[idd] project: organize driver sources by responsibility

Group the IDD sources and Visual Studio filters by subsystem.

Split the device and swap-chain implementations into focused units,
rename the context classes, and reduce header coupling.
This commit is contained in:
Geoffrey McRae
2026-08-07 14:38:36 +10:00
parent 3ddc199bec
commit 30a1383d5e
76 changed files with 5067 additions and 3923 deletions

View File

@@ -0,0 +1,337 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "CColorTransformEffect.h"
#include "CDebug.h"
#include <cstring>
using namespace PostProcessUtil;
namespace
{
enum TransferFunction : UINT
{
TRANSFER_LINEAR,
TRANSFER_SRGB,
TRANSFER_PQ,
};
bool CreateUploadBuffer(const ComPtr<ID3D12Device3>& device, size_t size,
ComPtr<ID3D12Resource>& resource)
{
D3D12_HEAP_PROPERTIES heapProps = {};
heapProps.Type = D3D12_HEAP_TYPE_UPLOAD;
D3D12_RESOURCE_DESC desc = {};
desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
desc.Width = size;
desc.Height = 1;
desc.DepthOrArraySize = 1;
desc.MipLevels = 1;
desc.SampleDesc.Count = 1;
desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
const HRESULT hr = device->CreateCommittedResource(&heapProps,
D3D12_HEAP_FLAG_NONE, &desc, D3D12_RESOURCE_STATE_GENERIC_READ,
nullptr, IID_PPV_ARGS(&resource));
return SUCCEEDED(hr);
}
bool Upload(const ComPtr<ID3D12Resource>& resource,
const void * data, size_t size)
{
void * dst = nullptr;
const D3D12_RANGE readRange = { 0, 0 };
if (FAILED(resource->Map(0, &readRange, &dst)))
return false;
std::memcpy(dst, data, size);
resource->Unmap(0, nullptr);
return true;
}
}
bool CColorTransformEffect::Init(const ComPtr<ID3D12Device3>& device)
{
D3D12_DESCRIPTOR_RANGE ranges[4] = {};
ranges[0].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_CBV;
ranges[0].NumDescriptors = 1;
ranges[0].BaseShaderRegister = 0;
ranges[0].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[1].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
ranges[1].NumDescriptors = 1;
ranges[1].BaseShaderRegister = 0;
ranges[1].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[2].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
ranges[2].NumDescriptors = 1;
ranges[2].BaseShaderRegister = 1;
ranges[2].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[3].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
ranges[3].NumDescriptors = 1;
ranges[3].BaseShaderRegister = 0;
ranges[3].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
const char * shader =
"cbuffer Constants : register(b0)\n"
"{\n"
" float4 ColorMatrix[3];\n"
" float Scalar;\n"
" uint MatrixEnabled;\n"
" uint LutEnabled;\n"
" uint InputTransfer;\n"
" uint OutputTransfer;\n"
"};\n"
"Texture2D<float4> src : register(t0);\n"
"Buffer<float4> lut : register(t1);\n"
"RWTexture2D<float4> dst : register(u0);\n"
"static const float PQ_m1 = 0.1593017578125;\n"
"static const float PQ_m2 = 78.84375;\n"
"static const float PQ_c1 = 0.8359375;\n"
"static const float PQ_c2 = 18.8515625;\n"
"static const float PQ_c3 = 18.6875;\n"
"float3 decode(float3 value)\n"
"{\n"
" if (InputTransfer == 1)\n"
" return lerp(value / 12.92, pow((value + 0.055) / 1.055, 2.4),\n"
" step(0.04045, value));\n"
" if (InputTransfer == 2)\n"
" {\n"
" float3 p = pow(max(value, 0.0), 1.0 / PQ_m2);\n"
" return pow(max(p - PQ_c1, 0.0) / max(PQ_c2 - PQ_c3 * p, 1e-6),\n"
" 1.0 / PQ_m1);\n"
" }\n"
" return value;\n"
"}\n"
"float3 encode(float3 value)\n"
"{\n"
" if (OutputTransfer == 1)\n"
" return lerp(value * 12.92, 1.055 * pow(max(value, 0.0), 1.0 / 2.4) - 0.055,\n"
" step(0.0031308, value));\n"
" if (OutputTransfer == 2)\n"
" {\n"
" float3 p = pow(max(value, 0.0), PQ_m1);\n"
" return pow((PQ_c1 + PQ_c2 * p) / (1.0 + PQ_c3 * p), PQ_m2);\n"
" }\n"
" return value;\n"
"}\n"
"float3 rgbToXYZ(float3 rgb)\n"
"{\n"
" if (InputTransfer == 2)\n"
" return float3(\n"
" dot(rgb, float3(0.6369580, 0.1446169, 0.1688810)),\n"
" dot(rgb, float3(0.2627002, 0.6779981, 0.0593017)),\n"
" dot(rgb, float3(0.0000000, 0.0280727, 1.0609851)));\n"
" return float3(\n"
" dot(rgb, float3(0.4123908, 0.3575843, 0.1804808)),\n"
" dot(rgb, float3(0.2126390, 0.7151687, 0.0721923)),\n"
" dot(rgb, float3(0.0193308, 0.1191948, 0.9505322)));\n"
"}\n"
"float3 xyzToRGB(float3 xyz)\n"
"{\n"
" if (OutputTransfer == 2)\n"
" return float3(\n"
" dot(xyz, float3( 1.7166512, -0.3556708, -0.2533663)),\n"
" dot(xyz, float3(-0.6666844, 1.6164812, 0.0157685)),\n"
" dot(xyz, float3( 0.0176399, -0.0427706, 0.9421031)));\n"
" return float3(\n"
" dot(xyz, float3( 3.2409699, -1.5373832, -0.4986108)),\n"
" dot(xyz, float3(-0.9692436, 1.8759675, 0.0415551)),\n"
" dot(xyz, float3( 0.0556301, -0.2039770, 1.0569715)));\n"
"}\n"
"float3 applyLut(float3 value)\n"
"{\n"
" float3 pos = saturate(value) * 4095.0;\n"
" uint3 lo = (uint3)floor(pos);\n"
" uint3 hi = min(lo + 1, 4095);\n"
" float3 f = frac(pos);\n"
" return float3(\n"
" lerp(lut[lo.r].r, lut[hi.r].r, f.r),\n"
" lerp(lut[lo.g].g, lut[hi.g].g, f.g),\n"
" lerp(lut[lo.b].b, lut[hi.b].b, f.b));\n"
"}\n"
"[numthreads(" POST_PROCESS_THREADS_STR ", " POST_PROCESS_THREADS_STR ", 1)]\n"
"void main(uint3 dt : SV_DispatchThreadID)\n"
"{\n"
" float4 pixel = src[dt.xy];\n"
" float3 value = decode(pixel.rgb);\n"
" if (MatrixEnabled != 0 || InputTransfer != OutputTransfer)\n"
" {\n"
" float3 xyz = rgbToXYZ(value);\n"
" if (MatrixEnabled != 0)\n"
" xyz = float3(dot(float4(xyz, 1.0), ColorMatrix[0]),\n"
" dot(float4(xyz, 1.0), ColorMatrix[1]),\n"
" dot(float4(xyz, 1.0), ColorMatrix[2])) * Scalar;\n"
" value = xyzToRGB(xyz);\n"
" }\n"
" if (InputTransfer == 0 && OutputTransfer == 2)\n"
" value *= 80.0 / 10000.0;\n"
" value = encode(value);\n"
" if (LutEnabled != 0)\n"
" value = applyLut(value);\n"
" dst[dt.xy] = float4(value, pixel.a);\n"
"}\n";
if (!InitCompute(device, ranges, ARRAYSIZE(ranges), nullptr, 0, shader))
return false;
const size_t constSize = AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
if (!CreateUploadBuffer(device, constSize, m_constBuffer) ||
!CreateUploadBuffer(device, sizeof(float) * 4096 * 4, m_lutBuffer))
{
DEBUG_ERROR("Failed to create color transform buffers");
return false;
}
return true;
}
PostProcessStatus CColorTransformEffect::SetFormat(
const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst)
{
if (!src.colorTransform || IsIdentityColorTransform(*src.colorTransform) ||
(!src.colorTransform->matrixEnabled && !src.colorTransform->lutEnabled))
return PostProcessStatus::BYPASS_EFFECT;
DXGI_FORMAT dstFormat;
FrameType frameType;
switch (src.desc.Format)
{
case DXGI_FORMAT_B8G8R8A8_UNORM:
dstFormat = DXGI_FORMAT_R8G8B8A8_UNORM;
frameType = FRAME_TYPE_RGBA;
break;
case DXGI_FORMAT_R8G8B8A8_UNORM:
case DXGI_FORMAT_R10G10B10A2_UNORM:
dstFormat = src.desc.Format;
frameType = src.format;
break;
case DXGI_FORMAT_R16G16B16A16_FLOAT:
// The client wire format is HDR10. Perform the XYZ adjustment before
// the BT.2020 rotation, and its LUT after PQ encoding, in one pass.
dstFormat = DXGI_FORMAT_R10G10B10A2_UNORM;
frameType = FRAME_TYPE_RGBA10;
break;
default:
DEBUG_ERROR("Unsupported color transform source format %u", src.desc.Format);
return PostProcessStatus::FAILED;
}
D3D12_RESOURCE_DESC desc = src.desc;
desc.Format = dstFormat;
desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
if (!m_dst || m_dst->GetDesc().Width != desc.Width ||
m_dst->GetDesc().Height != desc.Height ||
m_dst->GetDesc().Format != desc.Format)
{
if (!CreateDefaultTexture(device, desc, m_dst))
return PostProcessStatus::FAILED;
}
std::memcpy(m_consts.matrix, src.colorTransform->matrix,
sizeof(m_consts.matrix));
m_consts.scalar = src.colorTransform->scalar;
m_consts.matrixEnabled = src.colorTransform->matrixEnabled;
m_consts.lutEnabled = src.colorTransform->lutEnabled;
m_consts.inputTransfer = src.hdrPQ ? TRANSFER_PQ :
(src.hdr ? TRANSFER_LINEAR : TRANSFER_SRGB);
m_consts.outputTransfer = src.hdr ? TRANSFER_PQ : TRANSFER_SRGB;
std::memcpy(m_lut, src.colorTransform->lut, sizeof(m_lut));
m_uploadPending = true;
m_srcFormat = src.desc.Format;
m_dstFormat = dstFormat;
m_threadsX = ((unsigned)desc.Width + (Threads - 1)) / Threads;
m_threadsY = ((unsigned)desc.Height + (Threads - 1)) / Threads;
dst.desc = desc;
dst.format = frameType;
if (src.hdr)
dst.hdrPQ = true;
return PostProcessStatus::SUCCESS;
}
ComPtr<ID3D12Resource> CColorTransformEffect::Run(
const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects)
{
UNREFERENCED_PARAMETER(dirtyRects);
UNREFERENCED_PARAMETER(nbDirtyRects);
// The framebuffer-indexed compute slot waits for this chain's previous
// submission before Run is called, so this is the first point where its
// upload buffers are guaranteed not to be in use by the GPU.
if (m_uploadPending)
{
if (!Upload(m_constBuffer, &m_consts, sizeof(m_consts)) ||
!Upload(m_lutBuffer, m_lut, sizeof(m_lut)))
DEBUG_ERROR("Failed to upload display color transform");
else
m_uploadPending = false;
}
TransitionDst(commandList, D3D12_RESOURCE_STATE_COMMON,
D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
D3D12_CPU_DESCRIPTOR_HANDLE handle =
m_descHeap->GetCPUDescriptorHandleForHeapStart();
const UINT inc = device->GetDescriptorHandleIncrementSize(
D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
D3D12_CONSTANT_BUFFER_VIEW_DESC cbvDesc = {};
cbvDesc.BufferLocation = m_constBuffer->GetGPUVirtualAddress();
cbvDesc.SizeInBytes = (UINT)AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
device->CreateConstantBufferView(&cbvDesc, handle);
handle.ptr += inc;
D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
srvDesc.Format = m_srcFormat;
srvDesc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
srvDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
srvDesc.Texture2D.MipLevels = 1;
device->CreateShaderResourceView(src.Get(), &srvDesc, handle);
handle.ptr += inc;
D3D12_SHADER_RESOURCE_VIEW_DESC lutDesc = {};
lutDesc.Format = DXGI_FORMAT_R32G32B32A32_FLOAT;
lutDesc.ViewDimension = D3D12_SRV_DIMENSION_BUFFER;
lutDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
lutDesc.Buffer.NumElements = 4096;
device->CreateShaderResourceView(m_lutBuffer.Get(), &lutDesc, handle);
handle.ptr += inc;
D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
uavDesc.Format = m_dstFormat;
uavDesc.ViewDimension = D3D12_UAV_DIMENSION_TEXTURE2D;
device->CreateUnorderedAccessView(m_dst.Get(), nullptr, &uavDesc, handle);
Bind(commandList);
commandList->Dispatch(m_threadsX, m_threadsY, 1);
TransitionDst(commandList, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
D3D12_RESOURCE_STATE_COMMON);
return m_dst;
}

View File

@@ -0,0 +1,57 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#pragma once
#include "CComputeEffect.h"
class CColorTransformEffect : public CComputeEffect
{
private:
struct Consts
{
float matrix[3][4];
float scalar;
UINT matrixEnabled;
UINT lutEnabled;
UINT inputTransfer;
UINT outputTransfer;
} m_consts = {};
float m_lut[4096][4] = {};
bool m_uploadPending = false;
ComPtr<ID3D12Resource> m_constBuffer;
ComPtr<ID3D12Resource> m_lutBuffer;
DXGI_FORMAT m_srcFormat = DXGI_FORMAT_UNKNOWN;
DXGI_FORMAT m_dstFormat = DXGI_FORMAT_UNKNOWN;
public:
const char * GetName() const override { return "ColorTransform"; }
bool Init(const ComPtr<ID3D12Device3>& device);
PostProcessStatus SetFormat(const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst) override;
ComPtr<ID3D12Resource> Run(const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects) override;
};

View File

@@ -0,0 +1,195 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "CComputeEffect.h"
#include "CDebug.h"
#include <d3dcompiler.h>
#include <cstring>
namespace PostProcessUtil
{
static bool CreateDefaultResource(const ComPtr<ID3D12Device3>& device,
const D3D12_RESOURCE_DESC& desc, ComPtr<ID3D12Resource>& resource)
{
D3D12_HEAP_PROPERTIES heapProps = {};
heapProps.Type = D3D12_HEAP_TYPE_DEFAULT;
heapProps.CPUPageProperty = D3D12_CPU_PAGE_PROPERTY_UNKNOWN;
heapProps.MemoryPoolPreference = D3D12_MEMORY_POOL_UNKNOWN;
heapProps.CreationNodeMask = 1;
heapProps.VisibleNodeMask = 1;
HRESULT hr = device->CreateCommittedResource(
&heapProps,
D3D12_HEAP_FLAG_CREATE_NOT_ZEROED,
&desc,
D3D12_RESOURCE_STATE_COMMON,
nullptr,
IID_PPV_ARGS(&resource));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create post-processing destination resource");
return false;
}
return true;
}
bool CreateDefaultTexture(const ComPtr<ID3D12Device3>& device,
const D3D12_RESOURCE_DESC& desc, ComPtr<ID3D12Resource>& resource)
{
return CreateDefaultResource(device, desc, resource);
}
bool CreateDefaultBuffer(const ComPtr<ID3D12Device3>& device,
UINT64 size, ComPtr<ID3D12Resource>& resource)
{
D3D12_RESOURCE_DESC desc = {};
desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
desc.Width = size;
desc.Height = 1;
desc.DepthOrArraySize = 1;
desc.MipLevels = 1;
desc.SampleDesc.Count = 1;
desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
return CreateDefaultResource(device, desc, resource);
}
}
bool CComputeEffect::InitCompute(const ComPtr<ID3D12Device3>& device,
const D3D12_DESCRIPTOR_RANGE * ranges, UINT rangeCount,
const D3D12_STATIC_SAMPLER_DESC * samplers, UINT samplerCount,
const char * shader)
{
D3D12_ROOT_PARAMETER rootParam = {};
rootParam.ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
rootParam.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
rootParam.DescriptorTable.NumDescriptorRanges = rangeCount;
rootParam.DescriptorTable.pDescriptorRanges = ranges;
D3D12_VERSIONED_ROOT_SIGNATURE_DESC rootSignatureDesc = {};
rootSignatureDesc.Version = D3D_ROOT_SIGNATURE_VERSION_1;
rootSignatureDesc.Desc_1_0.NumParameters = 1;
rootSignatureDesc.Desc_1_0.pParameters = &rootParam;
rootSignatureDesc.Desc_1_0.NumStaticSamplers = samplerCount;
rootSignatureDesc.Desc_1_0.pStaticSamplers = samplers;
rootSignatureDesc.Desc_1_0.Flags = D3D12_ROOT_SIGNATURE_FLAG_NONE;
ComPtr<ID3DBlob> blob;
ComPtr<ID3DBlob> error;
HRESULT hr = D3D12SerializeVersionedRootSignature(
&rootSignatureDesc, &blob, &error);
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to serialize post-processing root signature");
if (error)
DEBUG_ERROR("%s", (const char *)error->GetBufferPointer());
return false;
}
hr = device->CreateRootSignature(
0,
blob->GetBufferPointer(),
blob->GetBufferSize(),
IID_PPV_ARGS(&m_rootSignature));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create post-processing root signature");
return false;
}
blob.Reset();
error.Reset();
hr = D3DCompile(
shader,
std::strlen(shader),
nullptr,
nullptr,
nullptr,
"main",
"cs_5_0",
0,
0,
&blob,
&error);
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to compile post-processing shader");
if (error)
DEBUG_ERROR("%s", (const char *)error->GetBufferPointer());
return false;
}
D3D12_COMPUTE_PIPELINE_STATE_DESC psoDesc = {};
psoDesc.pRootSignature = m_rootSignature.Get();
psoDesc.CS.pShaderBytecode = blob->GetBufferPointer();
psoDesc.CS.BytecodeLength = blob->GetBufferSize();
hr = device->CreateComputePipelineState(&psoDesc, IID_PPV_ARGS(&m_pso));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create post-processing PSO");
return false;
}
UINT descriptorCount = 0;
for (UINT i = 0; i < rangeCount; ++i)
descriptorCount += ranges[i].NumDescriptors;
D3D12_DESCRIPTOR_HEAP_DESC heapDesc = {};
heapDesc.Type = D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV;
heapDesc.NumDescriptors = descriptorCount;
heapDesc.Flags = D3D12_DESCRIPTOR_HEAP_FLAG_SHADER_VISIBLE;
hr = device->CreateDescriptorHeap(&heapDesc, IID_PPV_ARGS(&m_descHeap));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create post-processing descriptor heap");
return false;
}
return true;
}
void CComputeEffect::Bind(const ComPtr<ID3D12GraphicsCommandList>& commandList)
{
ID3D12DescriptorHeap * heaps[] = { m_descHeap.Get() };
commandList->SetDescriptorHeaps(1, heaps);
commandList->SetPipelineState(m_pso.Get());
commandList->SetComputeRootSignature(m_rootSignature.Get());
commandList->SetComputeRootDescriptorTable(
0, m_descHeap->GetGPUDescriptorHandleForHeapStart());
}
void CComputeEffect::TransitionDst(
const ComPtr<ID3D12GraphicsCommandList>& commandList,
D3D12_RESOURCE_STATES before, D3D12_RESOURCE_STATES after)
{
D3D12_RESOURCE_BARRIER barrier = {};
barrier.Type = D3D12_RESOURCE_BARRIER_TYPE_TRANSITION;
barrier.Flags = D3D12_RESOURCE_BARRIER_FLAG_NONE;
barrier.Transition.pResource = m_dst.Get();
barrier.Transition.StateBefore = before;
barrier.Transition.StateAfter = after;
barrier.Transition.Subresource = D3D12_RESOURCE_BARRIER_ALL_SUBRESOURCES;
commandList->ResourceBarrier(1, &barrier);
}

View File

@@ -0,0 +1,63 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#pragma once
#include "postprocess/CPostProcessor.h"
#define POST_PROCESS_THREADS_STR "8"
namespace PostProcessUtil
{
static constexpr unsigned Threads = 8;
template<typename T>
static constexpr T AlignTo(T value, T align)
{
return (value + (align - 1)) & ~(align - 1);
}
bool CreateDefaultTexture(const ComPtr<ID3D12Device3>& device,
const D3D12_RESOURCE_DESC& desc, ComPtr<ID3D12Resource>& resource);
bool CreateDefaultBuffer(const ComPtr<ID3D12Device3>& device,
UINT64 size, ComPtr<ID3D12Resource>& resource);
}
class CComputeEffect : public CPostProcessEffect
{
protected:
ComPtr<ID3D12RootSignature> m_rootSignature;
ComPtr<ID3D12PipelineState> m_pso;
ComPtr<ID3D12DescriptorHeap> m_descHeap;
ComPtr<ID3D12Resource> m_dst;
unsigned m_threadsX = 0;
unsigned m_threadsY = 0;
bool InitCompute(const ComPtr<ID3D12Device3>& device,
const D3D12_DESCRIPTOR_RANGE * ranges, UINT rangeCount,
const D3D12_STATIC_SAMPLER_DESC * samplers, UINT samplerCount,
const char * shader);
void Bind(const ComPtr<ID3D12GraphicsCommandList>& commandList);
void TransitionDst(const ComPtr<ID3D12GraphicsCommandList>& commandList,
D3D12_RESOURCE_STATES before, D3D12_RESOURCE_STATES after);
};

View File

@@ -0,0 +1,278 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "CDownsampleEffect.h"
#include "CDebug.h"
#include "config/CSettings.h"
#include <algorithm>
#include <cmath>
#include <cwchar>
#include <cwctype>
#include <cstring>
using namespace PostProcessUtil;
bool CDownsampleEffect::ParseRules(const std::wstring& value)
{
m_rules.clear();
if (value.empty())
return false;
size_t pos = 0;
while (pos < value.size())
{
size_t comma = value.find(L',', pos);
std::wstring token = value.substr(pos,
comma == std::wstring::npos ? std::wstring::npos : comma - pos);
while (!token.empty() && std::iswspace(token.front()))
token.erase(token.begin());
while (!token.empty() && std::iswspace(token.back()))
token.pop_back();
if (!token.empty())
{
Rule rule = {};
const wchar_t * start = token.c_str();
if (*start == L'>')
{
rule.greater = true;
++start;
}
if (swscanf_s(start, L"%ux%u:%ux%u",
&rule.x, &rule.y, &rule.targetX, &rule.targetY) != 4)
{
DEBUG_ERROR("Unable to parse IDD downsample rule");
m_rules.clear();
return false;
}
DEBUG_INFO("idd:downsample rule: %ux%u -> %ux%u%s",
rule.x, rule.y, rule.targetX, rule.targetY,
rule.greater ? " (greater-than)" : "");
m_rules.push_back(rule);
}
if (comma == std::wstring::npos)
break;
pos = comma + 1;
}
return !m_rules.empty();
}
const CDownsampleEffect::Rule * CDownsampleEffect::MatchRule(
unsigned width, unsigned height) const
{
const Rule * match = nullptr;
for (const auto& rule : m_rules)
if (( rule.greater && (width > rule.x || height > rule.y)) ||
(!rule.greater && (width == rule.x && height == rule.y)))
match = &rule;
return match;
}
bool CDownsampleEffect::Init(const ComPtr<ID3D12Device3>& device)
{
if (!ParseRules(g_settings.ReadStringValue(L"Downsample")))
return false;
D3D12_STATIC_SAMPLER_DESC sampler = {};
sampler.Filter = D3D12_FILTER_MIN_MAG_MIP_LINEAR;
sampler.AddressU = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
sampler.AddressV = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
sampler.AddressW = D3D12_TEXTURE_ADDRESS_MODE_CLAMP;
sampler.ComparisonFunc = D3D12_COMPARISON_FUNC_NEVER;
sampler.MaxLOD = D3D12_FLOAT32_MAX;
sampler.ShaderRegister = 0;
sampler.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
D3D12_DESCRIPTOR_RANGE ranges[3] = {};
ranges[0].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_CBV;
ranges[0].NumDescriptors = 1;
ranges[0].BaseShaderRegister = 0;
ranges[0].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[1].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
ranges[1].NumDescriptors = 1;
ranges[1].BaseShaderRegister = 0;
ranges[1].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[2].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
ranges[2].NumDescriptors = 1;
ranges[2].BaseShaderRegister = 0;
ranges[2].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
const char * shader =
"cbuffer Constants : register(b0)\n"
"{\n"
" float Width;\n"
" float Height;\n"
"};\n"
"Texture2D <float4> src : register(t0);\n"
"RWTexture2D<float4> dst : register(u0);\n"
"SamplerState ss : register(s0);\n"
"[numthreads(" POST_PROCESS_THREADS_STR ", " POST_PROCESS_THREADS_STR ", 1)]\n"
"void main(uint3 dt : SV_DispatchThreadID)\n"
"{\n"
" dst[dt.xy] = src.SampleLevel(ss,\n"
" float2(\n"
" (float(dt.x) + 0.5f) / Width,\n"
" (float(dt.y) + 0.5f) / Height),\n"
" 0);\n"
"}\n";
if (!InitCompute(device, ranges, ARRAYSIZE(ranges), &sampler, 1, shader))
return false;
D3D12_HEAP_PROPERTIES heapProps = {};
heapProps.Type = D3D12_HEAP_TYPE_UPLOAD;
D3D12_RESOURCE_DESC desc = {};
desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
desc.Width = AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
desc.Height = 1;
desc.DepthOrArraySize = 1;
desc.MipLevels = 1;
desc.SampleDesc.Count = 1;
desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
HRESULT hr = device->CreateCommittedResource(&heapProps,
D3D12_HEAP_FLAG_NONE, &desc, D3D12_RESOURCE_STATE_GENERIC_READ,
nullptr, IID_PPV_ARGS(&m_constBuffer));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create Downsample constant buffer");
return false;
}
return true;
}
PostProcessStatus CDownsampleEffect::SetFormat(
const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst)
{
const Rule * rule = MatchRule((unsigned)src.desc.Width, src.desc.Height);
if (!rule ||
(rule->targetX == src.desc.Width && rule->targetY == src.desc.Height))
return PostProcessStatus::BYPASS_EFFECT;
D3D12_RESOURCE_DESC desc = src.desc;
desc.Width = rule->targetX;
desc.Height = rule->targetY;
desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
if (!CreateDefaultTexture(device, desc, m_dst))
return PostProcessStatus::FAILED;
m_consts.width = (float)rule->targetX;
m_consts.height = (float)rule->targetY;
void * data = nullptr;
D3D12_RANGE readRange = { 0, 0 };
HRESULT hr = m_constBuffer->Map(0, &readRange, &data);
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to map Downsample constant buffer");
return PostProcessStatus::FAILED;
}
std::memcpy(data, &m_consts, sizeof(m_consts));
m_constBuffer->Unmap(0, nullptr);
m_threadsX = ((unsigned)desc.Width + (Threads - 1)) / Threads;
m_threadsY = ((unsigned)desc.Height + (Threads - 1)) / Threads;
m_format = src.desc.Format;
m_scaleX = (double)desc.Width / src.desc.Width;
m_scaleY = (double)desc.Height / src.desc.Height;
m_width = (unsigned)desc.Width;
m_height = desc.Height;
dst.desc = desc;
dst.width = (unsigned)desc.Width;
dst.height = desc.Height;
return PostProcessStatus::SUCCESS;
}
void CDownsampleEffect::AdjustDamage(RECT dirtyRects[], unsigned * nbDirtyRects)
{
for (RECT * rect = dirtyRects; rect < dirtyRects + *nbDirtyRects; ++rect)
{
unsigned width = (unsigned)std::ceil((double)(rect->right - rect->left) * m_scaleX);
unsigned height = (unsigned)std::ceil((double)(rect->bottom - rect->top ) * m_scaleY);
rect->left = (LONG)max(0.0, std::floor((double)rect->left * m_scaleX));
rect->right = (LONG)min((double)m_width , (double)rect->left + width);
rect->top = (LONG)max(0.0, std::floor((double)rect->top * m_scaleY));
rect->bottom = (LONG)min((double)m_height, (double)rect->top + height);
if (rect->left > 0 ) rect->left -= 1;
if (rect->top > 0 ) rect->top -= 1;
if (rect->right < (LONG)m_width ) rect->right += 1;
if (rect->bottom < (LONG)m_height ) rect->bottom += 1;
}
}
ComPtr<ID3D12Resource> CDownsampleEffect::Run(
const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects)
{
UNREFERENCED_PARAMETER(dirtyRects);
UNREFERENCED_PARAMETER(nbDirtyRects);
TransitionDst(commandList, D3D12_RESOURCE_STATE_COMMON,
D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
D3D12_CPU_DESCRIPTOR_HANDLE handle =
m_descHeap->GetCPUDescriptorHandleForHeapStart();
const UINT inc = device->GetDescriptorHandleIncrementSize(
D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
D3D12_CONSTANT_BUFFER_VIEW_DESC cbvDesc = {};
cbvDesc.BufferLocation = m_constBuffer->GetGPUVirtualAddress();
cbvDesc.SizeInBytes = (UINT)AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
device->CreateConstantBufferView(&cbvDesc, handle);
handle.ptr += inc;
D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
srvDesc.Format = m_format;
srvDesc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
srvDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
srvDesc.Texture2D.MipLevels = 1;
device->CreateShaderResourceView(src.Get(), &srvDesc, handle);
handle.ptr += inc;
D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
uavDesc.Format = m_format;
uavDesc.ViewDimension = D3D12_UAV_DIMENSION_TEXTURE2D;
device->CreateUnorderedAccessView(m_dst.Get(), nullptr, &uavDesc, handle);
Bind(commandList);
commandList->Dispatch(m_threadsX, m_threadsY, 1);
TransitionDst(commandList, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
D3D12_RESOURCE_STATE_COMMON);
return m_dst;
}

View File

@@ -0,0 +1,71 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#pragma once
#include "CComputeEffect.h"
#include <vector>
#include <string>
class CDownsampleEffect : public CComputeEffect
{
private:
struct Rule
{
bool greater = false;
unsigned x = 0;
unsigned y = 0;
unsigned targetX = 0;
unsigned targetY = 0;
};
struct Consts
{
float width;
float height;
} m_consts = {};
std::vector<Rule> m_rules;
ComPtr<ID3D12Resource> m_constBuffer;
DXGI_FORMAT m_format = DXGI_FORMAT_UNKNOWN;
double m_scaleX = 1.0;
double m_scaleY = 1.0;
unsigned m_width = 0;
unsigned m_height = 0;
bool ParseRules(const std::wstring& value);
const Rule * MatchRule(unsigned width, unsigned height) const;
public:
const char * GetName() const override { return "Downsample"; }
bool Init(const ComPtr<ID3D12Device3>& device);
PostProcessStatus SetFormat(const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst) override;
void AdjustDamage(RECT dirtyRects[], unsigned * nbDirtyRects) override;
ComPtr<ID3D12Resource> Run(const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects) override;
};

View File

@@ -0,0 +1,192 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "CHDR16to10Effect.h"
#include "CDebug.h"
#include <cstring>
using namespace PostProcessUtil;
bool CHDR16to10Effect::Init(const ComPtr<ID3D12Device3>& device)
{
D3D12_DESCRIPTOR_RANGE ranges[3] = {};
ranges[0].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_CBV;
ranges[0].NumDescriptors = 1;
ranges[0].BaseShaderRegister = 0;
ranges[0].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[1].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
ranges[1].NumDescriptors = 1;
ranges[1].BaseShaderRegister = 0;
ranges[1].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[2].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
ranges[2].NumDescriptors = 1;
ranges[2].BaseShaderRegister = 0;
ranges[2].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
const char * shader =
"cbuffer Constants : register(b0)\n"
"{\n"
" float ReferenceWhiteNits;\n"
"};\n"
"Texture2D<float4> src : register(t0);\n"
"RWTexture2D<float4> dst : register(u0);\n"
"static const float PQ_m1 = 0.1593017578125;\n"
"static const float PQ_m2 = 78.84375;\n"
"static const float PQ_c1 = 0.8359375;\n"
"static const float PQ_c2 = 18.8515625;\n"
"static const float PQ_c3 = 18.6875;\n"
"[numthreads(" POST_PROCESS_THREADS_STR ", " POST_PROCESS_THREADS_STR ", 1)]\n"
"void main(uint3 dt : SV_DispatchThreadID)\n"
"{\n"
" float3 linearValue = src[dt.xy].rgb * ReferenceWhiteNits;\n"
" // scRGB uses BT.709 primaries whereas HDR10/PQ output uses BT.2020.\n"
" // Rotate the gamut in linear light (BT.2087) BEFORE applying the PQ\n"
" // curve, otherwise the BT.709 values are later reinterpreted as BT.2020\n"
" // and saturated colours (most visibly red) are pushed outside their\n"
" // intended gamut.\n"
" float3 rec2020 = float3(\n"
" dot(linearValue, float3(0.6274039, 0.3292830, 0.0433131)),\n"
" dot(linearValue, float3(0.0690973, 0.9195404, 0.0113623)),\n"
" dot(linearValue, float3(0.0163914, 0.0880133, 0.8955953)));\n"
" // scRGB to PQ (ST.2084)\n"
" float3 Y = rec2020 / 10000.0;\n"
" float3 Ym1 = pow(max(Y, 0.0), PQ_m1);\n"
" float3 pq = pow((PQ_c1 + PQ_c2 * Ym1) / (1.0 + PQ_c3 * Ym1), PQ_m2);\n"
" dst[dt.xy] = float4(pq, src[dt.xy].a);\n"
"}\n";
if (!InitCompute(device, ranges, ARRAYSIZE(ranges), nullptr, 0, shader))
return false;
D3D12_HEAP_PROPERTIES heapProps = {};
heapProps.Type = D3D12_HEAP_TYPE_UPLOAD;
D3D12_RESOURCE_DESC desc = {};
desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
desc.Width = AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
desc.Height = 1;
desc.DepthOrArraySize = 1;
desc.MipLevels = 1;
desc.SampleDesc.Count = 1;
desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
HRESULT hr = device->CreateCommittedResource(&heapProps,
D3D12_HEAP_FLAG_NONE, &desc, D3D12_RESOURCE_STATE_GENERIC_READ,
nullptr, IID_PPV_ARGS(&m_constBuffer));
if (FAILED(hr))
{
DEBUG_ERROR_HR(hr, "Failed to create HDR16to10 constant buffer");
return false;
}
void * data = nullptr;
D3D12_RANGE readRange = { 0, 0 };
hr = m_constBuffer->Map(0, &readRange, &data);
if (FAILED(hr))
return false;
std::memcpy(data, &m_consts, sizeof(m_consts));
m_constBuffer->Unmap(0, nullptr);
return true;
}
PostProcessStatus CHDR16to10Effect::SetFormat(
const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst)
{
if (src.desc.Format != DXGI_FORMAT_R16G16B16A16_FLOAT || !src.hdr)
return PostProcessStatus::BYPASS_EFFECT;
D3D12_RESOURCE_DESC desc = src.desc;
desc.Format = DXGI_FORMAT_R10G10B10A2_UNORM;
desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
if (!CreateDefaultTexture(device, desc, m_dst))
return PostProcessStatus::FAILED;
m_threadsX = ((unsigned)desc.Width + (Threads - 1)) / Threads;
m_threadsY = ((unsigned)desc.Height + (Threads - 1)) / Threads;
dst.desc = desc;
dst.format = FRAME_TYPE_RGBA10;
dst.hdr = true;
dst.hdrPQ = true;
// Gamut conversion changes the signal's container primaries to BT.2020, but
// does not change the mastering display chromaticities described by ST 2086.
dst.hdrMetadata = src.hdrMetadata;
memcpy(dst.displayPrimary, src.displayPrimary, sizeof(dst.displayPrimary));
memcpy(dst.whitePoint , src.whitePoint , sizeof(dst.whitePoint ));
dst.maxDisplayLuminance = src.maxDisplayLuminance;
dst.minDisplayLuminance = src.minDisplayLuminance;
dst.maxContentLightLevel = src.maxContentLightLevel;
dst.maxFrameAverageLightLevel = src.maxFrameAverageLightLevel;
dst.sdrWhiteLevel = src.sdrWhiteLevel;
return PostProcessStatus::SUCCESS;
}
ComPtr<ID3D12Resource> CHDR16to10Effect::Run(
const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects)
{
UNREFERENCED_PARAMETER(dirtyRects);
UNREFERENCED_PARAMETER(nbDirtyRects);
TransitionDst(commandList, D3D12_RESOURCE_STATE_COMMON,
D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
D3D12_CPU_DESCRIPTOR_HANDLE handle =
m_descHeap->GetCPUDescriptorHandleForHeapStart();
const UINT inc = device->GetDescriptorHandleIncrementSize(
D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
D3D12_CONSTANT_BUFFER_VIEW_DESC cbvDesc = {};
cbvDesc.BufferLocation = m_constBuffer->GetGPUVirtualAddress();
cbvDesc.SizeInBytes = (UINT)AlignTo(sizeof(m_consts),
(size_t)D3D12_CONSTANT_BUFFER_DATA_PLACEMENT_ALIGNMENT);
device->CreateConstantBufferView(&cbvDesc, handle);
handle.ptr += inc;
D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
srvDesc.Format = DXGI_FORMAT_R16G16B16A16_FLOAT;
srvDesc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
srvDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
srvDesc.Texture2D.MipLevels = 1;
device->CreateShaderResourceView(src.Get(), &srvDesc, handle);
handle.ptr += inc;
D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
uavDesc.Format = DXGI_FORMAT_R10G10B10A2_UNORM;
uavDesc.ViewDimension = D3D12_UAV_DIMENSION_TEXTURE2D;
device->CreateUnorderedAccessView(m_dst.Get(), nullptr, &uavDesc, handle);
Bind(commandList);
commandList->Dispatch(m_threadsX, m_threadsY, 1);
TransitionDst(commandList, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
D3D12_RESOURCE_STATE_COMMON);
return m_dst;
}

View File

@@ -0,0 +1,46 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#pragma once
#include "CComputeEffect.h"
class CHDR16to10Effect : public CComputeEffect
{
private:
struct Consts
{
float ReferenceWhiteNits; // scRGB reference white in nits (typically 80)
} m_consts = { 80.0f };
ComPtr<ID3D12Resource> m_constBuffer;
public:
const char * GetName() const override { return "HDR16to10"; }
bool Init(const ComPtr<ID3D12Device3>& device);
PostProcessStatus SetFormat(const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst) override;
ComPtr<ID3D12Resource> Run(const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects) override;
};

View File

@@ -0,0 +1,577 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "CRGB24Effect.h"
#include "CDebug.h"
#include "config/CSettings.h"
#include "common/LGMPConfig.h"
#include <algorithm>
#include <climits>
#include <cstdint>
#include <cstring>
using namespace PostProcessUtil;
static_assert(D3D12_TEXTURE_DATA_PITCH_ALIGNMENT == 256,
"RGB24 shader row alignment must match D3D12");
struct CRGB24Effect::State
{
enum class Phase
{
DISABLED,
NATIVE_WARMUP,
NATIVE_SAMPLE,
PACKED_WARMUP,
PACKED_SAMPLE,
LOCKED_NATIVE,
LOCKED_PACKED,
};
struct FormatKey
{
D3D12_RESOURCE_DIMENSION resourceDimension = D3D12_RESOURCE_DIMENSION_UNKNOWN;
UINT64 resourceWidth = 0;
UINT resourceHeight = 0;
DXGI_FORMAT resourceFormat = DXGI_FORMAT_UNKNOWN;
unsigned width = 0;
unsigned height = 0;
FrameType format = FRAME_TYPE_INVALID;
bool hdr = false;
bool hdrPQ = false;
std::shared_ptr<const D12ColorTransform> colorTransform;
};
static const unsigned WarmupCount = LGMP_Q_FRAME_LEN;
static const unsigned SampleCount = 64;
static const unsigned TrimCount = SampleCount / 8;
SRWLOCK lock = SRWLOCK_INIT;
Phase phase = Phase::DISABLED;
FormatKey format = {};
bool formatValid = false;
uint64_t generation = 0;
unsigned warmups = 0;
unsigned sampleCount = 0;
uint64_t nativeMean = 0;
uint64_t samples[SampleCount] = {};
static bool IsEligible(const D12FrameFormat& format)
{
if (format.hdr ||
format.desc.Dimension != D3D12_RESOURCE_DIMENSION_TEXTURE2D ||
format.desc.Format != DXGI_FORMAT_B8G8R8A8_UNORM)
return false;
if (!format.colorTransform ||
(!format.colorTransform->matrixEnabled &&
!format.colorTransform->lutEnabled))
return true;
return IsIdentityColorTransform(*format.colorTransform);
}
bool WantsPackedLocked() const
{
return phase == Phase::PACKED_WARMUP ||
phase == Phase::PACKED_SAMPLE ||
phase == Phase::LOCKED_PACKED;
}
bool IsBenchmarkingLocked() const
{
return phase == Phase::NATIVE_WARMUP ||
phase == Phase::NATIVE_SAMPLE ||
phase == Phase::PACKED_WARMUP ||
phase == Phase::PACKED_SAMPLE;
}
void ResetStageLocked()
{
warmups = 0;
sampleCount = 0;
std::memset(samples, 0, sizeof(samples));
}
uint64_t TrimmedMeanLocked()
{
std::sort(samples, samples + SampleCount);
uint64_t total = 0;
for (unsigned i = TrimCount; i < SampleCount - TrimCount; ++i)
total += samples[i];
return total / (SampleCount - TrimCount * 2);
}
void Update(const D12FrameFormat& next)
{
AcquireSRWLockExclusive(&lock);
const bool formatChanged = !formatValid ||
format.resourceDimension != next.desc.Dimension ||
format.resourceWidth != next.desc.Width ||
format.resourceHeight != next.desc.Height ||
format.resourceFormat != next.desc.Format ||
format.width != next.width ||
format.height != next.height ||
format.format != next.format ||
format.hdr != next.hdr ||
format.hdrPQ != next.hdrPQ ||
format.colorTransform != next.colorTransform;
if (formatChanged)
{
format.resourceDimension = next.desc.Dimension;
format.resourceWidth = next.desc.Width;
format.resourceHeight = next.desc.Height;
format.resourceFormat = next.desc.Format;
format.width = next.width;
format.height = next.height;
format.format = next.format;
format.hdr = next.hdr;
format.hdrPQ = next.hdrPQ;
format.colorTransform = next.colorTransform;
formatValid = true;
phase = IsEligible(next) ?
Phase::NATIVE_WARMUP : Phase::DISABLED;
nativeMean = 0;
++generation;
ResetStageLocked();
}
switch (phase)
{
case Phase::NATIVE_WARMUP:
if (warmups >= WarmupCount)
{
phase = Phase::NATIVE_SAMPLE;
++generation;
ResetStageLocked();
}
break;
case Phase::NATIVE_SAMPLE:
if (sampleCount >= SampleCount)
{
nativeMean = TrimmedMeanLocked();
phase = Phase::PACKED_WARMUP;
++generation;
ResetStageLocked();
}
break;
case Phase::PACKED_WARMUP:
if (warmups >= WarmupCount)
{
phase = Phase::PACKED_SAMPLE;
++generation;
ResetStageLocked();
}
break;
case Phase::PACKED_SAMPLE:
if (sampleCount >= SampleCount)
{
const uint64_t packedMean = TrimmedMeanLocked();
const uint64_t relativeThreshold = nativeMean / 20;
const uint64_t threshold = relativeThreshold > 50000ULL ?
relativeThreshold : 50000ULL;
// Prefer the bandwidth saving unless native is meaningfully faster.
const bool usePacked = packedMean <= nativeMean ||
packedMean - nativeMean <= threshold;
DEBUG_INFO(
"RGB24 benchmark: native=%llu us, packed=%llu us, selected=%s",
(unsigned long long)(nativeMean / 1000),
(unsigned long long)(packedMean / 1000),
usePacked ? "packed" : "native");
phase = usePacked ?
Phase::LOCKED_PACKED : Phase::LOCKED_NATIVE;
++generation;
ResetStageLocked();
}
break;
default:
break;
}
ReleaseSRWLockExclusive(&lock);
}
bool WantsPacked()
{
AcquireSRWLockShared(&lock);
const bool result = WantsPackedLocked();
ReleaseSRWLockShared(&lock);
return result;
}
bool IsBenchmarking()
{
AcquireSRWLockShared(&lock);
const bool result = IsBenchmarkingLocked();
ReleaseSRWLockShared(&lock);
return result;
}
uint64_t GetTimingToken(bool packed)
{
AcquireSRWLockShared(&lock);
const uint64_t result = IsBenchmarkingLocked() &&
packed == WantsPackedLocked() ? generation : 0;
ReleaseSRWLockShared(&lock);
return result;
}
void Reject()
{
AcquireSRWLockExclusive(&lock);
if (WantsPackedLocked())
{
phase = Phase::LOCKED_NATIVE;
++generation;
ResetStageLocked();
}
ReleaseSRWLockExclusive(&lock);
}
void RecordTiming(uint64_t token, bool fullCopy, uint64_t totalTime)
{
AcquireSRWLockExclusive(&lock);
if (token == generation && fullCopy)
switch (phase)
{
case Phase::NATIVE_WARMUP:
case Phase::PACKED_WARMUP:
++warmups;
break;
case Phase::NATIVE_SAMPLE:
case Phase::PACKED_SAMPLE:
if (sampleCount < SampleCount)
samples[sampleCount++] = totalTime;
break;
default:
break;
}
ReleaseSRWLockExclusive(&lock);
}
};
bool CRGB24Effect::Init(const ComPtr<ID3D12Device3>& device)
{
if (!g_settings.ReadBoolValue(L"AllowRGB24", true))
return false;
D3D12_DESCRIPTOR_RANGE ranges[2] = {};
ranges[0].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
ranges[0].NumDescriptors = 1;
ranges[0].BaseShaderRegister = 0;
ranges[0].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
ranges[1].RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
ranges[1].NumDescriptors = 1;
ranges[1].BaseShaderRegister = 0;
ranges[1].OffsetInDescriptorsFromTableStart = D3D12_DESCRIPTOR_RANGE_OFFSET_APPEND;
const char * shader =
"Texture2D<float4> src : register(t0);\n"
"RWByteAddressBuffer dst : register(u0);\n"
"[numthreads(" POST_PROCESS_THREADS_STR ", " POST_PROCESS_THREADS_STR ", 1)]\n"
"void main(uint3 dt : SV_DispatchThreadID)\n"
"{\n"
" uint width, height;\n"
" src.GetDimensions(width, height);\n"
" uint rowBytes = width * 3;\n"
" uint dataWidth = ((rowBytes + 255) & ~255u) / 4;\n"
" if (dt.x >= dataWidth || dt.y >= height)\n"
" return;\n"
" uint rowOffset = dt.x * 4;\n"
" if (rowOffset >= rowBytes)\n"
" {\n"
" dst.Store((dt.y * dataWidth + dt.x) * 4, 0);\n"
" return;\n"
" }\n"
" uint firstX = (dt.x * 4) / 3;\n"
" uint secondX = firstX + 1;\n"
" float4 color0 = src[uint2(firstX, dt.y)];\n"
" float4 color3 = secondX < width ?\n"
" src[uint2(secondX, dt.y)] : 0.0f;\n"
" uint xmod3 = dt.x % 3;\n"
" float4 color1 = xmod3 <= 1 ? color0 : color3;\n"
" float4 color2 = xmod3 == 0 ? color0 : color3;\n"
" float4 packed = float4(\n"
" color0.bgr[xmod3], color1.grb[xmod3],\n"
" color2.rbg[xmod3], color3.bgr[xmod3]);\n"
" uint4 bytes = (uint4)(saturate(packed) * 255.0f + 0.5f);\n"
" uint value = bytes.x | (bytes.y << 8) |\n"
" (bytes.z << 16) | (bytes.w << 24);\n"
" dst.Store((dt.y * dataWidth + dt.x) * 4, value);\n"
"}\n";
if (!InitCompute(device, ranges, ARRAYSIZE(ranges), nullptr, 0, shader))
return false;
m_state = std::make_shared<State>();
return true;
}
void CRGB24Effect::ShareState(const CPostProcessEffect& other)
{
m_state = static_cast<const CRGB24Effect&>(other).m_state;
}
void CRGB24Effect::Update(const D12FrameFormat& format)
{
m_state->Update(format);
}
bool CRGB24Effect::NeedsReconfigure() const
{
return m_state->WantsPacked() != Enabled;
}
bool CRGB24Effect::RequiresFullDamage() const
{
return m_state->IsBenchmarking();
}
uint64_t CRGB24Effect::GetTimingToken() const
{
return m_state->GetTimingToken(Enabled);
}
void CRGB24Effect::RecordTiming(
uint64_t token, bool fullCopy, uint64_t totalTime)
{
m_state->RecordTiming(token, fullCopy, totalTime);
}
bool CRGB24Effect::GetCopyLayout(
unsigned * pitch, unsigned * dataHeight) const
{
*pitch = m_pitch;
*dataHeight = m_height;
return true;
}
static void GetCopySpan(const RECT& rect, unsigned width,
unsigned pitch, UINT64 * left, UINT64 * right)
{
// Damage stays in logical pixels until the packed copy is recorded.
*left = ((UINT64)rect.left * 3) & ~3ULL;
if (rect.left == 0 && rect.right == (LONG)width)
*right = pitch;
else
*right = ((UINT64)rect.right * 3 + 3) & ~3ULL;
}
bool CRGB24Effect::ShouldCopyFully(
const RECT dirtyRects[], unsigned nbDirtyRects) const
{
static const unsigned commandLimit = 256;
const uint64_t frameSize = (uint64_t)m_pitch * m_height;
uint64_t copiedBytes = 0;
unsigned commands = 0;
for (const RECT * rect = dirtyRects;
rect < dirtyRects + nbDirtyRects; ++rect)
{
UINT64 left;
UINT64 right;
GetCopySpan(*rect, m_width, m_pitch, &left, &right);
const unsigned rows = (unsigned)(rect->bottom - rect->top);
if (left == 0 && right == m_pitch)
{
++commands;
copiedBytes += (uint64_t)rows * m_pitch;
}
else
{
commands += rows;
copiedBytes += (right - left) * rows;
}
if (commands > commandLimit || copiedBytes >= frameSize)
return true;
}
return false;
}
void CRGB24Effect::CopyFrame(
const ComPtr<ID3D12GraphicsCommandList>& commandList,
ID3D12Resource * dst, ID3D12Resource * src,
const RECT dirtyRects[], unsigned nbDirtyRects, bool fullCopy) const
{
if (fullCopy)
{
commandList->CopyBufferRegion(
dst, 0, src, 0, (UINT64)m_pitch * m_height);
return;
}
for (const RECT * rect = dirtyRects;
rect < dirtyRects + nbDirtyRects; ++rect)
{
UINT64 left;
UINT64 right;
GetCopySpan(*rect, m_width, m_pitch, &left, &right);
if (left == 0 && right == m_pitch)
{
const UINT64 offset = (UINT64)rect->top * m_pitch;
const UINT64 size =
(UINT64)(rect->bottom - rect->top) * m_pitch;
commandList->CopyBufferRegion(dst, offset, src, offset, size);
continue;
}
for (LONG y = rect->top; y < rect->bottom; ++y)
{
const UINT64 offset = (UINT64)y * m_pitch + left;
commandList->CopyBufferRegion(
dst, offset, src, offset, right - left);
}
}
}
PostProcessStatus CRGB24Effect::SetFormat(const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst)
{
if (!m_state->WantsPacked())
{
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
if (src.desc.Dimension != D3D12_RESOURCE_DIMENSION_TEXTURE2D ||
src.desc.Format != DXGI_FORMAT_B8G8R8A8_UNORM ||
src.hdr)
{
DEBUG_WARN("RGB24 packing is unavailable for the current format");
m_state->Reject();
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
if (src.desc.Width >
(UINT64_MAX - (D3D12_TEXTURE_DATA_PITCH_ALIGNMENT - 1)) / 3)
{
m_state->Reject();
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
const UINT64 packedPitch = AlignTo<UINT64>(
src.desc.Width * 3, D3D12_TEXTURE_DATA_PITCH_ALIGNMENT);
if (!src.desc.Height || packedPitch > LONG_MAX ||
packedPitch > UINT64_MAX / src.desc.Height)
{
m_state->Reject();
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
const UINT64 bufferSize = packedPitch * src.desc.Height;
const UINT64 maxUAVSize =
(1ULL << D3D12_REQ_BUFFER_RESOURCE_TEXEL_COUNT_2_TO_EXP) *
sizeof(uint32_t);
if (bufferSize > UINT32_MAX || bufferSize > maxUAVSize)
{
m_state->Reject();
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
if (!CreateDefaultBuffer(device, bufferSize, m_dst))
{
m_state->Reject();
m_dst.Reset();
return PostProcessStatus::BYPASS_EFFECT;
}
const unsigned dataWidth = (unsigned)(packedPitch / 4);
m_threadsX = (dataWidth + (Threads - 1)) / Threads;
m_threadsY = (src.desc.Height + (Threads - 1)) / Threads;
m_width = (unsigned)src.desc.Width;
m_height = src.desc.Height;
m_pitch = (unsigned)packedPitch;
dst.desc = m_dst->GetDesc();
dst.dataWidth = dataWidth;
dst.dataHeight = src.desc.Height;
dst.pitch = m_pitch;
dst.format = FRAME_TYPE_BGR_32;
return PostProcessStatus::SUCCESS;
}
ComPtr<ID3D12Resource> CRGB24Effect::Run(const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects)
{
UNREFERENCED_PARAMETER(dirtyRects);
UNREFERENCED_PARAMETER(nbDirtyRects);
TransitionDst(commandList, D3D12_RESOURCE_STATE_COMMON,
D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
D3D12_CPU_DESCRIPTOR_HANDLE handle =
m_descHeap->GetCPUDescriptorHandleForHeapStart();
const UINT inc = device->GetDescriptorHandleIncrementSize(
D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
D3D12_SHADER_RESOURCE_VIEW_DESC srvDesc = {};
srvDesc.Format = DXGI_FORMAT_B8G8R8A8_UNORM;
srvDesc.ViewDimension = D3D12_SRV_DIMENSION_TEXTURE2D;
srvDesc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
srvDesc.Texture2D.MipLevels = 1;
device->CreateShaderResourceView(src.Get(), &srvDesc, handle);
handle.ptr += inc;
D3D12_UNORDERED_ACCESS_VIEW_DESC uavDesc = {};
uavDesc.Format = DXGI_FORMAT_R32_TYPELESS;
uavDesc.ViewDimension = D3D12_UAV_DIMENSION_BUFFER;
uavDesc.Buffer.NumElements = (UINT)(m_dst->GetDesc().Width / 4);
uavDesc.Buffer.StructureByteStride = 0;
uavDesc.Buffer.CounterOffsetInBytes = 0;
uavDesc.Buffer.Flags = D3D12_BUFFER_UAV_FLAG_RAW;
device->CreateUnorderedAccessView(m_dst.Get(), nullptr, &uavDesc, handle);
Bind(commandList);
commandList->Dispatch(m_threadsX, m_threadsY, 1);
TransitionDst(commandList, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
D3D12_RESOURCE_STATE_COMMON);
return m_dst;
}

View File

@@ -0,0 +1,62 @@
/**
* Looking Glass
* Copyright © 2017-2026 The Looking Glass Authors
* https://looking-glass.io
*
* This program is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License as published by the Free
* Software Foundation; either version 2 of the License, or (at your option)
* any later version.
*
* This program is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
* more details.
*
* You should have received a copy of the GNU General Public License along
* with this program; if not, write to the Free Software Foundation, Inc., 59
* Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#pragma once
#include "CComputeEffect.h"
class CRGB24Effect : public CComputeEffect
{
private:
struct State;
std::shared_ptr<State> m_state;
unsigned m_width = 0;
unsigned m_height = 0;
unsigned m_pitch = 0;
public:
const char * GetName() const override { return "RGB24"; }
bool Init(const ComPtr<ID3D12Device3>& device);
void ShareState(const CPostProcessEffect& other) override;
void Update(const D12FrameFormat& format) override;
bool NeedsReconfigure() const override;
bool RequiresFullDamage() const override;
uint64_t GetTimingToken() const override;
void RecordTiming(
uint64_t token, bool fullCopy, uint64_t totalTime) override;
bool GetCopyLayout(
unsigned * pitch, unsigned * dataHeight) const override;
bool ShouldCopyFully(
const RECT dirtyRects[], unsigned nbDirtyRects) const override;
void CopyFrame(
const ComPtr<ID3D12GraphicsCommandList>& commandList,
ID3D12Resource * dst, ID3D12Resource * src,
const RECT dirtyRects[], unsigned nbDirtyRects,
bool fullCopy) const override;
PostProcessStatus SetFormat(const ComPtr<ID3D12Device3>& device,
const D12FrameFormat& src, D12FrameFormat& dst) override;
ComPtr<ID3D12Resource> Run(const ComPtr<ID3D12Device3>& device,
const ComPtr<ID3D12GraphicsCommandList>& commandList,
const ComPtr<ID3D12Resource>& src, RECT dirtyRects[],
unsigned * nbDirtyRects) override;
};