Please sign in to comment.
No longer use ffast-math (#2960)
Summary: Regardless of whether we are compiling a model in Release or Debug mode, we should try to generate the same binary. Prior to this PR, for example the GraphOptimizer may have quantized a Constant Tensor payload slightly differently due to `ffast-math` when in Release vs. Debug. Pull Request resolved: #2960 Test Plan: All tests pass. Differential Revision: D15473529 Pulled By: jfix71 fbshipit-source-id: a1b21cb50d347de4ee0920f94135aad9b63b051d
- Loading branch information