<div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Nov 20, 2020 at 1:44 PM Mehdi AMINI via Phabricator <<a href="mailto:reviews@reviews.llvm.org">reviews@reviews.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">mehdi_amini added a comment.<br>
<br>
In D90554#2408163 <<a href="https://reviews.llvm.org/D90554#2408163" rel="noreferrer" target="_blank">https://reviews.llvm.org/D90554#2408163</a>>, @spatel wrote:<br>
<br>
> So to clarify policy, we reverted patches when (1) a test based on an experimental intrinsic was crashing and (2) that crash is easily reproducible independent of this patch as shown here:<br>
> 7ae346434 <<a href="https://reviews.llvm.org/rG7ae346434a5f51b81ebaeeb50bd5d97666ee288b" rel="noreferrer" target="_blank">https://reviews.llvm.org/rG7ae346434a5f51b81ebaeeb50bd5d97666ee288b</a>><br>
><br>
> I understand that we revert first and ask questions later, but should that be the rule for experimental code?<br>
<br>
In that case it seems like there was user-visible impact for some clang user according to @brooksmoses comment above? It isn't like an arbitrary Fuzzer was plugged to the system. IMO a revert is "low cost" and easy to re-land with a fix here, little downside to do this?<br></blockquote><div><br></div><div>Appreciated for sure. I'm not sure how "experimental" we want to consider the rounding math intrinsics at this point. It's an interesting question and probably should be raised on llvm-dev. If it were something not user visible or really experimental I probably wouldn't have reverted, but given it's a fairly common use case for developers to use the option it seemed on the "not really as experimental as it sounds" area. I think the intrinsics themselves might be experimental in the "we could change these radically" rather than "we don't really expect these to work".</div><div><br></div><div>-ericĀ </div></div></div>