<html><head><meta http-equiv="Content-Type" content="text/html; charset=us-ascii"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;" class="">Sorry about that, I reverted the patch in r329650 until I figure out the issue.<div class=""><br class=""></div><div class="">Michael<br class=""><div><br class=""><blockquote type="cite" class=""><div class="">On Apr 9, 2018, at 5:22 PM, Aleksey Shlyapnikov <<a href="mailto:alekseys@google.com" class="">alekseys@google.com</a>> wrote:</div><br class="Apple-interchange-newline"><div class=""><div dir="ltr" class="">Our Android and Windows bots are unhappy with these changes: <a href="http://lab.llvm.org:8011/builders/sanitizer-windows/builds/27045" class="">http://lab.llvm.org:8011/builders/sanitizer-windows/builds/27045</a> and <a href="http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-android/builds/9491" class="">http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-android/builds/9491</a><br class=""></div><div class="gmail_extra"><br class=""><div class="gmail_quote">On Mon, Apr 9, 2018 at 4:37 PM, Michael Zolotukhin via llvm-commits <span dir="ltr" class=""><<a href="mailto:llvm-commits@lists.llvm.org" target="_blank" class="">llvm-commits@lists.llvm.org</a>></span> wrote:<br class=""><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Author: mzolotukhin<br class="">
Date: Mon Apr 9 16:37:37 2018<br class="">
New Revision: 329644<br class="">
<br class="">
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=329644&view=rev" rel="noreferrer" target="_blank" class="">http://llvm.org/viewvc/llvm-<wbr class="">project?rev=329644&view=rev</a><br class="">
Log:<br class="">
[PR16756] Use SSAUpdaterBulk in JumpThreading.<br class="">
<br class="">
Summary:<br class="">
SSAUpdater is a bottleneck in JumpThreading, and this patch improves the<br class="">
situation by using SSAUpdaterBulk instead.<br class="">
<br class="">
Compile time impact: no noticable changes on CTMark, a big improvement<br class="">
on the test from PR16756.<br class="">
<br class="">
Reviewers: dberlin, davide, MatzeB<br class="">
<br class="">
Subscribers: llvm-commits, hiraditya<br class="">
<br class="">
Differential Revision: <a href="https://reviews.llvm.org/D44282" rel="noreferrer" target="_blank" class="">https://reviews.llvm.org/<wbr class="">D44282</a><br class="">
<br class="">
Modified:<br class="">
llvm/trunk/lib/Transforms/<wbr class="">Scalar/JumpThreading.cpp<br class="">
<br class="">
Modified: llvm/trunk/lib/Transforms/<wbr class="">Scalar/JumpThreading.cpp<br class="">
URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/JumpThreading.cpp?rev=329644&r1=329643&r2=329644&view=diff" rel="noreferrer" target="_blank" class="">http://llvm.org/viewvc/llvm-<wbr class="">project/llvm/trunk/lib/<wbr class="">Transforms/Scalar/<wbr class="">JumpThreading.cpp?rev=329644&<wbr class="">r1=329643&r2=329644&view=diff</a><br class="">
==============================<wbr class="">==============================<wbr class="">==================<br class="">
--- llvm/trunk/lib/Transforms/<wbr class="">Scalar/JumpThreading.cpp (original)<br class="">
+++ llvm/trunk/lib/Transforms/<wbr class="">Scalar/JumpThreading.cpp Mon Apr 9 16:37:37 2018<br class="">
@@ -66,6 +66,7 @@<br class="">
#include "llvm/Transforms/Utils/<wbr class="">BasicBlockUtils.h"<br class="">
#include "llvm/Transforms/Utils/<wbr class="">Cloning.h"<br class="">
#include "llvm/Transforms/Utils/<wbr class="">SSAUpdater.h"<br class="">
+#include "llvm/Transforms/Utils/<wbr class="">SSAUpdaterBulk.h"<br class="">
#include "llvm/Transforms/Utils/<wbr class="">ValueMapper.h"<br class="">
#include <algorithm><br class="">
#include <cassert><br class="">
@@ -1989,9 +1990,13 @@ bool JumpThreadingPass::ThreadEdge(<wbr class="">Basic<br class="">
// now have to update all uses of the value to use either the original value,<br class="">
// the cloned value, or some PHI derived value. This can require arbitrary<br class="">
// PHI insertion, of which we are prepared to do, clean these up now.<br class="">
- SSAUpdater SSAUpdate;<br class="">
+ SSAUpdaterBulk SSAUpdate;<br class="">
SmallVector<Use*, 16> UsesToRename;<br class="">
+<br class="">
+ unsigned VarNum = 0;<br class="">
for (Instruction &I : *BB) {<br class="">
+ UsesToRename.clear();<br class="">
+<br class="">
// Scan all uses of this instruction to see if it is used outside of its<br class="">
// block, and if so, record them in UsesToRename.<br class="">
for (Use &U : I.uses()) {<br class="">
@@ -2008,19 +2013,15 @@ bool JumpThreadingPass::ThreadEdge(<wbr class="">Basic<br class="">
// If there are no uses outside the block, we're done with this instruction.<br class="">
if (UsesToRename.empty())<br class="">
continue;<br class="">
+ SSAUpdate.AddVariable(VarNum, I.getName(), I.getType());<br class="">
<br class="">
- DEBUG(dbgs() << "JT: Renaming non-local uses of: " << I << "\n");<br class="">
-<br class="">
- // We found a use of I outside of BB. Rename all uses of I that are outside<br class="">
- // its block to be uses of the appropriate PHI node etc. See ValuesInBlocks<br class="">
- // with the two values we know.<br class="">
- SSAUpdate.Initialize(I.<wbr class="">getType(), I.getName());<br class="">
- SSAUpdate.AddAvailableValue(<wbr class="">BB, &I);<br class="">
- SSAUpdate.AddAvailableValue(<wbr class="">NewBB, ValueMapping[&I]);<br class="">
-<br class="">
- while (!UsesToRename.empty())<br class="">
- SSAUpdate.RewriteUse(*<wbr class="">UsesToRename.pop_back_val());<br class="">
- DEBUG(dbgs() << "\n");<br class="">
+ // We found a use of I outside of BB - we need to rename all uses of I that<br class="">
+ // are outside its block to be uses of the appropriate PHI node etc.<br class="">
+ SSAUpdate.AddAvailableValue(<wbr class="">VarNum, BB, &I);<br class="">
+ SSAUpdate.AddAvailableValue(<wbr class="">VarNum, NewBB, ValueMapping[&I]);<br class="">
+ for (auto U : UsesToRename)<br class="">
+ SSAUpdate.AddUse(VarNum, U);<br class="">
+ VarNum++;<br class="">
}<br class="">
<br class="">
// Ok, NewBB is good to go. Update the terminator of PredBB to jump to<br class="">
@@ -2037,6 +2038,10 @@ bool JumpThreadingPass::ThreadEdge(<wbr class="">Basic<br class="">
{DominatorTree::Insert, PredBB, NewBB},<br class="">
{DominatorTree::Delete, PredBB, BB}});<br class="">
<br class="">
+ // Apply all updates we queued with DDT and get the updated Dominator Tree.<br class="">
+ DominatorTree *DT = &DDT->flush();<br class="">
+ SSAUpdate.RewriteAllUses(DT);<br class="">
+<br class="">
// At this point, the IR is fully up to date and consistent. Do a quick scan<br class="">
// over the new instructions and zap any that are constants or dead. This<br class="">
// frequently happens because of phi translation.<br class="">
<br class="">
<br class="">
______________________________<wbr class="">_________________<br class="">
llvm-commits mailing list<br class="">
<a href="mailto:llvm-commits@lists.llvm.org" class="">llvm-commits@lists.llvm.org</a><br class="">
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank" class="">http://lists.llvm.org/cgi-bin/<wbr class="">mailman/listinfo/llvm-commits</a><br class="">
</blockquote></div><br class=""></div>
</div></blockquote></div><br class=""></div></body></html>