Unbox/Box Float64x2 and inline typed array loads and stores

Unbox/Box Float64x2 and inline typed array loads and stores

R=fschneider@google.com

Committed: https://code.google.com/p/dart/source/detail?r=32869

[Cutch, 2014-02-19 18:00:02]

+zra for the curlies and the ARM/MIPS changes.

[Florian Schneider, 2014-02-20 15:54:43]

Lgtm.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/flow_graph_optimi...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/flow_graph_optimi...
runtime/vm/flow_graph_optimizer.cc:1303: }
I think reformatting like this make the code less readable.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/flow_graph_optimi...
runtime/vm/flow_graph_optimizer.cc:1334: if (!ShouldInlineSimd()) {
Why? It does not improve readability in my view.

Quote from the C++ style guide: "Short conditional statements may be written on
one line if this enhances readability."

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_arm.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:992: __ LoadDFromOffset(dresult0,
index.reg(), 0);
Isn't there a Load-quad-word instruction on ARM? Could you maybe use vldmd
load-multiple instruction?

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3093: __ StoreDToOffset(value_even,
out_reg,
Use one store-multiple instead of two store instruction?

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3096: Float32x4::value_offset() +
2*kWordSize - kHeapObjectTag);
spaces around *

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3132: __ LoadDFromOffset(result_even,
value,
Load-multiple (vldmd) instead of 2 vldr? This would apply to the correctponding
Int32x4 and Float32x4 instructions as well.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_ia32.cc:955: if (class_id_ ==
kTypedDataFloat64x2ArrayCid) {
Left-over debug printing?

[Cutch, 2014-02-20 19:33:11]

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_arm.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:992: __ LoadDFromOffset(dresult0,
index.reg(), 0);
On 2014/02/20 15:54:44, Florian Schneider wrote:
> Isn't there a Load-quad-word instruction on ARM? Could you maybe use vldmd
> load-multiple instruction?

zra says there is but it can only be used if there is no offset, I think we can
use it some places but not others. I've added a TODO to look at this.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3096: Float32x4::value_offset() +
2*kWordSize - kHeapObjectTag);
On 2014/02/20 15:54:44, Florian Schneider wrote:
> spaces around *

Done here and elsewhere.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3132: __ LoadDFromOffset(result_even,
value,
On 2014/02/20 15:54:44, Florian Schneider wrote:
> Load-multiple (vldmd) instead of 2 vldr? This would apply to the
correctponding
> Int32x4 and Float32x4 instructions as well.

vldmd can't be used with an offset.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_ia32.cc:955: if (class_id_ ==
kTypedDataFloat64x2ArrayCid) {
On 2014/02/20 15:54:44, Florian Schneider wrote:
> Left-over debug printing?

Done.

[Cutch, 2014-02-20 20:42:11]

Committed patchset #3 manually as r32869 (presubmit successful).

[zra, 2014-02-20 22:46:40]

Test fails on ARM hardware.

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_arm.cc (right):

https://codereview.chromium.org/172653002/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm.cc:3094: Float32x4::value_offset() -
kHeapObjectTag);
Is this supposed to be Float64x2::value_offset()?