<div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Nov 15, 2021 at 2:27 PM Arthur Eubanks <<a href="mailto:aeubanks@google.com" target="_blank">aeubanks@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">`llc -O3` does not run the optimization pipeline on the IR, so IR-level optimizations aren't being run unless you use `opt -O3`.<div>`opt -O3` optimizes the IR.</div><div>`llc -O3` (mostly) enables MIR optimizations and better isel. But if the input IR isn't optimized then you lose most optimization opportunities.</div><div>So a typical `clang -O3` would be somewhat equivalent to running Clang's output IR through `opt -O3` then `llc -O3`.</div></div></blockquote><div><br></div><div>Thanks for pointing this out!</div><div><br></div><div>I built a local clang binary (to include a patch WIP) and ran `clang -v -x ir -S -O3 <file.ll>`. It indeed worked equivalently to a three-step operation (`opt` CLI with a patch | `opt` CLI to run CodeGenPrepare | `llc`)</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Nov 15, 2021 at 1:43 PM Mingming Liu <<a href="mailto:mingmingl@google.com" target="_blank">mingmingl@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">I used "llc -print-after-all -O3 <file.ll>" on this IR gives an assembly (<a href="https://godbolt.org/z/K6cszrPPf" target="_blank">https://godbolt.org/z/K6cszrPPf</a>), and codegenprepare indeed runs in `llc` (from `print-after-all` output)<div><br></div><div>The source of my confusion is:</div><div><ol><li>Running the same IR by `opt -O3 -codegenprepare` gives a more optimized IR (<a href="https://godbolt.org/z/fdqTGsqG4" target="_blank">https://godbolt.org/z/fdqTGsqG4</a>)</li><li>Piping the IR of step 1 (<a href="https://godbolt.org/z/544GMqaco" target="_blank">https://godbolt.org/z/544GMqaco</a>) to `llc -O3` gives a better assembly (tail call generated).</li></ol><div>I'm missing something here..</div></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Nov 15, 2021 at 1:00 PM Arthur Eubanks <<a href="mailto:aeubanks@google.com" target="_blank">aeubanks@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">`llc` should always run codegenprepare on IR before isel.</div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Nov 15, 2021 at 11:49 AM Mingming Liu <<a href="mailto:mingmingl@google.com" target="_blank">mingmingl@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div dir="ltr"><br></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Mon, Nov 15, 2021 at 10:34 AM Arthur Eubanks <<a href="mailto:aeubanks@google.com" target="_blank">aeubanks@google.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">`opt` is concerned about the optimization pipeline and `llc` is concerned about the codegen pipeline. codegenprepare is part of the codegen pipeline, not the optimization pipeline. We happen to be able to use `opt` to run codegenprepare on its own because of how legacy PM passes are structured and `llc` is not well suited to run individual IR passes. </div></blockquote><div><br></div><div>These all make sense to me.<br></div><div><br></div><div>(The following idea side-tracks from the original topic, but just brainstorming how to make the tools more friendly).</div><div><br></div><div>If it (piping `opt` and `llc` misses `CodeGenPrepare` and causes surprises) becomes a common question, `llc` tool might be enhanced by emitting a warning/hint to CLI users that the IR probably needs `CodeGenPrepare` pass (if input IR has metadata to record which middle-end passes ran)</div><div><br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">This wouldn't change even if we used the NPM for the codegen pipeline.</div></blockquote><div><br></div><div>I get the point that CodeGenPrepare could be supported in `opt` (w/ NPM) since `opt` does IR to IR transformations.</div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Nov 12, 2021 at 10:15 PM Mingming Liu via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr">Thank you so much Arthur and Yuanfang! These pointers are very educational.<div><br></div><div>Now I realize there are two questions</div><div>1) Use NPM for machine passes; this is the desired state <a href="https://lists.llvm.org/pipermail/llvm-dev/2020-July/143309.html" target="_blank">RFC</a> and <a href="https://reviews.llvm.org/D85168" target="_blank">D85168</a> tries to push forward.</div><div>2) Whether CodeGenPrepare should be enabled by default (e.g., user of opt CLI specifies an IR with sufficient target information, but doesn't enable CodeGenPrepare explicitly).</div><div><br></div><div>From <a href="https://llvm.org/docs/NewPassManager.html#status-of-the-new-and-legacy-pass-managers" target="_blank">https://llvm.org/docs/NewPassManager.html#status-of-the-new-and-legacy-pass-managers</a>, the preferred option is to not run CodeGenPrepare in the default settings (although users can still run it via specifying <i>-passes=codegenprepare</i>).</div><div><br></div><div>I could make sense of the pointers, and understood the rationales better now. </div><div><br></div><div>I'm curious if there were proposals to turn on CodeGenPrepare by default (if IR has sufficient target information). (didn't find one with <a href="https://www.google.com/search?q=llvm+rfc+turning+on+codegenpreare+opt&newwindow=1&sxsrf=AOaemvIqK3A44HhoAdT538LwKCQ_tbhq1g%3A1636783711790&ei=X1aPYcPSL8rU-gSnoq-IDg&oq=llvm+rfc+turning+on+codegenpreare+opt&gs_lcp=Cgdnd3Mtd2l6EAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsAMyBwgAEEcQsANKBAhBGABQAFgAYNYCaAFwAngAgAEAiAEAkgEAmAEAyAEIwAEB&sclient=gws-wiz&ved=0ahUKEwiD_tu91pT0AhVKqp4KHSfRC-EQ4dUDCA4&uact=5" target="_blank">this search query</a>)</div><div>The good thing is that, when someone (e.g., like me when ramping up on the llvm infra) pipes the <i>opt CLI</i> and <i>llc CLI </i>together, the machine assembly is closer to the machine assembly of Clang (in cpp to assembly mode).<br></div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">On Fri, Nov 12, 2021 at 2:17 PM <<a href="mailto:Yuanfang.Chen@sony.com" target="_blank">Yuanfang.Chen@sony.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">Hi Mingming,<br>
<br>
About the status of using the new pass manager for the codegen pipeline, the RFC was here (<a href="http://lists.llvm.org/pipermail/llvm-dev/2020-July/143309.html" rel="noreferrer" target="_blank">http://lists.llvm.org/pipermail/llvm-dev/2020-July/143309.html</a>) but there was no Bugzilla ticket for it, sorry! I've just created one <a href="https://bugs.llvm.org/show_bug.cgi?id=52493" rel="noreferrer" target="_blank">https://bugs.llvm.org/show_bug.cgi?id=52493</a> with updates for anyone who might be interested. I haven't been able to follow up on it for a while but a few in-flight patches are still relevant and in good shape (check PR52493). I'll see if I could push them forward in the near future.<br>
<br>
About codegen-prepare, I don't have much to add other than Arthur's answer, except that D85168 would enable the use case, although it has some dependencies so it's not like that it could be landed soon.<br>
<br>
HTH,<br>
- Yuanfang<br>
<br>
________________________________________<br>
From: llvm-dev <<a href="mailto:llvm-dev-bounces@lists.llvm.org" target="_blank">llvm-dev-bounces@lists.llvm.org</a>> on behalf of Mingming Liu via llvm-dev <<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a>><br>
Sent: Friday, November 12, 2021 10:26 AM<br>
To: <a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
Subject: [llvm-dev] status of CodeGen in new Pass Manager<br>
<br>
Hi,<br>
This is a newbie question around CodeGen related passes and the current status in new Pass Manager.<br>
<br>
From <a href="https://llvm.org/docs/NewPassManager.html#status-of-the-new-and-legacy-pass-managers" rel="noreferrer" target="_blank">https://llvm.org/docs/NewPassManager.html#status-of-the-new-and-legacy-pass-managers</a><<a href="https://urldefense.com/v3/__https://llvm.org/docs/NewPassManager.html*status-of-the-new-and-legacy-pass-managers__;Iw!!JmoZiZGBv3RvKRSx!tI8u93htbfzW8OQkAVIdBlQTDHabCnLJtB2D5fD_OjBuK1ACPDpumEw6GK_dphuBDA$" rel="noreferrer" target="_blank">https://urldefense.com/v3/__https://llvm.org/docs/NewPassManager.html*status-of-the-new-and-legacy-pass-managers__;Iw!!JmoZiZGBv3RvKRSx!tI8u93htbfzW8OQkAVIdBlQTDHabCnLJtB2D5fD_OjBuK1ACPDpumEw6GK_dphuBDA$</a>>, there are ongoing efforts to make the codegen pipeline work in the new Pass Manager (which is great!). Searching in the bug list (<a href="https://bugs.llvm.org/buglist.cgi?component=opt&list_id=226453&product=tools&query_format=advanced&resolution=---&short_desc=codegen&short_desc_type=allwordssubstr" rel="noreferrer" target="_blank">https://bugs.llvm.org/buglist.cgi?component=opt&list_id=226453&product=tools&query_format=advanced&resolution=---&short_desc=codegen&short_desc_type=allwordssubstr</a><<a href="https://urldefense.com/v3/__https://bugs.llvm.org/buglist.cgi?component=opt&list_id=226453&product=tools&query_format=advanced&resolution=---&short_desc=codegen&short_desc_type=allwordssubstr__;!!JmoZiZGBv3RvKRSx!tI8u93htbfzW8OQkAVIdBlQTDHabCnLJtB2D5fD_OjBuK1ACPDpumEw6GK-25d1S-w$" rel="noreferrer" target="_blank">https://urldefense.com/v3/__https://bugs.llvm.org/buglist.cgi?component=opt&list_id=226453&product=tools&query_format=advanced&resolution=---&short_desc=codegen&short_desc_type=allwordssubstr__;!!JmoZiZGBv3RvKRSx!tI8u93htbfzW8OQkAVIdBlQTDHabCnLJtB2D5fD_OjBuK1ACPDpumEw6GK-25d1S-w$</a>>) gives no result.<br>
<br>
I'm wondering if anyone has more information on the current status of CodeGen in the new Pass Manager (a tracking bug or other pointers)?<br>
<br>
The context is that, I'm using opt CLI (by default new PM is used), and surprised that codegenprepare pass doesn't run, so dig down and having more questions :-)<br>
<br>
Any related information will be appreciated!<br>
<br>
--<br>
Thanks,<br>
Mingming<br>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr"><div dir="ltr"><div><div dir="ltr"><div><div><font color="#555555" face="sans-serif" size="2">Thanks,</font></div><div><font color="#555555" face="sans-serif" size="2">Mingming</font></div></div></div></div></div></div>
_______________________________________________<br>
LLVM Developers mailing list<br>
<a href="mailto:llvm-dev@lists.llvm.org" target="_blank">llvm-dev@lists.llvm.org</a><br>
<a href="https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev" rel="noreferrer" target="_blank">https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev</a><br>
</blockquote></div>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr"><div dir="ltr"><div><div dir="ltr"><div><div><font color="#555555" face="sans-serif" size="2">Thanks,</font></div><div><font color="#555555" face="sans-serif" size="2">Mingming</font></div></div></div></div></div></div></div>
</blockquote></div>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr"><div dir="ltr"><div><div dir="ltr"><div><div><font color="#555555" face="sans-serif" size="2">Thanks,</font></div><div><font color="#555555" face="sans-serif" size="2">Mingming</font></div></div></div></div></div></div>
</blockquote></div>
</blockquote></div><br clear="all"><div><br></div>-- <br><div dir="ltr"><div dir="ltr"><div><div dir="ltr"><div><div><font color="#555555" face="sans-serif" size="2">Thanks,</font></div><div><font color="#555555" face="sans-serif" size="2">Mingming</font></div></div></div></div></div></div></div>