Tweak default max heap size constants for platforms with swap.

Tweak default max heap size constants for platforms with swap.

Configure platforms with swap memory with larger heap size by default. Also
introduce an additional "bucket" between 1GB-2GBs.

BUG=None
R=hpayer@chromium.org, jkummerow@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=17081

[rmcilroy, 2013-09-27 13:30:15]

As discussed with Jakob, this enables platforms with swap to be "one bucket up"
in max heap sizes.  It also introduces a 1-2GB bucket since previously the step
between 512MB and 2GB sizes was quite large.  PTAL.

[Hannes Payer (out of office), 2013-09-27 13:47:50]

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc
File src/defaults.cc (right):

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc#newcode46
src/defaults.cc:46: // be a multiple of Page::kPageSize.
You pump up the old space size for all platforms. This will change performance.
Please check performance, especially gc latency numbers.

https://codereview.chromium.org/24978006/diff/1/src/globals.h
File src/globals.h (right):

https://codereview.chromium.org/24978006/diff/1/src/globals.h#newcode286
src/globals.h:286: #if V8_OS_ANDROID
Would it make sense to add an OS call which checks dynamically for swap space?

[Jakob Kummerow, 2013-09-27 13:52:12]

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc
File src/defaults.cc (right):

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc#newcode50
src/defaults.cc:50: constraints->set_max_old_space_size(768 * lump_of_memory);
Let's leave this at 700. There's no benefit to having a power-of-2-ish limit for
old space.

https://codereview.chromium.org/24978006/diff/1/src/globals.h
File src/globals.h (right):

https://codereview.chromium.org/24978006/diff/1/src/globals.h#newcode287
src/globals.h:287: const bool kOsHasSwap = false;
Please move this definition to defaults.cc.

[rmcilroy, 2013-10-01 13:34:17]

I've re-structured the if/else branch to be more in line with my original
intention (as Jakob said, there the "> 2GB" check I was doing was effectively a
">= 4GB" check since there is not many systems with memory between 2-4GB.  PTAL.

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc
File src/defaults.cc (right):

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc#newcode46
src/defaults.cc:46: // be a multiple of Page::kPageSize.
On 2013/09/27 13:47:50, Hannes Payer wrote:
> You pump up the old space size for all platforms. This will change
performance.
> Please check performance, especially gc latency numbers.

Done.

https://codereview.chromium.org/24978006/diff/1/src/defaults.cc#newcode50
src/defaults.cc:50: constraints->set_max_old_space_size(768 * lump_of_memory);
On 2013/09/27 13:52:12, Jakob wrote:
> Let's leave this at 700. There's no benefit to having a power-of-2-ish limit
for
> old space.

Done.

https://codereview.chromium.org/24978006/diff/1/src/globals.h
File src/globals.h (right):

https://codereview.chromium.org/24978006/diff/1/src/globals.h#newcode286
src/globals.h:286: #if V8_OS_ANDROID
On 2013/09/27 13:47:50, Hannes Payer wrote:
> Would it make sense to add an OS call which checks dynamically for swap space?

I think we should avoid adding to the configuration matrix too much, so would
prefer this static check to a dynamic one, but if there are situations you can
think of where systems disable swap on platforms which provide it then I will
reconsider?

https://codereview.chromium.org/24978006/diff/1/src/globals.h#newcode287
src/globals.h:287: const bool kOsHasSwap = false;
On 2013/09/27 13:52:12, Jakob wrote:
> Please move this definition to defaults.cc.

Done.

[Jakob Kummerow, 2013-10-01 14:02:12]

lgtm

[rmcilroy, 2013-10-02 12:44:44]

On 2013/10/01 14:02:12, Jakob wrote:
> lgtm

Hannes, do you have any comments?

[Hannes Payer (out of office), 2013-10-02 12:59:17]

LGTM, with comments.

https://codereview.chromium.org/24978006/diff/1/src/globals.h
File src/globals.h (right):

https://codereview.chromium.org/24978006/diff/1/src/globals.h#newcode286
src/globals.h:286: #if V8_OS_ANDROID
On 2013/10/01 13:34:17, rmcilroy wrote:
> On 2013/09/27 13:47:50, Hannes Payer wrote:
> > Would it make sense to add an OS call which checks dynamically for swap
space?
> 
> I think we should avoid adding to the configuration matrix too much, so would
> prefer this static check to a dynamic one, but if there are situations you can
> think of where systems disable swap on platforms which provide it then I will
> reconsider?

OK, fine. In general I prefer probing the system for these properties over
static checks but it is not necessary to do that now.

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc
File src/defaults.cc (right):

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc#newcode55
src/defaults.cc:55: if (physical_memory <= 512ul * i::MB) {
This change will increase the maximum used heap size on low memory devices. I
hope this is intended.

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc#newcode61
src/defaults.cc:61: constraints->set_max_old_space_size(256 * lump_of_memory);
Same here.

[rmcilroy, 2013-10-02 13:05:56]

Could someone land this when convenient please.

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc
File src/defaults.cc (right):

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc#newcode55
src/defaults.cc:55: if (physical_memory <= 512ul * i::MB) {
On 2013/10/02 12:59:17, Hannes Payer wrote:
> This change will increase the maximum used heap size on low memory devices. I
> hope this is intended.

Yes, this is intended.

https://codereview.chromium.org/24978006/diff/15001/src/defaults.cc#newcode61
src/defaults.cc:61: constraints->set_max_old_space_size(256 * lump_of_memory);
On 2013/10/02 12:59:17, Hannes Payer wrote:
> Same here.

Also intended.  Thanks for checking!

[Hannes Payer (out of office), 2013-10-02 13:38:14]

Committed patchset #4 manually as r17081 (presubmit successful).

[Hannes Payer (out of office), 2013-10-02 13:38:20]

OK, I will do that.