[llvm-dev] RFC: Switching to the new pass manager by default

Thu Oct 19 04:50:46 PDT 2017

I just tried putting a relatively recent version of our merge branch
(r316139 + our local changes) through a small subset of the PS4 testing
with -fexperimental-new-pass-manager enabled.  In general, most tests
passed :).  There are a few failures I'd need to look at in more detail
though.  The most glaring are all our tests related to generating coverage
info:

$ cat 1.cpp
void foo(){}
$ ./build/bin/clang -c -fprofile-instr-generate -fcoverage-mapping 1.cpp
$ ./build/bin/clang -c -fprofile-instr-generate -fcoverage-mapping 1.cpp
-fexperimental-new-pass-manager
instrprof failed to lower an increment
UNREACHABLE executed at
/home/greg/public_git/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp:5893!

which I'm assuming is the same thing as was reported in
https://bugs.llvm.org/show_bug.cgi?id=33773 .  This would definitely be a
prerequisite for us to consider flipping the default.

There are a few tests looking for specific codegen that are seeing
something different.  At a glance, it looks like the output from clang's
-O2 looks roughly similar in some small examples with and without the new
PM, but clang's -O1 is producing very different output.  Whether this is an
issue, I can't say for sure (I've not tried benchmarking the -O1s against
each other yet).  Is this expected?

I've also got an example of a specific file that shows quite a large
compile time increase with the new PM:

Old PM:
   ---User Time---   --System Time--   --User+System--   ---Wall Time---
--- Name ---
  15.8125 ( 96.0%)   0.2969 ( 95.0%)  16.1094 ( 96.0%)  16.1059 ( 96.1%)
Code Generation Time
   0.6563 (  4.0%)   0.0156 (  5.0%)   0.6719 (  4.0%)   0.6609 (  3.9%)
LLVM IR Generation Time
  16.4688 (100.0%)   0.3125 (100.0%)  16.7813 (100.0%)  16.7668 (100.0%)
Total

New PM:
   ---User Time---   --System Time--   --User+System--   ---Wall Time---
--- Name ---
  67.0781 ( 99.1%)   0.3438 ( 95.7%)  67.4219 ( 99.0%)  67.4543 ( 99.0%)
Code Generation Time
   0.6406 (  0.9%)   0.0156 (  4.3%)   0.6563 (  1.0%)   0.6644 (  1.0%)
LLVM IR Generation Time
  67.7188 (100.0%)   0.3594 (100.0%)  68.0781 (100.0%)  68.1187 (100.0%)
Total

It's not an example I can easily share (and also not one I'm easily able to
verify in its present state on a branch without our local changes to make
sure they're not responsible somehow), but if it's of interest I can see
whether I can reduce it to something shareable and raise a bug.

If I get a chance, I'll try putting this through some PS4 game codebases to
see how build time and run time performance compare there, but I'm one of
the few manning the fort while most of the rest of the team is out having a
nice time at the Developers' Meeting this week, so I may not get that
chance for a bit.

Cheers,
-Greg

On 18 October 2017 at 07:50, Chandler Carruth via llvm-dev <
llvm-dev at lists.llvm.org> wrote:

> Greetings everyone!
>
> The new pass manager is getting extremely close to the point where I'm not
> aware of any significant outstanding work needed, and I'd like to see what
> else would be needed to enable it by default. Here are the current
> functionality I'm aware of outstanding:
>
> 1) Does not do non-trivial loop unswitching. Majority of this is in
> https://reviews.llvm.org/D34200 but will need one or two small follow-ups.
>
> 2) Currently, sanitizers don't work correctly with it. Thanks to the work
> of others, the missing infrastructure has been added and I'll send a patch
> to wire this up this week.
>
> 3) Missing support for 'optnone'. I've been working on this, but the
> existing testing wasn't as thorough as I wanted, so it is going slowly.
> I've got about 1/4 of this implemented and should have patches this week or
> next.
>
> 4) Missing opt-bisect (or similar) facility. This looks pretty trivial to
> add, but I've not even started. If anyone is interested in it, go for it.
> We might even be able to do something simpler using the generic debug
> counters and get equivalent functionality.
>
> ... that's it?
>
> Optimization quality / run-time performance:
> - We've been using it at Google extensively and are very happy with the
> optimization quality. Benchmarks look *very* good here.
> - More data from other users would be important.
> - You can try it out with `-fexperimental-new-pass-manager` to Clang
>
> Compile-time performance:
> - Sometimes *much* better due to cached analyses.
> - Sometimes worse, typically due to more / different inlining in turn
> running main pipeline (GVN + InstCombine) more times or over more code.
> - Overall somewhat a wash, but the increased compile times typically due
> to the optimizer "trying" harder, so not too concerning on our end.
> - Again, more feedback from other users good: `-fexperimental-new-pass-manager`
> to Clang
>
> Once the four missing things land, I'll also happily work on collecting
> some of the basics on the test-suite and CTMark. But I suspect more "in the
> wild" data would really be useful here given the significance of the change.
>
> Thoughts? What else (beyond the four items above and feedback on run-time
> and compile-time) would folks like to see?
>
> Once this happens, I'll also be preparing some batch, mechanical updates
> to the test suite to primarily use the new pass manager. Also there is lots
> of documentation updates that will be needed here.
>
> -Chandler
>
> PS: I'll be sending a note to cfe-dev as a "heads up" about this
> discussion as in some ways, the default flip is mostly a Clang default
> flip. But hopefully our doc updates will trigger this being "perceived" as
> the default for other frontends, and I'll try to reach out to other major
> frontends as well (Swift and Rust are on my radar, and I've already started
> talking with Philip Reames about their Falcon JIT).
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20171019/56b136ea/attachment-0001.html>