Relax assertion in StoreBuffer::FindPointersToNewSpaceInMapsRegion

Relax assertion in StoreBuffer::FindPointersToNewSpaceInMapsRegion

Since r21232, maps are now a non-power-of-two size and thus don't fit
evenly into a page.  No test case, as it is somewhat difficult to get
this condition to trigger.  It is possible with this case, however:

    var result;

    (function() {
        for (var i = 0; i < 1e7; i++) {
            result = (function*() { yield 1; })();
            result.foo = 1;
        }
    })();

R=hpayer@chromium.org
BUG=

Committed: https://code.google.com/p/v8/source/detail?r=21744

[Hannes Payer (out of office), 2014-05-20 05:49:49]

Please also add the regression test.

https://codereview.chromium.org/291913002/diff/1/src/store-buffer.cc
File src/store-buffer.cc (left):

https://codereview.chromium.org/291913002/diff/1/src/store-buffer.cc#oldcode451
src/store-buffer.cc:451: ASSERT(map_aligned_end == end);
can we change the assert to ASSERT(map_aligned_start <= map_aligned_end &&
map_aligned_end <= end);

[Jakob Kummerow, 2014-05-20 07:30:23]

On 2014/05/20 05:49:49, Hannes Payer wrote:
> Please also add the regression test.

Please don't.
That test runs for 34 seconds (ia32.release) / 38s (x64.release) on my machine,
which is way too long to be acceptable. If we can't come up with a regression
test that finishes much faster (max 500ms in Release, 1ms would be better), this
fix will have to go in without a regression test.

[Michael Starzinger, 2014-05-20 08:03:59]

+1 on Jakob's comment. I don't think regression tests that rely on (long)
running loopw scale.

[Hannes Payer (out of office), 2014-05-20 08:09:25]

On 2014/05/20 08:03:59, Michael Starzinger wrote:
> +1 on Jakob's comment. I don't think regression tests that rely on (long)
> running loopw scale.

Well, the test did not lead to any interesting insights, skip it.

[wingo, 2014-06-10 12:28:11]

On 2014/05/20 05:49:49, Hannes Payer wrote:
>
https://codereview.chromium.org/291913002/diff/1/src/store-buffer.cc#oldcode451
> src/store-buffer.cc:451: ASSERT(map_aligned_end == end);
> can we change the assert to ASSERT(map_aligned_start <= map_aligned_end &&
> map_aligned_end <= end);

Done, ptal.  I have verified that the given test case still fails on master and
passes on the branch.

[Hannes Payer (out of office), 2014-06-10 12:37:17]

LGTM

[Hannes Payer (out of office), 2014-06-10 12:49:21]

Committed patchset #2 manually as r21744 (presubmit successful).