Adjust memory limits.

Adjust memory limits.

- Introduce new constant: kBootCodeSizeMultiplier to handle the code size differences across the platforms.

- Increase memory limits due to larger code size on MIPS.

BUG=
R=danno@chromium.org, plind44@gmail.com

Committed: https://code.google.com/p/v8/source/detail?r=20825

[palfia, 2014-04-10 22:13:29]

PTAL.

I've added a MIPS #ifdef around the new values in test-mark-compact.cc. Do you
like this approach or should I just modify the values for all arch's?

[Paul Lind, 2014-04-10 22:35:01]

LGTM, but we need review from v8 team. Please note Ákos question above.

[danno, 2014-04-11 09:32:14]

https://codereview.chromium.org/234153002/diff/1/src/spaces.cc
File src/spaces.cc (right):

https://codereview.chromium.org/234153002/diff/1/src/spaces.cc#newcode1088
src/spaces.cc:1088: size = size * 15 / 10;  // Increase by 15% due to larger
code size.
I think this should be based in kCodeSizeMultiplier, and in that case, perhaps
it can be made cross platform again? e.g.

size = 480 * KB * (kCodeSizeMultiplier / 100);

[palfia, 2014-04-11 13:58:14]

On 2014/04/11 09:32:14, danno wrote:
> https://codereview.chromium.org/234153002/diff/1/src/spaces.cc
> File src/spaces.cc (right):
> 
> https://codereview.chromium.org/234153002/diff/1/src/spaces.cc#newcode1088
> src/spaces.cc:1088: size = size * 15 / 10;  // Increase by 15% due to larger
> code size.
> I think this should be based in kCodeSizeMultiplier, and in that case, perhaps
> it can be made cross platform again? e.g.
> 
> size = 480 * KB * (kCodeSizeMultiplier / 100);

Thanks! This is a nice solution, but if we do that we will increase the original
code space size by 42%-62% on all arch's. This would be almost OK for MIPS and
it would give us the necessary space, but it would waste the memory on other
arch's. Maybe we should scale down the 480 first?

[palfia, 2014-04-16 14:07:13]

Hi Danno,

I've measured the boot-up memory usage on (almost) all platforms:

- ia32
2776KB - 100%

- x64
3780KB - 136%

- arm (beagleboard xm)
2856KB - 102%

- mipsel (sigma8911)
3136KB - 113%

It seems that the actual memory usage has no relation to the kCodeSizeMultiplier
constant (ia32: 100%, arm, mips: 142, x64: 162) and I don't want change the
current kCodeSizeMultiplier constant, since it is well characterized for the
current usage.
Therefore, we think that the best solution would be to introduce a new constant
that we can safely use for both the first page code siza and the bootup memory
limits.
I've uploaded a new patchset with the new constant, PTAL.
Note: The values are not well-tested on all platforms yet.

[danno, 2014-04-16 14:41:54]

OK, this is an acceptable compromise. LGTM.

[palfia, 2014-04-16 21:57:58]

Committed patchset #2 manually as r20825 (presubmit successful).