ARM: Reduce amount of code generated for context save on deopt....

ARM: Reduce amount of code generated for context save on deopt.

BUG=none
TEST=none

Committed: http://code.google.com/p/v8/source/detail?r=7873

[Mads Ager (chromium), 2011-05-06 08:15:13]

Søren or Karl, could you look at this?

[Søren Thygesen Gjesse, 2011-05-06 10:12:37]

Do you think this optimization of the deoptimization code pays off? The code is
sitting at the bottom of a large jump-table and only exists in three copies.

But of cause it does not hurt.

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc
File src/arm/deoptimizer-arm.cc (right):

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc#newc...
src/arm/deoptimizer-arm.cc:552: // Save all general purpose registers before
messing with them.
Please also change the comment - it was already wrong before.

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc#newc...
src/arm/deoptimizer-arm.cc:553: DwVfpRegister first =
DwVfpRegister::FromAllocationIndex(0);
We should have an assert here checking that all double registers with allocation
indexes from 0 to DwVfpRegister::kNumAllocatableRegisters are between first and
last and that no "non-allocatable" registers are in the range. The mapping from
allocation index to register can be changed to any mapping.

[m.m.capewell, 2011-05-10 10:40:06]

On 2011/05/06 10:12:37, Søren Gjesse wrote:
> Do you think this optimization of the deoptimization code pays off? The code
is
> sitting at the bottom of a large jump-table and only exists in three copies.

We saw rather a lot of code generated for some deoptimizations, so thought it
was worthwhile tidying this up, and using the new vstm support in the process. I
agree that the performance impact is minimal.

[m.m.capewell, 2011-05-10 10:40:25]

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc
File src/arm/deoptimizer-arm.cc (right):

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc#newc...
src/arm/deoptimizer-arm.cc:552: // Save all general purpose registers before
messing with them.
On 2011/05/06 10:12:37, Søren Gjesse wrote:
> Please also change the comment - it was already wrong before.

Done.

http://codereview.chromium.org/6930038/diff/1/src/arm/deoptimizer-arm.cc#newc...
src/arm/deoptimizer-arm.cc:553: DwVfpRegister first =
DwVfpRegister::FromAllocationIndex(0);
On 2011/05/06 10:12:37, Søren Gjesse wrote:
> We should have an assert here checking that all double registers with
allocation
> indexes from 0 to DwVfpRegister::kNumAllocatableRegisters are between first
and
> last and that no "non-allocatable" registers are in the range. The mapping
from
> allocation index to register can be changed to any mapping.

Done.