llvm dev - Jun 2018 - Determine reason for failure at -O1

Hi Everyone,

We caught a report for a failed self test when using Clang 5.0 and 6.0
with -DDEBUG and -O1 (i.e., a "debug build"). The code is question is
located at https://github.com/weidai11/cryptopp/blob/master/cham-simd.cpp
. It is the SSSE3 code path for CHAM64.

Other optimizations levels are OK for Clang. GCC, ICC and MSVC are OK.
The code is valgrind, Sanitizer, Coverity and Enterprise Analysis
clean.

An objdump is available at https://pastebin.com/TAKH4eur but it is not
very helpful to me.

My question is, how can we begin to troubleshoot the issue?

Thanks in advance,

Jeff

I'd suggest approach this like any other bug in source code (not in the
compiler) - reduce the example, isolate the failure - until you either find
a source bug, or a small, standalone example that seems to demonstrate a
contradiction or bug in the way the compiler is interpreting the source.

On Sat, Jun 30, 2018 at 12:49 AM Jeffrey Walton via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Hi Everyone,
>
> We caught a report for a failed self test when using Clang 5.0 and 6.0
> with -DDEBUG and -O1 (i.e., a "debug build"). The code is
question is
> located at https://github.com/weidai11/cryptopp/blob/master/cham-simd.cpp
> . It is the SSSE3 code path for CHAM64.
>
> Other optimizations levels are OK for Clang. GCC, ICC and MSVC are OK.
> The code is valgrind, Sanitizer, Coverity and Enterprise Analysis
> clean.
>
> An objdump is available at https://pastebin.com/TAKH4eur but it is not
> very helpful to me.
>
> My question is, how can we begin to troubleshoot the issue?
>
> Thanks in advance,
>
> Jeff
> _______________________________________________
> 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/20180630/d6834ee4/attachment.html>

On Sat, Jun 30, 2018 at 2:37 PM, David Blaikie <dblaikie at gmail.com>
wrote:
> I'd suggest approach this like any other bug in source code (not in the
> compiler) - reduce the example, isolate the failure - until you either find
> a source bug, or a small, standalone example that seems to demonstrate a
> contradiction or bug in the way the compiler is interpreting the source.
Thanks Davis.

According to https://stackoverflow.com/q/15548023/608639 the list of
optimizations applied at -O1 include:

-globalopt -demanded-bits -branch-prob -inferattrs -ipsccp -dse
-loop-simplify -scoped-noalias -barrier -adce -deadargelim -memdep
-licm -globals-aa -rpo-functionattrs -basiccg -loop-idiom -forceattrs
-mem2reg -simplifycfg -early-cse -instcombine -sccp -loop-unswitch
-loop-vectorize -tailcallelim -functionattrs -loop-accesses -memcpyopt
-loop-deletion -reassociate -strip-dead-prototypes -loops -basicaa
-correlated-propagation -lcssa -domtree -always-inline -aa -block-freq
-float2int -lower-expect -sroa -loop-unroll
-alignment-from-assumptions -lazy-value-info -prune-eh -jump-threading
-loop-rotate -indvars -bdce -scalar-evolution -tbaa
-assumption-cache-tracker

I tried backing off some of the optimizations to begin to isolate:

$ CXX=clang++ CXXFLAGS="-g2 -O1 -no-loop-vectorize" make cham-simd.o
clang++ -g2 -O1 -no-loop-vectorize -fPIC -pthread -pipe -c cham-simd.cpp
clang-6.0: error: unknown argument: '-no-loop-vectorize'
make: *** [GNUmakefile:1069: cham-simd.o] Error 1

So I guess my question is, how do I disable a particular optimization?

Jeff

On Sat, Jun 30, 2018 at 3:49 AM, Jeffrey Walton <noloader at gmail.com>
wrote:
> Hi Everyone,
>
> We caught a report for a failed self test when using Clang 5.0 and 6.0
> with -DDEBUG and -O1 (i.e., a "debug build"). The code is
question is
> located at https://github.com/weidai11/cryptopp/blob/master/cham-simd.cpp
> . It is the SSSE3 code path for CHAM64.
>
> ...
We cleared the issue at
https://github.com/weidai11/cryptopp/commit/aa80c7d4acb6 . We did not
have a methodology. We teased-out the solution through knob turning.

The aa80c7d4acb6 change modifies a loop to perform 4
encryption/decryption rounds per iteration rather than 8. 8 rounds per
iteration decryption was bad. Dropping from 8 to 4 relieved some
register pressure by requiring 4 fewer loads of a mask used to shuffle
a key.

Thanks for the help and suggestions.

Jeff

Were you able to determine if it was a compiler bug? (an isolated example
would help to make that clear - it's hard to know from the workaround
whether that suppressed a legitimate compiler optimization that was
tickling some bug in the code or working around a bug in the compiler)

On Sun, Jul 1, 2018 at 8:07 AM Jeffrey Walton via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> On Sat, Jun 30, 2018 at 3:49 AM, Jeffrey Walton <noloader at
gmail.com>
> wrote:
> > Hi Everyone,
> >
> > We caught a report for a failed self test when using Clang 5.0 and 6.0
> > with -DDEBUG and -O1 (i.e., a "debug build"). The code is
question is
> > located at
> https://github.com/weidai11/cryptopp/blob/master/cham-simd.cpp
> > . It is the SSSE3 code path for CHAM64.
> >
> > ...
>
> We cleared the issue at
> https://github.com/weidai11/cryptopp/commit/aa80c7d4acb6 . We did not
> have a methodology. We teased-out the solution through knob turning.
>
> The aa80c7d4acb6 change modifies a loop to perform 4
> encryption/decryption rounds per iteration rather than 8. 8 rounds per
> iteration decryption was bad. Dropping from 8 to 4 relieved some
> register pressure by requiring 4 fewer loads of a mask used to shuffle
> a key.
>
> Thanks for the help and suggestions.
>
> Jeff
> _______________________________________________
> 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/20180701/7ebe6d88/attachment.html>