frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 12:54:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

[amd][lowering] hipify shim v2 headers (#134689)

Browse Source 
Summary: The default c_shim version was switched to 2 for HIP in D60674018. This results in some linking errors where shim function symbols are missing from the compiled .so file (eg. P1551186492) when building lowering benchmark scripts since the required files aren't included. Hipify the shim v2 generated header files as well since they're needed during codegen when the buck binaries are executed.

Reviewed By: frank-wei, zoranzhao, henryoier

Differential Revision: D61865202

Pull Request resolved: https://github.com/pytorch/pytorch/pull/134689
Approved by: https://github.com/zoranzhao
This commit is contained in:
Janet Yang
2024-08-28 21:53:24 +00:00
committed by PyTorch MergeBot
parent 7b3da5f297
commit 26ec06e45d
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
build.bzl
View File

@ -76,7 +76,7 @@ def define_targets(rules):
] + (["--static_dispatch_backend CPU"] if rules.is_cpu_static_dispatch_build() else []))
gen_aten_outs_cuda = (
GENERATED_H_CUDA + GENERATED_CPP_CUDA +
GENERATED_H_CUDA + GENERATED_CPP_CUDA + GENERATED_AOTI_CUDA_CPP +
aten_ufunc_generated_cuda_sources()
)
@ -320,5 +320,8 @@ GENERATED_AUTOGRAD_CPP = [
GENERATED_AOTI_CPP = [
"torch/csrc/inductor/aoti_torch/generated/c_shim_cpu.cpp",
]
GENERATED_AOTI_CUDA_CPP = [
"torch/csrc/inductor/aoti_torch/generated/c_shim_cuda.cpp",
]