This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [tree-ssa-lno] vectorizer patch.
- From: Richard Henderson <rth at redhat dot com>
- To: Dorit Naishlos <DORIT at il dot ibm dot com>
- Cc: gcc-patches at gcc dot gnu dot org, Pop Sébastian <pop at gauvain dot u-strasbg dot fr>
- Date: Fri, 2 Jan 2004 12:28:36 -0800
- Subject: Re: [tree-ssa-lno] vectorizer patch.
- References: <OF8010485C.A1ADECFB-ONC2256E0E.0052F0FF-C2256E0E.005B1C89@il.ibm.com>
On Thu, Jan 01, 2004 at 06:35:10PM +0200, Dorit Naishlos wrote:
> Test cases will follow shortly; The attached test file, if compiled with
> "-O2 -ftree-vectorize -fno-inline -maltivec", should emit:
> "
> vectorized 0 loops in function.
> vectorized 0 loops in function.
> vectorized 0 loops in function.
> vectorized 5 loops in function.
> vectorized 0 loops in function.
> vectorized 0 loops in function.
> vectorized 0 loops in function.
> "
> and it should run correctly, at least on powerpc-apple-darwin6.4 (where
> it's bootstrapping now).
FYI, adding the obvious
+#define UNITS_PER_SIMD_WORD \
+ (TARGET_SSE ? 16 : TARGET_MMX || TARGET_3DNOW ? 8 : 0)
to i386.h, and using -msse2 instead of -maltivec, results in some sort
of memory corruption with that test case (symptom is segv in free). I
didn't track it down beyond noting that valgrind complains a LOT before
it finally dies.
r~