Subject: optimizer bug?
To: None <email@example.com>
From: Danny Thomas <D.Thomas@vthrc.uq.edu.au>
Date: 04/04/1999 07:46:27
openSSL builds OK under NetBSD-i386K but fails several tests.
The compiler flags are:
-DTERMIOS -D_ANSI_SOURCE -O3 -fomit-frame-pointer -m486 -Wall
changing optimization to -O and rebuilding sees it pass all tests. Using
-O2 also fails. Are their known bugs with >O on egcs ? I don't think cc's
been changed since 1.3K, so it's unlikely to be updated for 1.4.
Could be related to PR7066 and it might be worthwhile for people to try
this on platforms other than x86. It doesn't surprise me that crypto
packages are usful in ferreting out some classes of compiler bugs - if only
because the compiled code is run through a battery of tests.
Although changing optimizer levels is an obvious thing to try when tests
fail, maybe their configure should check for egcs-1.1.1 and warn
appropriately (also openssl's quick install web-page skips the make test).
Maybe the OpenSSL distribution should get the O flags downgraded to what
works, or at least some warning in the docs.
FWIW I'm using an AMD-K6