Lines Matching full:rocm

3 HIP (ROCm) semantics
6 ROCm\ |trade| is AMD’s open source software platform for GPU-accelerated high
7 performance computing and machine learning. HIP is ROCm's C++ dialect designed
24     cuda0 = torch.device('cuda:0')  # 'rocm' or 'hip' are not valid, use 'cuda'
77 TensorFloat-32(TF32) on ROCm
80 TF32 is not supported on ROCm.
90 ``rocm-smi``. You can use :meth:`~torch.cuda.memory_allocated` and
120 Currently, only the "nccl" and "gloo" backends for torch.distributed are supported on ROCm.
135 ``#if defined(CUDA_VERSION) && CUDA_VERSION >= 11000`` to implicitly exclude ROCm/HIP,
137 use the following to not take the code path for ROCm/HIP:
141 Alternatively, if it is desired to take the code path for ROCm/HIP:
145 Or if it is desired to take the code path for ROCm/HIP only for specific HIP versions:
159 Kernel asserts are supported on ROCm, but they are disabled due to performance overhead. It can be …