<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"><head><meta http-equiv=Content-Type content="text/html; charset=utf-8"><meta name=Generator content="Microsoft Word 15 (filtered medium)"><style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0in;
margin-bottom:.0001pt;
font-size:12.0pt;
font-family:"Times New Roman",serif;}
a:link, span.MsoHyperlink
{mso-style-priority:99;
color:blue;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:purple;
text-decoration:underline;}
span.EmailStyle17
{mso-style-type:personal-reply;
font-family:"Calibri",sans-serif;
color:#1F497D;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri",sans-serif;}
@page WordSection1
{size:8.5in 11.0in;
margin:1.0in 1.0in 1.0in 1.0in;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]--></head><body lang=EN-US link=blue vlink=purple><div class=WordSection1><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>Most likely what is going on is that making isLegalAddImmediate() more precise is leading LoopStrengthReduce to make a different decision about some IV solution, which happens to be worse in this case.<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>I saw a bit of this in SPEC, but the net perf delta was positive, and the change itself seems good in isolation.<o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>I’ll investigate this case to see if my suspicion is correct, and if so if there’s anything we can do about it. <o:p></o:p></span></p><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'><o:p> </o:p></span></p><div style='mso-element:para-border-div;border:dashed #2F6FAB 1.0pt;padding:12.0pt 12.0pt 12.0pt 12.0pt;background:#F9F9F9'><p class=MsoNormal style='line-height:15.6pt;background:#F9F9F9;border:none;padding:0in'><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>--<o:p></o:p></span></p><p class=MsoNormal style='line-height:15.6pt;background:#F9F9F9;border:none;padding:0in'><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>Geoff Berry<o:p></o:p></span></p><p class=MsoNormal style='line-height:15.6pt;background:#F9F9F9;border:none;padding:0in'><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'>Employee of Qualcomm Innovation Center, Inc.<o:p></o:p></span></p><p class=MsoNormal style='line-height:15.6pt;background:#F9F9F9;border:none;padding:0in'><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project<o:p></o:p></span></p></div><p class=MsoNormal><span style='font-size:11.0pt;font-family:"Calibri",sans-serif;color:#1F497D'><o:p> </o:p></span></p><p class=MsoNormal><b><span style='font-size:11.0pt;font-family:"Calibri",sans-serif'>From:</span></b><span style='font-size:11.0pt;font-family:"Calibri",sans-serif'> James Molloy [mailto:james@jamesmolloy.co.uk] <br><b>Sent:</b> Tuesday, March 01, 2016 11:31 AM<br><b>To:</b> Geoff Berry; llvm-commits@lists.llvm.org<br><b>Subject:</b> Re: [llvm] r262248 - [AArch64] Fix isLegalAddImmediate() to return true for valid negative values.<o:p></o:p></span></p><p class=MsoNormal><o:p> </o:p></p><div><p class=MsoNormal>Hi Geoff,<o:p></o:p></p><div><p class=MsoNormal><o:p> </o:p></p></div><div><p class=MsoNormal>This commit caused a 10% regression in huffbench: <a href="http://llvm.org/perf/db_default/v4/nts/daily_report/2016/3/1?filter-machine-regex=aarch64%7Carm%7Cgreen">http://llvm.org/perf/db_default/v4/nts/daily_report/2016/3/1?filter-machine-regex=aarch64%7Carm%7Cgreen</a><o:p></o:p></p></div><div><p class=MsoNormal><o:p> </o:p></p></div><div><p class=MsoNormal>There are also many other tests in that report that have regressed that I suspect to be the same revision. We've also seen quite large swings (not all negative!) on internal/third party tests.<o:p></o:p></p></div><div><p class=MsoNormal><o:p> </o:p></p></div><div><p class=MsoNormal>Could you please take a look at what could be going on?<o:p></o:p></p></div><div><p class=MsoNormal><o:p> </o:p></p></div><div><p class=MsoNormal>Cheers,<o:p></o:p></p></div><div><p class=MsoNormal><o:p> </o:p></p></div><div><p class=MsoNormal>James<o:p></o:p></p></div></div><p class=MsoNormal><o:p> </o:p></p><div><div><p class=MsoNormal>On Mon, 29 Feb 2016 at 19:57 Geoff Berry via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a>> wrote:<o:p></o:p></p></div><blockquote style='border:none;border-left:solid #CCCCCC 1.0pt;padding:0in 0in 0in 6.0pt;margin-left:4.8pt;margin-right:0in'><p class=MsoNormal>Author: gberry<br>Date: Mon Feb 29 13:53:22 2016<br>New Revision: 262248<br><br>URL: <a href="http://llvm.org/viewvc/llvm-project?rev=262248&view=rev" target="_blank">http://llvm.org/viewvc/llvm-project?rev=262248&view=rev</a><br>Log:<br>[AArch64] Fix isLegalAddImmediate() to return true for valid negative values.<br><br>Reviewers: t.p.northover, jmolloy<br><br>Subscribers: mcrosier, aemerson, llvm-commits, rengolin<br><br>Differential Revision: <a href="http://reviews.llvm.org/D17463" target="_blank">http://reviews.llvm.org/D17463</a><br><br>Added:<br> llvm/trunk/test/CodeGen/AArch64/neg-imm.ll<br>Modified:<br> llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp<br><br>Modified: llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp?rev=262248&r1=262247&r2=262248&view=diff" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp?rev=262248&r1=262247&r2=262248&view=diff</a><br>==============================================================================<br>--- llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp (original)<br>+++ llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp Mon Feb 29 13:53:22 2016<br>@@ -7280,6 +7280,8 @@ EVT AArch64TargetLowering::getOptimalMem<br><br> // 12-bit optionally shifted immediates are legal for adds.<br> bool AArch64TargetLowering::isLegalAddImmediate(int64_t Immed) const {<br>+ // Same encoding for add/sub, just flip the sign.<br>+ Immed = std::abs(Immed);<br> if ((Immed >> 12) == 0 || ((Immed & 0xfff) == 0 && Immed >> 24 == 0))<br> return true;<br> return false;<br>@@ -7288,8 +7290,6 @@ bool AArch64TargetLowering::isLegalAddIm<br> // Integer comparisons are implemented with ADDS/SUBS, so the range of valid<br> // immediates is the same as for an add or a sub.<br> bool AArch64TargetLowering::isLegalICmpImmediate(int64_t Immed) const {<br>- if (Immed < 0)<br>- Immed *= -1;<br> return isLegalAddImmediate(Immed);<br> }<br><br><br>Added: llvm/trunk/test/CodeGen/AArch64/neg-imm.ll<br>URL: <a href="http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/neg-imm.ll?rev=262248&view=auto" target="_blank">http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/neg-imm.ll?rev=262248&view=auto</a><br>==============================================================================<br>--- llvm/trunk/test/CodeGen/AArch64/neg-imm.ll (added)<br>+++ llvm/trunk/test/CodeGen/AArch64/neg-imm.ll Mon Feb 29 13:53:22 2016<br>@@ -0,0 +1,46 @@<br>+; RUN: llc -mtriple=aarch64-linux-gnu -verify-machineinstrs -o - %s | FileCheck %s<br>+; LSR used to pick a sub-optimal solution due to the target responding<br>+; conservatively to isLegalAddImmediate for negative values.<br>+<br>+declare void @foo(i32)<br>+<br>+define void @test(i32 %px) {<br>+; CHECK-LABEL: test:<br>+; CHECK: // %entry<br>+; CHECK: subs<br>+; CHECK-NEXT: csel<br>+entry:<br>+ %sub = add nsw i32 %px, -1<br>+ %cmp = icmp slt i32 %px, 1<br>+ %.sub = select i1 %cmp, i32 0, i32 %sub<br>+ br label %for.body<br>+<br>+for.body:<br>+; CHECK: // %for.body<br>+; CHECK: cmp<br>+; CHECK-NEXT: b.eq<br>+; CHECK: // %if.then3<br>+ %x.015 = phi i32 [ %inc, %for.inc ], [ %.sub, %entry ]<br>+ %cmp2 = icmp eq i32 %x.015, %px<br>+ br i1 %cmp2, label %for.inc, label %if.then3<br>+<br>+if.then3:<br>+ tail call void @foo(i32 %x.015)<br>+ br label %for.inc<br>+<br>+for.inc:<br>+; CHECK: // %for.inc<br>+; CHECK: add<br>+; CHECK-NEXT: cmp<br>+; CHECK: b.le<br>+; CHECK: // %for.cond.cleanup<br>+ %inc = add nsw i32 %x.015, 1<br>+ %cmp1 = icmp sgt i32 %x.015, %px<br>+ br i1 %cmp1, label %for.cond.cleanup.loopexit, label %for.body<br>+<br>+for.cond.cleanup.loopexit:<br>+ br label %for.cond.cleanup<br>+<br>+for.cond.cleanup:<br>+ ret void<br>+}<br><br><br>_______________________________________________<br>llvm-commits mailing list<br><a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a><br><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><o:p></o:p></p></blockquote></div></div></body></html>