[Patch] Fix PR44328 - switch/case optimization produces an invalid lookup table index

["Ian Bolton" <Ian dot Bolton at arm dot com>]

This patch fixes PR44328, and provides a basic regression test for it.

The bug was caused by an underflow occurring in tree-switch-conversion.c,
when the lower bound for the valid switch range was subtracted from the
minimum value for the enum, thereby causing values beyond the enum's max
value to be possible.  Without this change, tree-vrp.c was optimising away
the range check because it seemed unnecessary based on the max value it
had for the index.

Note that the bug does not show on trunk unless you use -fstrict-enums.
My testing has been done on 4.4 branch and 4.5 branch for ARM.

I would appreciate some feedback on how to make the test better (e.g.
where to put it, should it be executed too?), and advice on whether
the patch should be aimed at 4.5 or trunk initially.

Cheers,
Ian


2010-08-12  Ian Bolton  <ian.bolton@arm.com>

	* tree-switch-conversion.c (gen_inbound_check): Ensure that the
	type for the conditional has wide enough range.

	* testsuite/gcc.target/arm/pr44328.C: New test.

Attachment: pr44328.patch
Description: Binary data