moonlight-stream/moonlight-qt
1
0
Fork 0
mirror of https://github.com/moonlight-stream/moonlight-qt.git synced 2025-07-01 23:35:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use optimized pixel shaders for the normal SDR and HDR colorspaces

Browse Source
This commit is contained in:
Cameron Gutman 2022-02-09 20:46:02 -06:00
parent be2af1d17a
commit 9a64c026ea
10 changed files with 193 additions and 73 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
app/resources.qrc
View File

@ -62,5 +62,7 @@
<file alias="d3d11_vertex.fxc">shaders/d3d11_vertex.fxc</file>
<file alias="d3d11_overlay_pixel.fxc">shaders/d3d11_overlay_pixel.fxc</file>
<file alias="d3d11_video_pixel.fxc">shaders/d3d11_video_pixel.fxc</file>
<file alias="d3d11_bt601lim_pixel.fxc">shaders/d3d11_bt601lim_pixel.fxc</file>
<file alias="d3d11_bt2020lim_pixel.fxc">shaders/d3d11_bt2020lim_pixel.fxc</file>
</qresource>
</RCC>

2
app/shaders/build_hlsl.bat
View File

@ -2,3 +2,5 @@ fxc /T vs_4_0_level_9_3 /Fo d3d11_vertex.fxc d3d11_vertex.hlsl
fxc /T ps_4_0_level_9_3 /Fo d3d11_overlay_pixel.fxc d3d11_overlay_pixel.hlsl
fxc /T ps_4_0_level_9_3 /Fo d3d11_video_pixel.fxc d3d11_video_pixel.hlsl
fxc /T ps_4_0_level_9_3 /Fo d3d11_bt601lim_pixel.fxc d3d11_bt601lim_pixel.hlsl
fxc /T ps_4_0_level_9_3 /Fo d3d11_bt2020lim_pixel.fxc d3d11_bt2020lim_pixel.hlsl

BIN
app/shaders/d3d11_bt2020lim_pixel.fxc Normal file
View File

Binary file not shown.

35
app/shaders/d3d11_bt2020lim_pixel.hlsl Normal file
View File

@ -0,0 +1,35 @@
Texture2D<min16float> luminancePlane : register(t0);
Texture2D<min16float2> chrominancePlane : register(t1);
SamplerState theSampler : register(s0);
static const min16float3x3 cscMatrix =
{
1.1644, 1.1644, 1.1644,
0.0, -0.1874, 2.1418,
1.6781, -0.6505, 0.0,
};
static const min16float3 offsets =
{
16.0 / 255.0, 128.0 / 255.0, 128.0 / 255.0
};
struct ShaderInput
{
min16float4 pos : SV_POSITION;
min16float2 tex : TEXCOORD0;
};
min16float4 main(ShaderInput input) : SV_TARGET
{
min16float3 yuv = min16float3(luminancePlane.Sample(theSampler, input.tex),
chrominancePlane.Sample(theSampler, input.tex));
// Subtract the YUV offset for limited vs full range
yuv -= offsets;
// Multiply by the conversion matrix for this colorspace
yuv = mul(yuv, cscMatrix);
return min16float4(yuv, 1.0);
}

BIN
app/shaders/d3d11_bt601lim_pixel.fxc Normal file
View File

Binary file not shown.

35
app/shaders/d3d11_bt601lim_pixel.hlsl Normal file
View File

@ -0,0 +1,35 @@
Texture2D<min16float> luminancePlane : register(t0);
Texture2D<min16float2> chrominancePlane : register(t1);
SamplerState theSampler : register(s0);
static const min16float3x3 cscMatrix =
{
1.1644, 1.1644, 1.1644,
0.0, -0.3917, 2.0172,
1.5960, -0.8129, 0.0,
};
static const min16float3 offsets =
{
16.0 / 255.0, 128.0 / 255.0, 128.0 / 255.0
};
struct ShaderInput
{
min16float4 pos : SV_POSITION;
min16float2 tex : TEXCOORD0;
};
min16float4 main(ShaderInput input) : SV_TARGET
{
min16float3 yuv = min16float3(luminancePlane.Sample(theSampler, input.tex),
chrominancePlane.Sample(theSampler, input.tex));
// Subtract the YUV offset for limited vs full range
yuv -= offsets;
// Multiply by the conversion matrix for this colorspace
yuv = mul(yuv, cscMatrix);
return min16float4(yuv, 1.0);
}

BIN
app/shaders/d3d11_video_pixel.fxc
View File

Binary file not shown.

2
app/shaders/d3d11_video_pixel.hlsl
View File

@ -25,5 +25,5 @@ min16float4 main(ShaderInput input) : SV_TARGET
// Multiply by the conversion matrix for this colorspace
yuv = mul(yuv, cscMatrix);
return min16float4(saturate(yuv), 1.0);
return min16float4(yuv, 1.0);
}

182
app/streaming/video/ffmpeg-renderers/d3d11va.cpp
View File

@ -81,7 +81,9 @@ D3D11VARenderer::D3D11VARenderer()
m_LastColorRange(AVCOL_RANGE_UNSPECIFIED),
m_AllowTearing(false),
m_FrameWaitableObject(nullptr),
m_VideoPixelShader(nullptr),
m_VideoGenericPixelShader(nullptr),
m_VideoBt601LimPixelShader(nullptr),
m_VideoBt2020LimPixelShader(nullptr),
m_VideoVertexBuffer(nullptr),
m_OverlayLock(0),
m_OverlayPixelShader(nullptr),
@ -101,7 +103,9 @@ D3D11VARenderer::~D3D11VARenderer()
SDL_DestroyMutex(m_ContextLock);
SAFE_COM_RELEASE(m_VideoVertexBuffer);
SAFE_COM_RELEASE(m_VideoPixelShader);
SAFE_COM_RELEASE(m_VideoBt2020LimPixelShader);
SAFE_COM_RELEASE(m_VideoBt601LimPixelShader);
SAFE_COM_RELEASE(m_VideoGenericPixelShader);
for (int i = 0; i < ARRAYSIZE(m_VideoTextureResourceViews); i++) {
SAFE_COM_RELEASE(m_VideoTextureResourceViews[i][0]);
@ -734,71 +738,88 @@ void D3D11VARenderer::renderOverlay(Overlay::OverlayType type)
overlayVertexBuffer->Release();
}
void D3D11VARenderer::updateColorConversionConstants(AVFrame* frame)
void D3D11VARenderer::bindColorConversion(AVFrame* frame)
{
// If nothing has changed since last frame, we're done
if (frame->colorspace == m_LastColorSpace && frame->color_range == m_LastColorRange) {
return;
// We have purpose-built shaders for the common Rec 601 (SDR) and Rec 2020 (HDR) cases
if (frame->color_range == AVCOL_RANGE_MPEG && frame->colorspace == AVCOL_SPC_SMPTE170M) {
m_DeviceContext->PSSetShader(m_VideoBt601LimPixelShader, nullptr, 0);
}
D3D11_BUFFER_DESC constDesc = {};
constDesc.ByteWidth = sizeof(CSC_CONST_BUF);
constDesc.Usage = D3D11_USAGE_IMMUTABLE;
constDesc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
constDesc.CPUAccessFlags = 0;
constDesc.MiscFlags = 0;
// This handles the case where the color range is unknown,
// so that we use Limited color range which is the default
// behavior for Moonlight.
CSC_CONST_BUF constBuf = {};
bool fullRange = (frame->color_range == AVCOL_RANGE_JPEG);
const float* rawCscMatrix;
switch (frame->colorspace) {
case AVCOL_SPC_SMPTE170M:
case AVCOL_SPC_BT470BG:
rawCscMatrix = fullRange ? k_CscMatrix_Bt601Full : k_CscMatrix_Bt601Lim;
break;
case AVCOL_SPC_BT709:
rawCscMatrix = fullRange ? k_CscMatrix_Bt709Full : k_CscMatrix_Bt709Lim;
break;
case AVCOL_SPC_BT2020_NCL:
case AVCOL_SPC_BT2020_CL:
rawCscMatrix = fullRange ? k_CscMatrix_Bt2020Full : k_CscMatrix_Bt2020Lim;
break;
default:
SDL_assert(false);
return;
}
// We need to adjust our raw CSC matrix to be column-major and with float3 vectors
// padded with a float in between each of them to adhere to HLSL requirements.
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
constBuf.cscMatrix[i * 4 + j] = rawCscMatrix[j * 3 + i];
}
}
// No adjustments are needed to the float[3] array of offsets, so it can just
// be copied with memcpy().
memcpy(constBuf.offsets,
fullRange ? k_Offsets_Full : k_Offsets_Lim,
sizeof(constBuf.offsets));
D3D11_SUBRESOURCE_DATA constData = {};
constData.pSysMem = &constBuf;
ID3D11Buffer* constantBuffer;
HRESULT hr = m_Device->CreateBuffer(&constDesc, &constData, &constantBuffer);
if (SUCCEEDED(hr)) {
m_DeviceContext->PSSetConstantBuffers(0, 1, &constantBuffer);
constantBuffer->Release();
else if (frame->color_range == AVCOL_RANGE_MPEG && frame->colorspace == AVCOL_SPC_BT2020_NCL) {
m_DeviceContext->PSSetShader(m_VideoBt2020LimPixelShader, nullptr, 0);
}
else {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
"ID3D11Device::CreateBuffer() failed: %x",
hr);
return;
// We'll need to use the generic shader for this colorspace and color range combo
m_DeviceContext->PSSetShader(m_VideoGenericPixelShader, nullptr, 0);
// If nothing has changed since last frame, we're done
if (frame->colorspace == m_LastColorSpace && frame->color_range == m_LastColorRange) {
return;
}
SDL_LogWarn(SDL_LOG_CATEGORY_APPLICATION,
"Falling back to generic video pixel shader for %d:%d",
frame->colorspace,
frame->color_range);
D3D11_BUFFER_DESC constDesc = {};
constDesc.ByteWidth = sizeof(CSC_CONST_BUF);
constDesc.Usage = D3D11_USAGE_IMMUTABLE;
constDesc.BindFlags = D3D11_BIND_CONSTANT_BUFFER;
constDesc.CPUAccessFlags = 0;
constDesc.MiscFlags = 0;
// This handles the case where the color range is unknown,
// so that we use Limited color range which is the default
// behavior for Moonlight.
CSC_CONST_BUF constBuf = {};
bool fullRange = (frame->color_range == AVCOL_RANGE_JPEG);
const float* rawCscMatrix;
switch (frame->colorspace) {
case AVCOL_SPC_SMPTE170M:
case AVCOL_SPC_BT470BG:
rawCscMatrix = fullRange ? k_CscMatrix_Bt601Full : k_CscMatrix_Bt601Lim;
break;
case AVCOL_SPC_BT709:
rawCscMatrix = fullRange ? k_CscMatrix_Bt709Full : k_CscMatrix_Bt709Lim;
break;
case AVCOL_SPC_BT2020_NCL:
case AVCOL_SPC_BT2020_CL:
rawCscMatrix = fullRange ? k_CscMatrix_Bt2020Full : k_CscMatrix_Bt2020Lim;
break;
default:
SDL_assert(false);
return;
}
// We need to adjust our raw CSC matrix to be column-major and with float3 vectors
// padded with a float in between each of them to adhere to HLSL requirements.
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
constBuf.cscMatrix[i * 4 + j] = rawCscMatrix[j * 3 + i];
}
}
// No adjustments are needed to the float[3] array of offsets, so it can just
// be copied with memcpy().
memcpy(constBuf.offsets,
fullRange ? k_Offsets_Full : k_Offsets_Lim,
sizeof(constBuf.offsets));
D3D11_SUBRESOURCE_DATA constData = {};
constData.pSysMem = &constBuf;
ID3D11Buffer* constantBuffer;
HRESULT hr = m_Device->CreateBuffer(&constDesc, &constData, &constantBuffer);
if (SUCCEEDED(hr)) {
m_DeviceContext->PSSetConstantBuffers(0, 1, &constantBuffer);
constantBuffer->Release();
}
else {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
"ID3D11Device::CreateBuffer() failed: %x",
hr);
return;
}
}
m_LastColorSpace = frame->colorspace;
@ -807,9 +828,6 @@ void D3D11VARenderer::updateColorConversionConstants(AVFrame* frame)
void D3D11VARenderer::renderVideo(AVFrame* frame)
{
// Update our CSC constants if the colorspace has changed
updateColorConversionConstants(frame);
// Bind video rendering vertex buffer
UINT stride = sizeof(VERTEX);
UINT offset = 0;
@ -826,8 +844,10 @@ void D3D11VARenderer::renderVideo(AVFrame* frame)
return;
}
// Bind video pixel shader and SRVs for this frame
m_DeviceContext->PSSetShader(m_VideoPixelShader, nullptr, 0);
// Bind our CSC shader (and constant buffer, if required)
bindColorConversion(frame);
// Bind SRVs for this frame
m_DeviceContext->PSSetShaderResources(0, 2, m_VideoTextureResourceViews[textureIndex]);
// Draw the video
@ -1140,7 +1160,31 @@ bool D3D11VARenderer::setupRenderingResources()
{
QByteArray videoPixelShaderBytecode = Path::readDataFile("d3d11_video_pixel.fxc");
hr = m_Device->CreatePixelShader(videoPixelShaderBytecode.constData(), videoPixelShaderBytecode.length(), nullptr, &m_VideoPixelShader);
hr = m_Device->CreatePixelShader(videoPixelShaderBytecode.constData(), videoPixelShaderBytecode.length(), nullptr, &m_VideoGenericPixelShader);
if (FAILED(hr)) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
"ID3D11Device::CreatePixelShader() failed: %x",
hr);
return false;
}
}
{
QByteArray videoPixelShaderBytecode = Path::readDataFile("d3d11_bt601lim_pixel.fxc");
hr = m_Device->CreatePixelShader(videoPixelShaderBytecode.constData(), videoPixelShaderBytecode.length(), nullptr, &m_VideoBt601LimPixelShader);
if (FAILED(hr)) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
"ID3D11Device::CreatePixelShader() failed: %x",
hr);
return false;
}
}
{
QByteArray videoPixelShaderBytecode = Path::readDataFile("d3d11_bt2020lim_pixel.fxc");
hr = m_Device->CreatePixelShader(videoPixelShaderBytecode.constData(), videoPixelShaderBytecode.length(), nullptr, &m_VideoBt2020LimPixelShader);
if (FAILED(hr)) {
SDL_LogError(SDL_LOG_CATEGORY_APPLICATION,
"ID3D11Device::CreatePixelShader() failed: %x",

6
app/streaming/video/ffmpeg-renderers/d3d11va.h
View File

@ -30,7 +30,7 @@ private:
bool setupRenderingResources();
bool setupTexturePoolViews(AVD3D11VAFramesContext* frameContext);
void renderOverlay(Overlay::OverlayType type);
void updateColorConversionConstants(AVFrame* frame);
void bindColorConversion(AVFrame* frame);
void renderVideo(AVFrame* frame);
bool checkDecoderSupport(IDXGIAdapter* adapter);
@ -52,7 +52,9 @@ private:
bool m_AllowTearing;
HANDLE m_FrameWaitableObject;
ID3D11PixelShader* m_VideoPixelShader;
ID3D11PixelShader* m_VideoGenericPixelShader;
ID3D11PixelShader* m_VideoBt601LimPixelShader;
ID3D11PixelShader* m_VideoBt2020LimPixelShader;
ID3D11Buffer* m_VideoVertexBuffer;
#define DECODER_BUFFER_POOL_SIZE 17