Re: [PATCH] Implement redundant store elimination, fix PR38513

[Ira Rosen <IRAR at il dot ibm dot com>]

gcc-patches-owner@gcc.gnu.org wrote on 30/12/2008 23:22:01:

> On Tue, 30 Dec 2008, Richard Guenther wrote:
>
> > On Tue, 30 Dec 2008, Richard Guenther wrote:
> >
> > >
> > > This implements redundant store elimination (just like postreload CSE

> > > does) from within FRE.  It triggers quite some time for C++ code and
> > > also breaks the following testcases for now
> > >
> > > FAIL: g++.dg/warn/Warray-bounds.C  (test for warnings, line 32)
> > > FAIL: g++.dg/warn/Warray-bounds.C  (test for warnings, line 43)
> > > FAIL: g++.dg/warn/Warray-bounds.C  (test for warnings, line 54)
> > >
> > > FAIL: gcc.dg/Warray-bounds.c  (test for warnings, line 32)
> > > FAIL: gcc.dg/Warray-bounds.c  (test for warnings, line 43)
> > > FAIL: gcc.dg/Warray-bounds.c  (test for warnings, line 54)
> > > FAIL: gcc.dg/vect/vect-35.c scan-tree-dump-times vect
> "vectorized 1 loops" 1
> > > FAIL: gcc.dg/vect/vect-multitypes-16.c scan-tree-dump-times vect
> > > "vectorized 1 loops" 1
> > > FAIL: gcc.dg/vect/vect-multitypes-17.c scan-tree-dump-times vect
> > > "vectorized 1 loops" 1
> > > FAIL: gcc.dg/vect/vect-reduc-dot-u8b.c scan-tree-dump-times vect
> > > "vectorized 1 loops" 1
> > > FAIL: gcc.dg/vect/slp-widen-mult-u8.c scan-tree-dump-times vect
> > > "vectorized 1 loops" 1
> > > FAIL: gcc.dg/vect/wrapv-vect-reduc-dot-s8b.c scan-tree-dump-times
vect
> > > "vectorized 1 loops" 1

There are additional loops that get vectorized with your patch. These are
initialization loops that we tried to prevent their vectorization by
inserting ifs. In the attached patch I replaced those loops with static
array initializations, except vect-35.c, where I changed the expected
number of vectorized loops. (Checked on powerpc-linux-gnu).

Ira

(See attached file: vect-tests.txt)

Attachment: vect-tests.txt
Description: Text document