Fix deoptimization from %_MathPow. Do not prepare the stack for a C...

Fix deoptimization from %_MathPow. Do not prepare the stack for a C
call and then bailout without performing the call.

In order to not duplicate code, convert both heap number and smi to a
double in an xmm register if possible. Then setup the stack and call
the C function.

BUG=http://code.google.com/p/v8/issues/detail?id=986


Committed: http://code.google.com/p/v8/source/detail?r=6032

[William Hesse, 2010-12-15 14:38:38]

LGTM, except that we shouldn't lose the integer power optimization.  If this is
committed, we should check to see if restoring the optimization improves
performance.

http://codereview.chromium.org/5900001/diff/5001/src/ia32/lithium-codegen-ia3...
File src/ia32/lithium-codegen-ia32.cc (right):

http://codereview.chromium.org/5900001/diff/5001/src/ia32/lithium-codegen-ia3...
src/ia32/lithium-codegen-ia32.cc:2285: __ jmp(&call);
Calling power_double_double rather than power_double_int here loses the
optimization for integer powers, which is much faster in many cases.  Integer
powers use multiplication and squaring, and double powers use log and exp.

[Mads Ager (chromium), 2010-12-15 14:43:20]

http://codereview.chromium.org/5900001/diff/5001/src/ia32/lithium-codegen-ia3...
File src/ia32/lithium-codegen-ia32.cc (right):

http://codereview.chromium.org/5900001/diff/5001/src/ia32/lithium-codegen-ia3...
src/ia32/lithium-codegen-ia32.cc:2285: __ jmp(&call);
On 2010/12/15 14:38:38, William Hesse wrote:
> Calling power_double_double rather than power_double_int here loses the
> optimization for integer powers, which is much faster in many cases.  Integer
> powers use multiplication and squaring, and double powers use log and exp.

Good point. I'll check performance. I didn't notice that two different functions
were called here. Thanks!