ShuriZma
/
duckstation
mirror of https://github.com/stenzek/duckstation
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

GPU: Split backend into Backend+Presenter

This commit is contained in:
Stenzek 2025-01-17 19:22:11 +10:00
parent 8f19ac2dee
commit d52bf795e4
No known key found for this signature in database
18 changed files with 1532 additions and 1389 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/core/CMakeLists.txt
View File

@ -57,6 +57,8 @@ add_library(core
gpu_hw_shadergen.h
gpu_hw_texture_cache.cpp
gpu_hw_texture_cache.h
gpu_presenter.cpp
gpu_presenter.h
gpu_shadergen.cpp
gpu_shadergen.h
gpu_sw.cpp

2
src/core/core.vcxproj
View File

@ -38,6 +38,7 @@
<ClCompile Include="gpu_dump.cpp" />
<ClCompile Include="gpu_hw_shadergen.cpp" />
<ClCompile Include="gpu_hw_texture_cache.cpp" />
<ClCompile Include="gpu_presenter.cpp" />
<ClCompile Include="gpu_shadergen.cpp" />
<ClCompile Include="gpu_sw.cpp" />
<ClCompile Include="gpu_sw_rasterizer.cpp" />
@ -116,6 +117,7 @@
<ClInclude Include="gpu_dump.h" />
<ClInclude Include="gpu_hw_shadergen.h" />
<ClInclude Include="gpu_hw_texture_cache.h" />
<ClInclude Include="gpu_presenter.h" />
<ClInclude Include="gpu_shadergen.h" />
<ClInclude Include="gpu_sw.h" />
<ClInclude Include="gpu_sw_rasterizer.h" />

2
src/core/core.vcxproj.filters
View File

@ -66,6 +66,7 @@
<ClCompile Include="jogcon.cpp" />
<ClCompile Include="pio.cpp" />
<ClCompile Include="gpu_thread.cpp" />
<ClCompile Include="gpu_presenter.cpp" />
</ItemGroup>
<ItemGroup>
<ClInclude Include="types.h" />
@ -142,6 +143,7 @@
<ClInclude Include="pio.h" />
<ClInclude Include="gpu_thread.h" />
<ClInclude Include="gpu_thread_commands.h" />
<ClInclude Include="gpu_presenter.h" />
</ItemGroup>
<ItemGroup>
<None Include="gpu_sw_rasterizer.inl" />

4
src/core/fullscreen_ui.cpp
View File

@ -8329,8 +8329,8 @@ LoadingScreenProgressCallback::~LoadingScreenProgressCallback()
}
else
{
// since this was pushing frames, we need to restore the context
GPUThread::Internal::RestoreContextAfterPresent();
// since this was pushing frames, we need to restore the context. do that by pushing a frame ourselves
GPUThread::Internal::DoRunIdle();
}
}

2
src/core/gpu.cpp
View File

@ -2062,7 +2062,7 @@ bool GPU::StartRecordingGPUDump(const char* path, u32 num_frames /* = 1 */)
// save screenshot to same location to identify it
GPUBackend::RenderScreenshotToFile(Path::ReplaceExtension(path, "png"), DisplayScreenshotMode::ScreenResolution, 85,
true, false);
false);
return true;
}

1265
src/core/gpu_backend.cpp
View File

File diff suppressed because it is too large Load Diff

111
src/core/gpu_backend.h
View File

@ -1,20 +1,13 @@
// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#pragma once
#include "gpu_thread_commands.h"
#include "util/gpu_device.h"
#include "common/heap_array.h"
#include "common/threading.h"
#include "gpu_thread_commands.h"
#include <atomic>
#include <condition_variable>
#include <memory>
#include <mutex>
#include <tuple>
class Error;
class SmallStringBase;
@ -25,6 +18,8 @@ class GPUPipeline;
struct GPUSettings;
class StateWrapper;
class GPUPresenter;
namespace System {
struct MemorySaveState;
}
@ -59,12 +54,12 @@ public:
static bool IsUsingHardwareBackend();
static std::unique_ptr<GPUBackend> CreateHardwareBackend();
static std::unique_ptr<GPUBackend> CreateSoftwareBackend();
static std::unique_ptr<GPUBackend> CreateHardwareBackend(GPUPresenter& presenter);
static std::unique_ptr<GPUBackend> CreateSoftwareBackend(GPUPresenter& presenter);
static bool RenderScreenshotToBuffer(u32 width, u32 height, bool postfx, Image* out_image);
static void RenderScreenshotToFile(const std::string_view path, DisplayScreenshotMode mode, u8 quality,
bool compress_on_thread, bool show_osd_message);
bool show_osd_message);
static bool BeginQueueFrame();
static void WaitForOneQueuedFrame();
@ -73,9 +68,11 @@ public:
static bool AllocateMemorySaveStates(std::span<System::MemorySaveState> states, Error* error);
public:
GPUBackend();
GPUBackend(GPUPresenter& presenter);
virtual ~GPUBackend();
ALWAYS_INLINE const GPUPresenter& GetPresenter() const { return m_presenter; }
virtual bool Initialize(bool upload_vram, Error* error);
virtual void UpdateSettings(const GPUSettings& old_settings);
@ -96,28 +93,13 @@ public:
/// Main command handler for GPU thread.
void HandleCommand(const GPUThreadCommand* cmd);
/// Draws the current display texture, with any post-processing.
GPUDevice::PresentResult PresentDisplay();
/// Helper function to save current display texture to PNG. Used for regtest.
bool WriteDisplayTextureToFile(std::string filename);
/// Helper function for computing screenshot bounds.
void CalculateScreenshotSize(DisplayScreenshotMode mode, u32* width, u32* height, GSVector4i* display_rect,
GSVector4i* draw_rect) const;
void GetStatsString(SmallStringBase& str) const;
void GetMemoryStatsString(SmallStringBase& str) const;
void ResetStatistics();
void UpdateStatistics(u32 frame_count);
protected:
enum : u32
{
DEINTERLACE_BUFFER_COUNT = 4,
};
/// Screen-aligned vertex type for various draw types.
struct ScreenVertex
{
float x;
@ -131,6 +113,18 @@ protected:
}
};
static void SetScreenQuadInputLayout(GPUPipeline::GraphicsConfig& config);
static GSVector4 GetScreenQuadClipSpaceCoordinates(const GSVector4i bounds, const GSVector2i rt_size);
static void DrawScreenQuad(const GSVector4i bounds, const GSVector2i rt_size,
const GSVector4 uv_bounds = GSVector4::cxpr(0.0f, 0.0f, 1.0f, 1.0f));
protected:
enum : u32
{
DEINTERLACE_BUFFER_COUNT = 4,
};
virtual void ReadVRAM(u32 x, u32 y, u32 width, u32 height) = 0;
virtual void FillVRAM(u32 x, u32 y, u32 width, u32 height, u32 color, bool interlaced_rendering,
u8 interlaced_display_field) = 0;
@ -156,67 +150,14 @@ protected:
virtual bool AllocateMemorySaveState(System::MemorySaveState& mss, Error* error) = 0;
virtual void DoMemoryState(StateWrapper& sw, System::MemorySaveState& mss) = 0;
static void SetScreenQuadInputLayout(GPUPipeline::GraphicsConfig& config);
static GSVector4 GetScreenQuadClipSpaceCoordinates(const GSVector4i bounds, const GSVector2i rt_size);
void DrawScreenQuad(const GSVector4i bounds, const GSVector2i rt_size,
const GSVector4 uv_bounds = GSVector4::cxpr(0.0f, 0.0f, 1.0f, 1.0f));
/// Helper function for computing the draw rectangle in a larger window.
void CalculateDrawRect(s32 window_width, s32 window_height, bool apply_rotation, bool apply_aspect_ratio,
GSVector4i* display_rect, GSVector4i* draw_rect) const;
/// Renders the display, optionally with postprocessing to the specified image.
bool RenderScreenshotToBuffer(u32 width, u32 height, const GSVector4i display_rect, const GSVector4i draw_rect,
bool postfx, Image* out_image);
bool CompileDisplayPipelines(bool display, bool deinterlace, bool chroma_smoothing, Error* error);
void HandleUpdateDisplayCommand(const GPUBackendUpdateDisplayCommand* cmd);
void HandleSubmitFrameCommand(const GPUBackendFramePresentationParameters* cmd);
void ClearDisplay();
void ClearDisplayTexture();
void SetDisplayTexture(GPUTexture* texture, GPUTexture* depth_buffer, s32 view_x, s32 view_y, s32 view_width,
s32 view_height);
GPUDevice::PresentResult RenderDisplay(GPUTexture* target, const GSVector4i display_rect, const GSVector4i draw_rect,
bool postfx);
/// Sends the current frame to media capture.
void SendDisplayToMediaCapture(MediaCapture* cap);
bool Deinterlace(u32 field);
bool DeinterlaceSetTargetSize(u32 width, u32 height, bool preserve);
void DestroyDeinterlaceTextures();
bool ApplyChromaSmoothing();
s32 m_display_width = 0;
s32 m_display_height = 0;
GPUPresenter& m_presenter;
GSVector4i m_clamped_drawing_area = {};
s32 m_display_origin_left = 0;
s32 m_display_origin_top = 0;
s32 m_display_vram_width = 0;
s32 m_display_vram_height = 0;
float m_display_pixel_aspect_ratio = 1.0f;
u32 m_current_deinterlace_buffer = 0;
std::unique_ptr<GPUPipeline> m_deinterlace_pipeline;
std::array<std::unique_ptr<GPUTexture>, DEINTERLACE_BUFFER_COUNT> m_deinterlace_buffers;
std::unique_ptr<GPUTexture> m_deinterlace_texture;
std::unique_ptr<GPUPipeline> m_chroma_smoothing_pipeline;
std::unique_ptr<GPUTexture> m_chroma_smoothing_texture;
std::unique_ptr<GPUPipeline> m_display_pipeline;
GPUTexture* m_display_texture = nullptr;
GPUTexture* m_display_depth_buffer = nullptr;
s32 m_display_texture_view_x = 0;
s32 m_display_texture_view_y = 0;
s32 m_display_texture_view_width = 0;
s32 m_display_texture_view_height = 0;
private:
static void ReleaseQueuedFrame();
};
namespace Host {

76
src/core/gpu_hw.cpp
View File

@ -6,6 +6,7 @@
#include "cpu_pgxp.h"
#include "gpu.h"
#include "gpu_hw_shadergen.h"
#include "gpu_presenter.h"
#include "gpu_sw_rasterizer.h"
#include "host.h"
#include "imgui_overlays.h"
@ -201,7 +202,7 @@ private:
};
} // namespace
GPU_HW::GPU_HW() : GPUBackend()
GPU_HW::GPU_HW(GPUPresenter& presenter) : GPUBackend(presenter)
{
#if defined(_DEBUG) || defined(_DEVEL)
s_draw_number = 0;
@ -438,6 +439,8 @@ void GPU_HW::UpdateSettings(const GPUSettings& old_settings)
{
GPUBackend::UpdateSettings(old_settings);
FlushRender();
const GPUDevice::Features features = g_gpu_device->GetFeatures();
const u8 resolution_scale = Truncate8(CalculateResolutionScale());
@ -480,8 +483,8 @@ void GPU_HW::UpdateSettings(const GPUSettings& old_settings)
{
Host::AddIconOSDMessage("ResolutionScaleChanged", ICON_FA_PAINT_BRUSH,
fmt::format(TRANSLATE_FS("GPU_HW", "Internal resolution set to {0}x ({1}x{2})."),
resolution_scale, m_display_width * resolution_scale,
resolution_scale * m_display_height),
resolution_scale, m_presenter.GetDisplayWidth() * resolution_scale,
m_presenter.GetDisplayHeight() * resolution_scale),
Host::OSD_INFO_DURATION);
}
@ -732,8 +735,9 @@ u32 GPU_HW::CalculateResolutionScale() const
else
{
// Auto scaling.
if (m_display_width == 0 || m_display_height == 0 || m_display_vram_width == 0 || m_display_vram_height == 0 ||
!m_display_texture || !g_gpu_device->HasMainSwapChain())
if (m_presenter.GetDisplayWidth() == 0 || m_presenter.GetDisplayHeight() == 0 ||
m_presenter.GetDisplayVRAMWidth() == 0 || m_presenter.GetDisplayVRAMHeight() == 0 ||
!m_presenter.HasDisplayTexture() || !g_gpu_device->HasMainSwapChain())
{
// When the system is starting and all borders crop is enabled, the registers are zero, and
// display_height therefore is also zero. Keep the existing resolution until it updates.
@ -742,18 +746,19 @@ u32 GPU_HW::CalculateResolutionScale() const
else
{
GSVector4i display_rect, draw_rect;
CalculateDrawRect(g_gpu_device->GetMainSwapChain()->GetWidth(), g_gpu_device->GetMainSwapChain()->GetHeight(),
true, true, &display_rect, &draw_rect);
m_presenter.CalculateDrawRect(g_gpu_device->GetMainSwapChain()->GetWidth(),
g_gpu_device->GetMainSwapChain()->GetHeight(), true, true, &display_rect,
&draw_rect);
// We use the draw rect to determine scaling. This way we match the resolution as best we can, regardless of the
// anamorphic aspect ratio.
const s32 draw_width = draw_rect.width();
const s32 draw_height = draw_rect.height();
scale = static_cast<u32>(
std::ceil(std::max(static_cast<float>(draw_width) / static_cast<float>(m_display_vram_width),
static_cast<float>(draw_height) / static_cast<float>(m_display_vram_height))));
std::ceil(std::max(static_cast<float>(draw_width) / static_cast<float>(m_presenter.GetDisplayVRAMWidth()),
static_cast<float>(draw_height) / static_cast<float>(m_presenter.GetDisplayVRAMHeight()))));
VERBOSE_LOG("Draw Size = {}x{}, VRAM Size = {}x{}, Preferred Scale = {}", draw_width, draw_height,
m_display_vram_width, m_display_vram_height, scale);
m_presenter.GetDisplayVRAMWidth(), m_presenter.GetDisplayVRAMHeight(), scale);
}
}
@ -1028,7 +1033,7 @@ void GPU_HW::DeactivateROV()
void GPU_HW::DestroyBuffers()
{
ClearDisplayTexture();
m_presenter.ClearDisplayTexture();
DebugAssert((m_batch_vertex_ptr != nullptr) == (m_batch_index_ptr != nullptr));
if (m_batch_vertex_ptr)
@ -3846,12 +3851,13 @@ void GPU_HW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
if (IsUsingMultisampling())
{
UpdateVRAMReadTexture(!m_vram_dirty_draw_rect.eq(INVALID_RECT), !m_vram_dirty_write_rect.eq(INVALID_RECT));
SetDisplayTexture(m_vram_read_texture.get(), nullptr, 0, 0, m_vram_read_texture->GetWidth(),
m_vram_read_texture->GetHeight());
m_presenter.SetDisplayTexture(m_vram_read_texture.get(), nullptr, 0, 0, m_vram_read_texture->GetWidth(),
m_vram_read_texture->GetHeight());
}
else
{
SetDisplayTexture(m_vram_texture.get(), nullptr, 0, 0, m_vram_texture->GetWidth(), m_vram_texture->GetHeight());
m_presenter.SetDisplayTexture(m_vram_texture.get(), nullptr, 0, 0, m_vram_texture->GetWidth(),
m_vram_texture->GetHeight());
}
return;
@ -3875,7 +3881,7 @@ void GPU_HW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
if (cmd->display_disabled)
{
ClearDisplayTexture();
m_presenter.ClearDisplayTexture();
return;
}
else if (!cmd->display_24bit && line_skip == 0 && !IsUsingMultisampling() &&
@ -3883,15 +3889,15 @@ void GPU_HW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
(scaled_vram_offset_y + scaled_display_height) <= m_vram_texture->GetHeight() &&
!PostProcessing::InternalChain.IsActive())
{
SetDisplayTexture(m_vram_texture.get(), depth_source, scaled_vram_offset_x, scaled_vram_offset_y,
scaled_display_width, scaled_display_height);
m_presenter.SetDisplayTexture(m_vram_texture.get(), depth_source, scaled_vram_offset_x, scaled_vram_offset_y,
scaled_display_width, scaled_display_height);
// Fast path if no copies are needed.
if (interlaced)
{
GL_INS("Deinterlace fast path");
drew_anything = true;
Deinterlace(interlaced_field);
m_presenter.Deinterlace(interlaced_field);
}
else
{
@ -3904,7 +3910,7 @@ void GPU_HW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
GPUTexture::Type::RenderTarget, GPUTexture::Format::RGBA8,
GPUTexture::Flags::None)) [[unlikely]]
{
ClearDisplayTexture();
m_presenter.ClearDisplayTexture();
return;
}
@ -3964,26 +3970,27 @@ void GPU_HW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
drew_anything = true;
SetDisplayTexture(m_vram_extract_texture.get(), depth_source ? m_vram_extract_depth_texture.get() : nullptr, 0, 0,
scaled_display_width, scaled_display_height);
m_presenter.SetDisplayTexture(m_vram_extract_texture.get(),
depth_source ? m_vram_extract_depth_texture.get() : nullptr, 0, 0,
scaled_display_width, scaled_display_height);
if (g_settings.display_24bit_chroma_smoothing)
{
if (ApplyChromaSmoothing())
if (m_presenter.ApplyChromaSmoothing())
{
if (interlaced)
Deinterlace(interlaced_field);
m_presenter.Deinterlace(interlaced_field);
}
}
else
{
if (interlaced)
Deinterlace(interlaced_field);
m_presenter.Deinterlace(interlaced_field);
}
}
if (m_downsample_mode != GPUDownsampleMode::Disabled && !cmd->display_24bit)
{
DebugAssert(m_display_texture);
DebugAssert(m_presenter.HasDisplayTexture());
DownsampleFramebuffer();
}
@ -4024,11 +4031,11 @@ void GPU_HW::OnBufferSwapped()
void GPU_HW::DownsampleFramebuffer()
{
GPUTexture* source = m_display_texture;
const u32 left = m_display_texture_view_x;
const u32 top = m_display_texture_view_y;
const u32 width = m_display_texture_view_width;
const u32 height = m_display_texture_view_height;
GPUTexture* source = m_presenter.GetDisplayTexture();
const u32 left = m_presenter.GetDisplayTextureViewX();
const u32 top = m_presenter.GetDisplayTextureViewY();
const u32 width = m_presenter.GetDisplayTextureViewWidth();
const u32 height = m_presenter.GetDisplayTextureViewHeight();
if (m_downsample_mode == GPUDownsampleMode::Adaptive)
DownsampleFramebufferAdaptive(source, left, top, width, height);
@ -4153,7 +4160,7 @@ void GPU_HW::DownsampleFramebufferAdaptive(GPUTexture* source, u32 left, u32 top
RestoreDeviceContext();
SetDisplayTexture(m_downsample_texture.get(), m_display_depth_buffer, 0, 0, width, height);
m_presenter.SetDisplayTexture(m_downsample_texture.get(), m_presenter.GetDisplayDepthBuffer(), 0, 0, width, height);
}
void GPU_HW::DownsampleFramebufferBoxFilter(GPUTexture* source, u32 left, u32 top, u32 width, u32 height)
@ -4185,10 +4192,11 @@ void GPU_HW::DownsampleFramebufferBoxFilter(GPUTexture* source, u32 left, u32 to
RestoreDeviceContext();
SetDisplayTexture(m_downsample_texture.get(), m_display_depth_buffer, 0, 0, ds_width, ds_height);
m_presenter.SetDisplayTexture(m_downsample_texture.get(), m_presenter.GetDisplayDepthBuffer(), 0, 0, ds_width,
ds_height);
}
std::unique_ptr<GPUBackend> GPUBackend::CreateHardwareBackend()
std::unique_ptr<GPUBackend> GPUBackend::CreateHardwareBackend(GPUPresenter& presenter)
{
return std::make_unique<GPU_HW>();
return std::make_unique<GPU_HW>(presenter);
}

2
src/core/gpu_hw.h
View File

@ -57,7 +57,7 @@ public:
GSVector4i::cxpr(std::numeric_limits<s32>::max(), std::numeric_limits<s32>::max(), std::numeric_limits<s32>::min(),
std::numeric_limits<s32>::min());
GPU_HW();
GPU_HW(GPUPresenter& presenter);
~GPU_HW() override;
bool Initialize(bool upload_vram, Error* error) override;

943
src/core/gpu_presenter.cpp Normal file
View File

@ -0,0 +1,943 @@
// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#include "gpu_presenter.h"
#include "fullscreen_ui.h"
#include "gpu.h"
#include "gpu_backend.h"
#include "gpu_shadergen.h"
#include "gpu_thread.h"
#include "gpu_thread_commands.h"
#include "host.h"
#include "imgui_overlays.h"
#include "performance_counters.h"
#include "save_state_version.h"
#include "settings.h"
#include "system.h"
#include "util/gpu_device.h"
#include "util/image.h"
#include "util/imgui_fullscreen.h"
#include "util/imgui_manager.h"
#include "util/media_capture.h"
#include "util/postprocessing.h"
#include "util/state_wrapper.h"
#include "common/align.h"
#include "common/error.h"
#include "common/file_system.h"
#include "common/gsvector_formatter.h"
#include "common/log.h"
#include "common/path.h"
#include "common/small_string.h"
#include "common/string_util.h"
#include "common/threading.h"
#include "common/timer.h"
#include <numbers>
LOG_CHANNEL(GPU);
static constexpr GPUTexture::Format DISPLAY_INTERNAL_POSTFX_FORMAT = GPUTexture::Format::RGBA8;
GPUPresenter::GPUPresenter() = default;
GPUPresenter::~GPUPresenter()
{
DestroyDeinterlaceTextures();
g_gpu_device->RecycleTexture(std::move(m_chroma_smoothing_texture));
}
bool GPUPresenter::Initialize(Error* error)
{
if (!CompileDisplayPipelines(true, true, g_gpu_settings.display_24bit_chroma_smoothing, error))
return false;
return true;
}
void GPUPresenter::UpdateSettings(const GPUSettings& old_settings)
{
if (g_gpu_settings.display_scaling != old_settings.display_scaling ||
g_gpu_settings.display_deinterlacing_mode != old_settings.display_deinterlacing_mode ||
g_gpu_settings.display_24bit_chroma_smoothing != old_settings.display_24bit_chroma_smoothing)
{
// Toss buffers on mode change.
if (g_gpu_settings.display_deinterlacing_mode != old_settings.display_deinterlacing_mode)
DestroyDeinterlaceTextures();
if (!CompileDisplayPipelines(
g_gpu_settings.display_scaling != old_settings.display_scaling,
g_gpu_settings.display_deinterlacing_mode != old_settings.display_deinterlacing_mode,
g_gpu_settings.display_24bit_chroma_smoothing != old_settings.display_24bit_chroma_smoothing, nullptr))
{
Panic("Failed to compile display pipeline on settings change.");
}
}
}
bool GPUPresenter::CompileDisplayPipelines(bool display, bool deinterlace, bool chroma_smoothing, Error* error)
{
const GPUShaderGen shadergen(g_gpu_device->GetRenderAPI(), g_gpu_device->GetFeatures().dual_source_blend,
g_gpu_device->GetFeatures().framebuffer_fetch);
GPUPipeline::GraphicsConfig plconfig;
plconfig.primitive = GPUPipeline::Primitive::Triangles;
plconfig.rasterization = GPUPipeline::RasterizationState::GetNoCullState();
plconfig.depth = GPUPipeline::DepthState::GetNoTestsState();
plconfig.blend = GPUPipeline::BlendState::GetNoBlendingState();
plconfig.geometry_shader = nullptr;
plconfig.depth_format = GPUTexture::Format::Unknown;
plconfig.samples = 1;
plconfig.per_sample_shading = false;
plconfig.render_pass_flags = GPUPipeline::NoRenderPassFlags;
if (display)
{
GPUBackend::SetScreenQuadInputLayout(plconfig);
plconfig.layout = GPUPipeline::Layout::SingleTextureAndPushConstants;
plconfig.SetTargetFormats(g_gpu_device->HasMainSwapChain() ? g_gpu_device->GetMainSwapChain()->GetFormat() :
GPUTexture::Format::RGBA8);
std::string vs = shadergen.GenerateDisplayVertexShader();
std::string fs;
switch (g_gpu_settings.display_scaling)
{
case DisplayScalingMode::BilinearSharp:
fs = shadergen.GenerateDisplaySharpBilinearFragmentShader();
break;
case DisplayScalingMode::BilinearSmooth:
case DisplayScalingMode::BilinearInteger:
fs = shadergen.GenerateDisplayFragmentShader(true, false);
break;
case DisplayScalingMode::Nearest:
case DisplayScalingMode::NearestInteger:
default:
fs = shadergen.GenerateDisplayFragmentShader(false, true);
break;
}
std::unique_ptr<GPUShader> vso =
g_gpu_device->CreateShader(GPUShaderStage::Vertex, shadergen.GetLanguage(), vs, error);
std::unique_ptr<GPUShader> fso =
g_gpu_device->CreateShader(GPUShaderStage::Fragment, shadergen.GetLanguage(), fs, error);
if (!vso || !fso)
return false;
GL_OBJECT_NAME(vso, "Display Vertex Shader");
GL_OBJECT_NAME_FMT(fso, "Display Fragment Shader [{}]",
Settings::GetDisplayScalingName(g_gpu_settings.display_scaling));
plconfig.vertex_shader = vso.get();
plconfig.fragment_shader = fso.get();
if (!(m_display_pipeline = g_gpu_device->CreatePipeline(plconfig, error)))
return false;
GL_OBJECT_NAME_FMT(m_display_pipeline, "Display Pipeline [{}]",
Settings::GetDisplayScalingName(g_gpu_settings.display_scaling));
}
plconfig.input_layout = {};
plconfig.primitive = GPUPipeline::Primitive::Triangles;
if (deinterlace)
{
std::unique_ptr<GPUShader> vso = g_gpu_device->CreateShader(GPUShaderStage::Vertex, shadergen.GetLanguage(),
shadergen.GenerateScreenQuadVertexShader(), error);
if (!vso)
return false;
GL_OBJECT_NAME(vso, "Deinterlace Vertex Shader");
plconfig.layout = GPUPipeline::Layout::SingleTextureAndPushConstants;
plconfig.vertex_shader = vso.get();
plconfig.SetTargetFormats(GPUTexture::Format::RGBA8);
switch (g_gpu_settings.display_deinterlacing_mode)
{
case DisplayDeinterlacingMode::Disabled:
case DisplayDeinterlacingMode::Progressive:
break;
case DisplayDeinterlacingMode::Weave:
{
std::unique_ptr<GPUShader> fso = g_gpu_device->CreateShader(
GPUShaderStage::Fragment, shadergen.GetLanguage(), shadergen.GenerateDeinterlaceWeaveFragmentShader(), error);
if (!fso)
return false;
GL_OBJECT_NAME(fso, "Weave Deinterlace Fragment Shader");
plconfig.layout = GPUPipeline::Layout::SingleTextureAndPushConstants;
plconfig.vertex_shader = vso.get();
plconfig.fragment_shader = fso.get();
if (!(m_deinterlace_pipeline = g_gpu_device->CreatePipeline(plconfig, error)))
return false;
GL_OBJECT_NAME(m_deinterlace_pipeline, "Weave Deinterlace Pipeline");
}
break;
case DisplayDeinterlacingMode::Blend:
{
std::unique_ptr<GPUShader> fso = g_gpu_device->CreateShader(
GPUShaderStage::Fragment, shadergen.GetLanguage(), shadergen.GenerateDeinterlaceBlendFragmentShader(), error);
if (!fso)
return false;
GL_OBJECT_NAME(fso, "Blend Deinterlace Fragment Shader");
plconfig.layout = GPUPipeline::Layout::MultiTextureAndPushConstants;
plconfig.vertex_shader = vso.get();
plconfig.fragment_shader = fso.get();
if (!(m_deinterlace_pipeline = g_gpu_device->CreatePipeline(plconfig, error)))
return false;
GL_OBJECT_NAME(m_deinterlace_pipeline, "Blend Deinterlace Pipeline");
}
break;
case DisplayDeinterlacingMode::Adaptive:
{
std::unique_ptr<GPUShader> fso =
g_gpu_device->CreateShader(GPUShaderStage::Fragment, shadergen.GetLanguage(),
shadergen.GenerateFastMADReconstructFragmentShader(), error);
if (!fso)
return false;
GL_OBJECT_NAME(fso, "FastMAD Reconstruct Fragment Shader");
plconfig.layout = GPUPipeline::Layout::MultiTextureAndPushConstants;
plconfig.fragment_shader = fso.get();
if (!(m_deinterlace_pipeline = g_gpu_device->CreatePipeline(plconfig, error)))
return false;
GL_OBJECT_NAME(m_deinterlace_pipeline, "FastMAD Reconstruct Pipeline");
}
break;
default:
UnreachableCode();
}
}
if (chroma_smoothing)
{
m_chroma_smoothing_pipeline.reset();
g_gpu_device->RecycleTexture(std::move(m_chroma_smoothing_texture));
if (g_gpu_settings.display_24bit_chroma_smoothing)
{
plconfig.layout = GPUPipeline::Layout::SingleTextureAndPushConstants;
plconfig.SetTargetFormats(GPUTexture::Format::RGBA8);
std::unique_ptr<GPUShader> vso = g_gpu_device->CreateShader(GPUShaderStage::Vertex, shadergen.GetLanguage(),
shadergen.GenerateScreenQuadVertexShader(), error);
std::unique_ptr<GPUShader> fso = g_gpu_device->CreateShader(
GPUShaderStage::Fragment, shadergen.GetLanguage(), shadergen.GenerateChromaSmoothingFragmentShader(), error);
if (!vso || !fso)
return false;
GL_OBJECT_NAME(vso, "Chroma Smoothing Vertex Shader");
GL_OBJECT_NAME(fso, "Chroma Smoothing Fragment Shader");
plconfig.vertex_shader = vso.get();
plconfig.fragment_shader = fso.get();
if (!(m_chroma_smoothing_pipeline = g_gpu_device->CreatePipeline(plconfig, error)))
return false;
GL_OBJECT_NAME(m_chroma_smoothing_pipeline, "Chroma Smoothing Pipeline");
}
}
return true;
}
void GPUPresenter::ClearDisplay()
{
ClearDisplayTexture();
// Just recycle the textures, it'll get re-fetched.
DestroyDeinterlaceTextures();
}
void GPUPresenter::ClearDisplayTexture()
{
m_display_texture = nullptr;
m_display_texture_view_x = 0;
m_display_texture_view_y = 0;
m_display_texture_view_width = 0;
m_display_texture_view_height = 0;
}
void GPUPresenter::SetDisplayParameters(u16 display_width, u16 display_height, u16 display_origin_left,
u16 display_origin_top, u16 display_vram_width, u16 display_vram_height,
float display_pixel_aspect_ratio)
{
m_display_width = display_width;
m_display_height = display_height;
m_display_origin_left = display_origin_left;
m_display_origin_top = display_origin_top;
m_display_vram_width = display_vram_width;
m_display_vram_height = display_vram_height;
m_display_pixel_aspect_ratio = display_pixel_aspect_ratio;
}
void GPUPresenter::SetDisplayTexture(GPUTexture* texture, GPUTexture* depth_buffer, s32 view_x, s32 view_y,
s32 view_width, s32 view_height)
{
DebugAssert(texture);
if (g_gpu_settings.display_auto_resize_window &&
(view_width != m_display_texture_view_width || view_height != m_display_texture_view_height))
{
Host::RunOnCPUThread([]() { System::RequestDisplaySize(); });
}
m_display_texture = texture;
m_display_depth_buffer = depth_buffer;
m_display_texture_view_x = view_x;
m_display_texture_view_y = view_y;
m_display_texture_view_width = view_width;
m_display_texture_view_height = view_height;
}
GPUDevice::PresentResult GPUPresenter::PresentDisplay()
{
if (!g_gpu_device->HasMainSwapChain())
return GPUDevice::PresentResult::SkipPresent;
GSVector4i display_rect;
GSVector4i draw_rect;
CalculateDrawRect(g_gpu_device->GetMainSwapChain()->GetWidth(), g_gpu_device->GetMainSwapChain()->GetHeight(),
!g_gpu_settings.gpu_show_vram, true, &display_rect, &draw_rect);
return RenderDisplay(nullptr, display_rect, draw_rect, !g_gpu_settings.gpu_show_vram);
}
GPUDevice::PresentResult GPUPresenter::RenderDisplay(GPUTexture* target, const GSVector4i display_rect,
const GSVector4i draw_rect, bool postfx)
{
GL_SCOPE_FMT("RenderDisplay: {}", draw_rect);
if (m_display_texture)
m_display_texture->MakeReadyForSampling();
// Internal post-processing.
GPUTexture* display_texture = m_display_texture;
s32 display_texture_view_x = m_display_texture_view_x;
s32 display_texture_view_y = m_display_texture_view_y;
s32 display_texture_view_width = m_display_texture_view_width;
s32 display_texture_view_height = m_display_texture_view_height;
if (postfx && display_texture && PostProcessing::InternalChain.IsActive() &&
PostProcessing::InternalChain.CheckTargets(DISPLAY_INTERNAL_POSTFX_FORMAT, display_texture_view_width,
display_texture_view_height))
{
DebugAssert(display_texture_view_x == 0 && display_texture_view_y == 0 &&
static_cast<s32>(display_texture->GetWidth()) == display_texture_view_width &&
static_cast<s32>(display_texture->GetHeight()) == display_texture_view_height);
// Now we can apply the post chain.
GPUTexture* post_output_texture = PostProcessing::InternalChain.GetOutputTexture();
if (PostProcessing::InternalChain.Apply(display_texture, m_display_depth_buffer, post_output_texture,
GSVector4i(0, 0, display_texture_view_width, display_texture_view_height),
display_texture_view_width, display_texture_view_height, m_display_width,
m_display_height) == GPUDevice::PresentResult::OK)
{
display_texture_view_x = 0;
display_texture_view_y = 0;
display_texture = post_output_texture;
display_texture->MakeReadyForSampling();
}
}
const GPUTexture::Format hdformat = target ? target->GetFormat() : g_gpu_device->GetMainSwapChain()->GetFormat();
const u32 target_width = target ? target->GetWidth() : g_gpu_device->GetMainSwapChain()->GetWidth();
const u32 target_height = target ? target->GetHeight() : g_gpu_device->GetMainSwapChain()->GetHeight();
const bool really_postfx = (postfx && PostProcessing::DisplayChain.IsActive() && g_gpu_device->HasMainSwapChain() &&
hdformat != GPUTexture::Format::Unknown && target_width > 0 && target_height > 0 &&
PostProcessing::DisplayChain.CheckTargets(hdformat, target_width, target_height));
const u32 real_target_width =
(target || really_postfx) ? target_width : g_gpu_device->GetMainSwapChain()->GetPostRotatedWidth();
const u32 real_target_height =
(target || really_postfx) ? target_height : g_gpu_device->GetMainSwapChain()->GetPostRotatedHeight();
GSVector4i real_draw_rect =
(target || really_postfx) ? draw_rect : g_gpu_device->GetMainSwapChain()->PreRotateClipRect(draw_rect);
if (really_postfx)
{
g_gpu_device->ClearRenderTarget(PostProcessing::DisplayChain.GetInputTexture(), GPUDevice::DEFAULT_CLEAR_COLOR);
g_gpu_device->SetRenderTarget(PostProcessing::DisplayChain.GetInputTexture());
}
else
{
if (target)
{
g_gpu_device->SetRenderTarget(target);
}
else
{
const GPUDevice::PresentResult pres = g_gpu_device->BeginPresent(g_gpu_device->GetMainSwapChain());
if (pres != GPUDevice::PresentResult::OK)
return pres;
}
}
if (display_texture)
{
bool texture_filter_linear = false;
struct alignas(16) Uniforms
{
float src_size[4];
float clamp_rect[4];
float params[4];
} uniforms;
std::memset(uniforms.params, 0, sizeof(uniforms.params));
switch (g_gpu_settings.display_scaling)
{
case DisplayScalingMode::Nearest:
case DisplayScalingMode::NearestInteger:
break;
case DisplayScalingMode::BilinearSmooth:
case DisplayScalingMode::BilinearInteger:
texture_filter_linear = true;
break;
case DisplayScalingMode::BilinearSharp:
{
texture_filter_linear = true;
uniforms.params[0] = std::max(
std::floor(static_cast<float>(draw_rect.width()) / static_cast<float>(m_display_texture_view_width)), 1.0f);
uniforms.params[1] = std::max(
std::floor(static_cast<float>(draw_rect.height()) / static_cast<float>(m_display_texture_view_height)), 1.0f);
uniforms.params[2] = 0.5f - 0.5f / uniforms.params[0];
uniforms.params[3] = 0.5f - 0.5f / uniforms.params[1];
}
break;
default:
UnreachableCode();
break;
}
g_gpu_device->SetPipeline(m_display_pipeline.get());
g_gpu_device->SetTextureSampler(
0, display_texture, texture_filter_linear ? g_gpu_device->GetLinearSampler() : g_gpu_device->GetNearestSampler());
// For bilinear, clamp to 0.5/SIZE-0.5 to avoid bleeding from the adjacent texels in VRAM. This is because
// 1.0 in UV space is not the bottom-right texel, but a mix of the bottom-right and wrapped/next texel.
const GSVector2 display_texture_size = GSVector2(display_texture->GetSizeVec());
const GSVector4 display_texture_size4 = GSVector4::xyxy(display_texture_size);
const GSVector4 uv_rect = GSVector4(GSVector4i(display_texture_view_x, display_texture_view_y,
display_texture_view_x + display_texture_view_width,
display_texture_view_y + display_texture_view_height)) /
display_texture_size4;
GSVector4::store<true>(uniforms.clamp_rect,
GSVector4(static_cast<float>(display_texture_view_x) + 0.5f,
static_cast<float>(display_texture_view_y) + 0.5f,
static_cast<float>(display_texture_view_x + display_texture_view_width) - 0.5f,
static_cast<float>(display_texture_view_y + display_texture_view_height) - 0.5f) /
display_texture_size4);
GSVector4::store<true>(uniforms.src_size,
GSVector4::xyxy(display_texture_size, GSVector2::cxpr(1.0f) / display_texture_size));
g_gpu_device->PushUniformBuffer(&uniforms, sizeof(uniforms));
g_gpu_device->SetViewport(0, 0, real_target_width, real_target_height);
g_gpu_device->SetScissor(g_gpu_device->UsesLowerLeftOrigin() ?
GPUDevice::FlipToLowerLeft(real_draw_rect, real_target_height) :
real_draw_rect);
GPUBackend::ScreenVertex* vertices;
u32 space;
u32 base_vertex;
g_gpu_device->MapVertexBuffer(sizeof(GPUBackend::ScreenVertex), 4, reinterpret_cast<void**>(&vertices), &space,
&base_vertex);
const WindowInfo::PreRotation surface_prerotation = (target || really_postfx) ?
WindowInfo::PreRotation::Identity :
g_gpu_device->GetMainSwapChain()->GetPreRotation();
const DisplayRotation uv_rotation = static_cast<DisplayRotation>(
(static_cast<u32>(g_gpu_settings.display_rotation) + static_cast<u32>(surface_prerotation)) %
static_cast<u32>(DisplayRotation::Count));
const GSVector4 xy =
GPUBackend::GetScreenQuadClipSpaceCoordinates(real_draw_rect, GSVector2i(real_target_width, real_target_height));
switch (uv_rotation)
{
case DisplayRotation::Normal:
vertices[0].Set(xy.xy(), uv_rect.xy());
vertices[1].Set(xy.zyzw().xy(), uv_rect.zyzw().xy());
vertices[2].Set(xy.xwzw().xy(), uv_rect.xwzw().xy());
vertices[3].Set(xy.zw(), uv_rect.zw());
break;
case DisplayRotation::Rotate90:
vertices[0].Set(xy.xy(), uv_rect.xwzw().xy());
vertices[1].Set(xy.zyzw().xy(), uv_rect.xy());
vertices[2].Set(xy.xwzw().xy(), uv_rect.zw());
vertices[3].Set(xy.zw(), uv_rect.zyzw().xy());
break;
case DisplayRotation::Rotate180:
vertices[0].Set(xy.xy(), uv_rect.xwzw().xy());
vertices[1].Set(xy.zyzw().xy(), uv_rect.zw());
vertices[2].Set(xy.xwzw().xy(), uv_rect.xy());
vertices[3].Set(xy.zw(), uv_rect.zyzw().xy());
break;
case DisplayRotation::Rotate270:
vertices[0].Set(xy.xy(), uv_rect.zyzw().xy());
vertices[1].Set(xy.zyzw().xy(), uv_rect.zw());
vertices[2].Set(xy.xwzw().xy(), uv_rect.xy());
vertices[3].Set(xy.zw(), uv_rect.xwzw().xy());
break;
DefaultCaseIsUnreachable();
}
g_gpu_device->UnmapVertexBuffer(sizeof(GPUBackend::ScreenVertex), 4);
g_gpu_device->Draw(4, base_vertex);
}
if (really_postfx)
{
DebugAssert(!g_gpu_settings.gpu_show_vram);
// "original size" in postfx includes padding.
const float upscale_x =
m_display_texture ? static_cast<float>(m_display_texture_view_width) / static_cast<float>(m_display_vram_width) :
1.0f;
const float upscale_y = m_display_texture ? static_cast<float>(m_display_texture_view_height) /
static_cast<float>(m_display_vram_height) :
1.0f;
const s32 orig_width = static_cast<s32>(std::ceil(static_cast<float>(m_display_width) * upscale_x));
const s32 orig_height = static_cast<s32>(std::ceil(static_cast<float>(m_display_height) * upscale_y));
return PostProcessing::DisplayChain.Apply(PostProcessing::DisplayChain.GetInputTexture(), nullptr, target,
display_rect, orig_width, orig_height, m_display_width, m_display_height);
}
else
{
return GPUDevice::PresentResult::OK;
}
}
void GPUPresenter::SendDisplayToMediaCapture(MediaCapture* cap)
{
GPUTexture* target = cap->GetRenderTexture();
if (!target) [[unlikely]]
{
WARNING_LOG("Failed to get video capture render texture.");
Host::RunOnCPUThread(&System::StopMediaCapture);
return;
}
const bool apply_aspect_ratio =
(g_gpu_settings.display_screenshot_mode != DisplayScreenshotMode::UncorrectedInternalResolution);
const bool postfx = (g_gpu_settings.display_screenshot_mode != DisplayScreenshotMode::InternalResolution);
GSVector4i display_rect, draw_rect;
CalculateDrawRect(target->GetWidth(), target->GetHeight(), !g_gpu_settings.gpu_show_vram, apply_aspect_ratio,
&display_rect, &draw_rect);
// Not cleared by RenderDisplay().
g_gpu_device->ClearRenderTarget(target, GPUDevice::DEFAULT_CLEAR_COLOR);
if (RenderDisplay(target, display_rect, draw_rect, postfx) != GPUDevice::PresentResult::OK ||
!cap->DeliverVideoFrame(target)) [[unlikely]]
{
WARNING_LOG("Failed to render/deliver video capture frame.");
Host::RunOnCPUThread(&System::StopMediaCapture);
return;
}
}
void GPUPresenter::DestroyDeinterlaceTextures()
{
for (std::unique_ptr<GPUTexture>& tex : m_deinterlace_buffers)
g_gpu_device->RecycleTexture(std::move(tex));
g_gpu_device->RecycleTexture(std::move(m_deinterlace_texture));
m_current_deinterlace_buffer = 0;
}
bool GPUPresenter::Deinterlace(u32 field)
{
GPUTexture* src = m_display_texture;
const u32 x = m_display_texture_view_x;
const u32 y = m_display_texture_view_y;
const u32 width = m_display_texture_view_width;
const u32 height = m_display_texture_view_height;
const auto copy_to_field_buffer = [&](u32 buffer) {
if (!g_gpu_device->ResizeTexture(&m_deinterlace_buffers[buffer], width, height, GPUTexture::Type::Texture,
src->GetFormat(), GPUTexture::Flags::None, false)) [[unlikely]]
{
return false;
}
GL_OBJECT_NAME_FMT(m_deinterlace_buffers[buffer], "Blend Deinterlace Buffer {}", buffer);
GL_INS_FMT("Copy {}x{} from {},{} to field buffer {}", width, height, x, y, buffer);
g_gpu_device->CopyTextureRegion(m_deinterlace_buffers[buffer].get(), 0, 0, 0, 0, m_display_texture, x, y, 0, 0,
width, height);
return true;
};
src->MakeReadyForSampling();
switch (g_gpu_settings.display_deinterlacing_mode)
{
case DisplayDeinterlacingMode::Disabled:
{
GL_INS("Deinterlacing disabled, displaying field texture");
return true;
}
case DisplayDeinterlacingMode::Weave:
{
GL_SCOPE_FMT("DeinterlaceWeave({{{},{}}}, {}x{}, field={})", x, y, width, height, field);
const u32 full_height = height * 2;
if (!DeinterlaceSetTargetSize(width, full_height, true)) [[unlikely]]
{
ClearDisplayTexture();
return false;
}
src->MakeReadyForSampling();
g_gpu_device->SetRenderTarget(m_deinterlace_texture.get());
g_gpu_device->SetPipeline(m_deinterlace_pipeline.get());
g_gpu_device->SetTextureSampler(0, src, g_gpu_device->GetNearestSampler());
const u32 uniforms[4] = {x, y, field, 0};
g_gpu_device->PushUniformBuffer(uniforms, sizeof(uniforms));
g_gpu_device->SetViewportAndScissor(0, 0, width, full_height);
g_gpu_device->Draw(3, 0);
m_deinterlace_texture->MakeReadyForSampling();
SetDisplayTexture(m_deinterlace_texture.get(), m_display_depth_buffer, 0, 0, width, full_height);
return true;
}
case DisplayDeinterlacingMode::Blend:
{
constexpr u32 NUM_BLEND_BUFFERS = 2;
GL_SCOPE_FMT("DeinterlaceBlend({{{},{}}}, {}x{}, field={})", x, y, width, height, field);
const u32 this_buffer = m_current_deinterlace_buffer;
m_current_deinterlace_buffer = (m_current_deinterlace_buffer + 1u) % NUM_BLEND_BUFFERS;
GL_INS_FMT("Current buffer: {}", this_buffer);
if (!DeinterlaceSetTargetSize(width, height, false) || !copy_to_field_buffer(this_buffer)) [[unlikely]]
{
ClearDisplayTexture();
return false;
}
copy_to_field_buffer(this_buffer);
// TODO: could be implemented with alpha blending instead..
g_gpu_device->InvalidateRenderTarget(m_deinterlace_texture.get());
g_gpu_device->SetRenderTarget(m_deinterlace_texture.get());
g_gpu_device->SetPipeline(m_deinterlace_pipeline.get());
g_gpu_device->SetTextureSampler(0, m_deinterlace_buffers[this_buffer].get(), g_gpu_device->GetNearestSampler());
g_gpu_device->SetTextureSampler(1, m_deinterlace_buffers[(this_buffer - 1) % NUM_BLEND_BUFFERS].get(),
g_gpu_device->GetNearestSampler());
g_gpu_device->SetViewportAndScissor(0, 0, width, height);
g_gpu_device->Draw(3, 0);
m_deinterlace_texture->MakeReadyForSampling();
SetDisplayTexture(m_deinterlace_texture.get(), m_display_depth_buffer, 0, 0, width, height);
return true;
}
case DisplayDeinterlacingMode::Adaptive:
{
GL_SCOPE_FMT("DeinterlaceAdaptive({{{},{}}}, {}x{}, field={})", x, y, width, height, field);
const u32 this_buffer = m_current_deinterlace_buffer;
const u32 full_height = height * 2;
m_current_deinterlace_buffer = (m_current_deinterlace_buffer + 1u) % DEINTERLACE_BUFFER_COUNT;
GL_INS_FMT("Current buffer: {}", this_buffer);
if (!DeinterlaceSetTargetSize(width, full_height, false) || !copy_to_field_buffer(this_buffer)) [[unlikely]]
{
ClearDisplayTexture();
return false;
}
g_gpu_device->SetRenderTarget(m_deinterlace_texture.get());
g_gpu_device->SetPipeline(m_deinterlace_pipeline.get());
g_gpu_device->SetTextureSampler(0, m_deinterlace_buffers[this_buffer].get(), g_gpu_device->GetNearestSampler());
g_gpu_device->SetTextureSampler(1, m_deinterlace_buffers[(this_buffer - 1) % DEINTERLACE_BUFFER_COUNT].get(),
g_gpu_device->GetNearestSampler());
g_gpu_device->SetTextureSampler(2, m_deinterlace_buffers[(this_buffer - 2) % DEINTERLACE_BUFFER_COUNT].get(),
g_gpu_device->GetNearestSampler());
g_gpu_device->SetTextureSampler(3, m_deinterlace_buffers[(this_buffer - 3) % DEINTERLACE_BUFFER_COUNT].get(),
g_gpu_device->GetNearestSampler());
const u32 uniforms[] = {field, full_height};
g_gpu_device->PushUniformBuffer(uniforms, sizeof(uniforms));
g_gpu_device->SetViewportAndScissor(0, 0, width, full_height);
g_gpu_device->Draw(3, 0);
m_deinterlace_texture->MakeReadyForSampling();
SetDisplayTexture(m_deinterlace_texture.get(), m_display_depth_buffer, 0, 0, width, full_height);
return true;
}
default:
UnreachableCode();
}
}
bool GPUPresenter::DeinterlaceSetTargetSize(u32 width, u32 height, bool preserve)
{
if (!g_gpu_device->ResizeTexture(&m_deinterlace_texture, width, height, GPUTexture::Type::RenderTarget,
GPUTexture::Format::RGBA8, GPUTexture::Flags::None, preserve)) [[unlikely]]
{
return false;
}
GL_OBJECT_NAME(m_deinterlace_texture, "Deinterlace target texture");
return true;
}
bool GPUPresenter::ApplyChromaSmoothing()
{
const u32 x = m_display_texture_view_x;
const u32 y = m_display_texture_view_y;
const u32 width = m_display_texture_view_width;
const u32 height = m_display_texture_view_height;
if (!g_gpu_device->ResizeTexture(&m_chroma_smoothing_texture, width, height, GPUTexture::Type::RenderTarget,
GPUTexture::Format::RGBA8, GPUTexture::Flags::None, false))
{
ClearDisplayTexture();
return false;
}
GL_OBJECT_NAME(m_chroma_smoothing_texture, "Chroma smoothing texture");
GL_SCOPE_FMT("ApplyChromaSmoothing({{{},{}}}, {}x{})", x, y, width, height);
m_display_texture->MakeReadyForSampling();
g_gpu_device->InvalidateRenderTarget(m_chroma_smoothing_texture.get());
g_gpu_device->SetRenderTarget(m_chroma_smoothing_texture.get());
g_gpu_device->SetPipeline(m_chroma_smoothing_pipeline.get());
g_gpu_device->SetTextureSampler(0, m_display_texture, g_gpu_device->GetNearestSampler());
const u32 uniforms[] = {x, y, width - 1, height - 1};
g_gpu_device->PushUniformBuffer(uniforms, sizeof(uniforms));
g_gpu_device->SetViewportAndScissor(0, 0, width, height);
g_gpu_device->Draw(3, 0);
m_chroma_smoothing_texture->MakeReadyForSampling();
SetDisplayTexture(m_chroma_smoothing_texture.get(), m_display_depth_buffer, 0, 0, width, height);
return true;
}
void GPUPresenter::CalculateDrawRect(s32 window_width, s32 window_height, bool apply_rotation, bool apply_aspect_ratio,
GSVector4i* display_rect, GSVector4i* draw_rect) const
{
const bool integer_scale = (g_gpu_settings.display_scaling == DisplayScalingMode::NearestInteger ||
g_gpu_settings.display_scaling == DisplayScalingMode::BilinearInteger);
const bool show_vram = g_gpu_settings.gpu_show_vram;
const u32 display_width = show_vram ? VRAM_WIDTH : m_display_width;
const u32 display_height = show_vram ? VRAM_HEIGHT : m_display_height;
const s32 display_origin_left = show_vram ? 0 : m_display_origin_left;
const s32 display_origin_top = show_vram ? 0 : m_display_origin_top;
const u32 display_vram_width = show_vram ? VRAM_WIDTH : m_display_vram_width;
const u32 display_vram_height = show_vram ? VRAM_HEIGHT : m_display_vram_height;
const float display_pixel_aspect_ratio = show_vram ? 1.0f : m_display_pixel_aspect_ratio;
GPU::CalculateDrawRect(window_width, window_height, display_width, display_height, display_origin_left,
display_origin_top, display_vram_width, display_vram_height, g_gpu_settings.display_rotation,
g_gpu_settings.display_alignment, display_pixel_aspect_ratio,
g_gpu_settings.display_stretch_vertically, integer_scale, display_rect, draw_rect);
}
bool GPUPresenter::PresentFrame(GPUPresenter* presenter, GPUBackend* backend, bool allow_skip_present, u64 present_time)
{
const bool skip_present = (!g_gpu_device->HasMainSwapChain() ||
(allow_skip_present && g_gpu_device->GetMainSwapChain()->ShouldSkipPresentingFrame() &&
presenter && presenter->m_skipped_present_count < MAX_SKIPPED_PRESENT_COUNT));
if (!skip_present)
{
// acquire for IO.MousePos and system state.
std::atomic_thread_fence(std::memory_order_acquire);
FullscreenUI::Render();
if (backend && System::IsValid())
ImGuiManager::RenderTextOverlays(backend);
ImGuiManager::RenderOverlayWindows();
ImGuiManager::RenderOSDMessages();
ImGuiFullscreen::RenderOverlays();
if (backend && System::GetState() == System::State::Running)
ImGuiManager::RenderSoftwareCursors();
ImGuiManager::RenderDebugWindows();
}
const GPUDevice::PresentResult pres =
skip_present ?
GPUDevice::PresentResult::SkipPresent :
(presenter ? presenter->PresentDisplay() : g_gpu_device->BeginPresent(g_gpu_device->GetMainSwapChain()));
if (pres == GPUDevice::PresentResult::OK)
{
if (presenter)
presenter->m_skipped_present_count = 0;
g_gpu_device->RenderImGui(g_gpu_device->GetMainSwapChain());
const GPUDevice::Features features = g_gpu_device->GetFeatures();
const bool scheduled_present = (present_time != 0);
const bool explicit_present = (scheduled_present && (features.explicit_present && !features.timed_present));
const bool timed_present = (scheduled_present && features.timed_present);
if (scheduled_present && !explicit_present)
{
// No explicit present support, simulate it with Flush.
g_gpu_device->FlushCommands();
SleepUntilPresentTime(present_time);
}
g_gpu_device->EndPresent(g_gpu_device->GetMainSwapChain(), explicit_present, timed_present ? present_time : 0);
if (g_gpu_device->IsGPUTimingEnabled())
PerformanceCounters::AccumulateGPUTime();
if (explicit_present)
{
SleepUntilPresentTime(present_time);
g_gpu_device->SubmitPresent(g_gpu_device->GetMainSwapChain());
}
}
else
{
if (presenter)
presenter->m_skipped_present_count++;
if (pres == GPUDevice::PresentResult::DeviceLost) [[unlikely]]
{
ERROR_LOG("GPU device lost during present.");
return false;
}
if (pres == GPUDevice::PresentResult::ExclusiveFullscreenLost) [[unlikely]]
{
WARNING_LOG("Lost exclusive fullscreen.");
Host::SetFullscreen(false);
}
if (!skip_present)
g_gpu_device->FlushCommands();
// Still need to kick ImGui or it gets cranky.
ImGui::EndFrame();
}
ImGuiManager::NewFrame();
return true;
}
void GPUPresenter::SleepUntilPresentTime(u64 present_time)
{
// Use a spinwait if we undersleep for all platforms except android.. don't want to burn battery.
// Linux also seems to do a much better job of waking up at the requested time.
#if !defined(__linux__) && !defined(__ANDROID__)
Timer::SleepUntil(present_time, true);
#else
Timer::SleepUntil(present_time, false);
#endif
}
bool GPUPresenter::RenderScreenshotToBuffer(u32 width, u32 height, const GSVector4i display_rect,
const GSVector4i draw_rect, bool postfx, Image* out_image)
{
const GPUTexture::Format hdformat =
g_gpu_device->HasMainSwapChain() ? g_gpu_device->GetMainSwapChain()->GetFormat() : GPUTexture::Format::RGBA8;
const ImageFormat image_format = GPUTexture::GetImageFormatForTextureFormat(hdformat);
if (image_format == ImageFormat::None)
return false;
auto render_texture = g_gpu_device->FetchAutoRecycleTexture(width, height, 1, 1, 1, GPUTexture::Type::RenderTarget,
hdformat, GPUTexture::Flags::None);
if (!render_texture)
return false;
g_gpu_device->ClearRenderTarget(render_texture.get(), GPUDevice::DEFAULT_CLEAR_COLOR);
// TODO: this should use copy shader instead.
RenderDisplay(render_texture.get(), display_rect, draw_rect, postfx);
Image image(width, height, image_format);
Error error;
std::unique_ptr<GPUDownloadTexture> dltex;
if (g_gpu_device->GetFeatures().memory_import)
{
dltex = g_gpu_device->CreateDownloadTexture(width, height, hdformat, image.GetPixels(), image.GetStorageSize(),
image.GetPitch(), &error);
}
if (!dltex)
{
if (!(dltex = g_gpu_device->CreateDownloadTexture(width, height, hdformat, &error)))
{
ERROR_LOG("Failed to create {}x{} download texture: {}", width, height, error.GetDescription());
return false;
}
}
dltex->CopyFromTexture(0, 0, render_texture.get(), 0, 0, width, height, 0, 0, false);
if (!dltex->ReadTexels(0, 0, width, height, image.GetPixels(), image.GetPitch()))
return false;
*out_image = std::move(image);
return true;
}
void GPUPresenter::CalculateScreenshotSize(DisplayScreenshotMode mode, u32* width, u32* height,
GSVector4i* display_rect, GSVector4i* draw_rect) const
{
const bool internal_resolution = (mode != DisplayScreenshotMode::ScreenResolution || g_gpu_settings.gpu_show_vram);
if (internal_resolution && m_display_texture_view_width != 0 && m_display_texture_view_height != 0)
{
if (mode == DisplayScreenshotMode::InternalResolution)
{
float f_width = static_cast<float>(m_display_texture_view_width);
float f_height = static_cast<float>(m_display_texture_view_height);
if (!g_gpu_settings.gpu_show_vram)
GPU::ApplyPixelAspectRatioToSize(m_display_pixel_aspect_ratio, &f_width, &f_height);
// DX11 won't go past 16K texture size.
const float max_texture_size = static_cast<float>(g_gpu_device->GetMaxTextureSize());
if (f_width > max_texture_size)
{
f_height = f_height / (f_width / max_texture_size);
f_width = max_texture_size;
}
if (f_height > max_texture_size)
{
f_height = max_texture_size;
f_width = f_width / (f_height / max_texture_size);
}
*width = static_cast<u32>(std::ceil(f_width));
*height = static_cast<u32>(std::ceil(f_height));
}
else // if (mode == DisplayScreenshotMode::UncorrectedInternalResolution)
{
*width = m_display_texture_view_width;
*height = m_display_texture_view_height;
}
// Remove padding, it's not part of the framebuffer.
*draw_rect = GSVector4i(0, 0, static_cast<s32>(*width), static_cast<s32>(*height));
*display_rect = *draw_rect;
}
else
{
*width = g_gpu_device->HasMainSwapChain() ? g_gpu_device->GetMainSwapChain()->GetWidth() : 1;
*height = g_gpu_device->HasMainSwapChain() ? g_gpu_device->GetMainSwapChain()->GetHeight() : 1;
CalculateDrawRect(*width, *height, true, !g_settings.gpu_show_vram, display_rect, draw_rect);
}
}

124
src/core/gpu_presenter.h Normal file
View File

@ -0,0 +1,124 @@
// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#pragma once
#include "util/gpu_device.h"
#include <memory>
class Error;
class Image;
class MediaCapture;
enum class DisplayScreenshotMode : u8;
class GPUBackend;
struct GPUSettings;
struct GPUBackendUpdateDisplayCommand;
struct GPUBackendFramePresentationParameters;
class GPUPresenter final
{
public:
GPUPresenter();
virtual ~GPUPresenter();
/// Main frame presenter - used both when a game is and is not running.
static bool PresentFrame(GPUPresenter* presenter, GPUBackend* backend, bool allow_skip_present, u64 present_time);
ALWAYS_INLINE s32 GetDisplayWidth() const { return m_display_width; }
ALWAYS_INLINE s32 GetDisplayHeight() const { return m_display_height; }
ALWAYS_INLINE s32 GetDisplayVRAMWidth() const { return m_display_vram_width; }
ALWAYS_INLINE s32 GetDisplayVRAMHeight() const { return m_display_vram_height; }
ALWAYS_INLINE s32 GetDisplayTextureViewX() const { return m_display_texture_view_x; }
ALWAYS_INLINE s32 GetDisplayTextureViewY() const { return m_display_texture_view_y; }
ALWAYS_INLINE s32 GetDisplayTextureViewWidth() const { return m_display_texture_view_width; }
ALWAYS_INLINE s32 GetDisplayTextureViewHeight() const { return m_display_texture_view_height; }
ALWAYS_INLINE GPUTexture* GetDisplayTexture() const { return m_display_texture; }
ALWAYS_INLINE GPUTexture* GetDisplayDepthBuffer() const { return m_display_depth_buffer; }
ALWAYS_INLINE bool HasDisplayTexture() const { return m_display_texture; }
bool Initialize(Error* error);
void UpdateSettings(const GPUSettings& old_settings);
void ClearDisplay();
void ClearDisplayTexture();
void SetDisplayParameters(u16 display_width, u16 display_height, u16 display_origin_left, u16 display_origin_top,
u16 display_vram_width, u16 display_vram_height, float display_pixel_aspect_ratio);
void SetDisplayTexture(GPUTexture* texture, GPUTexture* depth_buffer, s32 view_x, s32 view_y, s32 view_width,
s32 view_height);
bool Deinterlace(u32 field);
bool ApplyChromaSmoothing();
/// Helper function for computing the draw rectangle in a larger window.
void CalculateDrawRect(s32 window_width, s32 window_height, bool apply_rotation, bool apply_aspect_ratio,
GSVector4i* display_rect, GSVector4i* draw_rect) const;
/// Helper function for computing screenshot bounds.
void CalculateScreenshotSize(DisplayScreenshotMode mode, u32* width, u32* height, GSVector4i* display_rect,
GSVector4i* draw_rect) const;
/// Renders the display, optionally with postprocessing to the specified image.
bool RenderScreenshotToBuffer(u32 width, u32 height, const GSVector4i display_rect, const GSVector4i draw_rect,
bool postfx, Image* out_image);
/// Sends the current frame to media capture.
void SendDisplayToMediaCapture(MediaCapture* cap);
private:
enum : u32
{
DEINTERLACE_BUFFER_COUNT = 4,
MAX_SKIPPED_PRESENT_COUNT = 50,
};
static void SleepUntilPresentTime(u64 present_time);
/// Draws the current display texture, with any post-processing.
GPUDevice::PresentResult PresentDisplay();
bool CompileDisplayPipelines(bool display, bool deinterlace, bool chroma_smoothing, Error* error);
GPUDevice::PresentResult RenderDisplay(GPUTexture* target, const GSVector4i display_rect, const GSVector4i draw_rect,
bool postfx);
bool DeinterlaceSetTargetSize(u32 width, u32 height, bool preserve);
void DestroyDeinterlaceTextures();
s32 m_display_width = 0;
s32 m_display_height = 0;
s32 m_display_origin_left = 0;
s32 m_display_origin_top = 0;
s32 m_display_vram_width = 0;
s32 m_display_vram_height = 0;
float m_display_pixel_aspect_ratio = 1.0f;
u32 m_current_deinterlace_buffer = 0;
std::unique_ptr<GPUPipeline> m_deinterlace_pipeline;
std::array<std::unique_ptr<GPUTexture>, DEINTERLACE_BUFFER_COUNT> m_deinterlace_buffers;
std::unique_ptr<GPUTexture> m_deinterlace_texture;
std::unique_ptr<GPUPipeline> m_chroma_smoothing_pipeline;
std::unique_ptr<GPUTexture> m_chroma_smoothing_texture;
std::unique_ptr<GPUPipeline> m_display_pipeline;
GPUTexture* m_display_texture = nullptr;
GPUTexture* m_display_depth_buffer = nullptr;
s32 m_display_texture_view_x = 0;
s32 m_display_texture_view_y = 0;
s32 m_display_texture_view_width = 0;
s32 m_display_texture_view_height = 0;
u32 m_skipped_present_count = 0;
};
namespace Host {
/// Called at the end of the frame, before presentation.
void FrameDoneOnGPUThread(GPUPresenter* gpu_presenter, u32 frame_number);
} // namespace Host

27
src/core/gpu_sw.cpp
View File

@ -3,6 +3,7 @@
#include "gpu_sw.h"
#include "gpu.h"
#include "gpu_presenter.h"
#include "gpu_sw_rasterizer.h"
#include "settings.h"
#include "system_private.h"
@ -20,7 +21,9 @@
LOG_CHANNEL(GPU);
GPU_SW::GPU_SW() = default;
GPU_SW::GPU_SW(GPUPresenter& presenter) : GPUBackend(presenter)
{
}
GPU_SW::~GPU_SW() = default;
@ -209,7 +212,7 @@ GPUTexture* GPU_SW::GetDisplayTexture(u32 width, u32 height, GPUTexture::Format
if (!m_upload_texture || m_upload_texture->GetWidth() != width || m_upload_texture->GetHeight() != height ||
m_upload_texture->GetFormat() != format)
{
ClearDisplayTexture();
m_presenter.ClearDisplayTexture();
g_gpu_device->RecycleTexture(std::move(m_upload_texture));
m_upload_texture = g_gpu_device->FetchTexture(width, height, 1, 1, 1, GPUTexture::Type::Texture, format,
GPUTexture::Flags::AllowMap, nullptr, 0);
@ -388,7 +391,7 @@ void GPU_SW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
{
if (cmd->display_disabled)
{
ClearDisplayTexture();
m_presenter.ClearDisplayTexture();
return;
}
@ -407,15 +410,15 @@ void GPU_SW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
{
if (CopyOut(src_x, src_y, skip_x, width, height, line_skip, is_24bit))
{
SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, width, height);
m_presenter.SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, width, height);
if (is_24bit && g_settings.display_24bit_chroma_smoothing)
{
if (ApplyChromaSmoothing())
Deinterlace(field);
if (m_presenter.ApplyChromaSmoothing())
m_presenter.Deinterlace(field);
}
else
{
Deinterlace(field);
m_presenter.Deinterlace(field);
}
}
}
@ -423,20 +426,20 @@ void GPU_SW::UpdateDisplay(const GPUBackendUpdateDisplayCommand* cmd)
{
if (CopyOut(src_x, src_y, skip_x, width, height, 0, is_24bit))
{
SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, width, height);
m_presenter.SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, width, height);
if (is_24bit && g_settings.display_24bit_chroma_smoothing)
ApplyChromaSmoothing();
m_presenter.ApplyChromaSmoothing();
}
}
}
else
{
if (CopyOut(0, 0, 0, VRAM_WIDTH, VRAM_HEIGHT, 0, false))
SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, VRAM_WIDTH, VRAM_HEIGHT);
m_presenter.SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, VRAM_WIDTH, VRAM_HEIGHT);
}
}
std::unique_ptr<GPUBackend> GPUBackend::CreateSoftwareBackend()
std::unique_ptr<GPUBackend> GPUBackend::CreateSoftwareBackend(GPUPresenter& presenter)
{
return std::make_unique<GPU_SW>();
return std::make_unique<GPU_SW>(presenter);
}

2
src/core/gpu_sw.h
View File

@ -17,7 +17,7 @@
class GPU_SW final : public GPUBackend
{
public:
GPU_SW();
GPU_SW(GPUPresenter& presenter);
~GPU_SW() override;
bool Initialize(bool upload_vram, Error* error) override;

248
src/core/gpu_thread.cpp
View File

@ -5,6 +5,7 @@
#include "fullscreen_ui.h"
#include "gpu_backend.h"
#include "gpu_hw_texture_cache.h"
#include "gpu_presenter.h"
#include "gpu_thread_commands.h"
#include "gpu_types.h"
#include "host.h"
@ -42,7 +43,6 @@ enum : u32
{
COMMAND_QUEUE_SIZE = 16 * 1024 * 1024,
THRESHOLD_TO_WAKE_GPU = 65536,
MAX_SKIPPED_PRESENT_COUNT = 50
};
static constexpr s32 THREAD_WAKE_COUNT_CPU_THREAD_IS_WAITING = 0x40000000; // CPU thread needs waking
@ -76,19 +76,17 @@ static void DestroyDeviceOnThread(bool clear_fsui_state);
static void ResizeDisplayWindowOnThread(u32 width, u32 height, float scale);
static void UpdateDisplayWindowOnThread(bool fullscreen);
static void DisplayWindowResizedOnThread();
static void HandleGPUDeviceLost();
static void HandleExclusiveFullscreenLost();
static void ReconfigureOnThread(GPUThreadReconfigureCommand* cmd);
static bool CreateGPUBackendOnThread(GPURenderer renderer, bool upload_vram, Error* error);
static void DestroyGPUBackendOnThread();
static void DestroyGPUPresenterOnThread();
static bool PresentFrameAndRestoreContext();
static void UpdateSettingsOnThread(const GPUSettings& old_settings);
static void UpdateRunIdle();
static void SleepUntilPresentTime(Timer::Value present_time);
namespace {
struct ALIGN_TO_CACHE_LINE State
@ -109,8 +107,8 @@ struct ALIGN_TO_CACHE_LINE State
// Owned by GPU thread.
ALIGN_TO_CACHE_LINE std::unique_ptr<GPUBackend> gpu_backend;
ALIGN_TO_CACHE_LINE std::unique_ptr<GPUPresenter> gpu_presenter;
std::atomic<u32> command_fifo_read_ptr{0};
u32 skipped_present_count = 0;
u8 run_idle_reasons = 0;
bool run_idle_flag = false;
GPUVSyncMode requested_vsync = GPUVSyncMode::Disabled;
@ -528,7 +526,9 @@ void GPUThread::Internal::GPUThreadEntryPoint()
void GPUThread::Internal::DoRunIdle()
{
PresentFrame(false, 0);
if (!PresentFrameAndRestoreContext())
return;
if (!g_gpu_device->GetMainSwapChain()->IsVSyncModeBlocking())
g_gpu_device->GetMainSwapChain()->ThrottlePresentation();
}
@ -722,6 +722,9 @@ void GPUThread::DestroyDeviceOnThread(bool clear_fsui_state)
if (!g_gpu_device)
return;
// Presenter should be gone by this point
Assert(!s_state.gpu_presenter);
const bool has_window = g_gpu_device->HasMainSwapChain();
FullscreenUI::Shutdown(clear_fsui_state);
@ -738,64 +741,30 @@ void GPUThread::DestroyDeviceOnThread(bool clear_fsui_state)
std::atomic_thread_fence(std::memory_order_release);
}
void GPUThread::HandleGPUDeviceLost()
{
static Timer::Value s_last_gpu_reset_time = 0;
static constexpr float MIN_TIME_BETWEEN_RESETS = 15.0f;
// If we're constantly crashing on something in particular, we don't want to end up in an
// endless reset loop.. that'd probably end up leaking memory and/or crashing us for other
// reasons. So just abort in such case.
const Timer::Value current_time = Timer::GetCurrentValue();
if (s_last_gpu_reset_time != 0 &&
Timer::ConvertValueToSeconds(current_time - s_last_gpu_reset_time) < MIN_TIME_BETWEEN_RESETS)
{
Panic("Host GPU lost too many times, device is probably completely wedged.");
}
s_last_gpu_reset_time = current_time;
const bool is_fullscreen = Host::IsFullscreen();
// Device lost, something went really bad.
// Let's just toss out everything, and try to hobble on.
DestroyGPUBackendOnThread();
DestroyDeviceOnThread(false);
Error error;
if (!CreateDeviceOnThread(
Settings::GetRenderAPIForRenderer(s_state.requested_renderer.value_or(g_gpu_settings.gpu_renderer)),
is_fullscreen, true, &error) ||
(s_state.requested_renderer.has_value() &&
!CreateGPUBackendOnThread(s_state.requested_renderer.value(), true, &error)))
{
ERROR_LOG("Failed to recreate GPU device after loss: {}", error.GetDescription());
Panic("Failed to recreate GPU device after loss.");
return;
}
// First frame after reopening is definitely going to be trash, so skip it.
Host::AddIconOSDWarning(
"HostGPUDeviceLost", ICON_EMOJI_WARNING,
TRANSLATE_STR("System", "Host GPU device encountered an error and has recovered. This may cause broken rendering."),
Host::OSD_CRITICAL_ERROR_DURATION);
}
void GPUThread::HandleExclusiveFullscreenLost()
{
WARNING_LOG("Lost exclusive fullscreen.");
Host::SetFullscreen(false);
}
bool GPUThread::CreateGPUBackendOnThread(GPURenderer renderer, bool upload_vram, Error* error)
{
Error local_error;
// Create presenter if we don't already have one.
if (!s_state.gpu_presenter)
{
s_state.gpu_presenter = std::make_unique<GPUPresenter>();
if (!s_state.gpu_presenter->Initialize(&local_error))
{
ERROR_LOG("Failed to create presenter: {}", local_error.GetDescription());
Error::SetStringFmt(error, "Failed to create presenter: {}", local_error.GetDescription());
s_state.gpu_presenter.reset();
return false;
}
}
const bool is_hardware = (renderer != GPURenderer::Software);
if (is_hardware)
s_state.gpu_backend = GPUBackend::CreateHardwareBackend();
s_state.gpu_backend = GPUBackend::CreateHardwareBackend(*s_state.gpu_presenter);
else
s_state.gpu_backend = GPUBackend::CreateSoftwareBackend();
s_state.gpu_backend = GPUBackend::CreateSoftwareBackend(*s_state.gpu_presenter);
Error local_error;
bool okay = s_state.gpu_backend->Initialize(upload_vram, &local_error);
if (!okay)
{
@ -810,7 +779,7 @@ bool GPUThread::CreateGPUBackendOnThread(GPURenderer renderer, bool upload_vram,
Host::OSD_CRITICAL_ERROR_DURATION);
s_state.requested_renderer = GPURenderer::Software;
s_state.gpu_backend = GPUBackend::CreateSoftwareBackend();
s_state.gpu_backend = GPUBackend::CreateSoftwareBackend(*s_state.gpu_presenter);
okay = s_state.gpu_backend->Initialize(upload_vram, &local_error);
}
@ -825,7 +794,7 @@ bool GPUThread::CreateGPUBackendOnThread(GPURenderer renderer, bool upload_vram,
g_gpu_device->SetGPUTimingEnabled(g_gpu_settings.display_show_gpu_usage);
PostProcessing::Initialize();
ImGuiManager::UpdateDebugWindowConfig();
Internal::RestoreContextAfterPresent();
s_state.gpu_backend->RestoreDeviceContext();
SetRunIdleReason(RunIdleReason::NoGPUBackend, false);
std::atomic_thread_fence(std::memory_order_release);
return true;
@ -843,6 +812,7 @@ void GPUThread::ReconfigureOnThread(GPUThreadReconfigureCommand* cmd)
if (!cmd->renderer.has_value() && !s_state.requested_fullscreen_ui)
{
DestroyGPUBackendOnThread();
DestroyGPUPresenterOnThread();
DestroyDeviceOnThread(true);
return;
}
@ -875,6 +845,7 @@ void GPUThread::ReconfigureOnThread(GPUThreadReconfigureCommand* cmd)
if (cmd->force_recreate_device || !GPUDevice::IsSameRenderAPI(current_api, expected_api))
{
const bool fullscreen = cmd->fullscreen.value_or(Host::IsFullscreen());
DestroyGPUPresenterOnThread();
DestroyDeviceOnThread(false);
Error local_error;
@ -902,7 +873,11 @@ void GPUThread::ReconfigureOnThread(GPUThreadReconfigureCommand* cmd)
if (cmd->renderer.has_value())
{
// Do we want a renderer?
*cmd->out_result = CreateGPUBackendOnThread(cmd->renderer.value(), cmd->upload_vram, cmd->error_ptr);
if (!(*cmd->out_result = CreateGPUBackendOnThread(cmd->renderer.value(), cmd->upload_vram, cmd->error_ptr)))
{
// No point keeping the presenter around.
DestroyGPUPresenterOnThread();
}
}
else if (s_state.requested_fullscreen_ui)
{
@ -913,6 +888,9 @@ void GPUThread::ReconfigureOnThread(GPUThreadReconfigureCommand* cmd)
return;
}
// Don't need to present game frames anymore.
DestroyGPUPresenterOnThread();
// Don't need timing to run FSUI.
g_gpu_device->SetGPUTimingEnabled(false);
@ -926,6 +904,7 @@ void GPUThread::ReconfigureOnThread(GPUThreadReconfigureCommand* cmd)
else
{
// Device is no longer needed.
DestroyGPUBackendOnThread();
DestroyDeviceOnThread(true);
}
}
@ -945,6 +924,34 @@ void GPUThread::DestroyGPUBackendOnThread()
s_state.gpu_backend.reset();
}
void GPUThread::DestroyGPUPresenterOnThread()
{
if (!s_state.gpu_presenter)
return;
VERBOSE_LOG("Shutting down GPU presenter...");
// Should have no queued frames by this point. Backend can get replaced with null.
Assert(!s_state.gpu_backend);
Assert(GPUBackend::GetQueuedFrameCount() == 0);
s_state.gpu_presenter.reset();
}
bool GPUThread::PresentFrameAndRestoreContext()
{
if (s_state.gpu_backend)
s_state.gpu_backend->FlushRender();
if (!GPUPresenter::PresentFrame(s_state.gpu_presenter.get(), s_state.gpu_backend.get(), false, 0))
return false;
if (s_state.gpu_backend)
s_state.gpu_backend->RestoreDeviceContext();
return true;
}
void GPUThread::UpdateSettingsOnThread(const GPUSettings& old_settings)
{
if (g_gpu_device)
@ -964,11 +971,12 @@ void GPUThread::UpdateSettingsOnThread(const GPUSettings& old_settings)
PostProcessing::UpdateSettings();
s_state.gpu_presenter->UpdateSettings(old_settings);
s_state.gpu_backend->UpdateSettings(old_settings);
if (ImGuiManager::UpdateDebugWindowConfig() || (PostProcessing::DisplayChain.IsActive() && !IsSystemPaused()))
Internal::PresentFrame(false, 0);
s_state.gpu_backend->RestoreDeviceContext();
PresentFrameAndRestoreContext();
else
s_state.gpu_backend->RestoreDeviceContext();
}
}
@ -1057,7 +1065,7 @@ void GPUThread::UpdateSettings(bool gpu_settings_changed, bool device_settings_c
{
PostProcessing::UpdateSettings();
if (ImGuiManager::UpdateDebugWindowConfig() || (PostProcessing::DisplayChain.IsActive() && !IsSystemPaused()))
Internal::PresentFrame(false, 0);
PresentFrameAndRestoreContext();
}
});
}
@ -1180,8 +1188,8 @@ void GPUThread::DisplayWindowResizedOnThread()
{
// Hackity hack, on some systems, presenting a single frame isn't enough to actually get it
// displayed. Two seems to be good enough. Maybe something to do with direct scanout.
Internal::PresentFrame(false, 0);
Internal::PresentFrame(false, 0);
PresentFrameAndRestoreContext();
PresentFrameAndRestoreContext();
}
if (g_gpu_settings.gpu_resolution_scale == 0)
@ -1231,112 +1239,12 @@ void GPUThread::PresentCurrentFrame()
return;
}
Internal::PresentFrame(false, 0);
// But we shouldn't be not running idle without a GPU backend.
if (s_state.gpu_backend)
PresentFrameAndRestoreContext();
});
}
void GPUThread::SleepUntilPresentTime(Timer::Value present_time)
{
// Use a spinwait if we undersleep for all platforms except android.. don't want to burn battery.
// Linux also seems to do a much better job of waking up at the requested time.
#if !defined(__linux__) && !defined(__ANDROID__)
Timer::SleepUntil(present_time, true);
#else
Timer::SleepUntil(present_time, false);
#endif
}
void GPUThread::Internal::PresentFrame(bool allow_skip_present, u64 present_time)
{
if (s_state.gpu_backend)
s_state.gpu_backend->FlushRender();
const bool skip_present = (!g_gpu_device->HasMainSwapChain() ||
(allow_skip_present && g_gpu_device->GetMainSwapChain()->ShouldSkipPresentingFrame() &&
s_state.skipped_present_count < MAX_SKIPPED_PRESENT_COUNT));
if (!skip_present)
{
// acquire for IO.MousePos and system state.
std::atomic_thread_fence(std::memory_order_acquire);
FullscreenUI::Render();
if (s_state.gpu_backend && System::IsValid())
ImGuiManager::RenderTextOverlays(s_state.gpu_backend.get());
ImGuiManager::RenderOverlayWindows();
ImGuiManager::RenderOSDMessages();
ImGuiFullscreen::RenderOverlays();
if (s_state.gpu_backend && System::GetState() == System::State::Running)
ImGuiManager::RenderSoftwareCursors();
ImGuiManager::RenderDebugWindows();
}
const GPUDevice::PresentResult pres =
skip_present ? GPUDevice::PresentResult::SkipPresent :
(s_state.gpu_backend ? s_state.gpu_backend->PresentDisplay() :
g_gpu_device->BeginPresent(g_gpu_device->GetMainSwapChain()));
if (pres == GPUDevice::PresentResult::OK)
{
s_state.skipped_present_count = 0;
g_gpu_device->RenderImGui(g_gpu_device->GetMainSwapChain());
const GPUDevice::Features features = g_gpu_device->GetFeatures();
const bool scheduled_present = (present_time != 0);
const bool explicit_present = (scheduled_present && (features.explicit_present && !features.timed_present));
const bool timed_present = (scheduled_present && features.timed_present);
if (scheduled_present && !explicit_present)
{
// No explicit present support, simulate it with Flush.
g_gpu_device->FlushCommands();
SleepUntilPresentTime(present_time);
}
g_gpu_device->EndPresent(g_gpu_device->GetMainSwapChain(), explicit_present, timed_present ? present_time : 0);
if (g_gpu_device->IsGPUTimingEnabled())
PerformanceCounters::AccumulateGPUTime();
if (explicit_present)
{
SleepUntilPresentTime(present_time);
g_gpu_device->SubmitPresent(g_gpu_device->GetMainSwapChain());
}
}
else
{
s_state.skipped_present_count++;
if (pres == GPUDevice::PresentResult::DeviceLost) [[unlikely]]
HandleGPUDeviceLost();
else if (pres == GPUDevice::PresentResult::ExclusiveFullscreenLost)
HandleExclusiveFullscreenLost();
else if (!skip_present)
g_gpu_device->FlushCommands();
// Still need to kick ImGui or it gets cranky.
ImGui::EndFrame();
}
ImGuiManager::NewFrame();
RestoreContextAfterPresent();
}
void GPUThread::Internal::RestoreContextAfterPresent()
{
if (s_state.gpu_backend)
s_state.gpu_backend->RestoreDeviceContext();
}
bool GPUThread::GetRunIdleReason(RunIdleReason reason)
{
return (s_state.run_idle_reasons & static_cast<u8>(reason)) != 0;

3
src/core/gpu_thread.h
View File

@ -89,7 +89,6 @@ void PushCommandAndWakeThread(GPUThreadCommand* cmd);
void PushCommandAndSync(GPUThreadCommand* cmd, bool spin);
void SyncGPUThread(bool spin);
// NOTE: Only called by GPUBackend
namespace Internal {
const Threading::ThreadHandle& GetThreadHandle();
void ProcessStartup();
@ -97,8 +96,6 @@ void SetThreadEnabled(bool enabled);
void DoRunIdle();
void RequestShutdown();
void GPUThreadEntryPoint();
void PresentFrame(bool allow_skip_present, u64 present_time);
void RestoreContextAfterPresent();
} // namespace Internal
} // namespace GPUThread

10
src/core/system.cpp
View File

@ -19,6 +19,7 @@
#include "gpu_backend.h"
#include "gpu_dump.h"
#include "gpu_hw_texture_cache.h"
#include "gpu_presenter.h"
#include "gpu_thread.h"
#include "gte.h"
#include "host.h"
@ -5254,8 +5255,7 @@ std::string System::GetScreenshotPath(const char* extension)
return path;
}
void System::SaveScreenshot(const char* path, DisplayScreenshotMode mode, DisplayScreenshotFormat format, u8 quality,
bool compress_on_thread)
void System::SaveScreenshot(const char* path, DisplayScreenshotMode mode, DisplayScreenshotFormat format, u8 quality)
{
if (!IsValid())
return;
@ -5264,7 +5264,7 @@ void System::SaveScreenshot(const char* path, DisplayScreenshotMode mode, Displa
if (!path)
path = (auto_path = GetScreenshotPath(Settings::GetDisplayScreenshotFormatExtension(format))).c_str();
GPUBackend::RenderScreenshotToFile(path, mode, quality, compress_on_thread, true);
GPUBackend::RenderScreenshotToFile(path, mode, quality, true);
}
bool System::StartRecordingGPUDump(const char* path /*= nullptr*/, u32 num_frames /*= 0*/)
@ -5332,8 +5332,8 @@ bool System::StartMediaCapture(std::string path)
GSVector4i unused_display_rect, unused_draw_rect;
u32 video_width, video_height;
backend->CalculateScreenshotSize(DisplayScreenshotMode::InternalResolution, &video_width, &video_height,
&unused_display_rect, &unused_draw_rect);
backend->GetPresenter().CalculateScreenshotSize(DisplayScreenshotMode::InternalResolution, &video_width,
&video_height, &unused_display_rect, &unused_draw_rect);
// fire back to the CPU thread to actually start the capture
Host::RunOnCPUThread([path = std::move(path), capture_audio, video_width, video_height]() mutable {

2
src/core/system.h
View File

@ -391,7 +391,7 @@ void UpdateVolume();
/// Saves a screenshot to the specified file. If no file name is provided, one will be generated automatically.
void SaveScreenshot(const char* path = nullptr, DisplayScreenshotMode mode = g_settings.display_screenshot_mode,
DisplayScreenshotFormat format = g_settings.display_screenshot_format,
u8 quality = g_settings.display_screenshot_quality, bool compress_on_thread = true);
u8 quality = g_settings.display_screenshot_quality);
/// Starts/stops GPU dump/trace recording.
bool StartRecordingGPUDump(const char* path = nullptr, u32 num_frames = 1);

96
src/duckstation-regtest/regtest_host.cpp
View File

@ -1,4 +1,4 @@
// SPDX-FileCopyrightText: 2019-2024 Connor McLaughlin <stenzek@gmail.com>
// SPDX-FileCopyrightText: 2019-2025 Connor McLaughlin <stenzek@gmail.com>
// SPDX-License-Identifier: CC-BY-NC-ND-4.0
#include "core/achievements.h"
@ -8,6 +8,7 @@
#include "core/game_list.h"
#include "core/gpu.h"
#include "core/gpu_backend.h"
#include "core/gpu_presenter.h"
#include "core/gpu_thread.h"
#include "core/host.h"
#include "core/spu.h"
@ -32,6 +33,7 @@
#include "common/path.h"
#include "common/sha256_digest.h"
#include "common/string_util.h"
#include "common/threading.h"
#include "common/timer.h"
#include "fmt/format.h"
@ -52,7 +54,7 @@ static void HookSignals();
static bool SetFolders();
static bool SetNewDataRoot(const std::string& filename);
static void DumpSystemStateHashes();
static std::string GetFrameDumpFilename(u32 frame);
static std::string GetFrameDumpPath(u32 frame);
static void GPUThreadEntryPoint();
} // namespace RegTestHost
@ -400,8 +402,92 @@ void Host::DestroyAuxiliaryRenderWindow(AuxiliaryRenderWindowHandle handle, s32*
void Host::FrameDoneOnGPUThread(GPUBackend* gpu_backend, u32 frame_number)
{
if (s_frame_dump_interval > 0 && (s_frame_dump_interval == 1 || (frame_number % s_frame_dump_interval) == 0))
gpu_backend->WriteDisplayTextureToFile(RegTestHost::GetFrameDumpFilename(frame_number));
const GPUPresenter& presenter = gpu_backend->GetPresenter();
if (s_frame_dump_interval == 0 || (frame_number % s_frame_dump_interval) != 0 || !presenter.HasDisplayTexture())
return;
// Need to take a copy of the display texture.
GPUTexture* const read_texture = presenter.GetDisplayTexture();
const u32 read_x = static_cast<u32>(presenter.GetDisplayTextureViewX());
const u32 read_y = static_cast<u32>(presenter.GetDisplayTextureViewY());
const u32 read_width = static_cast<u32>(presenter.GetDisplayTextureViewWidth());
const u32 read_height = static_cast<u32>(presenter.GetDisplayTextureViewHeight());
const ImageFormat read_format = GPUTexture::GetImageFormatForTextureFormat(read_texture->GetFormat());
if (read_format == ImageFormat::None)
return;
Image image(read_width, read_height, read_format);
std::unique_ptr<GPUDownloadTexture> dltex;
if (g_gpu_device->GetFeatures().memory_import)
{
dltex = g_gpu_device->CreateDownloadTexture(read_width, read_height, read_texture->GetFormat(), image.GetPixels(),
image.GetStorageSize(), image.GetPitch());
}
if (!dltex)
{
if (!(dltex = g_gpu_device->CreateDownloadTexture(read_width, read_height, read_texture->GetFormat())))
{
ERROR_LOG("Failed to create {}x{} {} download texture", read_width, read_height,
GPUTexture::GetFormatName(read_texture->GetFormat()));
return;
}
}
dltex->CopyFromTexture(0, 0, read_texture, read_x, read_y, read_width, read_height, 0, 0, !dltex->IsImported());
if (!dltex->ReadTexels(0, 0, read_width, read_height, image.GetPixels(), image.GetPitch()))
{
ERROR_LOG("Failed to read {}x{} download texture", read_width, read_height);
gpu_backend->RestoreDeviceContext();
return;
}
// no more GPU calls
gpu_backend->RestoreDeviceContext();
Error error;
const std::string path = RegTestHost::GetFrameDumpPath(frame_number);
auto fp = FileSystem::OpenManagedCFile(path.c_str(), "wb", &error);
if (!fp)
{
ERROR_LOG("Can't open file '{}': {}", Path::GetFileName(path), error.GetDescription());
return;
}
System::QueueAsyncTask([path = std::move(path), fp = fp.release(), flip_y = g_gpu_device->UsesLowerLeftOrigin(),
image = std::move(image)]() mutable {
Error error;
if (flip_y)
image.FlipY();
if (image.GetFormat() != ImageFormat::RGBA8)
{
std::optional<Image> convert_image = image.ConvertToRGBA8(&error);
if (!convert_image.has_value())
{
ERROR_LOG("Failed to convert {} screenshot to RGBA8: {}", Image::GetFormatName(image.GetFormat()),
error.GetDescription());
image.Invalidate();
}
else
{
image = std::move(convert_image.value());
}
}
bool result = false;
if (image.IsValid())
{
image.SetAllPixelsOpaque();
result = image.SaveToFile(path.c_str(), fp, Image::DEFAULT_SAVE_QUALITY, &error);
if (!result)
ERROR_LOG("Failed to save screenshot to '{}': '{}'", Path::GetFileName(path), error.GetDescription());
}
std::fclose(fp);
return result;
});
}
void Host::OpenURL(std::string_view url)
@ -781,7 +867,7 @@ bool RegTestHost::SetNewDataRoot(const std::string& filename)
return true;
}
std::string RegTestHost::GetFrameDumpFilename(u32 frame)
std::string RegTestHost::GetFrameDumpPath(u32 frame)
{
return Path::Combine(EmuFolders::DataRoot, fmt::format("frame_{:05d}.png", frame));
}