<div dir="ltr">On Tue, Oct 2, 2018 at 3:11 PM Roman Lebedev <<a href="mailto:lebedev.ri@gmail.com">lebedev.ri@gmail.com</a>> wrote:<br><div class="gmail_quote"><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On Tue, Oct 2, 2018 at 10:06 PM Cameron McInally via llvm-dev<br>
<<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:<br>
><br>
> On Mon, Oct 1, 2018 at 5:41 PM Sanjay Patel <<a href="mailto:spatel@rotateright.com" target="_blank">spatel@rotateright.com</a>> wrote:<br>
>><br>
>> I don't see any controversy for the preliminary requirement of removing BinaryOperator::isFNeg() and friends, so start with that?<br>
>> That work may reveal other potential regressions that we can patch in advance too.<br>
><br>
><br>
> This is true and I will agree to do this work...<br>
><br>
>><br>
>> Other than that, I think there's really only a question of do we want 1 or both of fneg and fneg_constrained (and if we choose both, then I assume we'd also add fabs_constrained and copysign_constrained).<br>
><br>
><br>
> but this is the real goal. Doing the BinaryOperator::isFNeg() work is in vain if we don't have at least a conditional approval of an explicit FNEG IR instruction.<br>
><br>
> Would it be possible to obtain that conditional approval before work begins? That seems most prudent. </blockquote><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">
Will this affect (regress, pessimize) the current optimizations for<br>
non-strict cases?<br>
What about -ffast-math?<br></blockquote><div><br></div><div>The goal is for no regressions, but there may be problems to work through to get there.</div><div><br></div><div>-ffast-math allows for unsafe transformations, so should remain unchanged. </div></div></div>