Implements intrinsics for ARM.

Implements intrinsics for ARM.

R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=23311

[regis, 2013-05-28 22:01:44]

LGTM

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/assembler_arm_...
File runtime/vm/assembler_arm_test.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/assembler_arm_...
runtime/vm/assembler_arm_test.cc:235: __ eor(R0, R0, ShifterOperand(R0), VC);
mov(R0, ShifterOperand(0), VC) may be clearer than eor.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/flow_graph_com...
File runtime/vm/flow_graph_compiler_arm.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/flow_graph_com...
runtime/vm/flow_graph_compiler_arm.cc:1519: ASSERT(Utils::IsAbsoluteUint(12,
disp));
This assert will fire sooner or later.
I do not think we should implement this ElementAddressForIntIndex function on
ARM, because the range is too limited. It should be marked as UNREACHABLE like
ElementAddressForRegIndex below.
It is better to inline this function as needed (as is
ElementAddressForRegIndex).

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
File runtime/vm/intrinsifier_arm.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:160: __ mov(PC, ShifterOperand(LR), CC);
The use of mov to pc to return is deprecated. You should use bx(LR, CC).
We may still use mov in some assembler tests. We should change these at some
time.

There are 16 more occurrences to fix in this file.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:698: __ sdiv(TMP, left, right);  // TMP <- left /
right
Shouldn't you verify in the caller that integer division is supported by the
processor?

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:860: __ cmp(R1, ShifterOperand(R6));
You can make this sequence shorter as I did in my last cl:
  __ mov(IP, ShifterOperand(R1, LSL, R0));
  __ cmp(R1, ShifterOperand(IP, ASR, R0));

[zra, 2013-05-28 22:49:31]

Thanks!

I also changed mov(PC, LR) to bx(LR) in the assembler tests.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/assembler_arm_...
File runtime/vm/assembler_arm_test.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/assembler_arm_...
runtime/vm/assembler_arm_test.cc:235: __ eor(R0, R0, ShifterOperand(R0), VC);
On 2013/05/28 22:01:44, regis wrote:
> mov(R0, ShifterOperand(0), VC) may be clearer than eor.

Done.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/flow_graph_com...
File runtime/vm/flow_graph_compiler_arm.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/flow_graph_com...
runtime/vm/flow_graph_compiler_arm.cc:1519: ASSERT(Utils::IsAbsoluteUint(12,
disp));
On 2013/05/28 22:01:44, regis wrote:
> This assert will fire sooner or later.
> I do not think we should implement this ElementAddressForIntIndex function on
> ARM, because the range is too limited. It should be marked as UNREACHABLE like
> ElementAddressForRegIndex below.
> It is better to inline this function as needed (as is
> ElementAddressForRegIndex).  

Done.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
File runtime/vm/intrinsifier_arm.cc (right):

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:160: __ mov(PC, ShifterOperand(LR), CC);
On 2013/05/28 22:01:44, regis wrote:
> The use of mov to pc to return is deprecated. You should use bx(LR, CC).
> We may still use mov in some assembler tests. We should change these at some
> time.
> 
> There are 16 more occurrences to fix in this file.

Done.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:698: __ sdiv(TMP, left, right);  // TMP <- left /
right
On 2013/05/28 22:01:44, regis wrote:
> Shouldn't you verify in the caller that integer division is supported by the
> processor?

Yes. Done.

https://codereview.chromium.org/15822008/diff/16017/runtime/vm/intrinsifier_a...
runtime/vm/intrinsifier_arm.cc:860: __ cmp(R1, ShifterOperand(R6));
On 2013/05/28 22:01:44, regis wrote:
> You can make this sequence shorter as I did in my last cl:
>   __ mov(IP, ShifterOperand(R1, LSL, R0));
>   __ cmp(R1, ShifterOperand(IP, ASR, R0));

Done.

[zra, 2013-05-28 22:55:56]

Committed patchset #5 manually as r23311 (presubmit successful).