Increases the number of ARM floating-point registers.

Increases the number of ARM floating-point registers.

This is to be consistent with VFPv3_D32. Later, we'll detect
VFPv3_D16 at runtime.

R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=25352

[zra, 2013-07-23 17:13:37]

Found the bug from yesterday. vstmd and vldrd can only handle up to 16 registers
at once. Additionally, not all Q registers are backed by S registers, so some
SIMD IL instructions need a Q register specified which is backed by S registers.
These are the Q5, Q6 you'll see there in the summaries.

[regis, 2013-07-23 17:38:04]

LGTM with comments

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc
File runtime/vm/assembler_arm.cc (right):

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:923: ASSERT(last - first <= 16);
ASSERT((last - first + 1) <= 16);

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:932: ASSERT(last - first <= 16);
ditto

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2286: if (lastv - firstv > 16) {
if ((lastv - firstv + 1) > 16) {

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2287: ASSERT(kNumberOfDRegisters <= 32);
This assert seems overcautious.

I wonder if we should change the parameters of vstmd and vldmd to take a
register and a count, instead of a first and last register, which complicates
the call sites.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2312: if (lastv - firstv > 16) {
ditto

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_arm.cc (right):

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:2950: summary->set_in(0,
Location::FpuRegisterLocation(Q5));
Please, add a comment about which one of the instructions below require this
particular register.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:3012:
summary->set_out(Location::FpuRegisterLocation(Q6));
ditto

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:3182:
summary->set_out(Location::FpuRegisterLocation(Q6));
ditto

[zra, 2013-07-23 18:24:14]

Thanks!

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc
File runtime/vm/assembler_arm.cc (right):

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:923: ASSERT(last - first <= 16);
On 2013/07/23 17:38:04, regis wrote:
> ASSERT((last - first + 1) <= 16);

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:932: ASSERT(last - first <= 16);
On 2013/07/23 17:38:04, regis wrote:
> ditto

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2286: if (lastv - firstv > 16) {
On 2013/07/23 17:38:04, regis wrote:
> if ((lastv - firstv + 1) > 16) {

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2287: ASSERT(kNumberOfDRegisters <= 32);
On 2013/07/23 17:38:04, regis wrote:
> This assert seems overcautious.
> 
> I wonder if we should change the parameters of vstmd and vldmd to take a
> register and a count, instead of a first and last register, which complicates
> the call sites.

I changed vstmd and vldrd to take a count instead of a second register, and
added an assert there.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/assembler_arm.cc#n...
runtime/vm/assembler_arm.cc:2312: if (lastv - firstv > 16) {
On 2013/07/23 17:38:04, regis wrote:
> ditto

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
File runtime/vm/intermediate_language_arm.cc (right):

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:2950: summary->set_in(0,
Location::FpuRegisterLocation(Q5));
On 2013/07/23 17:38:04, regis wrote:
> Please, add a comment about which one of the instructions below require this
> particular register.

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:3012:
summary->set_out(Location::FpuRegisterLocation(Q6));
On 2013/07/23 17:38:04, regis wrote:
> ditto

Done.

https://codereview.chromium.org/19806005/diff/1/runtime/vm/intermediate_langu...
runtime/vm/intermediate_language_arm.cc:3182:
summary->set_out(Location::FpuRegisterLocation(Q6));
On 2013/07/23 17:38:04, regis wrote:
> ditto

Done.

[zra, 2013-07-23 18:25:49]

Committed patchset #2 manually as r25352 (presubmit successful).