ShuriZma
/
dolphin
mirror of https://github.com/dolphin-emu/dolphin.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Video: implement color correction to match the NTSC and PAL color spaces (and gamma) that GC and Wii targeted. 

To further increase the accuracy of the post process phase, I've added (scRGB) HDR support, which is necessary
to fully display the PAL and NTSC-J color spaces, and also to improve the quality of post process texture samplings and
do them in linear space instead of gamma space (which is very important when playing at low resolutions).
For SDR, the quality is also slightly increased, at least if any post process runs, as the buffer is now
R10G10B10A2 (on Vulkan, DX11 and DX12) if supported; previously it was R8G8B8A8 but the alpha bits were wasted.

Gamma correction is arguably the most important thing as Dolphin on Windows outputted in "sRGB" (implicitly)
as that's what Windows expects by default, though sRGB gamma is very different from the gamma commonly used
by video standards dating to the pre HDR era (roughly gamma 2.35).

Additionally, the addition of HDR support (which is pretty straight forward and minimal), added support for
our own custom AutoHDR shaders, which would allow us to achieve decent looking HDR in Dolphin games without
having to use SpecialK or Windows 11 AutoHDR. Both of which don't necessarily play nice with older games
with strongly different and simpler lighting. HDR should also be supported in Linux.
Development of my own AutoHDR shader is almost complete and will come next.

This has been carefully tested and there should be no regression in any of the different features that Dolphin
offers, like multisampling, stereo rendering, other post processes, etc etc.

Fixes: https://bugs.dolphin-emu.org/issues/8941

Co-authored-by: EndlesslyFlowering <EndlesslyFlowering@protonmail.com>
Co-authored-by: Dogway <lin_ares@hotmail.com>
This commit is contained in:
Filoppi 2023-06-10 11:48:05 +03:00
parent 0584f8ae2a
commit a2702c6e27
27 changed files with 755 additions and 141 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

86
Data/Sys/Shaders/default_pre_post_process.glsl Normal file
View File

@ -0,0 +1,86 @@
// References:
// https://www.unravel.com.au/understanding-color-spaces
// SMPTE 170M - BT.601 (NTSC-M) -> BT.709
mat3 from_NTSCM = transpose(mat3(
0.939497225737661, 0.0502268452914346, 0.0102759289709032,
0.0177558637510127, 0.965824605885027, 0.0164195303639603,
-0.00162163209967010, -0.00437400622653655, 1.00599563832621));
// ARIB TR-B9 (9300K+27MPCD with chromatic adaptation) (NTSC-J) -> BT.709
mat3 from_NTSCJ = transpose(mat3(
0.823613036967492, -0.0943227111084757, 0.00799341532931119,
0.0289258355537324, 1.02310733489462, 0.00243547111576797,
-0.00569501554980891, 0.0161828357559315, 1.22328453915712));
// EBU - BT.470BG/BT.601 (PAL) -> BT.709
mat3 from_PAL = transpose(mat3(
1.04408168421813, -0.0440816842181253, 0.000000000000000,
0.000000000000000, 1.00000000000000, 0.000000000000000,
0.000000000000000, 0.0118044782106489, 0.988195521789351));
float3 LinearTosRGBGamma(float3 color)
{
float a = 0.055;
for (int i = 0; i < 3; ++i)
{
float x = color[i];
if (x <= 0.0031308)
x = x * 12.92;
else
x = (1.0 + a) * pow(x, 1.0 / 2.4) - a;
color[i] = x;
}
return color;
}
void main()
{
// Note: sampling in gamma space is "wrong" if the source
// and target resolution don't match exactly.
// Fortunately at the moment here they always should but to do this correctly,
// we'd need to sample from 4 pixels, de-apply the gamma from each of these,
// and then do linear sampling on their corrected value.
float4 color = Sample();
// Convert to linear space to do any other kind of operation
color.rgb = pow(color.rgb, game_gamma.xxx);
if (OptionEnabled(correct_color_space))
{
if (game_color_space == 0)
color.rgb = color.rgb * from_NTSCM;
else if (game_color_space == 1)
color.rgb = color.rgb * from_NTSCJ;
else if (game_color_space == 2)
color.rgb = color.rgb * from_PAL;
}
if (OptionEnabled(hdr_output))
{
const float hdr_paper_white = hdr_paper_white_nits / hdr_sdr_white_nits;
color.rgb *= hdr_paper_white;
}
if (OptionEnabled(linear_space_output))
{
// Nothing to do here
}
// Correct the SDR gamma for sRGB (PC/Monitor) or ~2.2 (Common TV gamma)
else if (OptionEnabled(correct_gamma))
{
if (OptionEnabled(sdr_display_gamma_sRGB))
color.rgb = LinearTosRGBGamma(color.rgb);
else
color.rgb = pow(color.rgb, (1.0 / sdr_display_custom_gamma).xxx);
}
// Restore the original gamma without changes
else
{
color.rgb = pow(color.rgb, (1.0 / game_gamma).xxx);
}
SetOutput(color);
}

16
Source/Core/Core/Config/GraphicsSettings.cpp
View File

@ -128,6 +128,22 @@ const Info<bool> GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION{
{System::GFX, "Enhancements", "ArbitraryMipmapDetection"}, true};
const Info<float> GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION_THRESHOLD{
{System::GFX, "Enhancements", "ArbitraryMipmapDetectionThreshold"}, 14.0f};
const Info<bool> GFX_ENHANCE_HDR_OUTPUT{{System::GFX, "Enhancements", "HDROutput"}, false};
// Color.Correction
const Info<bool> GFX_CC_CORRECT_COLOR_SPACE{{System::GFX, "ColorCorrection", "CorrectColorSpace"},
false};
const Info<ColorCorrectionRegion> GFX_CC_GAME_COLOR_SPACE{
{System::GFX, "ColorCorrection", "GameColorSpace"}, ColorCorrectionRegion::SMPTE_NTSCM};
const Info<bool> GFX_CC_CORRECT_GAMMA{{System::GFX, "ColorCorrection", "CorrectGamma"}, false};
const Info<float> GFX_CC_GAME_GAMMA{{System::GFX, "ColorCorrection", "GameGamma"}, 2.35f};
const Info<bool> GFX_CC_SDR_DISPLAY_GAMMA_SRGB{
{System::GFX, "ColorCorrection", "SDRDisplayGammaSRGB"}, true};
const Info<float> GFX_CC_SDR_DISPLAY_CUSTOM_GAMMA{
{System::GFX, "ColorCorrection", "SDRDisplayCustomGamma"}, 2.2f};
const Info<float> GFX_CC_HDR_PAPER_WHITE_NITS{{System::GFX, "ColorCorrection", "HDRPaperWhiteNits"},
200.f};
// Graphics.Stereoscopy

21
Source/Core/Core/Config/GraphicsSettings.h
View File

@ -11,6 +11,7 @@ enum class AspectMode : int;
enum class ShaderCompilationMode : int;
enum class StereoMode : int;
enum class TextureFilteringMode : int;
enum class ColorCorrectionRegion : int;
enum class TriState : int;
namespace Config
@ -105,6 +106,26 @@ extern const Info<bool> GFX_ENHANCE_FORCE_TRUE_COLOR;
extern const Info<bool> GFX_ENHANCE_DISABLE_COPY_FILTER;
extern const Info<bool> GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION;
extern const Info<float> GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION_THRESHOLD;
extern const Info<bool> GFX_ENHANCE_HDR_OUTPUT;
// Color.Correction
static constexpr float GFX_CC_GAME_GAMMA_MIN = 2.2f;
static constexpr float GFX_CC_GAME_GAMMA_MAX = 2.8f;
static constexpr float GFX_CC_DISPLAY_GAMMA_MIN = 2.2f;
static constexpr float GFX_CC_DISPLAY_GAMMA_MAX = 2.4f;
static constexpr float GFX_CC_HDR_PAPER_WHITE_NITS_MIN = 80.f;
static constexpr float GFX_CC_HDR_PAPER_WHITE_NITS_MAX = 400.f;
extern const Info<bool> GFX_CC_CORRECT_COLOR_SPACE;
extern const Info<ColorCorrectionRegion> GFX_CC_GAME_COLOR_SPACE;
extern const Info<bool> GFX_CC_CORRECT_GAMMA;
extern const Info<float> GFX_CC_GAME_GAMMA;
extern const Info<bool> GFX_CC_SDR_DISPLAY_GAMMA_SRGB;
extern const Info<float> GFX_CC_SDR_DISPLAY_CUSTOM_GAMMA;
extern const Info<float> GFX_CC_HDR_PAPER_WHITE_NITS;
// Graphics.Stereoscopy

1
Source/Core/Core/DolphinAnalytics.cpp
View File

@ -380,6 +380,7 @@ void DolphinAnalytics::MakePerGameBuilder()
builder.AddData("cfg-gfx-internal-resolution", g_Config.iEFBScale);
builder.AddData("cfg-gfx-tc-samples", g_Config.iSafeTextureCache_ColorSamples);
builder.AddData("cfg-gfx-stereo-mode", static_cast<int>(g_Config.stereo_mode));
builder.AddData("cfg-gfx-hdr", static_cast<int>(g_Config.bHDR));
builder.AddData("cfg-gfx-per-pixel-lighting", g_Config.bEnablePixelLighting);
builder.AddData("cfg-gfx-shader-compilation-mode", GetShaderCompilationMode(g_Config));
builder.AddData("cfg-gfx-wait-for-shaders", g_Config.bWaitForShadersBeforeStarting);

7
Source/Core/VideoBackends/D3D/D3DBase.cpp
View File

@ -15,6 +15,7 @@
#include "VideoBackends/D3D/D3DState.h"
#include "VideoBackends/D3D/DXTexture.h"
#include "VideoBackends/D3DCommon/D3DCommon.h"
#include "VideoCommon/FramebufferManager.h"
#include "VideoCommon/VideoConfig.h"
namespace DX11
@ -202,12 +203,14 @@ std::vector<u32> GetAAModes(u32 adapter_index)
if (temp_feature_level == D3D_FEATURE_LEVEL_10_0)
return {};
const DXGI_FORMAT target_format =
D3DCommon::GetDXGIFormatForAbstractFormat(FramebufferManager::GetEFBColorFormat(), false);
std::vector<u32> aa_modes;
for (u32 samples = 1; samples <= D3D11_MAX_MULTISAMPLE_SAMPLE_COUNT; ++samples)
{
UINT quality_levels = 0;
if (SUCCEEDED(temp_device->CheckMultisampleQualityLevels(DXGI_FORMAT_R8G8B8A8_UNORM, samples,
&quality_levels)) &&
if (SUCCEEDED(
temp_device->CheckMultisampleQualityLevels(target_format, samples, &quality_levels)) &&
quality_levels > 0)
{
aa_modes.push_back(samples);

3
Source/Core/VideoBackends/D3D/D3DGfx.cpp
View File

@ -175,6 +175,9 @@ void Gfx::OnConfigChanged(u32 bits)
// Quad-buffer changes require swap chain recreation.
if (bits & CONFIG_CHANGE_BIT_STEREO_MODE && m_swap_chain)
m_swap_chain->SetStereo(SwapChain::WantsStereo());
if (bits & CONFIG_CHANGE_BIT_HDR && m_swap_chain)
m_swap_chain->SetHDR(SwapChain::WantsHDR());
}
void Gfx::CheckForSwapChainChanges()

1
Source/Core/VideoBackends/D3D/D3DMain.cpp
View File

@ -114,6 +114,7 @@ void VideoBackend::FillBackendInfo()
g_Config.backend_info.bSupportsSettingObjectNames = true;
g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
g_Config.backend_info.bSupportsDynamicVertexLoader = false;
g_Config.backend_info.bSupportsHDROutput = true;
g_Config.backend_info.Adapters = D3DCommon::GetAdapterNames();
g_Config.backend_info.AAModes = D3D::GetAAModes(g_Config.iAdapter);

2
Source/Core/VideoBackends/D3D/D3DSwapChain.cpp
View File

@ -21,7 +21,7 @@ std::unique_ptr<SwapChain> SwapChain::Create(const WindowSystemInfo& wsi)
{
std::unique_ptr<SwapChain> swap_chain =
std::make_unique<SwapChain>(wsi, D3D::dxgi_factory.Get(), D3D::device.Get());
if (!swap_chain->CreateSwapChain(WantsStereo()))
if (!swap_chain->CreateSwapChain(WantsStereo(), WantsHDR()))
return nullptr;
return swap_chain;

6
Source/Core/VideoBackends/D3D12/D3D12Gfx.cpp
View File

@ -421,6 +421,12 @@ void Gfx::OnConfigChanged(u32 bits)
m_swap_chain->SetStereo(SwapChain::WantsStereo());
}
if (m_swap_chain && bits & CONFIG_CHANGE_BIT_HDR)
{
ExecuteCommandList(true);
m_swap_chain->SetHDR(SwapChain::WantsHDR());
}
// Wipe sampler cache if force texture filtering or anisotropy changes.
if (bits & (CONFIG_CHANGE_BIT_ANISOTROPY | CONFIG_CHANGE_BIT_FORCE_TEXTURE_FILTERING))
{

2
Source/Core/VideoBackends/D3D12/D3D12SwapChain.cpp
View File

@ -22,7 +22,7 @@ std::unique_ptr<SwapChain> SwapChain::Create(const WindowSystemInfo& wsi)
{
std::unique_ptr<SwapChain> swap_chain = std::make_unique<SwapChain>(
wsi, g_dx_context->GetDXGIFactory(), g_dx_context->GetCommandQueue());
if (!swap_chain->CreateSwapChain(WantsStereo()))
if (!swap_chain->CreateSwapChain(WantsStereo(), WantsHDR()))
return nullptr;
return swap_chain;

5
Source/Core/VideoBackends/D3D12/DX12Context.cpp
View File

@ -16,6 +16,7 @@
#include "VideoBackends/D3D12/Common.h"
#include "VideoBackends/D3D12/D3D12StreamBuffer.h"
#include "VideoBackends/D3D12/DescriptorHeapManager.h"
#include "VideoCommon/FramebufferManager.h"
#include "VideoCommon/VideoConfig.h"
namespace DX12
@ -65,11 +66,13 @@ std::vector<u32> DXContext::GetAAModes(u32 adapter_index)
return {};
}
const DXGI_FORMAT target_format =
D3DCommon::GetDXGIFormatForAbstractFormat(FramebufferManager::GetEFBColorFormat(), false);
std::vector<u32> aa_modes;
for (u32 samples = 1; samples < D3D12_MAX_MULTISAMPLE_SAMPLE_COUNT; ++samples)
{
D3D12_FEATURE_DATA_MULTISAMPLE_QUALITY_LEVELS multisample_quality_levels = {};
multisample_quality_levels.Format = DXGI_FORMAT_R8G8B8A8_UNORM;
multisample_quality_levels.Format = target_format;
multisample_quality_levels.SampleCount = samples;
temp_device->CheckFeatureSupport(D3D12_FEATURE_MULTISAMPLE_QUALITY_LEVELS,

1
Source/Core/VideoBackends/D3D12/VideoBackend.cpp
View File

@ -90,6 +90,7 @@ void VideoBackend::FillBackendInfo()
g_Config.backend_info.bSupportsPartialMultisampleResolve = true;
g_Config.backend_info.bSupportsDynamicVertexLoader = true;
g_Config.backend_info.bSupportsVSLinePointExpand = true;
g_Config.backend_info.bSupportsHDROutput = true;
// We can only check texture support once we have a device.
if (g_dx_context)

80
Source/Core/VideoBackends/D3DCommon/SwapChain.cpp
View File

@ -51,13 +51,18 @@ bool SwapChain::WantsStereo()
return g_ActiveConfig.stereo_mode == StereoMode::QuadBuffer;
}
bool SwapChain::WantsHDR()
{
return g_ActiveConfig.bHDR;
}
u32 SwapChain::GetSwapChainFlags() const
{
// This flag is necessary if we want to use a flip-model swapchain without locking the framerate
return m_allow_tearing_supported ? DXGI_SWAP_CHAIN_FLAG_ALLOW_TEARING : 0;
}
bool SwapChain::CreateSwapChain(bool stereo)
bool SwapChain::CreateSwapChain(bool stereo, bool hdr)
{
RECT client_rc;
if (GetClientRect(static_cast<HWND>(m_wsi.render_surface), &client_rc))
@ -66,6 +71,9 @@ bool SwapChain::CreateSwapChain(bool stereo)
m_height = client_rc.bottom - client_rc.top;
}
m_stereo = false;
m_hdr = false;
// Try using the Win8 version if available.
Microsoft::WRL::ComPtr<IDXGIFactory2> dxgi_factory2;
HRESULT hr = m_dxgi_factory.As(&dxgi_factory2);
@ -81,6 +89,7 @@ bool SwapChain::CreateSwapChain(bool stereo)
swap_chain_desc.SampleDesc.Count = 1;
swap_chain_desc.SampleDesc.Quality = 0;
swap_chain_desc.Format = GetDXGIFormatForAbstractFormat(m_texture_format, false);
swap_chain_desc.Scaling = DXGI_SCALING_STRETCH;
swap_chain_desc.SwapEffect = DXGI_SWAP_EFFECT_FLIP_DISCARD;
swap_chain_desc.Stereo = stereo;
@ -108,6 +117,8 @@ bool SwapChain::CreateSwapChain(bool stereo)
// support the newer DXGI interface aren't going to support DX12 anyway.
if (FAILED(hr))
{
hdr = false;
DXGI_SWAP_CHAIN_DESC desc = {};
desc.BufferDesc.Width = m_width;
desc.BufferDesc.Height = m_height;
@ -138,6 +149,37 @@ bool SwapChain::CreateSwapChain(bool stereo)
WARN_LOG_FMT(VIDEO, "MakeWindowAssociation() failed: {}", Common::HRWrap(hr));
m_stereo = stereo;
if (hdr)
{
// Only try to activate HDR here, to avoid failing when creating the swapchain
// (we can't know if the format is supported upfront)
Microsoft::WRL::ComPtr<IDXGISwapChain4> swap_chain4;
hr = m_swap_chain->QueryInterface(IID_PPV_ARGS(&swap_chain4));
if (SUCCEEDED(hr))
{
UINT color_space_support = 0;
// Note that this should succeed even if HDR is not currently engaged on the monitor,
// but it should display fine nonetheless.
// We need to check for DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020 as checking for
// scRGB always returns false (DX bug).
hr = swap_chain4->CheckColorSpaceSupport(DXGI_COLOR_SPACE_RGB_FULL_G2084_NONE_P2020,
&color_space_support);
if (SUCCEEDED(hr) && (color_space_support & DXGI_SWAP_CHAIN_COLOR_SPACE_SUPPORT_FLAG_PRESENT))
{
hr = swap_chain4->ResizeBuffers(SWAP_CHAIN_BUFFER_COUNT, 0, 0,
GetDXGIFormatForAbstractFormat(m_texture_format_hdr, false),
GetSwapChainFlags());
if (SUCCEEDED(hr))
{
hr = swap_chain4->SetColorSpace1(DXGI_COLOR_SPACE_RGB_FULL_G10_NONE_P709);
if (SUCCEEDED(hr))
m_hdr = hdr;
}
}
}
}
if (!CreateSwapChainBuffers())
{
PanicAlertFmt("Failed to create swap chain buffers");
@ -164,12 +206,19 @@ bool SwapChain::ResizeSwapChain()
{
DestroySwapChainBuffers();
HRESULT hr = m_swap_chain->ResizeBuffers(SWAP_CHAIN_BUFFER_COUNT, 0, 0,
GetDXGIFormatForAbstractFormat(m_texture_format, false),
// The swap chain fills up the size of the window if no size is specified
HRESULT hr = m_swap_chain->ResizeBuffers(SWAP_CHAIN_BUFFER_COUNT, 0, 0, DXGI_FORMAT_UNKNOWN,
GetSwapChainFlags());
if (FAILED(hr))
WARN_LOG_FMT(VIDEO, "ResizeBuffers() failed: {}", Common::HRWrap(hr));
Microsoft::WRL::ComPtr<IDXGISwapChain4> swap_chain4;
hr = m_swap_chain->QueryInterface(IID_PPV_ARGS(&swap_chain4));
if (SUCCEEDED(hr))
hr = swap_chain4->SetColorSpace1(m_hdr ? DXGI_COLOR_SPACE_RGB_FULL_G10_NONE_P709 :
DXGI_COLOR_SPACE_RGB_FULL_G22_NONE_P709);
DXGI_SWAP_CHAIN_DESC desc;
if (SUCCEEDED(m_swap_chain->GetDesc(&desc)))
{
@ -186,10 +235,28 @@ void SwapChain::SetStereo(bool stereo)
return;
DestroySwapChain();
if (!CreateSwapChain(stereo))
// Do not try to re-activate HDR here if it had already failed
if (!CreateSwapChain(stereo, m_hdr))
{
PanicAlertFmt("Failed to switch swap chain stereo mode");
CreateSwapChain(false);
CreateSwapChain(false, false);
}
}
void SwapChain::SetHDR(bool hdr)
{
if (m_hdr == hdr)
return;
// NOTE: as an optimization here we could just call "ResizeSwapChain()"
// by adding some code to check if we could change the format to HDR.
DestroySwapChain();
// Do not try to re-activate stereo mode here if it had already failed
if (!CreateSwapChain(m_stereo, hdr))
{
PanicAlertFmt("Failed to switch swap chain SDR/HDR mode");
CreateSwapChain(false, false);
}
}
@ -249,7 +316,8 @@ bool SwapChain::ChangeSurface(void* native_handle)
{
DestroySwapChain();
m_wsi.render_surface = native_handle;
return CreateSwapChain(m_stereo);
// We only keep the swap chain settings (HDR/Stereo) that had successfully applied beofre
return CreateSwapChain(m_stereo, m_hdr);
}
} // namespace D3DCommon

14
Source/Core/VideoBackends/D3DCommon/SwapChain.h
View File

@ -25,8 +25,13 @@ public:
// Returns true if the stereo mode is quad-buffering.
static bool WantsStereo();
static bool WantsHDR();
IDXGISwapChain* GetDXGISwapChain() const { return m_swap_chain.Get(); }
AbstractTextureFormat GetFormat() const { return m_texture_format; }
AbstractTextureFormat GetFormat() const
{
return m_hdr ? m_texture_format_hdr : m_texture_format;
}
u32 GetWidth() const { return m_width; }
u32 GetHeight() const { return m_height; }
@ -43,10 +48,11 @@ public:
bool ChangeSurface(void* native_handle);
bool ResizeSwapChain();
void SetStereo(bool stereo);
void SetHDR(bool hdr);
protected:
u32 GetSwapChainFlags() const;
bool CreateSwapChain(bool stereo);
bool CreateSwapChain(bool stereo = false, bool hdr = false);
void DestroySwapChain();
virtual bool CreateSwapChainBuffers() = 0;
@ -56,12 +62,14 @@ protected:
Microsoft::WRL::ComPtr<IDXGIFactory> m_dxgi_factory;
Microsoft::WRL::ComPtr<IDXGISwapChain> m_swap_chain;
Microsoft::WRL::ComPtr<IUnknown> m_d3d_device;
AbstractTextureFormat m_texture_format = AbstractTextureFormat::RGBA8;
const AbstractTextureFormat m_texture_format = AbstractTextureFormat::RGB10_A2;
const AbstractTextureFormat m_texture_format_hdr = AbstractTextureFormat::RGBA16F;
u32 m_width = 1;
u32 m_height = 1;
bool m_stereo = false;
bool m_hdr = false;
bool m_allow_tearing_supported = false;
bool m_has_fullscreen = false;
bool m_fullscreen_request = false;

2
Source/Core/VideoBackends/Vulkan/VKGfx.cpp
View File

@ -399,7 +399,7 @@ void VKGfx::OnConfigChanged(u32 bits)
}
// For quad-buffered stereo we need to change the layer count, so recreate the swap chain.
if (m_swap_chain && bits & CONFIG_CHANGE_BIT_STEREO_MODE)
if (m_swap_chain && (bits & CONFIG_CHANGE_BIT_STEREO_MODE) || (bits & CONFIG_CHANGE_BIT_HDR))
{
ExecuteCommandBuffer(false, true);
m_swap_chain->RecreateSwapChain();

55
Source/Core/VideoBackends/Vulkan/VKSwapChain.cpp
View File

@ -154,7 +154,7 @@ bool SwapChain::SelectSurfaceFormat()
&format_count, surface_formats.data());
ASSERT(res == VK_SUCCESS);
// If there is a single undefined surface format, the device doesn't care, so we'll just use RGBA
// If there is a single undefined surface format, the device doesn't care, so we'll just use RGBA8
if (surface_formats[0].format == VK_FORMAT_UNDEFINED)
{
m_surface_format.format = VK_FORMAT_R8G8B8A8_UNORM;
@ -162,22 +162,61 @@ bool SwapChain::SelectSurfaceFormat()
return true;
}
// Try to find a suitable format.
const VkSurfaceFormatKHR* surface_format_RGBA8 = nullptr;
const VkSurfaceFormatKHR* surface_format_BGRA8 = nullptr;
const VkSurfaceFormatKHR* surface_format_RGB10_A2 = nullptr;
const VkSurfaceFormatKHR* surface_format_RGBA16F_scRGB = nullptr;
// Try to find all suitable formats.
for (const VkSurfaceFormatKHR& surface_format : surface_formats)
{
// Some drivers seem to return a SRGB format here (Intel Mesa).
// This results in gamma correction when presenting to the screen, which we don't want.
// Use a linear format instead, if this is the case.
// Some drivers seem to return a RGBA8 SRGB format here (Intel Mesa).
// Some other drivers return both a RGBA8 SRGB and UNORM formats (Nvidia).
// This results in gamma correction when presenting to the screen, which we don't want,
// because we already apply gamma ourselves, and we might not use sRGB gamma.
// Force using a linear format instead, if this is the case.
VkFormat format = VKTexture::GetLinearFormat(surface_format.format);
if (format == VK_FORMAT_R8G8B8A8_UNORM)
surface_format_RGBA8 = &surface_format;
else if (format == VK_FORMAT_B8G8R8A8_UNORM)
surface_format_BGRA8 = &surface_format;
else if (format == VK_FORMAT_A2B10G10R10_UNORM_PACK32 &&
surface_format.colorSpace == VK_COLOR_SPACE_SRGB_NONLINEAR_KHR)
surface_format_RGB10_A2 = &surface_format;
else if (format == VK_FORMAT_R16G16B16A16_SFLOAT &&
surface_format.colorSpace == VK_COLOR_SPACE_EXTENDED_SRGB_LINEAR_EXT)
surface_format_RGBA16F_scRGB = &surface_format;
else
continue;
}
const VkSurfaceFormatKHR* surface_format = nullptr;
// Pick the best format.
// "g_ActiveConfig" might not have been been updated yet.
if (g_Config.bHDR && surface_format_RGBA16F_scRGB)
surface_format = surface_format_RGBA16F_scRGB;
else if (surface_format_RGB10_A2)
surface_format = surface_format_RGB10_A2;
else if (surface_format_RGBA8)
surface_format = surface_format_RGBA8;
else if (surface_format_BGRA8)
surface_format = surface_format_BGRA8;
if (surface_format)
{
const VkFormat format = VKTexture::GetLinearFormat(surface_format->format);
if (format == VK_FORMAT_R8G8B8A8_UNORM)
m_texture_format = AbstractTextureFormat::RGBA8;
else if (format == VK_FORMAT_B8G8R8A8_UNORM)
m_texture_format = AbstractTextureFormat::BGRA8;
else
continue;
else if (format == VK_FORMAT_A2B10G10R10_UNORM_PACK32)
m_texture_format = AbstractTextureFormat::RGB10_A2;
else if (format == VK_FORMAT_R16G16B16A16_SFLOAT)
m_texture_format = AbstractTextureFormat::RGBA16F;
m_surface_format.format = format;
m_surface_format.colorSpace = surface_format.colorSpace;
m_surface_format.colorSpace = surface_format->colorSpace;
return true;
}

1
Source/Core/VideoBackends/Vulkan/VulkanContext.cpp
View File

@ -382,6 +382,7 @@ void VulkanContext::PopulateBackendInfo(VideoConfig* config)
config->backend_info.bSupportsPartialMultisampleResolve = true; // Assumed support.
config->backend_info.bSupportsDynamicVertexLoader = true; // Assumed support.
config->backend_info.bSupportsVSLinePointExpand = true; // Assumed support.
config->backend_info.bSupportsHDROutput = true; // Assumed support.
}
void VulkanContext::PopulateBackendInfoAdapters(VideoConfig* config, const GPUList& gpu_list)

2
Source/Core/VideoCommon/AbstractGfx.cpp
View File

@ -177,5 +177,5 @@ bool AbstractGfx::UseGeometryShaderForUI() const
// OpenGL doesn't render to a 2-layer backbuffer like D3D/Vulkan for quad-buffered stereo,
// instead drawing twice and the eye selected by glDrawBuffer() (see Presenter::RenderXFBToScreen)
return g_ActiveConfig.stereo_mode == StereoMode::QuadBuffer &&
g_ActiveConfig.backend_info.api_type != APIType::OpenGL;
!g_ActiveConfig.backend_info.bUsesExplictQuadBuffering;
}

4
Source/Core/VideoCommon/AbstractGfx.h
View File

@ -159,8 +159,8 @@ public:
// Called when the configuration changes, and backend structures need to be updated.
virtual void OnConfigChanged(u32 changed_bits);
// Returns true if a layer-expanding geometry shader should be used when rendering the user
// interface and final XFB.
// Returns true if a layer-expanding geometry shader should be used when rendering
// the user interface on the output buffer.
bool UseGeometryShaderForUI() const;
// Returns info about the main surface (aka backbuffer)

11
Source/Core/VideoCommon/FramebufferShaderGen.cpp
View File

@ -672,7 +672,7 @@ std::string GenerateImGuiVertexShader()
return code.GetBuffer();
}
std::string GenerateImGuiPixelShader()
std::string GenerateImGuiPixelShader(bool linear_space_output)
{
ShaderCode code;
EmitSamplerDeclarations(code, 0, 1, false);
@ -680,8 +680,13 @@ std::string GenerateImGuiPixelShader()
code.Write("{{\n"
" ocol0 = ");
EmitSampleTexture(code, 0, "float3(v_tex0.xy, 0.0)");
code.Write(" * v_col0;\n"
"}}\n");
// We approximate to gamma 2.2 instead of sRGB as it barely matters for this case.
// Note that if HDR is enabled, ideally we should multiply by
// the paper white brightness for readability.
if (linear_space_output)
code.Write(" * pow(v_col0, float4(2.2f, 2.2f, 2.2f, 1.0f));\n}}\n");
else
code.Write(" * v_col0;\n}}\n");
return code.GetBuffer();
}

2
Source/Core/VideoCommon/FramebufferShaderGen.h
View File

@ -24,6 +24,6 @@ std::string GenerateFormatConversionShader(EFBReinterpretType convtype, u32 samp
std::string GenerateTextureReinterpretShader(TextureFormat from_format, TextureFormat to_format);
std::string GenerateEFBRestorePixelShader();
std::string GenerateImGuiVertexShader();
std::string GenerateImGuiPixelShader();
std::string GenerateImGuiPixelShader(bool linear_space_output = false);
} // namespace FramebufferShaderGen

6
Source/Core/VideoCommon/OnScreenUI.cpp
View File

@ -120,11 +120,15 @@ bool OnScreenUI::RecompileImGuiPipeline()
return true;
}
const bool linear_space_output =
g_presenter->GetBackbufferFormat() == AbstractTextureFormat::RGBA16F;
std::unique_ptr<AbstractShader> vertex_shader = g_gfx->CreateShaderFromSource(
ShaderStage::Vertex, FramebufferShaderGen::GenerateImGuiVertexShader(),
"ImGui vertex shader");
std::unique_ptr<AbstractShader> pixel_shader = g_gfx->CreateShaderFromSource(
ShaderStage::Pixel, FramebufferShaderGen::GenerateImGuiPixelShader(), "ImGui pixel shader");
ShaderStage::Pixel, FramebufferShaderGen::GenerateImGuiPixelShader(linear_space_output),
"ImGui pixel shader");
if (!vertex_shader || !pixel_shader)
{
PanicAlertFmt("Failed to compile ImGui shaders");

491
Source/Core/VideoCommon/PostProcessing.cpp
View File

@ -33,7 +33,32 @@
namespace VideoCommon
{
static const char s_default_shader[] = "void main() { SetOutput(Sample()); }\n";
static const char s_empty_pixel_shader[] = "void main() { SetOutput(Sample()); }\n";
static const char s_default_pixel_shader_name[] = "default_pre_post_process";
// Keep the highest quality possible to avoid losing quality on subtle gamma conversions.
// RGBA16F should have enough quality even if we store colors in gamma space on it.
static const AbstractTextureFormat s_intermediary_buffer_format = AbstractTextureFormat::RGBA16F;
bool LoadShaderFromFile(const std::string& shader, const std::string& sub_dir,
std::string& out_code)
{
std::string path = File::GetUserPath(D_SHADERS_IDX) + sub_dir + shader + ".glsl";
if (!File::Exists(path))
{
// Fallback to shared user dir
path = File::GetSysDirectory() + SHADERS_DIR DIR_SEP + sub_dir + shader + ".glsl";
}
if (!File::ReadFileToString(path, out_code))
{
out_code = "";
ERROR_LOG_FMT(VIDEO, "Post-processing shader not found: {}", path);
return false;
}
return true;
}
PostProcessingConfiguration::PostProcessingConfiguration() = default;
@ -60,24 +85,16 @@ void PostProcessingConfiguration::LoadShader(const std::string& shader)
sub_dir = PASSIVE_DIR DIR_SEP;
}
// loading shader code
std::string code;
std::string path = File::GetUserPath(D_SHADERS_IDX) + sub_dir + shader + ".glsl";
if (!File::Exists(path))
if (!LoadShaderFromFile(shader, sub_dir, code))
{
// Fallback to shared user dir
path = File::GetSysDirectory() + SHADERS_DIR DIR_SEP + sub_dir + shader + ".glsl";
}
if (!File::ReadFileToString(path, code))
{
ERROR_LOG_FMT(VIDEO, "Post-processing shader not found: {}", path);
LoadDefaultShader();
return;
}
LoadOptions(code);
// Note that this will build the shaders with the custom options values users
// might have set in the settings
LoadOptionsConfiguration();
m_current_shader_code = code;
}
@ -86,7 +103,8 @@ void PostProcessingConfiguration::LoadDefaultShader()
{
m_options.clear();
m_any_options_dirty = false;
m_current_shader_code = s_default_shader;
m_current_shader = "";
m_current_shader_code = s_empty_pixel_shader;
}
void PostProcessingConfiguration::LoadOptions(const std::string& code)
@ -242,6 +260,7 @@ void PostProcessingConfiguration::LoadOptionsConfiguration()
ini.Load(File::GetUserPath(F_DOLPHINCONFIG_IDX));
std::string section = m_current_shader + "-options";
// We already expect all the options to be marked as "dirty" when we reach here
for (auto& it : m_options)
{
switch (it.second.m_type)
@ -375,6 +394,8 @@ static std::vector<std::string> GetShaders(const std::string& sub_dir = "")
{
std::string name;
SplitPath(path, nullptr, &name, nullptr);
if (name == s_default_pixel_shader_name)
continue;
result.push_back(name);
}
return result;
@ -409,8 +430,15 @@ bool PostProcessing::Initialize(AbstractTextureFormat format)
void PostProcessing::RecompileShader()
{
// Note: for simplicity we already recompile all the shaders
// and pipelines even if there might not be need to.
m_default_pipeline.reset();
m_pipeline.reset();
m_default_pixel_shader.reset();
m_pixel_shader.reset();
m_default_vertex_shader.reset();
m_vertex_shader.reset();
if (!CompilePixelShader())
return;
if (!CompileVertexShader())
@ -421,10 +449,27 @@ void PostProcessing::RecompileShader()
void PostProcessing::RecompilePipeline()
{
m_default_pipeline.reset();
m_pipeline.reset();
CompilePipeline();
}
bool PostProcessing::IsColorCorrectionActive() const
{
// We can skip the color correction pass if none of these settings are on
// (it might have still helped with gamma correct sampling, but it's not worth running it).
return g_ActiveConfig.color_correction.bCorrectColorSpace ||
g_ActiveConfig.color_correction.bCorrectGamma ||
m_framebuffer_format == AbstractTextureFormat::RGBA16F;
}
bool PostProcessing::NeedsIntermediaryBuffer() const
{
// If we have no user selected post process shader,
// there's no point in having an intermediary buffer doing nothing.
return !m_config.GetShader().empty();
}
void PostProcessing::BlitFromTexture(const MathUtil::Rectangle<int>& dst,
const MathUtil::Rectangle<int>& src,
const AbstractTexture* src_tex, int src_layer)
@ -435,69 +480,183 @@ void PostProcessing::BlitFromTexture(const MathUtil::Rectangle<int>& dst,
RecompilePipeline();
}
if (!m_pipeline)
return;
// By default all source layers will be copied into the respective target layers
const bool copy_all_layers = src_layer < 0;
src_layer = std::max(src_layer, 0);
FillUniformBuffer(src, src_tex, src_layer);
g_vertex_manager->UploadUtilityUniforms(m_uniform_staging_buffer.data(),
static_cast<u32>(m_uniform_staging_buffer.size()));
g_gfx->SetViewportAndScissor(
g_gfx->ConvertFramebufferRectangle(dst, g_gfx->GetCurrentFramebuffer()));
g_gfx->SetPipeline(m_pipeline.get());
g_gfx->SetTexture(0, src_tex);
MathUtil::Rectangle<int> src_rect = src;
g_gfx->SetSamplerState(0, RenderState::GetLinearSamplerState());
g_gfx->Draw(0, 3);
g_gfx->SetTexture(0, src_tex);
const bool is_color_correction_active = IsColorCorrectionActive();
const bool needs_intermediary_buffer = NeedsIntermediaryBuffer();
const AbstractPipeline* final_pipeline = m_pipeline.get();
std::vector<u8>* uniform_staging_buffer = &m_default_uniform_staging_buffer;
bool default_uniform_staging_buffer = true;
// Intermediary pass.
// We draw to a high quality intermediary texture for two reasons:
// -Keep quality for gamma and gamut conversions, and HDR output
// (low bit depths lose too much quality with gamma conversions)
// -We make a texture of the exact same res as the source one,
// because all the post process shaders we already had assume that
// the source texture size (EFB) is different from the swap chain
// texture size (which matches the window size).
if (m_default_pipeline && is_color_correction_active && needs_intermediary_buffer)
{
AbstractFramebuffer* const previous_framebuffer = g_gfx->GetCurrentFramebuffer();
// We keep the min number of layers as the render target,
// as in case of OpenGL, the source FBX will have two layers,
// but we will render onto two separate frame buffers (one by one),
// so it would be a waste to allocate two layers (see "bUsesExplictQuadBuffering").
const u32 target_layers = copy_all_layers ? src_tex->GetLayers() : 1;
if (!m_intermediary_frame_buffer || !m_intermediary_color_texture ||
m_intermediary_color_texture.get()->GetWidth() != static_cast<u32>(src_rect.GetWidth()) ||
m_intermediary_color_texture.get()->GetHeight() != static_cast<u32>(src_rect.GetHeight()) ||
m_intermediary_color_texture.get()->GetLayers() != target_layers)
{
const TextureConfig intermediary_color_texture_config(
src_rect.GetWidth(), src_rect.GetHeight(), 1, target_layers, src_tex->GetSamples(),
s_intermediary_buffer_format, AbstractTextureFlag_RenderTarget);
m_intermediary_color_texture = g_gfx->CreateTexture(intermediary_color_texture_config,
"Intermediary post process texture");
m_intermediary_frame_buffer =
g_gfx->CreateFramebuffer(m_intermediary_color_texture.get(), nullptr);
}
g_gfx->SetFramebuffer(m_intermediary_frame_buffer.get());
FillUniformBuffer(src_rect, src_tex, src_layer, g_gfx->GetCurrentFramebuffer()->GetRect(),
g_presenter->GetTargetRectangle(), uniform_staging_buffer->data(),
!default_uniform_staging_buffer);
g_vertex_manager->UploadUtilityUniforms(uniform_staging_buffer->data(),
static_cast<u32>(uniform_staging_buffer->size()));
g_gfx->SetViewportAndScissor(g_gfx->ConvertFramebufferRectangle(
m_intermediary_color_texture->GetRect(), m_intermediary_frame_buffer.get()));
g_gfx->SetPipeline(m_default_pipeline.get());
g_gfx->Draw(0, 3);
g_gfx->SetFramebuffer(previous_framebuffer);
src_rect = m_intermediary_color_texture->GetRect();
src_tex = m_intermediary_color_texture.get();
g_gfx->SetTexture(0, src_tex);
// The "m_intermediary_color_texture" has already copied
// from the specified source layer onto its first one.
// If we query for a layer that the source texture doesn't have,
// it will fall back on the first one anyway.
src_layer = 0;
uniform_staging_buffer = &m_uniform_staging_buffer;
default_uniform_staging_buffer = false;
}
else
{
// If we have no custom user shader selected, and color correction
// is active, directly run the fixed pipeline shader instead of
// doing two passes, with the second one doing nothing useful.
if (m_default_pipeline && is_color_correction_active)
{
final_pipeline = m_default_pipeline.get();
}
else
{
uniform_staging_buffer = &m_uniform_staging_buffer;
default_uniform_staging_buffer = false;
}
m_intermediary_frame_buffer.release();
m_intermediary_color_texture.release();
}
// TODO: ideally we'd do the user selected post process pass in the intermediary buffer in linear
// space (instead of gamma space), so the shaders could act more accurately (and sample in linear
// space), though that would break the look of some of current post processes we have, and thus is
// better avoided for now.
// Final pass, either a user selected shader or the default (fixed) shader.
if (final_pipeline)
{
FillUniformBuffer(src_rect, src_tex, src_layer, g_gfx->GetCurrentFramebuffer()->GetRect(),
g_presenter->GetTargetRectangle(), uniform_staging_buffer->data(),
!default_uniform_staging_buffer);
g_vertex_manager->UploadUtilityUniforms(uniform_staging_buffer->data(),
static_cast<u32>(uniform_staging_buffer->size()));
g_gfx->SetViewportAndScissor(
g_gfx->ConvertFramebufferRectangle(dst, g_gfx->GetCurrentFramebuffer()));
g_gfx->SetPipeline(final_pipeline);
g_gfx->Draw(0, 3);
}
}
std::string PostProcessing::GetUniformBufferHeader() const
std::string PostProcessing::GetUniformBufferHeader(bool user_post_process) const
{
std::ostringstream ss;
u32 unused_counter = 1;
ss << "UBO_BINDING(std140, 1) uniform PSBlock {\n";
// Builtin uniforms
ss << " float4 resolution;\n";
// Builtin uniforms:
ss << " float4 resolution;\n"; // Source resolution
ss << " float4 target_resolution;\n";
ss << " float4 window_resolution;\n";
// How many horizontal and vertical stereo views do we have? (set to 1 when we use layers instead)
ss << " int2 stereo_views;\n";
ss << " float4 src_rect;\n";
// The first (but not necessarily only) source layer we target
ss << " int src_layer;\n";
ss << " uint time;\n";
for (u32 i = 0; i < 2; i++)
ss << " uint ubo_align_" << unused_counter++ << "_;\n";
ss << "\n";
// Custom options/uniforms
for (const auto& it : m_config.GetOptions())
ss << " int correct_color_space;\n";
ss << " int game_color_space;\n";
ss << " int correct_gamma;\n";
ss << " float game_gamma;\n";
ss << " int sdr_display_gamma_sRGB;\n";
ss << " float sdr_display_custom_gamma;\n";
ss << " int linear_space_output;\n";
ss << " int hdr_output;\n";
ss << " float hdr_paper_white_nits;\n";
ss << " float hdr_sdr_white_nits;\n";
if (user_post_process)
{
if (it.second.m_type == PostProcessingConfiguration::ConfigurationOption::OptionType::Bool)
ss << "\n";
// Custom options/uniforms
for (const auto& it : m_config.GetOptions())
{
ss << fmt::format(" int {};\n", it.first);
for (u32 i = 0; i < 3; i++)
ss << " int ubo_align_" << unused_counter++ << "_;\n";
}
else if (it.second.m_type ==
PostProcessingConfiguration::ConfigurationOption::OptionType::Integer)
{
u32 count = static_cast<u32>(it.second.m_integer_values.size());
if (count == 1)
if (it.second.m_type == PostProcessingConfiguration::ConfigurationOption::OptionType::Bool)
{
ss << fmt::format(" int {};\n", it.first);
else
ss << fmt::format(" int{} {};\n", count, it.first);
for (u32 i = 0; i < 3; i++)
ss << " int ubo_align_" << unused_counter++ << "_;\n";
}
else if (it.second.m_type ==
PostProcessingConfiguration::ConfigurationOption::OptionType::Integer)
{
u32 count = static_cast<u32>(it.second.m_integer_values.size());
if (count == 1)
ss << fmt::format(" int {};\n", it.first);
else
ss << fmt::format(" int{} {};\n", count, it.first);
for (u32 i = count; i < 4; i++)
ss << " int ubo_align_" << unused_counter++ << "_;\n";
}
else if (it.second.m_type ==
PostProcessingConfiguration::ConfigurationOption::OptionType::Float)
{
u32 count = static_cast<u32>(it.second.m_float_values.size());
if (count == 1)
ss << fmt::format(" float {};\n", it.first);
else
ss << fmt::format(" float{} {};\n", count, it.first);
for (u32 i = count; i < 4; i++)
ss << " int ubo_align_" << unused_counter++ << "_;\n";
}
else if (it.second.m_type ==
PostProcessingConfiguration::ConfigurationOption::OptionType::Float)
{
u32 count = static_cast<u32>(it.second.m_float_values.size());
if (count == 1)
ss << fmt::format(" float {};\n", it.first);
else
ss << fmt::format(" float{} {};\n", count, it.first);
for (u32 i = count; i < 4; i++)
ss << " float ubo_align_" << unused_counter++ << "_;\n";
for (u32 i = count; i < 4; i++)
ss << " float ubo_align_" << unused_counter++ << "_;\n";
}
}
}
@ -505,11 +664,12 @@ std::string PostProcessing::GetUniformBufferHeader() const
return ss.str();
}
std::string PostProcessing::GetHeader() const
std::string PostProcessing::GetHeader(bool user_post_process) const
{
std::ostringstream ss;
ss << GetUniformBufferHeader();
ss << GetUniformBufferHeader(user_post_process);
ss << "SAMPLER_BINDING(0) uniform sampler2DArray samp0;\n";
ss << "SAMPLER_BINDING(1) uniform sampler2DArray samp1;\n";
if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
{
@ -530,6 +690,16 @@ float4 SampleLocation(float2 location) { return texture(samp0, float3(location,
float4 SampleLayer(int layer) { return texture(samp0, float3(v_tex0.xy, float(layer))); }
#define SampleOffset(offset) textureOffset(samp0, v_tex0, offset)
float2 GetTargetResolution()
{
return target_resolution.xy;
}
float2 GetInvTargetResolution()
{
return target_resolution.zw;
}
float2 GetWindowResolution()
{
return window_resolution.xy;
@ -585,7 +755,9 @@ std::string PostProcessing::GetFooter() const
bool PostProcessing::CompileVertexShader()
{
std::ostringstream ss;
ss << GetUniformBufferHeader();
// We never need the user selected post process custom uniforms in the vertex shader
const bool user_post_process = false;
ss << GetUniformBufferHeader(user_post_process);
if (g_ActiveConfig.backend_info.bSupportsGeometryShaders)
{
@ -605,16 +777,28 @@ bool PostProcessing::CompileVertexShader()
ss << " opos = float4(v_tex0.xy * float2(2.0f, -2.0f) + float2(-1.0f, 1.0f), 0.0f, 1.0f);\n";
ss << " v_tex0 = float3(src_rect.xy + (src_rect.zw * v_tex0.xy), float(src_layer));\n";
// Vulkan Y needs to be inverted on every pass
if (g_ActiveConfig.backend_info.api_type == APIType::Vulkan)
ss << " opos.y = -opos.y;\n";
std::string s2 = ss.str();
s2 += "}\n";
m_default_vertex_shader = g_gfx->CreateShaderFromSource(ShaderStage::Vertex, s2,
"Default post-processing vertex shader");
// OpenGL Y needs to be inverted once only (in the last pass)
if (g_ActiveConfig.backend_info.api_type == APIType::OpenGL)
ss << " opos.y = -opos.y;\n";
ss << "}\n";
m_vertex_shader =
g_gfx->CreateShaderFromSource(ShaderStage::Vertex, ss.str(), "Post-processing vertex shader");
if (!m_vertex_shader)
if (!m_default_vertex_shader || !m_vertex_shader)
{
PanicAlertFmt("Failed to compile post-processing vertex shader");
m_default_vertex_shader.reset();
m_vertex_shader.reset();
return false;
}
@ -623,44 +807,86 @@ bool PostProcessing::CompileVertexShader()
struct BuiltinUniforms
{
float resolution[4];
float window_resolution[4];
float src_rect[4];
// bools need to be represented as "s32"
std::array<float, 4> source_resolution;
std::array<float, 4> target_resolution;
std::array<float, 4> window_resolution;
std::array<float, 4> stereo_views;
std::array<float, 4> src_rect;
s32 src_layer;
u32 time;
u32 padding[2];
s32 correct_color_space;
s32 game_color_space;
s32 correct_gamma;
float game_gamma;
s32 sdr_display_gamma_sRGB;
float sdr_display_custom_gamma;
s32 linear_space_output;
s32 hdr_output;
float hdr_paper_white_nits;
float hdr_sdr_white_nits;
};
size_t PostProcessing::CalculateUniformsSize() const
size_t PostProcessing::CalculateUniformsSize(bool user_post_process) const
{
// Allocate a vec4 for each uniform to simplify allocation.
return sizeof(BuiltinUniforms) + m_config.GetOptions().size() * sizeof(float) * 4;
return sizeof(BuiltinUniforms) +
(user_post_process ? m_config.GetOptions().size() : 0) * sizeof(float) * 4;
}
void PostProcessing::FillUniformBuffer(const MathUtil::Rectangle<int>& src,
const AbstractTexture* src_tex, int src_layer)
const AbstractTexture* src_tex, int src_layer,
const MathUtil::Rectangle<int>& dst,
const MathUtil::Rectangle<int>& wnd, u8* buffer,
bool user_post_process)
{
const auto& window_rect = g_presenter->GetTargetRectangle();
const float rcp_src_width = 1.0f / src_tex->GetWidth();
const float rcp_src_height = 1.0f / src_tex->GetHeight();
BuiltinUniforms builtin_uniforms = {
{static_cast<float>(src_tex->GetWidth()), static_cast<float>(src_tex->GetHeight()),
rcp_src_width, rcp_src_height},
{static_cast<float>(window_rect.GetWidth()), static_cast<float>(window_rect.GetHeight()),
1.0f / static_cast<float>(window_rect.GetWidth()),
1.0f / static_cast<float>(window_rect.GetHeight())},
{static_cast<float>(src.left) * rcp_src_width, static_cast<float>(src.top) * rcp_src_height,
static_cast<float>(src.GetWidth()) * rcp_src_width,
static_cast<float>(src.GetHeight()) * rcp_src_height},
static_cast<s32>(src_layer),
static_cast<u32>(m_timer.ElapsedMs()),
};
u8* buf = m_uniform_staging_buffer.data();
std::memcpy(buf, &builtin_uniforms, sizeof(builtin_uniforms));
buf += sizeof(builtin_uniforms);
BuiltinUniforms builtin_uniforms;
builtin_uniforms.source_resolution = {static_cast<float>(src_tex->GetWidth()),
static_cast<float>(src_tex->GetHeight()), rcp_src_width,
rcp_src_height};
builtin_uniforms.target_resolution = {
static_cast<float>(dst.GetWidth()), static_cast<float>(dst.GetHeight()),
1.0f / static_cast<float>(dst.GetWidth()), 1.0f / static_cast<float>(dst.GetHeight())};
builtin_uniforms.window_resolution = {
static_cast<float>(wnd.GetWidth()), static_cast<float>(wnd.GetHeight()),
1.0f / static_cast<float>(wnd.GetWidth()), 1.0f / static_cast<float>(wnd.GetHeight())};
builtin_uniforms.src_rect = {static_cast<float>(src.left) * rcp_src_width,
static_cast<float>(src.top) * rcp_src_height,
static_cast<float>(src.GetWidth()) * rcp_src_width,
static_cast<float>(src.GetHeight()) * rcp_src_height};
builtin_uniforms.src_layer = static_cast<s32>(src_layer);
builtin_uniforms.time = static_cast<u32>(m_timer.ElapsedMs());
for (const auto& it : m_config.GetOptions())
// Color correction related uniforms.
// These are mainly used by the "m_default_pixel_shader",
// but should also be accessible to all other shaders.
builtin_uniforms.correct_color_space = g_ActiveConfig.color_correction.bCorrectColorSpace;
builtin_uniforms.game_color_space =
static_cast<int>(g_ActiveConfig.color_correction.game_color_space);
builtin_uniforms.correct_gamma = g_ActiveConfig.color_correction.bCorrectGamma;
builtin_uniforms.game_gamma = g_ActiveConfig.color_correction.fGameGamma;
builtin_uniforms.sdr_display_gamma_sRGB = g_ActiveConfig.color_correction.bSDRDisplayGammaSRGB;
builtin_uniforms.sdr_display_custom_gamma =
g_ActiveConfig.color_correction.fSDRDisplayCustomGamma;
// scRGB (RGBA16F) expects linear values as opposed to sRGB gamma
builtin_uniforms.linear_space_output = m_framebuffer_format == AbstractTextureFormat::RGBA16F;
// Implies ouput values can be beyond the 0-1 range
builtin_uniforms.hdr_output = m_framebuffer_format == AbstractTextureFormat::RGBA16F;
builtin_uniforms.hdr_paper_white_nits = g_ActiveConfig.color_correction.fHDRPaperWhiteNits;
// A value of 1 1 1 usually matches 80 nits in HDR
builtin_uniforms.hdr_sdr_white_nits = 80.f;
std::memcpy(buffer, &builtin_uniforms, sizeof(builtin_uniforms));
buffer += sizeof(builtin_uniforms);
if (!user_post_process)
return;
for (auto& it : m_config.GetOptions())
{
union
{
@ -688,53 +914,116 @@ void PostProcessing::FillUniformBuffer(const MathUtil::Rectangle<int>& src,
break;
}
std::memcpy(buf, &value, sizeof(value));
buf += sizeof(value);
it.second.m_dirty = false;
std::memcpy(buffer, &value, sizeof(value));
buffer += sizeof(value);
}
m_config.SetDirty(false);
}
bool PostProcessing::CompilePixelShader()
{
m_pipeline.reset();
m_default_pixel_shader.reset();
m_pixel_shader.reset();
// Generate GLSL and compile the new shader.
// Generate GLSL and compile the new shaders:
std::string default_pixel_shader_code;
if (LoadShaderFromFile(s_default_pixel_shader_name, "", default_pixel_shader_code))
{
m_default_pixel_shader = g_gfx->CreateShaderFromSource(
ShaderStage::Pixel, GetHeader(false) + default_pixel_shader_code + GetFooter(),
"Default post-processing pixel shader");
// We continue even if all of this failed, it doesn't matter
m_default_uniform_staging_buffer.resize(CalculateUniformsSize(false));
}
else
{
m_default_uniform_staging_buffer.resize(0);
}
m_config.LoadShader(g_ActiveConfig.sPostProcessingShader);
m_pixel_shader = g_gfx->CreateShaderFromSource(
ShaderStage::Pixel, GetHeader() + m_config.GetShaderCode() + GetFooter(),
fmt::format("Post-processing pixel shader: {}", m_config.GetShader()));
ShaderStage::Pixel, GetHeader(true) + m_config.GetShaderCode() + GetFooter(),
fmt::format("User post-processing pixel shader: {}", m_config.GetShader()));
if (!m_pixel_shader)
{
PanicAlertFmt("Failed to compile post-processing shader {}", m_config.GetShader());
PanicAlertFmt("Failed to compile user post-processing shader {}", m_config.GetShader());
// Use default shader.
m_config.LoadDefaultShader();
m_pixel_shader = g_gfx->CreateShaderFromSource(
ShaderStage::Pixel, GetHeader() + m_config.GetShaderCode() + GetFooter(),
"Default post-processing pixel shader");
ShaderStage::Pixel, GetHeader(true) + m_config.GetShaderCode() + GetFooter(),
"Default user post-processing pixel shader");
if (!m_pixel_shader)
{
m_uniform_staging_buffer.resize(0);
return false;
}
}
m_uniform_staging_buffer.resize(CalculateUniformsSize());
m_uniform_staging_buffer.resize(CalculateUniformsSize(true));
return true;
}
bool UseGeometryShaderForPostProcess(bool is_intermediary_buffer)
{
// We only return true on stereo modes that need to copy
// both source texture layers into the target texture layers.
// Any other case is handled manually with multiple copies, thus
// it doesn't need a geom shader.
switch (g_ActiveConfig.stereo_mode)
{
case StereoMode::QuadBuffer:
return !g_ActiveConfig.backend_info.bUsesExplictQuadBuffering;
case StereoMode::Anaglyph:
case StereoMode::Passive:
return is_intermediary_buffer;
case StereoMode::SBS:
case StereoMode::TAB:
case StereoMode::Off:
default:
return false;
}
}
bool PostProcessing::CompilePipeline()
{
// Not needed. Some backends don't like making pipelines with no targets,
// and in any case, we don't need to render anything if that happened.
if (m_framebuffer_format == AbstractTextureFormat::Undefined)
return true; // Not needed (some backends don't like making pipelines with no targets)
return true;
// If this is true, the "m_default_pipeline" won't be the only one that runs
const bool needs_intermediary_buffer = NeedsIntermediaryBuffer();
AbstractPipelineConfig config = {};
config.vertex_shader = m_vertex_shader.get();
config.geometry_shader =
g_gfx->UseGeometryShaderForUI() ? g_shader_cache->GetTexcoordGeometryShader() : nullptr;
config.pixel_shader = m_pixel_shader.get();
config.vertex_shader =
needs_intermediary_buffer ? m_vertex_shader.get() : m_default_vertex_shader.get();
// This geometry shader will take care of reading both layer 0 and 1 on the source texture,
// and writing to both layer 0 and 1 on the render target.
config.geometry_shader = UseGeometryShaderForPostProcess(needs_intermediary_buffer) ?
g_shader_cache->GetTexcoordGeometryShader() :
nullptr;
config.pixel_shader = m_default_pixel_shader.get();
config.rasterization_state = RenderState::GetNoCullRasterizationState(PrimitiveType::Triangles);
config.depth_state = RenderState::GetNoDepthTestingDepthState();
config.blending_state = RenderState::GetNoBlendingBlendState();
config.framebuffer_state = RenderState::GetColorFramebufferState(m_framebuffer_format);
config.framebuffer_state = RenderState::GetColorFramebufferState(
needs_intermediary_buffer ? s_intermediary_buffer_format : m_framebuffer_format);
config.usage = AbstractPipelineUsage::Utility;
// We continue even if it failed, it will be skipped later on
if (config.pixel_shader)
m_default_pipeline = g_gfx->CreatePipeline(config);
config.vertex_shader = m_default_vertex_shader.get();
config.geometry_shader = UseGeometryShaderForPostProcess(false) ?
g_shader_cache->GetTexcoordGeometryShader() :
nullptr;
config.pixel_shader = m_pixel_shader.get();
config.framebuffer_state = RenderState::GetColorFramebufferState(m_framebuffer_format);
m_pipeline = g_gfx->CreatePipeline(config);
if (!m_pipeline)
return false;

30
Source/Core/VideoCommon/PostProcessing.h
View File

@ -16,12 +16,14 @@
class AbstractPipeline;
class AbstractShader;
class AbstractTexture;
class AbstractFramebuffer;
namespace VideoCommon
{
class PostProcessingConfiguration
{
public:
// User defined post process
struct ConfigurationOption
{
enum class OptionType
@ -105,29 +107,43 @@ public:
void RecompilePipeline();
void BlitFromTexture(const MathUtil::Rectangle<int>& dst, const MathUtil::Rectangle<int>& src,
const AbstractTexture* src_tex, int src_layer);
const AbstractTexture* src_tex, int src_layer = -1);
bool IsColorCorrectionActive() const;
bool NeedsIntermediaryBuffer() const;
protected:
std::string GetUniformBufferHeader() const;
std::string GetHeader() const;
std::string GetUniformBufferHeader(bool user_post_process) const;
std::string GetHeader(bool user_post_process) const;
std::string GetFooter() const;
bool CompileVertexShader();
bool CompilePixelShader();
bool CompilePipeline();
size_t CalculateUniformsSize() const;
size_t CalculateUniformsSize(bool user_post_process) const;
void FillUniformBuffer(const MathUtil::Rectangle<int>& src, const AbstractTexture* src_tex,
int src_layer);
int src_layer, const MathUtil::Rectangle<int>& dst,
const MathUtil::Rectangle<int>& wnd, u8* buffer, bool user_post_process);
// Timer for determining our time value
Common::Timer m_timer;
PostProcessingConfiguration m_config;
// Dolphin fixed post process:
PostProcessingConfiguration::ConfigMap m_default_options;
std::unique_ptr<AbstractShader> m_default_vertex_shader;
std::unique_ptr<AbstractShader> m_default_pixel_shader;
std::unique_ptr<AbstractPipeline> m_default_pipeline;
std::unique_ptr<AbstractFramebuffer> m_intermediary_frame_buffer;
std::unique_ptr<AbstractTexture> m_intermediary_color_texture;
std::vector<u8> m_default_uniform_staging_buffer;
// User post process:
PostProcessingConfiguration m_config;
std::unique_ptr<AbstractShader> m_vertex_shader;
std::unique_ptr<AbstractShader> m_pixel_shader;
std::unique_ptr<AbstractPipeline> m_pipeline;
AbstractTextureFormat m_framebuffer_format = AbstractTextureFormat::Undefined;
std::vector<u8> m_uniform_staging_buffer;
AbstractTextureFormat m_framebuffer_format = AbstractTextureFormat::Undefined;
};
} // namespace VideoCommon

4
Source/Core/VideoCommon/Present.cpp
View File

@ -514,9 +514,11 @@ void Presenter::RenderXFBToScreen(const MathUtil::Rectangle<int>& target_rc,
m_post_processor->BlitFromTexture(left_rc, source_rc, source_texture, 0);
m_post_processor->BlitFromTexture(right_rc, source_rc, source_texture, 1);
}
// Every other case will be treated the same (stereo or not).
// If there's multiple source layers, they should all be copied.
else
{
m_post_processor->BlitFromTexture(target_rc, source_rc, source_texture, 0);
m_post_processor->BlitFromTexture(target_rc, source_rc, source_texture);
}
}

14
Source/Core/VideoCommon/VideoConfig.cpp
View File

@ -139,6 +139,15 @@ void VideoConfig::Refresh()
bArbitraryMipmapDetection = Config::Get(Config::GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION);
fArbitraryMipmapDetectionThreshold =
Config::Get(Config::GFX_ENHANCE_ARBITRARY_MIPMAP_DETECTION_THRESHOLD);
bHDR = Config::Get(Config::GFX_ENHANCE_HDR_OUTPUT);
color_correction.bCorrectColorSpace = Config::Get(Config::GFX_CC_CORRECT_COLOR_SPACE);
color_correction.game_color_space = Config::Get(Config::GFX_CC_GAME_COLOR_SPACE);
color_correction.bCorrectGamma = Config::Get(Config::GFX_CC_CORRECT_GAMMA);
color_correction.fGameGamma = Config::Get(Config::GFX_CC_GAME_GAMMA);
color_correction.bSDRDisplayGammaSRGB = Config::Get(Config::GFX_CC_SDR_DISPLAY_GAMMA_SRGB);
color_correction.fSDRDisplayCustomGamma = Config::Get(Config::GFX_CC_SDR_DISPLAY_CUSTOM_GAMMA);
color_correction.fHDRPaperWhiteNits = Config::Get(Config::GFX_CC_HDR_PAPER_WHITE_NITS);
stereo_mode = Config::Get(Config::GFX_STEREO_MODE);
iStereoDepth = Config::Get(Config::GFX_STEREO_DEPTH);
@ -263,6 +272,7 @@ void CheckForConfigChanges()
const AspectMode old_suggested_aspect_mode = g_ActiveConfig.suggested_aspect_mode;
const bool old_widescreen_hack = g_ActiveConfig.bWidescreenHack;
const auto old_post_processing_shader = g_ActiveConfig.sPostProcessingShader;
const auto old_hdr = g_ActiveConfig.bHDR;
UpdateActiveConfig();
FreeLook::UpdateActiveConfig();
@ -314,6 +324,8 @@ void CheckForConfigChanges()
changed_bits |= CONFIG_CHANGE_BIT_ASPECT_RATIO;
if (old_post_processing_shader != g_ActiveConfig.sPostProcessingShader)
changed_bits |= CONFIG_CHANGE_BIT_POST_PROCESSING_SHADER;
if (old_hdr != g_ActiveConfig.bHDR)
changed_bits |= CONFIG_CHANGE_BIT_HDR;
// No changes?
if (changed_bits == 0)
@ -323,7 +335,7 @@ void CheckForConfigChanges()
// Framebuffer changed?
if (changed_bits & (CONFIG_CHANGE_BIT_MULTISAMPLES | CONFIG_CHANGE_BIT_STEREO_MODE |
CONFIG_CHANGE_BIT_TARGET_SIZE))
CONFIG_CHANGE_BIT_TARGET_SIZE | CONFIG_CHANGE_BIT_HDR))
{
g_framebuffer_manager->RecreateEFBFramebuffer();
}

29
Source/Core/VideoCommon/VideoConfig.h
View File

@ -52,6 +52,13 @@ enum class TextureFilteringMode : int
Linear,
};
enum class ColorCorrectionRegion : int
{
SMPTE_NTSCM,
SYSTEMJ_NTSCJ,
EBU_PAL,
};
enum class TriState : int
{
Off,
@ -72,6 +79,7 @@ enum ConfigChangeBits : u32
CONFIG_CHANGE_BIT_BBOX = (1 << 7),
CONFIG_CHANGE_BIT_ASPECT_RATIO = (1 << 8),
CONFIG_CHANGE_BIT_POST_PROCESSING_SHADER = (1 << 9),
CONFIG_CHANGE_BIT_HDR = (1 << 10),
};
// NEVER inherit from this class.
@ -101,6 +109,26 @@ struct VideoConfig final
bool bDisableCopyFilter = false;
bool bArbitraryMipmapDetection = false;
float fArbitraryMipmapDetectionThreshold = 0;
bool bHDR = false;
// Color Correction
struct
{
// Color Space Correction:
bool bCorrectColorSpace = false;
ColorCorrectionRegion game_color_space = ColorCorrectionRegion::SMPTE_NTSCM;
// Gamma Correction:
bool bCorrectGamma = false;
float fGameGamma = 2.35f;
bool bSDRDisplayGammaSRGB = true;
// Custom gamma when the display is not sRGB
float fSDRDisplayCustomGamma = 2.2f;
// HDR:
// 200 is a good default value that matches the brightness of many SDR screens
float fHDRPaperWhiteNits = 200.f;
} color_correction;
// Information
bool bShowFPS = false;
@ -275,6 +303,7 @@ struct VideoConfig final
bool bSupportsDynamicVertexLoader = false;
bool bSupportsVSLinePointExpand = false;
bool bSupportsGLLayerInFS = true;
bool bSupportsHDROutput = false;
} backend_info;
// Utility