HIP
HIP copied to clipboard
ROCm
HIP: C++ Heterogeneous-Compute Interface for Portability
Is unified memory management in HIP supported? You said in https://github.com/ROCm-Developer-Tools/HIP/blob/master/docs/markdown/hip_faq.md that managed memory is not supported yet, otherwise in HIP/src/hip_memory.cpp the function hipMallocManaged() is implemented. If I use hipMallocManaged()...
For an example: To create an object lib for the nvcc compilation used as cuda_wrap_srcs add_subdirectory(cuda_app) cuda_include_directories(${PROJECT_SOURCE_DIR}/include) cuda_wrap_srcs(app OBJ app_cuda_backend_obj cuda_app/cuZFP.cu) app_cuda_backend_obj object created for cuda_app/cuZFP.cu . Example: https://github.com/LLNL/blt/issues/75 similarly...
Are Lambda Expressions in hipLaunchKernelGGL supported? During device code assembling I always have a segmentation fault with following kernel: ``` #define accelerator_forNB( iterator, num, nsimd, ... ) \ { \...
I have a cuda file and compiled it with **nvcc -cuda --output-file kernel.cpp kernel.cu** which gives me cpp file which generates the code which can be compiled by host compiler...
CUDA 9 `__shfl_sync` function is missing. I can use the deprecated `__shfl` but it would be be better to have the new function. Test code: ``` __global__ static void shflTest(int...
Is it possible to get the file `dump.kernel_input.ll` without following compilation procedure like the generation of host_binary / hsaco_binary. In other words, instead of `KMDUMPLLVM=1 hipcc main.cc -o useless.out`, I...
Regardless of HIP's ability to compile and on Nvidia machines with a negligible performance penalty on Nvidia machines, some developers prefer to keep both CUDA and HIP code in the...
I am attempting to port the QUDA library to HIP. I shall document here and unconverted references as I find them. I'm using cuda 9.2.148 and hip/1.5-cuda9 on SUMMIT.
hipMemPrefetchAsync is missing. I can live without that but it is nice to have the feature.
I'm undertaking a HIP port of the QUDA library. The strategy is to remove all CUDA specific data types and replace them with agnostic QUDA types, which will be converted...