[llvm] r342578 - [X86] Handle COPYs of physregs better (regalloc hints)
Jordan Rupprecht via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 1 13:18:24 PDT 2018
Thanks Jonas! We're no longer seeing any compile time regression. I
left a comment on the patch.
On Sat, Sep 29, 2018 at 11:14 AM Jonas Paulsson
<paulsson at linux.vnet.ibm.com> wrote:
>
> Hi Jordan,
>
> thanks for the test case. I think I found the reason for the compile
> time regression - please try https://reviews.llvm.org/D52316.
>
> /Jonas
>
>
> On 2018-09-27 21:52, Jordan Rupprecht wrote:
> > Sure,
> > I wasn't able to reduce the test case much -- the long compilation
> > time seems to be a factor of the length of the method, so a small
> > method wouldn't reproduce it.
> > $ time bin/llc /tmp/repro.ll -o /tmp/repro-r342577.o
> > bin/llc /tmp/repro.ll -o /tmp/repro-r342577.o 2.87s user 0.07s system
> > 99% cpu 2.942 total
> > $ time bin/llc /tmp/repro.ll -o /tmp/repro-r342578.o
> > bin/llc /tmp/repro.ll -o /tmp/repro-r342578.o 7.71s user 0.08s system
> > 99% cpu 7.795 total
> > I've attached repro.ll, but I generated it from the C++ snippet above,
> > with a copy of Eigen pulled from a few minutes ago. It wasn't slow
> > enough with the version of eigen at head, so I added a 35x35 dummy
> > matrix to the example.
> >
> > I've noticed a small sampling of other strange failures caused by this
> > commit as well, one of which just revealed an actual bug in a test,
> > but another which I haven't quite figured out yet -- I'm still trying
> > to see if it's a bug with this commit.
> >
> > On Thu, Sep 27, 2018 at 3:45 AM Jonas Paulsson
> > <paulsson at linux.vnet.ibm.com> wrote:
> >> Hi,
> >>
> >> I would be happy to take a look. Could you please post a preprocessed /
> >> bitcode (reduced if possible) test case, preferably llc input, with a
> >> command line?
> >>
> >> /Jonas
> >>
> >>
> >> On 2018-09-27 10:14, Simon Pilgrim wrote:
> >>> This is a question for Jonas - I just kept the x86 tests up to date
> >>> until the patch was accepted.
> >>>
> >>> On 27/09/2018 00:08, Jordan Rupprecht wrote:
> >>>> Hi Simon,
> >>>>
> >>>> We seem to have some extreme compile slowdowns in sanitizer (asan,
> >>>> msan, and tsan) modes after r342578. In one case, we've seen something
> >>>> built with tsan and -O2 go from 20 seconds to 12 minutes. It doesn't
> >>>> seem to be a general slowdown, just certain "interesting" pieces of
> >>>> code (heavy template users, generated code that has thousands of
> >>>> expressions, etc.)
> >>>>
> >>>> We've seen it in all sanitizer modes, but the tsan example mentioned
> >>>> can be distilled down to something like this:
> >>>>
> >>>> $ time bin/clang -c repro.cpp -I</path/to/eigen> -O2
> >>>> -fno-strict-aliasing -fsanitize=thread -o repro.o
> >>>> @r342577: 5.25s user 0.68s system 99% cpu 5.984 total
> >>>> @r342578: 9.50s user 0.26s system 99% cpu 9.811 total
> >>>>
> >>>> $ cat repro.cpp
> >>>> #include <Eigen/Core>
> >>>> Eigen::MatrixXi foo(int x) {
> >>>> Eigen::MatrixXi m;
> >>>> switch (x) {
> >>>> case 3:
> >>>> m.resize(3, 3);
> >>>> m << 1, 2, 3, //
> >>>> 1, 2, 3, //
> >>>> 1, 2, 3;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> break;
> >>>> case 5:
> >>>> m.resize(5, 5);
> >>>> m << 1, 2, 3, 4, 5, //
> >>>> 1, 2, 3, 4, 5, //
> >>>> 1, 2, 3, 4, 5, //
> >>>> 1, 2, 3, 4, 5, //
> >>>> 1, 2, 3, 4, 5;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> break;
> >>>> case 7:
> >>>> m.resize(7, 7);
> >>>> m << 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7, //
> >>>> 1, 2, 3, 4, 5, 6, 7;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> break;
> >>>> case 25:
> >>>> m.resize(25, 25);
> >>>> m << 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
> >>>> 17, 18, 19,
> >>>> 20, 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> break;
> >>>> case 30:
> >>>> m.resize(30, 30);
> >>>> m << 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16,
> >>>> 17, 18, 19,
> >>>> 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, //
> >>>> 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
> >>>> 18, 19, 20,
> >>>> 21, 22, 23, 24, 25, 26, 27, 28, 29, 30;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> m.array() -= 1;
> >>>> break;
> >>>> default:
> >>>> break;
> >>>> }
> >>>> return m;
> >>>> }
> >>>>
> >>>> In some cases, we can workaround this by rewriting the methods to be
> >>>> simpler. However, it'd be nice to understand what's going on -- is
> >>>> there a O(n^2) algorithm that enableMultipleCopyHints() is enabling
> >>>> that would cause large methods to be slow to compile?
> >>>>
> >>>> Thanks -- Jordan
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4849 bytes
Desc: S/MIME Cryptographic Signature
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20181001/69b712d9/attachment.bin>
More information about the llvm-commits
mailing list