tree 8be628b382d8df58813967a6b532b002c6c3a04c
parent 6120aa3718737dac685eb779ccc4539d4ba059d2
author Rachel Guo <guorachel@meta.com> 1719602178 +0000
committer PyTorch MergeBot <pytorchmergebot@users.noreply.github.com> 1719602178 +0000

[AOTI] Properly indent launchKernel calls in AOTInductor (#129616)

Summary:
There is a small cosmetic issue in the C++ wrapper file generated by AOTInductor - The launchKernel() call isn't properly indented.

Added indentation for launchKernel() code block call when there's a "if" condition. a.k.a when `grid_uses_symbolic_shapes` is `True`.

Test Plan:
Test cmd ran (in pytorch oss):

`TORCH_LOGS="output_code" TORCH_COMPILE_DEBUG=1 python test/inductor/test_aot_inductor.py -k test_zero_grid_with_backed_symbols_abi_compatible_cuda`

And then manually verified the output code generated in a path like
`/tmp/torchinductor_guorachel/coraisesuchpl3qabrazn7ydydszcit6lwpn7ckd3b4wej4rep5l/cba5g5ajeh5sym3tp5iqn7kkokimj7qqd4krs2rruhupbfqgppge.cpp`

Similarly, also verified for test case:`test_zero_grid_with_unbacked_symbols_abi_compatible_cuda`

Differential Revision: D58897157

Pull Request resolved: https://github.com/pytorch/pytorch/pull/129616
Approved by: https://github.com/ColinPeppler