| commit | 63653413749b99effcc48059d723b766080cf81c | |
| author | bsimmers <bsimmers@fb.com> | |
| Wed, 14 May 2014 23:53:14 +0000 (14 16:53 -0700) | ||
| committer | JoelMarcey <joelm@fb.com> | |
| Thu, 22 May 2014 18:15:41 +0000 (22 11:15 -0700) | ||
| tree | 01f8fd07cd4d3f802deb2edd922f0e3ede90909c | treesnapshot (tar.gz zip) |
| parent | 5f0be2910681e59b2a6ef4f8e31a6dda6604d74f | commitdiff |
Profile to find places we can guard on KindOfStaticString, take 2
Summary: If 100% of the Str values flowing through a guard in profiling
translations are KindOfStaticString, we change that guard to KindOfStaticString
in the optimized translation. In practice, most values are KindOfString with a
static _count but guarding on that is much more expensive and must be justified
with concrete savings. This diff just covers the cases that should be made
strictly better: the guard is the same size/cost, shouldn't fail any more often
than it used to, and the generated code will only differ in that there may be
fewer refcounting operations.
Reviewed By: @ottoni
Differential Revision: D1342007
Summary: If 100% of the Str values flowing through a guard in profiling
translations are KindOfStaticString, we change that guard to KindOfStaticString
in the optimized translation. In practice, most values are KindOfString with a
static _count but guarding on that is much more expensive and must be justified
with concrete savings. This diff just covers the cases that should be made
strictly better: the guard is the same size/cost, shouldn't fail any more often
than it used to, and the generated code will only differ in that there may be
fewer refcounting operations.
Reviewed By: @ottoni
Differential Revision: D1342007