Refactor bigint _sqrTo in preparation of intrinsification.

Refactor bigint _sqrTo in preparation of intrinsification.
Provide _sqrAdd intrinsic on ia32.

Committed: https://code.google.com/p/dart/source/detail?r=40623

[regis, 2014-09-23 19:26:11]

regis@google.com changed reviewers:
+ srdjan@google.com, vegorov@google.com

[regis, 2014-09-23 19:26:11]

Not for review, just for your information.

This cl causes a huge drop of performance on IA32 and therefore cannot be
committed without the corresponding intrinsic (_sqrAdd).

The drop could be due to the register allocation bug investigated by Slava. Not
clear.

[regis, 2014-09-23 23:38:51]

You can review now. The cl includes the intrinsic on ia32.

[srdjan, 2014-09-24 00:19:29]

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
File runtime/vm/intrinsifier_ia32.cc (right):

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
runtime/vm/intrinsifier_ia32.cc:969: 
You could define:
Address n_addr = Address(ESP, 2 * kWordSize)

and use it on various places below

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
runtime/vm/intrinsifier_ia32.cc:1055: __ j(NOT_EQUAL, &x_not_zero,
Assembler::kNearJump);
You could jump if equal 0 to a label just before the last ret in the code.

[regis, 2014-09-24 01:50:49]

Thanks!
I'll submit since you gave a LGTM via chat.

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
File runtime/vm/intrinsifier_ia32.cc (right):

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
runtime/vm/intrinsifier_ia32.cc:969: 
On 2014/09/24 00:19:28, srdjan wrote:
> You could define:
> Address n_addr = Address(ESP, 2 * kWordSize)
> 
> and use it on various places below

Done.
And it may not be such a good idea to untag n above the stack pointer (not
sure). I pushed an untagged copy as a local.

https://codereview.chromium.org/600533002/diff/20001/runtime/vm/intrinsifier_...
runtime/vm/intrinsifier_ia32.cc:1055: __ j(NOT_EQUAL, &x_not_zero,
Assembler::kNearJump);
On 2014/09/24 00:19:28, srdjan wrote:
> You could jump if equal 0 to a label just before the last ret in the code.

Done.

[regis, 2014-09-24 01:51:10]

Committed patchset #4 (id:60001) manually as r40623 (presubmit successful).

[srdjan, 2014-09-24 02:31:05]

lgtm