HIP – Convert CUDA to Portable C++

Re:HIP – Convert CUDA to Portable C++ by Wehrdo on Sunday March 18, 2018 @05:36PM (#16614535)

Could anyone who's used HIP comment on how it compares to programming in raw CUDA?

I considered using it for a project recently, but ultimately decided against it because I didn't need to be able to run on AMD systems.

[ Parent | on HN ]

Re:HIP – Convert CUDA to Portable C++ by waynecochran on Sunday March 18, 2018 @06:47PM (#16614948)

Wait ... if you write your program in HIP you get both CUDA and OpenCL (which includes AMD) for free right?

[ Parent | on HN ]

Re:HIP – Convert CUDA to Portable C++ by dragandj on Sunday March 18, 2018 @06:53PM (#16614979)

There is no OpenCL in the story there. AMD created HIP as a direct copy of CUDA. It then gets compiled to work on AMD hardware, or can be translated to CUDA.

AMD also supports OpenCL (which I prefer to both CUDA and HIP), but it's not connected to HIP.

[ Parent | on HN ]

Re:HIP – Convert CUDA to Portable C++ by TomV1971 on Sunday March 18, 2018 @07:27PM (#16615134)

Last time I checked, HIP didn’t support reading from textures.

That’s something that’s not only useful for pure graphics.

So it is (or was?) not a straight substitute for CUDA.

[ Parent | on HN ]