Question

I am trying to develop multi-streaming H.264 video player based on cudaDecodeD3D9 example from NVIDIA GPU Computing SDK 4.2.

Application works correctly with a few streams but it raises assertion (CUDA_ERROR_OUT_OF_MEMORY) in cuvidCreateDecoder function for 12 streams with resolution 800×600 or 9 streams with resolution 1920×1080. cudaMemGetInfo returns 387MB (for video card with 1GB) and 1.3Gb (for video card with 2GB) available memory. Does memory fragmentation cause this? How can I use available memory?

VideoDecoder::VideoDecoder(const CUVIDEOFORMAT & rVideoFormat, 
                       CUcontext &rContext, 
                       cudaVideoCreateFlags eCreateFlags, 
                       CUvideoctxlock &vidCtxLock) 
: m_VidCtxLock(vidCtxLock)
{
// get a copy of the CUDA context
m_Context          = rContext;
m_VideoCreateFlags = eCreateFlags;

// Fill the decoder-create-info struct from the given video-format struct.
memset(&oVideoDecodeCreateInfo_, 0, sizeof(CUVIDDECODECREATEINFO));
        // Create video decoder
oVideoDecodeCreateInfo_.CodecType           = rVideoFormat.codec;
oVideoDecodeCreateInfo_.ulWidth             = rVideoFormat.coded_width;
oVideoDecodeCreateInfo_.ulHeight            = rVideoFormat.coded_height;
oVideoDecodeCreateInfo_.ulNumDecodeSurfaces = FrameQueue::cnMaximumSize;

        // Limit decode memory to 24MB (16M pixels at 4:2:0 = 24M bytes)
while (oVideoDecodeCreateInfo_.ulNumDecodeSurfaces * rVideoFormat.coded_width * rVideoFormat.coded_height > 16*1024*1024)
{
    oVideoDecodeCreateInfo_.ulNumDecodeSurfaces--;
}
oVideoDecodeCreateInfo_.ChromaFormat        = rVideoFormat.chroma_format;
oVideoDecodeCreateInfo_.OutputFormat        = cudaVideoSurfaceFormat_NV12;
oVideoDecodeCreateInfo_.DeinterlaceMode     = cudaVideoDeinterlaceMode_Adaptive;

        // No scaling
oVideoDecodeCreateInfo_.ulTargetWidth       = oVideoDecodeCreateInfo_.ulWidth;
oVideoDecodeCreateInfo_.ulTargetHeight      = oVideoDecodeCreateInfo_.ulHeight;
oVideoDecodeCreateInfo_.ulNumOutputSurfaces = MAX_FRAME_COUNT;  // We won't simultaneously map more than 8 surfaces
oVideoDecodeCreateInfo_.ulCreationFlags     = m_VideoCreateFlags;
oVideoDecodeCreateInfo_.vidLock             = m_VidCtxLock;

size_t available, total;
cudaMemGetInfo(&available, &total);

        // create the decoder
CUresult oResult = cuvidCreateDecoder(&oDecoder_, &oVideoDecodeCreateInfo_);
assert(CUDA_SUCCESS == oResult);
}

Can cuvidCreateDecoder work with resolution over 1920×1080? When I try 2560×1920 stream cuvidCreateDecoder asserts CUDA_ERROR_INVALID_SOURCE.

My environment

• Hardware: NVidia GTX 550 Ti 1Gb, NVidia GT 610 2Gb, driver version 306.23
• Windows 7 x64
• Visual Studio 2010 SP1
• Windows SDK 7.1
• NVIDIA GPU Computing Toolkit v.4.2, v.5.0
• NVIDIA GPU Computing SDK 4.2.

Answer 1

For the memory issue, see this answer.

For the question on resolution, Compute Capability 2.0 and earlier GPUs do not support larger than HD resolution for cudaDecodeD3D9. This is why you can’t decode a 2560×1920 stream.

Kepler GPUs can support much larger resolutions.