Windows Desktop Duplication API taking a long time - c++

Question

I am using the windows desktop duplication API to record my screen in windows 10. I am however having some issues with performance. When playing a video using google chrome and attempting to record the time it takes to record the screen fluctuates from 15ms to 45ms. I wanted to be able to record at at least 30fps, and I know the desktop duplication api is capable of doing it. Anyways here is the code I used to actually capture the screen:
processor->hr = processor->lDeskDupl->AcquireNextFrame(0, &processor->lFrameInfo, &processor->lDesktopResource);
if (processor->hr == DXGI_ERROR_WAIT_TIMEOUT) {
processor->lDeskDupl->ReleaseFrame();
return false;
}
if (FAILED(processor->hr)) {
processor->lDeskDupl->ReleaseFrame();
return false;
}
// QI for ID3D11Texture2D
processor->hr = processor->lDesktopResource->QueryInterface(IID_PPV_ARGS(&processor->lAcquiredDesktopImage));
if (FAILED(processor->hr)) {
processor->lDeskDupl->ReleaseFrame();
return false;
}
processor->lDesktopResource.Release();
if (processor->lAcquiredDesktopImage == nullptr) {
processor->lDeskDupl->ReleaseFrame();
return false;
}
processor->lImmediateContext->CopyResource(processor->lGDIImage, processor->lAcquiredDesktopImage);
processor->lAcquiredDesktopImage.Release();
processor->lDeskDupl->ReleaseFrame();
// Copy image into CPU access texture
processor->lImmediateContext->CopyResource(processor->lDestImage, processor->lGDIImage);
// Copy from CPU access texture to bitmap buffer
D3D11_MAPPED_SUBRESOURCE resource;
processor->subresource = D3D11CalcSubresource(0, 0, 0);
processor->lImmediateContext->Map(processor->lDestImage, processor->subresource, D3D11_MAP_READ_WRITE, 0, &resource);
BYTE* sptr = reinterpret_cast<BYTE*>(resource.pData);
BYTE* dptr = processor->pBuf;
UINT lRowPitch = min(processor->lBmpRowPitch, resource.RowPitch);
for (int i = 0; i < processor->lOutputDuplDesc.ModeDesc.Height; i++) {
memcpy_s(dptr, processor->lBmpRowPitch, sptr, lRowPitch);
sptr += resource.RowPitch;
dptr += processor->lBmpRowPitch;
}
It is important to note that this is the specific section that is taking 15ms-45ms to complete every cycle. The memcpy loop at the bottom accounts for about 2ms of that time usually so I know that that is not responsible for the time it is taking here. Also AcquireNextFrame's timeout is set to zero so it returns nearly immediately. Any help would be greatly appreciated! The code pasted here was adapted from this: https://gist.github.com/Xirexel/a69ade44df0f70afd4a01c1c9d9e02cd

Answer 1

You're not using the API in optimal way. Read remarks in ReleaseFrame API documentation:
For performance reasons, we recommend that you release the frame just before you call the IDXGIOutputDuplication::AcquireNextFrame method to acquire the next frame. When the client does not own the frame, the operating system copies all desktop updates to the surface. This can result in wasted GPU cycles if the operating system updates the same region for each frame that occurs.
You are not doing what's written there, you release frames as soon as you copy.