<html><body><p><font size="2">Thanks for sharing the known problem.</font><br><br><font size="2">I think to solve the problem properly, we need to fully understand why that assumption about dominance is there and the implications of removing it. </font><br><br><font size="2">It would be good if you could be more specific about your idea of nullptr or SCEV_unknown (eg which function would return those values and when), but returning nullptr from getAddExpr or getSCEVAtScope may be problematic since they currently return valid pointers all the time and changing that would be error prone and would increase code complexity. Returning SCEV_Unknown from getAddExpr would seem plausible except that it would not allow for expression simplifications where recurrences over non-dominating loops can get canceled out. Having said that it may still be a reasonable middle-ground solution.</font><br><br><font size="2">Philip, do you have any thoughts on that?</font><br><br><font size="2">Bardia Mahjour<br></font><br><br><img width="16" height="16" src="cid:1__=8FBB0FDEDFC296B38f9e8a93df938690918c8FB@" border="0" alt="Inactive hide details for Jimmy Zhongduo Lin ---2020/04/16 08:39:34 PM---Hi Bardia, This is actually a long known problem: http"><font size="2" color="#424282">Jimmy Zhongduo Lin ---2020/04/16 08:39:34 PM---Hi Bardia, This is actually a long known problem: <a href="INVALID URI REMOVED">INVALID URI REMOVED</a></font><br><br><font size="2" color="#5F5F5F">From: </font><font size="2">Jimmy Zhongduo Lin <jimmy.zhongduo.lin@huawei.com></font><br><font size="2" color="#5F5F5F">To: </font><font size="2">Bardia Mahjour <bmahjour@ca.ibm.com></font><br><font size="2" color="#5F5F5F">Cc: </font><font size="2">Philip Reames <listmail@philipreames.com>, "llvm-dev@lists.llvm.org" <llvm-dev@lists.llvm.org></font><br><font size="2" color="#5F5F5F">Date: </font><font size="2">2020/04/16 08:39 PM</font><br><font size="2" color="#5F5F5F">Subject: </font><font size="2">[EXTERNAL] RE: [llvm-dev] Scalar Evolution Expressions Involving Sibling Loops</font><br><hr width="100%" size="2" align="left" noshade style="color:#8091A5; "><br><br><br><font color="#1F497D">Hi Bardia,</font><br><font color="#1F497D"> </font><br><font color="#1F497D">This is actually a long known problem: </font><a href="http://lists.llvm.org/pipermail/llvm-bugs/2017-July/056757.html"><u><font color="#0000FF" face="Times New Roman">http://lists.llvm.org/pipermail/llvm-bugs/2017-July/056757.html</font></u></a><font face="Times New Roman"> </font><br><font face="Times New Roman"> </font><br><font face="Times New Roman">As shown above, the problem gets triggered easily with scev-aa since it will compare two SCEVs from anywhere in the code, including your case of course. I believe the real problem is how to solve it properly. In the long run, checking satisfiesTotalOrder is too costly as it is duplicating part of the work in getAddExpr, but I agree it is way better than assertion error. Maybe SCEV_Unknown or nullptr can be used too. </font><br><font color="#1F497D"> </font><br><font color="#1F497D">Thanks,</font><br><font color="#1F497D">Jimmy</font><br><font color="#1F497D"> </font><br><b>From:</b> Bardia Mahjour [<a href="mailto:bmahjour@ca.ibm.com">mailto:bmahjour@ca.ibm.com</a>] <b><br>Sent:</b> Thursday, April 16, 2020 4:51 PM<b><br>To:</b> Jimmy Zhongduo Lin <jimmy.zhongduo.lin@huawei.com><b><br>Cc:</b> Philip Reames <listmail@philipreames.com>; llvm-dev@lists.llvm.org<b><br>Subject:</b> RE: [llvm-dev] Scalar Evolution Expressions Involving Sibling Loops<br><font face="Times New Roman"> </font><p><font size="2" face="Times New Roman">Hi Jimmy,</font><font face="Times New Roman"><br></font><font size="2" face="Times New Roman"><br>It's good to know that the problem is not specific to the case I ran into. May be you can provide your example as well, since Philip seems to be interested in some specific examples. If the assertion in getAddrExpr is deemed necessary, then I think a condition check would be the next best solution as it helps client code guard against such cases and make alternative arrangements to avoid an assertion or miscompile. </font><font face="Times New Roman"><br></font><font size="2" face="Times New Roman"><br>Bardia Mahjour<br>Compiler Optimizations<br>IBM Toronto Software Lab<br></font><font face="Times New Roman"><br><br><br></font><img src="cid:1__=8FBB0FDEDFC296B38f9e8a93df938690918c8FB@" width="16" height="16" alt="Inactive hide details for Jimmy Zhongduo Lin ---2020/04/16 04:34:24 PM---Hi Bardia, I am encountering a similar problem and tot"><font size="2" color="#424282" face="Times New Roman">Jimmy Zhongduo Lin ---2020/04/16 04:34:24 PM---Hi Bardia, I am encountering a similar problem and totally agree that getAddExpr shouldn't generate</font><font face="Times New Roman"><br></font><font size="2" color="#5F5F5F" face="Times New Roman"><br>From: </font><font size="2" face="Times New Roman">Jimmy Zhongduo Lin <</font><a href="mailto:jimmy.zhongduo.lin@huawei.com"><u><font size="2" color="#0000FF" face="Times New Roman">jimmy.zhongduo.lin@huawei.com</font></u></a><font size="2" face="Times New Roman">></font><font size="2" color="#5F5F5F" face="Times New Roman"><br>To: </font><font size="2" face="Times New Roman">Bardia Mahjour <</font><a href="mailto:bmahjour@ca.ibm.com"><u><font size="2" color="#0000FF" face="Times New Roman">bmahjour@ca.ibm.com</font></u></a><font size="2" face="Times New Roman">>, Philip Reames <</font><a href="mailto:listmail@philipreames.com"><u><font size="2" color="#0000FF" face="Times New Roman">listmail@philipreames.com</font></u></a><font size="2" face="Times New Roman">>, "</font><a href="mailto:llvm-dev@lists.llvm.org"><u><font size="2" color="#0000FF" face="Times New Roman">llvm-dev@lists.llvm.org</font></u></a><font size="2" face="Times New Roman">" <</font><a href="mailto:llvm-dev@lists.llvm.org"><u><font size="2" color="#0000FF" face="Times New Roman">llvm-dev@lists.llvm.org</font></u></a><font size="2" face="Times New Roman">></font><font size="2" color="#5F5F5F" face="Times New Roman"><br>Date: </font><font size="2" face="Times New Roman">2020/04/16 04:34 PM</font><font size="2" color="#5F5F5F" face="Times New Roman"><br>Subject: </font><font size="2" face="Times New Roman">[EXTERNAL] RE: [llvm-dev] Scalar Evolution Expressions Involving Sibling Loops</font><br><hr width="100%" size="2" align="left" noshade><br><font face="Times New Roman"><br><br></font><font color="#1F497D" face="Times New Roman"><br>Hi Bardia,</font><font face="Times New Roman"><br></font><font color="#1F497D" face="Times New Roman"><br>I am encountering a similar problem and totally agree that getAddExpr shouldn’t generate any assertion error or at least provide condition check. Even if this is something to avoid, would it be better to return nullptr instead of assertion error?</font><font face="Times New Roman"><br></font><font color="#1F497D" face="Times New Roman"><br>Thanks,<br>Jimmy</font><font face="Times New Roman"><br></font><b><font face="Times New Roman"><br>From:</font></b><font face="Times New Roman"> llvm-dev [</font><a href="mailto:llvm-dev-bounces@lists.llvm.org"><u><font color="#0000FF" face="Times New Roman">mailto:llvm-dev-bounces@lists.llvm.org</font></u></a><font face="Times New Roman">] </font><b><font face="Times New Roman">On Behalf Of </font></b><font face="Times New Roman">Bardia Mahjour via llvm-dev</font><b><font face="Times New Roman"><br>Sent:</font></b><font face="Times New Roman"> Monday, March 30, 2020 4:59 PM</font><b><font face="Times New Roman"><br>To:</font></b><font face="Times New Roman"> Philip Reames <</font><a href="mailto:listmail@philipreames.com"><u><font color="#0000FF" face="Times New Roman">listmail@philipreames.com</font></u></a><font face="Times New Roman">></font><b><font face="Times New Roman"><br>Cc:</font></b><font face="Times New Roman"> LLVM Development List <</font><a href="mailto:llvm-dev@lists.llvm.org"><u><font color="#0000FF" face="Times New Roman">llvm-dev@lists.llvm.org</font></u></a><font face="Times New Roman">></font><b><font face="Times New Roman"><br>Subject:</font></b><font face="Times New Roman"> Re: [llvm-dev] Scalar Evolution Expressions Involving Sibling Loops</font><p><font size="2" face="Courier New">> </font><font size="2" face="Times New Roman">I'm not following your example. If you have two sibling loops with the same parent, one will frequently, but not always dominate the other. Can you give a specific example of when forming a recurrence between two siblings (without one dominating the other), is useful? <br><br>The situation can happen with guarded loops or with a user guard like below:<br><br>if (c) {<br>for (i = 0; i < n; i++) <br>...<br>}<br>for (j = 0; j < n; j++)<br>...</font><font face="Times New Roman"><br></font><font size="2" face="Times New Roman"><br><br>The specific example that we ran into is described in </font><a href="https://reviews.llvm.org/D75628"><u><font size="2" color="#0000FF" face="Times New Roman">https://reviews.llvm.org/D75628</font></u></a><font size="2" face="Times New Roman">. Basically we have two triangular loops that are siblings and we'd like to run Banerjee MIV tests on the memory accesses in those loops. The loop looks like:<br><br>void foo(int *restrict A, int n1, int n2, int n3) {<br>for (int i1 = 0; i1 < n1; i1++) {<br>for (int i2 = 2; i2 < n2; i2++) {<br>for (int i3 = i2 + 1; i3 < n3; i3++) {<br>A[i2 + i3*n2] = 11;<br>}<br>}<br>for (int i4 = 2; i4 < n3; i4++) {<br>for (int i5 = 1; i5 < i4 - 1; i5++) {<br>A[i5] = 22;<br>}<br>}<br>}<br>}<br><br>To check the bounds of the dependence function we need to create a symbolic expression that involves AddRecs for i2 and i4.<br><br>Bardia Mahjour</font><font face="Times New Roman"><br><br><br><br></font><img src="cid:1__=8FBB0FDEDFC296B38f9e8a93df938690918c8FB@" width="16" height="16" alt="Inactive hide details for Philip Reames ---2020/03/30 02:50:45 PM---On 3/30/20 11:27 AM, Bardia Mahjour via llvm-dev wrote: >"><font size="2" color="#424282" face="Times New Roman">Philip Reames ---2020/03/30 02:50:45 PM---On 3/30/20 11:27 AM, Bardia Mahjour via llvm-dev wrote: ></font><font size="2" color="#5F5F5F" face="Times New Roman"><br><br>From: </font><font size="2" face="Times New Roman">Philip Reames <</font><a href="mailto:listmail@philipreames.com"><u><font size="2" color="#0000FF" face="Times New Roman">listmail@philipreames.com</font></u></a><font size="2" face="Times New Roman">></font><font size="2" color="#5F5F5F" face="Times New Roman"><br>To: </font><font size="2" face="Times New Roman">Bardia Mahjour <</font><a href="mailto:bmahjour@ca.ibm.com"><u><font size="2" color="#0000FF" face="Times New Roman">bmahjour@ca.ibm.com</font></u></a><font size="2" face="Times New Roman">>, LLVM Development List <</font><a href="mailto:llvm-dev@lists.llvm.org"><u><font size="2" color="#0000FF" face="Times New Roman">llvm-dev@lists.llvm.org</font></u></a><font size="2" face="Times New Roman">></font><font size="2" color="#5F5F5F" face="Times New Roman"><br>Date: </font><font size="2" face="Times New Roman">2020/03/30 02:50 PM</font><font size="2" color="#5F5F5F" face="Times New Roman"><br>Subject: </font><font size="2" face="Times New Roman">[EXTERNAL] Re: [llvm-dev] Scalar Evolution Expressions Involving Sibling Loops</font><br><hr width="100%" size="2" align="left" noshade><br><font face="Times New Roman"> </font><p><font face="Times New Roman">On 3/30/20 11:27 AM, Bardia Mahjour via llvm-dev wrote: </font><ul><ul><ul><ul><ul><ul><ul><ul><font size="2" face="Times New Roman">Forwarding to the dev list, in case others ran into similar issues and/or have input on this topic.<br><br>Bardia Mahjour</font><font size="2" color="#800080" face="Times New Roman"><br><br>----- Forwarded by Bardia Mahjour/Toronto/IBM on 2020/03/30 02:25 PM -----</font><font size="2" color="#5F5F5F" face="Times New Roman"><br><br>From: </font><font size="2" face="Times New Roman">Bardia Mahjour/Toronto/IBM</font><font size="2" color="#5F5F5F" face="Times New Roman"><br>To: </font><a href="mailto:listmail@philipreames.com"><u><font size="2" color="#0000FF" face="Times New Roman">listmail@philipreames.com</font></u></a><font size="2" color="#5F5F5F" face="Times New Roman"><br>Cc: </font><font size="2" face="Times New Roman">"Michael Kruse" </font><a href="mailto:llvm@meinersbur.de"><u><font size="2" color="#0000FF" face="Times New Roman"><llvm@meinersbur.de></font></u></a><font size="2" color="#5F5F5F" face="Times New Roman"><br>Date: </font><font size="2" face="Times New Roman">2020/03/26 11:47 AM</font><font size="2" color="#5F5F5F" face="Times New Roman"><br>Subject: </font><font size="2" face="Times New Roman">Scalar Evolution Expressions Involving Sibling Loops</font><br><hr width="100%" size="2" align="left" noshade><br><font size="2" face="Times New Roman"><br><br><br>Hi Philip,<br><br>I hope you are doing well. <br><br>We've recently run into an issue with SCEV in the context of dependence analysis, and would like your opinion on it. Background discussion can be found here </font><a href="https://reviews.llvm.org/D75628#inline-689527"><u><font size="2" color="#0000FF" face="Times New Roman">https://reviews.llvm.org/D75628#inline-689527</font></u></a><font size="2" face="Times New Roman">.<br><br>Basically in this case, the dependence equation requires us to symbolically create an expression involving two or more recurrences that recur with non-dominating loops (sibling loops). </font></ul></ul></ul></ul></ul></ul></ul></ul><font size="2" face="Times New Roman">I'm not following your example. If you have two sibling loops with the same parent, one will frequently, but not always dominate the other. Can you give a specific example of when forming a recurrence between two siblings (without one dominating the other), is useful? </font><ul><ul><ul><ul><ul><ul><ul><ul><font size="2" face="Times New Roman">Currently creating such a SCEV expression trips asserts in `</font><font size="2" face="Courier New">CompareSCEVComplexity</font><font size="2" face="Times New Roman">()` and `</font><font size="2" face="Courier New">isKnownViaInduction</font><font size="2" face="Times New Roman">()` saying that a given SCEV expression cannot be composed of recurrences that have no dominance relationship between them. <br><br>Michael tried explaining to me why there is this restriction about dominance, and I'm beginning to understand why such restriction may be necessary when evaluating or expanding SCEV expressions in outer scopes (eg. `</font><font size="2" face="Courier New">getSCEVAtScope</font><font size="2" face="Times New Roman">(nullptr)`) but I still don't understand why this restriction is imposed at construction. Shouldn't this restriction be asserted on when calling getSCEVAtScope instead of when creating AddRecs, given that simplification steps may remove identical terms involving non-dominating loops? </font></ul></ul></ul></ul></ul></ul></ul></ul><font size="2" face="Times New Roman">Well, SCEV construction is generally done to parallel IR. SSA requires dominance, so having the SCEV operands require dominance would seem like a reasonable thing. If you want to change this, you'll need to motivate the change. (i.e. see above question)</font><ul><ul><ul><ul><ul><ul><ul><ul><font size="2" face="Times New Roman"><br>We would appreciate any other insight you might have about this issue.<br><br>Regards,<br><br>Bardia Mahjour<br>Compiler Optimizations<br>IBM Toronto Software Lab</font><u><font color="#0000FF" face="Times New Roman"><br></font></u><a href="mailto:bmahjour@ca.ibm.com"><u><font size="2" color="#0000FF" face="Times New Roman">bmahjour@ca.ibm.com</font></u></a><font face="Times New Roman"><br></font><font size="2" face="Courier New"><br><br>_______________________________________________<br>LLVM Developers mailing list</font><u><font color="#0000FF" face="Times New Roman"><br></font></u><a href="mailto:llvm-dev@lists.llvm.org"><u><font size="2" color="#0000FF" face="Courier New">llvm-dev@lists.llvm.org</font></u></a><u><font color="#0000FF" face="Times New Roman"><br></font></u><a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev"><u><font size="2" color="#0000FF" face="Courier New">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</font></u></a></ul></ul></ul></ul></ul></ul></ul></ul><font face="Times New Roman"><br></font><br><br><BR>
</body></html>