How to improve the copying performance between video memory and system memory?

____2 · ‎05-13-2015

I got a problem when using Intel media SDK.

When my video decode program was running, I copied YUV vedio frame data from vedio memory(sdk's frame data struct) to system memory (my buffer,in system memory) by function memcpy(), but got a low speed( only 92 frames per sec.only copy operation, and 66 fps display on D3D9 at the sametime). That won't be a satisfactory to video processing because i need some time to make other instant special effects on video frames.

I did it like this:

m_pMFXAllocator->LockFrame(frame->Data.MemId, &frame->Data);
   memcpy(m_gpu_Y, frame->Data.Y, frame->Info.Height * frame->Data.Pitch);
   memcpy(m_gpu_UV, frame->Data.UV, frame->Info.Height * frame->Data.Pitch / 2);
   mfxU16 pitch = frame->Data.Pitch;
   m_pMFXAllocator->UnlockFrame(frame->Data.MemId, &frame->Data);

How can I quicken the copying speed? And I hope know more information in detail about the copy performance in this procedure.

Thanks!

Surbhi_M_Intel · ‎05-19-2015

Hi there,

I see you have another post for same question(https://software.intel.com/en-us/forums/topic/557837) which is being addressed by my colleague. So I am closing this post to avoid confusion.

Thanks,
Surbhi