Coerce to true_ctype in tensor_factory #7856
Conversation
|
Stack from ghstack (oldest at bottom): |
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/7856
Note: Links to docs will display an error until the docs builds have been completed. This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
swolchok
added
release notes: misc
| dynamism) {} | ||
| dynamism) { | ||
| // The only valid values for bool are 0 and 1; coerce! | ||
| if constexpr (std::is_same_v<true_ctype, bool>) { |
There was a problem hiding this comment.
It seems in ifdef aten section this transformation is unconditional while here it is applied only if it is bool. I dont fully understand.
There was a problem hiding this comment.
it doesn't build with bits16 and we don't have bit_cast, otherwise I would make it unconditional here and use bit_cast.
There was a problem hiding this comment.
I am guessing bit_cast not being usable is related to it available only in c++20.
There was a problem hiding this comment.
yes; we will have access to c10::bit_cast once I get back to code sharing but currently we don't have it.
Partial fix for #7748.
* Coerce to true_ctype in tensor_factory (#7856) This should fix the problem where attempts to test bool are often wonky in OSS and fail UBSAN internally; it is undefined behavior to store a value other than 0 or 1 for type bool. * Support Half/BFloat16 in prod operator (#7857) Partial fix for #7748.
* Coerce to true_ctype in tensor_factory (pytorch#7856) This should fix the problem where attempts to test bool are often wonky in OSS and fail UBSAN internally; it is undefined behavior to store a value other than 0 or 1 for type bool. * Support Half/BFloat16 in prod operator (pytorch#7857) Partial fix for pytorch#7748.
This should fix the problem where attempts to test
boolare often wonky in OSS and fail UBSAN internally; it is undefined behavior to store a value other than 0 or 1 for typebool.