Eliminates more large immediates from x64 code.

Eliminates more large immediates from x64 code.

R=fschneider@google.com

Committed: https://code.google.com/p/dart/source/detail?r=28067

[zra, 2013-09-27 23:40:55]

Starting out with immediates generated for the IL and the flow graph, as much as
possible. Stubs and intrinsics are a bit more tricky since a pool-pointer is not
currently always loaded.

[Florian Schneider, 2013-09-30 11:48:35]

Can you explain why is PP not valid in the intrinsic code?

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64.cc
File runtime/vm/assembler_x64.cc (right):

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2455: // It's a 64-bit constant and we're not in the
VM isolate, so load form
/s/form/from/

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2476: LoadImmediate(TMP, imm, pp);
This does not emit the most compact code in the case of

(pp == kNoRegister) || (Isolate::Current() == Dart::vm_isolate())

where it could be 

just movq(dst, imm) instead of

movq(TMP, imm);
movq(dst, TMP);

[zra, 2013-09-30 18:01:08]

All the intrinsics have some PP loaded, and can therefore access Null, True and
False objects even if they don't explicitly load a pool pointer themselves.
Without explicitly loading their own pool-pointer, though, the intrinsics can't
add new things to the object pool. I want to see what the performance impact of
doing that is in isolation, so I'd like to leave it for a forthcoming CL.

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64.cc
File runtime/vm/assembler_x64.cc (right):

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2455: // It's a 64-bit constant and we're not in the
VM isolate, so load form
On 2013/09/30 11:48:35, Florian Schneider wrote:
> /s/form/from/

Done.

https://codereview.chromium.org/25118002/diff/22001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2476: LoadImmediate(TMP, imm, pp);
On 2013/09/30 11:48:35, Florian Schneider wrote:
> This does not emit the most compact code in the case of
> 
> (pp == kNoRegister) || (Isolate::Current() == Dart::vm_isolate())
> 
> where it could be 
> 
> just movq(dst, imm) instead of
> 
> movq(TMP, imm);
> movq(dst, TMP);

Done.

[Florian Schneider, 2013-09-30 18:43:35]

LGTM.

https://codereview.chromium.org/25118002/diff/32001/runtime/vm/assembler_x64.cc
File runtime/vm/assembler_x64.cc (right):

https://codereview.chromium.org/25118002/diff/32001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2475: (Isolate::Current() == Dart::vm_isolate())) {
These 3 conditions seem to be repeated a couple of times. Would it be worth to
encapsulate them in a helper function?

This would make it easy to use the same test for all the functions like
AddImmediate, CompareImmediate, etc.

[zra, 2013-09-30 19:59:30]

Thanks!

https://codereview.chromium.org/25118002/diff/32001/runtime/vm/assembler_x64.cc
File runtime/vm/assembler_x64.cc (right):

https://codereview.chromium.org/25118002/diff/32001/runtime/vm/assembler_x64....
runtime/vm/assembler_x64.cc:2475: (Isolate::Current() == Dart::vm_isolate())) {
On 2013/09/30 18:43:35, Florian Schneider wrote:
> These 3 conditions seem to be repeated a couple of times. Would it be worth to
> encapsulate them in a helper function?
> 
> This would make it easy to use the same test for all the functions like
> AddImmediate, CompareImmediate, etc.

Done.

[zra, 2013-09-30 20:19:29]

Committed patchset #6 manually as r28067 (presubmit successful).