| Message ID | 1440700507.4911.14.camel@oc8801110288.ibm.com |
|---|---|
| State | New |
| Headers | show |
On 08/27/2015 12:35 PM, Bill Schmidt wrote: > Hi, > > I observed today that { dg-require-effective-target vect_double } was > unintentionally excluding all powerpc* targets. This patch corrects the > situation by enabling vect_double when the VSX architectural support is > present, which provides the vector double type. As a result of this > change, I observed one XPASS. This is due to vect_no_align being used > where { vect_no_align && { ! vect_hw_misalign } } is actually desired. > (I made a number of similar changes back in April, but didn't catch this > one at the time.) > > Tested on powerpc64-unknown-linux-gnu and powerpc64le-unknown-linux-gnu > with no regressions. Is this ok for trunk? > > Thanks, > Bill > > > 2015-08-27 Bill Schmidt <wschmidt@linux.vnet.ibm.com> > > * lib/target-supports.exp (check-effective_target_vect_double): > Enable for Power targets with VSX hardware available. > * gfortran.dg/vect/O3-pr49957.f: Replace vect_no_align with > vect_no_align && { ! vect_hw_misalign }. OK. jeff
Index: gcc/testsuite/gfortran.dg/vect/O3-pr49957.f =================================================================== --- gcc/testsuite/gfortran.dg/vect/O3-pr49957.f (revision 227258) +++ gcc/testsuite/gfortran.dg/vect/O3-pr49957.f (working copy) @@ -13,4 +13,4 @@ enddo return end -! { dg-final { scan-tree-dump "vectorized 1 loops" "vect" { xfail vect_no_align } } } +! { dg-final { scan-tree-dump "vectorized 1 loops" "vect" { xfail { vect_no_align && { ! vect_hw_misalign } } } } } Index: gcc/testsuite/lib/target-supports.exp =================================================================== --- gcc/testsuite/lib/target-supports.exp (revision 227258) +++ gcc/testsuite/lib/target-supports.exp (working copy) @@ -3711,6 +3711,8 @@ proc check_effective_target_vect_double { } { } } elseif { [istarget spu-*-*] } { set et_vect_double_saved 1 + } elseif { [istarget powerpc*-*-*] && [check_vsx_hw_available] } { + set et_vect_double_saved 1 } }
Hi, I observed today that { dg-require-effective-target vect_double } was unintentionally excluding all powerpc* targets. This patch corrects the situation by enabling vect_double when the VSX architectural support is present, which provides the vector double type. As a result of this change, I observed one XPASS. This is due to vect_no_align being used where { vect_no_align && { ! vect_hw_misalign } } is actually desired. (I made a number of similar changes back in April, but didn't catch this one at the time.) Tested on powerpc64-unknown-linux-gnu and powerpc64le-unknown-linux-gnu with no regressions. Is this ok for trunk? Thanks, Bill 2015-08-27 Bill Schmidt <wschmidt@linux.vnet.ibm.com> * lib/target-supports.exp (check-effective_target_vect_double): Enable for Power targets with VSX hardware available. * gfortran.dg/vect/O3-pr49957.f: Replace vect_no_align with vect_no_align && { ! vect_hw_misalign }.