Hands-On GPU Computing with Python

CUDA-like memory management with HIP

Just as conventional C++ uses standard C functions to handle memory, HIP also has memory-management functions that it uses in a similar manner. C++ code is referred to as host code, while HIP code is known as device code. The following functions are frequently used with HIP when developing or porting AMD/NVIDIA GPU programs and during code execution through device kernels:

 

CUDA version	HIP version of the same functions
cudaMalloc	hipMalloc
cudaMemcpy	hipMemcpy
cudaFree	hipFree

 

 

Similar to CUDA, hipMemcpyHostToDevice and hipMemcpyHostToDevice on HIP correspond to what cudaMemcpyHostToDevice and cudaMemcpyDeviceToHost are meant to do on CUDA.