MIPS: port Implement ICs for FastDoubleArray loads and stores

MIPS: port Implement ICs for FastDoubleArray loads and stores

Ported r8637 (72bd791)

BUG=
TEST=

Committed: http://code.google.com/p/v8/source/detail?r=8686

[danno, 2011-07-18 14:36:12]

LGTM with a comment nit. If you upload an updated version I'll commit it.

http://codereview.chromium.org/7386003/diff/1003/src/mips/stub-cache-mips.cc
File src/mips/stub-cache-mips.cc (right):

http://codereview.chromium.org/7386003/diff/1003/src/mips/stub-cache-mips.cc#...
src/mips/stub-cache-mips.cc:4372: //  -- a3    : scratch
This comment is out of date with the renames of the register names below. It's
also wrong in ARM, I'll fix it.

[Paul Lind, 2011-07-20 04:43:54]

http://codereview.chromium.org/7386003/diff/1003/src/mips/stub-cache-mips.cc
File src/mips/stub-cache-mips.cc (right):

http://codereview.chromium.org/7386003/diff/1003/src/mips/stub-cache-mips.cc#...
src/mips/stub-cache-mips.cc:4372: //  -- a3    : scratch
On 2011/07/18 14:36:12, danno wrote:
> This comment is out of date with the renames of the register names below. It's
> also wrong in ARM, I'll fix it.

I updated the comment for the named scratch regs. However, this opens up
potential for more register "state" comment updates throughout this file. 

This function uses a couple FPU registers as scratch, which are not commented
(as arm uses uncommented VFP scratch regs). 

Many other functions declare input regs but do not declare scratch regs, whether
named or un-named. Examples: GenerateLoadFastElement(),
GenerateLoadFastDoubleElement(), GenerateStoreExternalArray(), many others.

It is probably a good idea to comment these, but I've not done so here. We will
follow your lead on this.

[danno, 2011-07-20 07:08:55]

Thanks, I'll land this.

I personally think that the register comments at the beginning are only
marginally usefully if you declare useful named register variables, and in that
case the comments are only useful to document the in/out calling convention. I
am wary of comments that duplicate the obvious and just create an opportunity
for something to get out of date. 

It certainly would be nice and perhaps a bit more maintainable if we generally
used descriptive register aliases everywhere, but we're a long way from that.
Besides, I'm as guilty as the next guy of not being 100% consistent about this
in the past, so I'm not going to get pedantic about it.