GoaLitiuM/FlaxEngine
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Refactor UAV slots binding when rendering from PS into UAV

Browse Source
This commit is contained in:
Wojtek Figat
2021-10-07 14:15:45 +02:00
parent bdd2dbb719
commit 88cf1bcd28
9 changed files with 38 additions and 116 deletions
Download Patch File Download Diff File Expand all files Collapse all files

99
Source/Engine/GraphicsDevice/DirectX/DX11/GPUContextDX11.cpp
View File

@@ -49,7 +49,6 @@ GPUContextDX11::GPUContextDX11(GPUDeviceDX11* device, ID3D11DeviceContext* conte
, _omDirtyFlag(false)
, _rtCount(0)
, _rtDepth(nullptr)
, _uaOutput(nullptr)
, _srDirtyFlag(false)
, _uaDirtyFlag(false)
, _cbDirtyFlag(false)
@@ -61,9 +60,9 @@ GPUContextDX11::GPUContextDX11(GPUDeviceDX11* device, ID3D11DeviceContext* conte
#endif
// Only DirectX 11 supports more than 1 UAV
_maxUASlotsForCS = GPU_MAX_UA_BINDED;
_maxUASlots = GPU_MAX_UA_BINDED;
if (_device->GetRendererType() != RendererType::DirectX11)
_maxUASlotsForCS = 1;
_maxUASlots = 1;
}
GPUContextDX11::~GPUContextDX11()
@@ -86,7 +85,6 @@ void GPUContextDX11::FrameBegin()
_rtCount = 0;
_currentState = nullptr;
_rtDepth = nullptr;
_uaOutput = nullptr;
Platform::MemoryClear(_rtHandles, sizeof(_rtHandles));
Platform::MemoryClear(_srHandles, sizeof(_srHandles));
Platform::MemoryClear(_uaHandles, sizeof(_uaHandles));
@@ -180,12 +178,11 @@ void GPUContextDX11::ClearUA(GPUBuffer* buf, const Vector4& value)
void GPUContextDX11::ResetRenderTarget()
{
if (_rtCount != 0 || _uaOutput || _rtDepth)
if (_rtCount != 0 || _rtDepth)
{
_omDirtyFlag = true;
_rtCount = 0;
_rtDepth = nullptr;
_uaOutput = nullptr;
Platform::MemoryClear(_rtHandles, sizeof(_rtHandles));
@@ -200,13 +197,12 @@ void GPUContextDX11::SetRenderTarget(GPUTextureView* rt)
ID3D11RenderTargetView* rtv = rtDX11 ? rtDX11->RTV() : nullptr;
int32 newRtCount = rtv ? 1 : 0;
if (_rtCount != newRtCount || _rtHandles[0] != rtv || _rtDepth != nullptr || _uaOutput)
if (_rtCount != newRtCount || _rtHandles[0] != rtv || _rtDepth != nullptr)
{
_omDirtyFlag = true;
_rtCount = newRtCount;
_rtDepth = nullptr;
_rtHandles[0] = rtv;
_uaOutput = nullptr;
}
}
@@ -219,13 +215,12 @@ void GPUContextDX11::SetRenderTarget(GPUTextureView* depthBuffer, GPUTextureView
ID3D11DepthStencilView* dsv = depthBufferDX11 ? depthBufferDX11->DSV() : nullptr;
int32 newRtCount = rtv ? 1 : 0;
if (_rtCount != newRtCount || _rtHandles[0] != rtv || _rtDepth != dsv || _uaOutput)
if (_rtCount != newRtCount || _rtHandles[0] != rtv || _rtDepth != dsv)
{
_omDirtyFlag = true;
_rtCount = newRtCount;
_rtDepth = dsv;
_rtHandles[0] = rtv;
_uaOutput = nullptr;
}
}
@@ -244,35 +239,15 @@ void GPUContextDX11::SetRenderTarget(GPUTextureView* depthBuffer, const Span<GPU
}
int32 rtvsSize = sizeof(ID3D11RenderTargetView*) * rts.Length();
if (_rtCount != rts.Length() || _rtDepth != dsv || _uaOutput || Platform::MemoryCompare(_rtHandles, rtvs, rtvsSize) != 0)
if (_rtCount != rts.Length() || _rtDepth != dsv || Platform::MemoryCompare(_rtHandles, rtvs, rtvsSize) != 0)
{
_omDirtyFlag = true;
_rtCount = rts.Length();
_rtDepth = dsv;
_uaOutput = nullptr;
Platform::MemoryCopy(_rtHandles, rtvs, rtvsSize);
}
}
void GPUContextDX11::SetRenderTarget(GPUTextureView* rt, GPUBuffer* uaOutput)
{
auto rtDX11 = reinterpret_cast<GPUTextureViewDX11*>(rt);
auto uaOutputDX11 = reinterpret_cast<GPUBufferDX11*>(uaOutput);
ID3D11RenderTargetView* rtv = rtDX11 ? rtDX11->RTV() : nullptr;
ID3D11UnorderedAccessView* uav = uaOutputDX11 ? ((GPUBufferViewDX11*)uaOutputDX11->View())->UAV() : nullptr;
int32 newRtCount = rtv ? 1 : 0;
if (_rtCount != newRtCount || _rtHandles[0] != rtv || _rtDepth != nullptr || _uaOutput != uav)
{
_omDirtyFlag = true;
_rtCount = newRtCount;
_rtDepth = nullptr;
_rtHandles[0] = rtv;
_uaOutput = uav;
}
}
void GPUContextDX11::ResetSR()
{
_srDirtyFlag = false;
@@ -291,7 +266,8 @@ void GPUContextDX11::ResetUA()
_uaDirtyFlag = false;
Platform::MemoryClear(_uaHandles, sizeof(_uaHandles));
_context->CSSetUnorderedAccessViews(0, _maxUASlotsForCS, _uaHandles, nullptr);
_context->CSSetUnorderedAccessViews(0, _maxUASlots, _uaHandles, nullptr);
_context->OMSetRenderTargetsAndUnorderedAccessViews(D3D11_KEEP_RENDER_TARGETS_AND_DEPTH_STENCIL, nullptr, nullptr, 0, 0, nullptr, nullptr);
}
void GPUContextDX11::ResetCB()
@@ -791,10 +767,8 @@ void GPUContextDX11::CopySubresource(GPUResource* dstResource, uint32 dstSubreso
void GPUContextDX11::flushSRVs()
{
// Check if need to flush shader resources
if (_srDirtyFlag)
{
// Clear flag
_srDirtyFlag = false;
// Flush with the driver
@@ -816,24 +790,23 @@ void GPUContextDX11::flushSRVs()
void GPUContextDX11::flushUAVs()
{
// Check if need to flush unordered access
if (_uaDirtyFlag)
{
// Clear flag
_uaDirtyFlag = false;
// Flush with the driver
uint32 initialCounts[GPU_MAX_UA_BINDED] = { 0 };
_context->CSSetUnorderedAccessViews(0, _maxUASlotsForCS, _uaHandles, initialCounts);
if (CurrentCS)
_context->CSSetUnorderedAccessViews(0, _maxUASlots, _uaHandles, initialCounts);
else
_context->OMSetRenderTargetsAndUnorderedAccessViews(D3D11_KEEP_RENDER_TARGETS_AND_DEPTH_STENCIL, nullptr, nullptr, _rtCount, _maxUASlots - _rtCount, _uaHandles + _rtCount, initialCounts);
}
}
void GPUContextDX11::flushCBs()
{
// Check if need to flush constant buffers
if (_cbDirtyFlag)
{
// Clear flag
_cbDirtyFlag = false;
// Flush with the driver
@@ -855,47 +828,29 @@ void GPUContextDX11::flushCBs()
void GPUContextDX11::flushOM()
{
// Check if need to flush output merger state or/and unordered access views
if (_omDirtyFlag)
{
#if _DEBUG
// Validate binded render targets amount
int32 rtCount = 0;
for (int i = 0; i < ARRAY_COUNT(_rtHandles) && i < _rtCount; i++)
{
if (_rtHandles[i] != nullptr)
rtCount++;
else
break;
}
ASSERT(rtCount == _rtCount);
#endif
// Check if don't use UAVs and set output merger render targets table
if (_uaOutput == nullptr)
{
_context->OMSetRenderTargets(_rtCount, _rtHandles, _rtDepth);
}
else
{
// Note: current dx11 content implementation assumes that there must be one or more render targets binded in order to use uav output
ASSERT(_rtCount > 0);
uint32 initialCounts[1] = { 0 };
// TODO: set -1 if buffer had no ResetCounter call since last time
_context->OMSetRenderTargetsAndUnorderedAccessViews(_rtCount, _rtHandles, _rtDepth, _rtCount, 1, &_uaOutput, initialCounts);
}
// Clear flag
_omDirtyFlag = false;
int32 uaCount = 0;
for (int32 i = _maxUASlots - 1; i >= 0; i--)
{
if (_uaHandles[i])
{
uaCount = i + 1;
break;
}
}
// Flush with the driver
if (uaCount > 0)
_context->OMSetRenderTargetsAndUnorderedAccessViews(_rtCount, _rtHandles, _rtDepth, 0, D3D11_KEEP_UNORDERED_ACCESS_VIEWS, nullptr, nullptr);
else
_context->OMSetRenderTargets(_rtCount, _rtHandles, _rtDepth);
}
}
void GPUContextDX11::onDrawCall()
{
ASSERT(_currentState);
// Flush
flushCBs();
flushSRVs();
flushUAVs();

4
Source/Engine/GraphicsDevice/DirectX/DX11/GPUContextDX11.h
View File

@@ -23,14 +23,13 @@ private:
#if GPU_ALLOW_PROFILE_EVENTS
ID3DUserDefinedAnnotation* _userDefinedAnnotations;
#endif
int32 _maxUASlotsForCS;
int32 _maxUASlots;
// Output Merger
bool _omDirtyFlag;
int32 _rtCount;
ID3D11DepthStencilView* _rtDepth;
ID3D11RenderTargetView* _rtHandles[GPU_MAX_RT_BINDED];
ID3D11UnorderedAccessView* _uaOutput;
// Shader Resources
bool _srDirtyFlag;
@@ -113,7 +112,6 @@ public:
void SetRenderTarget(GPUTextureView* rt) override;
void SetRenderTarget(GPUTextureView* depthBuffer, GPUTextureView* rt) override;
void SetRenderTarget(GPUTextureView* depthBuffer, const Span<GPUTextureView*>& rts) override;
void SetRenderTarget(GPUTextureView* rt, GPUBuffer* uaOutput) override;
void ResetSR() override;
void ResetUA() override;
void ResetCB() override;