Program Listing for File directx-utilities.h

↰ Return to documentation for file (include\utility\directx-utilities.h)

/******************************************************************************
© Intel Corporation.

This software and the related documents are Intel copyrighted materials,
and your use of them is governed by the express license under which they
were provided to you ("License"). Unless the License provides otherwise,
you may not use, modify, copy, publish, distribute, disclose or transmit
this software or the related documents without Intel's prior written
permission.


 This software and the related documents are provided as is, with no express
or implied warranties, other than those that are expressly stated in the
License.

******************************************************************************/
#pragma once

#include "assert.h"
#include "guid_hash.h"

#include <d3d9.h>
#include <d3d10_1.h>
#include <d3d11_4.h>
#include <d3d12.h>
#include <dxgi1_6.h>

#include <atlbase.h>

#include <cstdint>
#include <functional>
#include <map>
#include <memory>
#include <string>
#include <type_traits>
#include <unordered_map>
#include <vector>

// @brief Generic test for success on any directx creation call
#define UNSUCCESSFUL(hr) (FAILED(hr) || hr == S_FALSE)

void AssertIfFailed(HRESULT result);

namespace gpa {
namespace utility {
namespace directx {

CComPtr<ID3D12Device> GetDevice(ID3D12DeviceChild* deviceChild);

class DirectXPrivateData : virtual public IUnknown
{
public:
    virtual ~DirectXPrivateData();

    virtual HRESULT STDMETHODCALLTYPE QueryInterface(REFIID riid, void** ppvObject);
    virtual ULONG STDMETHODCALLTYPE AddRef();
    virtual ULONG STDMETHODCALLTYPE Release();
};

// {374D3F62-2ACE-4DF6-B605-9A62B2EA793A}
static const GUID sAssociatedCaptureKeyGUID = {0x374d3f62, 0x2ace, 0x4df6, {0xb6, 0x5, 0x9a, 0x62, 0xb2, 0xea, 0x79, 0x3a}};

// {90455112-A6EC-415F-8CE0-2830405824BF}
static const GUID sOutObjectCaptureKeyGUID = {0x90455112, 0xa6ec, 0x415f, {0x8c, 0xe0, 0x28, 0x30, 0x40, 0x58, 0x24, 0xbf}};

// {0ED71FEC-DF0F-4C69-A809-80FF4B8B53F8}
static const GUID sHwndSwapChainGUID = {0xed71fec, 0xdf0f, 0x4c69, {0xa8, 0x9, 0x80, 0xff, 0x4b, 0x8b, 0x53, 0xf8}};

// {2ED93005-E5AC-4EFC-90F7-091EE2D7884F}
static const GUID sExtensionResourceDataContainerGUID = {0x2ed93005, 0xe5ac, 0x4efc, {0x90, 0xf7, 0x9, 0x1e, 0xe2, 0xd7, 0x88, 0x4f}};

// {2ED93005-E5AC-4EFC-90F7-091EE2D6884F}
static const GUID sResourceAllocationInfoGUID = {0x2ed93005, 0xe5ac, 0x4efc, {0x90, 0xf7, 0x9, 0x1e, 0xe2, 0xd6, 0x88, 0x4f}};

template<class Type>
constexpr bool has_private_data =
    std::is_convertible<Type, ID3D11Device*>::value ||
    std::is_convertible<Type, ID3D11DeviceChild*>::value ||
    std::is_convertible<Type, IDXGIObject*>::value ||
    std::is_convertible<Type, ID3D12Object*>::value;

struct D3DResourceDesc
{
    uint32_t Width;
    uint32_t Height;
    uint32_t Depth;
    uint32_t MipLevels;
    uint32_t ArraySize;
    DXGI_FORMAT Format;
    DXGI_SAMPLE_DESC SampleDesc;

    uint32_t CompressedWidth;
    uint32_t CompressedHeight;
    uint32_t FormatStride;
};

struct D3D10ResourceDesc : public D3DResourceDesc
{
    D3D10_RESOURCE_DIMENSION Dimension;
    D3D10_USAGE Usage;
    uint32_t BindFlags;
    uint32_t CPUAccessFlags;
    uint32_t MiscFlags;
};

struct D3D11ResourceDesc : public D3DResourceDesc
{
    D3D11_RESOURCE_DIMENSION Dimension;
    D3D11_USAGE Usage;
    uint32_t BindFlags;
    uint32_t CPUAccessFlags;
    uint32_t MiscFlags;
};

struct D3D12ResourceDesc : public D3DResourceDesc
{
    D3D12_RESOURCE_DIMENSION Dimension;
    UINT64 Alignment;
    D3D12_TEXTURE_LAYOUT Layout;
    D3D12_RESOURCE_FLAGS Flags;
};

class D3DPrivateDataInterface : public IUnknown
{
public:
    virtual ~D3DPrivateDataInterface() = default;

private:
    HRESULT STDMETHODCALLTYPE QueryInterface(REFIID, void**) override
    {
        return E_NOINTERFACE;
    }

    ULONG STDMETHODCALLTYPE AddRef() override
    {
        return 1;
    }

    ULONG STDMETHODCALLTYPE Release() override
    {
        delete this;
        return 0;
    }
};

// @brief Basic implementation of IUnknown
//        Can be used with any class that derives from IUnknown ( specify IUnknownDerivedClass )
template<class IUnknownDerivedClass = IUnknown>
class IUnknownImpl : public IUnknownDerivedClass
{
    ULONG mRefCount = 1;

public:
    virtual ~IUnknownImpl() = default;

    virtual HRESULT STDMETHODCALLTYPE QueryInterface(REFIID riid, void** ppvObject)
    {
        if (!ppvObject) {
            return E_INVALIDARG;
        }

        if (InlineIsEqualGUID(riid, __uuidof(IUnknownDerivedClass)) || InlineIsEqualGUID(riid, __uuidof(IUnknown))) {
            ++mRefCount;
            *ppvObject = (void*)this;
            return S_OK;
        }

        return E_FAIL;
    }

    virtual ULONG STDMETHODCALLTYPE AddRef()
    {
        return ++mRefCount;
    }

    virtual ULONG STDMETHODCALLTYPE Release()
    {
        if (--mRefCount == 0) {
            delete this;
            return 0;
        }
        return mRefCount;
    }
};

// @brief Basic implementation of ID3D12Object
//        Can be used with any class that derives from ID3D12Object ( specify ID3D12ObjectDerivedClass )
template<class ID3D12ObjectDerivedClass = ID3D12Object>
class ID3D12ObjectImpl : public IUnknownImpl<ID3D12ObjectDerivedClass>
{
    struct PrivateDataEntry
    {
        std::vector<uint8_t> privateDataBlob;
        IUnknown* privateDataInterface = nullptr;
    };
    std::unordered_map<GUID, PrivateDataEntry> mPrivateData;

public:
    virtual HRESULT STDMETHODCALLTYPE GetPrivateData(REFGUID guid, UINT* pDataSize, void* pData)
    {
        if (!pDataSize) {
            return E_INVALIDARG;
        }

        auto found = mPrivateData.find(guid);
        if (found == mPrivateData.end()) {
            *pDataSize = 0;
            return DXGI_ERROR_NOT_FOUND;
        }

        PrivateDataEntry& foundEntry = found->second;
        uint32_t dataToCopySize = 0;
        void* dataToCopy = nullptr;
        if (foundEntry.privateDataInterface) {
            dataToCopySize = sizeof(IUnknown*);
            dataToCopy = &foundEntry.privateDataInterface;
            ASSERT(foundEntry.privateDataBlob.empty());
        } else {
            dataToCopySize = (uint32_t)foundEntry.privateDataBlob.size();
            dataToCopy = (void*)foundEntry.privateDataBlob.data();
        }

        uint32_t pDataAvailableSize = *pDataSize;
        *pDataSize = dataToCopySize;

        if (!pData) {
            return S_OK;
        }

        if (pDataAvailableSize < dataToCopySize) {
            return DXGI_ERROR_MORE_DATA;
        }

        memcpy(pData, dataToCopy, dataToCopySize);
        if (foundEntry.privateDataInterface) {
            foundEntry.privateDataInterface->AddRef();
        }

        return S_OK;
    }

    virtual HRESULT STDMETHODCALLTYPE SetPrivateData(REFGUID guid, UINT DataSize, const void* pData)
    {
        if (pData == nullptr) {
            // SetPrivateData with null pData returns S_FALSE, if entry had not been provided earlier
            return mPrivateData.erase(guid) ? S_OK : S_FALSE;
        }

        auto& entry = mPrivateData[guid];
        entry.privateDataBlob.resize(DataSize);
        memcpy(entry.privateDataBlob.data(), pData, DataSize);
        if (entry.privateDataInterface) {
            entry.privateDataInterface->Release();
            entry.privateDataInterface = nullptr;
        }
        return S_OK;
    }

    virtual HRESULT STDMETHODCALLTYPE SetPrivateDataInterface(REFGUID guid, const IUnknown* pData)
    {
        auto& entry = mPrivateData[guid];
        entry.privateDataBlob.clear();

        // No idea why the argument is const. Makes no sense, since the object needs to accept AddRef and Release calls
        IUnknown* mutablePData = const_cast<IUnknown*>(pData);
        if (mutablePData) {
            mutablePData->AddRef();
        }
        if (entry.privateDataInterface) {
            entry.privateDataInterface->Release();
        }
        entry.privateDataInterface = mutablePData;
        return S_OK;
    }

    virtual HRESULT STDMETHODCALLTYPE SetName(LPCWSTR Name)
    {
        return SetPrivateData(WKPDID_D3DDebugObjectNameW, (uint32_t)wcslen(Name), Name);
    }
};

/*
This class exists solely to facilitate deferred capture (and, by extension, subcapture)
of Intel D3D12 extensions. The extension mechanism can and does overwrite the data contained
in the resources it uses, so if we wait until a state capture is performed to extract resource
data from the driver, it's too late -- the data has already been altered. So for extension
resources only, we save a copy of the original data into one of these, which is attached to
the resource via SetPrivateDataInterface. If one of these is present during writing of the
state capture, we use this data instead of data obtained from the current state of the
resource.

Additionally, some extension scenarios (such as 64-bit emulated atomics) delete the
resource once the extension has been initialized, and so there would not be data available
at state capture time to copy down from the GPU, and so this object is used to keep the resource
data around for those extension scenarios as well.
*/
struct DataUpdate
{
    std::vector<uint8_t> data;
    uint64_t timestamp;
};

typedef std::vector<DataUpdate> DataUpdates;
typedef DataUpdates::iterator UpdateIter;
typedef DataUpdates::const_iterator UpdateIterConst;

class D3DResourceDataContainerImpl
{
public:
    D3DResourceDataContainerImpl(D3D12_RESOURCE_DESC const& desc);

    void AppendDataUpdate(uint64_t timestamp, void* data, size_t byteLen);
    size_t UpdateCount() const;
    D3D12_RESOURCE_DESC const& Desc() const;

    UpdateIter begin() { return mDataUpdates.begin(); }
    UpdateIterConst cbegin() const { return mDataUpdates.cbegin(); }
    UpdateIter end() { return mDataUpdates.end(); }
    UpdateIterConst cend() const { return mDataUpdates.cend(); }

private:
    DataUpdates mDataUpdates;
    D3D12_RESOURCE_DESC mDesc;
};

class D3DResourceDataContainer : public D3DPrivateDataInterface
{
public:
    D3DResourceDataContainer(D3D12_RESOURCE_DESC const& desc);
    std::shared_ptr<D3DResourceDataContainerImpl> mImpl;

    void AppendDataUpdate(uint64_t timestamp, void* data, size_t byteLen);
    size_t UpdateCount() const;
    D3D12_RESOURCE_DESC const& Desc() const;

    UpdateIter begin() { return mImpl->begin(); }
    UpdateIterConst cbegin() const { return mImpl->cbegin(); }
    UpdateIter end() { return mImpl->end(); }
    UpdateIterConst cend() const { return mImpl->cend(); }
};

template<typename DeviceType>
void DoWithDevice(ID3D12DeviceChild* object, std::function<void(DeviceType*)> callback)
{
    CComPtr<DeviceType> device;
    object->GetDevice(IID_PPV_ARGS(&device));
    if (device) {
        callback(device);
    }
}

// Unique identifier for a shader that can be used in Shader records.
// This is only valid for Ray generation shaders, Hit groups, Miss shaders and Callable shaders.
struct D3D12ShaderIdentifier
{
    uint64_t id[4] = {};

    explicit D3D12ShaderIdentifier(uint64_t* data);
    D3D12ShaderIdentifier& operator=(const D3D12ShaderIdentifier& other);
    bool operator<(const D3D12ShaderIdentifier& other) const;
    bool operator==(const D3D12ShaderIdentifier& other) const;
    bool operator!=(const D3D12ShaderIdentifier& other) const;
    explicit operator bool() const;
};
static_assert(D3D12_SHADER_IDENTIFIER_SIZE_IN_BYTES == sizeof(D3D12ShaderIdentifier), "Shader identifier size must be equal 32 bytes");

uint64_t GetCaptureKey(ID3D12Object* object, const GUID& guid);
uint64_t GetCaptureKey(ID3D12Object* object);
uint64_t GetCreatingObjectCaptureKey(ID3D12Device* device);

void GetSubresourceDesc(ID3D10Resource* resource, uint32_t subresource, D3D10ResourceDesc* desc);
void GetSubresourceDesc(ID3D11Resource* resource, uint32_t subresource, D3D11ResourceDesc* desc);
void GetSubresourceDesc(ID3D12Resource* resource, uint32_t subresource, D3D12ResourceDesc* desc);
size_t GetD3D11SubresourceDataSize(D3D11ResourceDesc const& desc, uint32_t rowPitch, uint32_t depthPitch);
size_t GetD3D11SubresourceDataSize(ID3D11Resource* resource, uint32_t subresource, uint32_t rowPitch, uint32_t depthPitch);
uint32_t CalculateMipLevels(uint32_t const width, uint32_t const height, uint32_t const depth);
uint32_t GetFormatBitCount(DXGI_FORMAT const format);
uint32_t GetFormatByteCount(DXGI_FORMAT const format);
uint32_t GetFormatStride(DXGI_FORMAT const format, uint32_t const width);
uint32_t GetFormatComponentSize(DXGI_FORMAT const format);
uint32_t Align(uint32_t value, uint32_t alignment);
uint64_t Align(uint64_t value, uint64_t alignment);
uint32_t CalculateSubresourceSize(uint32_t width, uint32_t height, uint32_t depth, DXGI_FORMAT format, uint32_t alignment);
bool IsFormatCompressed(DXGI_FORMAT const format);
void GetCompressedSize(uint32_t* width, uint32_t* height);
bool IsSRGBFormat(DXGI_FORMAT const format);
D3D12_RESOURCE_STATES TruncateResourceState(D3D12_COMMAND_LIST_TYPE commandListType, D3D12_RESOURCE_STATES state);
UINT GetVendorId(decltype(&CreateDXGIFactory1) createDXGIFactory1, ID3D12Device* device);

void CopyTextureRegion(decltype(&CreateDXGIFactory1) createDXGIFactory1, ID3D12Device* device, ID3D12GraphicsCommandList* commandList,
                       D3D12_TEXTURE_COPY_LOCATION* src, D3D12_TEXTURE_COPY_LOCATION* dst, D3D12_RESOURCE_DIMENSION dimension);

std::wstring GetObjectName(ID3D12Object* object);
std::wstring GetObjectName(ID3D11DeviceChild* object);

template<typename DescType>
inline uint32_t GetArraySize(DescType const* desc)
{
    return desc->ArraySize;
}
template<>
inline uint32_t GetArraySize(D3D10_BUFFER_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetArraySize(D3D11_BUFFER_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetArraySize(D3D10_TEXTURE3D_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetArraySize(D3D11_TEXTURE3D_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetArraySize(D3D11_TEXTURE3D_DESC1 const* /*desc*/)
{
    return 1;
}

template<typename DescType>
inline uint32_t GetMipLevels(DescType const* desc)
{
    return desc->MipLevels;
}
template<>
inline uint32_t GetMipLevels(D3D10_BUFFER_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetMipLevels(D3D11_BUFFER_DESC const* /*desc*/)
{
    return 1;
}

template<typename DescType>
inline uint32_t GetWidth(DescType const* desc)
{
    return desc->Width;
}
template<>
inline uint32_t GetWidth(D3D10_BUFFER_DESC const* desc)
{
    return desc->ByteWidth;
}
template<>
inline uint32_t GetWidth(D3D11_BUFFER_DESC const* desc)
{
    return desc->ByteWidth;
}

template<typename DescType>
inline uint32_t GetHeight(DescType const* desc)
{
    return desc->Height;
}
template<>
inline uint32_t GetHeight(D3D10_BUFFER_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetHeight(D3D11_BUFFER_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetHeight(D3D10_TEXTURE1D_DESC const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetHeight(D3D11_TEXTURE1D_DESC const* /*desc*/)
{
    return 1;
}

template<typename DescType>
inline uint32_t GetDepth(DescType const* /*desc*/)
{
    return 1;
}
template<>
inline uint32_t GetDepth(D3D10_TEXTURE3D_DESC const* desc)
{
    return desc->Depth;
}
template<>
inline uint32_t GetDepth(D3D11_TEXTURE3D_DESC const* desc)
{
    return desc->Depth;
}
template<>
inline uint32_t GetDepth(D3D11_TEXTURE3D_DESC1 const* desc)
{
    return desc->Depth;
}

template<typename DescType>
inline DXGI_FORMAT GetFormat(DescType const* desc)
{
    return desc->Format;
}
template<>
inline DXGI_FORMAT GetFormat(D3D10_BUFFER_DESC const* /*desc*/)
{
    return DXGI_FORMAT_R8_TYPELESS;
}
template<>
inline DXGI_FORMAT GetFormat(D3D11_BUFFER_DESC const* /*desc*/)
{
    return DXGI_FORMAT_R8_TYPELESS;
}

std::string ConvertPIXEventDataToString(UINT metadataType, const void* pData, UINT size);

template<typename DataType, typename ObjectType>
DataType* GetPrivateObject(ObjectType* object, bool force)
{
    static_assert(std::is_base_of<IUnknown, DataType>::value, "data must be inherited from IUnknown");
    static_assert(std::is_base_of<IUnknown, ObjectType>::value, "object must be inherited from IUnknown");

    if (!object) {
        return nullptr;
    }

    DataType* data = nullptr;
    UINT size = sizeof(data);
    HRESULT hr = object->GetPrivateData(__uuidof(DataType), &size, &data);
    if (FAILED(hr) && force) {
        data = new DataType(object);
        object->SetPrivateDataInterface(__uuidof(DataType), data);
    }
    return data;
}

inline size_t AlignTo(size_t value)
{
    return ((value + sizeof(uint64_t) - 1) / sizeof(uint64_t)) * sizeof(uint64_t);
};

struct RootSignaturePatchTable
{
    enum ParamType {
        INVALID = 0,
        DESCRIPTOR_TABLE = 1,
        RESOURCE_VIEW = 2,
    };

    struct PatchableParam
    {
        ParamType type;
        uint32_t offset;
        std::vector<D3D12_DESCRIPTOR_RANGE1> descriptorRanges;  // Available if type is DESCRIPTOR_TABLE
    };

    std::vector<PatchableParam> patchableParams;
};

class __declspec(uuid("483863D0-A5FC-4CED-A38D-3B093CF950DC"))
    D3D12StateObjectExportDataFromDDI : public D3DPrivateDataInterface
{
public:
    D3D12StateObjectExportDataFromDDI(ID3D12StateObject*) {}

    std::map<std::wstring, RootSignaturePatchTable>& GetExportNameToPatchTable()
    {
        return mExportNameToPatchTable;
    }

    std::map<std::wstring, RootSignaturePatchTable> const& GetExportNameToPatchTable() const
    {
        return mExportNameToPatchTable;
    }

private:
    std::map<std::wstring, RootSignaturePatchTable> mExportNameToPatchTable;
};

// Aligns the |address| to the nearest multiple of |alignment| then returns the number of alignment bytes needed for alignment.
D3D12_GPU_VIRTUAL_ADDRESS AlignVATo(D3D12_GPU_VIRTUAL_ADDRESS address, uint64_t alignment, uint64_t* bytesToAlign = nullptr);

// Helper to get pointer to D3D12_RAYTRACING_GEOMETRY_DESC regardless of the geometry layout specified in the AS input.
// https://microsoft.github.io/DirectX-Specs/d3d/Raytracing.html#d3d12_raytracing_geometry_desc
D3D12_RAYTRACING_GEOMETRY_DESC const* GetRaytracingGeometryDescAtIndex(const D3D12_BUILD_RAYTRACING_ACCELERATION_STRUCTURE_INPUTS& blasInputs, uint32_t indexOfGeometryDesc);

}  // namespace directx
}  // namespace utility
}  // namespace gpa

// Intel D3D12 extensions mark committed resource data as extension-
// specific by setting the top bit to 1 -- we need to filter that out
// when creating page tracking data structures
#define INTC_EXT_MASK 0x8000000000000000
#define INTC_EXT_FILTER(s) ((s) & ~(INTC_EXT_MASK))