[llvm] r217144 - Enable noalias metadata by default and swap the order of the SLP and Loop vectorizers by default.

[Yi Kong]

Hi Hal,

Loop vectorizer is found to cause the regression. The LLVM IR
immediately after the loop vectorizer may be malformed, which is
unveiled by EarlyCSE pass enabled by '-fslp-vectorize-aggressive'.

I've filed a bug report against it.

Thanks,
Yi

On 26 September 2014 01:10, Hal Finkel <hfinkel at anl.gov> wrote:
> Hi Yi,
>
> This is almost likely a bug in BBVectorize. Can you please file a bug report.
>
> Out of curiosity, are there still cases where -fslp-vectorize-aggressive provides additional speedup over -O3 (which should now include -fslp-vectorize)?
>
> Thanks again,
> Hal
>
> ----- Original Message -----
>> From: "Yi Kong" <kongy.dev at gmail.com>
>> To: "James Molloy" <james.molloy at arm.com>
>> Cc: llvm-commits at cs.uiuc.edu
>> Sent: Thursday, September 25, 2014 6:58:24 PM
>> Subject: Re: [llvm] r217144 - Enable noalias metadata by default and swap the order of the SLP and Loop vectorizers
>> by default.
>>
>> Hi James,
>>
>> This patch is causing a bunch of failures in LNT test suite, when
>> compiling with `-O3 -fslp-vectorize-aggressive`. Clang fails when
>> trying to do the CSE after BBVectorize.
>>
>> -Yi
>>
>> New Failures - Compile Time
>> MultiSource/Applications/ClamAV/clamscan
>> MultiSource/Applications/JM/ldecod/ldecod
>> MultiSource/Applications/JM/lencod/lencod
>> MultiSource/Applications/oggenc/oggenc
>> MultiSource/Benchmarks/7zip/7zip-benchmark
>> MultiSource/Benchmarks/FreeBench/neural/neural
>> MultiSource/Benchmarks/MallocBench/espresso/espresso
>> MultiSource/Benchmarks/MiBench/consumer-lame/consumer-lame
>> SingleSource/Benchmarks/CoyoteBench/lpbench
>>
>> Error message:
>> $ clang lpbench.c --target=aarch64-linux-gnu -O3
>> -fslp-vectorize-aggressive
>> Instruction does not dominate all uses!
>>   %cmp.zero241 = icmp eq i64 %n.vec239, 0
>>   br i1 %cmp.zero241, label %middle.block, label %vector.memcheck
>> Instruction does not dominate all uses!
>>   %n.vec239 = and i64 %0, -4
>>   %37 = icmp eq i64 %index.next, %n.vec239
>> Instruction does not dominate all uses!
>>   %n.vec239 = and i64 %0, -4
>>   %resume.val = phi i64 [ 0, %overflow.checked ], [ 0,
>> %vector.memcheck ], [ %n.vec239, %vector.body ]
>> Instruction does not dominate all uses!
>>   %cmp.zero241 = icmp eq i64 %n.vec239, 0
>>   br i1 %cmp.zero241, label %middle.block184, label
>>   %vector.memcheck205
>> Instruction does not dominate all uses!
>>   %n.vec239 = and i64 %0, -4
>>   %56 = icmp eq i64 %index.next211, %n.vec239
>> Instruction does not dominate all uses!
>>   %n.vec239 = and i64 %0, -4
>>   %resume.val206 = phi i64 [ 0, %overflow.checked192 ], [ 0,
>> %vector.memcheck205 ], [ %n.vec239, %vector.body183 ]
>> fatal error: error in backend: Broken function found, compilation
>> aborted!
>> clang: error: clang frontend command failed with exit code 70 (use -v
>> to see invocation)
>> clang version 3.6.0 (trunk 217400)
>> Target: aarch64--linux-gnu
>> Thread model: posix
>> clang: note: diagnostic msg: PLEASE submit a bug report to
>> http://llvm.org/bugs/ and include the crash backtrace, preprocessed
>> source, and associated run script.
>> clang: note: diagnostic msg:
>> ********************
>>
>> PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
>> Preprocessed source(s) and associated run script(s) are located at:
>> clang: note: diagnostic msg: /tmp/lpbench-8701bf.c
>> clang: note: diagnostic msg: /tmp/lpbench-8701bf.sh
>> clang: note: diagnostic msg:
>>
>> ********************
>>
>> On 4 September 2014 14:23, James Molloy <james.molloy at arm.com> wrote:
>> > Author: jamesm
>> > Date: Thu Sep  4 08:23:08 2014
>> > New Revision: 217144
>> >
>> > URL: http://llvm.org/viewvc/llvm-project?rev=217144&view=rev
>> > Log:
>> > Enable noalias metadata by default and swap the order of the SLP
>> > and Loop vectorizers by default.
>> >
>> > After some time maturing, hopefully the flags themselves will be
>> > removed.
>> >
>> > Modified:
>> >     llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp
>> >     llvm/trunk/lib/Transforms/Utils/InlineFunction.cpp
>> >
>> > Modified: llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp?rev=217144&r1=217143&r2=217144&view=diff
>> > ==============================================================================
>> > --- llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp (original)
>> > +++ llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp Thu Sep  4
>> > 08:23:08 2014
>> > @@ -62,7 +62,7 @@ static cl::opt<bool> RunLoadCombine("com
>> >
>> >  static cl::opt<bool>
>> >  RunSLPAfterLoopVectorization("run-slp-after-loop-vectorization",
>> > -  cl::init(false), cl::Hidden,
>> > +  cl::init(true), cl::Hidden,
>> >    cl::desc("Run the SLP vectorizer (and BB vectorizer) after the
>> >    Loop "
>> >             "vectorizer instead of before"));
>> >
>> >
>> > Modified: llvm/trunk/lib/Transforms/Utils/InlineFunction.cpp
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/InlineFunction.cpp?rev=217144&r1=217143&r2=217144&view=diff
>> > ==============================================================================
>> > --- llvm/trunk/lib/Transforms/Utils/InlineFunction.cpp (original)
>> > +++ llvm/trunk/lib/Transforms/Utils/InlineFunction.cpp Thu Sep  4
>> > 08:23:08 2014
>> > @@ -42,7 +42,7 @@
>> >  using namespace llvm;
>> >
>> >  static cl::opt<bool>
>> > -EnableNoAliasConversion("enable-noalias-to-md-conversion",
>> > cl::init(false),
>> > +EnableNoAliasConversion("enable-noalias-to-md-conversion",
>> > cl::init(true),
>> >    cl::Hidden,
>> >    cl::desc("Convert noalias attributes to metadata during
>> >    inlining."));
>> >
>> >
>> >
>> > _______________________________________________
>> > llvm-commits mailing list
>> > llvm-commits at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>
>
> --
> Hal Finkel
> Assistant Computational Scientist
> Leadership Computing Facility
> Argonne National Laboratory