On 19/02/15 16:31 +0100, Hans-Peter Nilsson wrote:
Lately,
26_numerics/random/binomial_distribution/operators/values.cc has
started to FAIL on trunk with a timeout for my autotester for
cris-elf, a soft-float simulator target running a cgen-generated
simulator on a six-year-old x86_64-linux-gnu host. The reason
it's started to fail in the last few days are not clear, except
that they're probably local: I've eliminated code quality issues
and the test-case itself hasn't changed in about 1.5 years (a
few more details in the PR).
Anyway, the test-case is by appearance a few test-cases bundled
together, which seem to be just as well tested individually
considering the apparent intent of the test and considering that
the last non-comment change was to add the last two sub-tests.
(A split-up is also what's done in the past, in preference to
cutting down the test with ifdefs for simulator targets.) So, I
split out all-but-the-first subtest from values.cc.
For those interested, the individual runtimes (total M:S) for
the plain simulator runs on that machine, together with the
cycle counts (when using --cris-cycles=all) after the split-up are:
values: 0:51.131 4011197245
values2: 3:16.39 15456247210
values3: 1:11.76 5633161989
values4: 0:20.944 1626754775
values5: 3:24.28 16163068067
(I just wanted to make sure the longest run wasn't over half the
default time out, i.e. not over five minutes.)
Ok to commit?
Yes, OK - thanks.
