Support for SDIV and MLS ARM instructions, and implement DoModI using them

Support for SDIV and MLS ARM instructions, and implement DoModI using them.
Also added support for the runtime detection to check if hardware supports SDIV/UDIV
Other new opportunities to exploit SDIV/UDIV will be done in separate issues.

Committed: https://code.google.com/p/v8/source/detail?r=12646

[danno, 2012-09-27 15:28:09]

In general, SDIV support seems like a reasonable thing to add, but it needs to
be modeled like other optional processor instructions, like VFP3.

http://codereview.chromium.org/10977051/diff/1/build/common.gypi
File build/common.gypi (right):

http://codereview.chromium.org/10977051/diff/1/build/common.gypi#newcode49
build/common.gypi:49: 'v8_can_use_integer_division%': 'true',
I think this flag is mis-named. can_use_sdiv_instructions or something similar
is better, since flags expressing similar functionality (VFP) clearly reference
something that is ARM-specific.

also, I we can't land this in the current form as a compile-time flag, since
there is only one binary that gets distributed on ARM

http://codereview.chromium.org/10977051/diff/1/src/arm/disasm-arm.cc
File src/arm/disasm-arm.cc (right):

http://codereview.chromium.org/10977051/diff/1/src/arm/disasm-arm.cc#newcode985
src/arm/disasm-arm.cc:985: #if CAN_USE_INTEGER_DIVISION
Need to be a runtime check rather than a compile-time check, introduced in the
same way as VFP3 support works.

http://codereview.chromium.org/10977051/diff/1/src/arm/lithium-codegen-arm.cc
File src/arm/lithium-codegen-arm.cc (right):

http://codereview.chromium.org/10977051/diff/1/src/arm/lithium-codegen-arm.cc...
src/arm/lithium-codegen-arm.cc:982: #if CAN_USE_INTEGER_DIVISION
Need to be a runtime check rather than a compile-time check, introduced in the
same way as VFP3 support works.

http://codereview.chromium.org/10977051/diff/1/src/arm/lithium-codegen-arm.cc...
src/arm/lithium-codegen-arm.cc:1110: void
LCodeGen::EmitSignedIntegerDivisionByConstant(
Are there not cases in here that might also benefit from sdiv?

http://codereview.chromium.org/10977051/diff/1/src/arm/lithium-codegen-arm.cc...
src/arm/lithium-codegen-arm.cc:1288: void
LCodeGen::DoMathFloorOfDiv(LMathFloorOfDiv* instr) {
This could definitely benefit from sdiv, too.

[danno, 2012-10-01 21:16:17]

LGTM, I'll land this for you

http://codereview.chromium.org/10977051/diff/17002/src/v8globals.h
File src/v8globals.h (right):

http://codereview.chromium.org/10977051/diff/17002/src/v8globals.h#newcode441
src/v8globals.h:441: SUDIV = 5,   // ARM
I changed this to 4 for the commit, I think keeping the sequence for a platform
is more consistent than uniqueness over all platforms, which is already not
guaranteed.

[acleung, 2012-10-08 22:16:03]

On 2012/10/01 21:16:17, danno wrote:
> LGTM, I'll land this for you
> 
> http://codereview.chromium.org/10977051/diff/17002/src/v8globals.h
> File src/v8globals.h (right):
> 
> http://codereview.chromium.org/10977051/diff/17002/src/v8globals.h#newcode441
> src/v8globals.h:441: SUDIV = 5,   // ARM
> I changed this to 4 for the commit, I think keeping the sequence for a
platform
> is more consistent than uniqueness over all platforms, which is already not
> guaranteed.

Has someone tried this in Android Chrome? I ran the V8 benchmark and it hangs on
Crypto.