<div class="socmaildefaultfont" dir="ltr" style="font-family:Arial, Helvetica, sans-serif;font-size:10pt" ><div dir="ltr" > </div>
<div dir="ltr" >Hi Philip,<br><br>I have run a -print-after-all and I'm going to send you the IR before and after the Loop Vectorize pass. Hopefully this will help.<br>Let me know if you need more information or if you want me to run something on my end.<br> </div>
<div dir="ltr" >Best,</div>
<div dir="ltr" >Stefan</div>
<div dir="ltr" > </div>
<blockquote data-history-content-modified="1" dir="ltr" style="border-left:solid #aaaaaa 2px; margin-left:5px; padding-left:5px; direction:ltr; margin-right:0px" >----- Original message -----<br>From: Philip Reames <listmail@philipreames.com><br>To: Nemanja Ivanovic <nemanjai@ca.ibm.com><br>Cc: akuegel@google.com, benny.kra@gmail.com, llvm-commits@lists.llvm.org, llvmlistbot@llvm.org, LLVM on Power <powerllvm@ca.ibm.com>, Stefan Pintilie <stefanp@ca.ibm.com><br>Subject: [EXTERNAL] Re: [llvm] 7fe41ac - Revert "[LV] Unconditionally branch from middle to scalar preheader if the scalar loop must execute"<br>Date: Tue, May 25, 2021 5:38 PM<br> <br><!-- BaNnErBlUrFlE-HeAdEr-start --> <!-- BaNnErBlUrFlE-HeAdEr-end --> <!--Notes ACF
<meta http-equiv="Content-Type" content="text/html; charset=utf8" >--> <!-- BaNnErBlUrFlE-BoDy-start --> <!-- Preheader Text : BEGIN --> <span style="display:none !important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" > If you have the before and after IR, that's really all I probably need. If you share those, I can take the investigation from there. Philip On 5/25/21 2:12 PM, Nemanja Ivanovic wrote: </span> <!-- Preheader Text : END --> <!-- Email Banner : BEGIN --> <span style="display:none !important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" >ZjQcmQRYFpfptBannerStart</span> <!--[if ((ie)|(mso))]>
<table border="0" cellspacing="0" cellpadding="0" width="100%" style="padding: 3px 0px 16px 0px;" ><tr><td>
<table class="pfptBannerTableMSO" border="0" cellspacing="0" cellpadding="0" style="padding: 0px 10px 5px 6px; width: 100%;border-radius:4px;border-top:4px solid #90a4ae;background-color:#d0d8dc;" ><tr><td valign="top" >
<table align="left" border="0" cellspacing="0" cellpadding="0" style="padding: 4px 8px 4px 8px" >
<tr><td><span class="pfptTitleMSO" style="color:#000000 !important;font-family: 'Arial', sans-serif;font-weight:bold !important;font-size:14px !important;" >
This Message Is From an External Sender
</span></td></tr>
<tr><td><span class="pfptSubtitleMSO" style="color:#000000 !important;font-weight:normal !important;font-family: 'Arial', sans-serif; font-size:12px !important;" >
This message came from outside your organization.
</span></td></tr>
</table>
</td></tr></table>
</td></tr></table>
<![endif]--> <!--[if !((ie)|(mso))]-->
<div style="margin:16px 0px 16px 0px; padding:8px 16px 8px 16px; border-radius: 4px; min-width: 200px;background-color: #d0d8dc; border-top: 4px solid #90a4ae;" ><div style="display: inline-block; margin: 0px 0px 1px 0px; max-width: 600px;" ><div style="color:#000000 !important;font-family: 'Arial', sans-serif !important;font-weight:bold !important;font-size:14px !important;line-height:18px;display:block;" >This Message Is From an External Sender</div>
<div style="color:#000000 !important;font-weight:normal !important;font-family: 'Arial', sans-serif !important;font-size:12px !important;line-height:18px;margin-top:2px;display:block" >This message came from outside your organization.</div></div></div><!--[endif]--> <span style="display:none !important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" >ZjQcmQRYFpfptBannerEnd</span><br><!-- Email Banner : END --> <!-- BaNnErBlUrFlE-BoDy-end -->
<p>If you have the before and after IR, that's really all I probably need. If you share those, I can take the investigation from there.</p>
<p>Philip</p>
<div>On 5/25/21 2:12 PM, Nemanja Ivanovic wrote:</div>
<blockquote cite="mid:OFC99B9E23.4F6BE4A3-ON002586E0.00745021-002586E0.00748300@notes.na.collabserv.com" type="cite" ><!--Notes ACF
<meta http-equiv="content-type" content="text/html; charset=utf8" >-->
<div dir="ltr" style="font-family:Arial, Helvetica, sans-serif;font-size:10pt" ><div dir="ltr" >Hi Philip,</div>
<div dir="ltr" >I am sorry about the late reply - this was a 4 day weekend here in Toronto. We kind of started with looking at the difference in the post-LV IR with and without your patch for the SingleSource test case. Stefan is currently looking at it and we're hoping to figure out what's going on soon. Would it help you if we shared the two IR files with you to get your opinion on what might be happening as well?</div>
<div dir="ltr" > </div>
<div dir="ltr" >Nemanja Ivanovic<br>LLVM PPC Backend Development<br>IBM Toronto Lab<br>Email: <a href="mailto:nemanjai@ca.ibm.com" target="_blank">nemanjai@ca.ibm.com</a><br>Phone: 905-413-3388</div>
<div dir="ltr" > </div>
<div dir="ltr" > </div>
<blockquote data-history-content-modified="1" dir="ltr" style="border-left:solid #aaaaaa 2px; margin-left:5px; padding-left:5px; direction:ltr; margin-right:0px" >----- Original message -----<br>From: Philip Reames <a href="mailto:listmail@philipreames.com" target="_blank"><listmail@philipreames.com></a><br>To: Nemanja Ivanovic <a href="mailto:nemanjai@ca.ibm.com" target="_blank"><nemanjai@ca.ibm.com></a><br>Cc: <a href="mailto:akuegel@google.com" target="_blank">akuegel@google.com</a>, <a href="mailto:benny.kra@gmail.com" target="_blank">benny.kra@gmail.com</a>, <a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>, <a href="mailto:llvmlistbot@llvm.org" target="_blank">llvmlistbot@llvm.org</a>, LLVM on Power <a href="mailto:powerllvm@ca.ibm.com" target="_blank"><powerllvm@ca.ibm.com></a><br>Subject: [EXTERNAL] Re: [llvm] 7fe41ac - Revert "[LV] Unconditionally branch from middle to scalar preheader if the scalar loop must execute"<br>Date: Wed, May 19, 2021 2:15 PM<br> <br><!--Notes ACF
<meta http-equiv="Content-Type" content="text/html; charset=utf8" >-->
<p>Looking at the various failures, I see both LE and BE bots failing. The BE bot shows a miscompile in one of the test suite benchmarks. The LE bots appear to be showing a crash while using a stage1 build clang to build stage2 clang. </p>
<p>LE example: <a href="https://lab.llvm.org/buildbot#builders/19/builds/4236" moz-do-not-send="true" target="_blank">https://lab.llvm.org/buildbot#builders/19/builds/4236</a><br>BE example: <a href="https://lab.llvm.org/buildbot#builders/100/builds/5762" moz-do-not-send="true" target="_blank">https://lab.llvm.org/buildbot#builders/100/builds/5762</a></p>
<p>So both are showing miscompiles, just with different symptoms. Frankly, the BE looks easier to debug (much code making it into the miscompiled binary.)</p>
<p>Oddly, pretty much only PPC bots are failing. The one exception is a stage2 failure on AArch64 (<a href="https://lab.llvm.org/buildbot/#/builders/111/builds/2027" moz-do-not-send="true" target="_blank">https://lab.llvm.org/buildbot/#/builders/111/builds/2027</a>) which looks similar to the LE failure above. </p>
<p>Given this appears to be target specific, I am *guessing* there's some vectorizer hook which is causing a different codepath to be executed. Before we start trying to get me access to hardware, do you have any guesses on what that hook might be? If you can give me a good hint on where to look, I suspect I can probably find the issue that way. </p>
<p>Philip</p>
<div>On 5/18/21 3:07 AM, Nemanja Ivanovic wrote:</div>
<blockquote cite="mid:OF443062AC.F5998286-ON002586D9.00373B50-002586D9.0037A584@notes.na.collabserv.com" type="cite" ><!--Notes ACF
<meta http-equiv="content-type" content="text/html; charset=utf8" >-->
<div dir="ltr" style="font-family:Arial, Helvetica, sans-serif;font-size:10pt" ><div dir="ltr" >Hi Philip,</div>
<div dir="ltr" >I am not sure what happened with your first attempt to contact us and how we missed it. We would be more than happy to help you debug this issue. Do you know if this only affects the big endian bot or if it also fails on little endian bots? In the latter case, we can certainly provide access to a little endian machine hosted at OSU/OSL. In the former case, we don't have a machine available and we'll have to do the debugging and report to you (which might take a bit longer).</div>
<div dir="ltr" > </div>
<div dir="ltr" >Nemanja Ivanovic<br>LLVM PPC Backend Development<br>IBM Toronto Lab<br>Email: <a href="mailto:nemanjai@ca.ibm.com" moz-do-not-send="true" target="_blank">nemanjai@ca.ibm.com</a><br>Phone: 905-413-3388</div>
<div dir="ltr" > </div>
<div dir="ltr" > </div>
<blockquote data-history-content-modified="1" data-history-expanded="1" dir="ltr" style="border-left:solid #aaaaaa 2px; margin-left:5px; padding-left:5px; direction:ltr; margin-right:0px" >----- Original message -----<br>From: Philip Reames <a href="mailto:listmail@philipreames.com" moz-do-not-send="true" target="_blank"><listmail@philipreames.com></a><br>To: Adrian Kuegel <a href="mailto:akuegel@google.com" moz-do-not-send="true" target="_blank"><akuegel@google.com></a><br>Cc: Benjamin Kramer <a href="mailto:benny.kra@gmail.com" moz-do-not-send="true" target="_blank"><benny.kra@gmail.com></a>, Adrian Kuegel <a href="mailto:llvmlistbot@llvm.org" moz-do-not-send="true" target="_blank"><llvmlistbot@llvm.org></a>, llvm-commits <a href="mailto:llvm-commits@lists.llvm.org" moz-do-not-send="true" target="_blank"><llvm-commits@lists.llvm.org></a>, <a href="mailto:powerllvm@ca.ibm.com" moz-do-not-send="true" target="_blank">powerllvm@ca.ibm.com</a><br>Subject: [EXTERNAL] Re: [llvm] 7fe41ac - Revert "[LV] Unconditionally branch from middle to scalar preheader if the scalar loop must execute"<br>Date: Mon, May 17, 2021 11:59 PM<br> <br><!-- BaNnErBlUrFlE-HeAdEr-start --> <!-- BaNnErBlUrFlE-HeAdEr-end --> <!--Notes ACF
<meta http-equiv="Content-Type" content="text/html; charset=utf8" >--> <!-- BaNnErBlUrFlE-BoDy-start --> <!-- Preheader Text : BEGIN --> <span style="display:none!important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" > I tried another cycle to see if this had been resolved, but am still seeing build bot failures, nearly exclusively on PPC. (And one arm self host bot.) @PPC Bot Owner - I need help reducing a test case for the failure seen here: </span> <!-- Preheader Text : END --> <!-- Email Banner : BEGIN --> <span style="display:none!important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" >ZjQcmQRYFpfptBannerStart</span> <!--[if ((ie)|(mso))]>
<table border="0" cellspacing="0" cellpadding="0" width="100%" style="padding: 3px 0px 16px 0px;" ><tr><td>
<table class="pfptBannerTableMSO" border="0" cellspacing="0" cellpadding="0" style="padding: 0px 10px 5px 6px; width: 100%;border-radius:4px;border-top:4px solid #90a4ae;background-color:#d0d8dc;" ><tr><td valign="top" >
<table align="left" border="0" cellspacing="0" cellpadding="0" style="padding: 4px 8px 4px 8px" >
<tr><td><span class="pfptTitleMSO" style="color:#000000 !important;font-family: 'Arial', sans-serif;font-weight:bold !important;font-size:14px !important;" >
This Message Is From an External Sender
</span></td></tr>
<tr><td><span class="pfptSubtitleMSO" style="color:#000000 !important;font-weight:normal !important;font-family: 'Arial', sans-serif; font-size:12px !important;" >
This message came from outside your organization.
</span></td></tr>
</table>
</td></tr></table>
</td></tr></table>
<![endif]--> <!--[if !((ie)|(mso))]-->
<div style="margin:16px 0px 16px 0px; padding:8px 16px 8px 16px; border-radius: 4px; min-width: 200px;background-color: #d0d8dc; border-top: 4px solid #90a4ae;" ><div style="display: inline-block; margin: 0px 0px 1px 0px; max-width: 600px;" ><div style="color:#000000 !important;font-family: 'Arial', sans-serif !important;font-weight:bold !important;font-size:14px !important;line-height:18px;display:block;" >This Message Is From an External Sender</div>
<div style="color:#000000 !important;font-weight:normal !important;font-family: 'Arial', sans-serif !important;font-size:12px !important;line-height:18px;margin-top:2px;display:block" >This message came from outside your organization.</div></div></div><!--[endif]--> <span style="display:none!important;visibility:hidden;mso-hide:all;font-size:1px;color:#ffffff;line-height:1px;max-height:0px;max-width:0px;opacity:0;overflow:hidden;" >ZjQcmQRYFpfptBannerEnd</span><br><!-- Email Banner : END --> <!-- BaNnErBlUrFlE-BoDy-end -->
<p>I tried another cycle to see if this had been resolved, but am still seeing build bot failures, nearly exclusively on PPC. (And one arm self host bot.) </p>
<p>@PPC Bot Owner - I need help reducing a test case for the failure seen here: <a href="https://lab.llvm.org/buildbot#builders/100/builds/5762" moz-do-not-send="true" target="_blank">https://lab.llvm.org/buildbot#builders/100/builds/5762</a>. I am stuck, and unable to make progress for nearly 3 months now. I would greatly appreciate help. </p>
<p>Philip</p>
<div>On 5/5/21 11:59 PM, Adrian Kuegel wrote:</div>
<blockquote cite="mid:CAHhV=OX3NbAExy4YsQWLH0fC-zXiG_VpToAtB72NhkUwwYekRQ@mail.gmail.com" type="cite" ><!--Notes ACF
<meta http-equiv="content-type" content="text/html; charset=utf8" >-->
<div dir="ltr" ><div dir="ltr" >Sounds good to me, and thanks for sending the heads up :)</div>
<div><div dir="ltr" >On Wed, May 5, 2021 at 10:01 PM Philip Reames <<a href="mailto:listmail@philipreames.com" moz-do-not-send="true" target="_blank">listmail@philipreames.com</a>> wrote:</div>
<blockquote style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex" >FYI, I'm going to try recommitting this without changes in a day or so.<br><br>I never heard back from a PPC bot owner, and I don't have enough<br>information to really debug anything from the builtbot log. I did run<br>across a latent issue which this patch may very well have exposed at<br>much higher frequency; the previous patch in this series (which is much<br>more restrictive) does appear to have increased frequency. That was<br>worked around in 80e80250. My educated guess is that same issue<br>triggered the miscompile seen on the ppc bot, but that is more of a<br>guess than I'd really prefer.<br><br>I'm going to submit this during off hours, and watch the bots fairly<br>closely after submit. Hopefully this either cycles clean or I get a<br>better clue as to what the root issue is.<br><br>Philip<br><br>On 2/8/21 9:09 PM, Philip Reames via llvm-commits wrote:<br>> Ben,<br>><br>> Thanks for the clarification. The log does not make the fact this is<br>> an execution failure obvious.<br>><br>> No, I don't have access to a PPC machine.<br>><br>> I am going to need some assistance from the bot owner on this. At a<br>> minimum, IR for the test in question (before optimization, but on<br>> target platform) seems like a reasonable ask.<br>><br>> I strongly suspect this change is simply exposing another latent<br>> issue. Or at least, I've reviewed the change and don't see anything<br>> likely to cause runtime crashes w/o also tripping compiler asserts.<br>><br>> Philip<br>><br>><br>> On 2/8/21 5:21 AM, Benjamin Kramer wrote:<br>>> `execution_time` failures mean that the bot succeeded building a test<br>>> but it failed when running it. I'm relatively certain that this is the<br>>> same issue Adrian is seeing -- binaries segfaulting early on PPC.<br>>><br>>> The bot log output isn't helpful at all for investigating why this is<br>>> happening. Do you happen to have access to a PPC machine?<br>>><br>>> On Fri, Feb 5, 2021 at 6:05 PM Philip Reames via llvm-commits<br>>> <<a href="mailto:llvm-commits@lists.llvm.org" moz-do-not-send="true" target="_blank">llvm-commits@lists.llvm.org</a>> wrote:<br>>>> Adrian,<br>>>><br>>>> I'm going to need you to provide a bit more information here. The test<br>>>> failure in stage1 was fixed at the time you reverted this patch. The<br>>>> remaining failure in the bot is very unclear. What is a execution_time<br>>>> failure? From the log output, the "failing" run finished in 0.5<br>>>> seconds,<br>>>> whereas the previous "succeeding" run finished in 11 seconds. Without<br>>>> further context, I'd say that's no failure.<br>>>><br>>>> I'll also note that I did not receive email from this bot. I received<br>>>> notice from the various other bots and fixed the ARM test issue, but<br>>>> unless I missed it in with the others, this bot is not notifying.<br>>>><br>>>> In general, I'm a fan of fast reverts, but I have to admit, this one<br>>>> appears borderline at the moment.<br>>>><br>>>> Philip<br>>>><br>>>> On 2/5/21 3:53 AM, Adrian Kuegel via llvm-commits wrote:<br>>>>> Author: Adrian Kuegel<br>>>>> Date: 2021-02-05T12:51:03+01:00<br>>>>> New Revision: 7fe41ac3dff2d44c3d2c31b28554fbe4a86eaa6c<br>>>>><br>>>>> URL:<br>>>>> <a href="https://github.com/llvm/llvm-project/commit/7fe41ac3dff2d44c3d2c31b28554fbe4a86eaa6c" moz-do-not-send="true" rel="noreferrer" target="_blank">https://github.com/llvm/llvm-project/commit/7fe41ac3dff2d44c3d2c31b28554fbe4a86eaa6c</a><br>>>>> DIFF:<br>>>>> <a href="https://github.com/llvm/llvm-project/commit/7fe41ac3dff2d44c3d2c31b28554fbe4a86eaa6c.diff" moz-do-not-send="true" rel="noreferrer" target="_blank">https://github.com/llvm/llvm-project/commit/7fe41ac3dff2d44c3d2c31b28554fbe4a86eaa6c.diff</a><br>>>>><br>>>>> LOG: Revert "[LV] Unconditionally branch from middle to scalar<br>>>>> preheader if the scalar loop must execute"<br>>>>><br>>>>> This reverts commit 3e5ce49e5371ce4feadbf97dd5c2b652d9db3d1d.<br>>>>><br>>>>> Tests started failing on PPC, for example:<br>>>>> <a href="http://lab.llvm.org:8011/#/builders/105/builds/5569" moz-do-not-send="true" rel="noreferrer" target="_blank">http://lab.llvm.org:8011/#/builders/105/builds/5569</a><br>>>>><br>>>>> Added:<br>>>>><br>>>>><br>>>>> Modified:<br>>>>> llvm/lib/Transforms/Utils/LoopVersioning.cpp<br>>>>> llvm/lib/Transforms/Vectorize/LoopVectorize.cpp<br>>>>> llvm/test/Transforms/LoopVectorize/first-order-recurrence-complex.ll<br>>>>> llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll<br>>>>> llvm/test/Transforms/LoopVectorize/loop-form.ll<br>>>>><br>>>>> Removed:<br>>>>><br>>>>><br>>>>><br>>>>> ################################################################################<br>>>>><br>>>>> diff --git a/llvm/lib/Transforms/Utils/LoopVersioning.cpp<br>>>>> b/llvm/lib/Transforms/Utils/LoopVersioning.cpp<br>>>>> index 8a89158788cf..de4fb446fdf2 100644<br>>>>> --- a/llvm/lib/Transforms/Utils/LoopVersioning.cpp<br>>>>> +++ b/llvm/lib/Transforms/Utils/LoopVersioning.cpp<br>>>>> @@ -44,11 +44,11 @@ LoopVersioning::LoopVersioning(const<br>>>>> LoopAccessInfo &LAI,<br>>>>> AliasChecks(Checks.begin(), Checks.end()),<br>>>>> Preds(LAI.getPSE().getUnionPredicate()), LAI(LAI), LI(LI),<br>>>>> DT(DT),<br>>>>> SE(SE) {<br>>>>> + assert(L->getUniqueExitBlock() && "No single exit block");<br>>>>> }<br>>>>><br>>>>> void LoopVersioning::versionLoop(<br>>>>> const SmallVectorImpl<Instruction *> &DefsUsedOutside) {<br>>>>> - assert(VersionedLoop->getUniqueExitBlock() && "No single exit<br>>>>> block");<br>>>>> assert(VersionedLoop->isLoopSimplifyForm() &&<br>>>>> "Loop is not in loop-simplify form");<br>>>>><br>>>>><br>>>>> diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp<br>>>>> b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp<br>>>>> index 3277842edbfe..6bce0caeb36f 100644<br>>>>> --- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp<br>>>>> +++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp<br>>>>> @@ -852,7 +852,7 @@ class InnerLoopVectorizer {<br>>>>> /// Middle Block between the vector and the scalar.<br>>>>> BasicBlock *LoopMiddleBlock;<br>>>>><br>>>>> - /// The unique ExitBlock of the scalar loop if one exists. Note<br>>>>> that<br>>>>> + /// The (unique) ExitBlock of the scalar loop. Note that<br>>>>> /// there can be multiple exiting edges reaching this block.<br>>>>> BasicBlock *LoopExitBlock;<br>>>>><br>>>>> @@ -3147,13 +3147,9 @@ void<br>>>>> InnerLoopVectorizer::emitMinimumIterationCountCheck(Loop *L,<br>>>>> DT->getNode(Bypass)->getIDom()) &&<br>>>>> "TC check is expected to dominate Bypass");<br>>>>><br>>>>> - // Update dominator for Bypass & LoopExit (if needed).<br>>>>> + // Update dominator for Bypass & LoopExit.<br>>>>> DT->changeImmediateDominator(Bypass, TCCheckBlock);<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // If there is an epilogue which must run, there's no edge<br>>>>> from the<br>>>>> - // middle block to exit blocks and thus no need to update the<br>>>>> immediate<br>>>>> - // dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock, TCCheckBlock);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock, TCCheckBlock);<br>>>>><br>>>>> ReplaceInstWithInst(<br>>>>> TCCheckBlock->getTerminator(),<br>>>>> @@ -3192,11 +3188,7 @@ void<br>>>>> InnerLoopVectorizer::emitSCEVChecks(Loop *L, BasicBlock *Bypass) {<br>>>>> // Update dominator only if this is first RT check.<br>>>>> if (LoopBypassBlocks.empty()) {<br>>>>> DT->changeImmediateDominator(Bypass, SCEVCheckBlock);<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // If there is an epilogue which must run, there's no edge<br>>>>> from the<br>>>>> - // middle block to exit blocks and thus no need to update<br>>>>> the immediate<br>>>>> - // dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock, SCEVCheckBlock);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock, SCEVCheckBlock);<br>>>>> }<br>>>>><br>>>>> ReplaceInstWithInst(<br>>>>> @@ -3252,11 +3244,7 @@ void<br>>>>> InnerLoopVectorizer::emitMemRuntimeChecks(Loop *L, BasicBlock<br>>>>> *Bypass) {<br>>>>> // Update dominator only if this is first RT check.<br>>>>> if (LoopBypassBlocks.empty()) {<br>>>>> DT->changeImmediateDominator(Bypass, MemCheckBlock);<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // If there is an epilogue which must run, there's no edge<br>>>>> from the<br>>>>> - // middle block to exit blocks and thus no need to update<br>>>>> the immediate<br>>>>> - // dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock, MemCheckBlock);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock, MemCheckBlock);<br>>>>> }<br>>>>><br>>>>> Instruction *FirstCheckInst;<br>>>>> @@ -3381,10 +3369,9 @@ Value<br>>>>> *InnerLoopVectorizer::emitTransformedIndex(<br>>>>> Loop *InnerLoopVectorizer::createVectorLoopSkeleton(StringRef<br>>>>> Prefix) {<br>>>>> LoopScalarBody = OrigLoop->getHeader();<br>>>>> LoopVectorPreHeader = OrigLoop->getLoopPreheader();<br>>>>> + LoopExitBlock = OrigLoop->getUniqueExitBlock();<br>>>>> + assert(LoopExitBlock && "Must have an exit block");<br>>>>> assert(LoopVectorPreHeader && "Invalid loop structure");<br>>>>> - LoopExitBlock = OrigLoop->getUniqueExitBlock(); // may be nullptr<br>>>>> - assert((LoopExitBlock || Cost->requiresScalarEpilogue()) &&<br>>>>> - "multiple exit loop without required epilogue?");<br>>>>><br>>>>> LoopMiddleBlock =<br>>>>> SplitBlock(LoopVectorPreHeader,<br>>>>> LoopVectorPreHeader->getTerminator(), DT,<br>>>>> @@ -3393,20 +3380,12 @@ Loop<br>>>>> *InnerLoopVectorizer::createVectorLoopSkeleton(StringRef Prefix) {<br>>>>> SplitBlock(LoopMiddleBlock,<br>>>>> LoopMiddleBlock->getTerminator(), DT, LI,<br>>>>> nullptr, Twine(Prefix) + "<a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>");<br>>>>><br>>>>> + // Set up branch from middle block to the exit and scalar<br>>>>> preheader blocks.<br>>>>> + // completeLoopSkeleton will update the condition to use an<br>>>>> iteration check,<br>>>>> + // if required to decide whether to execute the remainder.<br>>>>> + BranchInst *BrInst =<br>>>>> + BranchInst::Create(LoopExitBlock, LoopScalarPreHeader,<br>>>>> Builder.getTrue());<br>>>>> auto *ScalarLatchTerm =<br>>>>> OrigLoop->getLoopLatch()->getTerminator();<br>>>>> -<br>>>>> - // Set up the middle block terminator. Two cases:<br>>>>> - // 1) If we know that we must execute the scalar epilogue, emit an<br>>>>> - // unconditional branch.<br>>>>> - // 2) Otherwise, we must have a single unique exit block (due to<br>>>>> how we<br>>>>> - // implement the multiple exit case). In this case, set up a<br>>>>> conditonal<br>>>>> - // branch from the middle block to the loop scalar preheader,<br>>>>> and the<br>>>>> - // exit block. completeLoopSkeleton will update the<br>>>>> condition to use an<br>>>>> - // iteration check, if required to decide whether to execute<br>>>>> the remainder.<br>>>>> - BranchInst *BrInst = Cost->requiresScalarEpilogue() ?<br>>>>> - BranchInst::Create(LoopScalarPreHeader) :<br>>>>> - BranchInst::Create(LoopExitBlock, LoopScalarPreHeader,<br>>>>> - Builder.getTrue());<br>>>>> BrInst->setDebugLoc(ScalarLatchTerm->getDebugLoc());<br>>>>> ReplaceInstWithInst(LoopMiddleBlock->getTerminator(), BrInst);<br>>>>><br>>>>> @@ -3418,11 +3397,7 @@ Loop<br>>>>> *InnerLoopVectorizer::createVectorLoopSkeleton(StringRef Prefix) {<br>>>>> nullptr, nullptr, Twine(Prefix) + "vector.body");<br>>>>><br>>>>> // Update dominator for loop exit.<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // If there is an epilogue which must run, there's no edge<br>>>>> from the<br>>>>> - // middle block to exit blocks and thus no need to update the<br>>>>> immediate<br>>>>> - // dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock, LoopMiddleBlock);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock, LoopMiddleBlock);<br>>>>><br>>>>> // Create and register the new vector loop.<br>>>>> Loop *Lp = LI->AllocateLoop();<br>>>>> @@ -3519,14 +3494,10 @@ BasicBlock<br>>>>> *InnerLoopVectorizer::completeLoopSkeleton(Loop *L,<br>>>>> auto *ScalarLatchTerm =<br>>>>> OrigLoop->getLoopLatch()->getTerminator();<br>>>>><br>>>>> // Add a check in the middle block to see if we have completed<br>>>>> - // all of the iterations in the first vector loop. Three cases:<br>>>>> - // 1) If we require a scalar epilogue, there is no conditional<br>>>>> branch as<br>>>>> - // we unconditionally branch to the scalar preheader. Do<br>>>>> nothing.<br>>>>> - // 2) If (N - N%VF) == N, then we *don't* need to run the<br>>>>> remainder.<br>>>>> - // Thus if tail is to be folded, we know we don't need to run<br>>>>> the<br>>>>> - // remainder and we can use the previous value for the<br>>>>> condition (true).<br>>>>> - // 3) Otherwise, construct a runtime check.<br>>>>> - if (!Cost->requiresScalarEpilogue() &&<br>>>>> !Cost->foldTailByMasking()) {<br>>>>> + // all of the iterations in the first vector loop.<br>>>>> + // If (N - N%VF) == N, then we *don't* need to run the remainder.<br>>>>> + // If tail is to be folded, we know we don't need to run the<br>>>>> remainder.<br>>>>> + if (!Cost->foldTailByMasking()) {<br>>>>> Instruction *CmpN = CmpInst::Create(Instruction::ICmp,<br>>>>> CmpInst::ICMP_EQ,<br>>>>> Count, VectorTripCount,<br>>>>> "cmp.n",<br>>>>> LoopMiddleBlock->getTerminator());<br>>>>> @@ -3590,17 +3561,17 @@ BasicBlock<br>>>>> *InnerLoopVectorizer::createVectorizedLoopSkeleton() {<br>>>>> | [ ]_| <-- vector loop.<br>>>>> | |<br>>>>> | v<br>>>>> - \ -[ ] <--- middle-block.<br>>>>> - \/ |<br>>>>> - /\ v<br>>>>> - | ->[ ] <--- new preheader.<br>>>>> + | -[ ] <--- middle-block.<br>>>>> + | / |<br>>>>> + | / v<br>>>>> + -|- >[ ] <--- new preheader.<br>>>>> | |<br>>>>> - (opt) v <-- edge from middle to exit iff epilogue is not<br>>>>> required.<br>>>>> + | v<br>>>>> | [ ] \<br>>>>> - | [ ]_| <-- old scalar loop to handle remainder (scalar<br>>>>> epilogue).<br>>>>> + | [ ]_| <-- old scalar loop to handle remainder.<br>>>>> \ |<br>>>>> \ v<br>>>>> - >[ ] <-- exit block(s).<br>>>>> + >[ ] <-- exit block.<br>>>>> ...<br>>>>> */<br>>>>><br>>>>> @@ -4021,18 +3992,13 @@ void<br>>>>> InnerLoopVectorizer::fixVectorizedLoop() {<br>>>>> // Forget the original basic block.<br>>>>> PSE.getSE()->forgetLoop(OrigLoop);<br>>>>><br>>>>> - // If we inserted an edge from the middle block to the unique<br>>>>> exit block,<br>>>>> - // update uses outside the loop (phis) to account for the newly<br>>>>> inserted<br>>>>> - // edge.<br>>>>> - if (!Cost->requiresScalarEpilogue()) {<br>>>>> - // Fix-up external users of the induction variables.<br>>>>> - for (auto &Entry : Legal->getInductionVars())<br>>>>> - fixupIVUsers(Entry.first, Entry.second,<br>>>>> - getOrCreateVectorTripCount(LI->getLoopFor(LoopVectorBody)),<br>>>>> - IVEndValues[Entry.first], LoopMiddleBlock);<br>>>>> + // Fix-up external users of the induction variables.<br>>>>> + for (auto &Entry : Legal->getInductionVars())<br>>>>> + fixupIVUsers(Entry.first, Entry.second,<br>>>>> + getOrCreateVectorTripCount(LI->getLoopFor(LoopVectorBody)),<br>>>>> + IVEndValues[Entry.first], LoopMiddleBlock);<br>>>>><br>>>>> - fixLCSSAPHIs();<br>>>>> - }<br>>>>> + fixLCSSAPHIs();<br>>>>> for (Instruction *PI : PredicatedInstructions)<br>>>>> sinkScalarOperands(&*PI);<br>>>>><br>>>>> @@ -4250,13 +4216,12 @@ void<br>>>>> InnerLoopVectorizer::fixFirstOrderRecurrence(PHINode *Phi) {<br>>>>> // recurrence in the exit block, and then add an edge for the<br>>>>> middle block.<br>>>>> // Note that LCSSA does not imply single entry when the<br>>>>> original scalar loop<br>>>>> // had multiple exiting edges (as we always run the last<br>>>>> iteration in the<br>>>>> - // scalar epilogue); in that case, there is no edge from middle<br>>>>> to exit and<br>>>>> - // and thus no phis which needed updated.<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - for (PHINode &LCSSAPhi : LoopExitBlock->phis())<br>>>>> - if (any_of(LCSSAPhi.incoming_values(),<br>>>>> - [Phi](Value *V) { return V == Phi; }))<br>>>>> - LCSSAPhi.addIncoming(ExtractForPhiUsedOutsideLoop,<br>>>>> LoopMiddleBlock);<br>>>>> + // scalar epilogue); in that case, the exiting path through<br>>>>> middle will be<br>>>>> + // dynamically dead and the value picked for the phi doesn't<br>>>>> matter.<br>>>>> + for (PHINode &LCSSAPhi : LoopExitBlock->phis())<br>>>>> + if (any_of(LCSSAPhi.incoming_values(),<br>>>>> + [Phi](Value *V) { return V == Phi; }))<br>>>>> + LCSSAPhi.addIncoming(ExtractForPhiUsedOutsideLoop,<br>>>>> LoopMiddleBlock);<br>>>>> }<br>>>>><br>>>>> void InnerLoopVectorizer::fixReduction(PHINode *Phi) {<br>>>>> @@ -4421,11 +4386,10 @@ void<br>>>>> InnerLoopVectorizer::fixReduction(PHINode *Phi) {<br>>>>> // We know that the loop is in LCSSA form. We need to update<br>>>>> the PHI nodes<br>>>>> // in the exit blocks. See comment on analogous loop in<br>>>>> // fixFirstOrderRecurrence for a more complete explaination of<br>>>>> the logic.<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - for (PHINode &LCSSAPhi : LoopExitBlock->phis())<br>>>>> - if (any_of(LCSSAPhi.incoming_values(),<br>>>>> - [LoopExitInst](Value *V) { return V ==<br>>>>> LoopExitInst; }))<br>>>>> - LCSSAPhi.addIncoming(ReducedPartRdx, LoopMiddleBlock);<br>>>>> + for (PHINode &LCSSAPhi : LoopExitBlock->phis())<br>>>>> + if (any_of(LCSSAPhi.incoming_values(),<br>>>>> + [LoopExitInst](Value *V) { return V ==<br>>>>> LoopExitInst; }))<br>>>>> + LCSSAPhi.addIncoming(ReducedPartRdx, LoopMiddleBlock);<br>>>>><br>>>>> // Fix the scalar loop reduction variable with the incoming<br>>>>> reduction sum<br>>>>> // from the vector body and from the backedge value.<br>>>>> @@ -8074,11 +8038,7 @@ BasicBlock<br>>>>> *EpilogueVectorizerMainLoop::emitMinimumIterationCountCheck(<br>>>>><br>>>>> // Update dominator for Bypass & LoopExit.<br>>>>> DT->changeImmediateDominator(Bypass, TCCheckBlock);<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // For loops with multiple exits, there's no edge from the<br>>>>> middle block<br>>>>> - // to exit blocks (as the epilogue must run) and thus no<br>>>>> need to update<br>>>>> - // the immediate dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock, TCCheckBlock);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock, TCCheckBlock);<br>>>>><br>>>>> LoopBypassBlocks.push_back(TCCheckBlock);<br>>>>><br>>>>> @@ -8142,12 +8102,7 @@<br>>>>> EpilogueVectorizerEpilogueLoop::createEpilogueVectorizedLoopSkeleton()<br>>>>> {<br>>>>><br>>>>> DT->changeImmediateDominator(LoopScalarPreHeader,<br>>>>> EPI.EpilogueIterationCountCheck);<br>>>>> - if (!Cost->requiresScalarEpilogue())<br>>>>> - // If there is an epilogue which must run, there's no edge<br>>>>> from the<br>>>>> - // middle block to exit blocks and thus no need to update the<br>>>>> immediate<br>>>>> - // dominator of the exit blocks.<br>>>>> - DT->changeImmediateDominator(LoopExitBlock,<br>>>>> - EPI.EpilogueIterationCountCheck);<br>>>>> + DT->changeImmediateDominator(LoopExitBlock,<br>>>>> EPI.EpilogueIterationCountCheck);<br>>>>><br>>>>> // Keep track of bypass blocks, as they feed start values to<br>>>>> the induction<br>>>>> // phis in the scalar loop preheader.<br>>>>><br>>>>> diff --git<br>>>>> a/llvm/test/Transforms/LoopVectorize/first-order-recurrence-complex.ll<br>>>>> b/llvm/test/Transforms/LoopVectorize/first-order-recurrence-complex.ll<br>>>>> index ec280bf5d5e4..7d4a3c5c9935 100644<br>>>>> ---<br>>>>> a/llvm/test/Transforms/LoopVectorize/first-order-recurrence-complex.ll<br>>>>> +++<br>>>>> b/llvm/test/Transforms/LoopVectorize/first-order-recurrence-complex.ll<br>>>>> @@ -471,9 +471,10 @@ define i16 @multiple_exit(i16* %p, i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]<br>>>>> ; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement<br>>>>> <4 x i16> [[WIDE_LOAD]], i32 3<br>>>>> ; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] =<br>>>>> extractelement <4 x i16> [[WIDE_LOAD]], i32 2<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i16 [ 0,<br>>>>> [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]<br>>>>> @@ -485,14 +486,14 @@ define i16 @multiple_exit(i16* %p, i32 %n) {<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> ; CHECK-NEXT: [[REC_NEXT]] = load i16, i16* [[B]], align 2<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: store i16 [[SCALAR_RECUR]], i16* [[B]], align 4<br>>>>> ; CHECK-NEXT: [[INC]] = add nsw i32 [[I]], 1<br>>>>> ; CHECK-NEXT: [[CMP2:%.*]] = icmp slt i32 [[I]], 2096<br>>>>> ; CHECK-NEXT: br i1 [[CMP2]], label [[FOR_COND]], label<br>>>>> [[IF_END]], [[LOOP7:!llvm.loop !.*]]<br>>>>> ; CHECK: if.end:<br>>>>> -; CHECK-NEXT: [[REC_LCSSA:%.*]] = phi i16 [ [[SCALAR_RECUR]],<br>>>>> [[FOR_BODY]] ], [ [[SCALAR_RECUR]], [[FOR_COND]] ]<br>>>>> +; CHECK-NEXT: [[REC_LCSSA:%.*]] = phi i16 [ [[SCALAR_RECUR]],<br>>>>> [[FOR_BODY]] ], [ [[SCALAR_RECUR]], [[FOR_COND]] ], [<br>>>>> [[VECTOR_RECUR_EXTRACT_FOR_PHI]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: ret i16 [[REC_LCSSA]]<br>>>>> ;<br>>>>> entry:<br>>>>> @@ -557,9 +558,10 @@ define i16 @multiple_exit2(i16* %p, i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP8:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]<br>>>>> ; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement<br>>>>> <4 x i16> [[WIDE_LOAD]], i32 3<br>>>>> ; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] =<br>>>>> extractelement <4 x i16> [[WIDE_LOAD]], i32 2<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i16 [ 0,<br>>>>> [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]<br>>>>> @@ -571,14 +573,14 @@ define i16 @multiple_exit2(i16* %p, i32 %n) {<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> ; CHECK-NEXT: [[REC_NEXT]] = load i16, i16* [[B]], align 2<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: store i16 [[SCALAR_RECUR]], i16* [[B]], align 4<br>>>>> ; CHECK-NEXT: [[INC]] = add nsw i32 [[I]], 1<br>>>>> ; CHECK-NEXT: [[CMP2:%.*]] = icmp slt i32 [[I]], 2096<br>>>>> ; CHECK-NEXT: br i1 [[CMP2]], label [[FOR_COND]], label<br>>>>> [[IF_END]], [[LOOP9:!llvm.loop !.*]]<br>>>>> ; CHECK: if.end:<br>>>>> -; CHECK-NEXT: [[REC_LCSSA:%.*]] = phi i16 [ [[SCALAR_RECUR]],<br>>>>> [[FOR_COND]] ], [ 10, [[FOR_BODY]] ]<br>>>>> +; CHECK-NEXT: [[REC_LCSSA:%.*]] = phi i16 [ [[SCALAR_RECUR]],<br>>>>> [[FOR_COND]] ], [ 10, [[FOR_BODY]] ], [<br>>>>> [[VECTOR_RECUR_EXTRACT_FOR_PHI]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: ret i16 [[REC_LCSSA]]<br>>>>> ;<br>>>>> entry:<br>>>>><br>>>>> diff --git<br>>>>> a/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll<br>>>>> b/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll<br>>>>> index f0ba677348ab..0d4bdf0ecac3 100644<br>>>>> --- a/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll<br>>>>> +++ b/llvm/test/Transforms/LoopVectorize/interleaved-accesses.ll<br>>>>> @@ -447,7 +447,7 @@ define void @even_load_static_tc(i32* noalias<br>>>>> nocapture readonly %A, i32* noalia<br>>>>> ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 508<br>>>>> ; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP12:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_COND_CLEANUP:%.*]],<br>>>>> label [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 1016,<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_BODY:%.*]]<br>>>>> @@ -463,7 +463,7 @@ define void @even_load_static_tc(i32* noalias<br>>>>> nocapture readonly %A, i32* noalia<br>>>>> ; CHECK-NEXT: store i32 [[MUL]], i32* [[ARRAYIDX2]], align 4<br>>>>> ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64<br>>>>> [[INDVARS_IV]], 2<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[INDVARS_IV]], 1022<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[FOR_COND_CLEANUP:%.*]], [[LOOP13:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[FOR_COND_CLEANUP]], [[LOOP13:!llvm.loop !.*]]<br>>>>> ;<br>>>>> entry:<br>>>>> br label %for.body<br>>>>> @@ -528,7 +528,7 @@ define void @even_load_dynamic_tc(i32* noalias<br>>>>> nocapture readonly %A, i32* noali<br>>>>> ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP14:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_COND_CLEANUP:%.*]],<br>>>>> label [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_BODY:%.*]]<br>>>>> @@ -544,7 +544,7 @@ define void @even_load_dynamic_tc(i32* noalias<br>>>>> nocapture readonly %A, i32* noali<br>>>>> ; CHECK-NEXT: store i32 [[MUL]], i32* [[ARRAYIDX2]], align 4<br>>>>> ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64<br>>>>> [[INDVARS_IV]], 2<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[INDVARS_IV_NEXT]],<br>>>>> [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[FOR_COND_CLEANUP:%.*]], [[LOOP15:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[FOR_COND_CLEANUP]], [[LOOP15:!llvm.loop !.*]]<br>>>>> ;<br>>>>> entry:<br>>>>> br label %for.body<br>>>>> @@ -973,7 +973,7 @@ define void @PR27626_0(%pair.i32 *%p, i32 %z,<br>>>>> i64 %n) {<br>>>>> ; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP24:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_BODY:%.*]]<br>>>>> @@ -985,7 +985,7 @@ define void @PR27626_0(%pair.i32 *%p, i32 %z,<br>>>>> i64 %n) {<br>>>>> ; CHECK-NEXT: store i32 [[Z]], i32* [[P_I_Y]], align 4<br>>>>> ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1<br>>>>> ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END:%.*]], [[LOOP25:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END]], [[LOOP25:!llvm.loop !.*]]<br>>>>> ; CHECK: for.end:<br>>>>> ; CHECK-NEXT: ret void<br>>>>> ;<br>>>>> @@ -1066,7 +1066,7 @@ define i32 @PR27626_1(%pair.i32 *%p, i64 %n) {<br>>>>> ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <4 x i32><br>>>>> [[BIN_RDX]], <4 x i32> poison, <4 x i32> <i32 1, i32 undef, i32<br>>>>> undef, i32 undef><br>>>>> ; CHECK-NEXT: [[BIN_RDX4:%.*]] = add <4 x i32> [[BIN_RDX]],<br>>>>> [[RDX_SHUF3]]<br>>>>> ; CHECK-NEXT: [[TMP19:%.*]] = extractelement <4 x i32><br>>>>> [[BIN_RDX4]], i32 0<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP19]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]<br>>>>> @@ -1081,9 +1081,10 @@ define i32 @PR27626_1(%pair.i32 *%p, i64 %n) {<br>>>>> ; CHECK-NEXT: [[TMP21]] = add nsw i32 [[TMP20]], [[S]]<br>>>>> ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1<br>>>>> ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END:%.*]], [[LOOP27:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END]], [[LOOP27:!llvm.loop !.*]]<br>>>>> ; CHECK: for.end:<br>>>>> -; CHECK-NEXT: ret i32 [[TMP21]]<br>>>>> +; CHECK-NEXT: [[TMP22:%.*]] = phi i32 [ [[TMP21]], [[FOR_BODY]]<br>>>>> ], [ [[TMP19]], [[MIDDLE_BLOCK]] ]<br>>>>> +; CHECK-NEXT: ret i32 [[TMP22]]<br>>>>> ;<br>>>>> entry:<br>>>>> br label %for.body<br>>>>> @@ -1162,7 +1163,7 @@ define void @PR27626_2(%pair.i32 *%p, i64 %n,<br>>>>> i32 %z) {<br>>>>> ; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP28:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_BODY:%.*]]<br>>>>> @@ -1176,7 +1177,7 @@ define void @PR27626_2(%pair.i32 *%p, i64 %n,<br>>>>> i32 %z) {<br>>>>> ; CHECK-NEXT: store i32 [[TMP21]], i32* [[P_I_Y]], align 4<br>>>>> ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1<br>>>>> ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END:%.*]], [[LOOP29:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END]], [[LOOP29:!llvm.loop !.*]]<br>>>>> ; CHECK: for.end:<br>>>>> ; CHECK-NEXT: ret void<br>>>>> ;<br>>>>> @@ -1263,7 +1264,7 @@ define i32 @PR27626_3(%pair.i32 *%p, i64 %n,<br>>>>> i32 %z) {<br>>>>> ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <4 x i32><br>>>>> [[BIN_RDX]], <4 x i32> poison, <4 x i32> <i32 1, i32 undef, i32<br>>>>> undef, i32 undef><br>>>>> ; CHECK-NEXT: [[BIN_RDX4:%.*]] = add <4 x i32> [[BIN_RDX]],<br>>>>> [[RDX_SHUF3]]<br>>>>> ; CHECK-NEXT: [[TMP22:%.*]] = extractelement <4 x i32><br>>>>> [[BIN_RDX4]], i32 0<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: br i1 false, label [[FOR_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP22]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]<br>>>>> @@ -1281,9 +1282,10 @@ define i32 @PR27626_3(%pair.i32 *%p, i64 %n,<br>>>>> i32 %z) {<br>>>>> ; CHECK-NEXT: [[TMP25]] = add nsw i32 [[TMP24]], [[S]]<br>>>>> ; CHECK-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1<br>>>>> ; CHECK-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END:%.*]], [[LOOP31:!llvm.loop !.*]]<br>>>>> +; CHECK-NEXT: br i1 [[COND]], label [[FOR_BODY]], label<br>>>>> [[FOR_END]], [[LOOP31:!llvm.loop !.*]]<br>>>>> ; CHECK: for.end:<br>>>>> -; CHECK-NEXT: ret i32 [[TMP25]]<br>>>>> +; CHECK-NEXT: [[TMP26:%.*]] = phi i32 [ [[TMP25]], [[FOR_BODY]]<br>>>>> ], [ [[TMP22]], [[MIDDLE_BLOCK]] ]<br>>>>> +; CHECK-NEXT: ret i32 [[TMP26]]<br>>>>> ;<br>>>>> entry:<br>>>>> br label %for.body<br>>>>><br>>>>> diff --git a/llvm/test/Transforms/LoopVectorize/loop-form.ll<br>>>>> b/llvm/test/Transforms/LoopVectorize/loop-form.ll<br>>>>> index f32002fae2b6..91780789088b 100644<br>>>>> --- a/llvm/test/Transforms/LoopVectorize/loop-form.ll<br>>>>> +++ b/llvm/test/Transforms/LoopVectorize/loop-form.ll<br>>>>> @@ -146,14 +146,15 @@ define void @early_exit(i16* %p, i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP1]], [[N_VEC]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_COND:%.*]]<br>>>>> ; CHECK: for.cond:<br>>>>> ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]],<br>>>>> [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: [[IPROM:%.*]] = sext i32 [[I]] to i64<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> @@ -285,14 +286,15 @@ define void @multiple_unique_exit(i16* %p,<br>>>>> i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_COND:%.*]]<br>>>>> ; CHECK: for.cond:<br>>>>> ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]],<br>>>>> [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: [[IPROM:%.*]] = sext i32 [[I]] to i64<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> @@ -372,14 +374,17 @@ define i32 @multiple_unique_exit2(i16* %p,<br>>>>> i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP8:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]<br>>>>> +; CHECK-NEXT: [[IND_ESCAPE:%.*]] = sub i32 [[N_VEC]], 1<br>>>>> +; CHECK-NEXT: [[IND_ESCAPE1:%.*]] = sub i32 [[N_VEC]], 1<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_COND:%.*]]<br>>>>> ; CHECK: for.cond:<br>>>>> ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]],<br>>>>> [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: [[IPROM:%.*]] = sext i32 [[I]] to i64<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> @@ -388,7 +393,7 @@ define i32 @multiple_unique_exit2(i16* %p, i32<br>>>>> %n) {<br>>>>> ; CHECK-NEXT: [[CMP2:%.*]] = icmp slt i32 [[I]], 2096<br>>>>> ; CHECK-NEXT: br i1 [[CMP2]], label [[FOR_COND]], label<br>>>>> [[IF_END]], [[LOOP9:!llvm.loop !.*]]<br>>>>> ; CHECK: if.end:<br>>>>> -; CHECK-NEXT: [[I_LCSSA:%.*]] = phi i32 [ [[I]], [[FOR_BODY]]<br>>>>> ], [ [[I]], [[FOR_COND]] ]<br>>>>> +; CHECK-NEXT: [[I_LCSSA:%.*]] = phi i32 [ [[I]], [[FOR_BODY]]<br>>>>> ], [ [[I]], [[FOR_COND]] ], [ [[IND_ESCAPE1]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: ret i32 [[I_LCSSA]]<br>>>>> ;<br>>>>> ; TAILFOLD-LABEL: @multiple_unique_exit2(<br>>>>> @@ -461,14 +466,15 @@ define i32 @multiple_unique_exit3(i16* %p,<br>>>>> i32 %n) {<br>>>>> ; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i32 [[INDEX_NEXT]],<br>>>>> [[N_VEC]]<br>>>>> ; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP10:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[TMP2]], [[N_VEC]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[IF_END:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]],<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[FOR_COND:%.*]]<br>>>>> ; CHECK: for.cond:<br>>>>> ; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[BC_RESUME_VAL]],<br>>>>> [[SCALAR_PH]] ], [ [[INC:%.*]], [[FOR_BODY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[I]], [[N]]<br>>>>> -; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label<br>>>>> [[IF_END:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[IF_END]]<br>>>>> ; CHECK: for.body:<br>>>>> ; CHECK-NEXT: [[IPROM:%.*]] = sext i32 [[I]] to i64<br>>>>> ; CHECK-NEXT: [[B:%.*]] = getelementptr inbounds i16, i16*<br>>>>> [[P]], i64 [[IPROM]]<br>>>>> @@ -477,7 +483,7 @@ define i32 @multiple_unique_exit3(i16* %p, i32<br>>>>> %n) {<br>>>>> ; CHECK-NEXT: [[CMP2:%.*]] = icmp slt i32 [[I]], 2096<br>>>>> ; CHECK-NEXT: br i1 [[CMP2]], label [[FOR_COND]], label<br>>>>> [[IF_END]], [[LOOP11:!llvm.loop !.*]]<br>>>>> ; CHECK: if.end:<br>>>>> -; CHECK-NEXT: [[EXIT:%.*]] = phi i32 [ 0, [[FOR_COND]] ], [ 1,<br>>>>> [[FOR_BODY]] ]<br>>>>> +; CHECK-NEXT: [[EXIT:%.*]] = phi i32 [ 0, [[FOR_COND]] ], [ 1,<br>>>>> [[FOR_BODY]] ], [ 0, [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: ret i32 [[EXIT]]<br>>>>> ;<br>>>>> ; TAILFOLD-LABEL: @multiple_unique_exit3(<br>>>>> @@ -994,7 +1000,8 @@ define void @scalar_predication(float* %addr) {<br>>>>> ; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 200<br>>>>> ; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]],<br>>>>> label [[VECTOR_BODY]], [[LOOP12:!llvm.loop !.*]]<br>>>>> ; CHECK: middle.block:<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 201, 200<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 200,<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]<br>>>>> @@ -1002,7 +1009,7 @@ define void @scalar_predication(float* %addr) {<br>>>>> ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]],<br>>>>> [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]<br>>>>> ; CHECK-NEXT: [[GEP:%.*]] = getelementptr float, float*<br>>>>> [[ADDR]], i64 [[IV]]<br>>>>> ; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV]], 200<br>>>>> -; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT:%.*]], label<br>>>>> [[LOOP_BODY:%.*]]<br>>>>> +; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label<br>>>>> [[LOOP_BODY:%.*]]<br>>>>> ; CHECK: loop.body:<br>>>>> ; CHECK-NEXT: [[TMP11:%.*]] = load float, float* [[GEP]],<br>>>>> align 4<br>>>>> ; CHECK-NEXT: [[PRED:%.*]] = fcmp oeq float [[TMP11]],<br>>>>> 0.000000e+00<br>>>>> @@ -1088,7 +1095,8 @@ define i32 @me_reduction(i32* %addr) {<br>>>>> ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <2 x i32><br>>>>> [[TMP5]], <2 x i32> poison, <2 x i32> <i32 1, i32 undef><br>>>>> ; CHECK-NEXT: [[BIN_RDX:%.*]] = add <2 x i32> [[TMP5]],<br>>>>> [[RDX_SHUF]]<br>>>>> ; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x i32><br>>>>> [[BIN_RDX]], i32 0<br>>>>> -; CHECK-NEXT: br label [[SCALAR_PH]]<br>>>>> +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 201, 200<br>>>>> +; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label<br>>>>> [[SCALAR_PH]]<br>>>>> ; CHECK: <a href="http://scalar.ph" moz-do-not-send="true" rel="noreferrer" target="_blank">scalar.ph</a>:<br>>>>> ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 200,<br>>>>> [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]<br>>>>> ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]]<br>>>>> ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]<br>>>>> @@ -1098,7 +1106,7 @@ define i32 @me_reduction(i32* %addr) {<br>>>>> ; CHECK-NEXT: [[ACCUM:%.*]] = phi i32 [ [[BC_MERGE_RDX]],<br>>>>> [[SCALAR_PH]] ], [ [[ACCUM_NEXT:%.*]], [[LOOP_LATCH]] ]<br>>>>> ; CHECK-NEXT: [[GEP:%.*]] = getelementptr i32, i32* [[ADDR]],<br>>>>> i64 [[IV]]<br>>>>> ; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV]], 200<br>>>>> -; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT:%.*]], label<br>>>>> [[LOOP_LATCH]]<br>>>>> +; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label<br>>>>> [[LOOP_LATCH]]<br>>>>> ; CHECK: loop.latch:<br>>>>> ; CHECK-NEXT: [[TMP8:%.*]] = load i32, i32* [[GEP]], align 4<br>>>>> ; CHECK-NEXT: [[ACCUM_NEXT]] = add i32 [[ACCUM]], [[TMP8]]<br>>>>> @@ -1106,7 +1114,7 @@ define i32 @me_reduction(i32* %addr) {<br>>>>> ; CHECK-NEXT: [[EXITCOND2_NOT:%.*]] = icmp eq i64 [[IV]], 400<br>>>>> ; CHECK-NEXT: br i1 [[EXITCOND2_NOT]], label [[EXIT]], label<br>>>>> [[LOOP_HEADER]], [[LOOP15:!llvm.loop !.*]]<br>>>>> ; CHECK: exit:<br>>>>> -; CHECK-NEXT: [[LCSSA:%.*]] = phi i32 [ 0, [[LOOP_HEADER]] ], [<br>>>>> [[ACCUM_NEXT]], [[LOOP_LATCH]] ]<br>>>>> +; CHECK-NEXT: [[LCSSA:%.*]] = phi i32 [ 0, [[LOOP_HEADER]] ], [<br>>>>> [[ACCUM_NEXT]], [[LOOP_LATCH]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ]<br>>>>> ; CHECK-NEXT: ret i32 [[LCSSA]]<br>>>>> ;<br>>>>> ; TAILFOLD-LABEL: @me_reduction(<br>>>>><br>>>>><br>>>>><br>>>>> _______________________________________________<br>>>>> llvm-commits mailing list<br>>>>> <a href="mailto:llvm-commits@lists.llvm.org" moz-do-not-send="true" target="_blank">llvm-commits@lists.llvm.org</a><br>>>>> <a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" moz-do-not-send="true" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>>>> _______________________________________________<br>>>> llvm-commits mailing list<br>>>> <a href="mailto:llvm-commits@lists.llvm.org" moz-do-not-send="true" target="_blank">llvm-commits@lists.llvm.org</a><br>>>> <a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" moz-do-not-send="true" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>> _______________________________________________<br>> llvm-commits mailing list<br>> <a href="mailto:llvm-commits@lists.llvm.org" moz-do-not-send="true" target="_blank">llvm-commits@lists.llvm.org</a><br>> <a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" moz-do-not-send="true" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a></blockquote></div></div></blockquote></blockquote>
<div dir="ltr" > </div></div></blockquote></blockquote>
<div dir="ltr" > </div></div></blockquote></blockquote>
<div dir="ltr" > </div></div>
<BR>