[rocm] F.embedding reports invalid configuration argument #130806
Description
🐛 Describe the bug
F.embedding will crash with relatively large tensor input on the AMD GPU:
input = torch.randint(low=0, high=16032, size=[131072], device="cuda")
w = torch.randn([16032, 16384], device="cuda")
torch.nn.functional.embedding(input, w)
RuntimeError: HIP error: invalid configuration argument
Versions
top of tree
cc @jeffdaily @sunway513 @jithunnair-amd @pruthvistony @ROCmSupport @dllehr-amd @jataylo @hongxiayang
Metadata
Metadata
Assignees
Labels
Type
Projects
Status