How to convert the memory requested by new and malloc into the memory that cuda can use by device

[pangxie]

After I use a third-party library for audio decoding, I need to perform mixing calculation on the decoded audio data. I want to implement the mixing calculation process in cuda. However, the decoded memory is requested using malloc, and cuda cannot use it. How can I convert the memory requested by malloc to the memory that cuda can use without requesting new memory?

[AastaLLL]

Hi,

To access via GPU, the memory need to be pinned memory and cannot be paged out.
So you will need to allocate it with cudaMallocHost.

For other memory types, please check the following document:

CUDA for Tegra :: CUDA Toolkit Documentation

Thanks.