<html><head><meta http-equiv="Content-Type" content="text/html; charset=utf-8"></head><body style="word-wrap: break-word; -webkit-nbsp-mode: space; line-break: after-white-space;" class="">Hi Marina,<br class=""><div><br class=""><blockquote type="cite" class=""><div class="">On Oct 2, 2017, at 11:11 AM, Yatsina, Marina <<a href="mailto:marina.yatsina@intel.com" class="">marina.yatsina@intel.com</a>> wrote:</div><br class="Apple-interchange-newline"><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Hi Quentin,<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><br class="">Let’s currently focus on the “bad eviction” patch which is under review, as the other issue (“spill cost of<span class="Apple-converted-space"> </span></span><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">local interval</span><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">”) is still half-baked and will probably need more refinement<span class="Apple-converted-space"> </span></span><span style="font-size: 11pt; font-family: Wingdings; color: rgb(31, 73, 125);" class="">J</span><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><br class=""></span></div></div></div></blockquote><div><br class=""></div><div>Sounds good.</div><br class=""><blockquote type="cite" class=""><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><br class=""><o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">The compilation time measurements I mentioned refer to the time of the whole compilation.</span></div></div></div></blockquote><div><br class=""></div><div>Okay. 4% is a sizable chunk of time. Therefore, I would suggest to hide this new heuristic behind a TargetLowering hook and a command line option so that people can decide whether or not they want to pay the extract cost when the patch lands.</div><div>The default would be off, then we should an email on llvm-dev to ask people to try it to see if we want/could change that default.</div><br class=""><blockquote type="cite" class=""><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">I’ve ran the measurements a couple of times again and I see that my compilation times are not consistent.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">For example, a workload that compiles ~4 minutes - at one run compiles 30% worse with my patch and at another run compiles 5% better with my patch.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Workloads that have short compile time (10-20 seconds) also show such big variations.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">The performance gains are consistent and I see them both for workloads that run 10 seconds and for workloads that run 5 minutes.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">* The gains and the size of the gain depend very much on the revision you are checking – you may see a very nice gain at revision X, but if at revision X+1 somebody changed something in an earlier pass which affects the llvm ir/machine ir, it may change the register allocation and now these bad eviction chains don’t happen anymore for this workload, so this patch will no longer show gains.</span></div></div></div></blockquote><div><br class=""></div><div>Okay.</div><br class=""><blockquote type="cite" class=""><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">The additional prediction I’ve added only happens if the candidate will create local intervals and only for those intervals.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">The check itself of whether a local interval will be created does not add additional complexity.<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">If the compile time is still a concern - can you check the compile on your systems and let me know if you see significant issues with my patch?</span></div></div></div></blockquote><div><br class=""></div><div>Let us go with the “hidden behind an option” for now and do broader measurements later. I don’t have time to do it now.</div><br class=""><blockquote type="cite" class=""><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Also, do you have any comments regarding the patch itself?</span></div></div></div></blockquote><div><br class=""></div><div>I have to find some time for the details, but one high level (nitpick) comment is your please double check your comments, some were not sentences (like no period at the end).</div><div><br class=""></div><div>Cheers,</div><div>-Quentin </div><br class=""><blockquote type="cite" class=""><div class=""><div class="WordSection1" style="page: WordSection1; font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;"><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Thanks,<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Marina<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><a name="_MailEndCompose" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""><o:p class=""> </o:p></span></a></div><div class=""><div style="border-style: solid none none; border-top-width: 1pt; border-top-color: rgb(225, 225, 225); padding: 3pt 0cm 0cm;" class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><b class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif;" class="">From:</span></b><span style="font-size: 11pt; font-family: Calibri, sans-serif;" class=""><span class="Apple-converted-space"> </span><a href="mailto:qcolombet@apple.com" class="">qcolombet@apple.com</a> [<a href="mailto:qcolombet@apple.com" class="">mailto:qcolombet@apple.com</a>]<span class="Apple-converted-space"> </span><br class=""><b class="">Sent:</b><span class="Apple-converted-space"> </span>Thursday, September 28, 2017 19:45<br class=""><b class="">To:</b><span class="Apple-converted-space"> </span>Yatsina, Marina <<a href="mailto:marina.yatsina@intel.com" class="">marina.yatsina@intel.com</a>><br class=""><b class="">Cc:</b><span class="Apple-converted-space"> </span><a href="mailto:reviews+D35816+public+4cab5a74a0b7bc06@reviews.llvm.org" class="">reviews+D35816+public+4cab5a74a0b7bc06@reviews.llvm.org</a>; <a href="mailto:stoklund@2pi.dk" class="">stoklund@2pi.dk</a>; Matthias Braun <<a href="mailto:matze@braunis.de" class="">matze@braunis.de</a>>; llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" class="">llvm-commits@lists.llvm.org</a>><br class=""><b class="">Subject:</b><span class="Apple-converted-space"> </span>Re: [PATCH] D35816: [Greedy RegAlloc] Add logic to greedy reg alloc to avoid bad eviction chains<o:p class=""></o:p></span></div></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Hi Maria,<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Thanks for the numbers.<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Couple more questions.<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">On Sep 27, 2017, at 4:19 PM, Yatsina, Marina <<a href="mailto:marina.yatsina@intel.com" style="color: purple; text-decoration: underline;" class="">marina.yatsina@intel.com</a>> wrote:<o:p class=""></o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">According to my measurement the patch in the review (considering if the local interval might cause "bad" eviction) gives to 2-13% gain in eembc, coremark-pro and geekbench.</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">I have a regression of about 2.5% which seem to be related to different decisions taken later down the road because now the allocation is changed. It seems to expose another not optimal decision of the register allocator and probably worth investigating, but unrelated to the problem I'm trying to solve now.</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">compile time - improvement of up to 17.4x, regression up to 9.5x, geomean - 1.8% more compile time.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Is this the pass itself of the whole compiler?<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Also, could you share the raw data, if we are talking minutes in one case and seconds in the other case (and vice-versa) this is not really comparable.<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""> </span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">The initial patch I have for considering if the local interval might spill gives up to 50% gain.</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Here I have a few regressions of 2-4.5% which I'm investigating.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Ditto, raw numbers would be helpful.<o:p class=""></o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">compile time - improvement of up to 8x, regression up to 6.8x, geomean - 14% more compile time.</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""> </span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">When combining both patches: gains of up to 50%, regression of up to 5.24%.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">compile time - improvement of up to 6x, regression up to 5.7x, geomean - 4.3% more compile time.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Ditto and same question about pass vs. whole compile time.<o:p class=""></o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""> </span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">I will try to improve the compile time for the spill consideration, this should also positively affect the compile time of the combination of the 2 patches together.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Thanks,<o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">-Quentin<br class=""><br class=""><br class=""><o:p class=""></o:p></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""> </span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Thanks,</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class="">Marina</span><o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif; color: rgb(31, 73, 125);" class=""> </span><o:p class=""></o:p></div></div><div class=""><div style="border-style: solid none none; border-top-width: 1pt; border-top-color: rgb(225, 225, 225); padding: 3pt 0cm 0cm;" class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><b class=""><span style="font-size: 11pt; font-family: Calibri, sans-serif;" class="">From:</span></b><span class="apple-converted-space"><span style="font-size: 11pt; font-family: Calibri, sans-serif;" class=""> </span></span><span style="font-size: 11pt; font-family: Calibri, sans-serif;" class=""><a href="mailto:qcolombet@apple.com" style="color: purple; text-decoration: underline;" class="">qcolombet@apple.com</a><span class="Apple-converted-space"> </span>[<a href="mailto:qcolombet@apple.com" style="color: purple; text-decoration: underline;" class="">mailto:qcolombet@apple.com</a>]<span class="apple-converted-space"> </span><br class=""><b class="">Sent:</b><span class="apple-converted-space"> </span>Monday, September 25, 2017 23:10<br class=""><b class="">To:</b><span class="apple-converted-space"> </span>Yatsina, Marina <<a href="mailto:marina.yatsina@intel.com" style="color: purple; text-decoration: underline;" class="">marina.yatsina@intel.com</a>><br class=""><b class="">Cc:</b><span class="apple-converted-space"> </span><a href="mailto:reviews+D35816+public+4cab5a74a0b7bc06@reviews.llvm.org" style="color: purple; text-decoration: underline;" class="">reviews+D35816+public+4cab5a74a0b7bc06@reviews.llvm.org</a>;<span class="Apple-converted-space"> </span><a href="mailto:stoklund@2pi.dk" style="color: purple; text-decoration: underline;" class="">stoklund@2pi.dk</a>; Matthias Braun <<a href="mailto:matze@braunis.de" style="color: purple; text-decoration: underline;" class="">matze@braunis.de</a>>; llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" style="color: purple; text-decoration: underline;" class="">llvm-commits@lists.llvm.org</a>>; Quentin Colombet <<a href="mailto:qcolombet@apple.com" style="color: purple; text-decoration: underline;" class="">qcolombet@apple.com</a>><br class=""><b class="">Subject:</b><span class="apple-converted-space"> </span>Re: [PATCH] D35816: [Greedy RegAlloc] Add logic to greedy reg alloc to avoid bad eviction chains</span><o:p class=""></o:p></div></div></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""> <o:p class=""></o:p></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""> <o:p class=""></o:p></div></div><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><br class=""><o:p class=""></o:p></div></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">On Sep 25, 2017, at 10:44 AM, Quentin Colombet via llvm-commits <<a href="mailto:llvm-commits@lists.llvm.org" style="color: purple; text-decoration: underline;" class=""><span style="color: purple;" class="">llvm-commits@lists.llvm.org</span></a>> wrote:<o:p class=""></o:p></div></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""> <o:p class=""></o:p></div></div><div class=""><blockquote style="margin-top: 5pt; margin-bottom: 5pt; font-variant-caps: normal; text-align: start; -webkit-text-stroke-width: 0px; word-spacing: 0px;" class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class=""><br class="">On Sep 25, 2017, at 8:46 AM, Marina Yatsina via Phabricator <<a href="mailto:reviews@reviews.llvm.org" style="color: purple; text-decoration: underline;" class=""><span style="color: purple;" class="">reviews@reviews.llvm.org</span></a>> wrote:<br class=""><br class="">myatsina added a comment.<br class=""><br class="">Hi Quentin,<br class=""><br class="">I wouldn’t say my patch tries to avoid splitting, but rather tries to improve the calculation of the spill weight of split candidates:<br class="">When the register allocator decides to do a region split, it looks for the best physical register candidate for the split.<br class="">The best candidate is the one that will cause the minimal spill cost.<br class="">When calculating the spill cost of each candidate the algorithm takes into account interferences in the entrance/exist of the basic blocks.<br class="">However, there may be interference local to a basic block, which later, during the split itself will cause the creation of a new local interval (which will be local to the basic block) on top of the “by reg” and “by stack” intervals which are created during the split.<br class="">The algorithm currently ignores the fact that this local interval may cause spills (and thus may increase the spill weight of this candidate for the split).<br class=""><br class="">My solution is to try to predict if this split candidate will case the creation of local intervals and if they in turn will cause spills, and add their spill weights to the total weight.<br class="">By doing so, I try to make the spill weight calculation of each candidate more accurate and allow the algorithm to choose a more suitable candidate.<br class=""><br class="">If a local interval is created then we have a few options for its allocation:<br class=""><br class="">1. The interval will be allocated to some free reg – no additional spill cost needed.<br class="">2. The interval may cause an eviction – in some cases this eviction is "bad" and guaranteed to causes a spill (it’s “bad” when you’re evicting the interval that evicted you, kind of like a cat and mouse game - somebody must loose here) - in this patch I’m trying to predict if it’s "bad" or not, and incorporate the spill weight of this interval.<br class="">3. The interval may spill – I’ve already encountered a case where the new local interval is in a hot loop and ends us spilling around all uses – this spill cost wasn’t considered when the candidate was chosen. I have a solution for this case which is based on parts of this patch.<br class="">4. The interval may split – I guess there might be some spill cost to consider here as well, but I didn’t explore this case yet.<br class=""><br class="">I did see nice performance results with my current solution.<span class="apple-converted-space"> </span></span><o:p class=""></o:p></div></div></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class=""><br class="">I can totally see that. I wonder if adding 1K LOC is worth it and in particular if we can’t have the same result with fewer LOC.<br class="">My main concern is we try to predict yet another thing. Again the approach itself as you describe it makes sense, but the trade off complexity reward is not clear to me. In particular, what is the compile time impact, what are the regressions and so on.</span><o:p class=""></o:p></div></div></div></blockquote><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""> <o:p class=""></o:p></div></div></div><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class="">Assuming this all looks good, I’ll give the patch another look, so please share your numbers.<o:p class=""></o:p></div></div></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><br class=""><br class=""><br class=""><o:p class=""></o:p></div></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><div class=""><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class=""><br class=""><br class=""><br class=""></span><o:p class=""></o:p></div></div><blockquote style="margin-top: 5pt; margin-bottom: 5pt;" class=""><p class="MsoNormal" style="margin: 0cm 0cm 12pt; font-size: 12pt; font-family: "Times New Roman", serif;"><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class="">I will try to look into the hint reconciling as well, but I do think that the current spill weight calculation of the split candidates is not accurate enough and we need to consider the affects of those local intervals.<br class=""><br class="">Thanks,<br class="">Marina<br class=""><br class=""><br class="">Repository:<br class="">rL LLVM<br class=""><br class=""><a href="https://reviews.llvm.org/D35816" style="color: purple; text-decoration: underline;" class=""><span style="color: purple;" class="">https://reviews.llvm.org/D35816</span></a><br class=""><br class=""><br class=""><br class=""></span><o:p class=""></o:p></p></blockquote><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class=""><br class="">_______________________________________________<br class="">llvm-commits mailing list<br class=""></span><a href="mailto:llvm-commits@lists.llvm.org" style="color: purple; text-decoration: underline;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif; color: purple;" class="">llvm-commits@lists.llvm.org</span></a><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class=""><br class=""></span><a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" style="color: purple; text-decoration: underline;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif; color: purple;" class="">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</span></a><o:p class=""></o:p></div></div></div></blockquote></div><div class=""><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""> <o:p class=""></o:p></div></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif; font-variant-caps: normal; text-align: start; -webkit-text-stroke-width: 0px; word-spacing: 0px;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class="">---------------------------------------------------------------------<br class="">Intel Israel (74) Limited<o:p class=""></o:p></span></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif; font-variant-caps: normal; text-align: start; -webkit-text-stroke-width: 0px; word-spacing: 0px;" class=""><span style="font-size: 9pt; font-family: Helvetica, sans-serif;" class="">This e-mail and any attachments may contain confidential material for<br class="">the sole use of the intended recipient(s). Any review or distribution<br class="">by others is strictly prohibited. If you are not the intended<br class="">recipient, please contact the sender and delete all copies.<o:p class=""></o:p></span></div></div></blockquote></div><div style="margin: 0cm 0cm 0.0001pt; font-size: 12pt; font-family: "Times New Roman", serif;" class=""><o:p class=""> </o:p></div></div><p style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;" class="">---------------------------------------------------------------------<br class="">Intel Israel (74) Limited</p><p style="font-family: Helvetica; font-size: 12px; font-style: normal; font-variant-caps: normal; font-weight: normal; letter-spacing: normal; text-align: start; text-indent: 0px; text-transform: none; white-space: normal; word-spacing: 0px; -webkit-text-stroke-width: 0px;" class="">This e-mail and any attachments may contain confidential material for<br class="">the sole use of the intended recipient(s). Any review or distribution<br class="">by others is strictly prohibited. If you are not the intended<br class="">recipient, please contact the sender and delete all copies.</p></div></blockquote></div><br class=""></body></html>