Improve the assembly code for power function with integer exponential on x64
authoryangguo@chromium.org <yangguo@chromium.org@ce2b1a6d-e550-0410-aec6-3dcde31c8c00>
Tue, 18 Sep 2012 08:04:31 +0000 (08:04 +0000)
committeryangguo@chromium.org <yangguo@chromium.org@ce2b1a6d-e550-0410-aec6-3dcde31c8c00>
Tue, 18 Sep 2012 08:04:31 +0000 (08:04 +0000)
commitcf85bfd72516f0cbb44e3d7deab1a1c07f57c8b1
tree2ee92f02086c5171d05ebbc28dfaef57ef83f5a7
parent36dd23aa1b24f1b30defd7037f948610a7d0c767
Improve the assembly code for power function with integer exponential on x64

The change removes one unused multiply and reschedules
the shift, multiply and jump instructions to reduce
stall. Experiment shows it improve about 20% performance
on x64 for exponetials from about 100 to 2000.

Review URL: https://chromiumcodereview.appspot.com/10939013
Patch from Xi Qian <xi.qian@intel.com>.

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@12535 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
src/ia32/code-stubs-ia32.cc
src/x64/code-stubs-x64.cc