[SOLVED] Convert NV12 to BGR by NVIDIA Performance Primitives

Convert NV12 to BGR by NVIDIA Performance Primitives

I'm trying to convert NV12 image to BGR by npp, but in the final array i have zeroes.

int lumaStepBytes, chromaStepBytes;
int rgbStepBytes;
      
auto dpNV12LumaFrame = nppiMalloc_8u_C1(dec.GetWidth(), dec.GetHeight(), &lumaStepBytes);
auto dpNV12ChromaFrame = nppiMalloc_8u_C1(dec.GetWidth(), dec.GetChromaHeight(), &chromaStepBytes);
auto dpBGRFrame = nppiMalloc_8u_C3(dec.GetWidth(), dec.GetHeight(), &rgbStepBytes);
            
cudaMemcpy2D(dpNV12LumaFrame, lumaStepBytes, pFrame, dec.GetWidth(),
             dec.GetWidth(), dec.GetHeight(), cudaMemcpyKind::cudaMemcpyHostToDevice);
cudaMemcpy2D(dpNV12ChromaFrame, chromaStepBytes, pFrame + dec.GetLumaPlaneSize(), dec.GetWidth(),
             dec.GetWidth(), dec.GetChromaHeight(), cudaMemcpyKind::cudaMemcpyHostToDevice);

Npp8u *planesAddres[2];
planesAddres[0] = dpNV12LumaFrame;
planesAddres[1] = dpNV12ChromaFrame;

nppiNV12ToBGR_8u_P2C3R(planesAddres, lumaStepBytes,
                       dpBGRFrame, rgbStepBytes,
                       {dec.GetWidth(), dec.GetHeight()});

res.m_data.resize(dec.GetWidth() * dec.GetHeight() * 3);
cudaMemcpy2D(res.m_data.data(), dec.GetWidth(), dpBGRFrame, rgbStepBytes,
             dec.GetWidth(), dec.GetHeight(), cudaMemcpyKind::cudaMemcpyDeviceToHost);
nppiFree(dpBGRFrame);
nppiFree(dpNV12ChromaFrame);
nppiFree(dpNV12LumaFrame);

dec is a video decoder which gives pFrame in NV12 format and provide additional information about that, like offsets, dimensions, NV12 planes, etc. The same result I have if I use cu... and cuda... functions for allocating without alignment.

Do anybody have any ideas about the problem?

Solution

the problem went away when I changed the video card. GeForce 740 to 1080