[v8-dev] Implementing Math.floor and Math.sqrt for crankshaft.

Implementing Math.floor and Math.sqrt for crankshaft.

BUG=none
TEST=none

[Rodolph Perfetta, 2011-01-13 11:37:53]

replaces issue 6124005 http://codereview.chromium.org/6124005/

[Mads Ager (chromium), 2011-01-13 12:17:52]

LGTM

[Mads Ager (chromium), 2011-01-13 12:25:54]

http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h
File src/arm/assembler-arm.h (right):

http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h#newcode171
src/arm/assembler-arm.h:171: // d1 has also been excluded from allocation to be
used as a scratch
I already landed this, but why is this in this patch? I don't see any use of
double_scratch()?

[Søren Thygesen Gjesse, 2011-01-13 12:33:06]

On 2011/01/13 12:25:54, Mads Ager wrote:
> http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h
> File src/arm/assembler-arm.h (right):
> 
>
http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h#newcode171
> src/arm/assembler-arm.h:171: // d1 has also been excluded from allocation to
be
> used as a scratch
> I already landed this, but why is this in this patch? I don't see any use of
> double_scratch()?

Adding to Mads comment how about only having scratch_double0() as d0 and then
use scratch_double0().high() or scratch_double0().low() for single precision
scratch registers?

[Rodolph Perfetta, 2011-01-13 13:38:55]

On 2011/01/13 12:33:06, Søren Gjesse wrote:
> On 2011/01/13 12:25:54, Mads Ager wrote:
> > http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h
> > File src/arm/assembler-arm.h (right):
> > 
> >
>
http://codereview.chromium.org/6250002/diff/1/src/arm/assembler-arm.h#newcode171
> > src/arm/assembler-arm.h:171: // d1 has also been excluded from allocation to
> be
> > used as a scratch
> > I already landed this, but why is this in this patch? I don't see any use of
> > double_scratch()?

There is no use of double_scratch(), I had submitted this code in a previous
upload, and at the time it was suggested I should add double_scratch() at the
same time as single_scratch().
 
> Adding to Mads comment how about only having scratch_double0() as d0 and then
> use scratch_double0().high() or scratch_double0().low() for single precision
> scratch registers?

that's fine by me, the only reason I have two double registers for scratch is if
one need a single and double scratch register which don't overlap.

I am happy to upload a new change for it, just let me know.