Yuanfang Chen

豆瓣已经关了API? On Mon, Sep 24, 2018 at 4:45 AM BabyFoxy wrote: > 能麻烦抽点时间修改一下吗？谢谢 > > — > You are receiving this because you are subscribed to this thread. > Reply...

*** Bug llvm/llvm-project#40384 has been marked as a duplicate of this bug. ***

FYI: D117356 fixed the original test case. However, I'll leave this open since the derefinement issue persists.

Try adding proper properties for the builtin here https://github.com/llvm/llvm-project/blob/6710b21d46987257b9b97105c9128c63eec2fb57/llvm/include/llvm/IR/IntrinsicsPowerPC.td#L1656-L1657 The list of properties is here: https://github.com/llvm/llvm-project/blob/43dc319049516c586f5ece974e618baf40d737a5/llvm/include/llvm/IR/Intrinsics.td#L17-L163

Sorry. This is definitely out of my expertise. One more thing I noticed is that the instruction has side effects. https://pages.cs.wisc.edu/~tesch/doc/mtspr.html If `__builtin_ppc_mtspr` has side effects but mtudscr does not,...

That is plausible. If it is deterministic, maybe file a ticket against gtest?

@efriedma-quic Thanks for the pointer and the cherry-picks. I added a check in the LIT that could prevent this for unit tests and it would issue warnings if a gtest...

> What's our policy on DRs where we implemented the obvious fix even if the committee hasn't yet voted that in? Most of the time, "Open" DRs have no solutions....

I think the test case should be accepted. When picking the overloaded non-template function, constraints partial ordering applies by https://eel.is/c++draft/over.match.best#general-2.6

This is similar to https://github.com/llvm/llvm-project/issues/44304