I know we've talked about AMD GPU support a lot and so far it has not worked out well. Maybe their new "HIP" in the Open Compute / Boltzmann / ROCm 1.3 initiative could be worth a new try? Usig it to port CAFFE from CUDA to AMD-something was presumably surprisingly easy and faster than OpenCL out of the box.

MrS

---

Scanning for our furry friends since Jan 2002

Multiscale Lab, the lab of gpugrid, some times ago published this tool: SWAN, porting cuda to opencl...

Maybe their new "HIP" in the Open Compute / Boltzmann / ROCm 1.3 initiative could be worth a new try?

Now ROCm is at 1.5 version.....(with full support to opencl)

Apparently there is another way to compile CUDA codes to OpenCL. I understand the project team is pretty busy keeping up with the current crunching power, though.

MrS

---

Scanning for our furry friends since Jan 2002

Maybe their new "HIP" in the Open Compute / Boltzmann / ROCm 1.3 initiative could be worth a new try?

With the latest version (1.4) there are a lot of new features, bugfix, documentation, etc
HIP

Very interesting
rocRAND

Another interesting news about converting Cuda into OpenCl
LLVM Cuda to HIP

With the latest version (1.4) there are a lot of new features, bugfix, documentation, etc
HIP

Uh, i forget the new version of HIP (1.5) with new features

- Support threadIdx, blockIdx, blockDim directly (no need for hipify conversions in kernels.) HIP Kernel syntax is now identical to CUDA kernel syntax - no need for extra parms or conversions.
- Remove need for manual destructors
- Provide printf in device code
- Support for globals when using module API
- hipify-clang now supports using newer versions of clang
- HIP texture support equivalent to CUDA texture driver APIs

etc

Another interesting news about converting Cuda into OpenCl
LLVM Cuda to HIP

Another big step to cuda-opencl conversion tool
HypSYCL

ReSYCLator seems to be interesting.
And, also, an interesting article about Cuda, OpenCl and Sycl

First release of HipCl

Sycl on Cuda

Will it work with a Wrapper (ACEMD3)? That could pose some issues.

Sycl on Cuda

Will it work with a Wrapper (ACEMD3)? That could pose some issues.

Why not? With this big batch, Amd hardware, maybe, is welcome.

Another big step: hipCl 0.9

You can win 1000$ (and others prizes) porting CUDA code to SYCL: The Great Cross Architecture Challenge

AMD released Orochi to dynamic runtime switching between radeon HIP & NVIDIA CUDA