projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 5f66348) | patch
Separate SerializeToCubin from GPUTransforms.
authorArtem Belevich <tra@google.com>
Fri, 17 Mar 2023 20:47:46 +0000 (13:47 -0700)
committerArtem Belevich <tra@google.com>
Fri, 17 Mar 2023 21:08:53 +0000 (14:08 -0700)
commitfea8faa705b43e23508bc9fef842c62929786110
tree4cbfddecda2ff967181df6e8eda0729f9630aa67tree | snapshot
parent5f66348e59aa7ce5e5780a972b3875268c45d57ccommit | diff
Separate SerializeToCubin from GPUTransforms.

SerializeToCubin depends on CUDA at *runtime* which is undesirable for MLIR's
general use case, as compilation should be doable on any host, regardless of
whether it has a GPU.

SerializeToCubin is needed to run some GPU tests, so when we build mlir-opt,
SerializeToCubin pass is linked in directly into mlir-opt.

Differential Revision: https://reviews.llvm.org/D146330
utils/bazel/llvm-project-overlay/mlir/BUILD.bazel diff | blob | history
Domain: System / Toolchain;