<div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Fri, May 11, 2018 at 2:37 PM, Hiroshi Yamauchi <span dir="ltr"><<a href="mailto:yamauchi@google.com" target="_blank">yamauchi@google.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:arial,helvetica,sans-serif"><br></div><br><div class="gmail_quote"><div><div class="gmail-m_2423541276025245453h5"><div dir="ltr">On Thu, May 10, 2018 at 12:49 PM Daniel Berlin <<a href="mailto:dberlin@dberlin.org" target="_blank">dberlin@dberlin.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"><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Thu, May 10, 2018 at 12:05 PM, Hiroshi Yamauchi <span dir="ltr"><<a href="mailto:yamauchi@google.com" target="_blank">yamauchi@google.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:arial,helvetica,sans-serif"><br></div><br><div class="gmail_quote"><span><div dir="ltr">On Wed, May 9, 2018 at 8:24 PM Daniel Berlin <<a href="mailto:dberlin@dberlin.org" target="_blank">dberlin@dberlin.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"><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Wed, May 9, 2018 at 10:39 AM, Hiroshi Yamauchi <span dir="ltr"><<a href="mailto:yamauchi@google.com" target="_blank">yamauchi@google.com</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div style="font-family:arial,helvetica,sans-serif"><br></div><br><div class="gmail_quote"><span class="gmail-m_2423541276025245453m_-9018700274548918466gmail-m_2085822097580644453m_402989712542517733gmail-m_8307257963026143428gmail-"><div dir="ltr">On Tue, May 8, 2018 at 11:15 AM Daniel Berlin <<a href="mailto:dberlin@dberlin.org" target="_blank">dberlin@dberlin.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"><div dir="ltr"><br><div class="gmail_extra"><br><div class="gmail_quote">On Tue, May 8, 2018 at 10:38 AM, Hiroshi Yamauchi via llvm-dev <span dir="ltr"><<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">(<div style="color:rgb(34,34,34);font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;text-decoration-style:initial;text-decoration-color:initial;font-family:arial,helvetica,sans-serif;display:inline">​I came across this issue in the context of</div><span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"> </span><a href="https://reviews.llvm.org/D46336" style="color:rgb(17,85,204);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px" target="_blank">D46336</a><span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">.</span><div style="font-family:arial,helvetica,sans-serif;display:inline">​ ​</div>Thanks, Sanjay, for starting this discussion.)<br><div><br></div><div>If <div style="font-family:arial,helvetica,sans-serif;display:inline">​we will</div> move <span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><div style="font-family:arial,helvetica,sans-serif;display:inline">​reassociation, </div></span>or keep additional ones<div style="font-family:arial,helvetica,sans-serif;display:inline">​,​</div> out of instcombine, <div style="font-family:arial,helvetica,sans-serif;display:inline">​open questions for me would be</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​:</div><br><br>1. Since -reassociate isn't a fixed point pass,</div></div></blockquote><div><br></div><div>This is fixable, fwiw, without fixpointing it.</div></div></div></div></blockquote><div><br></div></span><div><div style="font-family:arial,helvetica,sans-serif">How?</div></div></div></div></blockquote><div><br></div><div>Depends on specifically which part you would like to know about ;)</div></div></div></div></blockquote><div><br></div></span><div><div style="font-family:arial,helvetica,sans-serif">​Maybe I misunderstood what you meant by "This is fixable". Did you mean that we won't somehow need to fixpoint between instcombine and reassociate, or that <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">the specific motivating examples from the above differentials are foldable without fixpointing?</span></div></div></div></div></blockquote><div><br></div><div>If by fixpointing you mean "fixpointing reassociate and instcombine", then yes, that is fixable without fixpointing reassociate and instcombine, but would require rewriting instcombine :)</div><div> <br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div style="font-family:arial,helvetica,sans-serif"><br></div><div style="font-family:arial,helvetica,sans-serif"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">If the latter, that may be the case. The concern was that we may encounter examples that may need many more iterations, if not fixpointing. As long as it's feasible to fixpoint between <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">instcombine and reassociate, it seems to work, but I guess that would <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">probably </span>need some pass management change.</span></span></div><div><div class="gmail-m_2423541276025245453m_-9018700274548918466gmail-m_2085822097580644453h5"><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><span class="gmail-m_2423541276025245453m_-9018700274548918466gmail-m_2085822097580644453m_402989712542517733gmail-m_8307257963026143428gmail-"><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div> we might need to repeat "-instcombine -reassociate" multiple times to <div style="font-family:arial,helvetica,sans-serif;display:inline">​fold</div> down to what we want (relating to <a href="https://reviews.llvm.org/D46336#1087082" target="_blank">my comment here</a>). I assumed this isn't not what we want to do<div style="font-family:arial,helvetica,sans-serif;display:inline">​? My impression is we don't do a fixed-point with passes?</div></div></div></blockquote><div><br></div><div>Well, i mean there is no practical difference between passes that we fixpoint externally and fixpoint internally.</div></div></div></div></blockquote></span><div><div style="font-family:arial,helvetica,sans-serif">​​</div><div style="font-family:arial,helvetica,sans-serif"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">I had the following in mind: Does the pass manager support fixpointing externally? Is there any performance difference? Are people okay with that in general?</span></div><div style="font-family:arial,helvetica,sans-serif"><br></div><div style="font-family:arial,helvetica,sans-serif">But if <span style="font-family:sans-serif">there is no practical difference</span>, I don't see any problem with that :)</div></div><span class="gmail-m_2423541276025245453m_-9018700274548918466gmail-m_2085822097580644453m_402989712542517733gmail-m_8307257963026143428gmail-"><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div> <br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div> <br></div></div></blockquote><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div>2. <div style="font-family:arial,helvetica,sans-serif;display:inline">​Since -<span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociate needs to come up with one operand order (at least currently as the only <span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociate </span>pass), would there exist a single, unique operand order that would enable all <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociative/commutative</span> foldings that we want? </span></span></div></div></div></blockquote><div><br></div><div>In what way?</div><div>Are you asking whether there is a single reassociation order that makes all foldings occur in the same operation or something?<br>I don't feel like i understand what you are asking.<br></div></div></div></div></blockquote><div><br></div></span><div><div><span style="font-family:arial,helvetica,sans-serif">Does this rephrase help: with the motivating examples (like and-of-shifts or bit check patterns) from the above differentials in mind, can we come up with a single <span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociation order that solves all those and all the others that may come up in the future? Would we need different <span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociation orders to fold different patterns?</span></span></span></div></div><span class="gmail-m_2423541276025245453m_-9018700274548918466gmail-m_2085822097580644453m_402989712542517733gmail-m_8307257963026143428gmail-"><div></div></span></div></div></blockquote><div><br></div><div>It doesn't quite help.</div><div>When stated that generally, there can be no such ordering at all, that's easy to prove.  It is a statically undecidable problem.</div><div><br></div><div>There is however, a different question and answer to a few related problems that maybe you are really asking?<br>1. Is there a way to determine and apply the a maximal or nearly-maximal set of folds/graph transforms that could be applied to a given set of code in a sane and principled way -> yes</div><div><br></div><div>(see, e.g., <a href="http://www.cs.cornell.edu/~ross/publications/eqsat/" target="_blank">http://www.cs.cornell.ed<wbr>u/~ross/publications/eqsat/</a>)</div><div><br></div><div>2. Is there a way to determine all expressions in the program as it exists that are equivalent or equivalent under constant time constant folding/reassociation, in a reasonable time bound -> yes</div><div><br></div><div>(not a single easy link, happy to talk about it)</div><div><br></div><div>Your original question is basically equivalent to</div><div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial">Is there a way to determine all expressions in the program as it exists that are equivalent or could be made equivalent through any type of folding that one can think up?</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial">The answer to that is "no", it's provable that this is not statically decidable, so the time bound doesn't matter :)</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial">You have to limit the possible folding/evaluation you apply in various ways to make this decidable, and then further limit it to make the time bound reasonable.</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial">This all quickly devolves into herbrand equivalence and it's variations.</div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br></div><div style="color:rgb(34,34,34);font-family:arial,sans-serif;font-size:small;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial"><br></div></div></div></div></div></blockquote><div><span style="font-family:arial,helvetica,sans-serif">​</span></div></div></div><div><div><font face="arial, helvetica, sans-serif">Let me try one more time :) </font><span style="font-family:arial,helvetica,sans-serif">May we need multiple reassociate passes to fold different </span><span style="font-family:arial,helvetica,sans-serif">reassociative patterns?</span></div><div><font face="arial, helvetica, sans-serif"><br></font></div><div><font face="arial, helvetica, sans-serif">A longer version: If Sanjay wants a particular reassociative pattern to be folded (D45842), Omer wants another particular </font><span style="font-family:arial,helvetica,sans-serif">reassociative <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">pattern</span> to be folded (</span><font face="arial, helvetica, sans-serif">D41574)</font><span style="font-family:arial,helvetica,sans-serif">, and I want yet another </span><font face="arial, helvetica, sans-serif">​</font><span style="font-family:arial,helvetica,sans-serif">particular </span><span style="font-family:arial,helvetica,sans-serif">reassociat<wbr>ive <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">pattern</span> to be folded (</span><font face="arial, helvetica, sans-serif">D46336)</font><span style="font-family:arial,helvetica,sans-serif">, would we potentially need three different reassociate passes with each combined with instcombine, rather than just one that may be able to <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">somehow </span>handle those cases in one shot, (assuming we don't want to put those in instcombine)?</span></div><div><span style="font-family:arial,helvetica,sans-serif"><br></span></div><div><span style="font-family:arial,helvetica,sans-serif">And it sounds like the answer is yes?</span></div></div></div></div></blockquote><div><br></div><div>If you take the current instcombine as a base, then yes, that is correct.</div></div></div></div></blockquote><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_extra"><div class="gmail_quote"><div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>If you are willing to rearchitect instcombine, the answer is no, it's possible to do this all in a single pass in a relatively sane way.</div></div></div></div></blockquote><div><div style="font-family:arial,helvetica,sans-serif"><br></div></div></div></div><div style="font-family:arial,helvetica,sans-serif">I assume by <span style="color:rgb(34,34,34);font-family:sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">rearchitect, you mean a major rewrite as per this comment</span><span style="font-family:sans-serif">: </span><span style="font-family:sans-serif">"</span><span style="font-family:sans-serif">Is there a way to determine all expressions in the program as it exists that are equivalent or equivalent under constant time constant folding/reassociation, in a reasonable time bound -> yes". </span><span style="font-family:sans-serif">Any pointer or time to chat?</span></div></div></div></blockquote><div><br></div><div>I'm happy to do both.</div><div> </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div style="font-family:arial,helvetica,sans-serif">​</div><div><div><div style="font-family:arial,helvetica,sans-serif;display:inline">I think that an approach like ​</div>D<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>4<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>6<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>3<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>3<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>6<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​ / </div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>D<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>4<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>6<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>5<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>9<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div>5<div style="font-family:arial,helvetica,sans-serif;display:inline">​​</div><div style="font-family:arial,helvetica,sans-serif;display:inline">​​ has a merit: it would adds a bit of complexity, but would not<span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"> require:</span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><br></span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">1. a major rewrite of instcombine,</span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">2. writing multiple (potentially many) reassociate passes and figuring out how to fixpoint them with instcombine, <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">or</span></span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">3. writing a self-contained folding pass for a specific pattern</span></span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline"><br></span></div></div><div><div style="font-family:arial,helvetica,sans-serif;display:inline"><span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">If you look at the diffs in the existing .ll files in </span></div><font face="arial, helvetica, sans-serif">D46336<div style="font-family:arial,helvetica,sans-serif;display:inline">​, it helps fold some previously-unfolded <span style="color:rgb(34,34,34);font-family:arial,helvetica,sans-serif;font-size:13px;font-style:normal;font-variant-ligatures:normal;font-variant-caps:normal;font-weight:400;letter-spacing:normal;text-align:start;text-indent:0px;text-transform:none;white-space:normal;word-spacing:0px;background-color:rgb(255,255,255);text-decoration-style:initial;text-decoration-color:initial;float:none;display:inline">reassociat<wbr>ion patterns beyond the bit check patterns that it originally targeted.</span></div></font></div></div><div><br></div></div></div></blockquote><div>Sure, and it does so by  adding another O(N) cost to evaluation in each case. Instcombine doesn't even do lazy reevaluation through tracking dependencies, so it'll do so a lot of times as well.</div><div><br></div><div>To me, that's not a good tradeoff, especially given how slow instcombine is *already*.  The code it produces is "good enough" to stop for a while and do something else and not suffer horribly in performance.[1]</div><div><br></div><div>Let me ask a different question:<br><br></div><div>At what point would anyone here be willing to stop adding things to instcombine and start doing something else instead, instead of waiting for someone else to do it?</div><div>As far as i can tell, the answer is: "never", which makes most of these discussions just pointless rehashes as we slowly repeat the same disaster that became  gcc's instruction combiner  :)</div><div><br></div><div>If the answer is "something", great, i'll set a mail filter and ignore these threads until that something happens :)</div><div><br></div><div>Personally, in my experience people will never do more here unless pushed somewhat, or the thing becomes such a complete disaster no one wants to touch it.</div><div><br></div><div>[1]  Last year i computed the "improvement in performance on applications" due to instcombine for a bunch of google apps and open source apps that had easy to use benchmarks (IE I isolated about two years of instcombine changes and made them to a current compiler piece by piece while measuring performance).</div><div>I also computed the compile time increase in single instcombine passes over the same time period.</div><div><br></div><div>On x86, but the numbers basically said we were basically gaining nearly nothing for high cost.  IE our drive for better looking output does not appear to translate into any real gains that i can find.  Either improvements to other opts hid them, or they simply didn't matter on the processors i tested on.</div><div><br></div><div>Certainly, apps/workloads/architectures may vary here, and my goal is not to claim it's all worthless.</div><div>My actual goal in all of this was to get a sense of whether my perspective on instcombine was still "reasonable", not to do a true scientific exploration :)</div><div>I didn't have time/energy/etc to run it elsewhere, and again, my goal was not to give certainty/try to give exact percentages.</div><div><br></div><div>--Dan</div><div><br></div><div><br></div><div><br></div><div><br></div><div><br></div></div></div></div>