<html><body><p><font size="2">Hi Evgeny,</font><br><br><font size="2">Yes, please do. It was our hope that folks would verify the impact of the partial inliner on the platforms they're currently working on.</font><br><br><font size="2">Cheers,</font><br><br><font size="2">Graham Yiu<br>LLVM Compiler Development<br>IBM Toronto Software Lab<br>Office: (905) 413-4077 C2-707/8200/Markham<br>Email: gyiu@ca.ibm.com</font><br><br><img width="16" height="16" src="cid:1__=8FBB0B5EDFC8B8D28f9e8a93df938690918c8FB@" border="0" alt="Inactive hide details for Evgeny Astigeevich ---11/03/2017 12:18:05 PM---Hi, We'd like to check impact on armv7m and armv6m tar"><font size="2" color="#424282">Evgeny Astigeevich ---11/03/2017 12:18:05 PM---Hi, We'd like to check impact on armv7m and armv6m targets, especially code size. We have not tried</font><br><br><font size="2" color="#5F5F5F">From: </font><font size="2">Evgeny Astigeevich <Evgeny.Astigeevich@arm.com></font><br><font size="2" color="#5F5F5F">To: </font><font size="2">Tobias Grosser <tobias.grosser@inf.ethz.ch>, Graham Yiu <gyiu@ca.ibm.com></font><br><font size="2" color="#5F5F5F">Cc: </font><font size="2">"junbuml@codeaurora.org" <junbuml@codeaurora.org>, "llvm-dev@lists.llvm.org" <llvm-dev@lists.llvm.org>, nd <nd@arm.com></font><br><font size="2" color="#5F5F5F">Date: </font><font size="2">11/03/2017 12:18 PM</font><br><font size="2" color="#5F5F5F">Subject: </font><font size="2">Re: [llvm-dev] [RFC] Enable Partial Inliner by default</font><br><hr width="100%" size="2" align="left" noshade style="color:#8091A5; "><br><br><br><tt><font size="2">Hi,<br><br><br><br>We'd like to check impact on armv7m and armv6m targets, especially code size. We have not tried the partial inliner on them.<br><br><br><br>Could a decision to turn it on by default wait for results?<br><br><br><br>Thanks,<br><br>Evgeny Astigeevich<br><br>The Arm Compiler Optimization team<br><br><br><br>-----Original Message-----<br><br>From: llvm-dev <llvm-dev-bounces@lists.llvm.org> on behalf of Tobias Grosser via llvm-dev <llvm-dev@lists.llvm.org><br><br>Reply-To: Tobias Grosser <tobias.grosser@inf.ethz.ch><br><br>Date: Thursday, 2 November 2017 at 23:32<br><br>To: Graham Yiu <gyiu@ca.ibm.com>, "llvm-dev@lists.llvm.org" <llvm-dev@lists.llvm.org><br><br>Cc: "junbuml@codeaurora.org" <junbuml@codeaurora.org><br><br>Subject: Re: [llvm-dev] [RFC] Enable Partial Inliner by default<br><br><br><br>Hi Graham,<br><br><br><br>I think this is a good idea. It is also useful for libquantum, where<br><br>together with some other changes, it enables Polly to perform libfusion.<br><br><br><br>The ARM people also played with the partial inliner and might have<br><br>feedback.<br><br><br><br>Best,<br><br>Tobias<br><br><br><br>On Thu, Nov 2, 2017, at 23:05, Graham Yiu via llvm-dev wrote:<br><br>> <br><br>> Forgot to add that all experiments were done with '-O3 -m64<br><br>> -fexperimental-new-pass-manager'.<br><br>> <br><br>> Graham Yiu<br><br>> LLVM Compiler Development<br><br>> IBM Toronto Software Lab<br><br>> Office: (905) 413-4077 C2-707/8200/Markham<br><br>> Email: gyiu@ca.ibm.com<br><br>> <br><br>> <br><br>> <br><br>> From: Graham Yiu/Toronto/IBM<br><br>> To: llvm-dev@lists.llvm.org<br><br>> Cc: junbuml@codeaurora.org, xinliangli@gmail.com<br><br>> Date: 11/02/2017 05:26 PM<br><br>> Subject: [RFC] Enable Partial Inliner by default<br><br>> <br><br>> <br><br>> Hello,<br><br>> <br><br>> I'd like to propose turning on the partial inliner<br><br>> (-enable-partial-inlining) by default.<br><br>> <br><br>> We've seen small gains on SPEC2006/2017 runtimes as well as lnt<br><br>> compile-times with a 2nd stage bootstrap of LLVM. We also saw positive<br><br>> gains on our internal workloads.<br><br>> <br><br>> -------------------------------------<br><br>> Brief description of Partial Inlining<br><br>> -------------------------------------<br><br>> A pass in opt that runs after the normal inlining pass. Looks for<br><br>> branches<br><br>> to a return block in the entry and immediate successor blocks of a<br><br>> function. If found, it outlines the rest of the function using the<br><br>> CodeExtractor. It then attempts to inline the leftover entry block (and<br><br>> possibly one or more of its successors) to all its callers. This<br><br>> effectively peels the early return block(s) into the caller, which could<br><br>> be<br><br>> executed without incurring the call overhead of the function just to<br><br>> return<br><br>> immediately. Inlining and call overhead cost, as well as branch<br><br>> probabilities of the return block(s) are taken into account before<br><br>> inlining<br><br>> is done. If inlining is not successful, then the changes are discarded.<br><br>> <br><br>> eg.<br><br>> <br><br>> void foo() {<br><br>> bar();<br><br>> // rest of the code in foo<br><br>> }<br><br>> <br><br>> void bar() {<br><br>> if (X)<br><br>> return;<br><br>> // rest of code (to be outlined)<br><br>> }<br><br>> <br><br>> After Partial Inlining:<br><br>> <br><br>> void foo() {<br><br>> if (!X)<br><br>> bar.outlined();<br><br>> // rest of the code in foo<br><br>> }<br><br>> <br><br>> void bar.outlined() {<br><br>> // rest of the code in bar<br><br>> }<br><br>> <br><br>> <br><br>> Here are the numbers on a Power8 PPCLE running Ubuntu 15.04 in ST-mode<br><br>> <br><br>> ----------------------------------------------<br><br>> Runtime performance (speed)<br><br>> ----------------------------------------------<br><br>> Workload Improvement<br><br>> -------- -----------<br><br>> SPEC2006(C/C++) 0.06% (geomean)<br><br>> SPEC2017(C/C++) 0.10% (geomean)<br><br>> ----------------------------------------------<br><br>> Compile time performance for Bootstrapped LLVM<br><br>> ----------------------------------------------<br><br>> Workload Improvement<br><br>> -------- -----------<br><br>> SPEC2006(C/C++) 0.41% (cumulative)<br><br>> SPEC2017(C/C++) -0.16% (cumulative)<br><br>> lnt 0.61% (geomean)<br><br>> ----------------------------------------------<br><br>> Compile time performance<br><br>> ----------------------------------------------<br><br>> Workload Increase<br><br>> -------- --------<br><br>> SPEC2006(C/C++) 1.31% (cumulative)<br><br>> SPEC2017(C/C++) 0.25% (cumulative)<br><br>> ----------------------------------------------<br><br>> Code size<br><br>> ----------------------------------------------<br><br>> Workload Increase<br><br>> -------- --------<br><br>> SPEC2006(C/C++) 3.90% (geomean)<br><br>> SPEC2017(C/C++) 1.05% (geomean)<br><br>> <br><br>> NOTE1: Code size increase in SPEC2006 was mainly attributed to benchmark<br><br>> "astar", which increased by 86%. Removing this outlier, we get a more<br><br>> reasonable increase of 0.58%.<br><br>> <br><br>> NOTE2: There is a patch up for review on Phabricator to enhance the<br><br>> partial<br><br>> inliner with the presence of profiling information (<br><br>> </font></tt><tt><font size="2"><a href="https://urldefense.proofpoint.com/v2/url?u=https-3A__reviews.llvm.org_D38190&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=6o17wydYZM0l4kPAb3l3cJ95JRPoYb-3l4sHv-R0GaA&e=">https://urldefense.proofpoint.com/v2/url?u=https-3A__reviews.llvm.org_D38190&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=6o17wydYZM0l4kPAb3l3cJ95JRPoYb-3l4sHv-R0GaA&e=</a></font></tt><tt><font size="2">).<br><br>> <br><br>> <br><br>> Graham Yiu<br><br>> LLVM Compiler Development<br><br>> IBM Toronto Software Lab<br><br>> Office: (905) 413-4077 C2-707/8200/Markham<br><br>> Email: gyiu@ca.ibm.com<br><br>> <br><br>> _______________________________________________<br><br>> LLVM Developers mailing list<br><br>> llvm-dev@lists.llvm.org<br><br>> </font></tt><tt><font size="2"><a href="https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.llvm.org_cgi-2Dbin_mailman_listinfo_llvm-2Ddev&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=_WAS3iXS9l627yoGcLCkw5IMyoeBRXAb3ShcSIW5qjk&e=">https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.llvm.org_cgi-2Dbin_mailman_listinfo_llvm-2Ddev&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=_WAS3iXS9l627yoGcLCkw5IMyoeBRXAb3ShcSIW5qjk&e=</a></font></tt><tt><font size="2"><br><br>> Email had 1 attachment:<br><br>> + graycol.gif<br><br>> 1k (image/gif)<br><br>_______________________________________________<br><br>LLVM Developers mailing list<br><br>llvm-dev@lists.llvm.org<br><br></font></tt><tt><font size="2"><a href="https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.llvm.org_cgi-2Dbin_mailman_listinfo_llvm-2Ddev&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=_WAS3iXS9l627yoGcLCkw5IMyoeBRXAb3ShcSIW5qjk&e=">https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.llvm.org_cgi-2Dbin_mailman_listinfo_llvm-2Ddev&d=DwIGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=4ST7e3kMd0GTi3w9ByK5Cw&m=sY89ox2ivgmox5Vg311rAsEr4WFT-o-LRopDU9e7rl0&s=_WAS3iXS9l627yoGcLCkw5IMyoeBRXAb3ShcSIW5qjk&e=</a></font></tt><tt><font size="2"><br><br><br><br><br><br><br></font></tt><br><br><BR>
</body></html>