Exverge/suyu
3
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

VideoCore: Initial Setup for the Resolution Scaler.

Browse source
This commit is contained in:
Fernando Sahmkow 2021-07-18 18:40:14 +02:00
parent 720970c4c1
commit 22f4b290b6
11 changed files with 255 additions and 18 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
src/common/settings.cpp
View file

@ -47,6 +47,7 @@ void LogSettings() {
log_setting("System_TimeZoneIndex", values.time_zone_index.GetValue());
log_setting("Core_UseMultiCore", values.use_multi_core.GetValue());
log_setting("CPU_Accuracy", values.cpu_accuracy.GetValue());
log_setting("Renderer_UseResolutionScaling", values.resolution_setup.GetValue());
log_setting("Renderer_UseResolutionFactor", values.resolution_factor.GetValue());
log_setting("Renderer_UseSpeedLimit", values.use_speed_limit.GetValue());
log_setting("Renderer_SpeedLimit", values.speed_limit.GetValue());

18
src/common/settings.h
View file

@ -52,6 +52,22 @@ enum class NvdecEmulation : u32 {
GPU = 2,
};
enum class ResolutionSetup : u32 {
Res1_2X = 0,
Res3_4X = 1,
Res1X = 2,
Res3_2K = 3,
Res2X = 4,
Res3X = 5,
};
struct ResolutionScalingInfo {
u32 up_scale{2};
u32 down_shift{0};
f32 up_factor{2.0f};
f32 down_factor{0.5f};
};
/** The BasicSetting class is a simple resource manager. It defines a label and default value
* alongside the actual value of the setting for simpler and less-error prone use with frontend
* configurations. Setting a default value and label is required, though subclasses may deviate from
@ -451,6 +467,8 @@ struct Values {
"disable_shader_loop_safety_checks"};
Setting<int> vulkan_device{0, "vulkan_device"};
ResolutionScalingInfo resolution_info{};
Setting<ResolutionSetup> resolution_setup{ResolutionSetup::Res1X, "resolution_setup"};
Setting<u16> resolution_factor{1, "resolution_factor"};
// *nix platforms may have issues with the borderless windowed fullscreen mode.
// Default to exclusive fullscreen on these platforms for now.

1
src/video_core/dirty_flags.h
View file

@ -29,6 +29,7 @@ enum : u8 {
ColorBuffer6,
ColorBuffer7,
ZetaBuffer,
Rescale,
VertexBuffers,
VertexBuffer0,

16
src/video_core/renderer_opengl/gl_texture_cache.cpp
View file

@ -849,6 +849,22 @@ void Image::CopyImageToBuffer(const VideoCommon::BufferImageCopy& copy, size_t b
}
}
void Image::ScaleUp() {
if (True(flags & ImageFlagBits::Rescaled)) {
return;
}
flags |= ImageFlagBits::Rescaled;
UNIMPLEMENTED();
}
void Image::ScaleDown() {
if (False(flags & ImageFlagBits::Rescaled)) {
return;
}
flags &= ~ImageFlagBits::Rescaled;
UNIMPLEMENTED();
}
ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info,
ImageId image_id_, Image& image)
: VideoCommon::ImageViewBase{info, image.info, image_id_}, views{runtime.null_image_views} {

8
src/video_core/renderer_opengl/gl_texture_cache.h
View file

@ -72,6 +72,8 @@ public:
StateTracker& state_tracker);
~TextureCacheRuntime();
void Init() {}
void Finish();
ImageBufferMap UploadStagingBuffer(size_t size);
@ -110,6 +112,8 @@ public:
bool HasNativeASTC() const noexcept;
void TickFrame() {}
private:
struct StagingBuffers {
explicit StagingBuffers(GLenum storage_flags_, GLenum map_flags_);
@ -185,6 +189,10 @@ public:
return gl_type;
}
bool ScaleUp();
bool ScaleDown();
private:
void CopyBufferToImage(const VideoCommon::BufferImageCopy& copy, size_t buffer_offset);

81
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View file

@ -32,6 +32,7 @@ using Tegra::Engines::Fermi2D;
using Tegra::Texture::SwizzleSource;
using Tegra::Texture::TextureMipmapFilter;
using VideoCommon::BufferImageCopy;
using VideoCommon::ImageFlagBits;
using VideoCommon::ImageInfo;
using VideoCommon::ImageType;
using VideoCommon::SubresourceRange;
@ -123,7 +124,8 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
}
}
[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info) {
[[nodiscard]] VkImageCreateInfo MakeImageCreateInfo(const Device& device, const ImageInfo& info,
u32 up, u32 down) {
const PixelFormat format = StorageFormat(info.format);
const auto format_info = MaxwellToVK::SurfaceFormat(device, FormatType::Optimal, false, format);
VkImageCreateFlags flags = VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT;
@ -142,9 +144,9 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
.imageType = ConvertImageType(info.type),
.format = format_info.format,
.extent{
.width = info.size.width >> samples_x,
.height = info.size.height >> samples_y,
.depth = info.size.depth,
.width = ((info.size.width << up) >> down) >> samples_x,
.height = ((info.size.height << up) >> down) >> samples_y,
.depth = (info.size.depth << up) >> down,
},
.mipLevels = static_cast<u32>(info.resources.levels),
.arrayLayers = static_cast<u32>(info.resources.layers),
@ -158,11 +160,12 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
};
}
[[nodiscard]] vk::Image MakeImage(const Device& device, const ImageInfo& info) {
[[nodiscard]] vk::Image MakeImage(const Device& device, const ImageInfo& info, u32 up = 0,
u32 down = 0) {
if (info.type == ImageType::Buffer) {
return vk::Image{};
}
return device.GetLogical().CreateImage(MakeImageCreateInfo(device, info));
return device.GetLogical().CreateImage(MakeImageCreateInfo(device, info, up, down));
}
[[nodiscard]] VkImageAspectFlags ImageAspectMask(PixelFormat format) {
@ -590,6 +593,11 @@ struct RangedBarrierRange {
}
} // Anonymous namespace
void TextureCacheRuntime::Init() {
resolution = Settings::values.resolution_info;
is_rescaling_on = resolution.up_scale != 1 || resolution.down_shift != 0;
}
void TextureCacheRuntime::Finish() {
scheduler.Finish();
}
@ -840,20 +848,26 @@ u64 TextureCacheRuntime::GetDeviceLocalMemory() const {
return device.GetDeviceLocalMemory();
}
Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_addr_,
void TextureCacheRuntime::TickFrame() {
prescaled_images.Tick();
prescaled_commits.Tick();
prescaled_views.Tick();
}
Image::Image(TextureCacheRuntime& runtime_, const ImageInfo& info_, GPUVAddr gpu_addr_,
VAddr cpu_addr_)
: VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime.scheduler},
image(MakeImage(runtime.device, info)),
commit(runtime.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)),
aspect_mask(ImageAspectMask(info.format)) {
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
: VideoCommon::ImageBase(info_, gpu_addr_, cpu_addr_), scheduler{&runtime_.scheduler},
image(MakeImage(runtime_.device, info)),
commit(runtime_.memory_allocator.Commit(image, MemoryUsage::DeviceLocal)),
aspect_mask(ImageAspectMask(info.format)), runtime{&runtime_} {
if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
if (Settings::values.accelerate_astc.GetValue()) {
flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
} else {
flags |= VideoCommon::ImageFlagBits::Converted;
}
}
if (runtime.device.HasDebuggingToolAttached()) {
if (runtime->device.HasDebuggingToolAttached()) {
image.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
@ -861,8 +875,8 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
.pNext = nullptr,
.usage = VK_IMAGE_USAGE_STORAGE_BIT,
};
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
const auto& device = runtime.device.GetLogical();
if (IsPixelFormatASTC(info.format) && !runtime->device.IsOptimalAstcSupported()) {
const auto& device = runtime->device.GetLogical();
storage_image_views.reserve(info.resources.levels);
for (s32 level = 0; level < info.resources.levels; ++level) {
storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{
@ -907,6 +921,10 @@ void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImag
}
void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
const bool is_rescaled = True(flags & ImageFlagBits::Rescaled);
if (is_rescaled) {
ScaleDown();
}
std::vector vk_copies = TransformBufferImageCopies(copies, map.offset, aspect_mask);
scheduler->RequestOutsideRenderPassOperationContext();
scheduler->Record([buffer = map.buffer, image = *image, aspect_mask = aspect_mask,
@ -959,6 +977,39 @@ void Image::DownloadMemory(const StagingBufferRef& map, std::span<const BufferIm
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
0, memory_write_barrier, nullptr, image_write_barrier);
});
if (is_rescaled) {
ScaleUp();
}
}
void Image::ScaleUp() {
if (True(flags & ImageFlagBits::Rescaled)) {
return;
}
ASSERT(info.type != ImageType::Linear);
if (!runtime->is_rescaling_on) {
flags |= ImageFlagBits::Rescaled;
return;
}
flags |= ImageFlagBits::Rescaled;
scaling_count++;
ASSERT(scaling_count < 10);
return;
}
void Image::ScaleDown() {
if (False(flags & ImageFlagBits::Rescaled)) {
return;
}
ASSERT(info.type != ImageType::Linear);
if (!runtime->is_rescaling_on) {
flags &= ~ImageFlagBits::Rescaled;
return;
}
flags &= ~ImageFlagBits::Rescaled;
scaling_count++;
ASSERT(scaling_count < 10);
return;
}
ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewInfo& info,

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

@ -6,7 +6,9 @@
#include <span>
#include "common/settings.h"
#include "shader_recompiler/shader_info.h"
#include "video_core/delayed_destruction_ring.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/texture_cache/image_view_base.h"
#include "video_core/texture_cache/texture_cache_base.h"
@ -15,6 +17,7 @@
namespace Vulkan {
using VideoCommon::DelayedDestructionRing;
using VideoCommon::ImageId;
using VideoCommon::NUM_RT;
using VideoCommon::Region2D;
@ -39,6 +42,14 @@ struct TextureCacheRuntime {
BlitImageHelper& blit_image_helper;
ASTCDecoderPass& astc_decoder_pass;
RenderPassCache& render_pass_cache;
static constexpr size_t TICKS_TO_DESTROY = 6;
DelayedDestructionRing<vk::Image, TICKS_TO_DESTROY> prescaled_images;
DelayedDestructionRing<MemoryCommit, TICKS_TO_DESTROY> prescaled_commits;
DelayedDestructionRing<vk::ImageView, TICKS_TO_DESTROY> prescaled_views;
Settings::ResolutionScalingInfo resolution;
bool is_rescaling_on{};
void Init();
void Finish();
@ -74,6 +85,8 @@ struct TextureCacheRuntime {
return true;
}
void TickFrame();
u64 GetDeviceLocalMemory() const;
};
@ -113,6 +126,10 @@ public:
return std::exchange(initialized, true);
}
void ScaleUp();
void ScaleDown();
private:
VKScheduler* scheduler;
vk::Image image;
@ -121,6 +138,8 @@ private:
std::vector<vk::ImageView> storage_image_views;
VkImageAspectFlags aspect_mask = 0;
bool initialized = false;
TextureCacheRuntime* runtime;
u32 scaling_count{};
};
class ImageView : public VideoCommon::ImageViewBase {

4
src/video_core/texture_cache/image_base.h
View file

@ -33,6 +33,10 @@ enum class ImageFlagBits : u32 {
///< garbage collection priority
Alias = 1 << 11, ///< This image has aliases and has priority on garbage
///< collection
// Rescaler
Rescaled = 1 << 12,
RescaleChecked = 1 << 13,
};
DECLARE_ENUM_FLAG_OPERATORS(ImageFlagBits)

2
src/video_core/texture_cache/image_info.h
View file

@ -15,7 +15,7 @@ using Tegra::Texture::TICEntry;
using VideoCore::Surface::PixelFormat;
struct ImageInfo {
explicit ImageInfo() = default;
ImageInfo() = default;
explicit ImageInfo(const TICEntry& config) noexcept;
explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs, size_t index) noexcept;
explicit ImageInfo(const Tegra::Engines::Maxwell3D::Regs& regs) noexcept;

113
src/video_core/texture_cache/texture_cache.h
View file

@ -35,6 +35,7 @@ TextureCache<P>::TextureCache(Runtime& runtime_, VideoCore::RasterizerInterface&
Tegra::MemoryManager& gpu_memory_)
: runtime{runtime_}, rasterizer{rasterizer_}, maxwell3d{maxwell3d_},
kepler_compute{kepler_compute_}, gpu_memory{gpu_memory_} {
runtime.Init();
// Configure null sampler
TSCEntry sampler_descriptor{};
sampler_descriptor.min_filter.Assign(Tegra::Texture::TextureFilter::Linear);
@ -103,6 +104,7 @@ void TextureCache<P>::TickFrame() {
sentenced_images.Tick();
sentenced_framebuffers.Tick();
sentenced_image_view.Tick();
runtime.TickFrame();
++frame_tick;
}
@ -208,18 +210,63 @@ void TextureCache<P>::UpdateRenderTargets(bool is_clear) {
const bool force = flags[Dirty::RenderTargetControl];
flags[Dirty::RenderTargetControl] = false;
bool can_rescale = true;
std::array<ImageId, NUM_RT> tmp_color_images{};
ImageId tmp_depth_image{};
const auto check_rescale = [&](ImageViewId view_id, ImageId& id_save) {
if (view_id) {
const auto& view = slot_image_views[view_id];
const auto image_id = view.image_id;
id_save = image_id;
auto& image = slot_images[image_id];
can_rescale &= ImageCanRescale(image);
} else {
id_save = CORRUPT_ID;
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
if (flags[Dirty::ColorBuffer0 + index] || force) {
flags[Dirty::ColorBuffer0 + index] = false;
BindRenderTarget(&color_buffer_id, FindColorBuffer(index, is_clear));
}
PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id));
check_rescale(color_buffer_id, tmp_color_images[index]);
}
if (flags[Dirty::ZetaBuffer] || force) {
flags[Dirty::ZetaBuffer] = false;
BindRenderTarget(&render_targets.depth_buffer_id, FindDepthBuffer(is_clear));
}
check_rescale(render_targets.depth_buffer_id, tmp_depth_image);
if (can_rescale) {
const auto scale_up = [this](ImageId image_id) {
if (image_id != CORRUPT_ID) {
Image& image = slot_images[image_id];
image.ScaleUp();
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
scale_up(tmp_color_images[index]);
}
scale_up(tmp_depth_image);
} else {
const auto scale_down = [this](ImageId image_id) {
if (image_id != CORRUPT_ID) {
Image& image = slot_images[image_id];
image.ScaleDown();
}
};
for (size_t index = 0; index < NUM_RT; ++index) {
scale_down(tmp_color_images[index]);
}
scale_down(tmp_depth_image);
}
// Rescale End
for (size_t index = 0; index < NUM_RT; ++index) {
ImageViewId& color_buffer_id = render_targets.color_buffer_ids[index];
PrepareImageView(color_buffer_id, true, is_clear && IsFullClear(color_buffer_id));
}
const ImageViewId depth_buffer_id = render_targets.depth_buffer_id;
PrepareImageView(depth_buffer_id, true, is_clear && IsFullClear(depth_buffer_id));
@ -623,6 +670,31 @@ ImageId TextureCache<P>::FindImage(const ImageInfo& info, GPUVAddr gpu_addr,
return image_id;
}
template <class P>
bool TextureCache<P>::ImageCanRescale(Image& image) {
if (True(image.flags & ImageFlagBits::Rescaled) ||
True(image.flags & ImageFlagBits::RescaleChecked)) {
return true;
}
const auto& info = image.info;
const bool can_this_rescale =
(info.type == ImageType::e1D || info.type == ImageType::e2D) && info.block.depth == 0;
if (!can_this_rescale) {
image.flags &= ~ImageFlagBits::RescaleChecked;
return false;
}
image.flags |= ImageFlagBits::RescaleChecked;
for (const auto& alias : image.aliased_images) {
Image& other_image = slot_images[alias.id];
if (!ImageCanRescale(other_image)) {
image.flags &= ~ImageFlagBits::RescaleChecked;
return false;
}
}
image.flags &= ~ImageFlagBits::RescaleChecked;
return true;
}
template <class P>
ImageId TextureCache<P>::InsertImage(const ImageInfo& info, GPUVAddr gpu_addr,
RelaxedOptions options) {
@ -660,12 +732,18 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
std::vector<ImageId> right_aliased_ids;
std::unordered_set<ImageId> ignore_textures;
std::vector<ImageId> bad_overlap_ids;
std::vector<ImageId> all_siblings;
const bool this_is_linear = info.type == ImageType::Linear;
const auto region_check = [&](ImageId overlap_id, ImageBase& overlap) {
if (True(overlap.flags & ImageFlagBits::Remapped)) {
ignore_textures.insert(overlap_id);
return;
}
if (info.type == ImageType::Linear) {
const bool overlap_is_linear = overlap.info.type == ImageType::Linear;
if (this_is_linear != overlap_is_linear) {
return;
}
if (this_is_linear && overlap_is_linear) {
if (info.pitch == overlap.info.pitch && gpu_addr == overlap.gpu_addr) {
// Alias linear images with the same pitch
left_aliased_ids.push_back(overlap_id);
@ -681,6 +759,7 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
cpu_addr = solution->cpu_addr;
new_info.resources = solution->resources;
overlap_ids.push_back(overlap_id);
all_siblings.push_back(overlap_id);
return;
}
static constexpr auto options = RelaxedOptions::Size | RelaxedOptions::Format;
@ -688,10 +767,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
if (IsSubresource(new_info, overlap, gpu_addr, options, broken_views, native_bgr)) {
left_aliased_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::Alias;
all_siblings.push_back(overlap_id);
} else if (IsSubresource(overlap.info, new_image_base, overlap.gpu_addr, options,
broken_views, native_bgr)) {
right_aliased_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::Alias;
all_siblings.push_back(overlap_id);
} else {
bad_overlap_ids.push_back(overlap_id);
overlap.flags |= ImageFlagBits::BadOverlap;
@ -709,8 +790,36 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
}
};
ForEachSparseImageInRegion(gpu_addr, size_bytes, region_check_gpu);
bool can_rescale =
(info.type == ImageType::e1D || info.type == ImageType::e2D) && info.block.depth == 0;
for (const ImageId sibling_id : all_siblings) {
if (!can_rescale) {
break;
}
Image& sibling = slot_images[sibling_id];
can_rescale &= ImageCanRescale(sibling);
}
if (can_rescale) {
for (const ImageId sibling_id : all_siblings) {
Image& sibling = slot_images[sibling_id];
sibling.ScaleUp();
}
} else {
for (const ImageId sibling_id : all_siblings) {
Image& sibling = slot_images[sibling_id];
sibling.ScaleDown();
}
}
const ImageId new_image_id = slot_images.insert(runtime, new_info, gpu_addr, cpu_addr);
Image& new_image = slot_images[new_image_id];
if (can_rescale) {
new_image.ScaleUp();
} else {
new_image.ScaleDown();
}
if (!gpu_memory.IsContinousRange(new_image.gpu_addr, new_image.guest_size_bytes)) {
new_image.flags |= ImageFlagBits::Sparse;

10
src/video_core/texture_cache/texture_cache_base.h
View file

@ -142,6 +142,14 @@ public:
const Tegra::Engines::Fermi2D::Surface& src,
const Tegra::Engines::Fermi2D::Config& copy);
/// Invalidate the contents of the color buffer index
/// These contents become unspecified, the cache can assume aggressive optimizations.
void InvalidateColorBuffer(size_t index);
/// Invalidate the contents of the depth buffer
/// These contents become unspecified, the cache can assume aggressive optimizations.
void InvalidateDepthBuffer();
/// Try to find a cached image view in the given CPU address
[[nodiscard]] ImageView* TryFindFramebufferImageView(VAddr cpu_addr);
@ -318,6 +326,8 @@ private:
/// Returns true if the current clear parameters clear the whole image of a given image view
[[nodiscard]] bool IsFullClear(ImageViewId id);
bool ImageCanRescale(Image& image);
Runtime& runtime;
VideoCore::RasterizerInterface& rasterizer;
Tegra::Engines::Maxwell3D& maxwell3d;