On 23.12.2013 16:24, H.J. Lu wrote:
On Sun, Dec 22, 2013 at 10:52 PM, Andrey Belevantsev <a...@ispras.ru> wrote:
Hello,
As described in the PR, the ICE reason was the typo made when introducing
calls to add_hard_reg_set. Fixed by the first attached patch, bootstrapped
and tested on both ia64 and x86_64, committed as obvious.
The test case is very sensitive to the scheduler decisions (e.g. it didn't
fail on trunk but only on the revision reported for me), so instead of
adding the test I have put in the code two asserts checking that we can
always schedule the fence instruction as is. This hunk was tested together
with the first but committed separately.
Testcase is very small. Why not add it?
Frankly, I think that the chances of this test uncovering similar issues in
the future are very small. It needs lots of options to make it trigger and
even with this a specific revision. The chance of triggering the asserts I
added on another code is much higher. In the past, I have also avoided to
add tests that require 5+ options to trigger the issue, adding only those
that have some hope on more ore less reliably reproducing the required
issue. The best solution of course is having an infrastructure to test the
specific scheduler decisions, which we don't have.
You are welcome to add the test if you feel so strongly about us needing it.
Andrey
