http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51084
Bug #: 51084
Summary: bounds checking not optimized to a single comparison
Classification: Unclassified
Product: gcc
Version: 4.6.2
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
AssignedTo: unassig...@gcc.gnu.org
ReportedBy: egg...@gnu.org
Host: x86-64
Target: x86-64
Build: x86-64
Created attachment 25788
--> http://gcc.gnu.org/bugzilla/attachment.cgi?id=25788
code for pure0 and pure1 is slower than for pure2 and pure3
I ran into this problem when tuning GNU Emacs; see
<http://lists.gnu.org/archive/html/emacs-devel/2011-11/msg00145.html>.
Emacs checks whether a pointer P points into an
array A by doing the equivalent of
(A <= P && P < &A[sizeof A]). This can be
done with a subtraction followed by a single
comparison, but GCC generates slower code that does
two comparisons and a conditional branch.
The attached source code illustrates the problem.
When compiled using "gcc -O2 -S", the functions
pure0 and pure1 should generate code as fast as
that generated by pure2 and pure3. But the slower
code is generated.
