Disable msvc vectorization in GC get_promoted_bytes #100309
Conversation
|
Tagging subscribers to this area: @dotnet/gc |
| @@ -24211,6 +24211,9 @@ size_t gc_heap::get_promoted_bytes() | |||
|
|
|||
| dprintf (3, ("h%d getting surv", heap_number)); | |||
| size_t promoted = 0; | |||
| #ifdef _MSC_VER | |||
| #pragma loop(no_vector) | |||
There was a problem hiding this comment.
Add a link to the MSVC bug that this is working around?
There was a problem hiding this comment.
Yes, I think I will add it to the description and merge as soon as the tests finish so that this stops impacting everyone, and then I can add a comment when the timing doesn't matter (and hopefully we'll know more -- latest toolchains don't repro this anymore, but we'd like to know if it was directly fixed. Andy filed https://developercommunity.visualstudio.com/t/Bad-codegen-in-coreclr-GC-code/10625094)
There was a problem hiding this comment.
https://devdiv.visualstudio.com/DevDiv/_git/msvc/commit/e72c28af106a0efbc93eb67b7b21c433e8749424
I haven't checked the timing yet. We might be able to remove the workaround already.
There was a problem hiding this comment.
We've stopped using preview compilers. The codegen bug was fixed. This workaround was removed in #101995.
|
|
…dotnet#100309)" This partially reverts commit 765ca4e. (reverts the workaround but keeps the test)
This fixes the code generation rather than trying to address the symptoms of tests failing. Therefore, test Callback_Svr can be enabled again.
See https://developercommunity.visualstudio.com/t/Bad-codegen-in-coreclr-GC-code/10625094. Issue may have been fixed already as later toolsets don't seem to show the issue, but this hasn't been verified.