, {});
cmdbuf.CopyBuffer(buffer, staging, VkBufferCopy{offset, 0, size});
});
scheduler.Finish();
@ -129,17 +145,30 @@ void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t
std::size_t dst_offset, std::size_t size) {
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([src_buffer = *src->GetHandle(), dst_buffer = *dst->GetHandle(), src_offset,
dst_offset, size](auto cmdbuf, auto& dld) {
cmdbuf.copyBuffer(src_buffer, dst_buffer, {{src_offset, dst_offset, size}}, dld);
cmdbuf.pipelineBarrier(
vk::PipelineStageFlagBits::eTransfer, UploadPipelineStage, {}, {},
{vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferRead,
vk::AccessFlagBits::eShaderWrite, VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED, src_buffer, src_offset, size),
vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite, UploadAccessBarriers,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, dst_buffer,
dst_offset, size)},
{}, dld);
dst_offset, size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size});
std::array<VkBufferMemoryBarrier, 2> barriers;
barriers[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barriers[0].pNext = nullptr;
barriers[0].srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
barriers[0].dstAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barriers[0].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[0].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[0].buffer = src_buffer;
barriers[0].offset = src_offset;
barriers[0].size = size;
barriers[1].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barriers[1].pNext = nullptr;
barriers[1].srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
barriers[1].dstAccessMask = UPLOAD_ACCESS_BARRIERS;
barriers[1].srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[1].dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barriers[1].buffer = dst_buffer;
barriers[1].offset = dst_offset;
barriers[1].size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, UPLOAD_PIPELINE_STAGE, 0, {},
barriers, {});
});
}

12
src/video_core/renderer_vulkan/vk_buffer_cache.h
View file

@ -11,11 +11,11 @@
#include "common/common_types.h"
#include "video_core/buffer_cache/buffer_cache.h"
#include "video_core/rasterizer_cache.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_stream_buffer.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Core {
class System;
@ -33,8 +33,8 @@ public:
VAddr cpu_addr, std::size_t size);
~CachedBufferBlock();
const vk::Buffer* GetHandle() const {
return &*buffer.handle;
const VkBuffer* GetHandle() const {
return buffer.handle.address();
}
private:
@ -43,21 +43,21 @@ private:
using Buffer = std::shared_ptr<CachedBufferBlock>;
class VKBufferCache final : public VideoCommon::BufferCache<Buffer, vk::Buffer, VKStreamBuffer> {
class VKBufferCache final : public VideoCommon::BufferCache<Buffer, VkBuffer, VKStreamBuffer> {
public:
explicit VKBufferCache(VideoCore::RasterizerInterface& rasterizer, Core::System& system,
const VKDevice& device, VKMemoryManager& memory_manager,
VKScheduler& scheduler, VKStagingBufferPool& staging_pool);
~VKBufferCache();
const vk::Buffer* GetEmptyBuffer(std::size_t size) override;
const VkBuffer* GetEmptyBuffer(std::size_t size) override;
protected:
void WriteBarrier() override {}
Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
const vk::Buffer* ToHandle(const Buffer& buffer) override;
const VkBuffer* ToHandle(const Buffer& buffer) override;
void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
const u8* data) override;

235
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View file

@ -10,13 +10,13 @@
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -114,6 +114,35 @@ constexpr u8 quad_array[] = {
0xf9, 0x00, 0x02, 0x00, 0x4c, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x4b, 0x00, 0x00, 0x00,
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
VkDescriptorSetLayoutBinding BuildQuadArrayPassDescriptorSetLayoutBinding() {
VkDescriptorSetLayoutBinding binding;
binding.binding = 0;
binding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
binding.descriptorCount = 1;
binding.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
binding.pImmutableSamplers = nullptr;
return binding;
}
VkDescriptorUpdateTemplateEntryKHR BuildQuadArrayPassDescriptorUpdateTemplateEntry() {
VkDescriptorUpdateTemplateEntryKHR entry;
entry.dstBinding = 0;
entry.dstArrayElement = 0;
entry.descriptorCount = 1;
entry.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
entry.offset = 0;
entry.stride = sizeof(DescriptorUpdateEntry);
return entry;
}
VkPushConstantRange BuildQuadArrayPassPushConstantRange() {
VkPushConstantRange range;
range.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
range.offset = 0;
range.size = sizeof(u32);
return range;
}
// Uint8 SPIR-V module. Generated from the "shaders/" directory.
constexpr u8 uint8_pass[] = {
0x03, 0x02, 0x23, 0x07, 0x00, 0x00, 0x01, 0x00, 0x07, 0x00, 0x08, 0x00, 0x2f, 0x00, 0x00, 0x00,
@ -191,53 +220,111 @@ constexpr u8 uint8_pass[] = {
0xf9, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00, 0xf8, 0x00, 0x02, 0x00, 0x1d, 0x00, 0x00, 0x00,
0xfd, 0x00, 0x01, 0x00, 0x38, 0x00, 0x01, 0x00};
std::array<VkDescriptorSetLayoutBinding, 2> BuildUint8PassDescriptorSetBindings() {
std::array<VkDescriptorSetLayoutBinding, 2> bindings;
bindings[0].binding = 0;
bindings[0].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
bindings[0].descriptorCount = 1;
bindings[0].stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
bindings[0].pImmutableSamplers = nullptr;
bindings[1].binding = 1;
bindings[1].descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
bindings[1].descriptorCount = 1;
bindings[1].stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
bindings[1].pImmutableSamplers = nullptr;
return bindings;
}
VkDescriptorUpdateTemplateEntryKHR BuildUint8PassDescriptorUpdateTemplateEntry() {
VkDescriptorUpdateTemplateEntryKHR entry;
entry.dstBinding = 0;
entry.dstArrayElement = 0;
entry.descriptorCount = 2;
entry.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
entry.offset = 0;
entry.stride = sizeof(DescriptorUpdateEntry);
return entry;
}
} // Anonymous namespace
VKComputePass::VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
const std::vector<vk::DescriptorUpdateTemplateEntry>& templates,
const std::vector<vk::PushConstantRange> push_constants,
std::size_t code_size, const u8* code) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
vk::Span<VkPushConstantRange> push_constants, std::size_t code_size,
const u8* code) {
VkDescriptorSetLayoutCreateInfo descriptor_layout_ci;
descriptor_layout_ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
descriptor_layout_ci.pNext = nullptr;
descriptor_layout_ci.flags = 0;
descriptor_layout_ci.bindingCount = bindings.size();
descriptor_layout_ci.pBindings = bindings.data();
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(descriptor_layout_ci);
const vk::DescriptorSetLayoutCreateInfo descriptor_layout_ci(
{}, static_cast<u32>(bindings.size()), bindings.data());
descriptor_set_layout = dev.createDescriptorSetLayoutUnique(descriptor_layout_ci, nullptr, dld);
const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &*descriptor_set_layout,
static_cast<u32>(push_constants.size()),
push_constants.data());
layout = dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
VkPipelineLayoutCreateInfo pipeline_layout_ci;
pipeline_layout_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
pipeline_layout_ci.pNext = nullptr;
pipeline_layout_ci.flags = 0;
pipeline_layout_ci.setLayoutCount = 1;
pipeline_layout_ci.pSetLayouts = descriptor_set_layout.address();
pipeline_layout_ci.pushConstantRangeCount = push_constants.size();
pipeline_layout_ci.pPushConstantRanges = push_constants.data();
layout = device.GetLogical().CreatePipelineLayout(pipeline_layout_ci);
if (!templates.empty()) {
const vk::DescriptorUpdateTemplateCreateInfo template_ci(
{}, static_cast<u32>(templates.size()), templates.data(),
vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
vk::PipelineBindPoint::eGraphics, *layout, 0);
descriptor_template = dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
VkDescriptorUpdateTemplateCreateInfoKHR template_ci;
template_ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR;
template_ci.pNext = nullptr;
template_ci.flags = 0;
template_ci.descriptorUpdateEntryCount = templates.size();
template_ci.pDescriptorUpdateEntries = templates.data();
template_ci.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR;
template_ci.descriptorSetLayout = *descriptor_set_layout;
template_ci.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
template_ci.pipelineLayout = *layout;
template_ci.set = 0;
descriptor_template = device.GetLogical().CreateDescriptorUpdateTemplateKHR(template_ci);
descriptor_allocator.emplace(descriptor_pool, *descriptor_set_layout);
}
auto code_copy = std::make_unique<u32[]>(code_size / sizeof(u32) + 1);
std::memcpy(code_copy.get(), code, code_size);
const vk::ShaderModuleCreateInfo module_ci({}, code_size, code_copy.get());
module = dev.createShaderModuleUnique(module_ci, nullptr, dld);
const vk::PipelineShaderStageCreateInfo stage_ci({}, vk::ShaderStageFlagBits::eCompute, *module,
"main", nullptr);
VkShaderModuleCreateInfo module_ci;
module_ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
module_ci.pNext = nullptr;
module_ci.flags = 0;
module_ci.codeSize = code_size;
module_ci.pCode = code_copy.get();
module = device.GetLogical().CreateShaderModule(module_ci);
const vk::ComputePipelineCreateInfo pipeline_ci({}, stage_ci, *layout, nullptr, 0);
pipeline = dev.createComputePipelineUnique(nullptr, pipeline_ci, nullptr, dld);
VkComputePipelineCreateInfo pipeline_ci;
pipeline_ci.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO;
pipeline_ci.pNext = nullptr;
pipeline_ci.flags = 0;
pipeline_ci.layout = *layout;
pipeline_ci.basePipelineHandle = nullptr;
pipeline_ci.basePipelineIndex = 0;
VkPipelineShaderStageCreateInfo& stage_ci = pipeline_ci.stage;
stage_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage_ci.pNext = nullptr;
stage_ci.flags = 0;
stage_ci.stage = VK_SHADER_STAGE_COMPUTE_BIT;
stage_ci.module = *module;
stage_ci.pName = "main";
stage_ci.pSpecializationInfo = nullptr;
pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci);
}
VKComputePass::~VKComputePass() = default;
vk::DescriptorSet VKComputePass::CommitDescriptorSet(
VKUpdateDescriptorQueue& update_descriptor_queue, VKFence& fence) {
VkDescriptorSet VKComputePass::CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue,
VKFence& fence) {
if (!descriptor_template) {
return {};
return nullptr;
}
const auto set = descriptor_allocator->Commit(fence);
update_descriptor_queue.Send(*descriptor_template, set);
@ -248,25 +335,21 @@ QuadArrayPass::QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
VKDescriptorPool& descriptor_pool,
VKStagingBufferPool& staging_buffer_pool,
VKUpdateDescriptorQueue& update_descriptor_queue)
: VKComputePass(device, descriptor_pool,
{vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eStorageBuffer, 1,
vk::ShaderStageFlagBits::eCompute, nullptr)},
{vk::DescriptorUpdateTemplateEntry(0, 0, 1, vk::DescriptorType::eStorageBuffer,
0, sizeof(DescriptorUpdateEntry))},
{vk::PushConstantRange(vk::ShaderStageFlagBits::eCompute, 0, sizeof(u32))},
std::size(quad_array), quad_array),
: VKComputePass(device, descriptor_pool, BuildQuadArrayPassDescriptorSetLayoutBinding(),
BuildQuadArrayPassDescriptorUpdateTemplateEntry(),
BuildQuadArrayPassPushConstantRange(), std::size(quad_array), quad_array),
scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
update_descriptor_queue{update_descriptor_queue} {}
QuadArrayPass::~QuadArrayPass() = default;
std::pair<const vk::Buffer&, vk::DeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
std::pair<const VkBuffer*, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
const u32 num_triangle_vertices = num_vertices * 6 / 4;
const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(&*buffer.handle, 0, staging_size);
update_descriptor_queue.AddBuffer(buffer.handle.address(), 0, staging_size);
const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
scheduler.RequestOutsideRenderPassOperationContext();
@ -274,66 +357,72 @@ std::pair<const vk::Buffer&, vk::DeviceSize> QuadArrayPass::Assemble(u32 num_ver
ASSERT(num_vertices % 4 == 0);
const u32 num_quads = num_vertices / 4;
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, num_quads,
first, set](auto cmdbuf, auto& dld) {
first, set](vk::CommandBuffer cmdbuf) {
constexpr u32 dispatch_size = 1024;
cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline, dld);
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, layout, 0, {set}, {}, dld);
cmdbuf.pushConstants(layout, vk::ShaderStageFlagBits::eCompute, 0, sizeof(first), &first,
dld);
cmdbuf.dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1, dld);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_COMPUTE_BIT, 0, sizeof(first), &first);
cmdbuf.Dispatch(Common::AlignUp(num_quads, dispatch_size) / dispatch_size, 1, 1);
const vk::BufferMemoryBarrier barrier(
vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eVertexAttributeRead,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer, 0,
static_cast<vk::DeviceSize>(num_quads) * 6 * sizeof(u32));
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eVertexInput, {}, {}, {barrier}, {}, dld);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = 0;
barrier.size = static_cast<VkDeviceSize>(num_quads) * 6 * sizeof(u32);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {});
});
return {*buffer.handle, 0};
return {buffer.handle.address(), 0};
}
Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
VKDescriptorPool& descriptor_pool, VKStagingBufferPool& staging_buffer_pool,
VKUpdateDescriptorQueue& update_descriptor_queue)
: VKComputePass(device, descriptor_pool,
{vk::DescriptorSetLayoutBinding(0, vk::DescriptorType::eStorageBuffer, 1,
vk::ShaderStageFlagBits::eCompute, nullptr),
vk::DescriptorSetLayoutBinding(1, vk::DescriptorType::eStorageBuffer, 1,
vk::ShaderStageFlagBits::eCompute, nullptr)},
{vk::DescriptorUpdateTemplateEntry(0, 0, 2, vk::DescriptorType::eStorageBuffer,
0, sizeof(DescriptorUpdateEntry))},
{}, std::size(uint8_pass), uint8_pass),
: VKComputePass(device, descriptor_pool, BuildUint8PassDescriptorSetBindings(),
BuildUint8PassDescriptorUpdateTemplateEntry(), {}, std::size(uint8_pass),
uint8_pass),
scheduler{scheduler}, staging_buffer_pool{staging_buffer_pool},
update_descriptor_queue{update_descriptor_queue} {}
Uint8Pass::~Uint8Pass() = default;
std::pair<const vk::Buffer*, u64> Uint8Pass::Assemble(u32 num_vertices, vk::Buffer src_buffer,
u64 src_offset) {
std::pair<const VkBuffer*, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
u64 src_offset) {
const auto staging_size = static_cast<u32>(num_vertices * sizeof(u16));
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(&src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(&*buffer.handle, 0, staging_size);
update_descriptor_queue.AddBuffer(buffer.handle.address(), 0, staging_size);
const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([layout = *layout, pipeline = *pipeline, buffer = *buffer.handle, set,
num_vertices](auto cmdbuf, auto& dld) {
num_vertices](vk::CommandBuffer cmdbuf) {
constexpr u32 dispatch_size = 1024;
cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline, dld);
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, layout, 0, {set}, {}, dld);
cmdbuf.dispatch(Common::AlignUp(num_vertices, dispatch_size) / dispatch_size, 1, 1, dld);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, 0, set, {});
cmdbuf.Dispatch(Common::AlignUp(num_vertices, dispatch_size) / dispatch_size, 1, 1);
const vk::BufferMemoryBarrier barrier(
vk::AccessFlagBits::eShaderWrite, vk::AccessFlagBits::eVertexAttributeRead,
VK_QUEUE_FAMILY_IGNORED, VK_QUEUE_FAMILY_IGNORED, buffer, 0,
static_cast<vk::DeviceSize>(num_vertices) * sizeof(u16));
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eVertexInput, {}, {}, {barrier}, {}, dld);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
barrier.dstAccessMask = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.buffer = buffer;
barrier.offset = 0;
barrier.size = static_cast<VkDeviceSize>(num_vertices * sizeof(u16));
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
});
return {&*buffer.handle, 0};
return {buffer.handle.address(), 0};
}
} // namespace Vulkan

29
src/video_core/renderer_vulkan/vk_compute_pass.h
View file

@ -8,8 +8,8 @@
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -22,24 +22,24 @@ class VKUpdateDescriptorQueue;
class VKComputePass {
public:
explicit VKComputePass(const VKDevice& device, VKDescriptorPool& descriptor_pool,
const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
const std::vector<vk::DescriptorUpdateTemplateEntry>& templates,
const std::vector<vk::PushConstantRange> push_constants,
std::size_t code_size, const u8* code);
vk::Span<VkDescriptorSetLayoutBinding> bindings,
vk::Span<VkDescriptorUpdateTemplateEntryKHR> templates,
vk::Span<VkPushConstantRange> push_constants, std::size_t code_size,
const u8* code);
~VKComputePass();
protected:
vk::DescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue,
VKFence& fence);
VkDescriptorSet CommitDescriptorSet(VKUpdateDescriptorQueue& update_descriptor_queue,
VKFence& fence);
UniqueDescriptorUpdateTemplate descriptor_template;
UniquePipelineLayout layout;
UniquePipeline pipeline;
vk::DescriptorUpdateTemplateKHR descriptor_template;
vk::PipelineLayout layout;
vk::Pipeline pipeline;
private:
UniqueDescriptorSetLayout descriptor_set_layout;
vk::DescriptorSetLayout descriptor_set_layout;
std::optional<DescriptorAllocator> descriptor_allocator;
UniqueShaderModule module;
vk::ShaderModule module;
};
class QuadArrayPass final : public VKComputePass {
@ -50,7 +50,7 @@ public:
VKUpdateDescriptorQueue& update_descriptor_queue);
~QuadArrayPass();
std::pair<const vk::Buffer&, vk::DeviceSize> Assemble(u32 num_vertices, u32 first);
std::pair<const VkBuffer*, VkDeviceSize> Assemble(u32 num_vertices, u32 first);
private:
VKScheduler& scheduler;
@ -65,8 +65,7 @@ public:
VKUpdateDescriptorQueue& update_descriptor_queue);
~Uint8Pass();
std::pair<const vk::Buffer*, u64> Assemble(u32 num_vertices, vk::Buffer src_buffer,
u64 src_offset);
std::pair<const VkBuffer*, u64> Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset);
private:
VKScheduler& scheduler;

127
src/video_core/renderer_vulkan/vk_compute_pipeline.cpp
View file

@ -5,7 +5,6 @@
#include <memory>
#include <vector>
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_device.h"
@ -14,6 +13,7 @@
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -30,7 +30,7 @@ VKComputePipeline::VKComputePipeline(const VKDevice& device, VKScheduler& schedu
VKComputePipeline::~VKComputePipeline() = default;
vk::DescriptorSet VKComputePipeline::CommitDescriptorSet() {
VkDescriptorSet VKComputePipeline::CommitDescriptorSet() {
if (!descriptor_template) {
return {};
}
@ -39,74 +39,109 @@ vk::DescriptorSet VKComputePipeline::CommitDescriptorSet() {
return set;
}
UniqueDescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
std::vector<vk::DescriptorSetLayoutBinding> bindings;
vk::DescriptorSetLayout VKComputePipeline::CreateDescriptorSetLayout() const {
std::vector<VkDescriptorSetLayoutBinding> bindings;
u32 binding = 0;
const auto AddBindings = [&](vk::DescriptorType descriptor_type, std::size_t num_entries) {
const auto add_bindings = [&](VkDescriptorType descriptor_type, std::size_t num_entries) {
// TODO(Rodrigo): Maybe make individual bindings here?
for (u32 bindpoint = 0; bindpoint < static_cast<u32>(num_entries); ++bindpoint) {
bindings.emplace_back(binding++, descriptor_type, 1, vk::ShaderStageFlagBits::eCompute,
nullptr);
VkDescriptorSetLayoutBinding& entry = bindings.emplace_back();
entry.binding = binding++;
entry.descriptorType = descriptor_type;
entry.descriptorCount = 1;
entry.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT;
entry.pImmutableSamplers = nullptr;
}
};
AddBindings(vk::DescriptorType::eUniformBuffer, entries.const_buffers.size());
AddBindings(vk::DescriptorType::eStorageBuffer, entries.global_buffers.size());
AddBindings(vk::DescriptorType::eUniformTexelBuffer, entries.texel_buffers.size());
AddBindings(vk::DescriptorType::eCombinedImageSampler, entries.samplers.size());
AddBindings(vk::DescriptorType::eStorageImage, entries.images.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, entries.const_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, entries.global_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, entries.texel_buffers.size());
add_bindings(VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, entries.samplers.size());
add_bindings(VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, entries.images.size());
const vk::DescriptorSetLayoutCreateInfo descriptor_set_layout_ci(
{}, static_cast<u32>(bindings.size()), bindings.data());
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createDescriptorSetLayoutUnique(descriptor_set_layout_ci, nullptr, dld);
VkDescriptorSetLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.bindingCount = static_cast<u32>(bindings.size());
ci.pBindings = bindings.data();
return device.GetLogical().CreateDescriptorSetLayout(ci);
}
UniquePipelineLayout VKComputePipeline::CreatePipelineLayout() const {
const vk::PipelineLayoutCreateInfo layout_ci({}, 1, &*descriptor_set_layout, 0, nullptr);
const auto dev = device.GetLogical();
return dev.createPipelineLayoutUnique(layout_ci, nullptr, device.GetDispatchLoader());
vk::PipelineLayout VKComputePipeline::CreatePipelineLayout() const {
VkPipelineLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.setLayoutCount = 1;
ci.pSetLayouts = descriptor_set_layout.address();
ci.pushConstantRangeCount = 0;
ci.pPushConstantRanges = nullptr;
return device.GetLogical().CreatePipelineLayout(ci);
}
UniqueDescriptorUpdateTemplate VKComputePipeline::CreateDescriptorUpdateTemplate() const {
std::vector<vk::DescriptorUpdateTemplateEntry> template_entries;
vk::DescriptorUpdateTemplateKHR VKComputePipeline::CreateDescriptorUpdateTemplate() const {
std::vector<VkDescriptorUpdateTemplateEntryKHR> template_entries;
u32 binding = 0;
u32 offset = 0;
FillDescriptorUpdateTemplateEntries(entries, binding, offset, template_entries);
if (template_entries.empty()) {
// If the shader doesn't use descriptor sets, skip template creation.
return UniqueDescriptorUpdateTemplate{};
return {};
}
const vk::DescriptorUpdateTemplateCreateInfo template_ci(
{}, static_cast<u32>(template_entries.size()), template_entries.data(),
vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
vk::PipelineBindPoint::eGraphics, *layout, DESCRIPTOR_SET);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
VkDescriptorUpdateTemplateCreateInfoKHR ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR;
ci.pNext = nullptr;
ci.flags = 0;
ci.descriptorUpdateEntryCount = static_cast<u32>(template_entries.size());
ci.pDescriptorUpdateEntries = template_entries.data();
ci.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR;
ci.descriptorSetLayout = *descriptor_set_layout;
ci.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
ci.pipelineLayout = *layout;
ci.set = DESCRIPTOR_SET;
return device.GetLogical().CreateDescriptorUpdateTemplateKHR(ci);
}
UniqueShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
const vk::ShaderModuleCreateInfo module_ci({}, code.size() * sizeof(u32), code.data());
const auto dev = device.GetLogical();
return dev.createShaderModuleUnique(module_ci, nullptr, device.GetDispatchLoader());
vk::ShaderModule VKComputePipeline::CreateShaderModule(const std::vector<u32>& code) const {
VkShaderModuleCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.codeSize = code.size() * sizeof(u32);
ci.pCode = code.data();
return device.GetLogical().CreateShaderModule(ci);
}
UniquePipeline VKComputePipeline::CreatePipeline() const {
vk::PipelineShaderStageCreateInfo shader_stage_ci({}, vk::ShaderStageFlagBits::eCompute,
*shader_module, "main", nullptr);
vk::PipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
vk::Pipeline VKComputePipeline::CreatePipeline() const {
VkComputePipelineCreateInfo ci;
VkPipelineShaderStageCreateInfo& stage_ci = ci.stage;
stage_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage_ci.pNext = nullptr;
stage_ci.flags = 0;
stage_ci.stage = VK_SHADER_STAGE_COMPUTE_BIT;
stage_ci.module = *shader_module;
stage_ci.pName = "main";
stage_ci.pSpecializationInfo = nullptr;
VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
subgroup_size_ci.sType =
VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT;
subgroup_size_ci.pNext = nullptr;
subgroup_size_ci.requiredSubgroupSize = GuestWarpSize;
if (entries.uses_warps && device.IsGuestWarpSizeSupported(vk::ShaderStageFlagBits::eCompute)) {
shader_stage_ci.pNext = &subgroup_size_ci;
if (entries.uses_warps && device.IsGuestWarpSizeSupported(VK_SHADER_STAGE_COMPUTE_BIT)) {
stage_ci.pNext = &subgroup_size_ci;
}
const vk::ComputePipelineCreateInfo create_info({}, shader_stage_ci, *layout, {}, 0);
const auto dev = device.GetLogical();
return dev.createComputePipelineUnique({}, create_info, nullptr, device.GetDispatchLoader());
ci.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.layout = *layout;
ci.basePipelineHandle = nullptr;
ci.basePipelineIndex = 0;
return device.GetLogical().CreateComputePipeline(ci);
}
} // namespace Vulkan

30
src/video_core/renderer_vulkan/vk_compute_pipeline.h
View file

@ -7,9 +7,9 @@
#include <memory>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -25,42 +25,42 @@ public:
const SPIRVShader& shader);
~VKComputePipeline();
vk::DescriptorSet CommitDescriptorSet();
VkDescriptorSet CommitDescriptorSet();
vk::Pipeline GetHandle() const {
VkPipeline GetHandle() const {
return *pipeline;
}
vk::PipelineLayout GetLayout() const {
VkPipelineLayout GetLayout() const {
return *layout;
}
const ShaderEntries& GetEntries() {
const ShaderEntries& GetEntries() const {
return entries;
}
private:
UniqueDescriptorSetLayout CreateDescriptorSetLayout() const;
vk::DescriptorSetLayout CreateDescriptorSetLayout() const;
UniquePipelineLayout CreatePipelineLayout() const;
vk::PipelineLayout CreatePipelineLayout() const;
UniqueDescriptorUpdateTemplate CreateDescriptorUpdateTemplate() const;
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate() const;
UniqueShaderModule CreateShaderModule(const std::vector<u32>& code) const;
vk::ShaderModule CreateShaderModule(const std::vector<u32>& code) const;
UniquePipeline CreatePipeline() const;
vk::Pipeline CreatePipeline() const;
const VKDevice& device;
VKScheduler& scheduler;
ShaderEntries entries;
UniqueDescriptorSetLayout descriptor_set_layout;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
UniquePipelineLayout layout;
UniqueDescriptorUpdateTemplate descriptor_template;
UniqueShaderModule shader_module;
UniquePipeline pipeline;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
vk::ShaderModule shader_module;
vk::Pipeline pipeline;
};
} // namespace Vulkan

88
src/video_core/renderer_vulkan/vk_descriptor_pool.cpp
View file

@ -6,10 +6,10 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -17,19 +17,18 @@ namespace Vulkan {
constexpr std::size_t SETS_GROW_RATE = 0x20;
DescriptorAllocator::DescriptorAllocator(VKDescriptorPool& descriptor_pool,
vk::DescriptorSetLayout layout)
VkDescriptorSetLayout layout)
: VKFencedPool{SETS_GROW_RATE}, descriptor_pool{descriptor_pool}, layout{layout} {}
DescriptorAllocator::~DescriptorAllocator() = default;
vk::DescriptorSet DescriptorAllocator::Commit(VKFence& fence) {
return *descriptors[CommitResource(fence)];
VkDescriptorSet DescriptorAllocator::Commit(VKFence& fence) {
const std::size_t index = CommitResource(fence);
return descriptors_allocations[index / SETS_GROW_RATE][index % SETS_GROW_RATE];
}
void DescriptorAllocator::Allocate(std::size_t begin, std::size_t end) {
auto new_sets = descriptor_pool.AllocateDescriptors(layout, end - begin);
descriptors.insert(descriptors.end(), std::make_move_iterator(new_sets.begin()),
std::make_move_iterator(new_sets.end()));
descriptors_allocations.push_back(descriptor_pool.AllocateDescriptors(layout, end - begin));
}
VKDescriptorPool::VKDescriptorPool(const VKDevice& device)
@ -37,53 +36,50 @@ VKDescriptorPool::VKDescriptorPool(const VKDevice& device)
VKDescriptorPool::~VKDescriptorPool() = default;
vk::DescriptorPool VKDescriptorPool::AllocateNewPool() {
vk::DescriptorPool* VKDescriptorPool::AllocateNewPool() {
static constexpr u32 num_sets = 0x20000;
static constexpr vk::DescriptorPoolSize pool_sizes[] = {
{vk::DescriptorType::eUniformBuffer, num_sets * 90},
{vk::DescriptorType::eStorageBuffer, num_sets * 60},
{vk::DescriptorType::eUniformTexelBuffer, num_sets * 64},
{vk::DescriptorType::eCombinedImageSampler, num_sets * 64},
{vk::DescriptorType::eStorageImage, num_sets * 40}};
static constexpr VkDescriptorPoolSize pool_sizes[] = {
{VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER, num_sets * 90},
{VK_DESCRIPTOR_TYPE_STORAGE_BUFFER, num_sets * 60},
{VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER, num_sets * 64},
{VK_DESCRIPTOR_TYPE_STORAGE_IMAGE, num_sets * 40}};
const vk::DescriptorPoolCreateInfo create_info(
vk::DescriptorPoolCreateFlagBits::eFreeDescriptorSet, num_sets,
static_cast<u32>(std::size(pool_sizes)), std::data(pool_sizes));
const auto dev = device.GetLogical();
return *pools.emplace_back(
dev.createDescriptorPoolUnique(create_info, nullptr, device.GetDispatchLoader()));
VkDescriptorPoolCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT;
ci.maxSets = num_sets;
ci.poolSizeCount = static_cast<u32>(std::size(pool_sizes));
ci.pPoolSizes = std::data(pool_sizes);
return &pools.emplace_back(device.GetLogical().CreateDescriptorPool(ci));
}
std::vector<UniqueDescriptorSet> VKDescriptorPool::AllocateDescriptors(
vk::DescriptorSetLayout layout, std::size_t count) {
std::vector layout_copies(count, layout);
vk::DescriptorSetAllocateInfo allocate_info(active_pool, static_cast<u32>(count),
layout_copies.data());
vk::DescriptorSets VKDescriptorPool::AllocateDescriptors(VkDescriptorSetLayout layout,
std::size_t count) {
const std::vector layout_copies(count, layout);
VkDescriptorSetAllocateInfo ai;
ai.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO;
ai.pNext = nullptr;
ai.descriptorPool = **active_pool;
ai.descriptorSetCount = static_cast<u32>(count);
ai.pSetLayouts = layout_copies.data();
std::vector<vk::DescriptorSet> sets(count);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
switch (const auto result = dev.allocateDescriptorSets(&allocate_info, sets.data(), dld)) {
case vk::Result::eSuccess:
break;
case vk::Result::eErrorOutOfPoolMemory:
active_pool = AllocateNewPool();
allocate_info.descriptorPool = active_pool;
if (dev.allocateDescriptorSets(&allocate_info, sets.data(), dld) == vk::Result::eSuccess) {
break;
}
[[fallthrough]];
default:
vk::throwResultException(result, "vk::Device::allocateDescriptorSetsUnique");
vk::DescriptorSets sets = active_pool->Allocate(ai);
if (!sets.IsOutOfPoolMemory()) {
return sets;
}
vk::PoolFree deleter(dev, active_pool, dld);
std::vector<UniqueDescriptorSet> unique_sets;
unique_sets.reserve(count);
for (const auto set : sets) {
unique_sets.push_back(UniqueDescriptorSet{set, deleter});
// Our current pool is out of memory. Allocate a new one and retry
active_pool = AllocateNewPool();
ai.descriptorPool = **active_pool;
sets = active_pool->Allocate(ai);
if (!sets.IsOutOfPoolMemory()) {
return sets;
}
return unique_sets;
// After allocating a new pool, we are out of memory again. We can't handle this from here.
throw vk::Exception(VK_ERROR_OUT_OF_POOL_MEMORY);
}
} // namespace Vulkan

19
src/video_core/renderer_vulkan/vk_descriptor_pool.h
View file

@ -8,8 +8,8 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -17,21 +17,21 @@ class VKDescriptorPool;
class DescriptorAllocator final : public VKFencedPool {
public:
explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, vk::DescriptorSetLayout layout);
explicit DescriptorAllocator(VKDescriptorPool& descriptor_pool, VkDescriptorSetLayout layout);
~DescriptorAllocator() override;
DescriptorAllocator(const DescriptorAllocator&) = delete;
vk::DescriptorSet Commit(VKFence& fence);
VkDescriptorSet Commit(VKFence& fence);
protected:
void Allocate(std::size_t begin, std::size_t end) override;
private:
VKDescriptorPool& descriptor_pool;
const vk::DescriptorSetLayout layout;
const VkDescriptorSetLayout layout;
std::vector<UniqueDescriptorSet> descriptors;
std::vector<vk::DescriptorSets> descriptors_allocations;
};
class VKDescriptorPool final {
@ -42,15 +42,14 @@ public:
~VKDescriptorPool();
private:
vk::DescriptorPool AllocateNewPool();
vk::DescriptorPool* AllocateNewPool();
std::vector<UniqueDescriptorSet> AllocateDescriptors(vk::DescriptorSetLayout layout,
std::size_t count);
vk::DescriptorSets AllocateDescriptors(VkDescriptorSetLayout layout, std::size_t count);
const VKDevice& device;
std::vector<UniqueDescriptorPool> pools;
vk::DescriptorPool active_pool;
std::vector<vk::DescriptorPool> pools;
vk::DescriptorPool* active_pool;
};
} // namespace Vulkan

562
src/video_core/renderer_vulkan/vk_device.cpp
View file

@ -6,15 +6,15 @@
#include <chrono>
#include <cstdlib>
#include <optional>
#include <set>
#include <string_view>
#include <thread>
#include <unordered_set>
#include <vector>
#include "common/assert.h"
#include "core/settings.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -22,49 +22,43 @@ namespace {
namespace Alternatives {
constexpr std::array Depth24UnormS8Uint = {vk::Format::eD32SfloatS8Uint,
vk::Format::eD16UnormS8Uint, vk::Format{}};
constexpr std::array Depth16UnormS8Uint = {vk::Format::eD24UnormS8Uint,
vk::Format::eD32SfloatS8Uint, vk::Format{}};
constexpr std::array Depth24UnormS8_UINT = {VK_FORMAT_D32_SFLOAT_S8_UINT,
VK_FORMAT_D16_UNORM_S8_UINT, VkFormat{}};
constexpr std::array Depth16UnormS8_UINT = {VK_FORMAT_D24_UNORM_S8_UINT,
VK_FORMAT_D32_SFLOAT_S8_UINT, VkFormat{}};
} // namespace Alternatives
constexpr std::array REQUIRED_EXTENSIONS = {
VK_KHR_SWAPCHAIN_EXTENSION_NAME,
VK_KHR_16BIT_STORAGE_EXTENSION_NAME,
VK_KHR_8BIT_STORAGE_EXTENSION_NAME,
VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME,
VK_KHR_DESCRIPTOR_UPDATE_TEMPLATE_EXTENSION_NAME,
VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME,
VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME,
VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME,
VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME,
};
template <typename T>
void SetNext(void**& next, T& data) {
*next = &data;
next = &data.pNext;
}
template <typename T>
T GetFeatures(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dld) {
vk::PhysicalDeviceFeatures2 features;
T extension_features;
features.pNext = &extension_features;
physical.getFeatures2(&features, dld);
return extension_features;
}
template <typename T>
T GetProperties(vk::PhysicalDevice physical, const vk::DispatchLoaderDynamic& dld) {
vk::PhysicalDeviceProperties2 properties;
T extension_properties;
properties.pNext = &extension_properties;
physical.getProperties2(&properties, dld);
return extension_properties;
}
constexpr const vk::Format* GetFormatAlternatives(vk::Format format) {
constexpr const VkFormat* GetFormatAlternatives(VkFormat format) {
switch (format) {
case vk::Format::eD24UnormS8Uint:
return Alternatives::Depth24UnormS8Uint.data();
case vk::Format::eD16UnormS8Uint:
return Alternatives::Depth16UnormS8Uint.data();
case VK_FORMAT_D24_UNORM_S8_UINT:
return Alternatives::Depth24UnormS8_UINT.data();
case VK_FORMAT_D16_UNORM_S8_UINT:
return Alternatives::Depth16UnormS8_UINT.data();
default:
return nullptr;
}
}
vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties, FormatType format_type) {
VkFormatFeatureFlags GetFormatFeatures(VkFormatProperties properties, FormatType format_type) {
switch (format_type) {
case FormatType::Linear:
return properties.linearTilingFeatures;
@ -77,79 +71,220 @@ vk::FormatFeatureFlags GetFormatFeatures(vk::FormatProperties properties, Format
}
}
std::unordered_map<VkFormat, VkFormatProperties> GetFormatProperties(
vk::PhysicalDevice physical, const vk::InstanceDispatch& dld) {
static constexpr std::array formats{VK_FORMAT_A8B8G8R8_UNORM_PACK32,
VK_FORMAT_A8B8G8R8_UINT_PACK32,
VK_FORMAT_A8B8G8R8_SNORM_PACK32,
VK_FORMAT_A8B8G8R8_SRGB_PACK32,
VK_FORMAT_B5G6R5_UNORM_PACK16,
VK_FORMAT_A2B10G10R10_UNORM_PACK32,
VK_FORMAT_A1R5G5B5_UNORM_PACK16,
VK_FORMAT_R32G32B32A32_SFLOAT,
VK_FORMAT_R32G32B32A32_UINT,
VK_FORMAT_R32G32_SFLOAT,
VK_FORMAT_R32G32_UINT,
VK_FORMAT_R16G16B16A16_UINT,
VK_FORMAT_R16G16B16A16_SNORM,
VK_FORMAT_R16G16B16A16_UNORM,
VK_FORMAT_R16G16_UNORM,
VK_FORMAT_R16G16_SNORM,
VK_FORMAT_R16G16_SFLOAT,
VK_FORMAT_R16_UNORM,
VK_FORMAT_R8G8B8A8_SRGB,
VK_FORMAT_R8G8_UNORM,
VK_FORMAT_R8G8_SNORM,
VK_FORMAT_R8_UNORM,
VK_FORMAT_R8_UINT,
VK_FORMAT_B10G11R11_UFLOAT_PACK32,
VK_FORMAT_R32_SFLOAT,
VK_FORMAT_R32_UINT,
VK_FORMAT_R32_SINT,
VK_FORMAT_R16_SFLOAT,
VK_FORMAT_R16G16B16A16_SFLOAT,
VK_FORMAT_B8G8R8A8_UNORM,
VK_FORMAT_R4G4B4A4_UNORM_PACK16,
VK_FORMAT_D32_SFLOAT,
VK_FORMAT_D16_UNORM,
VK_FORMAT_D16_UNORM_S8_UINT,
VK_FORMAT_D24_UNORM_S8_UINT,
VK_FORMAT_D32_SFLOAT_S8_UINT,
VK_FORMAT_BC1_RGBA_UNORM_BLOCK,
VK_FORMAT_BC2_UNORM_BLOCK,
VK_FORMAT_BC3_UNORM_BLOCK,
VK_FORMAT_BC4_UNORM_BLOCK,
VK_FORMAT_BC5_UNORM_BLOCK,
VK_FORMAT_BC5_SNORM_BLOCK,
VK_FORMAT_BC7_UNORM_BLOCK,
VK_FORMAT_BC6H_UFLOAT_BLOCK,
VK_FORMAT_BC6H_SFLOAT_BLOCK,
VK_FORMAT_BC1_RGBA_SRGB_BLOCK,
VK_FORMAT_BC2_SRGB_BLOCK,
VK_FORMAT_BC3_SRGB_BLOCK,
VK_FORMAT_BC7_SRGB_BLOCK,
VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
VK_FORMAT_ASTC_5x5_UNORM_BLOCK,
VK_FORMAT_ASTC_5x5_SRGB_BLOCK,
VK_FORMAT_ASTC_10x8_UNORM_BLOCK,
VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
VK_FORMAT_ASTC_6x6_UNORM_BLOCK,
VK_FORMAT_ASTC_6x6_SRGB_BLOCK,
VK_FORMAT_ASTC_10x10_UNORM_BLOCK,
VK_FORMAT_ASTC_10x10_SRGB_BLOCK,
VK_FORMAT_ASTC_12x12_UNORM_BLOCK,
VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
VK_FORMAT_ASTC_8x6_UNORM_BLOCK,
VK_FORMAT_ASTC_8x6_SRGB_BLOCK,
VK_FORMAT_ASTC_6x5_UNORM_BLOCK,
VK_FORMAT_ASTC_6x5_SRGB_BLOCK,
VK_FORMAT_E5B9G9R9_UFLOAT_PACK32};
std::unordered_map<VkFormat, VkFormatProperties> format_properties;
for (const auto format : formats) {
format_properties.emplace(format, physical.GetFormatProperties(format));
}
return format_properties;
}
} // Anonymous namespace
VKDevice::VKDevice(const vk::DispatchLoaderDynamic& dld, vk::PhysicalDevice physical,
vk::SurfaceKHR surface)
: dld{dld}, physical{physical}, properties{physical.getProperties(dld)},
format_properties{GetFormatProperties(dld, physical)} {
VKDevice::VKDevice(VkInstance instance, vk::PhysicalDevice physical, VkSurfaceKHR surface,
const vk::InstanceDispatch& dld)
: dld{dld}, physical{physical}, properties{physical.GetProperties()},
format_properties{GetFormatProperties(physical, dld)} {
SetupFamilies(surface);
SetupFeatures();
}
VKDevice::~VKDevice() = default;
bool VKDevice::Create(vk::Instance instance) {
bool VKDevice::Create() {
const auto queue_cis = GetDeviceQueueCreateInfos();
const std::vector extensions = LoadExtensions();
vk::PhysicalDeviceFeatures2 features2;
VkPhysicalDeviceFeatures2 features2;
features2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2;
features2.pNext = nullptr;
void** next = &features2.pNext;
auto& features = features2.features;
features.vertexPipelineStoresAndAtomics = true;
features.robustBufferAccess = false;
features.fullDrawIndexUint32 = false;
features.imageCubeArray = false;
features.independentBlend = true;
features.depthClamp = true;
features.samplerAnisotropy = true;
features.largePoints = true;
features.multiViewport = true;
features.depthBiasClamp = true;
features.geometryShader = true;
features.tessellationShader = true;
features.sampleRateShading = false;
features.dualSrcBlend = false;
features.logicOp = false;
features.multiDrawIndirect = false;
features.drawIndirectFirstInstance = false;
features.depthClamp = true;
features.depthBiasClamp = true;
features.fillModeNonSolid = false;
features.depthBounds = false;
features.wideLines = false;
features.largePoints = true;
features.alphaToOne = false;
features.multiViewport = true;
features.samplerAnisotropy = true;
features.textureCompressionETC2 = false;
features.textureCompressionASTC_LDR = is_optimal_astc_supported;
features.textureCompressionBC = false;
features.occlusionQueryPrecise = true;
features.pipelineStatisticsQuery = false;
features.vertexPipelineStoresAndAtomics = true;
features.fragmentStoresAndAtomics = true;
features.shaderTessellationAndGeometryPointSize = false;
features.shaderImageGatherExtended = true;
features.shaderStorageImageExtendedFormats = false;
features.shaderStorageImageMultisample = false;
features.shaderStorageImageReadWithoutFormat = is_formatless_image_load_supported;
features.shaderStorageImageWriteWithoutFormat = true;
features.textureCompressionASTC_LDR = is_optimal_astc_supported;
features.shaderUniformBufferArrayDynamicIndexing = false;
features.shaderSampledImageArrayDynamicIndexing = false;
features.shaderStorageBufferArrayDynamicIndexing = false;
features.shaderStorageImageArrayDynamicIndexing = false;
features.shaderClipDistance = false;
features.shaderCullDistance = false;
features.shaderFloat64 = false;
features.shaderInt64 = false;
features.shaderInt16 = false;
features.shaderResourceResidency = false;
features.shaderResourceMinLod = false;
features.sparseBinding = false;
features.sparseResidencyBuffer = false;
features.sparseResidencyImage2D = false;
features.sparseResidencyImage3D = false;
features.sparseResidency2Samples = false;
features.sparseResidency4Samples = false;
features.sparseResidency8Samples = false;
features.sparseResidency16Samples = false;
features.sparseResidencyAliased = false;
features.variableMultisampleRate = false;
features.inheritedQueries = false;
vk::PhysicalDevice16BitStorageFeaturesKHR bit16_storage;
VkPhysicalDevice16BitStorageFeaturesKHR bit16_storage;
bit16_storage.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR;
bit16_storage.pNext = nullptr;
bit16_storage.storageBuffer16BitAccess = false;
bit16_storage.uniformAndStorageBuffer16BitAccess = true;
bit16_storage.storagePushConstant16 = false;
bit16_storage.storageInputOutput16 = false;
SetNext(next, bit16_storage);
vk::PhysicalDevice8BitStorageFeaturesKHR bit8_storage;
VkPhysicalDevice8BitStorageFeaturesKHR bit8_storage;
bit8_storage.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR;
bit8_storage.pNext = nullptr;
bit8_storage.storageBuffer8BitAccess = false;
bit8_storage.uniformAndStorageBuffer8BitAccess = true;
bit8_storage.storagePushConstant8 = false;
SetNext(next, bit8_storage);
vk::PhysicalDeviceHostQueryResetFeaturesEXT host_query_reset;
VkPhysicalDeviceHostQueryResetFeaturesEXT host_query_reset;
host_query_reset.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT;
host_query_reset.hostQueryReset = true;
SetNext(next, host_query_reset);
vk::PhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8;
if (is_float16_supported) {
float16_int8.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR;
float16_int8.pNext = nullptr;
float16_int8.shaderFloat16 = true;
float16_int8.shaderInt8 = false;
SetNext(next, float16_int8);
} else {
LOG_INFO(Render_Vulkan, "Device doesn't support float16 natively");
}
vk::PhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout;
VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR std430_layout;
if (khr_uniform_buffer_standard_layout) {
std430_layout.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR;
std430_layout.pNext = nullptr;
std430_layout.uniformBufferStandardLayout = true;
SetNext(next, std430_layout);
} else {
LOG_INFO(Render_Vulkan, "Device doesn't support packed UBOs");
}
vk::PhysicalDeviceIndexTypeUint8FeaturesEXT index_type_uint8;
VkPhysicalDeviceIndexTypeUint8FeaturesEXT index_type_uint8;
if (ext_index_type_uint8) {
index_type_uint8.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT;
index_type_uint8.pNext = nullptr;
index_type_uint8.indexTypeUint8 = true;
SetNext(next, index_type_uint8);
} else {
LOG_INFO(Render_Vulkan, "Device doesn't support uint8 indexes");
}
vk::PhysicalDeviceTransformFeedbackFeaturesEXT transform_feedback;
VkPhysicalDeviceTransformFeedbackFeaturesEXT transform_feedback;
if (ext_transform_feedback) {
transform_feedback.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT;
transform_feedback.pNext = nullptr;
transform_feedback.transformFeedback = true;
transform_feedback.geometryStreams = true;
SetNext(next, transform_feedback);
@ -161,60 +296,48 @@ bool VKDevice::Create(vk::Instance instance) {
LOG_INFO(Render_Vulkan, "Device doesn't support depth range unrestricted");
}
vk::DeviceCreateInfo device_ci({}, static_cast<u32>(queue_cis.size()), queue_cis.data(), 0,
nullptr, static_cast<u32>(extensions.size()), extensions.data(),
nullptr);
device_ci.pNext = &features2;
vk::Device unsafe_logical;
if (physical.createDevice(&device_ci, nullptr, &unsafe_logical, dld) != vk::Result::eSuccess) {
LOG_CRITICAL(Render_Vulkan, "Logical device failed to be created!");
logical = vk::Device::Create(physical, queue_cis, extensions, features2, dld);
if (!logical) {
LOG_ERROR(Render_Vulkan, "Failed to create logical device");
return false;
}
dld.init(instance, dld.vkGetInstanceProcAddr, unsafe_logical);
logical = UniqueDevice(unsafe_logical, {nullptr, dld});
CollectTelemetryParameters();
graphics_queue = logical->getQueue(graphics_family, 0, dld);
present_queue = logical->getQueue(present_family, 0, dld);
graphics_queue = logical.GetQueue(graphics_family);
present_queue = logical.GetQueue(present_family);
return true;
}
vk::Format VKDevice::GetSupportedFormat(vk::Format wanted_format,
vk::FormatFeatureFlags wanted_usage,
FormatType format_type) const {
VkFormat VKDevice::GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
FormatType format_type) const {
if (IsFormatSupported(wanted_format, wanted_usage, format_type)) {
return wanted_format;
}
// The wanted format is not supported by hardware, search for alternatives
const vk::Format* alternatives = GetFormatAlternatives(wanted_format);
const VkFormat* alternatives = GetFormatAlternatives(wanted_format);
if (alternatives == nullptr) {
UNREACHABLE_MSG("Format={} with usage={} and type={} has no defined alternatives and host "
"hardware does not support it",
vk::to_string(wanted_format), vk::to_string(wanted_usage),
static_cast<u32>(format_type));
wanted_format, wanted_usage, format_type);
return wanted_format;
}
std::size_t i = 0;
for (vk::Format alternative = alternatives[0]; alternative != vk::Format{};
alternative = alternatives[++i]) {
for (VkFormat alternative = *alternatives; alternative; alternative = alternatives[++i]) {
if (!IsFormatSupported(alternative, wanted_usage, format_type)) {
continue;
}
LOG_WARNING(Render_Vulkan,
"Emulating format={} with alternative format={} with usage={} and type={}",
static_cast<u32>(wanted_format), static_cast<u32>(alternative),
static_cast<u32>(wanted_usage), static_cast<u32>(format_type));
wanted_format, alternative, wanted_usage, format_type);
return alternative;
}
// No alternatives found, panic
UNREACHABLE_MSG("Format={} with usage={} and type={} is not supported by the host hardware and "
"doesn't support any of the alternatives",
static_cast<u32>(wanted_format), static_cast<u32>(wanted_usage),
static_cast<u32>(format_type));
wanted_format, wanted_usage, format_type);
return wanted_format;
}
@ -228,38 +351,39 @@ void VKDevice::ReportLoss() const {
return;
}
[[maybe_unused]] const std::vector data = graphics_queue.getCheckpointDataNV(dld);
[[maybe_unused]] const std::vector data = graphics_queue.GetCheckpointDataNV(dld);
// Catch here in debug builds (or with optimizations disabled) the last graphics pipeline to be
// executed. It can be done on a debugger by evaluating the expression:
// *(VKGraphicsPipeline*)data[0]
}
bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features) const {
bool VKDevice::IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const {
// Disable for now to avoid converting ASTC twice.
static constexpr std::array astc_formats = {
vk::Format::eAstc4x4UnormBlock, vk::Format::eAstc4x4SrgbBlock,
vk::Format::eAstc5x4UnormBlock, vk::Format::eAstc5x4SrgbBlock,
vk::Format::eAstc5x5UnormBlock, vk::Format::eAstc5x5SrgbBlock,
vk::Format::eAstc6x5UnormBlock, vk::Format::eAstc6x5SrgbBlock,
vk::Format::eAstc6x6UnormBlock, vk::Format::eAstc6x6SrgbBlock,
vk::Format::eAstc8x5UnormBlock, vk::Format::eAstc8x5SrgbBlock,
vk::Format::eAstc8x6UnormBlock, vk::Format::eAstc8x6SrgbBlock,
vk::Format::eAstc8x8UnormBlock, vk::Format::eAstc8x8SrgbBlock,
vk::Format::eAstc10x5UnormBlock, vk::Format::eAstc10x5SrgbBlock,
vk::Format::eAstc10x6UnormBlock, vk::Format::eAstc10x6SrgbBlock,
vk::Format::eAstc10x8UnormBlock, vk::Format::eAstc10x8SrgbBlock,
vk::Format::eAstc10x10UnormBlock, vk::Format::eAstc10x10SrgbBlock,
vk::Format::eAstc12x10UnormBlock, vk::Format::eAstc12x10SrgbBlock,
vk::Format::eAstc12x12UnormBlock, vk::Format::eAstc12x12SrgbBlock};
VK_FORMAT_ASTC_4x4_UNORM_BLOCK, VK_FORMAT_ASTC_4x4_SRGB_BLOCK,
VK_FORMAT_ASTC_5x4_UNORM_BLOCK, VK_FORMAT_ASTC_5x4_SRGB_BLOCK,
VK_FORMAT_ASTC_5x5_UNORM_BLOCK, VK_FORMAT_ASTC_5x5_SRGB_BLOCK,
VK_FORMAT_ASTC_6x5_UNORM_BLOCK, VK_FORMAT_ASTC_6x5_SRGB_BLOCK,
VK_FORMAT_ASTC_6x6_UNORM_BLOCK, VK_FORMAT_ASTC_6x6_SRGB_BLOCK,
VK_FORMAT_ASTC_8x5_UNORM_BLOCK, VK_FORMAT_ASTC_8x5_SRGB_BLOCK,
VK_FORMAT_ASTC_8x6_UNORM_BLOCK, VK_FORMAT_ASTC_8x6_SRGB_BLOCK,
VK_FORMAT_ASTC_8x8_UNORM_BLOCK, VK_FORMAT_ASTC_8x8_SRGB_BLOCK,
VK_FORMAT_ASTC_10x5_UNORM_BLOCK, VK_FORMAT_ASTC_10x5_SRGB_BLOCK,
VK_FORMAT_ASTC_10x6_UNORM_BLOCK, VK_FORMAT_ASTC_10x6_SRGB_BLOCK,
VK_FORMAT_ASTC_10x8_UNORM_BLOCK, VK_FORMAT_ASTC_10x8_SRGB_BLOCK,
VK_FORMAT_ASTC_10x10_UNORM_BLOCK, VK_FORMAT_ASTC_10x10_SRGB_BLOCK,
VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK,
VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
};
if (!features.textureCompressionASTC_LDR) {
return false;
}
const auto format_feature_usage{
vk::FormatFeatureFlagBits::eSampledImage | vk::FormatFeatureFlagBits::eBlitSrc |
vk::FormatFeatureFlagBits::eBlitDst | vk::FormatFeatureFlagBits::eTransferSrc |
vk::FormatFeatureFlagBits::eTransferDst};
VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT | VK_FORMAT_FEATURE_BLIT_SRC_BIT |
VK_FORMAT_FEATURE_BLIT_DST_BIT | VK_FORMAT_FEATURE_TRANSFER_SRC_BIT |
VK_FORMAT_FEATURE_TRANSFER_DST_BIT};
for (const auto format : astc_formats) {
const auto format_properties{physical.getFormatProperties(format, dld)};
const auto format_properties{physical.GetFormatProperties(format)};
if (!(format_properties.optimalTilingFeatures & format_feature_usage)) {
return false;
}
@ -267,61 +391,49 @@ bool VKDevice::IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features
return true;
}
bool VKDevice::IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlags wanted_usage,
bool VKDevice::IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
FormatType format_type) const {
const auto it = format_properties.find(wanted_format);
if (it == format_properties.end()) {
UNIMPLEMENTED_MSG("Unimplemented format query={}", vk::to_string(wanted_format));
UNIMPLEMENTED_MSG("Unimplemented format query={}", wanted_format);
return true;
}
const auto supported_usage = GetFormatFeatures(it->second, format_type);
return (supported_usage & wanted_usage) == wanted_usage;
}
bool VKDevice::IsSuitable(vk::PhysicalDevice physical, vk::SurfaceKHR surface,
const vk::DispatchLoaderDynamic& dld) {
static constexpr std::array required_extensions = {
VK_KHR_SWAPCHAIN_EXTENSION_NAME,
VK_KHR_16BIT_STORAGE_EXTENSION_NAME,
VK_KHR_8BIT_STORAGE_EXTENSION_NAME,
VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME,
VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME,
VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME,
VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME,
VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME,
};
bool VKDevice::IsSuitable(vk::PhysicalDevice physical, VkSurfaceKHR surface) {
bool is_suitable = true;
std::bitset<required_extensions.size()> available_extensions{};
std::bitset<REQUIRED_EXTENSIONS.size()> available_extensions;
for (const auto& prop : physical.enumerateDeviceExtensionProperties(nullptr, dld)) {
for (std::size_t i = 0; i < required_extensions.size(); ++i) {
for (const auto& prop : physical.EnumerateDeviceExtensionProperties()) {
for (std::size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) {
if (available_extensions[i]) {
continue;
}
available_extensions[i] =
required_extensions[i] == std::string_view{prop.extensionName};
const std::string_view name{prop.extensionName};
available_extensions[i] = name == REQUIRED_EXTENSIONS[i];
}
}
if (!available_extensions.all()) {
for (std::size_t i = 0; i < required_extensions.size(); ++i) {
for (std::size_t i = 0; i < REQUIRED_EXTENSIONS.size(); ++i) {
if (available_extensions[i]) {
continue;
}
LOG_ERROR(Render_Vulkan, "Missing required extension: {}", required_extensions[i]);
LOG_ERROR(Render_Vulkan, "Missing required extension: {}", REQUIRED_EXTENSIONS[i]);
is_suitable = false;
}
}
bool has_graphics{}, has_present{};
const auto queue_family_properties = physical.getQueueFamilyProperties(dld);
const std::vector queue_family_properties = physical.GetQueueFamilyProperties();
for (u32 i = 0; i < static_cast<u32>(queue_family_properties.size()); ++i) {
const auto& family = queue_family_properties[i];
if (family.queueCount == 0) {
continue;
}
has_graphics |=
(family.queueFlags & vk::QueueFlagBits::eGraphics) != static_cast<vk::QueueFlagBits>(0);
has_present |= physical.getSurfaceSupportKHR(i, surface, dld) != 0;
has_graphics |= family.queueFlags & VK_QUEUE_GRAPHICS_BIT;
has_present |= physical.GetSurfaceSupportKHR(i, surface);
}
if (!has_graphics || !has_present) {
LOG_ERROR(Render_Vulkan, "Device lacks a graphics and present queue");
@ -329,7 +441,7 @@ bool VKDevice::IsSuitable(vk::PhysicalDevice physical, vk::SurfaceKHR surface,
}
// TODO(Rodrigo): Check if the device matches all requeriments.
const auto properties{physical.getProperties(dld)};
const auto properties{physical.GetProperties()};
const auto& limits{properties.limits};
constexpr u32 required_ubo_size = 65536;
@ -346,7 +458,7 @@ bool VKDevice::IsSuitable(vk::PhysicalDevice physical, vk::SurfaceKHR surface,
is_suitable = false;
}
const auto features{physical.getFeatures(dld)};
const auto features{physical.GetFeatures()};
const std::array feature_report = {
std::make_pair(features.vertexPipelineStoresAndAtomics, "vertexPipelineStoresAndAtomics"),
std::make_pair(features.independentBlend, "independentBlend"),
@ -380,7 +492,7 @@ bool VKDevice::IsSuitable(vk::PhysicalDevice physical, vk::SurfaceKHR surface,
std::vector<const char*> VKDevice::LoadExtensions() {
std::vector<const char*> extensions;
const auto Test = [&](const vk::ExtensionProperties& extension,
const auto Test = [&](const VkExtensionProperties& extension,
std::optional<std::reference_wrapper<bool>> status, const char* name,
bool push) {
if (extension.extensionName != std::string_view(name)) {
@ -394,22 +506,13 @@ std::vector<const char*> VKDevice::LoadExtensions() {
}
};
extensions.reserve(15);
extensions.push_back(VK_KHR_SWAPCHAIN_EXTENSION_NAME);
extensions.push_back(VK_KHR_16BIT_STORAGE_EXTENSION_NAME);
extensions.push_back(VK_KHR_8BIT_STORAGE_EXTENSION_NAME);
extensions.push_back(VK_KHR_DRIVER_PROPERTIES_EXTENSION_NAME);
extensions.push_back(VK_EXT_VERTEX_ATTRIBUTE_DIVISOR_EXTENSION_NAME);
extensions.push_back(VK_EXT_SHADER_SUBGROUP_BALLOT_EXTENSION_NAME);
extensions.push_back(VK_EXT_SHADER_SUBGROUP_VOTE_EXTENSION_NAME);
extensions.push_back(VK_EXT_HOST_QUERY_RESET_EXTENSION_NAME);
extensions.reserve(7 + REQUIRED_EXTENSIONS.size());
extensions.insert(extensions.begin(), REQUIRED_EXTENSIONS.begin(), REQUIRED_EXTENSIONS.end());
[[maybe_unused]] const bool nsight =
std::getenv("NVTX_INJECTION64_PATH") || std::getenv("NSIGHT_LAUNCHED");
bool has_khr_shader_float16_int8{};
bool has_ext_subgroup_size_control{};
bool has_ext_transform_feedback{};
for (const auto& extension : physical.enumerateDeviceExtensionProperties(nullptr, dld)) {
for (const auto& extension : physical.EnumerateDeviceExtensionProperties()) {
Test(extension, khr_uniform_buffer_standard_layout,
VK_KHR_UNIFORM_BUFFER_STANDARD_LAYOUT_EXTENSION_NAME, true);
Test(extension, has_khr_shader_float16_int8, VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME,
@ -429,38 +532,67 @@ std::vector<const char*> VKDevice::LoadExtensions() {
}
}
VkPhysicalDeviceFeatures2KHR features;
features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2_KHR;
VkPhysicalDeviceProperties2KHR properties;
properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR;
if (has_khr_shader_float16_int8) {
is_float16_supported =
GetFeatures<vk::PhysicalDeviceFloat16Int8FeaturesKHR>(physical, dld).shaderFloat16;
VkPhysicalDeviceFloat16Int8FeaturesKHR float16_int8_features;
float16_int8_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR;
float16_int8_features.pNext = nullptr;
features.pNext = &float16_int8_features;
physical.GetFeatures2KHR(features);
is_float16_supported = float16_int8_features.shaderFloat16;
extensions.push_back(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME);
}
if (has_ext_subgroup_size_control) {
const auto features =
GetFeatures<vk::PhysicalDeviceSubgroupSizeControlFeaturesEXT>(physical, dld);
const auto properties =
GetProperties<vk::PhysicalDeviceSubgroupSizeControlPropertiesEXT>(physical, dld);
VkPhysicalDeviceSubgroupSizeControlFeaturesEXT subgroup_features;
subgroup_features.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT;
subgroup_features.pNext = nullptr;
features.pNext = &subgroup_features;
physical.GetFeatures2KHR(features);
is_warp_potentially_bigger = properties.maxSubgroupSize > GuestWarpSize;
VkPhysicalDeviceSubgroupSizeControlPropertiesEXT subgroup_properties;
subgroup_properties.sType =
VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_PROPERTIES_EXT;
subgroup_properties.pNext = nullptr;
properties.pNext = &subgroup_properties;
physical.GetProperties2KHR(properties);
if (features.subgroupSizeControl && properties.minSubgroupSize <= GuestWarpSize &&
properties.maxSubgroupSize >= GuestWarpSize) {
is_warp_potentially_bigger = subgroup_properties.maxSubgroupSize > GuestWarpSize;
if (subgroup_features.subgroupSizeControl &&
subgroup_properties.minSubgroupSize <= GuestWarpSize &&
subgroup_properties.maxSubgroupSize >= GuestWarpSize) {
extensions.push_back(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
guest_warp_stages = properties.requiredSubgroupSizeStages;
guest_warp_stages = subgroup_properties.requiredSubgroupSizeStages;
}
} else {
is_warp_potentially_bigger = true;
}
if (has_ext_transform_feedback) {
const auto features =
GetFeatures<vk::PhysicalDeviceTransformFeedbackFeaturesEXT>(physical, dld);
const auto properties =
GetProperties<vk::PhysicalDeviceTransformFeedbackPropertiesEXT>(physical, dld);
VkPhysicalDeviceTransformFeedbackFeaturesEXT tfb_features;
tfb_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT;
tfb_features.pNext = nullptr;
features.pNext = &tfb_features;
physical.GetFeatures2KHR(features);
if (features.transformFeedback && features.geometryStreams &&
properties.maxTransformFeedbackStreams >= 4 && properties.maxTransformFeedbackBuffers &&
properties.transformFeedbackQueries && properties.transformFeedbackDraw) {
VkPhysicalDeviceTransformFeedbackPropertiesEXT tfb_properties;
tfb_properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT;
tfb_properties.pNext = nullptr;
properties.pNext = &tfb_properties;
physical.GetProperties2KHR(properties);
if (tfb_features.transformFeedback && tfb_features.geometryStreams &&
tfb_properties.maxTransformFeedbackStreams >= 4 &&
tfb_properties.maxTransformFeedbackBuffers && tfb_properties.transformFeedbackQueries &&
tfb_properties.transformFeedbackDraw) {
extensions.push_back(VK_EXT_TRANSFORM_FEEDBACK_EXTENSION_NAME);
ext_transform_feedback = true;
}
@ -469,10 +601,10 @@ std::vector<const char*> VKDevice::LoadExtensions() {
return extensions;
}
void VKDevice::SetupFamilies(vk::SurfaceKHR surface) {
void VKDevice::SetupFamilies(VkSurfaceKHR surface) {
std::optional<u32> graphics_family_, present_family_;
const auto queue_family_properties = physical.getQueueFamilyProperties(dld);
const std::vector queue_family_properties = physical.GetQueueFamilyProperties();
for (u32 i = 0; i < static_cast<u32>(queue_family_properties.size()); ++i) {
if (graphics_family_ && present_family_)
break;
@ -481,10 +613,10 @@ void VKDevice::SetupFamilies(vk::SurfaceKHR surface) {
if (queue_family.queueCount == 0)
continue;
if (queue_family.queueFlags & vk::QueueFlagBits::eGraphics) {
if (queue_family.queueFlags & VK_QUEUE_GRAPHICS_BIT) {
graphics_family_ = i;
}
if (physical.getSurfaceSupportKHR(i, surface, dld)) {
if (physical.GetSurfaceSupportKHR(i, surface)) {
present_family_ = i;
}
}
@ -495,120 +627,48 @@ void VKDevice::SetupFamilies(vk::SurfaceKHR surface) {
}
void VKDevice::SetupFeatures() {
const auto supported_features{physical.getFeatures(dld)};
const auto supported_features{physical.GetFeatures()};
is_formatless_image_load_supported = supported_features.shaderStorageImageReadWithoutFormat;
is_optimal_astc_supported = IsOptimalAstcSupported(supported_features);
}
void VKDevice::CollectTelemetryParameters() {
const auto driver = GetProperties<vk::PhysicalDeviceDriverPropertiesKHR>(physical, dld);
VkPhysicalDeviceDriverPropertiesKHR driver;
driver.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR;
driver.pNext = nullptr;
VkPhysicalDeviceProperties2KHR properties;
properties.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROPERTIES_2_KHR;
properties.pNext = &driver;
physical.GetProperties2KHR(properties);
driver_id = driver.driverID;
vendor_name = driver.driverName;
const auto extensions = physical.enumerateDeviceExtensionProperties(nullptr, dld);
const std::vector extensions = physical.EnumerateDeviceExtensionProperties();
reported_extensions.reserve(std::size(extensions));
for (const auto& extension : extensions) {
reported_extensions.push_back(extension.extensionName);
}
}
std::vector<vk::DeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() const {
static const float QUEUE_PRIORITY = 1.0f;
std::vector<VkDeviceQueueCreateInfo> VKDevice::GetDeviceQueueCreateInfos() const {
static constexpr float QUEUE_PRIORITY = 1.0f;
std::set<u32> unique_queue_families = {graphics_family, present_family};
std::vector<vk::DeviceQueueCreateInfo> queue_cis;
std::unordered_set<u32> unique_queue_families = {graphics_family, present_family};
std::vector<VkDeviceQueueCreateInfo> queue_cis;
for (u32 queue_family : unique_queue_families)
queue_cis.push_back({{}, queue_family, 1, &QUEUE_PRIORITY});
for (const u32 queue_family : unique_queue_families) {
VkDeviceQueueCreateInfo& ci = queue_cis.emplace_back();
ci.sType = VK_STRUCTURE_TYPE_DEVICE_QUEUE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.queueFamilyIndex = queue_family;
ci.queueCount = 1;
ci.pQueuePriorities = &QUEUE_PRIORITY;
}
return queue_cis;
}
std::unordered_map<vk::Format, vk::FormatProperties> VKDevice::GetFormatProperties(
const vk::DispatchLoaderDynamic& dld, vk::PhysicalDevice physical) {
static constexpr std::array formats{vk::Format::eA8B8G8R8UnormPack32,
vk::Format::eA8B8G8R8UintPack32,
vk::Format::eA8B8G8R8SnormPack32,
vk::Format::eA8B8G8R8SrgbPack32,
vk::Format::eB5G6R5UnormPack16,
vk::Format::eA2B10G10R10UnormPack32,
vk::Format::eA1R5G5B5UnormPack16,
vk::Format::eR32G32B32A32Sfloat,
vk::Format::eR32G32B32A32Uint,
vk::Format::eR32G32Sfloat,
vk::Format::eR32G32Uint,
vk::Format::eR16G16B16A16Uint,
vk::Format::eR16G16B16A16Snorm,
vk::Format::eR16G16B16A16Unorm,
vk::Format::eR16G16Unorm,
vk::Format::eR16G16Snorm,
vk::Format::eR16G16Sfloat,
vk::Format::eR16Unorm,
vk::Format::eR8G8B8A8Srgb,
vk::Format::eR8G8Unorm,
vk::Format::eR8G8Snorm,
vk::Format::eR8Unorm,
vk::Format::eR8Uint,
vk::Format::eB10G11R11UfloatPack32,
vk::Format::eR32Sfloat,
vk::Format::eR32Uint,
vk::Format::eR32Sint,
vk::Format::eR16Sfloat,
vk::Format::eR16G16B16A16Sfloat,
vk::Format::eB8G8R8A8Unorm,
vk::Format::eR4G4B4A4UnormPack16,
vk::Format::eD32Sfloat,
vk::Format::eD16Unorm,
vk::Format::eD16UnormS8Uint,
vk::Format::eD24UnormS8Uint,
vk::Format::eD32SfloatS8Uint,
vk::Format::eBc1RgbaUnormBlock,
vk::Format::eBc2UnormBlock,
vk::Format::eBc3UnormBlock,
vk::Format::eBc4UnormBlock,
vk::Format::eBc5UnormBlock,
vk::Format::eBc5SnormBlock,
vk::Format::eBc7UnormBlock,
vk::Format::eBc6HUfloatBlock,
vk::Format::eBc6HSfloatBlock,
vk::Format::eBc1RgbaSrgbBlock,
vk::Format::eBc2SrgbBlock,
vk::Format::eBc3SrgbBlock,
vk::Format::eBc7SrgbBlock,
vk::Format::eAstc4x4UnormBlock,
vk::Format::eAstc4x4SrgbBlock,
vk::Format::eAstc5x4UnormBlock,
vk::Format::eAstc5x4SrgbBlock,
vk::Format::eAstc5x5UnormBlock,
vk::Format::eAstc5x5SrgbBlock,
vk::Format::eAstc6x5UnormBlock,
vk::Format::eAstc6x5SrgbBlock,
vk::Format::eAstc6x6UnormBlock,
vk::Format::eAstc6x6SrgbBlock,
vk::Format::eAstc8x5UnormBlock,
vk::Format::eAstc8x5SrgbBlock,
vk::Format::eAstc8x6UnormBlock,
vk::Format::eAstc8x6SrgbBlock,
vk::Format::eAstc8x8UnormBlock,
vk::Format::eAstc8x8SrgbBlock,
vk::Format::eAstc10x5UnormBlock,
vk::Format::eAstc10x5SrgbBlock,
vk::Format::eAstc10x6UnormBlock,
vk::Format::eAstc10x6SrgbBlock,
vk::Format::eAstc10x8UnormBlock,
vk::Format::eAstc10x8SrgbBlock,
vk::Format::eAstc10x10UnormBlock,
vk::Format::eAstc10x10SrgbBlock,
vk::Format::eAstc12x10UnormBlock,
vk::Format::eAstc12x10SrgbBlock,
vk::Format::eAstc12x12UnormBlock,
vk::Format::eAstc12x12SrgbBlock,
vk::Format::eE5B9G9R9UfloatPack32};
std::unordered_map<vk::Format, vk::FormatProperties> format_properties;
for (const auto format : formats) {
format_properties.emplace(format, physical.getFormatProperties(format, dld));
}
return format_properties;
}
} // namespace Vulkan

78
src/video_core/renderer_vulkan/vk_device.h
View file

@ -8,8 +8,9 @@
#include <string_view>
#include <unordered_map>
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -22,12 +23,12 @@ const u32 GuestWarpSize = 32;
/// Handles data specific to a physical device.
class VKDevice final {
public:
explicit VKDevice(const vk::DispatchLoaderDynamic& dld, vk::PhysicalDevice physical,
vk::SurfaceKHR surface);
explicit VKDevice(VkInstance instance, vk::PhysicalDevice physical, VkSurfaceKHR surface,
const vk::InstanceDispatch& dld);
~VKDevice();
/// Initializes the device. Returns true on success.
bool Create(vk::Instance instance);
bool Create();
/**
* Returns a format supported by the device for the passed requeriments.
@ -36,20 +37,20 @@ public:
* @param format_type Format type usage.
* @returns A format supported by the device.
*/
vk::Format GetSupportedFormat(vk::Format wanted_format, vk::FormatFeatureFlags wanted_usage,
FormatType format_type) const;
VkFormat GetSupportedFormat(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
FormatType format_type) const;
/// Reports a device loss.
void ReportLoss() const;
/// Returns the dispatch loader with direct function pointers of the device.
const vk::DispatchLoaderDynamic& GetDispatchLoader() const {
const vk::DeviceDispatch& GetDispatchLoader() const {
return dld;
}
/// Returns the logical device.
vk::Device GetLogical() const {
return logical.get();
const vk::Device& GetLogical() const {
return logical;
}
/// Returns the physical device.
@ -79,7 +80,7 @@ public:
/// Returns true if the device is integrated with the host CPU.
bool IsIntegrated() const {
return properties.deviceType == vk::PhysicalDeviceType::eIntegratedGpu;
return properties.deviceType == VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU;
}
/// Returns the current Vulkan API version provided in Vulkan-formatted version numbers.
@ -98,27 +99,27 @@ public:
}
/// Returns the driver ID.
vk::DriverIdKHR GetDriverID() const {
VkDriverIdKHR GetDriverID() const {
return driver_id;
}
/// Returns uniform buffer alignment requeriment.
vk::DeviceSize GetUniformBufferAlignment() const {
VkDeviceSize GetUniformBufferAlignment() const {
return properties.limits.minUniformBufferOffsetAlignment;
}
/// Returns storage alignment requeriment.
vk::DeviceSize GetStorageBufferAlignment() const {
VkDeviceSize GetStorageBufferAlignment() const {
return properties.limits.minStorageBufferOffsetAlignment;
}
/// Returns the maximum range for storage buffers.
vk::DeviceSize GetMaxStorageBufferRange() const {
VkDeviceSize GetMaxStorageBufferRange() const {
return properties.limits.maxStorageBufferRange;
}
/// Returns the maximum size for push constants.
vk::DeviceSize GetMaxPushConstantsSize() const {
VkDeviceSize GetMaxPushConstantsSize() const {
return properties.limits.maxPushConstantsSize;
}
@ -138,8 +139,8 @@ public:
}
/// Returns true if the device can be forced to use the guest warp size.
bool IsGuestWarpSizeSupported(vk::ShaderStageFlagBits stage) const {
return (guest_warp_stages & stage) != vk::ShaderStageFlags{};
bool IsGuestWarpSizeSupported(VkShaderStageFlagBits stage) const {
return guest_warp_stages & stage;
}
/// Returns true if formatless image load is supported.
@ -188,15 +189,14 @@ public:
}
/// Checks if the physical device is suitable.
static bool IsSuitable(vk::PhysicalDevice physical, vk::SurfaceKHR surface,
const vk::DispatchLoaderDynamic& dld);
static bool IsSuitable(vk::PhysicalDevice physical, VkSurfaceKHR surface);
private:
/// Loads extensions into a vector and stores available ones in this object.
std::vector<const char*> LoadExtensions();
/// Sets up queue families.
void SetupFamilies(vk::SurfaceKHR surface);
void SetupFamilies(VkSurfaceKHR surface);
/// Sets up device features.
void SetupFeatures();
@ -205,32 +205,28 @@ private:
void CollectTelemetryParameters();
/// Returns a list of queue initialization descriptors.
std::vector<vk::DeviceQueueCreateInfo> GetDeviceQueueCreateInfos() const;
std::vector<VkDeviceQueueCreateInfo> GetDeviceQueueCreateInfos() const;
/// Returns true if ASTC textures are natively supported.
bool IsOptimalAstcSupported(const vk::PhysicalDeviceFeatures& features) const;
bool IsOptimalAstcSupported(const VkPhysicalDeviceFeatures& features) const;
/// Returns true if a format is supported.
bool IsFormatSupported(vk::Format wanted_format, vk::FormatFeatureFlags wanted_usage,
bool IsFormatSupported(VkFormat wanted_format, VkFormatFeatureFlags wanted_usage,
FormatType format_type) const;
/// Returns the device properties for Vulkan formats.
static std::unordered_map<vk::Format, vk::FormatProperties> GetFormatProperties(
const vk::DispatchLoaderDynamic& dld, vk::PhysicalDevice physical);
vk::DispatchLoaderDynamic dld; ///< Device function pointers.
vk::PhysicalDevice physical; ///< Physical device.
vk::PhysicalDeviceProperties properties; ///< Device properties.
UniqueDevice logical; ///< Logical device.
vk::Queue graphics_queue; ///< Main graphics queue.
vk::Queue present_queue; ///< Main present queue.
u32 graphics_family{}; ///< Main graphics queue family index.
u32 present_family{}; ///< Main present queue family index.
vk::DriverIdKHR driver_id{}; ///< Driver ID.
vk::ShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced.ed
bool is_optimal_astc_supported{}; ///< Support for native ASTC.
bool is_float16_supported{}; ///< Support for float16 arithmetics.
bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest.
vk::DeviceDispatch dld; ///< Device function pointers.
vk::PhysicalDevice physical; ///< Physical device.
VkPhysicalDeviceProperties properties; ///< Device properties.
vk::Device logical; ///< Logical device.
vk::Queue graphics_queue; ///< Main graphics queue.
vk::Queue present_queue; ///< Main present queue.
u32 graphics_family{}; ///< Main graphics queue family index.
u32 present_family{}; ///< Main present queue family index.
VkDriverIdKHR driver_id{}; ///< Driver ID.
VkShaderStageFlags guest_warp_stages{}; ///< Stages where the guest warp size can be forced.ed
bool is_optimal_astc_supported{}; ///< Support for native ASTC.
bool is_float16_supported{}; ///< Support for float16 arithmetics.
bool is_warp_potentially_bigger{}; ///< Host warp size can be bigger than guest.
bool is_formatless_image_load_supported{}; ///< Support for shader image read without format.
bool khr_uniform_buffer_standard_layout{}; ///< Support for std430 on UBOs.
bool ext_index_type_uint8{}; ///< Support for VK_EXT_index_type_uint8.
@ -244,7 +240,7 @@ private:
std::vector<std::string> reported_extensions; ///< Reported Vulkan extensions.
/// Format properties dictionary.
std::unordered_map<vk::Format, vk::FormatProperties> format_properties;
std::unordered_map<VkFormat, VkFormatProperties> format_properties;
};
} // namespace Vulkan

360
src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp
View file

@ -2,11 +2,13 @@
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include <array>
#include <cstring>
#include <vector>
#include "common/assert.h"
#include "common/common_types.h"
#include "common/microprofile.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
@ -16,6 +18,7 @@
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -23,21 +26,26 @@ MICROPROFILE_DECLARE(Vulkan_PipelineCache);
namespace {
vk::StencilOpState GetStencilFaceState(const FixedPipelineState::StencilFace& face) {
return vk::StencilOpState(MaxwellToVK::StencilOp(face.action_stencil_fail),
MaxwellToVK::StencilOp(face.action_depth_pass),
MaxwellToVK::StencilOp(face.action_depth_fail),
MaxwellToVK::ComparisonOp(face.test_func), 0, 0, 0);
VkStencilOpState GetStencilFaceState(const FixedPipelineState::StencilFace& face) {
VkStencilOpState state;
state.failOp = MaxwellToVK::StencilOp(face.action_stencil_fail);
state.passOp = MaxwellToVK::StencilOp(face.action_depth_pass);
state.depthFailOp = MaxwellToVK::StencilOp(face.action_depth_fail);
state.compareOp = MaxwellToVK::ComparisonOp(face.test_func);
state.compareMask = 0;
state.writeMask = 0;
state.reference = 0;
return state;
}
bool SupportsPrimitiveRestart(vk::PrimitiveTopology topology) {
bool SupportsPrimitiveRestart(VkPrimitiveTopology topology) {
static constexpr std::array unsupported_topologies = {
vk::PrimitiveTopology::ePointList,
vk::PrimitiveTopology::eLineList,
vk::PrimitiveTopology::eTriangleList,
vk::PrimitiveTopology::eLineListWithAdjacency,
vk::PrimitiveTopology::eTriangleListWithAdjacency,
vk::PrimitiveTopology::ePatchList};
VK_PRIMITIVE_TOPOLOGY_POINT_LIST,
VK_PRIMITIVE_TOPOLOGY_LINE_LIST,
VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST,
VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY,
VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY,
VK_PRIMITIVE_TOPOLOGY_PATCH_LIST};
return std::find(std::begin(unsupported_topologies), std::end(unsupported_topologies),
topology) == std::end(unsupported_topologies);
}
@ -49,7 +57,7 @@ VKGraphicsPipeline::VKGraphicsPipeline(const VKDevice& device, VKScheduler& sche
VKUpdateDescriptorQueue& update_descriptor_queue,
VKRenderPassCache& renderpass_cache,
const GraphicsPipelineCacheKey& key,
const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
const SPIRVProgram& program)
: device{device}, scheduler{scheduler}, fixed_state{key.fixed_state}, hash{key.Hash()},
descriptor_set_layout{CreateDescriptorSetLayout(bindings)},
@ -63,7 +71,7 @@ VKGraphicsPipeline::VKGraphicsPipeline(const VKDevice& device, VKScheduler& sche
VKGraphicsPipeline::~VKGraphicsPipeline() = default;
vk::DescriptorSet VKGraphicsPipeline::CommitDescriptorSet() {
VkDescriptorSet VKGraphicsPipeline::CommitDescriptorSet() {
if (!descriptor_template) {
return {};
}
@ -72,27 +80,32 @@ vk::DescriptorSet VKGraphicsPipeline::CommitDescriptorSet() {
return set;
}
UniqueDescriptorSetLayout VKGraphicsPipeline::CreateDescriptorSetLayout(
const std::vector<vk::DescriptorSetLayoutBinding>& bindings) const {
const vk::DescriptorSetLayoutCreateInfo descriptor_set_layout_ci(
{}, static_cast<u32>(bindings.size()), bindings.data());
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createDescriptorSetLayoutUnique(descriptor_set_layout_ci, nullptr, dld);
vk::DescriptorSetLayout VKGraphicsPipeline::CreateDescriptorSetLayout(
vk::Span<VkDescriptorSetLayoutBinding> bindings) const {
VkDescriptorSetLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.bindingCount = bindings.size();
ci.pBindings = bindings.data();
return device.GetLogical().CreateDescriptorSetLayout(ci);
}
UniquePipelineLayout VKGraphicsPipeline::CreatePipelineLayout() const {
const vk::PipelineLayoutCreateInfo pipeline_layout_ci({}, 1, &*descriptor_set_layout, 0,
nullptr);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createPipelineLayoutUnique(pipeline_layout_ci, nullptr, dld);
vk::PipelineLayout VKGraphicsPipeline::CreatePipelineLayout() const {
VkPipelineLayoutCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.setLayoutCount = 1;
ci.pSetLayouts = descriptor_set_layout.address();
ci.pushConstantRangeCount = 0;
ci.pPushConstantRanges = nullptr;
return device.GetLogical().CreatePipelineLayout(ci);
}
UniqueDescriptorUpdateTemplate VKGraphicsPipeline::CreateDescriptorUpdateTemplate(
vk::DescriptorUpdateTemplateKHR VKGraphicsPipeline::CreateDescriptorUpdateTemplate(
const SPIRVProgram& program) const {
std::vector<vk::DescriptorUpdateTemplateEntry> template_entries;
std::vector<VkDescriptorUpdateTemplateEntry> template_entries;
u32 binding = 0;
u32 offset = 0;
for (const auto& stage : program) {
@ -102,38 +115,47 @@ UniqueDescriptorUpdateTemplate VKGraphicsPipeline::CreateDescriptorUpdateTemplat
}
if (template_entries.empty()) {
// If the shader doesn't use descriptor sets, skip template creation.
return UniqueDescriptorUpdateTemplate{};
return {};
}
const vk::DescriptorUpdateTemplateCreateInfo template_ci(
{}, static_cast<u32>(template_entries.size()), template_entries.data(),
vk::DescriptorUpdateTemplateType::eDescriptorSet, *descriptor_set_layout,
vk::PipelineBindPoint::eGraphics, *layout, DESCRIPTOR_SET);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createDescriptorUpdateTemplateUnique(template_ci, nullptr, dld);
VkDescriptorUpdateTemplateCreateInfoKHR ci;
ci.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_UPDATE_TEMPLATE_CREATE_INFO_KHR;
ci.pNext = nullptr;
ci.flags = 0;
ci.descriptorUpdateEntryCount = static_cast<u32>(template_entries.size());
ci.pDescriptorUpdateEntries = template_entries.data();
ci.templateType = VK_DESCRIPTOR_UPDATE_TEMPLATE_TYPE_DESCRIPTOR_SET_KHR;
ci.descriptorSetLayout = *descriptor_set_layout;
ci.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
ci.pipelineLayout = *layout;
ci.set = DESCRIPTOR_SET;
return device.GetLogical().CreateDescriptorUpdateTemplateKHR(ci);
}
std::vector<UniqueShaderModule> VKGraphicsPipeline::CreateShaderModules(
std::vector<vk::ShaderModule> VKGraphicsPipeline::CreateShaderModules(
const SPIRVProgram& program) const {
std::vector<UniqueShaderModule> modules;
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
VkShaderModuleCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
std::vector<vk::ShaderModule> modules;
modules.reserve(Maxwell::MaxShaderStage);
for (std::size_t i = 0; i < Maxwell::MaxShaderStage; ++i) {
const auto& stage = program[i];
if (!stage) {
continue;
}
const vk::ShaderModuleCreateInfo module_ci({}, stage->code.size() * sizeof(u32),
stage->code.data());
modules.emplace_back(dev.createShaderModuleUnique(module_ci, nullptr, dld));
ci.codeSize = stage->code.size() * sizeof(u32);
ci.pCode = stage->code.data();
modules.push_back(device.GetLogical().CreateShaderModule(ci));
}
return modules;
}
UniquePipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
const SPIRVProgram& program) const {
vk::Pipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& renderpass_params,
const SPIRVProgram& program) const {
const auto& vi = fixed_state.vertex_input;
const auto& ia = fixed_state.input_assembly;
const auto& ds = fixed_state.depth_stencil;
@ -141,19 +163,26 @@ UniquePipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& render
const auto& ts = fixed_state.tessellation;
const auto& rs = fixed_state.rasterizer;
std::vector<vk::VertexInputBindingDescription> vertex_bindings;
std::vector<vk::VertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors;
std::vector<VkVertexInputBindingDescription> vertex_bindings;
std::vector<VkVertexInputBindingDivisorDescriptionEXT> vertex_binding_divisors;
for (std::size_t i = 0; i < vi.num_bindings; ++i) {
const auto& binding = vi.bindings[i];
const bool instanced = binding.divisor != 0;
const auto rate = instanced ? vk::VertexInputRate::eInstance : vk::VertexInputRate::eVertex;
vertex_bindings.emplace_back(binding.index, binding.stride, rate);
const auto rate = instanced ? VK_VERTEX_INPUT_RATE_INSTANCE : VK_VERTEX_INPUT_RATE_VERTEX;
auto& vertex_binding = vertex_bindings.emplace_back();
vertex_binding.binding = binding.index;
vertex_binding.stride = binding.stride;
vertex_binding.inputRate = rate;
if (instanced) {
vertex_binding_divisors.emplace_back(binding.index, binding.divisor);
auto& binding_divisor = vertex_binding_divisors.emplace_back();
binding_divisor.binding = binding.index;
binding_divisor.divisor = binding.divisor;
}
}
std::vector<vk::VertexInputAttributeDescription> vertex_attributes;
std::vector<VkVertexInputAttributeDescription> vertex_attributes;
const auto& input_attributes = program[0]->entries.attributes;
for (std::size_t i = 0; i < vi.num_attributes; ++i) {
const auto& attribute = vi.attributes[i];
@ -161,109 +190,194 @@ UniquePipeline VKGraphicsPipeline::CreatePipeline(const RenderPassParams& render
// Skip attributes not used by the vertex shaders.
continue;
}
vertex_attributes.emplace_back(attribute.index, attribute.buffer,
MaxwellToVK::VertexFormat(attribute.type, attribute.size),
attribute.offset);
auto& vertex_attribute = vertex_attributes.emplace_back();
vertex_attribute.location = attribute.index;
vertex_attribute.binding = attribute.buffer;
vertex_attribute.format = MaxwellToVK::VertexFormat(attribute.type, attribute.size);
vertex_attribute.offset = attribute.offset;
}
vk::PipelineVertexInputStateCreateInfo vertex_input_ci(
{}, static_cast<u32>(vertex_bindings.size()), vertex_bindings.data(),
static_cast<u32>(vertex_attributes.size()), vertex_attributes.data());
VkPipelineVertexInputStateCreateInfo vertex_input_ci;
vertex_input_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
vertex_input_ci.pNext = nullptr;
vertex_input_ci.flags = 0;
vertex_input_ci.vertexBindingDescriptionCount = static_cast<u32>(vertex_bindings.size());
vertex_input_ci.pVertexBindingDescriptions = vertex_bindings.data();
vertex_input_ci.vertexAttributeDescriptionCount = static_cast<u32>(vertex_attributes.size());
vertex_input_ci.pVertexAttributeDescriptions = vertex_attributes.data();
const vk::PipelineVertexInputDivisorStateCreateInfoEXT vertex_input_divisor_ci(
static_cast<u32>(vertex_binding_divisors.size()), vertex_binding_divisors.data());
VkPipelineVertexInputDivisorStateCreateInfoEXT input_divisor_ci;
input_divisor_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT;
input_divisor_ci.pNext = nullptr;
input_divisor_ci.vertexBindingDivisorCount = static_cast<u32>(vertex_binding_divisors.size());
input_divisor_ci.pVertexBindingDivisors = vertex_binding_divisors.data();
if (!vertex_binding_divisors.empty()) {
vertex_input_ci.pNext = &vertex_input_divisor_ci;
vertex_input_ci.pNext = &input_divisor_ci;
}
const auto primitive_topology = MaxwellToVK::PrimitiveTopology(device, ia.topology);
const vk::PipelineInputAssemblyStateCreateInfo input_assembly_ci(
{}, primitive_topology,
ia.primitive_restart_enable && SupportsPrimitiveRestart(primitive_topology));
VkPipelineInputAssemblyStateCreateInfo input_assembly_ci;
input_assembly_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
input_assembly_ci.pNext = nullptr;
input_assembly_ci.flags = 0;
input_assembly_ci.topology = MaxwellToVK::PrimitiveTopology(device, ia.topology);
input_assembly_ci.primitiveRestartEnable =
ia.primitive_restart_enable && SupportsPrimitiveRestart(input_assembly_ci.topology);
const vk::PipelineTessellationStateCreateInfo tessellation_ci({}, ts.patch_control_points);
VkPipelineTessellationStateCreateInfo tessellation_ci;
tessellation_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO;
tessellation_ci.pNext = nullptr;
tessellation_ci.flags = 0;
tessellation_ci.patchControlPoints = ts.patch_control_points;
const vk::PipelineViewportStateCreateInfo viewport_ci({}, Maxwell::NumViewports, nullptr,
Maxwell::NumViewports, nullptr);
VkPipelineViewportStateCreateInfo viewport_ci;
viewport_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO;
viewport_ci.pNext = nullptr;
viewport_ci.flags = 0;
viewport_ci.viewportCount = Maxwell::NumViewports;
viewport_ci.pViewports = nullptr;
viewport_ci.scissorCount = Maxwell::NumViewports;
viewport_ci.pScissors = nullptr;
// TODO(Rodrigo): Find out what's the default register value for front face
const vk::PipelineRasterizationStateCreateInfo rasterizer_ci(
{}, rs.depth_clamp_enable, false, vk::PolygonMode::eFill,
rs.cull_enable ? MaxwellToVK::CullFace(rs.cull_face) : vk::CullModeFlagBits::eNone,
MaxwellToVK::FrontFace(rs.front_face), rs.depth_bias_enable, 0.0f, 0.0f, 0.0f, 1.0f);
VkPipelineRasterizationStateCreateInfo rasterization_ci;
rasterization_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO;
rasterization_ci.pNext = nullptr;
rasterization_ci.flags = 0;
rasterization_ci.depthClampEnable = rs.depth_clamp_enable;
rasterization_ci.rasterizerDiscardEnable = VK_FALSE;
rasterization_ci.polygonMode = VK_POLYGON_MODE_FILL;
rasterization_ci.cullMode =
rs.cull_enable ? MaxwellToVK::CullFace(rs.cull_face) : VK_CULL_MODE_NONE;
rasterization_ci.frontFace = MaxwellToVK::FrontFace(rs.front_face);
rasterization_ci.depthBiasEnable = rs.depth_bias_enable;
rasterization_ci.depthBiasConstantFactor = 0.0f;
rasterization_ci.depthBiasClamp = 0.0f;
rasterization_ci.depthBiasSlopeFactor = 0.0f;
rasterization_ci.lineWidth = 1.0f;
const vk::PipelineMultisampleStateCreateInfo multisampling_ci(
{}, vk::SampleCountFlagBits::e1, false, 0.0f, nullptr, false, false);
VkPipelineMultisampleStateCreateInfo multisample_ci;
multisample_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
multisample_ci.pNext = nullptr;
multisample_ci.flags = 0;
multisample_ci.rasterizationSamples = VK_SAMPLE_COUNT_1_BIT;
multisample_ci.sampleShadingEnable = VK_FALSE;
multisample_ci.minSampleShading = 0.0f;
multisample_ci.pSampleMask = nullptr;
multisample_ci.alphaToCoverageEnable = VK_FALSE;
multisample_ci.alphaToOneEnable = VK_FALSE;
const vk::CompareOp depth_test_compare = ds.depth_test_enable
? MaxwellToVK::ComparisonOp(ds.depth_test_function)
: vk::CompareOp::eAlways;
VkPipelineDepthStencilStateCreateInfo depth_stencil_ci;
depth_stencil_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO;
depth_stencil_ci.pNext = nullptr;
depth_stencil_ci.flags = 0;
depth_stencil_ci.depthTestEnable = ds.depth_test_enable;
depth_stencil_ci.depthWriteEnable = ds.depth_write_enable;
depth_stencil_ci.depthCompareOp = ds.depth_test_enable
? MaxwellToVK::ComparisonOp(ds.depth_test_function)
: VK_COMPARE_OP_ALWAYS;
depth_stencil_ci.depthBoundsTestEnable = ds.depth_bounds_enable;
depth_stencil_ci.stencilTestEnable = ds.stencil_enable;
depth_stencil_ci.front = GetStencilFaceState(ds.front_stencil);
depth_stencil_ci.back = GetStencilFaceState(ds.back_stencil);
depth_stencil_ci.minDepthBounds = 0.0f;
depth_stencil_ci.maxDepthBounds = 0.0f;
const vk::PipelineDepthStencilStateCreateInfo depth_stencil_ci(
{}, ds.depth_test_enable, ds.depth_write_enable, depth_test_compare, ds.depth_bounds_enable,
ds.stencil_enable, GetStencilFaceState(ds.front_stencil),
GetStencilFaceState(ds.back_stencil), 0.0f, 0.0f);
std::array<vk::PipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
std::array<VkPipelineColorBlendAttachmentState, Maxwell::NumRenderTargets> cb_attachments;
const std::size_t num_attachments =
std::min(cd.attachments_count, renderpass_params.color_attachments.size());
for (std::size_t i = 0; i < num_attachments; ++i) {
constexpr std::array component_table{
vk::ColorComponentFlagBits::eR, vk::ColorComponentFlagBits::eG,
vk::ColorComponentFlagBits::eB, vk::ColorComponentFlagBits::eA};
static constexpr std::array component_table = {
VK_COLOR_COMPONENT_R_BIT, VK_COLOR_COMPONENT_G_BIT, VK_COLOR_COMPONENT_B_BIT,
VK_COLOR_COMPONENT_A_BIT};
const auto& blend = cd.attachments[i];
vk::ColorComponentFlags color_components{};
VkColorComponentFlags color_components = 0;
for (std::size_t j = 0; j < component_table.size(); ++j) {
if (blend.components[j])
if (blend.components[j]) {
color_components |= component_table[j];
}
}
cb_attachments[i] = vk::PipelineColorBlendAttachmentState(
blend.enable, MaxwellToVK::BlendFactor(blend.src_rgb_func),
MaxwellToVK::BlendFactor(blend.dst_rgb_func),
MaxwellToVK::BlendEquation(blend.rgb_equation),
MaxwellToVK::BlendFactor(blend.src_a_func), MaxwellToVK::BlendFactor(blend.dst_a_func),
MaxwellToVK::BlendEquation(blend.a_equation), color_components);
VkPipelineColorBlendAttachmentState& attachment = cb_attachments[i];
attachment.blendEnable = blend.enable;
attachment.srcColorBlendFactor = MaxwellToVK::BlendFactor(blend.src_rgb_func);
attachment.dstColorBlendFactor = MaxwellToVK::BlendFactor(blend.dst_rgb_func);
attachment.colorBlendOp = MaxwellToVK::BlendEquation(blend.rgb_equation);
attachment.srcAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.src_a_func);
attachment.dstAlphaBlendFactor = MaxwellToVK::BlendFactor(blend.dst_a_func);
attachment.alphaBlendOp = MaxwellToVK::BlendEquation(blend.a_equation);
attachment.colorWriteMask = color_components;
}
const vk::PipelineColorBlendStateCreateInfo color_blending_ci({}, false, vk::LogicOp::eCopy,
static_cast<u32>(num_attachments),
cb_attachments.data(), {});
constexpr std::array dynamic_states = {
vk::DynamicState::eViewport, vk::DynamicState::eScissor,
vk::DynamicState::eDepthBias, vk::DynamicState::eBlendConstants,
vk::DynamicState::eDepthBounds, vk::DynamicState::eStencilCompareMask,
vk::DynamicState::eStencilWriteMask, vk::DynamicState::eStencilReference};
const vk::PipelineDynamicStateCreateInfo dynamic_state_ci(
{}, static_cast<u32>(dynamic_states.size()), dynamic_states.data());
VkPipelineColorBlendStateCreateInfo color_blend_ci;
color_blend_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO;
color_blend_ci.pNext = nullptr;
color_blend_ci.flags = 0;
color_blend_ci.logicOpEnable = VK_FALSE;
color_blend_ci.logicOp = VK_LOGIC_OP_COPY;
color_blend_ci.attachmentCount = static_cast<u32>(num_attachments);
color_blend_ci.pAttachments = cb_attachments.data();
std::memset(color_blend_ci.blendConstants, 0, sizeof(color_blend_ci.blendConstants));
vk::PipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
static constexpr std::array dynamic_states = {
VK_DYNAMIC_STATE_VIEWPORT, VK_DYNAMIC_STATE_SCISSOR,
VK_DYNAMIC_STATE_DEPTH_BIAS, VK_DYNAMIC_STATE_BLEND_CONSTANTS,
VK_DYNAMIC_STATE_DEPTH_BOUNDS, VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
VK_DYNAMIC_STATE_STENCIL_WRITE_MASK, VK_DYNAMIC_STATE_STENCIL_REFERENCE};
VkPipelineDynamicStateCreateInfo dynamic_state_ci;
dynamic_state_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO;
dynamic_state_ci.pNext = nullptr;
dynamic_state_ci.flags = 0;
dynamic_state_ci.dynamicStateCount = static_cast<u32>(dynamic_states.size());
dynamic_state_ci.pDynamicStates = dynamic_states.data();
VkPipelineShaderStageRequiredSubgroupSizeCreateInfoEXT subgroup_size_ci;
subgroup_size_ci.sType =
VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_REQUIRED_SUBGROUP_SIZE_CREATE_INFO_EXT;
subgroup_size_ci.pNext = nullptr;
subgroup_size_ci.requiredSubgroupSize = GuestWarpSize;
std::vector<vk::PipelineShaderStageCreateInfo> shader_stages;
std::vector<VkPipelineShaderStageCreateInfo> shader_stages;
std::size_t module_index = 0;
for (std::size_t stage = 0; stage < Maxwell::MaxShaderStage; ++stage) {
if (!program[stage]) {
continue;
}
const auto stage_enum = static_cast<Tegra::Engines::ShaderType>(stage);
const auto vk_stage = MaxwellToVK::ShaderStage(stage_enum);
auto& stage_ci = shader_stages.emplace_back(vk::PipelineShaderStageCreateFlags{}, vk_stage,
*modules[module_index++], "main", nullptr);
if (program[stage]->entries.uses_warps && device.IsGuestWarpSizeSupported(vk_stage)) {
VkPipelineShaderStageCreateInfo& stage_ci = shader_stages.emplace_back();
stage_ci.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO;
stage_ci.pNext = nullptr;
stage_ci.flags = 0;
stage_ci.stage = MaxwellToVK::ShaderStage(static_cast<Tegra::Engines::ShaderType>(stage));
stage_ci.module = *modules[module_index++];
stage_ci.pName = "main";
stage_ci.pSpecializationInfo = nullptr;
if (program[stage]->entries.uses_warps && device.IsGuestWarpSizeSupported(stage_ci.stage)) {
stage_ci.pNext = &subgroup_size_ci;
}
}
const vk::GraphicsPipelineCreateInfo create_info(
{}, static_cast<u32>(shader_stages.size()), shader_stages.data(), &vertex_input_ci,
&input_assembly_ci, &tessellation_ci, &viewport_ci, &rasterizer_ci, &multisampling_ci,
&depth_stencil_ci, &color_blending_ci, &dynamic_state_ci, *layout, renderpass, 0, {}, 0);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createGraphicsPipelineUnique(nullptr, create_info, nullptr, dld);
VkGraphicsPipelineCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.stageCount = static_cast<u32>(shader_stages.size());
ci.pStages = shader_stages.data();
ci.pVertexInputState = &vertex_input_ci;
ci.pInputAssemblyState = &input_assembly_ci;
ci.pTessellationState = &tessellation_ci;
ci.pViewportState = &viewport_ci;
ci.pRasterizationState = &rasterization_ci;
ci.pMultisampleState = &multisample_ci;
ci.pDepthStencilState = &depth_stencil_ci;
ci.pColorBlendState = &color_blend_ci;
ci.pDynamicState = &dynamic_state_ci;
ci.layout = *layout;
ci.renderPass = renderpass;
ci.subpass = 0;
ci.basePipelineHandle = nullptr;
ci.basePipelineIndex = 0;
return device.GetLogical().CreateGraphicsPipeline(ci);
}
} // namespace Vulkan

38
src/video_core/renderer_vulkan/vk_graphics_pipeline.h
View file

@ -11,12 +11,12 @@
#include <vector>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -39,52 +39,52 @@ public:
VKUpdateDescriptorQueue& update_descriptor_queue,
VKRenderPassCache& renderpass_cache,
const GraphicsPipelineCacheKey& key,
const std::vector<vk::DescriptorSetLayoutBinding>& bindings,
vk::Span<VkDescriptorSetLayoutBinding> bindings,
const SPIRVProgram& program);
~VKGraphicsPipeline();
vk::DescriptorSet CommitDescriptorSet();
VkDescriptorSet CommitDescriptorSet();
vk::Pipeline GetHandle() const {
VkPipeline GetHandle() const {
return *pipeline;
}
vk::PipelineLayout GetLayout() const {
VkPipelineLayout GetLayout() const {
return *layout;
}
vk::RenderPass GetRenderPass() const {
VkRenderPass GetRenderPass() const {
return renderpass;
}
private:
UniqueDescriptorSetLayout CreateDescriptorSetLayout(
const std::vector<vk::DescriptorSetLayoutBinding>& bindings) const;
vk::DescriptorSetLayout CreateDescriptorSetLayout(
vk::Span<VkDescriptorSetLayoutBinding> bindings) const;
UniquePipelineLayout CreatePipelineLayout() const;
vk::PipelineLayout CreatePipelineLayout() const;
UniqueDescriptorUpdateTemplate CreateDescriptorUpdateTemplate(
vk::DescriptorUpdateTemplateKHR CreateDescriptorUpdateTemplate(
const SPIRVProgram& program) const;
std::vector<UniqueShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
std::vector<vk::ShaderModule> CreateShaderModules(const SPIRVProgram& program) const;
UniquePipeline CreatePipeline(const RenderPassParams& renderpass_params,
const SPIRVProgram& program) const;
vk::Pipeline CreatePipeline(const RenderPassParams& renderpass_params,
const SPIRVProgram& program) const;
const VKDevice& device;
VKScheduler& scheduler;
const FixedPipelineState fixed_state;
const u64 hash;
UniqueDescriptorSetLayout descriptor_set_layout;
vk::DescriptorSetLayout descriptor_set_layout;
DescriptorAllocator descriptor_allocator;
VKUpdateDescriptorQueue& update_descriptor_queue;
UniquePipelineLayout layout;
UniqueDescriptorUpdateTemplate descriptor_template;
std::vector<UniqueShaderModule> modules;
vk::PipelineLayout layout;
vk::DescriptorUpdateTemplateKHR descriptor_template;
std::vector<vk::ShaderModule> modules;
vk::RenderPass renderpass;
UniquePipeline pipeline;
VkRenderPass renderpass;
vk::Pipeline pipeline;
};
} // namespace Vulkan

61
src/video_core/renderer_vulkan/vk_image.cpp
View file

@ -6,22 +6,21 @@
#include <vector>
#include "common/assert.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_image.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
VKImage::VKImage(const VKDevice& device, VKScheduler& scheduler,
const vk::ImageCreateInfo& image_ci, vk::ImageAspectFlags aspect_mask)
VKImage::VKImage(const VKDevice& device, VKScheduler& scheduler, const VkImageCreateInfo& image_ci,
VkImageAspectFlags aspect_mask)
: device{device}, scheduler{scheduler}, format{image_ci.format}, aspect_mask{aspect_mask},
image_num_layers{image_ci.arrayLayers}, image_num_levels{image_ci.mipLevels} {
UNIMPLEMENTED_IF_MSG(image_ci.queueFamilyIndexCount != 0,
"Queue family tracking is not implemented");
const auto dev = device.GetLogical();
image = dev.createImageUnique(image_ci, nullptr, device.GetDispatchLoader());
image = device.GetLogical().CreateImage(image_ci);
const u32 num_ranges = image_num_layers * image_num_levels;
barriers.resize(num_ranges);
@ -31,8 +30,8 @@ VKImage::VKImage(const VKDevice& device, VKScheduler& scheduler,
VKImage::~VKImage() = default;
void VKImage::Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access,
vk::ImageLayout new_layout) {
VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
VkImageLayout new_layout) {
if (!HasChanged(base_layer, num_layers, base_level, num_levels, new_access, new_layout)) {
return;
}
@ -43,9 +42,21 @@ void VKImage::Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num
const u32 layer = base_layer + layer_it;
const u32 level = base_level + level_it;
auto& state = GetSubrangeState(layer, level);
barriers[cursor] = vk::ImageMemoryBarrier(
state.access, new_access, state.layout, new_layout, VK_QUEUE_FAMILY_IGNORED,
VK_QUEUE_FAMILY_IGNORED, *image, {aspect_mask, level, 1, layer, 1});
auto& barrier = barriers[cursor];
barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = state.access;
barrier.dstAccessMask = new_access;
barrier.oldLayout = state.layout;
barrier.newLayout = new_layout;
barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
barrier.image = *image;
barrier.subresourceRange.aspectMask = aspect_mask;
barrier.subresourceRange.baseMipLevel = level;
barrier.subresourceRange.levelCount = 1;
barrier.subresourceRange.baseArrayLayer = layer;
barrier.subresourceRange.layerCount = 1;
state.access = new_access;
state.layout = new_layout;
}
@ -53,16 +64,16 @@ void VKImage::Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([barriers = barriers, cursor](auto cmdbuf, auto& dld) {
scheduler.Record([barriers = barriers, cursor](vk::CommandBuffer cmdbuf) {
// TODO(Rodrigo): Implement a way to use the latest stage across subresources.
constexpr auto stage_stub = vk::PipelineStageFlagBits::eAllCommands;
cmdbuf.pipelineBarrier(stage_stub, stage_stub, {}, 0, nullptr, 0, nullptr,
static_cast<u32>(cursor), barriers.data(), dld);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, {}, {},
vk::Span(barriers.data(), cursor));
});
}
bool VKImage::HasChanged(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
vk::AccessFlags new_access, vk::ImageLayout new_layout) noexcept {
VkAccessFlags new_access, VkImageLayout new_layout) noexcept {
const bool is_full_range = base_layer == 0 && num_layers == image_num_layers &&
base_level == 0 && num_levels == image_num_levels;
if (!is_full_range) {
@ -91,11 +102,21 @@ bool VKImage::HasChanged(u32 base_layer, u32 num_layers, u32 base_level, u32 num
void VKImage::CreatePresentView() {
// Image type has to be 2D to be presented.
const vk::ImageViewCreateInfo image_view_ci({}, *image, vk::ImageViewType::e2D, format, {},
{aspect_mask, 0, 1, 0, 1});
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
present_view = dev.createImageViewUnique(image_view_ci, nullptr, dld);
VkImageViewCreateInfo image_view_ci;
image_view_ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
image_view_ci.pNext = nullptr;
image_view_ci.flags = 0;
image_view_ci.image = *image;
image_view_ci.viewType = VK_IMAGE_VIEW_TYPE_2D;
image_view_ci.format = format;
image_view_ci.components = {VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY};
image_view_ci.subresourceRange.aspectMask = aspect_mask;
image_view_ci.subresourceRange.baseMipLevel = 0;
image_view_ci.subresourceRange.levelCount = 1;
image_view_ci.subresourceRange.baseArrayLayer = 0;
image_view_ci.subresourceRange.layerCount = 1;
present_view = device.GetLogical().CreateImageView(image_view_ci);
}
VKImage::SubrangeState& VKImage::GetSubrangeState(u32 layer, u32 level) noexcept {

40
src/video_core/renderer_vulkan/vk_image.h
View file

@ -8,7 +8,7 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -18,16 +18,16 @@ class VKScheduler;
class VKImage {
public:
explicit VKImage(const VKDevice& device, VKScheduler& scheduler,
const vk::ImageCreateInfo& image_ci, vk::ImageAspectFlags aspect_mask);
const VkImageCreateInfo& image_ci, VkImageAspectFlags aspect_mask);
~VKImage();
/// Records in the passed command buffer an image transition and updates the state of the image.
void Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access,
vk::ImageLayout new_layout);
VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
VkImageLayout new_layout);
/// Returns a view compatible with presentation, the image has to be 2D.
vk::ImageView GetPresentView() {
VkImageView GetPresentView() {
if (!present_view) {
CreatePresentView();
}
@ -35,28 +35,28 @@ public:
}
/// Returns the Vulkan image handler.
vk::Image GetHandle() const {
return *image;
const vk::Image& GetHandle() const {
return image;
}
/// Returns the Vulkan format for this image.
vk::Format GetFormat() const {
VkFormat GetFormat() const {
return format;
}
/// Returns the Vulkan aspect mask.
vk::ImageAspectFlags GetAspectMask() const {
VkImageAspectFlags GetAspectMask() const {
return aspect_mask;
}
private:
struct SubrangeState final {
vk::AccessFlags access{}; ///< Current access bits.
vk::ImageLayout layout = vk::ImageLayout::eUndefined; ///< Current image layout.
VkAccessFlags access = 0; ///< Current access bits.
VkImageLayout layout = VK_IMAGE_LAYOUT_UNDEFINED; ///< Current image layout.
};
bool HasChanged(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
vk::AccessFlags new_access, vk::ImageLayout new_layout) noexcept;
VkAccessFlags new_access, VkImageLayout new_layout) noexcept;
/// Creates a presentation view.
void CreatePresentView();
@ -67,16 +67,16 @@ private:
const VKDevice& device; ///< Device handler.
VKScheduler& scheduler; ///< Device scheduler.
const vk::Format format; ///< Vulkan format.
const vk::ImageAspectFlags aspect_mask; ///< Vulkan aspect mask.
const u32 image_num_layers; ///< Number of layers.
const u32 image_num_levels; ///< Number of mipmap levels.
const VkFormat format; ///< Vulkan format.
const VkImageAspectFlags aspect_mask; ///< Vulkan aspect mask.
const u32 image_num_layers; ///< Number of layers.
const u32 image_num_levels; ///< Number of mipmap levels.
UniqueImage image; ///< Image handle.
UniqueImageView present_view; ///< Image view compatible with presentation.
vk::Image image; ///< Image handle.
vk::ImageView present_view; ///< Image view compatible with presentation.
std::vector<vk::ImageMemoryBarrier> barriers; ///< Pool of barriers.
std::vector<SubrangeState> subrange_states; ///< Current subrange state.
std::vector<VkImageMemoryBarrier> barriers; ///< Pool of barriers.
std::vector<SubrangeState> subrange_states; ///< Current subrange state.
bool state_diverged = false; ///< True when subresources mismatch in layout.
};

104
src/video_core/renderer_vulkan/vk_memory_manager.cpp
View file

@ -11,9 +11,9 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -30,17 +30,11 @@ u64 GetAllocationChunkSize(u64 required_size) {
class VKMemoryAllocation final {
public:
explicit VKMemoryAllocation(const VKDevice& device, vk::DeviceMemory memory,
vk::MemoryPropertyFlags properties, u64 allocation_size, u32 type)
: device{device}, memory{memory}, properties{properties}, allocation_size{allocation_size},
shifted_type{ShiftType(type)} {}
VkMemoryPropertyFlags properties, u64 allocation_size, u32 type)
: device{device}, memory{std::move(memory)}, properties{properties},
allocation_size{allocation_size}, shifted_type{ShiftType(type)} {}
~VKMemoryAllocation() {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
dev.free(memory, nullptr, dld);
}
VKMemoryCommit Commit(vk::DeviceSize commit_size, vk::DeviceSize alignment) {
VKMemoryCommit Commit(VkDeviceSize commit_size, VkDeviceSize alignment) {
auto found = TryFindFreeSection(free_iterator, allocation_size,
static_cast<u64>(commit_size), static_cast<u64>(alignment));
if (!found) {
@ -73,9 +67,8 @@ public:
}
/// Returns whether this allocation is compatible with the arguments.
bool IsCompatible(vk::MemoryPropertyFlags wanted_properties, u32 type_mask) const {
return (wanted_properties & properties) != vk::MemoryPropertyFlagBits(0) &&
(type_mask & shifted_type) != 0;
bool IsCompatible(VkMemoryPropertyFlags wanted_properties, u32 type_mask) const {
return (wanted_properties & properties) && (type_mask & shifted_type) != 0;
}
private:
@ -111,11 +104,11 @@ private:
return std::nullopt;
}
const VKDevice& device; ///< Vulkan device.
const vk::DeviceMemory memory; ///< Vulkan memory allocation handler.
const vk::MemoryPropertyFlags properties; ///< Vulkan properties.
const u64 allocation_size; ///< Size of this allocation.
const u32 shifted_type; ///< Stored Vulkan type of this allocation, shifted.
const VKDevice& device; ///< Vulkan device.
const vk::DeviceMemory memory; ///< Vulkan memory allocation handler.
const VkMemoryPropertyFlags properties; ///< Vulkan properties.
const u64 allocation_size; ///< Size of this allocation.
const u32 shifted_type; ///< Stored Vulkan type of this allocation, shifted.
/// Hints where the next free region is likely going to be.
u64 free_iterator{};
@ -125,22 +118,20 @@ private:
};
VKMemoryManager::VKMemoryManager(const VKDevice& device)
: device{device}, properties{device.GetPhysical().getMemoryProperties(
device.GetDispatchLoader())},
: device{device}, properties{device.GetPhysical().GetMemoryProperties()},
is_memory_unified{GetMemoryUnified(properties)} {}
VKMemoryManager::~VKMemoryManager() = default;
VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& requirements,
VKMemoryCommit VKMemoryManager::Commit(const VkMemoryRequirements& requirements,
bool host_visible) {
const u64 chunk_size = GetAllocationChunkSize(requirements.size);
// When a host visible commit is asked, search for host visible and coherent, otherwise search
// for a fast device local type.
const vk::MemoryPropertyFlags wanted_properties =
host_visible
? vk::MemoryPropertyFlagBits::eHostVisible | vk::MemoryPropertyFlagBits::eHostCoherent
: vk::MemoryPropertyFlagBits::eDeviceLocal;
const VkMemoryPropertyFlags wanted_properties =
host_visible ? VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT
: VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT;
if (auto commit = TryAllocCommit(requirements, wanted_properties)) {
return commit;
@ -161,23 +152,19 @@ VKMemoryCommit VKMemoryManager::Commit(const vk::MemoryRequirements& requirement
return commit;
}
VKMemoryCommit VKMemoryManager::Commit(vk::Buffer buffer, bool host_visible) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
auto commit = Commit(dev.getBufferMemoryRequirements(buffer, dld), host_visible);
dev.bindBufferMemory(buffer, commit->GetMemory(), commit->GetOffset(), dld);
VKMemoryCommit VKMemoryManager::Commit(const vk::Buffer& buffer, bool host_visible) {
auto commit = Commit(device.GetLogical().GetBufferMemoryRequirements(*buffer), host_visible);
buffer.BindMemory(commit->GetMemory(), commit->GetOffset());
return commit;
}
VKMemoryCommit VKMemoryManager::Commit(vk::Image image, bool host_visible) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
auto commit = Commit(dev.getImageMemoryRequirements(image, dld), host_visible);
dev.bindImageMemory(image, commit->GetMemory(), commit->GetOffset(), dld);
VKMemoryCommit VKMemoryManager::Commit(const vk::Image& image, bool host_visible) {
auto commit = Commit(device.GetLogical().GetImageMemoryRequirements(*image), host_visible);
image.BindMemory(commit->GetMemory(), commit->GetOffset());
return commit;
}
bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask,
bool VKMemoryManager::AllocMemory(VkMemoryPropertyFlags wanted_properties, u32 type_mask,
u64 size) {
const u32 type = [&] {
for (u32 type_index = 0; type_index < properties.memoryTypeCount; ++type_index) {
@ -191,24 +178,26 @@ bool VKMemoryManager::AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32
return 0U;
}();
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
// Try to allocate found type.
const vk::MemoryAllocateInfo memory_ai(size, type);
vk::DeviceMemory memory;
if (const auto res = dev.allocateMemory(&memory_ai, nullptr, &memory, dld);
res != vk::Result::eSuccess) {
LOG_CRITICAL(Render_Vulkan, "Device allocation failed with code {}!", vk::to_string(res));
VkMemoryAllocateInfo memory_ai;
memory_ai.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
memory_ai.pNext = nullptr;
memory_ai.allocationSize = size;
memory_ai.memoryTypeIndex = type;
vk::DeviceMemory memory = device.GetLogical().TryAllocateMemory(memory_ai);
if (!memory) {
LOG_CRITICAL(Render_Vulkan, "Device allocation failed!");
return false;
}
allocations.push_back(
std::make_unique<VKMemoryAllocation>(device, memory, wanted_properties, size, type));
allocations.push_back(std::make_unique<VKMemoryAllocation>(device, std::move(memory),
wanted_properties, size, type));
return true;
}
VKMemoryCommit VKMemoryManager::TryAllocCommit(const vk::MemoryRequirements& requirements,
vk::MemoryPropertyFlags wanted_properties) {
VKMemoryCommit VKMemoryManager::TryAllocCommit(const VkMemoryRequirements& requirements,
VkMemoryPropertyFlags wanted_properties) {
for (auto& allocation : allocations) {
if (!allocation->IsCompatible(wanted_properties, requirements.memoryTypeBits)) {
continue;
@ -220,10 +209,9 @@ VKMemoryCommit VKMemoryManager::TryAllocCommit(const vk::MemoryRequirements& req
return {};
}
/*static*/ bool VKMemoryManager::GetMemoryUnified(
const vk::PhysicalDeviceMemoryProperties& properties) {
bool VKMemoryManager::GetMemoryUnified(const VkPhysicalDeviceMemoryProperties& properties) {
for (u32 heap_index = 0; heap_index < properties.memoryHeapCount; ++heap_index) {
if (!(properties.memoryHeaps[heap_index].flags & vk::MemoryHeapFlagBits::eDeviceLocal)) {
if (!(properties.memoryHeaps[heap_index].flags & VK_MEMORY_HEAP_DEVICE_LOCAL_BIT)) {
// Memory is considered unified when heaps are device local only.
return false;
}
@ -232,23 +220,19 @@ VKMemoryCommit VKMemoryManager::TryAllocCommit(const vk::MemoryRequirements& req
}
VKMemoryCommitImpl::VKMemoryCommitImpl(const VKDevice& device, VKMemoryAllocation* allocation,
vk::DeviceMemory memory, u64 begin, u64 end)
: device{device}, interval{begin, end}, memory{memory}, allocation{allocation} {}
const vk::DeviceMemory& memory, u64 begin, u64 end)
: device{device}, memory{memory}, interval{begin, end}, allocation{allocation} {}
VKMemoryCommitImpl::~VKMemoryCommitImpl() {
allocation->Free(this);
}
MemoryMap VKMemoryCommitImpl::Map(u64 size, u64 offset_) const {
const auto dev = device.GetLogical();
const auto address = reinterpret_cast<u8*>(
dev.mapMemory(memory, interval.first + offset_, size, {}, device.GetDispatchLoader()));
return MemoryMap{this, address};
return MemoryMap{this, memory.Map(interval.first + offset_, size)};
}
void VKMemoryCommitImpl::Unmap() const {
const auto dev = device.GetLogical();
dev.unmapMemory(memory, device.GetDispatchLoader());
memory.Unmap();
}
MemoryMap VKMemoryCommitImpl::Map() const {

34
src/video_core/renderer_vulkan/vk_memory_manager.h
View file

@ -8,7 +8,7 @@
#include <utility>
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -32,13 +32,13 @@ public:
* memory. When passing false, it will try to allocate device local memory.
* @returns A memory commit.
*/
VKMemoryCommit Commit(const vk::MemoryRequirements& reqs, bool host_visible);
VKMemoryCommit Commit(const VkMemoryRequirements& reqs, bool host_visible);
/// Commits memory required by the buffer and binds it.
VKMemoryCommit Commit(vk::Buffer buffer, bool host_visible);
VKMemoryCommit Commit(const vk::Buffer& buffer, bool host_visible);
/// Commits memory required by the image and binds it.
VKMemoryCommit Commit(vk::Image image, bool host_visible);
VKMemoryCommit Commit(const vk::Image& image, bool host_visible);
/// Returns true if the memory allocations are done always in host visible and coherent memory.
bool IsMemoryUnified() const {
@ -47,18 +47,18 @@ public:
private:
/// Allocates a chunk of memory.
bool AllocMemory(vk::MemoryPropertyFlags wanted_properties, u32 type_mask, u64 size);
bool AllocMemory(VkMemoryPropertyFlags wanted_properties, u32 type_mask, u64 size);
/// Tries to allocate a memory commit.
VKMemoryCommit TryAllocCommit(const vk::MemoryRequirements& requirements,
vk::MemoryPropertyFlags wanted_properties);
VKMemoryCommit TryAllocCommit(const VkMemoryRequirements& requirements,
VkMemoryPropertyFlags wanted_properties);
/// Returns true if the device uses an unified memory model.
static bool GetMemoryUnified(const vk::PhysicalDeviceMemoryProperties& properties);
static bool GetMemoryUnified(const VkPhysicalDeviceMemoryProperties& properties);
const VKDevice& device; ///< Device handler.
const vk::PhysicalDeviceMemoryProperties properties; ///< Physical device properties.
const bool is_memory_unified; ///< True if memory model is unified.
const VKDevice& device; ///< Device handler.
const VkPhysicalDeviceMemoryProperties properties; ///< Physical device properties.
const bool is_memory_unified; ///< True if memory model is unified.
std::vector<std::unique_ptr<VKMemoryAllocation>> allocations; ///< Current allocations.
};
@ -68,7 +68,7 @@ class VKMemoryCommitImpl final {
public:
explicit VKMemoryCommitImpl(const VKDevice& device, VKMemoryAllocation* allocation,
vk::DeviceMemory memory, u64 begin, u64 end);
const vk::DeviceMemory& memory, u64 begin, u64 end);
~VKMemoryCommitImpl();
/// Maps a memory region and returns a pointer to it.
@ -80,13 +80,13 @@ public:
MemoryMap Map() const;
/// Returns the Vulkan memory handler.
vk::DeviceMemory GetMemory() const {
return memory;
VkDeviceMemory GetMemory() const {
return *memory;
}
/// Returns the start position of the commit relative to the allocation.
vk::DeviceSize GetOffset() const {
return static_cast<vk::DeviceSize>(interval.first);
VkDeviceSize GetOffset() const {
return static_cast<VkDeviceSize>(interval.first);
}
private:
@ -94,8 +94,8 @@ private:
void Unmap() const;
const VKDevice& device; ///< Vulkan device.
const vk::DeviceMemory& memory; ///< Vulkan device memory handler.
std::pair<u64, u64> interval{}; ///< Interval where the commit exists.
vk::DeviceMemory memory; ///< Vulkan device memory handler.
VKMemoryAllocation* allocation{}; ///< Pointer to the large memory allocation.
};

93
src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
View file

@ -13,7 +13,6 @@
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/memory_manager.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pipeline.h"
@ -26,6 +25,7 @@
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/shader/compiler_settings.h"
namespace Vulkan {
@ -36,12 +36,11 @@ using Tegra::Engines::ShaderType;
namespace {
// C++20's using enum
constexpr auto eUniformBuffer = vk::DescriptorType::eUniformBuffer;
constexpr auto eStorageBuffer = vk::DescriptorType::eStorageBuffer;
constexpr auto eUniformTexelBuffer = vk::DescriptorType::eUniformTexelBuffer;
constexpr auto eCombinedImageSampler = vk::DescriptorType::eCombinedImageSampler;
constexpr auto eStorageImage = vk::DescriptorType::eStorageImage;
constexpr VkDescriptorType UNIFORM_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER;
constexpr VkDescriptorType STORAGE_BUFFER = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
constexpr VkDescriptorType UNIFORM_TEXEL_BUFFER = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER;
constexpr VkDescriptorType COMBINED_IMAGE_SAMPLER = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
constexpr VkDescriptorType STORAGE_IMAGE = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
constexpr VideoCommon::Shader::CompilerSettings compiler_settings{
VideoCommon::Shader::CompileDepth::FullDecompile};
@ -126,32 +125,37 @@ ShaderType GetShaderType(Maxwell::ShaderProgram program) {
}
}
template <vk::DescriptorType descriptor_type, class Container>
void AddBindings(std::vector<vk::DescriptorSetLayoutBinding>& bindings, u32& binding,
vk::ShaderStageFlags stage_flags, const Container& container) {
template <VkDescriptorType descriptor_type, class Container>
void AddBindings(std::vector<VkDescriptorSetLayoutBinding>& bindings, u32& binding,
VkShaderStageFlags stage_flags, const Container& container) {
const u32 num_entries = static_cast<u32>(std::size(container));
for (std::size_t i = 0; i < num_entries; ++i) {
u32 count = 1;
if constexpr (descriptor_type == eCombinedImageSampler) {
if constexpr (descriptor_type == VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER) {
// Combined image samplers can be arrayed.
count = container[i].Size();
}
bindings.emplace_back(binding++, descriptor_type, count, stage_flags, nullptr);
VkDescriptorSetLayoutBinding& entry = bindings.emplace_back();
entry.binding = binding++;
entry.descriptorType = descriptor_type;
entry.descriptorCount = count;
entry.stageFlags = stage_flags;
entry.pImmutableSamplers = nullptr;
}
}
u32 FillDescriptorLayout(const ShaderEntries& entries,
std::vector<vk::DescriptorSetLayoutBinding>& bindings,
std::vector<VkDescriptorSetLayoutBinding>& bindings,
Maxwell::ShaderProgram program_type, u32 base_binding) {
const ShaderType stage = GetStageFromProgram(program_type);
const vk::ShaderStageFlags flags = MaxwellToVK::ShaderStage(stage);
const VkShaderStageFlags flags = MaxwellToVK::ShaderStage(stage);
u32 binding = base_binding;
AddBindings<eUniformBuffer>(bindings, binding, flags, entries.const_buffers);
AddBindings<eStorageBuffer>(bindings, binding, flags, entries.global_buffers);
AddBindings<eUniformTexelBuffer>(bindings, binding, flags, entries.texel_buffers);
AddBindings<eCombinedImageSampler>(bindings, binding, flags, entries.samplers);
AddBindings<eStorageImage>(bindings, binding, flags, entries.images);
AddBindings<UNIFORM_BUFFER>(bindings, binding, flags, entries.const_buffers);
AddBindings<STORAGE_BUFFER>(bindings, binding, flags, entries.global_buffers);
AddBindings<UNIFORM_TEXEL_BUFFER>(bindings, binding, flags, entries.texel_buffers);
AddBindings<COMBINED_IMAGE_SAMPLER>(bindings, binding, flags, entries.samplers);
AddBindings<STORAGE_IMAGE>(bindings, binding, flags, entries.images);
return binding;
}
@ -318,7 +322,7 @@ void VKPipelineCache::Unregister(const Shader& shader) {
RasterizerCache::Unregister(shader);
}
std::pair<SPIRVProgram, std::vector<vk::DescriptorSetLayoutBinding>>
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>>
VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
const auto& fixed_state = key.fixed_state;
auto& memory_manager = system.GPU().MemoryManager();
@ -335,7 +339,7 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
specialization.ndc_minus_one_to_one = fixed_state.rasterizer.ndc_minus_one_to_one;
SPIRVProgram program;
std::vector<vk::DescriptorSetLayoutBinding> bindings;
std::vector<VkDescriptorSetLayoutBinding> bindings;
for (std::size_t index = 0; index < Maxwell::MaxShaderProgram; ++index) {
const auto program_enum = static_cast<Maxwell::ShaderProgram>(index);
@ -371,32 +375,49 @@ VKPipelineCache::DecompileShaders(const GraphicsPipelineCacheKey& key) {
return {std::move(program), std::move(bindings)};
}
template <vk::DescriptorType descriptor_type, class Container>
void AddEntry(std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries, u32& binding,
template <VkDescriptorType descriptor_type, class Container>
void AddEntry(std::vector<VkDescriptorUpdateTemplateEntry>& template_entries, u32& binding,
u32& offset, const Container& container) {
static constexpr u32 entry_size = static_cast<u32>(sizeof(DescriptorUpdateEntry));
const u32 count = static_cast<u32>(std::size(container));
if constexpr (descriptor_type == eCombinedImageSampler) {
if constexpr (descriptor_type == COMBINED_IMAGE_SAMPLER) {
for (u32 i = 0; i < count; ++i) {
const u32 num_samplers = container[i].Size();
template_entries.emplace_back(binding, 0, num_samplers, descriptor_type, offset,
entry_size);
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding;
entry.dstArrayElement = 0;
entry.descriptorCount = num_samplers;
entry.descriptorType = descriptor_type;
entry.offset = offset;
entry.stride = entry_size;
++binding;
offset += num_samplers * entry_size;
}
return;
}
if constexpr (descriptor_type == eUniformTexelBuffer) {
if constexpr (descriptor_type == UNIFORM_TEXEL_BUFFER) {
// Nvidia has a bug where updating multiple uniform texels at once causes the driver to
// crash.
for (u32 i = 0; i < count; ++i) {
template_entries.emplace_back(binding + i, 0, 1, descriptor_type,
offset + i * entry_size, entry_size);
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding + i;
entry.dstArrayElement = 0;
entry.descriptorCount = 1;
entry.descriptorType = descriptor_type;
entry.offset = offset + i * entry_size;
entry.stride = entry_size;
}
} else if (count > 0) {
template_entries.emplace_back(binding, 0, count, descriptor_type, offset, entry_size);
VkDescriptorUpdateTemplateEntry& entry = template_entries.emplace_back();
entry.dstBinding = binding;
entry.dstArrayElement = 0;
entry.descriptorCount = count;
entry.descriptorType = descriptor_type;
entry.offset = offset;
entry.stride = entry_size;
}
offset += count * entry_size;
binding += count;
@ -404,12 +425,12 @@ void AddEntry(std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries,
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries) {
AddEntry<eUniformBuffer>(template_entries, offset, binding, entries.const_buffers);
AddEntry<eStorageBuffer>(template_entries, offset, binding, entries.global_buffers);
AddEntry<eUniformTexelBuffer>(template_entries, offset, binding, entries.texel_buffers);
AddEntry<eCombinedImageSampler>(template_entries, offset, binding, entries.samplers);
AddEntry<eStorageImage>(template_entries, offset, binding, entries.images);
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries) {
AddEntry<UNIFORM_BUFFER>(template_entries, offset, binding, entries.const_buffers);
AddEntry<STORAGE_BUFFER>(template_entries, offset, binding, entries.global_buffers);
AddEntry<UNIFORM_TEXEL_BUFFER>(template_entries, offset, binding, entries.texel_buffers);
AddEntry<COMBINED_IMAGE_SAMPLER>(template_entries, offset, binding, entries.samplers);
AddEntry<STORAGE_IMAGE>(template_entries, offset, binding, entries.images);
}
} // namespace Vulkan

6
src/video_core/renderer_vulkan/vk_pipeline_cache.h
View file

@ -19,12 +19,12 @@
#include "video_core/engines/const_buffer_engine_interface.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/rasterizer_cache.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_graphics_pipeline.h"
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_shader_decompiler.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/shader/registry.h"
#include "video_core/shader/shader_ir.h"
#include "video_core/surface.h"
@ -172,7 +172,7 @@ protected:
void FlushObjectInner(const Shader& object) override {}
private:
std::pair<SPIRVProgram, std::vector<vk::DescriptorSetLayoutBinding>> DecompileShaders(
std::pair<SPIRVProgram, std::vector<VkDescriptorSetLayoutBinding>> DecompileShaders(
const GraphicsPipelineCacheKey& key);
Core::System& system;
@ -194,6 +194,6 @@ private:
void FillDescriptorUpdateTemplateEntries(
const ShaderEntries& entries, u32& binding, u32& offset,
std::vector<vk::DescriptorUpdateTemplateEntry>& template_entries);
std::vector<VkDescriptorUpdateTemplateEntryKHR>& template_entries);
} // namespace Vulkan

56
src/video_core/renderer_vulkan/vk_query_cache.cpp
View file

@ -8,19 +8,19 @@
#include <utility>
#include <vector>
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
namespace {
constexpr std::array QUERY_TARGETS = {vk::QueryType::eOcclusion};
constexpr std::array QUERY_TARGETS = {VK_QUERY_TYPE_OCCLUSION};
constexpr vk::QueryType GetTarget(VideoCore::QueryType type) {
constexpr VkQueryType GetTarget(VideoCore::QueryType type) {
return QUERY_TARGETS[static_cast<std::size_t>(type)];
}
@ -35,29 +35,34 @@ void QueryPool::Initialize(const VKDevice& device_, VideoCore::QueryType type_)
type = type_;
}
std::pair<vk::QueryPool, std::uint32_t> QueryPool::Commit(VKFence& fence) {
std::pair<VkQueryPool, u32> QueryPool::Commit(VKFence& fence) {
std::size_t index;
do {
index = CommitResource(fence);
} while (usage[index]);
usage[index] = true;
return {*pools[index / GROW_STEP], static_cast<std::uint32_t>(index % GROW_STEP)};
return {*pools[index / GROW_STEP], static_cast<u32>(index % GROW_STEP)};
}
void QueryPool::Allocate(std::size_t begin, std::size_t end) {
usage.resize(end);
const auto dev = device->GetLogical();
const u32 size = static_cast<u32>(end - begin);
const vk::QueryPoolCreateInfo query_pool_ci({}, GetTarget(type), size, {});
pools.push_back(dev.createQueryPoolUnique(query_pool_ci, nullptr, device->GetDispatchLoader()));
VkQueryPoolCreateInfo query_pool_ci;
query_pool_ci.sType = VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO;
query_pool_ci.pNext = nullptr;
query_pool_ci.flags = 0;
query_pool_ci.queryType = GetTarget(type);
query_pool_ci.queryCount = static_cast<u32>(end - begin);
query_pool_ci.pipelineStatistics = 0;
pools.push_back(device->GetLogical().CreateQueryPool(query_pool_ci));
}
void QueryPool::Reserve(std::pair<vk::QueryPool, std::uint32_t> query) {
void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) {
const auto it =
std::find_if(std::begin(pools), std::end(pools),
[query_pool = query.first](auto& pool) { return query_pool == *pool; });
std::find_if(pools.begin(), pools.end(), [query_pool = query.first](vk::QueryPool& pool) {
return query_pool == *pool;
});
ASSERT(it != std::end(pools));
const std::ptrdiff_t pool_index = std::distance(std::begin(pools), it);
@ -76,12 +81,11 @@ VKQueryCache::VKQueryCache(Core::System& system, VideoCore::RasterizerInterface&
VKQueryCache::~VKQueryCache() = default;
std::pair<vk::QueryPool, std::uint32_t> VKQueryCache::AllocateQuery(VideoCore::QueryType type) {
std::pair<VkQueryPool, u32> VKQueryCache::AllocateQuery(VideoCore::QueryType type) {
return query_pools[static_cast<std::size_t>(type)].Commit(scheduler.GetFence());
}
void VKQueryCache::Reserve(VideoCore::QueryType type,
std::pair<vk::QueryPool, std::uint32_t> query) {
void VKQueryCache::Reserve(VideoCore::QueryType type, std::pair<VkQueryPool, u32> query) {
query_pools[static_cast<std::size_t>(type)].Reserve(query);
}
@ -89,10 +93,10 @@ HostCounter::HostCounter(VKQueryCache& cache, std::shared_ptr<HostCounter> depen
VideoCore::QueryType type)
: VideoCommon::HostCounterBase<VKQueryCache, HostCounter>{std::move(dependency)}, cache{cache},
type{type}, query{cache.AllocateQuery(type)}, ticks{cache.Scheduler().Ticks()} {
const auto dev = cache.Device().GetLogical();
cache.Scheduler().Record([dev, query = query](vk::CommandBuffer cmdbuf, auto& dld) {
dev.resetQueryPoolEXT(query.first, query.second, 1, dld);
cmdbuf.beginQuery(query.first, query.second, vk::QueryControlFlagBits::ePrecise, dld);
const vk::Device* logical = &cache.Device().GetLogical();
cache.Scheduler().Record([logical, query = query](vk::CommandBuffer cmdbuf) {
logical->ResetQueryPoolEXT(query.first, query.second, 1);
cmdbuf.BeginQuery(query.first, query.second, VK_QUERY_CONTROL_PRECISE_BIT);
});
}
@ -101,22 +105,16 @@ HostCounter::~HostCounter() {
}
void HostCounter::EndQuery() {
cache.Scheduler().Record([query = query](auto cmdbuf, auto& dld) {
cmdbuf.endQuery(query.first, query.second, dld);
});
cache.Scheduler().Record(
[query = query](vk::CommandBuffer cmdbuf) { cmdbuf.EndQuery(query.first, query.second); });
}
u64 HostCounter::BlockingQuery() const {
if (ticks >= cache.Scheduler().Ticks()) {
cache.Scheduler().Flush();
}
const auto dev = cache.Device().GetLogical();
const auto& dld = cache.Device().GetDispatchLoader();
u64 value;
dev.getQueryPoolResults(query.first, query.second, 1, sizeof(value), &value, sizeof(value),
vk::QueryResultFlagBits::e64 | vk::QueryResultFlagBits::eWait, dld);
return value;
return cache.Device().GetLogical().GetQueryResult<u64>(
query.first, query.second, VK_QUERY_RESULT_64_BIT | VK_QUERY_RESULT_WAIT_BIT);
}
} // namespace Vulkan

14
src/video_core/renderer_vulkan/vk_query_cache.h
View file

@ -12,8 +12,8 @@
#include "common/common_types.h"
#include "video_core/query_cache.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace VideoCore {
class RasterizerInterface;
@ -36,9 +36,9 @@ public:
void Initialize(const VKDevice& device, VideoCore::QueryType type);
std::pair<vk::QueryPool, std::uint32_t> Commit(VKFence& fence);
std::pair<VkQueryPool, u32> Commit(VKFence& fence);
void Reserve(std::pair<vk::QueryPool, std::uint32_t> query);
void Reserve(std::pair<VkQueryPool, u32> query);
protected:
void Allocate(std::size_t begin, std::size_t end) override;
@ -49,7 +49,7 @@ private:
const VKDevice* device = nullptr;
VideoCore::QueryType type = {};
std::vector<UniqueQueryPool> pools;
std::vector<vk::QueryPool> pools;
std::vector<bool> usage;
};
@ -61,9 +61,9 @@ public:
const VKDevice& device, VKScheduler& scheduler);
~VKQueryCache();
std::pair<vk::QueryPool, std::uint32_t> AllocateQuery(VideoCore::QueryType type);
std::pair<VkQueryPool, u32> AllocateQuery(VideoCore::QueryType type);
void Reserve(VideoCore::QueryType type, std::pair<vk::QueryPool, std::uint32_t> query);
void Reserve(VideoCore::QueryType type, std::pair<VkQueryPool, u32> query);
const VKDevice& Device() const noexcept {
return device;
@ -91,7 +91,7 @@ private:
VKQueryCache& cache;
const VideoCore::QueryType type;
const std::pair<vk::QueryPool, std::uint32_t> query;
const std::pair<VkQueryPool, u32> query;
const u64 ticks;
};

281
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -19,7 +19,6 @@
#include "core/memory.h"
#include "video_core/engines/kepler_compute.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/renderer_vulkan.h"
@ -39,6 +38,7 @@
#include "video_core/renderer_vulkan/vk_state_tracker.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -60,32 +60,39 @@ namespace {
constexpr auto ComputeShaderIndex = static_cast<std::size_t>(Tegra::Engines::ShaderType::Compute);
vk::Viewport GetViewportState(const VKDevice& device, const Maxwell& regs, std::size_t index) {
const auto& viewport = regs.viewport_transform[index];
const float x = viewport.translate_x - viewport.scale_x;
const float y = viewport.translate_y - viewport.scale_y;
const float width = viewport.scale_x * 2.0f;
const float height = viewport.scale_y * 2.0f;
VkViewport GetViewportState(const VKDevice& device, const Maxwell& regs, std::size_t index) {
const auto& src = regs.viewport_transform[index];
VkViewport viewport;
viewport.x = src.translate_x - src.scale_x;
viewport.y = src.translate_y - src.scale_y;
viewport.width = src.scale_x * 2.0f;
viewport.height = src.scale_y * 2.0f;
const float reduce_z = regs.depth_mode == Maxwell::DepthMode::MinusOneToOne;
float near = viewport.translate_z - viewport.scale_z * reduce_z;
float far = viewport.translate_z + viewport.scale_z;
viewport.minDepth = src.translate_z - src.scale_z * reduce_z;
viewport.maxDepth = src.translate_z + src.scale_z;
if (!device.IsExtDepthRangeUnrestrictedSupported()) {
near = std::clamp(near, 0.0f, 1.0f);
far = std::clamp(far, 0.0f, 1.0f);
viewport.minDepth = std::clamp(viewport.minDepth, 0.0f, 1.0f);
viewport.maxDepth = std::clamp(viewport.maxDepth, 0.0f, 1.0f);
}
return vk::Viewport(x, y, width != 0 ? width : 1.0f, height != 0 ? height : 1.0f, near, far);
return viewport;
}
constexpr vk::Rect2D GetScissorState(const Maxwell& regs, std::size_t index) {
const auto& scissor = regs.scissor_test[index];
if (!scissor.enable) {
return {{0, 0}, {INT32_MAX, INT32_MAX}};
VkRect2D GetScissorState(const Maxwell& regs, std::size_t index) {
const auto& src = regs.scissor_test[index];
VkRect2D scissor;
if (src.enable) {
scissor.offset.x = static_cast<s32>(src.min_x);
scissor.offset.y = static_cast<s32>(src.min_y);
scissor.extent.width = src.max_x - src.min_x;
scissor.extent.height = src.max_y - src.min_y;
} else {
scissor.offset.x = 0;
scissor.offset.y = 0;
scissor.extent.width = std::numeric_limits<s32>::max();
scissor.extent.height = std::numeric_limits<s32>::max();
}
const u32 width = scissor.max_x - scissor.min_x;
const u32 height = scissor.max_y - scissor.min_y;
return {{static_cast<s32>(scissor.min_x), static_cast<s32>(scissor.min_y)}, {width, height}};
return scissor;
}
std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
@ -97,8 +104,8 @@ std::array<GPUVAddr, Maxwell::MaxShaderProgram> GetShaderAddresses(
return addresses;
}
void TransitionImages(const std::vector<ImageView>& views, vk::PipelineStageFlags pipeline_stage,
vk::AccessFlags access) {
void TransitionImages(const std::vector<ImageView>& views, VkPipelineStageFlags pipeline_stage,
VkAccessFlags access) {
for (auto& [view, layout] : views) {
view->Transition(*layout, pipeline_stage, access);
}
@ -127,13 +134,13 @@ Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry
class BufferBindings final {
public:
void AddVertexBinding(const vk::Buffer* buffer, vk::DeviceSize offset) {
void AddVertexBinding(const VkBuffer* buffer, VkDeviceSize offset) {
vertex.buffer_ptrs[vertex.num_buffers] = buffer;
vertex.offsets[vertex.num_buffers] = offset;
++vertex.num_buffers;
}
void SetIndexBinding(const vk::Buffer* buffer, vk::DeviceSize offset, vk::IndexType type) {
void SetIndexBinding(const VkBuffer* buffer, VkDeviceSize offset, VkIndexType type) {
index.buffer = buffer;
index.offset = offset;
index.type = type;
@ -217,14 +224,14 @@ private:
// Some of these fields are intentionally left uninitialized to avoid initializing them twice.
struct {
std::size_t num_buffers = 0;
std::array<const vk::Buffer*, Maxwell::NumVertexArrays> buffer_ptrs;
std::array<vk::DeviceSize, Maxwell::NumVertexArrays> offsets;
std::array<const VkBuffer*, Maxwell::NumVertexArrays> buffer_ptrs;
std::array<VkDeviceSize, Maxwell::NumVertexArrays> offsets;
} vertex;
struct {
const vk::Buffer* buffer = nullptr;
vk::DeviceSize offset;
vk::IndexType type;
const VkBuffer* buffer = nullptr;
VkDeviceSize offset;
VkIndexType type;
} index;
template <std::size_t N>
@ -243,38 +250,35 @@ private:
return;
}
std::array<vk::Buffer, N> buffers;
std::array<VkBuffer, N> buffers;
std::transform(vertex.buffer_ptrs.begin(), vertex.buffer_ptrs.begin() + N, buffers.begin(),
[](const auto ptr) { return *ptr; });
std::array<vk::DeviceSize, N> offsets;
std::array<VkDeviceSize, N> offsets;
std::copy(vertex.offsets.begin(), vertex.offsets.begin() + N, offsets.begin());
if constexpr (is_indexed) {
// Indexed draw
scheduler.Record([buffers, offsets, index_buffer = *index.buffer,
index_offset = index.offset,
index_type = index.type](auto cmdbuf, auto& dld) {
cmdbuf.bindIndexBuffer(index_buffer, index_offset, index_type, dld);
cmdbuf.bindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data(),
dld);
index_type = index.type](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(index_buffer, index_offset, index_type);
cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data());
});
} else {
// Array draw
scheduler.Record([buffers, offsets](auto cmdbuf, auto& dld) {
cmdbuf.bindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data(),
dld);
scheduler.Record([buffers, offsets](vk::CommandBuffer cmdbuf) {
cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data());
});
}
}
};
void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf,
const vk::DispatchLoaderDynamic& dld) const {
void RasterizerVulkan::DrawParameters::Draw(vk::CommandBuffer cmdbuf) const {
if (is_indexed) {
cmdbuf.drawIndexed(num_vertices, num_instances, 0, base_vertex, base_instance, dld);
cmdbuf.DrawIndexed(num_vertices, num_instances, 0, base_vertex, base_instance);
} else {
cmdbuf.draw(num_vertices, num_instances, base_vertex, base_instance, dld);
cmdbuf.Draw(num_vertices, num_instances, base_vertex, base_instance);
}
}
@ -337,7 +341,7 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
const auto renderpass = pipeline.GetRenderPass();
const auto [framebuffer, render_area] = ConfigureFramebuffers(renderpass);
scheduler.RequestRenderpass({renderpass, framebuffer, {{0, 0}, render_area}, 0, nullptr});
scheduler.RequestRenderpass(renderpass, framebuffer, render_area);
UpdateDynamicStates();
@ -345,19 +349,19 @@ void RasterizerVulkan::Draw(bool is_indexed, bool is_instanced) {
if (device.IsNvDeviceDiagnosticCheckpoints()) {
scheduler.Record(
[&pipeline](auto cmdbuf, auto& dld) { cmdbuf.setCheckpointNV(&pipeline, dld); });
[&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(&pipeline); });
}
BeginTransformFeedback();
const auto pipeline_layout = pipeline.GetLayout();
const auto descriptor_set = pipeline.CommitDescriptorSet();
scheduler.Record([pipeline_layout, descriptor_set, draw_params](auto cmdbuf, auto& dld) {
scheduler.Record([pipeline_layout, descriptor_set, draw_params](vk::CommandBuffer cmdbuf) {
if (descriptor_set) {
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eGraphics, pipeline_layout,
DESCRIPTOR_SET, 1, &descriptor_set, 0, nullptr, dld);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline_layout,
DESCRIPTOR_SET, descriptor_set, {});
}
draw_params.Draw(cmdbuf, dld);
draw_params.Draw(cmdbuf);
});
EndTransformFeedback();
@ -389,48 +393,54 @@ void RasterizerVulkan::Clear() {
DEBUG_ASSERT(texceptions.none());
SetupImageTransitions(0, color_attachments, zeta_attachment);
const vk::RenderPass renderpass = renderpass_cache.GetRenderPass(GetRenderPassParams(0));
const VkRenderPass renderpass = renderpass_cache.GetRenderPass(GetRenderPassParams(0));
const auto [framebuffer, render_area] = ConfigureFramebuffers(renderpass);
scheduler.RequestRenderpass({renderpass, framebuffer, {{0, 0}, render_area}, 0, nullptr});
scheduler.RequestRenderpass(renderpass, framebuffer, render_area);
const auto& scissor = regs.scissor_test[0];
const vk::Offset2D scissor_offset(scissor.min_x, scissor.min_y);
vk::Extent2D scissor_extent{scissor.max_x - scissor.min_x, scissor.max_y - scissor.min_y};
scissor_extent.width = std::min(scissor_extent.width, render_area.width);
scissor_extent.height = std::min(scissor_extent.height, render_area.height);
const u32 layer = regs.clear_buffers.layer;
const vk::ClearRect clear_rect({scissor_offset, scissor_extent}, layer, 1);
VkClearRect clear_rect;
clear_rect.baseArrayLayer = regs.clear_buffers.layer;
clear_rect.layerCount = 1;
clear_rect.rect = GetScissorState(regs, 0);
clear_rect.rect.extent.width = std::min(clear_rect.rect.extent.width, render_area.width);
clear_rect.rect.extent.height = std::min(clear_rect.rect.extent.height, render_area.height);
if (use_color) {
const std::array clear_color = {regs.clear_color[0], regs.clear_color[1],
regs.clear_color[2], regs.clear_color[3]};
const vk::ClearValue clear_value{clear_color};
VkClearValue clear_value;
std::memcpy(clear_value.color.float32, regs.clear_color, sizeof(regs.clear_color));
const u32 color_attachment = regs.clear_buffers.RT;
scheduler.Record([color_attachment, clear_value, clear_rect](auto cmdbuf, auto& dld) {
const vk::ClearAttachment attachment(vk::ImageAspectFlagBits::eColor, color_attachment,
clear_value);
cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld);
scheduler.Record([color_attachment, clear_value, clear_rect](vk::CommandBuffer cmdbuf) {
VkClearAttachment attachment;
attachment.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
attachment.colorAttachment = color_attachment;
attachment.clearValue = clear_value;
cmdbuf.ClearAttachments(attachment, clear_rect);
});
}
if (!use_depth && !use_stencil) {
return;
}
vk::ImageAspectFlags aspect_flags;
VkImageAspectFlags aspect_flags = 0;
if (use_depth) {
aspect_flags |= vk::ImageAspectFlagBits::eDepth;
aspect_flags |= VK_IMAGE_ASPECT_DEPTH_BIT;
}
if (use_stencil) {
aspect_flags |= vk::ImageAspectFlagBits::eStencil;
aspect_flags |= VK_IMAGE_ASPECT_STENCIL_BIT;
}
scheduler.Record([clear_depth = regs.clear_depth, clear_stencil = regs.clear_stencil,
clear_rect, aspect_flags](auto cmdbuf, auto& dld) {
const vk::ClearDepthStencilValue clear_zeta(clear_depth, clear_stencil);
const vk::ClearValue clear_value{clear_zeta};
const vk::ClearAttachment attachment(aspect_flags, 0, clear_value);
cmdbuf.clearAttachments(1, &attachment, 1, &clear_rect, dld);
clear_rect, aspect_flags](vk::CommandBuffer cmdbuf) {
VkClearValue clear_value;
clear_value.depthStencil.depth = clear_depth;
clear_value.depthStencil.stencil = clear_stencil;
VkClearAttachment attachment;
attachment.aspectMask = aspect_flags;
attachment.colorAttachment = 0;
attachment.clearValue.depthStencil.depth = clear_depth;
attachment.clearValue.depthStencil.stencil = clear_stencil;
cmdbuf.ClearAttachments(attachment, clear_rect);
});
}
@ -463,24 +473,24 @@ void RasterizerVulkan::DispatchCompute(GPUVAddr code_addr) {
buffer_cache.Unmap();
TransitionImages(sampled_views, vk::PipelineStageFlagBits::eComputeShader,
vk::AccessFlagBits::eShaderRead);
TransitionImages(image_views, vk::PipelineStageFlagBits::eComputeShader,
vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite);
TransitionImages(sampled_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT);
TransitionImages(image_views, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
if (device.IsNvDeviceDiagnosticCheckpoints()) {
scheduler.Record(
[&pipeline](auto cmdbuf, auto& dld) { cmdbuf.setCheckpointNV(nullptr, dld); });
[&pipeline](vk::CommandBuffer cmdbuf) { cmdbuf.SetCheckpointNV(nullptr); });
}
scheduler.Record([grid_x = launch_desc.grid_dim_x, grid_y = launch_desc.grid_dim_y,
grid_z = launch_desc.grid_dim_z, pipeline_handle = pipeline.GetHandle(),
layout = pipeline.GetLayout(),
descriptor_set = pipeline.CommitDescriptorSet()](auto cmdbuf, auto& dld) {
cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline_handle, dld);
cmdbuf.bindDescriptorSets(vk::PipelineBindPoint::eCompute, layout, DESCRIPTOR_SET, 1,
&descriptor_set, 0, nullptr, dld);
cmdbuf.dispatch(grid_x, grid_y, grid_z, dld);
descriptor_set = pipeline.CommitDescriptorSet()](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, pipeline_handle);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, layout, DESCRIPTOR_SET,
descriptor_set, {});
cmdbuf.Dispatch(grid_x, grid_y, grid_z);
});
}
@ -625,13 +635,13 @@ bool RasterizerVulkan::WalkAttachmentOverlaps(const CachedSurfaceView& attachmen
continue;
}
overlap = true;
*layout = vk::ImageLayout::eGeneral;
*layout = VK_IMAGE_LAYOUT_GENERAL;
}
return overlap;
}
std::tuple<vk::Framebuffer, vk::Extent2D> RasterizerVulkan::ConfigureFramebuffers(
vk::RenderPass renderpass) {
std::tuple<VkFramebuffer, VkExtent2D> RasterizerVulkan::ConfigureFramebuffers(
VkRenderPass renderpass) {
FramebufferCacheKey key{renderpass, std::numeric_limits<u32>::max(),
std::numeric_limits<u32>::max(), std::numeric_limits<u32>::max()};
@ -658,15 +668,20 @@ std::tuple<vk::Framebuffer, vk::Extent2D> RasterizerVulkan::ConfigureFramebuffer
const auto [fbentry, is_cache_miss] = framebuffer_cache.try_emplace(key);
auto& framebuffer = fbentry->second;
if (is_cache_miss) {
const vk::FramebufferCreateInfo framebuffer_ci(
{}, key.renderpass, static_cast<u32>(key.views.size()), key.views.data(), key.width,
key.height, key.layers);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
framebuffer = dev.createFramebufferUnique(framebuffer_ci, nullptr, dld);
VkFramebufferCreateInfo framebuffer_ci;
framebuffer_ci.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
framebuffer_ci.pNext = nullptr;
framebuffer_ci.flags = 0;
framebuffer_ci.renderPass = key.renderpass;
framebuffer_ci.attachmentCount = static_cast<u32>(key.views.size());
framebuffer_ci.pAttachments = key.views.data();
framebuffer_ci.width = key.width;
framebuffer_ci.height = key.height;
framebuffer_ci.layers = key.layers;
framebuffer = device.GetLogical().CreateFramebuffer(framebuffer_ci);
}
return {*framebuffer, vk::Extent2D{key.width, key.height}};
return {*framebuffer, VkExtent2D{key.width, key.height}};
}
RasterizerVulkan::DrawParameters RasterizerVulkan::SetupGeometry(FixedPipelineState& fixed_state,
@ -714,10 +729,9 @@ void RasterizerVulkan::SetupShaderDescriptors(
void RasterizerVulkan::SetupImageTransitions(
Texceptions texceptions, const std::array<View, Maxwell::NumRenderTargets>& color_attachments,
const View& zeta_attachment) {
TransitionImages(sampled_views, vk::PipelineStageFlagBits::eAllGraphics,
vk::AccessFlagBits::eShaderRead);
TransitionImages(image_views, vk::PipelineStageFlagBits::eAllGraphics,
vk::AccessFlagBits::eShaderRead | vk::AccessFlagBits::eShaderWrite);
TransitionImages(sampled_views, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_ACCESS_SHADER_READ_BIT);
TransitionImages(image_views, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT,
VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT);
for (std::size_t rt = 0; rt < std::size(color_attachments); ++rt) {
const auto color_attachment = color_attachments[rt];
@ -725,19 +739,19 @@ void RasterizerVulkan::SetupImageTransitions(
continue;
}
const auto image_layout =
texceptions[rt] ? vk::ImageLayout::eGeneral : vk::ImageLayout::eColorAttachmentOptimal;
color_attachment->Transition(
image_layout, vk::PipelineStageFlagBits::eColorAttachmentOutput,
vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite);
texceptions[rt] ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
color_attachment->Transition(image_layout, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT);
}
if (zeta_attachment != nullptr) {
const auto image_layout = texceptions[ZETA_TEXCEPTION_INDEX]
? vk::ImageLayout::eGeneral
: vk::ImageLayout::eDepthStencilAttachmentOptimal;
zeta_attachment->Transition(image_layout, vk::PipelineStageFlagBits::eLateFragmentTests,
vk::AccessFlagBits::eDepthStencilAttachmentRead |
vk::AccessFlagBits::eDepthStencilAttachmentWrite);
? VK_IMAGE_LAYOUT_GENERAL
: VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
zeta_attachment->Transition(image_layout, VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT);
}
}
@ -773,9 +787,9 @@ void RasterizerVulkan::BeginTransformFeedback() {
const std::size_t size = binding.buffer_size;
const auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
scheduler.Record([buffer = *buffer, offset = offset, size](auto cmdbuf, auto& dld) {
cmdbuf.bindTransformFeedbackBuffersEXT(0, {buffer}, {offset}, {size}, dld);
cmdbuf.beginTransformFeedbackEXT(0, {}, {}, dld);
scheduler.Record([buffer = *buffer, offset = offset, size](vk::CommandBuffer cmdbuf) {
cmdbuf.BindTransformFeedbackBuffersEXT(0, 1, &buffer, &offset, &size);
cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr);
});
}
@ -786,7 +800,7 @@ void RasterizerVulkan::EndTransformFeedback() {
}
scheduler.Record(
[](auto cmdbuf, auto& dld) { cmdbuf.endTransformFeedbackEXT(0, {}, {}, dld); });
[](vk::CommandBuffer cmdbuf) { cmdbuf.EndTransformFeedbackEXT(0, 0, nullptr, nullptr); });
}
void RasterizerVulkan::SetupVertexArrays(FixedPipelineState::VertexInput& vertex_input,
@ -837,7 +851,7 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
} else {
const auto [buffer, offset] =
quad_array_pass.Assemble(params.num_vertices, params.base_vertex);
buffer_bindings.SetIndexBinding(&buffer, offset, vk::IndexType::eUint32);
buffer_bindings.SetIndexBinding(buffer, offset, VK_INDEX_TYPE_UINT32);
params.base_vertex = 0;
params.num_vertices = params.num_vertices * 6 / 4;
params.is_indexed = true;
@ -1022,7 +1036,7 @@ void RasterizerVulkan::SetupTexture(const Tegra::Texture::FullTextureInfo& textu
update_descriptor_queue.AddSampledImage(sampler, image_view);
const auto image_layout = update_descriptor_queue.GetLastImageLayout();
*image_layout = vk::ImageLayout::eShaderReadOnlyOptimal;
*image_layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
sampled_views.push_back(ImageView{std::move(view), image_layout});
}
@ -1039,7 +1053,7 @@ void RasterizerVulkan::SetupImage(const Tegra::Texture::TICEntry& tic, const Ima
update_descriptor_queue.AddImage(image_view);
const auto image_layout = update_descriptor_queue.GetLastImageLayout();
*image_layout = vk::ImageLayout::eGeneral;
*image_layout = VK_IMAGE_LAYOUT_GENERAL;
image_views.push_back(ImageView{std::move(view), image_layout});
}
@ -1056,9 +1070,7 @@ void RasterizerVulkan::UpdateViewportsState(Tegra::Engines::Maxwell3D::Regs& reg
GetViewportState(device, regs, 10), GetViewportState(device, regs, 11),
GetViewportState(device, regs, 12), GetViewportState(device, regs, 13),
GetViewportState(device, regs, 14), GetViewportState(device, regs, 15)};
scheduler.Record([viewports](auto cmdbuf, auto& dld) {
cmdbuf.setViewport(0, static_cast<u32>(viewports.size()), viewports.data(), dld);
});
scheduler.Record([viewports](vk::CommandBuffer cmdbuf) { cmdbuf.SetViewport(0, viewports); });
}
void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs) {
@ -1072,9 +1084,7 @@ void RasterizerVulkan::UpdateScissorsState(Tegra::Engines::Maxwell3D::Regs& regs
GetScissorState(regs, 9), GetScissorState(regs, 10), GetScissorState(regs, 11),
GetScissorState(regs, 12), GetScissorState(regs, 13), GetScissorState(regs, 14),
GetScissorState(regs, 15)};
scheduler.Record([scissors](auto cmdbuf, auto& dld) {
cmdbuf.setScissor(0, static_cast<u32>(scissors.size()), scissors.data(), dld);
});
scheduler.Record([scissors](vk::CommandBuffer cmdbuf) { cmdbuf.SetScissor(0, scissors); });
}
void RasterizerVulkan::UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs) {
@ -1082,8 +1092,8 @@ void RasterizerVulkan::UpdateDepthBias(Tegra::Engines::Maxwell3D::Regs& regs) {
return;
}
scheduler.Record([constant = regs.polygon_offset_units, clamp = regs.polygon_offset_clamp,
factor = regs.polygon_offset_factor](auto cmdbuf, auto& dld) {
cmdbuf.setDepthBias(constant, clamp, factor / 2.0f, dld);
factor = regs.polygon_offset_factor](vk::CommandBuffer cmdbuf) {
cmdbuf.SetDepthBias(constant, clamp, factor / 2.0f);
});
}
@ -1093,9 +1103,8 @@ void RasterizerVulkan::UpdateBlendConstants(Tegra::Engines::Maxwell3D::Regs& reg
}
const std::array blend_color = {regs.blend_color.r, regs.blend_color.g, regs.blend_color.b,
regs.blend_color.a};
scheduler.Record([blend_color](auto cmdbuf, auto& dld) {
cmdbuf.setBlendConstants(blend_color.data(), dld);
});
scheduler.Record(
[blend_color](vk::CommandBuffer cmdbuf) { cmdbuf.SetBlendConstants(blend_color.data()); });
}
void RasterizerVulkan::UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs) {
@ -1103,7 +1112,7 @@ void RasterizerVulkan::UpdateDepthBounds(Tegra::Engines::Maxwell3D::Regs& regs)
return;
}
scheduler.Record([min = regs.depth_bounds[0], max = regs.depth_bounds[1]](
auto cmdbuf, auto& dld) { cmdbuf.setDepthBounds(min, max, dld); });
vk::CommandBuffer cmdbuf) { cmdbuf.SetDepthBounds(min, max); });
}
void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs) {
@ -1116,24 +1125,24 @@ void RasterizerVulkan::UpdateStencilFaces(Tegra::Engines::Maxwell3D::Regs& regs)
[front_ref = regs.stencil_front_func_ref, front_write_mask = regs.stencil_front_mask,
front_test_mask = regs.stencil_front_func_mask, back_ref = regs.stencil_back_func_ref,
back_write_mask = regs.stencil_back_mask,
back_test_mask = regs.stencil_back_func_mask](auto cmdbuf, auto& dld) {
back_test_mask = regs.stencil_back_func_mask](vk::CommandBuffer cmdbuf) {
// Front face
cmdbuf.setStencilReference(vk::StencilFaceFlagBits::eFront, front_ref, dld);
cmdbuf.setStencilWriteMask(vk::StencilFaceFlagBits::eFront, front_write_mask, dld);
cmdbuf.setStencilCompareMask(vk::StencilFaceFlagBits::eFront, front_test_mask, dld);
cmdbuf.SetStencilReference(VK_STENCIL_FACE_FRONT_BIT, front_ref);
cmdbuf.SetStencilWriteMask(VK_STENCIL_FACE_FRONT_BIT, front_write_mask);
cmdbuf.SetStencilCompareMask(VK_STENCIL_FACE_FRONT_BIT, front_test_mask);
// Back face
cmdbuf.setStencilReference(vk::StencilFaceFlagBits::eBack, back_ref, dld);
cmdbuf.setStencilWriteMask(vk::StencilFaceFlagBits::eBack, back_write_mask, dld);
cmdbuf.setStencilCompareMask(vk::StencilFaceFlagBits::eBack, back_test_mask, dld);
cmdbuf.SetStencilReference(VK_STENCIL_FACE_BACK_BIT, back_ref);
cmdbuf.SetStencilWriteMask(VK_STENCIL_FACE_BACK_BIT, back_write_mask);
cmdbuf.SetStencilCompareMask(VK_STENCIL_FACE_BACK_BIT, back_test_mask);
});
} else {
// Front face defines both faces
scheduler.Record([ref = regs.stencil_back_func_ref, write_mask = regs.stencil_back_mask,
test_mask = regs.stencil_back_func_mask](auto cmdbuf, auto& dld) {
cmdbuf.setStencilReference(vk::StencilFaceFlagBits::eFrontAndBack, ref, dld);
cmdbuf.setStencilWriteMask(vk::StencilFaceFlagBits::eFrontAndBack, write_mask, dld);
cmdbuf.setStencilCompareMask(vk::StencilFaceFlagBits::eFrontAndBack, test_mask, dld);
test_mask = regs.stencil_back_func_mask](vk::CommandBuffer cmdbuf) {
cmdbuf.SetStencilReference(VK_STENCIL_FACE_FRONT_AND_BACK, ref);
cmdbuf.SetStencilWriteMask(VK_STENCIL_FACE_FRONT_AND_BACK, write_mask);
cmdbuf.SetStencilCompareMask(VK_STENCIL_FACE_FRONT_AND_BACK, test_mask);
});
}
}

15
src/video_core/renderer_vulkan/vk_rasterizer.h
View file

@ -17,7 +17,6 @@
#include "video_core/memory_manager.h"
#include "video_core/rasterizer_accelerated.h"
#include "video_core/rasterizer_interface.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/fixed_pipeline_state.h"
#include "video_core/renderer_vulkan/vk_buffer_cache.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
@ -32,6 +31,7 @@
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Core {
class System;
@ -49,11 +49,10 @@ namespace Vulkan {
struct VKScreenInfo;
using ImageViewsPack =
boost::container::static_vector<vk::ImageView, Maxwell::NumRenderTargets + 1>;
using ImageViewsPack = boost::container::static_vector<VkImageView, Maxwell::NumRenderTargets + 1>;
struct FramebufferCacheKey {
vk::RenderPass renderpass{};
VkRenderPass renderpass{};
u32 width = 0;
u32 height = 0;
u32 layers = 0;
@ -101,7 +100,7 @@ class BufferBindings;
struct ImageView {
View view;
vk::ImageLayout* layout = nullptr;
VkImageLayout* layout = nullptr;
};
class RasterizerVulkan final : public VideoCore::RasterizerAccelerated {
@ -137,7 +136,7 @@ public:
private:
struct DrawParameters {
void Draw(vk::CommandBuffer cmdbuf, const vk::DispatchLoaderDynamic& dld) const;
void Draw(vk::CommandBuffer cmdbuf) const;
u32 base_instance = 0;
u32 num_instances = 0;
@ -154,7 +153,7 @@ private:
Texceptions UpdateAttachments();
std::tuple<vk::Framebuffer, vk::Extent2D> ConfigureFramebuffers(vk::RenderPass renderpass);
std::tuple<VkFramebuffer, VkExtent2D> ConfigureFramebuffers(VkRenderPass renderpass);
/// Setups geometry buffers and state.
DrawParameters SetupGeometry(FixedPipelineState& fixed_state, BufferBindings& buffer_bindings,
@ -272,7 +271,7 @@ private:
u32 draw_counter = 0;
// TODO(Rodrigo): Invalidate on image destruction
std::unordered_map<FramebufferCacheKey, UniqueFramebuffer> framebuffer_cache;
std::unordered_map<FramebufferCacheKey, vk::Framebuffer> framebuffer_cache;
};
} // namespace Vulkan

116
src/video_core/renderer_vulkan/vk_renderpass_cache.cpp
View file

@ -6,10 +6,10 @@
#include <vector>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_renderpass_cache.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -17,7 +17,7 @@ VKRenderPassCache::VKRenderPassCache(const VKDevice& device) : device{device} {}
VKRenderPassCache::~VKRenderPassCache() = default;
vk::RenderPass VKRenderPassCache::GetRenderPass(const RenderPassParams& params) {
VkRenderPass VKRenderPassCache::GetRenderPass(const RenderPassParams& params) {
const auto [pair, is_cache_miss] = cache.try_emplace(params);
auto& entry = pair->second;
if (is_cache_miss) {
@ -26,9 +26,9 @@ vk::RenderPass VKRenderPassCache::GetRenderPass(const RenderPassParams& params)
return *entry;
}
UniqueRenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& params) const {
std::vector<vk::AttachmentDescription> descriptors;
std::vector<vk::AttachmentReference> color_references;
vk::RenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& params) const {
std::vector<VkAttachmentDescription> descriptors;
std::vector<VkAttachmentReference> color_references;
for (std::size_t rt = 0; rt < params.color_attachments.size(); ++rt) {
const auto attachment = params.color_attachments[rt];
@ -39,16 +39,25 @@ UniqueRenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& par
// TODO(Rodrigo): Add eMayAlias when it's needed.
const auto color_layout = attachment.is_texception
? vk::ImageLayout::eGeneral
: vk::ImageLayout::eColorAttachmentOptimal;
descriptors.emplace_back(vk::AttachmentDescriptionFlagBits::eMayAlias, format.format,
vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eLoad,
vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eDontCare,
vk::AttachmentStoreOp::eDontCare, color_layout, color_layout);
color_references.emplace_back(static_cast<u32>(rt), color_layout);
? VK_IMAGE_LAYOUT_GENERAL
: VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
VkAttachmentDescription& descriptor = descriptors.emplace_back();
descriptor.flags = VK_ATTACHMENT_DESCRIPTION_MAY_ALIAS_BIT;
descriptor.format = format.format;
descriptor.samples = VK_SAMPLE_COUNT_1_BIT;
descriptor.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
descriptor.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
descriptor.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
descriptor.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
descriptor.initialLayout = color_layout;
descriptor.finalLayout = color_layout;
VkAttachmentReference& reference = color_references.emplace_back();
reference.attachment = static_cast<u32>(rt);
reference.layout = color_layout;
}
vk::AttachmentReference zeta_attachment_ref;
VkAttachmentReference zeta_attachment_ref;
if (params.has_zeta) {
const auto format =
MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, params.zeta_pixel_format);
@ -56,45 +65,68 @@ UniqueRenderPass VKRenderPassCache::CreateRenderPass(const RenderPassParams& par
static_cast<u32>(params.zeta_pixel_format));
const auto zeta_layout = params.zeta_texception
? vk::ImageLayout::eGeneral
: vk::ImageLayout::eDepthStencilAttachmentOptimal;
descriptors.emplace_back(vk::AttachmentDescriptionFlags{}, format.format,
vk::SampleCountFlagBits::e1, vk::AttachmentLoadOp::eLoad,
vk::AttachmentStoreOp::eStore, vk::AttachmentLoadOp::eLoad,
vk::AttachmentStoreOp::eStore, zeta_layout, zeta_layout);
zeta_attachment_ref =
vk::AttachmentReference(static_cast<u32>(params.color_attachments.size()), zeta_layout);
? VK_IMAGE_LAYOUT_GENERAL
: VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL;
VkAttachmentDescription& descriptor = descriptors.emplace_back();
descriptor.flags = 0;
descriptor.format = format.format;
descriptor.samples = VK_SAMPLE_COUNT_1_BIT;
descriptor.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
descriptor.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
descriptor.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
descriptor.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
descriptor.initialLayout = zeta_layout;
descriptor.finalLayout = zeta_layout;
zeta_attachment_ref.attachment = static_cast<u32>(params.color_attachments.size());
zeta_attachment_ref.layout = zeta_layout;
}
const vk::SubpassDescription subpass_description(
{}, vk::PipelineBindPoint::eGraphics, 0, nullptr, static_cast<u32>(color_references.size()),
color_references.data(), nullptr, params.has_zeta ? &zeta_attachment_ref : nullptr, 0,
nullptr);
VkSubpassDescription subpass_description;
subpass_description.flags = 0;
subpass_description.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass_description.inputAttachmentCount = 0;
subpass_description.pInputAttachments = nullptr;
subpass_description.colorAttachmentCount = static_cast<u32>(color_references.size());
subpass_description.pColorAttachments = color_references.data();
subpass_description.pResolveAttachments = nullptr;
subpass_description.pDepthStencilAttachment = params.has_zeta ? &zeta_attachment_ref : nullptr;
subpass_description.preserveAttachmentCount = 0;
subpass_description.pPreserveAttachments = nullptr;
vk::AccessFlags access;
vk::PipelineStageFlags stage;
VkAccessFlags access = 0;
VkPipelineStageFlags stage = 0;
if (!color_references.empty()) {
access |=
vk::AccessFlagBits::eColorAttachmentRead | vk::AccessFlagBits::eColorAttachmentWrite;
stage |= vk::PipelineStageFlagBits::eColorAttachmentOutput;
access |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
stage |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
}
if (params.has_zeta) {
access |= vk::AccessFlagBits::eDepthStencilAttachmentRead |
vk::AccessFlagBits::eDepthStencilAttachmentWrite;
stage |= vk::PipelineStageFlagBits::eLateFragmentTests;
access |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
stage |= VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
}
const vk::SubpassDependency subpass_dependency(VK_SUBPASS_EXTERNAL, 0, stage, stage, {}, access,
{});
VkSubpassDependency subpass_dependency;
subpass_dependency.srcSubpass = VK_SUBPASS_EXTERNAL;
subpass_dependency.dstSubpass = 0;
subpass_dependency.srcStageMask = stage;
subpass_dependency.dstStageMask = stage;
subpass_dependency.srcAccessMask = 0;
subpass_dependency.dstAccessMask = access;
subpass_dependency.dependencyFlags = 0;
const vk::RenderPassCreateInfo create_info({}, static_cast<u32>(descriptors.size()),
descriptors.data(), 1, &subpass_description, 1,
&subpass_dependency);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createRenderPassUnique(create_info, nullptr, dld);
VkRenderPassCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.attachmentCount = static_cast<u32>(descriptors.size());
ci.pAttachments = descriptors.data();
ci.subpassCount = 1;
ci.pSubpasses = &subpass_description;
ci.dependencyCount = 1;
ci.pDependencies = &subpass_dependency;
return device.GetLogical().CreateRenderPass(ci);
}
} // namespace Vulkan

8
src/video_core/renderer_vulkan/vk_renderpass_cache.h
View file

@ -12,7 +12,7 @@
#include <boost/functional/hash.hpp>
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/surface.h"
namespace Vulkan {
@ -85,13 +85,13 @@ public:
explicit VKRenderPassCache(const VKDevice& device);
~VKRenderPassCache();
vk::RenderPass GetRenderPass(const RenderPassParams& params);
VkRenderPass GetRenderPass(const RenderPassParams& params);
private:
UniqueRenderPass CreateRenderPass(const RenderPassParams& params) const;
vk::RenderPass CreateRenderPass(const RenderPassParams& params) const;
const VKDevice& device;
std::unordered_map<RenderPassParams, UniqueRenderPass> cache;
std::unordered_map<RenderPassParams, vk::RenderPass> cache;
};
} // namespace Vulkan

85
src/video_core/renderer_vulkan/vk_resource_manager.cpp
View file

@ -6,83 +6,83 @@
#include <optional>
#include "common/assert.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
namespace {
// TODO(Rodrigo): Fine tune these numbers.
constexpr std::size_t COMMAND_BUFFER_POOL_SIZE = 0x1000;
constexpr std::size_t FENCES_GROW_STEP = 0x40;
VkFenceCreateInfo BuildFenceCreateInfo() {
VkFenceCreateInfo fence_ci;
fence_ci.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO;
fence_ci.pNext = nullptr;
fence_ci.flags = 0;
return fence_ci;
}
} // Anonymous namespace
class CommandBufferPool final : public VKFencedPool {
public:
CommandBufferPool(const VKDevice& device)
: VKFencedPool(COMMAND_BUFFER_POOL_SIZE), device{device} {}
void Allocate(std::size_t begin, std::size_t end) override {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const u32 graphics_family = device.GetGraphicsFamily();
auto pool = std::make_unique<Pool>();
// Command buffers are going to be commited, recorded, executed every single usage cycle.
// They are also going to be reseted when commited.
const auto pool_flags = vk::CommandPoolCreateFlagBits::eTransient |
vk::CommandPoolCreateFlagBits::eResetCommandBuffer;
const vk::CommandPoolCreateInfo cmdbuf_pool_ci(pool_flags, graphics_family);
pool->handle = dev.createCommandPoolUnique(cmdbuf_pool_ci, nullptr, dld);
VkCommandPoolCreateInfo command_pool_ci;
command_pool_ci.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO;
command_pool_ci.pNext = nullptr;
command_pool_ci.flags =
VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT;
command_pool_ci.queueFamilyIndex = device.GetGraphicsFamily();
const vk::CommandBufferAllocateInfo cmdbuf_ai(*pool->handle,
vk::CommandBufferLevel::ePrimary,
static_cast<u32>(COMMAND_BUFFER_POOL_SIZE));
pool->cmdbufs =
dev.allocateCommandBuffersUnique<std::allocator<UniqueCommandBuffer>>(cmdbuf_ai, dld);
pools.push_back(std::move(pool));
Pool& pool = pools.emplace_back();
pool.handle = device.GetLogical().CreateCommandPool(command_pool_ci);
pool.cmdbufs = pool.handle.Allocate(COMMAND_BUFFER_POOL_SIZE);
}
vk::CommandBuffer Commit(VKFence& fence) {
VkCommandBuffer Commit(VKFence& fence) {
const std::size_t index = CommitResource(fence);
const auto pool_index = index / COMMAND_BUFFER_POOL_SIZE;
const auto sub_index = index % COMMAND_BUFFER_POOL_SIZE;
return *pools[pool_index]->cmdbufs[sub_index];
return pools[pool_index].cmdbufs[sub_index];
}
private:
struct Pool {
UniqueCommandPool handle;
std::vector<UniqueCommandBuffer> cmdbufs;
vk::CommandPool handle;
vk::CommandBuffers cmdbufs;
};
const VKDevice& device;
std::vector<std::unique_ptr<Pool>> pools;
std::vector<Pool> pools;
};
VKResource::VKResource() = default;
VKResource::~VKResource() = default;
VKFence::VKFence(const VKDevice& device, UniqueFence handle)
: device{device}, handle{std::move(handle)} {}
VKFence::VKFence(const VKDevice& device)
: device{device}, handle{device.GetLogical().CreateFence(BuildFenceCreateInfo())} {}
VKFence::~VKFence() = default;
void VKFence::Wait() {
static constexpr u64 timeout = std::numeric_limits<u64>::max();
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
switch (const auto result = dev.waitForFences(1, &*handle, true, timeout, dld)) {
case vk::Result::eSuccess:
switch (const VkResult result = handle.Wait()) {
case VK_SUCCESS:
return;
case vk::Result::eErrorDeviceLost:
case VK_ERROR_DEVICE_LOST:
device.ReportLoss();
[[fallthrough]];
default:
vk::throwResultException(result, "vk::waitForFences");
throw vk::Exception(result);
}
}
@ -107,13 +107,11 @@ bool VKFence::Tick(bool gpu_wait, bool owner_wait) {
return false;
}
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
if (gpu_wait) {
// Wait for the fence if it has been requested.
dev.waitForFences({*handle}, true, std::numeric_limits<u64>::max(), dld);
(void)handle.Wait();
} else {
if (dev.getFenceStatus(*handle, dld) != vk::Result::eSuccess) {
if (handle.GetStatus() != VK_SUCCESS) {
// Vulkan fence is not ready, not much it can do here
return false;
}
@ -126,7 +124,7 @@ bool VKFence::Tick(bool gpu_wait, bool owner_wait) {
protected_resources.clear();
// Prepare fence for reusage.
dev.resetFences({*handle}, dld);
handle.Reset();
is_used = false;
return true;
}
@ -299,21 +297,16 @@ VKFence& VKResourceManager::CommitFence() {
return *found_fence;
}
vk::CommandBuffer VKResourceManager::CommitCommandBuffer(VKFence& fence) {
VkCommandBuffer VKResourceManager::CommitCommandBuffer(VKFence& fence) {
return command_buffer_pool->Commit(fence);
}
void VKResourceManager::GrowFences(std::size_t new_fences_count) {
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const vk::FenceCreateInfo fence_ci;
const std::size_t previous_size = fences.size();
fences.resize(previous_size + new_fences_count);
std::generate(fences.begin() + previous_size, fences.end(), [&]() {
return std::make_unique<VKFence>(device, dev.createFenceUnique(fence_ci, nullptr, dld));
});
std::generate(fences.begin() + previous_size, fences.end(),
[this] { return std::make_unique<VKFence>(device); });
}
} // namespace Vulkan

10
src/video_core/renderer_vulkan/vk_resource_manager.h
View file

@ -7,7 +7,7 @@
#include <cstddef>
#include <memory>
#include <vector>
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -42,7 +42,7 @@ class VKFence {
friend class VKResourceManager;
public:
explicit VKFence(const VKDevice& device, UniqueFence handle);
explicit VKFence(const VKDevice& device);
~VKFence();
/**
@ -69,7 +69,7 @@ public:
void RedirectProtection(VKResource* old_resource, VKResource* new_resource) noexcept;
/// Retreives the fence.
operator vk::Fence() const {
operator VkFence() const {
return *handle;
}
@ -87,7 +87,7 @@ private:
bool Tick(bool gpu_wait, bool owner_wait);
const VKDevice& device; ///< Device handler
UniqueFence handle; ///< Vulkan fence
vk::Fence handle; ///< Vulkan fence
std::vector<VKResource*> protected_resources; ///< List of resources protected by this fence
bool is_owned = false; ///< The fence has been commited but not released yet.
bool is_used = false; ///< The fence has been commited but it has not been checked to be free.
@ -181,7 +181,7 @@ public:
VKFence& CommitFence();
/// Commits an unused command buffer and protects it with a fence.
vk::CommandBuffer CommitCommandBuffer(VKFence& fence);
VkCommandBuffer CommitCommandBuffer(VKFence& fence);
private:
/// Allocates new fences.

76
src/video_core/renderer_vulkan/vk_sampler_cache.cpp
View file

@ -7,64 +7,64 @@
#include <unordered_map>
#include "common/assert.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_sampler_cache.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/textures/texture.h"
namespace Vulkan {
static std::optional<vk::BorderColor> TryConvertBorderColor(std::array<float, 4> color) {
namespace {
VkBorderColor ConvertBorderColor(std::array<float, 4> color) {
// TODO(Rodrigo): Manage integer border colors
if (color == std::array<float, 4>{0, 0, 0, 0}) {
return vk::BorderColor::eFloatTransparentBlack;
return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
} else if (color == std::array<float, 4>{0, 0, 0, 1}) {
return vk::BorderColor::eFloatOpaqueBlack;
return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
} else if (color == std::array<float, 4>{1, 1, 1, 1}) {
return vk::BorderColor::eFloatOpaqueWhite;
return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
}
if (color[0] + color[1] + color[2] > 1.35f) {
// If color elements are brighter than roughly 0.5 average, use white border
return VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE;
} else if (color[3] > 0.5f) {
return VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
} else {
if (color[0] + color[1] + color[2] > 1.35f) {
// If color elements are brighter than roughly 0.5 average, use white border
return vk::BorderColor::eFloatOpaqueWhite;
}
if (color[3] > 0.5f) {
return vk::BorderColor::eFloatOpaqueBlack;
}
return vk::BorderColor::eFloatTransparentBlack;
return VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
}
}
} // Anonymous namespace
VKSamplerCache::VKSamplerCache(const VKDevice& device) : device{device} {}
VKSamplerCache::~VKSamplerCache() = default;
UniqueSampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc) const {
const float max_anisotropy{tsc.GetMaxAnisotropy()};
const bool has_anisotropy{max_anisotropy > 1.0f};
const auto border_color{tsc.GetBorderColor()};
const auto vk_border_color{TryConvertBorderColor(border_color)};
constexpr bool unnormalized_coords{false};
const vk::SamplerCreateInfo sampler_ci(
{}, MaxwellToVK::Sampler::Filter(tsc.mag_filter),
MaxwellToVK::Sampler::Filter(tsc.min_filter),
MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter),
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter),
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter),
MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter), tsc.GetLodBias(),
has_anisotropy, max_anisotropy, tsc.depth_compare_enabled,
MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func), tsc.GetMinLod(),
tsc.GetMaxLod(), vk_border_color.value_or(vk::BorderColor::eFloatTransparentBlack),
unnormalized_coords);
const auto& dld{device.GetDispatchLoader()};
const auto dev{device.GetLogical()};
return dev.createSamplerUnique(sampler_ci, nullptr, dld);
vk::Sampler VKSamplerCache::CreateSampler(const Tegra::Texture::TSCEntry& tsc) const {
VkSamplerCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.magFilter = MaxwellToVK::Sampler::Filter(tsc.mag_filter);
ci.minFilter = MaxwellToVK::Sampler::Filter(tsc.min_filter);
ci.mipmapMode = MaxwellToVK::Sampler::MipmapMode(tsc.mipmap_filter);
ci.addressModeU = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_u, tsc.mag_filter);
ci.addressModeV = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_v, tsc.mag_filter);
ci.addressModeW = MaxwellToVK::Sampler::WrapMode(device, tsc.wrap_p, tsc.mag_filter);
ci.mipLodBias = tsc.GetLodBias();
ci.anisotropyEnable = tsc.GetMaxAnisotropy() > 1.0f ? VK_TRUE : VK_FALSE;
ci.maxAnisotropy = tsc.GetMaxAnisotropy();
ci.compareEnable = tsc.depth_compare_enabled;
ci.compareOp = MaxwellToVK::Sampler::DepthCompareFunction(tsc.depth_compare_func);
ci.minLod = tsc.GetMinLod();
ci.maxLod = tsc.GetMaxLod();
ci.borderColor = ConvertBorderColor(tsc.GetBorderColor());
ci.unnormalizedCoordinates = VK_FALSE;
return device.GetLogical().CreateSampler(ci);
}
vk::Sampler VKSamplerCache::ToSamplerType(const UniqueSampler& sampler) const {
VkSampler VKSamplerCache::ToSamplerType(const vk::Sampler& sampler) const {
return *sampler;
}

8
src/video_core/renderer_vulkan/vk_sampler_cache.h
View file

@ -4,7 +4,7 @@
#pragma once
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/sampler_cache.h"
#include "video_core/textures/texture.h"
@ -12,15 +12,15 @@ namespace Vulkan {
class VKDevice;
class VKSamplerCache final : public VideoCommon::SamplerCache<vk::Sampler, UniqueSampler> {
class VKSamplerCache final : public VideoCommon::SamplerCache<VkSampler, vk::Sampler> {
public:
explicit VKSamplerCache(const VKDevice& device);
~VKSamplerCache();
protected:
UniqueSampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
vk::Sampler CreateSampler(const Tegra::Texture::TSCEntry& tsc) const override;
vk::Sampler ToSamplerType(const UniqueSampler& sampler) const override;
VkSampler ToSamplerType(const vk::Sampler& sampler) const override;
private:
const VKDevice& device;

87
src/video_core/renderer_vulkan/vk_scheduler.cpp
View file

@ -10,23 +10,22 @@
#include "common/assert.h"
#include "common/microprofile.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_query_cache.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_state_tracker.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
MICROPROFILE_DECLARE(Vulkan_WaitForWorker);
void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf,
const vk::DispatchLoaderDynamic& dld) {
void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf) {
auto command = first;
while (command != nullptr) {
auto next = command->GetNext();
command->Execute(cmdbuf, dld);
command->Execute(cmdbuf);
command->~Command();
command = next;
}
@ -51,7 +50,7 @@ VKScheduler::~VKScheduler() {
worker_thread.join();
}
void VKScheduler::Flush(bool release_fence, vk::Semaphore semaphore) {
void VKScheduler::Flush(bool release_fence, VkSemaphore semaphore) {
SubmitExecution(semaphore);
if (release_fence) {
current_fence->Release();
@ -59,7 +58,7 @@ void VKScheduler::Flush(bool release_fence, vk::Semaphore semaphore) {
AllocateNewContext();
}
void VKScheduler::Finish(bool release_fence, vk::Semaphore semaphore) {
void VKScheduler::Finish(bool release_fence, VkSemaphore semaphore) {
SubmitExecution(semaphore);
current_fence->Wait();
if (release_fence) {
@ -89,17 +88,34 @@ void VKScheduler::DispatchWork() {
AcquireNewChunk();
}
void VKScheduler::RequestRenderpass(const vk::RenderPassBeginInfo& renderpass_bi) {
if (state.renderpass && renderpass_bi == *state.renderpass) {
void VKScheduler::RequestRenderpass(VkRenderPass renderpass, VkFramebuffer framebuffer,
VkExtent2D render_area) {
if (renderpass == state.renderpass && framebuffer == state.framebuffer &&
render_area.width == state.render_area.width &&
render_area.height == state.render_area.height) {
return;
}
const bool end_renderpass = state.renderpass.has_value();
state.renderpass = renderpass_bi;
Record([renderpass_bi, end_renderpass](auto cmdbuf, auto& dld) {
const bool end_renderpass = state.renderpass != nullptr;
state.renderpass = renderpass;
state.framebuffer = framebuffer;
state.render_area = render_area;
VkRenderPassBeginInfo renderpass_bi;
renderpass_bi.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO;
renderpass_bi.pNext = nullptr;
renderpass_bi.renderPass = renderpass;
renderpass_bi.framebuffer = framebuffer;
renderpass_bi.renderArea.offset.x = 0;
renderpass_bi.renderArea.offset.y = 0;
renderpass_bi.renderArea.extent = render_area;
renderpass_bi.clearValueCount = 0;
renderpass_bi.pClearValues = nullptr;
Record([renderpass_bi, end_renderpass](vk::CommandBuffer cmdbuf) {
if (end_renderpass) {
cmdbuf.endRenderPass(dld);
cmdbuf.EndRenderPass();
}
cmdbuf.beginRenderPass(renderpass_bi, vk::SubpassContents::eInline, dld);
cmdbuf.BeginRenderPass(renderpass_bi, VK_SUBPASS_CONTENTS_INLINE);
});
}
@ -107,13 +123,13 @@ void VKScheduler::RequestOutsideRenderPassOperationContext() {
EndRenderPass();
}
void VKScheduler::BindGraphicsPipeline(vk::Pipeline pipeline) {
void VKScheduler::BindGraphicsPipeline(VkPipeline pipeline) {
if (state.graphics_pipeline == pipeline) {
return;
}
state.graphics_pipeline = pipeline;
Record([pipeline](auto cmdbuf, auto& dld) {
cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline, dld);
Record([pipeline](vk::CommandBuffer cmdbuf) {
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
});
}
@ -126,37 +142,50 @@ void VKScheduler::WorkerThread() {
}
auto extracted_chunk = std::move(chunk_queue.Front());
chunk_queue.Pop();
extracted_chunk->ExecuteAll(current_cmdbuf, device.GetDispatchLoader());
extracted_chunk->ExecuteAll(current_cmdbuf);
chunk_reserve.Push(std::move(extracted_chunk));
} while (!quit);
}
void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
void VKScheduler::SubmitExecution(VkSemaphore semaphore) {
EndPendingOperations();
InvalidateState();
WaitWorker();
std::unique_lock lock{mutex};
const auto queue = device.GetGraphicsQueue();
const auto& dld = device.GetDispatchLoader();
current_cmdbuf.end(dld);
current_cmdbuf.End();
const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1U : 0U,
&semaphore);
queue.submit({submit_info}, static_cast<vk::Fence>(*current_fence), dld);
VkSubmitInfo submit_info;
submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO;
submit_info.pNext = nullptr;
submit_info.waitSemaphoreCount = 0;
submit_info.pWaitSemaphores = nullptr;
submit_info.pWaitDstStageMask = nullptr;
submit_info.commandBufferCount = 1;
submit_info.pCommandBuffers = current_cmdbuf.address();
submit_info.signalSemaphoreCount = semaphore ? 1 : 0;
submit_info.pSignalSemaphores = &semaphore;
device.GetGraphicsQueue().Submit(submit_info, *current_fence);
}
void VKScheduler::AllocateNewContext() {
++ticks;
VkCommandBufferBeginInfo cmdbuf_bi;
cmdbuf_bi.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO;
cmdbuf_bi.pNext = nullptr;
cmdbuf_bi.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT;
cmdbuf_bi.pInheritanceInfo = nullptr;
std::unique_lock lock{mutex};
current_fence = next_fence;
next_fence = &resource_manager.CommitFence();
current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit},
device.GetDispatchLoader());
current_cmdbuf = vk::CommandBuffer(resource_manager.CommitCommandBuffer(*current_fence),
device.GetDispatchLoader());
current_cmdbuf.Begin(cmdbuf_bi);
// Enable counters once again. These are disabled when a command buffer is finished.
if (query_cache) {
query_cache->UpdateCounters();
@ -177,8 +206,8 @@ void VKScheduler::EndRenderPass() {
if (!state.renderpass) {
return;
}
state.renderpass = std::nullopt;
Record([](auto cmdbuf, auto& dld) { cmdbuf.endRenderPass(dld); });
state.renderpass = nullptr;
Record([](vk::CommandBuffer cmdbuf) { cmdbuf.EndRenderPass(); });
}
void VKScheduler::AcquireNewChunk() {

29
src/video_core/renderer_vulkan/vk_scheduler.h
View file

@ -13,7 +13,7 @@
#include <utility>
#include "common/common_types.h"
#include "common/threadsafe_queue.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -49,10 +49,10 @@ public:
~VKScheduler();
/// Sends the current execution context to the GPU.
void Flush(bool release_fence = true, vk::Semaphore semaphore = nullptr);
void Flush(bool release_fence = true, VkSemaphore semaphore = nullptr);
/// Sends the current execution context to the GPU and waits for it to complete.
void Finish(bool release_fence = true, vk::Semaphore semaphore = nullptr);
void Finish(bool release_fence = true, VkSemaphore semaphore = nullptr);
/// Waits for the worker thread to finish executing everything. After this function returns it's
/// safe to touch worker resources.
@ -62,14 +62,15 @@ public:
void DispatchWork();
/// Requests to begin a renderpass.
void RequestRenderpass(const vk::RenderPassBeginInfo& renderpass_bi);
void RequestRenderpass(VkRenderPass renderpass, VkFramebuffer framebuffer,
VkExtent2D render_area);
/// Requests the current executino context to be able to execute operations only allowed outside
/// of a renderpass.
void RequestOutsideRenderPassOperationContext();
/// Binds a pipeline to the current execution context.
void BindGraphicsPipeline(vk::Pipeline pipeline);
void BindGraphicsPipeline(VkPipeline pipeline);
/// Assigns the query cache.
void SetQueryCache(VKQueryCache& query_cache_) {
@ -101,8 +102,7 @@ private:
public:
virtual ~Command() = default;
virtual void Execute(vk::CommandBuffer cmdbuf,
const vk::DispatchLoaderDynamic& dld) const = 0;
virtual void Execute(vk::CommandBuffer cmdbuf) const = 0;
Command* GetNext() const {
return next;
@ -125,9 +125,8 @@ private:
TypedCommand(TypedCommand&&) = delete;
TypedCommand& operator=(TypedCommand&&) = delete;
void Execute(vk::CommandBuffer cmdbuf,
const vk::DispatchLoaderDynamic& dld) const override {
command(cmdbuf, dld);
void Execute(vk::CommandBuffer cmdbuf) const override {
command(cmdbuf);
}
private:
@ -136,7 +135,7 @@ private:
class CommandChunk final {
public:
void ExecuteAll(vk::CommandBuffer cmdbuf, const vk::DispatchLoaderDynamic& dld);
void ExecuteAll(vk::CommandBuffer cmdbuf);
template <typename T>
bool Record(T& command) {
@ -175,7 +174,7 @@ private:
void WorkerThread();
void SubmitExecution(vk::Semaphore semaphore);
void SubmitExecution(VkSemaphore semaphore);
void AllocateNewContext();
@ -198,8 +197,10 @@ private:
VKFence* next_fence = nullptr;
struct State {
std::optional<vk::RenderPassBeginInfo> renderpass;
vk::Pipeline graphics_pipeline;
VkRenderPass renderpass = nullptr;
VkFramebuffer framebuffer = nullptr;
VkExtent2D render_area = {0, 0};
VkPipeline graphics_pipeline = nullptr;
} state;
std::unique_ptr<CommandChunk> chunk;

2
src/video_core/renderer_vulkan/vk_shader_decompiler.cpp
View file

@ -801,7 +801,7 @@ private:
if (IsOutputAttributeArray()) {
const u32 num = GetNumOutputVertices();
type = TypeArray(type, Constant(t_uint, num));
if (device.GetDriverID() != vk::DriverIdKHR::eIntelProprietaryWindows) {
if (device.GetDriverID() != VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR) {
// Intel's proprietary driver fails to setup defaults for arrayed output
// attributes.
varying_default = ConstantComposite(type, std::vector(num, varying_default));

20
src/video_core/renderer_vulkan/vk_shader_util.cpp
View file

@ -8,27 +8,25 @@
#include "common/alignment.h"
#include "common/assert.h"
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
UniqueShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data) {
vk::ShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data) {
// Avoid undefined behavior by copying to a staging allocation
ASSERT(code_size % sizeof(u32) == 0);
const auto data = std::make_unique<u32[]>(code_size / sizeof(u32));
std::memcpy(data.get(), code_data, code_size);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const vk::ShaderModuleCreateInfo shader_ci({}, code_size, data.get());
vk::ShaderModule shader_module;
if (dev.createShaderModule(&shader_ci, nullptr, &shader_module, dld) != vk::Result::eSuccess) {
UNREACHABLE_MSG("Shader module failed to build!");
}
return UniqueShaderModule(shader_module, vk::ObjectDestroy(dev, nullptr, dld));
VkShaderModuleCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.codeSize = code_size;
ci.pCode = data.get();
return device.GetLogical().CreateShaderModule(ci);
}
} // namespace Vulkan

4
src/video_core/renderer_vulkan/vk_shader_util.h
View file

@ -6,12 +6,12 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
class VKDevice;
UniqueShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data);
vk::ShaderModule BuildShader(const VKDevice& device, std::size_t code_size, const u8* code_data);
} // namespace Vulkan

25
src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
View file

@ -13,6 +13,7 @@
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -71,17 +72,23 @@ VKBuffer* VKStagingBufferPool::TryGetReservedBuffer(std::size_t size, bool host_
}
VKBuffer& VKStagingBufferPool::CreateStagingBuffer(std::size_t size, bool host_visible) {
const auto usage =
vk::BufferUsageFlagBits::eTransferSrc | vk::BufferUsageFlagBits::eTransferDst |
vk::BufferUsageFlagBits::eUniformBuffer | vk::BufferUsageFlagBits::eStorageBuffer |
vk::BufferUsageFlagBits::eIndexBuffer;
const u32 log2 = Common::Log2Ceil64(size);
const vk::BufferCreateInfo buffer_ci({}, 1ULL << log2, usage, vk::SharingMode::eExclusive, 0,
nullptr);
const auto dev = device.GetLogical();
VkBufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.size = 1ULL << log2;
ci.usage = VK_BUFFER_USAGE_TRANSFER_SRC_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT |
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT |
VK_BUFFER_USAGE_INDEX_BUFFER_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
auto buffer = std::make_unique<VKBuffer>();
buffer->handle = dev.createBufferUnique(buffer_ci, nullptr, device.GetDispatchLoader());
buffer->commit = memory_manager.Commit(*buffer->handle, host_visible);
buffer->handle = device.GetLogical().CreateBuffer(ci);
buffer->commit = memory_manager.Commit(buffer->handle, host_visible);
auto& entries = GetCache(host_visible)[log2].entries;
return *entries.emplace_back(std::move(buffer), scheduler.GetFence(), epoch).buffer;

4
src/video_core/renderer_vulkan/vk_staging_buffer_pool.h
View file

@ -11,9 +11,9 @@
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -22,7 +22,7 @@ class VKFenceWatch;
class VKScheduler;
struct VKBuffer final {
UniqueBuffer handle;
vk::Buffer handle;
VKMemoryCommit commit;
};

62
src/video_core/renderer_vulkan/vk_stream_buffer.cpp
View file

@ -9,11 +9,11 @@
#include "common/alignment.h"
#include "common/assert.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_stream_buffer.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -25,8 +25,8 @@ constexpr u64 WATCHES_RESERVE_CHUNK = 0x1000;
constexpr u64 STREAM_BUFFER_SIZE = 256 * 1024 * 1024;
std::optional<u32> FindMemoryType(const VKDevice& device, u32 filter,
vk::MemoryPropertyFlags wanted) {
const auto properties = device.GetPhysical().getMemoryProperties(device.GetDispatchLoader());
VkMemoryPropertyFlags wanted) {
const auto properties = device.GetPhysical().GetMemoryProperties();
for (u32 i = 0; i < properties.memoryTypeCount; i++) {
if (!(filter & (1 << i))) {
continue;
@ -35,13 +35,13 @@ std::optional<u32> FindMemoryType(const VKDevice& device, u32 filter,
return i;
}
}
return {};
return std::nullopt;
}
} // Anonymous namespace
VKStreamBuffer::VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
vk::BufferUsageFlags usage)
VkBufferUsageFlags usage)
: device{device}, scheduler{scheduler} {
CreateBuffers(usage);
ReserveWatches(current_watches, WATCHES_INITIAL_RESERVE);
@ -78,17 +78,13 @@ std::tuple<u8*, u64, bool> VKStreamBuffer::Map(u64 size, u64 alignment) {
invalidated = true;
}
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
const auto pointer = reinterpret_cast<u8*>(dev.mapMemory(*memory, offset, size, {}, dld));
return {pointer, offset, invalidated};
return {memory.Map(offset, size), offset, invalidated};
}
void VKStreamBuffer::Unmap(u64 size) {
ASSERT_MSG(size <= mapped_size, "Reserved size is too small");
const auto dev = device.GetLogical();
dev.unmapMemory(*memory, device.GetDispatchLoader());
memory.Unmap();
offset += size;
@ -101,30 +97,42 @@ void VKStreamBuffer::Unmap(u64 size) {
watch.fence.Watch(scheduler.GetFence());
}
void VKStreamBuffer::CreateBuffers(vk::BufferUsageFlags usage) {
const vk::BufferCreateInfo buffer_ci({}, STREAM_BUFFER_SIZE, usage, vk::SharingMode::eExclusive,
0, nullptr);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
buffer = dev.createBufferUnique(buffer_ci, nullptr, dld);
void VKStreamBuffer::CreateBuffers(VkBufferUsageFlags usage) {
VkBufferCreateInfo buffer_ci;
buffer_ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
buffer_ci.pNext = nullptr;
buffer_ci.flags = 0;
buffer_ci.size = STREAM_BUFFER_SIZE;
buffer_ci.usage = usage;
buffer_ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
buffer_ci.queueFamilyIndexCount = 0;
buffer_ci.pQueueFamilyIndices = nullptr;
const auto requirements = dev.getBufferMemoryRequirements(*buffer, dld);
const auto& dev = device.GetLogical();
buffer = dev.CreateBuffer(buffer_ci);
const auto& dld = device.GetDispatchLoader();
const auto requirements = dev.GetBufferMemoryRequirements(*buffer);
// Prefer device local host visible allocations (this should hit AMD's pinned memory).
auto type = FindMemoryType(device, requirements.memoryTypeBits,
vk::MemoryPropertyFlagBits::eHostVisible |
vk::MemoryPropertyFlagBits::eHostCoherent |
vk::MemoryPropertyFlagBits::eDeviceLocal);
auto type =
FindMemoryType(device, requirements.memoryTypeBits,
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
if (!type) {
// Otherwise search for a host visible allocation.
type = FindMemoryType(device, requirements.memoryTypeBits,
vk::MemoryPropertyFlagBits::eHostVisible |
vk::MemoryPropertyFlagBits::eHostCoherent);
VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
VK_MEMORY_PROPERTY_HOST_COHERENT_BIT);
ASSERT_MSG(type, "No host visible and coherent memory type found");
}
const vk::MemoryAllocateInfo alloc_ci(requirements.size, *type);
memory = dev.allocateMemoryUnique(alloc_ci, nullptr, dld);
VkMemoryAllocateInfo memory_ai;
memory_ai.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
memory_ai.pNext = nullptr;
memory_ai.allocationSize = requirements.size;
memory_ai.memoryTypeIndex = *type;
dev.bindBufferMemory(*buffer, *memory, 0, dld);
memory = dev.AllocateMemory(memory_ai);
buffer.BindMemory(*memory, 0);
}
void VKStreamBuffer::ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size) {

18
src/video_core/renderer_vulkan/vk_stream_buffer.h
View file

@ -9,7 +9,7 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -21,7 +21,7 @@ class VKScheduler;
class VKStreamBuffer final {
public:
explicit VKStreamBuffer(const VKDevice& device, VKScheduler& scheduler,
vk::BufferUsageFlags usage);
VkBufferUsageFlags usage);
~VKStreamBuffer();
/**
@ -35,7 +35,7 @@ public:
/// Ensures that "size" bytes of memory are available to the GPU, potentially recording a copy.
void Unmap(u64 size);
vk::Buffer GetHandle() const {
VkBuffer GetHandle() const {
return *buffer;
}
@ -46,20 +46,18 @@ private:
};
/// Creates Vulkan buffer handles committing the required the required memory.
void CreateBuffers(vk::BufferUsageFlags usage);
void CreateBuffers(VkBufferUsageFlags usage);
/// Increases the amount of watches available.
void ReserveWatches(std::vector<Watch>& watches, std::size_t grow_size);
void WaitPendingOperations(u64 requested_upper_bound);
const VKDevice& device; ///< Vulkan device manager.
VKScheduler& scheduler; ///< Command scheduler.
const vk::AccessFlags access; ///< Access usage of this stream buffer.
const vk::PipelineStageFlags pipeline_stage; ///< Pipeline usage of this stream buffer.
const VKDevice& device; ///< Vulkan device manager.
VKScheduler& scheduler; ///< Command scheduler.
UniqueBuffer buffer; ///< Mapped buffer.
UniqueDeviceMemory memory; ///< Memory allocation.
vk::Buffer buffer; ///< Mapped buffer.
vk::DeviceMemory memory; ///< Memory allocation.
u64 offset{}; ///< Buffer iterator.
u64 mapped_size{}; ///< Size reserved for the current copy.

155
src/video_core/renderer_vulkan/vk_swapchain.cpp
View file

@ -11,69 +11,64 @@
#include "common/logging/log.h"
#include "core/core.h"
#include "core/frontend/framebuffer_layout.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_resource_manager.h"
#include "video_core/renderer_vulkan/vk_swapchain.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
namespace {
vk::SurfaceFormatKHR ChooseSwapSurfaceFormat(const std::vector<vk::SurfaceFormatKHR>& formats,
bool srgb) {
if (formats.size() == 1 && formats[0].format == vk::Format::eUndefined) {
vk::SurfaceFormatKHR format;
format.format = vk::Format::eB8G8R8A8Unorm;
format.colorSpace = vk::ColorSpaceKHR::eSrgbNonlinear;
VkSurfaceFormatKHR ChooseSwapSurfaceFormat(vk::Span<VkSurfaceFormatKHR> formats, bool srgb) {
if (formats.size() == 1 && formats[0].format == VK_FORMAT_UNDEFINED) {
VkSurfaceFormatKHR format;
format.format = VK_FORMAT_B8G8R8A8_UNORM;
format.colorSpace = VK_COLOR_SPACE_SRGB_NONLINEAR_KHR;
return format;
}
const auto& found = std::find_if(formats.begin(), formats.end(), [srgb](const auto& format) {
const auto request_format = srgb ? vk::Format::eB8G8R8A8Srgb : vk::Format::eB8G8R8A8Unorm;
const auto request_format = srgb ? VK_FORMAT_B8G8R8A8_SRGB : VK_FORMAT_B8G8R8A8_UNORM;
return format.format == request_format &&
format.colorSpace == vk::ColorSpaceKHR::eSrgbNonlinear;
format.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR;
});
return found != formats.end() ? *found : formats[0];
}
vk::PresentModeKHR ChooseSwapPresentMode(const std::vector<vk::PresentModeKHR>& modes) {
VkPresentModeKHR ChooseSwapPresentMode(vk::Span<VkPresentModeKHR> modes) {
// Mailbox doesn't lock the application like fifo (vsync), prefer it
const auto& found = std::find_if(modes.begin(), modes.end(), [](const auto& mode) {
return mode == vk::PresentModeKHR::eMailbox;
});
return found != modes.end() ? *found : vk::PresentModeKHR::eFifo;
const auto found = std::find(modes.begin(), modes.end(), VK_PRESENT_MODE_MAILBOX_KHR);
return found != modes.end() ? *found : VK_PRESENT_MODE_FIFO_KHR;
}
vk::Extent2D ChooseSwapExtent(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width,
u32 height) {
VkExtent2D ChooseSwapExtent(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height) {
constexpr auto undefined_size{std::numeric_limits<u32>::max()};
if (capabilities.currentExtent.width != undefined_size) {
return capabilities.currentExtent;
}
vk::Extent2D extent = {width, height};
VkExtent2D extent;
extent.width = std::max(capabilities.minImageExtent.width,
std::min(capabilities.maxImageExtent.width, extent.width));
std::min(capabilities.maxImageExtent.width, width));
extent.height = std::max(capabilities.minImageExtent.height,
std::min(capabilities.maxImageExtent.height, extent.height));
std::min(capabilities.maxImageExtent.height, height));
return extent;
}
} // Anonymous namespace
VKSwapchain::VKSwapchain(vk::SurfaceKHR surface, const VKDevice& device)
VKSwapchain::VKSwapchain(VkSurfaceKHR surface, const VKDevice& device)
: surface{surface}, device{device} {}
VKSwapchain::~VKSwapchain() = default;
void VKSwapchain::Create(u32 width, u32 height, bool srgb) {
const auto& dld = device.GetDispatchLoader();
const auto physical_device = device.GetPhysical();
const auto capabilities{physical_device.getSurfaceCapabilitiesKHR(surface, dld)};
const auto capabilities{physical_device.GetSurfaceCapabilitiesKHR(surface)};
if (capabilities.maxImageExtent.width == 0 || capabilities.maxImageExtent.height == 0) {
return;
}
device.GetLogical().waitIdle(dld);
device.GetLogical().WaitIdle();
Destroy();
CreateSwapchain(capabilities, width, height, srgb);
@ -84,10 +79,8 @@ void VKSwapchain::Create(u32 width, u32 height, bool srgb) {
}
void VKSwapchain::AcquireNextImage() {
const auto dev{device.GetLogical()};
const auto& dld{device.GetDispatchLoader()};
dev.acquireNextImageKHR(*swapchain, std::numeric_limits<u64>::max(),
*present_semaphores[frame_index], {}, &image_index, dld);
device.GetLogical().AcquireNextImageKHR(*swapchain, std::numeric_limits<u64>::max(),
*present_semaphores[frame_index], {}, &image_index);
if (auto& fence = fences[image_index]; fence) {
fence->Wait();
@ -96,29 +89,37 @@ void VKSwapchain::AcquireNextImage() {
}
}
bool VKSwapchain::Present(vk::Semaphore render_semaphore, VKFence& fence) {
const vk::Semaphore present_semaphore{*present_semaphores[frame_index]};
const std::array<vk::Semaphore, 2> semaphores{present_semaphore, render_semaphore};
const u32 wait_semaphore_count{render_semaphore ? 2U : 1U};
const auto& dld{device.GetDispatchLoader()};
bool VKSwapchain::Present(VkSemaphore render_semaphore, VKFence& fence) {
const VkSemaphore present_semaphore{*present_semaphores[frame_index]};
const std::array<VkSemaphore, 2> semaphores{present_semaphore, render_semaphore};
const auto present_queue{device.GetPresentQueue()};
bool recreated = false;
const vk::PresentInfoKHR present_info(wait_semaphore_count, semaphores.data(), 1,
&swapchain.get(), &image_index, {});
switch (const auto result = present_queue.presentKHR(&present_info, dld); result) {
case vk::Result::eSuccess:
VkPresentInfoKHR present_info;
present_info.sType = VK_STRUCTURE_TYPE_PRESENT_INFO_KHR;
present_info.pNext = nullptr;
present_info.waitSemaphoreCount = render_semaphore ? 2U : 1U;
present_info.pWaitSemaphores = semaphores.data();
present_info.swapchainCount = 1;
present_info.pSwapchains = swapchain.address();
present_info.pImageIndices = &image_index;
present_info.pResults = nullptr;
switch (const VkResult result = present_queue.Present(present_info)) {
case VK_SUCCESS:
break;
case vk::Result::eErrorOutOfDateKHR:
case VK_SUBOPTIMAL_KHR:
LOG_DEBUG(Render_Vulkan, "Suboptimal swapchain");
break;
case VK_ERROR_OUT_OF_DATE_KHR:
if (current_width > 0 && current_height > 0) {
Create(current_width, current_height, current_srgb);
recreated = true;
}
break;
default:
LOG_CRITICAL(Render_Vulkan, "Vulkan failed to present swapchain due to {}!",
vk::to_string(result));
UNREACHABLE();
LOG_CRITICAL(Render_Vulkan, "Failed to present with error {}", vk::ToString(result));
break;
}
ASSERT(fences[image_index] == nullptr);
@ -132,74 +133,92 @@ bool VKSwapchain::HasFramebufferChanged(const Layout::FramebufferLayout& framebu
return framebuffer.width != current_width || framebuffer.height != current_height;
}
void VKSwapchain::CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width,
void VKSwapchain::CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width,
u32 height, bool srgb) {
const auto& dld{device.GetDispatchLoader()};
const auto physical_device{device.GetPhysical()};
const auto formats{physical_device.getSurfaceFormatsKHR(surface, dld)};
const auto present_modes{physical_device.getSurfacePresentModesKHR(surface, dld)};
const auto formats{physical_device.GetSurfaceFormatsKHR(surface)};
const auto present_modes{physical_device.GetSurfacePresentModesKHR(surface)};
const vk::SurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats, srgb)};
const vk::PresentModeKHR present_mode{ChooseSwapPresentMode(present_modes)};
const VkSurfaceFormatKHR surface_format{ChooseSwapSurfaceFormat(formats, srgb)};
const VkPresentModeKHR present_mode{ChooseSwapPresentMode(present_modes)};
u32 requested_image_count{capabilities.minImageCount + 1};
if (capabilities.maxImageCount > 0 && requested_image_count > capabilities.maxImageCount) {
requested_image_count = capabilities.maxImageCount;
}
vk::SwapchainCreateInfoKHR swapchain_ci(
{}, surface, requested_image_count, surface_format.format, surface_format.colorSpace, {}, 1,
vk::ImageUsageFlagBits::eColorAttachment, {}, {}, {}, capabilities.currentTransform,
vk::CompositeAlphaFlagBitsKHR::eOpaque, present_mode, false, {});
VkSwapchainCreateInfoKHR swapchain_ci;
swapchain_ci.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR;
swapchain_ci.pNext = nullptr;
swapchain_ci.flags = 0;
swapchain_ci.surface = surface;
swapchain_ci.minImageCount = requested_image_count;
swapchain_ci.imageFormat = surface_format.format;
swapchain_ci.imageColorSpace = surface_format.colorSpace;
swapchain_ci.imageArrayLayers = 1;
swapchain_ci.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
swapchain_ci.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
swapchain_ci.queueFamilyIndexCount = 0;
swapchain_ci.pQueueFamilyIndices = nullptr;
swapchain_ci.preTransform = capabilities.currentTransform;
swapchain_ci.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR;
swapchain_ci.presentMode = present_mode;
swapchain_ci.clipped = VK_FALSE;
swapchain_ci.oldSwapchain = nullptr;
const u32 graphics_family{device.GetGraphicsFamily()};
const u32 present_family{device.GetPresentFamily()};
const std::array<u32, 2> queue_indices{graphics_family, present_family};
if (graphics_family != present_family) {
swapchain_ci.imageSharingMode = vk::SharingMode::eConcurrent;
swapchain_ci.imageSharingMode = VK_SHARING_MODE_CONCURRENT;
swapchain_ci.queueFamilyIndexCount = static_cast<u32>(queue_indices.size());
swapchain_ci.pQueueFamilyIndices = queue_indices.data();
} else {
swapchain_ci.imageSharingMode = vk::SharingMode::eExclusive;
swapchain_ci.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE;
}
// Request the size again to reduce the possibility of a TOCTOU race condition.
const auto updated_capabilities = physical_device.getSurfaceCapabilitiesKHR(surface, dld);
const auto updated_capabilities = physical_device.GetSurfaceCapabilitiesKHR(surface);
swapchain_ci.imageExtent = ChooseSwapExtent(updated_capabilities, width, height);
// Don't add code within this and the swapchain creation.
const auto dev{device.GetLogical()};
swapchain = dev.createSwapchainKHRUnique(swapchain_ci, nullptr, dld);
swapchain = device.GetLogical().CreateSwapchainKHR(swapchain_ci);
extent = swapchain_ci.imageExtent;
current_width = extent.width;
current_height = extent.height;
current_srgb = srgb;
images = dev.getSwapchainImagesKHR(*swapchain, dld);
images = swapchain.GetImages();
image_count = static_cast<u32>(images.size());
image_format = surface_format.format;
}
void VKSwapchain::CreateSemaphores() {
const auto dev{device.GetLogical()};
const auto& dld{device.GetDispatchLoader()};
present_semaphores.resize(image_count);
for (std::size_t i = 0; i < image_count; i++) {
present_semaphores[i] = dev.createSemaphoreUnique({}, nullptr, dld);
}
std::generate(present_semaphores.begin(), present_semaphores.end(),
[this] { return device.GetLogical().CreateSemaphore(); });
}
void VKSwapchain::CreateImageViews() {
const auto dev{device.GetLogical()};
const auto& dld{device.GetDispatchLoader()};
VkImageViewCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
// ci.image
ci.viewType = VK_IMAGE_VIEW_TYPE_2D;
ci.format = image_format;
ci.components = {VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY,
VK_COMPONENT_SWIZZLE_IDENTITY, VK_COMPONENT_SWIZZLE_IDENTITY};
ci.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
ci.subresourceRange.baseMipLevel = 0;
ci.subresourceRange.levelCount = 1;
ci.subresourceRange.baseArrayLayer = 0;
ci.subresourceRange.layerCount = 1;
image_views.resize(image_count);
for (std::size_t i = 0; i < image_count; i++) {
const vk::ImageViewCreateInfo image_view_ci({}, images[i], vk::ImageViewType::e2D,
image_format, {},
{vk::ImageAspectFlagBits::eColor, 0, 1, 0, 1});
image_views[i] = dev.createImageViewUnique(image_view_ci, nullptr, dld);
ci.image = images[i];
image_views[i] = device.GetLogical().CreateImageView(ci);
}
}

32
src/video_core/renderer_vulkan/vk_swapchain.h
View file

@ -7,7 +7,7 @@
#include <vector>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Layout {
struct FramebufferLayout;
@ -20,7 +20,7 @@ class VKFence;
class VKSwapchain {
public:
explicit VKSwapchain(vk::SurfaceKHR surface, const VKDevice& device);
explicit VKSwapchain(VkSurfaceKHR surface, const VKDevice& device);
~VKSwapchain();
/// Creates (or recreates) the swapchain with a given size.
@ -31,12 +31,12 @@ public:
/// Presents the rendered image to the swapchain. Returns true when the swapchains had to be
/// recreated. Takes responsability for the ownership of fence.
bool Present(vk::Semaphore render_semaphore, VKFence& fence);
bool Present(VkSemaphore render_semaphore, VKFence& fence);
/// Returns true when the framebuffer layout has changed.
bool HasFramebufferChanged(const Layout::FramebufferLayout& framebuffer) const;
const vk::Extent2D& GetSize() const {
VkExtent2D GetSize() const {
return extent;
}
@ -48,15 +48,15 @@ public:
return image_index;
}
vk::Image GetImageIndex(std::size_t index) const {
VkImage GetImageIndex(std::size_t index) const {
return images[index];
}
vk::ImageView GetImageViewIndex(std::size_t index) const {
VkImageView GetImageViewIndex(std::size_t index) const {
return *image_views[index];
}
vk::Format GetImageFormat() const {
VkFormat GetImageFormat() const {
return image_format;
}
@ -65,30 +65,30 @@ public:
}
private:
void CreateSwapchain(const vk::SurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
void CreateSwapchain(const VkSurfaceCapabilitiesKHR& capabilities, u32 width, u32 height,
bool srgb);
void CreateSemaphores();
void CreateImageViews();
void Destroy();
const vk::SurfaceKHR surface;
const VkSurfaceKHR surface;
const VKDevice& device;
UniqueSwapchainKHR swapchain;
vk::SwapchainKHR swapchain;
std::size_t image_count{};
std::vector<vk::Image> images;
std::vector<UniqueImageView> image_views;
std::vector<UniqueFramebuffer> framebuffers;
std::vector<VkImage> images;
std::vector<vk::ImageView> image_views;
std::vector<vk::Framebuffer> framebuffers;
std::vector<VKFence*> fences;
std::vector<UniqueSemaphore> present_semaphores;
std::vector<vk::Semaphore> present_semaphores;
u32 image_index{};
u32 frame_index{};
vk::Format image_format{};
vk::Extent2D extent{};
VkFormat image_format{};
VkExtent2D extent{};
u32 current_width{};
u32 current_height{};

356
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View file

@ -17,7 +17,6 @@
#include "core/memory.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/morton.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
@ -25,6 +24,7 @@
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/surface.h"
#include "video_core/textures/convert.h"
@ -39,18 +39,18 @@ using VideoCore::Surface::SurfaceTarget;
namespace {
vk::ImageType SurfaceTargetToImage(SurfaceTarget target) {
VkImageType SurfaceTargetToImage(SurfaceTarget target) {
switch (target) {
case SurfaceTarget::Texture1D:
case SurfaceTarget::Texture1DArray:
return vk::ImageType::e1D;
return VK_IMAGE_TYPE_1D;
case SurfaceTarget::Texture2D:
case SurfaceTarget::Texture2DArray:
case SurfaceTarget::TextureCubemap:
case SurfaceTarget::TextureCubeArray:
return vk::ImageType::e2D;
return VK_IMAGE_TYPE_2D;
case SurfaceTarget::Texture3D:
return vk::ImageType::e3D;
return VK_IMAGE_TYPE_3D;
case SurfaceTarget::TextureBuffer:
UNREACHABLE();
return {};
@ -59,35 +59,35 @@ vk::ImageType SurfaceTargetToImage(SurfaceTarget target) {
return {};
}
vk::ImageAspectFlags PixelFormatToImageAspect(PixelFormat pixel_format) {
VkImageAspectFlags PixelFormatToImageAspect(PixelFormat pixel_format) {
if (pixel_format < PixelFormat::MaxColorFormat) {
return vk::ImageAspectFlagBits::eColor;
return VK_IMAGE_ASPECT_COLOR_BIT;
} else if (pixel_format < PixelFormat::MaxDepthFormat) {
return vk::ImageAspectFlagBits::eDepth;
return VK_IMAGE_ASPECT_DEPTH_BIT;
} else if (pixel_format < PixelFormat::MaxDepthStencilFormat) {
return vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil;
return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
} else {
UNREACHABLE_MSG("Invalid pixel format={}", static_cast<u32>(pixel_format));
return vk::ImageAspectFlagBits::eColor;
UNREACHABLE_MSG("Invalid pixel format={}", static_cast<int>(pixel_format));
return VK_IMAGE_ASPECT_COLOR_BIT;
}
}
vk::ImageViewType GetImageViewType(SurfaceTarget target) {
VkImageViewType GetImageViewType(SurfaceTarget target) {
switch (target) {
case SurfaceTarget::Texture1D:
return vk::ImageViewType::e1D;
return VK_IMAGE_VIEW_TYPE_1D;
case SurfaceTarget::Texture2D:
return vk::ImageViewType::e2D;
return VK_IMAGE_VIEW_TYPE_2D;
case SurfaceTarget::Texture3D:
return vk::ImageViewType::e3D;
return VK_IMAGE_VIEW_TYPE_3D;
case SurfaceTarget::Texture1DArray:
return vk::ImageViewType::e1DArray;
return VK_IMAGE_VIEW_TYPE_1D_ARRAY;
case SurfaceTarget::Texture2DArray:
return vk::ImageViewType::e2DArray;
return VK_IMAGE_VIEW_TYPE_2D_ARRAY;
case SurfaceTarget::TextureCubemap:
return vk::ImageViewType::eCube;
return VK_IMAGE_VIEW_TYPE_CUBE;
case SurfaceTarget::TextureCubeArray:
return vk::ImageViewType::eCubeArray;
return VK_IMAGE_VIEW_TYPE_CUBE_ARRAY;
case SurfaceTarget::TextureBuffer:
break;
}
@ -95,73 +95,88 @@ vk::ImageViewType GetImageViewType(SurfaceTarget target) {
return {};
}
UniqueBuffer CreateBuffer(const VKDevice& device, const SurfaceParams& params,
std::size_t host_memory_size) {
vk::Buffer CreateBuffer(const VKDevice& device, const SurfaceParams& params,
std::size_t host_memory_size) {
// TODO(Rodrigo): Move texture buffer creation to the buffer cache
const vk::BufferCreateInfo buffer_ci({}, host_memory_size,
vk::BufferUsageFlagBits::eUniformTexelBuffer |
vk::BufferUsageFlagBits::eTransferSrc |
vk::BufferUsageFlagBits::eTransferDst,
vk::SharingMode::eExclusive, 0, nullptr);
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
return dev.createBufferUnique(buffer_ci, nullptr, dld);
VkBufferCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.size = static_cast<VkDeviceSize>(host_memory_size);
ci.usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_SRC_BIT |
VK_BUFFER_USAGE_TRANSFER_DST_BIT;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
return device.GetLogical().CreateBuffer(ci);
}
vk::BufferViewCreateInfo GenerateBufferViewCreateInfo(const VKDevice& device,
const SurfaceParams& params,
vk::Buffer buffer,
std::size_t host_memory_size) {
VkBufferViewCreateInfo GenerateBufferViewCreateInfo(const VKDevice& device,
const SurfaceParams& params, VkBuffer buffer,
std::size_t host_memory_size) {
ASSERT(params.IsBuffer());
const auto format =
MaxwellToVK::SurfaceFormat(device, FormatType::Buffer, params.pixel_format).format;
return vk::BufferViewCreateInfo({}, buffer, format, 0, host_memory_size);
VkBufferViewCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.buffer = buffer;
ci.format = MaxwellToVK::SurfaceFormat(device, FormatType::Buffer, params.pixel_format).format;
ci.offset = 0;
ci.range = static_cast<VkDeviceSize>(host_memory_size);
return ci;
}
vk::ImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceParams& params) {
constexpr auto sample_count = vk::SampleCountFlagBits::e1;
constexpr auto tiling = vk::ImageTiling::eOptimal;
VkImageCreateInfo GenerateImageCreateInfo(const VKDevice& device, const SurfaceParams& params) {
ASSERT(!params.IsBuffer());
const auto [format, attachable, storage] =
MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, params.pixel_format);
auto image_usage = vk::ImageUsageFlagBits::eSampled | vk::ImageUsageFlagBits::eTransferDst |
vk::ImageUsageFlagBits::eTransferSrc;
VkImageCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.imageType = SurfaceTargetToImage(params.target);
ci.format = format;
ci.mipLevels = params.num_levels;
ci.arrayLayers = static_cast<u32>(params.GetNumLayers());
ci.samples = VK_SAMPLE_COUNT_1_BIT;
ci.tiling = VK_IMAGE_TILING_OPTIMAL;
ci.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
ci.queueFamilyIndexCount = 0;
ci.pQueueFamilyIndices = nullptr;
ci.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
ci.usage = VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
if (attachable) {
image_usage |= params.IsPixelFormatZeta() ? vk::ImageUsageFlagBits::eDepthStencilAttachment
: vk::ImageUsageFlagBits::eColorAttachment;
ci.usage |= params.IsPixelFormatZeta() ? VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT
: VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
}
if (storage) {
image_usage |= vk::ImageUsageFlagBits::eStorage;
ci.usage |= VK_IMAGE_USAGE_STORAGE_BIT;
}
vk::ImageCreateFlags flags;
vk::Extent3D extent;
switch (params.target) {
case SurfaceTarget::TextureCubemap:
case SurfaceTarget::TextureCubeArray:
flags |= vk::ImageCreateFlagBits::eCubeCompatible;
ci.flags |= VK_IMAGE_CREATE_CUBE_COMPATIBLE_BIT;
[[fallthrough]];
case SurfaceTarget::Texture1D:
case SurfaceTarget::Texture1DArray:
case SurfaceTarget::Texture2D:
case SurfaceTarget::Texture2DArray:
extent = vk::Extent3D(params.width, params.height, 1);
ci.extent = {params.width, params.height, 1};
break;
case SurfaceTarget::Texture3D:
extent = vk::Extent3D(params.width, params.height, params.depth);
ci.extent = {params.width, params.height, params.depth};
break;
case SurfaceTarget::TextureBuffer:
UNREACHABLE();
}
return vk::ImageCreateInfo(flags, SurfaceTargetToImage(params.target), format, extent,
params.num_levels, static_cast<u32>(params.GetNumLayers()),
sample_count, tiling, image_usage, vk::SharingMode::eExclusive, 0,
nullptr, vk::ImageLayout::eUndefined);
return ci;
}
} // Anonymous namespace
@ -175,15 +190,13 @@ CachedSurface::CachedSurface(Core::System& system, const VKDevice& device,
memory_manager{memory_manager}, scheduler{scheduler}, staging_pool{staging_pool} {
if (params.IsBuffer()) {
buffer = CreateBuffer(device, params, host_memory_size);
commit = memory_manager.Commit(*buffer, false);
commit = memory_manager.Commit(buffer, false);
const auto buffer_view_ci =
GenerateBufferViewCreateInfo(device, params, *buffer, host_memory_size);
format = buffer_view_ci.format;
const auto dev = device.GetLogical();
const auto& dld = device.GetDispatchLoader();
buffer_view = dev.createBufferViewUnique(buffer_view_ci, nullptr, dld);
buffer_view = device.GetLogical().CreateBufferView(buffer_view_ci);
} else {
const auto image_ci = GenerateImageCreateInfo(device, params);
format = image_ci.format;
@ -221,16 +234,15 @@ void CachedSurface::DownloadTexture(std::vector<u8>& staging_buffer) {
// We can't copy images to buffers inside a renderpass
scheduler.RequestOutsideRenderPassOperationContext();
FullTransition(vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferRead,
vk::ImageLayout::eTransferSrcOptimal);
FullTransition(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_READ_BIT,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
const auto& buffer = staging_pool.GetUnusedBuffer(host_memory_size, true);
// TODO(Rodrigo): Do this in a single copy
for (u32 level = 0; level < params.num_levels; ++level) {
scheduler.Record([image = image->GetHandle(), buffer = *buffer.handle,
copy = GetBufferImageCopy(level)](auto cmdbuf, auto& dld) {
cmdbuf.copyImageToBuffer(image, vk::ImageLayout::eTransferSrcOptimal, buffer, {copy},
dld);
scheduler.Record([image = *image->GetHandle(), buffer = *buffer.handle,
copy = GetBufferImageCopy(level)](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyImageToBuffer(image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, buffer, copy);
});
}
scheduler.Finish();
@ -257,15 +269,27 @@ void CachedSurface::UploadBuffer(const std::vector<u8>& staging_buffer) {
std::memcpy(src_buffer.commit->Map(host_memory_size), staging_buffer.data(), host_memory_size);
scheduler.Record([src_buffer = *src_buffer.handle, dst_buffer = *buffer,
size = host_memory_size](auto cmdbuf, auto& dld) {
const vk::BufferCopy copy(0, 0, size);
cmdbuf.copyBuffer(src_buffer, dst_buffer, {copy}, dld);
size = host_memory_size](vk::CommandBuffer cmdbuf) {
VkBufferCopy copy;
copy.srcOffset = 0;
copy.dstOffset = 0;
copy.size = size;
cmdbuf.CopyBuffer(src_buffer, dst_buffer, copy);
cmdbuf.pipelineBarrier(
vk::PipelineStageFlagBits::eTransfer, vk::PipelineStageFlagBits::eVertexShader, {}, {},
{vk::BufferMemoryBarrier(vk::AccessFlagBits::eTransferWrite,
vk::AccessFlagBits::eShaderRead, 0, 0, dst_buffer, 0, size)},
{}, dld);
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
barrier.pNext = nullptr;
barrier.srcAccessMask = VK_PIPELINE_STAGE_TRANSFER_BIT;
barrier.dstAccessMask = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT;
barrier.srcQueueFamilyIndex = VK_ACCESS_TRANSFER_WRITE_BIT;
barrier.dstQueueFamilyIndex = VK_ACCESS_SHADER_READ_BIT;
barrier.srcQueueFamilyIndex = 0;
barrier.dstQueueFamilyIndex = 0;
barrier.buffer = dst_buffer;
barrier.offset = 0;
barrier.size = size;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT,
0, {}, barrier, {});
});
}
@ -273,43 +297,49 @@ void CachedSurface::UploadImage(const std::vector<u8>& staging_buffer) {
const auto& src_buffer = staging_pool.GetUnusedBuffer(host_memory_size, true);
std::memcpy(src_buffer.commit->Map(host_memory_size), staging_buffer.data(), host_memory_size);
FullTransition(vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferWrite,
vk::ImageLayout::eTransferDstOptimal);
FullTransition(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
for (u32 level = 0; level < params.num_levels; ++level) {
vk::BufferImageCopy copy = GetBufferImageCopy(level);
if (image->GetAspectMask() ==
(vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil)) {
vk::BufferImageCopy depth = copy;
vk::BufferImageCopy stencil = copy;
depth.imageSubresource.aspectMask = vk::ImageAspectFlagBits::eDepth;
stencil.imageSubresource.aspectMask = vk::ImageAspectFlagBits::eStencil;
scheduler.Record([buffer = *src_buffer.handle, image = image->GetHandle(), depth,
stencil](auto cmdbuf, auto& dld) {
cmdbuf.copyBufferToImage(buffer, image, vk::ImageLayout::eTransferDstOptimal,
{depth, stencil}, dld);
const VkBufferImageCopy copy = GetBufferImageCopy(level);
if (image->GetAspectMask() == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
scheduler.Record([buffer = *src_buffer.handle, image = *image->GetHandle(),
copy](vk::CommandBuffer cmdbuf) {
std::array<VkBufferImageCopy, 2> copies = {copy, copy};
copies[0].imageSubresource.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
copies[1].imageSubresource.aspectMask = VK_IMAGE_ASPECT_STENCIL_BIT;
cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
copies);
});
} else {
scheduler.Record([buffer = *src_buffer.handle, image = image->GetHandle(),
copy](auto cmdbuf, auto& dld) {
cmdbuf.copyBufferToImage(buffer, image, vk::ImageLayout::eTransferDstOptimal,
{copy}, dld);
scheduler.Record([buffer = *src_buffer.handle, image = *image->GetHandle(),
copy](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBufferToImage(buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
});
}
}
}
vk::BufferImageCopy CachedSurface::GetBufferImageCopy(u32 level) const {
const u32 vk_depth = params.target == SurfaceTarget::Texture3D ? params.GetMipDepth(level) : 1;
const std::size_t mip_offset = params.GetHostMipmapLevelOffset(level, is_converted);
return vk::BufferImageCopy(
mip_offset, 0, 0,
{image->GetAspectMask(), level, 0, static_cast<u32>(params.GetNumLayers())}, {0, 0, 0},
{params.GetMipWidth(level), params.GetMipHeight(level), vk_depth});
VkBufferImageCopy CachedSurface::GetBufferImageCopy(u32 level) const {
VkBufferImageCopy copy;
copy.bufferOffset = params.GetHostMipmapLevelOffset(level, is_converted);
copy.bufferRowLength = 0;
copy.bufferImageHeight = 0;
copy.imageSubresource.aspectMask = image->GetAspectMask();
copy.imageSubresource.mipLevel = level;
copy.imageSubresource.baseArrayLayer = 0;
copy.imageSubresource.layerCount = static_cast<u32>(params.GetNumLayers());
copy.imageOffset.x = 0;
copy.imageOffset.y = 0;
copy.imageOffset.z = 0;
copy.imageExtent.width = params.GetMipWidth(level);
copy.imageExtent.height = params.GetMipHeight(level);
copy.imageExtent.depth =
params.target == SurfaceTarget::Texture3D ? params.GetMipDepth(level) : 1;
return copy;
}
vk::ImageSubresourceRange CachedSurface::GetImageSubresourceRange() const {
VkImageSubresourceRange CachedSurface::GetImageSubresourceRange() const {
return {image->GetAspectMask(), 0, params.num_levels, 0,
static_cast<u32>(params.GetNumLayers())};
}
@ -321,12 +351,12 @@ CachedSurfaceView::CachedSurfaceView(const VKDevice& device, CachedSurface& surf
aspect_mask{surface.GetAspectMask()}, device{device}, surface{surface},
base_layer{params.base_layer}, num_layers{params.num_layers}, base_level{params.base_level},
num_levels{params.num_levels}, image_view_type{image ? GetImageViewType(params.target)
: vk::ImageViewType{}} {}
: VK_IMAGE_VIEW_TYPE_1D} {}
CachedSurfaceView::~CachedSurfaceView() = default;
vk::ImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y_source,
SwizzleSource z_source, SwizzleSource w_source) {
VkImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource y_source,
SwizzleSource z_source, SwizzleSource w_source) {
const u32 swizzle = EncodeSwizzle(x_source, y_source, z_source, w_source);
if (last_image_view && last_swizzle == swizzle) {
return last_image_view;
@ -351,37 +381,45 @@ vk::ImageView CachedSurfaceView::GetHandle(SwizzleSource x_source, SwizzleSource
// Games can sample depth or stencil values on textures. This is decided by the swizzle value on
// hardware. To emulate this on Vulkan we specify it in the aspect.
vk::ImageAspectFlags aspect = aspect_mask;
if (aspect == (vk::ImageAspectFlagBits::eDepth | vk::ImageAspectFlagBits::eStencil)) {
VkImageAspectFlags aspect = aspect_mask;
if (aspect == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
UNIMPLEMENTED_IF(x_source != SwizzleSource::R && x_source != SwizzleSource::G);
const bool is_first = x_source == SwizzleSource::R;
switch (params.pixel_format) {
case VideoCore::Surface::PixelFormat::Z24S8:
case VideoCore::Surface::PixelFormat::Z32FS8:
aspect = is_first ? vk::ImageAspectFlagBits::eDepth : vk::ImageAspectFlagBits::eStencil;
aspect = is_first ? VK_IMAGE_ASPECT_DEPTH_BIT : VK_IMAGE_ASPECT_STENCIL_BIT;
break;
case VideoCore::Surface::PixelFormat::S8Z24:
aspect = is_first ? vk::ImageAspectFlagBits::eStencil : vk::ImageAspectFlagBits::eDepth;
aspect = is_first ? VK_IMAGE_ASPECT_STENCIL_BIT : VK_IMAGE_ASPECT_DEPTH_BIT;
break;
default:
aspect = vk::ImageAspectFlagBits::eDepth;
aspect = VK_IMAGE_ASPECT_DEPTH_BIT;
UNIMPLEMENTED();
}
// Vulkan doesn't seem to understand swizzling of a depth stencil image, use identity
swizzle_x = vk::ComponentSwizzle::eR;
swizzle_y = vk::ComponentSwizzle::eG;
swizzle_z = vk::ComponentSwizzle::eB;
swizzle_w = vk::ComponentSwizzle::eA;
swizzle_x = VK_COMPONENT_SWIZZLE_R;
swizzle_y = VK_COMPONENT_SWIZZLE_G;
swizzle_z = VK_COMPONENT_SWIZZLE_B;
swizzle_w = VK_COMPONENT_SWIZZLE_A;
}
const vk::ImageViewCreateInfo image_view_ci(
{}, surface.GetImageHandle(), image_view_type, surface.GetImage().GetFormat(),
{swizzle_x, swizzle_y, swizzle_z, swizzle_w},
{aspect, base_level, num_levels, base_layer, num_layers});
VkImageViewCreateInfo ci;
ci.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
ci.pNext = nullptr;
ci.flags = 0;
ci.image = surface.GetImageHandle();
ci.viewType = image_view_type;
ci.format = surface.GetImage().GetFormat();
ci.components = {swizzle_x, swizzle_y, swizzle_z, swizzle_w};
ci.subresourceRange.aspectMask = aspect;
ci.subresourceRange.baseMipLevel = base_level;
ci.subresourceRange.levelCount = num_levels;
ci.subresourceRange.baseArrayLayer = base_layer;
ci.subresourceRange.layerCount = num_layers;
image_view = device.GetLogical().CreateImageView(ci);
const auto dev = device.GetLogical();
image_view = dev.createImageViewUnique(image_view_ci, nullptr, device.GetDispatchLoader());
return last_image_view = *image_view;
}
@ -418,25 +456,36 @@ void VKTextureCache::ImageCopy(Surface& src_surface, Surface& dst_surface,
scheduler.RequestOutsideRenderPassOperationContext();
src_surface->Transition(copy_params.source_z, copy_params.depth, copy_params.source_level, 1,
vk::PipelineStageFlagBits::eTransfer, vk::AccessFlagBits::eTransferRead,
vk::ImageLayout::eTransferSrcOptimal);
dst_surface->Transition(
dst_base_layer, num_layers, copy_params.dest_level, 1, vk::PipelineStageFlagBits::eTransfer,
vk::AccessFlagBits::eTransferWrite, vk::ImageLayout::eTransferDstOptimal);
VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_READ_BIT,
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL);
dst_surface->Transition(dst_base_layer, num_layers, copy_params.dest_level, 1,
VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_WRITE_BIT,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
const vk::ImageSubresourceLayers src_subresource(
src_surface->GetAspectMask(), copy_params.source_level, copy_params.source_z, num_layers);
const vk::ImageSubresourceLayers dst_subresource(
dst_surface->GetAspectMask(), copy_params.dest_level, dst_base_layer, num_layers);
const vk::Offset3D src_offset(copy_params.source_x, copy_params.source_y, 0);
const vk::Offset3D dst_offset(copy_params.dest_x, copy_params.dest_y, dst_offset_z);
const vk::Extent3D extent(copy_params.width, copy_params.height, extent_z);
const vk::ImageCopy copy(src_subresource, src_offset, dst_subresource, dst_offset, extent);
const vk::Image src_image = src_surface->GetImageHandle();
const vk::Image dst_image = dst_surface->GetImageHandle();
scheduler.Record([src_image, dst_image, copy](auto cmdbuf, auto& dld) {
cmdbuf.copyImage(src_image, vk::ImageLayout::eTransferSrcOptimal, dst_image,
vk::ImageLayout::eTransferDstOptimal, {copy}, dld);
VkImageCopy copy;
copy.srcSubresource.aspectMask = src_surface->GetAspectMask();
copy.srcSubresource.mipLevel = copy_params.source_level;
copy.srcSubresource.baseArrayLayer = copy_params.source_z;
copy.srcSubresource.layerCount = num_layers;
copy.srcOffset.x = copy_params.source_x;
copy.srcOffset.y = copy_params.source_y;
copy.srcOffset.z = 0;
copy.dstSubresource.aspectMask = dst_surface->GetAspectMask();
copy.dstSubresource.mipLevel = copy_params.dest_level;
copy.dstSubresource.baseArrayLayer = dst_base_layer;
copy.dstSubresource.layerCount = num_layers;
copy.dstOffset.x = copy_params.dest_x;
copy.dstOffset.y = copy_params.dest_y;
copy.dstOffset.z = dst_offset_z;
copy.extent.width = copy_params.width;
copy.extent.height = copy_params.height;
copy.extent.depth = extent_z;
const VkImage src_image = src_surface->GetImageHandle();
const VkImage dst_image = dst_surface->GetImageHandle();
scheduler.Record([src_image, dst_image, copy](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, copy);
});
}
@ -445,25 +494,34 @@ void VKTextureCache::ImageBlit(View& src_view, View& dst_view,
// We can't blit inside a renderpass
scheduler.RequestOutsideRenderPassOperationContext();
src_view->Transition(vk::ImageLayout::eTransferSrcOptimal, vk::PipelineStageFlagBits::eTransfer,
vk::AccessFlagBits::eTransferRead);
dst_view->Transition(vk::ImageLayout::eTransferDstOptimal, vk::PipelineStageFlagBits::eTransfer,
vk::AccessFlagBits::eTransferWrite);
src_view->Transition(VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_READ_BIT);
dst_view->Transition(VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_ACCESS_TRANSFER_WRITE_BIT);
VkImageBlit blit;
blit.srcSubresource = src_view->GetImageSubresourceLayers();
blit.srcOffsets[0].x = copy_config.src_rect.left;
blit.srcOffsets[0].y = copy_config.src_rect.top;
blit.srcOffsets[0].z = 0;
blit.srcOffsets[1].x = copy_config.src_rect.right;
blit.srcOffsets[1].y = copy_config.src_rect.bottom;
blit.srcOffsets[1].z = 1;
blit.dstSubresource = dst_view->GetImageSubresourceLayers();
blit.dstOffsets[0].x = copy_config.dst_rect.left;
blit.dstOffsets[0].y = copy_config.dst_rect.top;
blit.dstOffsets[0].z = 0;
blit.dstOffsets[1].x = copy_config.dst_rect.right;
blit.dstOffsets[1].y = copy_config.dst_rect.bottom;
blit.dstOffsets[1].z = 1;
const auto& cfg = copy_config;
const auto src_top_left = vk::Offset3D(cfg.src_rect.left, cfg.src_rect.top, 0);
const auto src_bot_right = vk::Offset3D(cfg.src_rect.right, cfg.src_rect.bottom, 1);
const auto dst_top_left = vk::Offset3D(cfg.dst_rect.left, cfg.dst_rect.top, 0);
const auto dst_bot_right = vk::Offset3D(cfg.dst_rect.right, cfg.dst_rect.bottom, 1);
const vk::ImageBlit blit(src_view->GetImageSubresourceLayers(), {src_top_left, src_bot_right},
dst_view->GetImageSubresourceLayers(), {dst_top_left, dst_bot_right});
const bool is_linear = copy_config.filter == Tegra::Engines::Fermi2D::Filter::Linear;
scheduler.Record([src_image = src_view->GetImage(), dst_image = dst_view->GetImage(), blit,
is_linear](auto cmdbuf, auto& dld) {
cmdbuf.blitImage(src_image, vk::ImageLayout::eTransferSrcOptimal, dst_image,
vk::ImageLayout::eTransferDstOptimal, {blit},
is_linear ? vk::Filter::eLinear : vk::Filter::eNearest, dld);
is_linear](vk::CommandBuffer cmdbuf) {
cmdbuf.BlitImage(src_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, dst_image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, blit,
is_linear ? VK_FILTER_LINEAR : VK_FILTER_NEAREST);
});
}

64
src/video_core/renderer_vulkan/vk_texture_cache.h
View file

@ -13,10 +13,10 @@
#include "common/math_util.h"
#include "video_core/gpu.h"
#include "video_core/rasterizer_cache.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_image.h"
#include "video_core/renderer_vulkan/vk_memory_manager.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/wrapper.h"
#include "video_core/texture_cache/surface_base.h"
#include "video_core/texture_cache/texture_cache.h"
#include "video_core/textures/decoders.h"
@ -60,15 +60,15 @@ public:
void UploadTexture(const std::vector<u8>& staging_buffer) override;
void DownloadTexture(std::vector<u8>& staging_buffer) override;
void FullTransition(vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access,
vk::ImageLayout new_layout) {
void FullTransition(VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
VkImageLayout new_layout) {
image->Transition(0, static_cast<u32>(params.GetNumLayers()), 0, params.num_levels,
new_stage_mask, new_access, new_layout);
}
void Transition(u32 base_layer, u32 num_layers, u32 base_level, u32 num_levels,
vk::PipelineStageFlags new_stage_mask, vk::AccessFlags new_access,
vk::ImageLayout new_layout) {
VkPipelineStageFlags new_stage_mask, VkAccessFlags new_access,
VkImageLayout new_layout) {
image->Transition(base_layer, num_layers, base_level, num_levels, new_stage_mask,
new_access, new_layout);
}
@ -81,15 +81,15 @@ public:
return *image;
}
vk::Image GetImageHandle() const {
return image->GetHandle();
VkImage GetImageHandle() const {
return *image->GetHandle();
}
vk::ImageAspectFlags GetAspectMask() const {
VkImageAspectFlags GetAspectMask() const {
return image->GetAspectMask();
}
vk::BufferView GetBufferViewHandle() const {
VkBufferView GetBufferViewHandle() const {
return *buffer_view;
}
@ -104,9 +104,9 @@ private:
void UploadImage(const std::vector<u8>& staging_buffer);
vk::BufferImageCopy GetBufferImageCopy(u32 level) const;
VkBufferImageCopy GetBufferImageCopy(u32 level) const;
vk::ImageSubresourceRange GetImageSubresourceRange() const;
VkImageSubresourceRange GetImageSubresourceRange() const;
Core::System& system;
const VKDevice& device;
@ -116,11 +116,11 @@ private:
VKStagingBufferPool& staging_pool;
std::optional<VKImage> image;
UniqueBuffer buffer;
UniqueBufferView buffer_view;
vk::Buffer buffer;
vk::BufferView buffer_view;
VKMemoryCommit commit;
vk::Format format;
VkFormat format = VK_FORMAT_UNDEFINED;
};
class CachedSurfaceView final : public VideoCommon::ViewBase {
@ -129,16 +129,16 @@ public:
const ViewParams& params, bool is_proxy);
~CachedSurfaceView();
vk::ImageView GetHandle(Tegra::Texture::SwizzleSource x_source,
Tegra::Texture::SwizzleSource y_source,
Tegra::Texture::SwizzleSource z_source,
Tegra::Texture::SwizzleSource w_source);
VkImageView GetHandle(Tegra::Texture::SwizzleSource x_source,
Tegra::Texture::SwizzleSource y_source,
Tegra::Texture::SwizzleSource z_source,
Tegra::Texture::SwizzleSource w_source);
bool IsSameSurface(const CachedSurfaceView& rhs) const {
return &surface == &rhs.surface;
}
vk::ImageView GetHandle() {
VkImageView GetHandle() {
return GetHandle(Tegra::Texture::SwizzleSource::R, Tegra::Texture::SwizzleSource::G,
Tegra::Texture::SwizzleSource::B, Tegra::Texture::SwizzleSource::A);
}
@ -159,24 +159,24 @@ public:
return buffer_view;
}
vk::Image GetImage() const {
VkImage GetImage() const {
return image;
}
vk::BufferView GetBufferView() const {
VkBufferView GetBufferView() const {
return buffer_view;
}
vk::ImageSubresourceRange GetImageSubresourceRange() const {
VkImageSubresourceRange GetImageSubresourceRange() const {
return {aspect_mask, base_level, num_levels, base_layer, num_layers};
}
vk::ImageSubresourceLayers GetImageSubresourceLayers() const {
VkImageSubresourceLayers GetImageSubresourceLayers() const {
return {surface.GetAspectMask(), base_level, base_layer, num_layers};
}
void Transition(vk::ImageLayout new_layout, vk::PipelineStageFlags new_stage_mask,
vk::AccessFlags new_access) const {
void Transition(VkImageLayout new_layout, VkPipelineStageFlags new_stage_mask,
VkAccessFlags new_access) const {
surface.Transition(base_layer, num_layers, base_level, num_levels, new_stage_mask,
new_access, new_layout);
}
@ -196,9 +196,9 @@ private:
// Store a copy of these values to avoid double dereference when reading them
const SurfaceParams params;
const vk::Image image;
const vk::BufferView buffer_view;
const vk::ImageAspectFlags aspect_mask;
const VkImage image;
const VkBufferView buffer_view;
const VkImageAspectFlags aspect_mask;
const VKDevice& device;
CachedSurface& surface;
@ -206,12 +206,12 @@ private:
const u32 num_layers;
const u32 base_level;
const u32 num_levels;
const vk::ImageViewType image_view_type;
const VkImageViewType image_view_type;
vk::ImageView last_image_view;
u32 last_swizzle{};
VkImageView last_image_view = nullptr;
u32 last_swizzle = 0;
std::unordered_map<u32, UniqueImageView> view_cache;
std::unordered_map<u32, vk::ImageView> view_cache;
};
class VKTextureCache final : public TextureCacheBase {

18
src/video_core/renderer_vulkan/vk_update_descriptor.cpp
View file

@ -7,10 +7,10 @@
#include "common/assert.h"
#include "common/logging/log.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/vk_device.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -27,8 +27,8 @@ void VKUpdateDescriptorQueue::Acquire() {
entries.clear();
}
void VKUpdateDescriptorQueue::Send(vk::DescriptorUpdateTemplate update_template,
vk::DescriptorSet set) {
void VKUpdateDescriptorQueue::Send(VkDescriptorUpdateTemplateKHR update_template,
VkDescriptorSet set) {
if (payload.size() + entries.size() >= payload.max_size()) {
LOG_WARNING(Render_Vulkan, "Payload overflow, waiting for worker thread");
scheduler.WaitWorker();
@ -37,21 +37,21 @@ void VKUpdateDescriptorQueue::Send(vk::DescriptorUpdateTemplate update_template,
const auto payload_start = payload.data() + payload.size();
for (const auto& entry : entries) {
if (const auto image = std::get_if<vk::DescriptorImageInfo>(&entry)) {
if (const auto image = std::get_if<VkDescriptorImageInfo>(&entry)) {
payload.push_back(*image);
} else if (const auto buffer = std::get_if<Buffer>(&entry)) {
payload.emplace_back(*buffer->buffer, buffer->offset, buffer->size);
} else if (const auto texel = std::get_if<vk::BufferView>(&entry)) {
} else if (const auto texel = std::get_if<VkBufferView>(&entry)) {
payload.push_back(*texel);
} else {
UNREACHABLE();
}
}
scheduler.Record([dev = device.GetLogical(), payload_start, set,
update_template]([[maybe_unused]] auto cmdbuf, auto& dld) {
dev.updateDescriptorSetWithTemplate(set, update_template, payload_start, dld);
});
scheduler.Record(
[payload_start, set, update_template, logical = &device.GetLogical()](vk::CommandBuffer) {
logical->UpdateDescriptorSet(set, update_template, payload_start);
});
}
} // namespace Vulkan

42
src/video_core/renderer_vulkan/vk_update_descriptor.h
View file

@ -9,7 +9,7 @@
#include <boost/container/static_vector.hpp>
#include "common/common_types.h"
#include "video_core/renderer_vulkan/declarations.h"
#include "video_core/renderer_vulkan/wrapper.h"
namespace Vulkan {
@ -20,18 +20,18 @@ class DescriptorUpdateEntry {
public:
explicit DescriptorUpdateEntry() : image{} {}
DescriptorUpdateEntry(vk::DescriptorImageInfo image) : image{image} {}
DescriptorUpdateEntry(VkDescriptorImageInfo image) : image{image} {}
DescriptorUpdateEntry(vk::Buffer buffer, vk::DeviceSize offset, vk::DeviceSize size)
DescriptorUpdateEntry(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size)
: buffer{buffer, offset, size} {}
DescriptorUpdateEntry(vk::BufferView texel_buffer) : texel_buffer{texel_buffer} {}
DescriptorUpdateEntry(VkBufferView texel_buffer) : texel_buffer{texel_buffer} {}
private:
union {
vk::DescriptorImageInfo image;
vk::DescriptorBufferInfo buffer;
vk::BufferView texel_buffer;
VkDescriptorImageInfo image;
VkDescriptorBufferInfo buffer;
VkBufferView texel_buffer;
};
};
@ -44,37 +44,35 @@ public:
void Acquire();
void Send(vk::DescriptorUpdateTemplate update_template, vk::DescriptorSet set);
void Send(VkDescriptorUpdateTemplateKHR update_template, VkDescriptorSet set);
void AddSampledImage(vk::Sampler sampler, vk::ImageView image_view) {
entries.emplace_back(vk::DescriptorImageInfo{sampler, image_view, {}});
void AddSampledImage(VkSampler sampler, VkImageView image_view) {
entries.emplace_back(VkDescriptorImageInfo{sampler, image_view, {}});
}
void AddImage(vk::ImageView image_view) {
entries.emplace_back(vk::DescriptorImageInfo{{}, image_view, {}});
void AddImage(VkImageView image_view) {
entries.emplace_back(VkDescriptorImageInfo{{}, image_view, {}});
}
void AddBuffer(const vk::Buffer* buffer, u64 offset, std::size_t size) {
void AddBuffer(const VkBuffer* buffer, u64 offset, std::size_t size) {
entries.push_back(Buffer{buffer, offset, size});
}
void AddTexelBuffer(vk::BufferView texel_buffer) {
void AddTexelBuffer(VkBufferView texel_buffer) {
entries.emplace_back(texel_buffer);
}
vk::ImageLayout* GetLastImageLayout() {
return &std::get<vk::DescriptorImageInfo>(entries.back()).imageLayout;
VkImageLayout* GetLastImageLayout() {
return &std::get<VkDescriptorImageInfo>(entries.back()).imageLayout;
}
private:
struct Buffer {
const vk::Buffer* buffer{};
u64 offset{};
std::size_t size{};
const VkBuffer* buffer = nullptr;
u64 offset = 0;
std::size_t size = 0;
};
using Variant = std::variant<vk::DescriptorImageInfo, Buffer, vk::BufferView>;
// Old gcc versions don't consider this trivially copyable.
// static_assert(std::is_trivially_copyable_v<Variant>);
using Variant = std::variant<VkDescriptorImageInfo, Buffer, VkBufferView>;
const VKDevice& device;
VKScheduler& scheduler;