Ping? Any action before the next 0.9.7/0.9.8 releases? As I've mentioned,
there is a long and sordid history of bugs/assumptions in Solaris -xdepend
optimization, so the trivial solution is to simply apply the fix and be
done with it for Solaris cc. Getting a 'fix' from Sun, or finding the
lines offensive to the optimizer, seems like an exercise in futility...
>>>>> The fix is trivial; in the solaris-x86-cc line of Configure, you must
>>>>> replace -fast -xO5 with -fast -xdepend=no (note: -xO5 was implied by
>>>>> the -fast argument.) The implicit -xdepend=yes implied by -fast was
>>>>> causing this failure.
We now have confirmation that this fixes problems seen in solaris 9 and 10,
multiple cc versions, both x86 and sparc hardware. Please apply :)
Bill
William A. Rowe, Jr. wrote:
Kyle Hamilton wrote:
Have you filed a bug with Sun about this issue?
No, because the specific loop that failed hasn't been identified, so it's
difficult to give them a precise reproduction case.
Moreso, we need to patch this given the large number of Solaris cc
compilers
deployed. Loop-data optimizations are a historical sore point with solaris
compilers, appearing to have many artificats from overly agressive
optimization.
Research "-xdepend" bugs in the solaris CC release notes, any flavor
from the
SunStudio 8 through 11.
Bill
On 2/27/06, William A. Rowe, Jr. <[EMAIL PROTECTED]> wrote:
Glad this licked it [yes I was that guest, forgot to sign my note,
sorry.]
One thing about -xdepend is that Solaris cc 5.x is unrolling one of our
loops incorrectly. If someone wants to simplify, try reducing the
complex
for (;;) and while () loops to avoid (;;x++, y++) or (;x[c++] +=x;)
sorts
of grammers which have apparently confounded the optimizer.
Bill
[EMAIL PROTECTED] via RT wrote:
via RT schrieb:
The fix is trivial; in the solaris-x86-cc line of Configure, you must
replace -fast -xO5 with -fast -xdepend=no (note: -xO5 was implied by
the -fast argument.) The implicit -xdepend=yes implied by -fast was
causing this failure.
There may be further code fixes in the evp code to ensure that data
dependencies in the loops can be unrolled and optimized.
excellent!
now it works fine
thanks a lot !
______________________________________________________________________
OpenSSL Project http://www.openssl.org
Development Mailing List openssl-dev@openssl.org
Automated List Manager [EMAIL PROTECTED]
