Huge John The Ripper regression with Clang 10 trunk

[davidbolvansky]

Bugzilla Link	44410
Resolution	WORKSFORME
Resolved on	Feb 25, 2020 06:56
Version	unspecified
OS	Linux
Blocks	#43756
CC	@alexey-bataev,@dwblaikie,@fhahn,@zmodem,@hfinkel,@jdoerfert,@RKSimon

Extended Description

See:
https://www.phoronix.com/scan.php?page=article&item=gcc-clang-3960x&num=3

[alexey-bataev]

See:
https://www.phoronix.com/scan.php?page=article&item=gcc-clang-3960x&num=3

I saw this report already. Hard to say what is the cause of this. We did not change the functionality of the OpenMP significantly, mostly the new features were implemented. Will try to investigate it later.

[alexey-bataev]

Compared the resulst of John the Ripper for trunk and 9.0. No significant difference. Built at O3 with OpenMP support, powerpc64le-unknown-linux. Seems to me, internal problem of the testsuite. Seems like ./configure script was unable to detect support for OpenMP and built the code wihtout OpenMP support or something like that. Already saw some incorrectly reported results from phoronix for OpenMP in clang a few years ago.

[zmodem]

So, are we ready to close this as wontfix?

[alexey-bataev]

So, are we ready to close this as wontfix?

I think so, if there are no other opinions

[davidbolvansky]

Yeah, not a clang/llvm bug

[alexey-bataev]

Yeah, not a clang/llvm bug

Then closing as not a bug

[davidbolvansky]

mentioned in issue #43756