https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126996
Bug ID: 126996
Summary: [17 Regression] SLP patterns ICE after
g:cd962857ea3d8db2e7ed08f98df4152e7739e8cd
Product: gcc
Version: 17.0
Status: UNCONFIRMED
Keywords: ice-on-valid-code
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: tnfchris at gcc dot gnu.org
Target Milestone: ---
Target: aarch64*
subroutine c1f4kb ( ido, l1, cc, in1)
real cc(in1,l1,ido,4)
if ( na == 1 ) then
do k = 1, l1
ti1 = cc(2,k,1,1)-cc(2,k,1,3)
tr4 = cc(2,k,1,4)-cc(2,k,1,2)
tr1 = cc(1,k,1,1)-cc(1,k,1,3)
ti4 = cc(1,k,1,2)-cc(1,k,1,4)
cc(1,k,1,2) = tr1
cc(1,k,1,4) = tr1-tr4
cc(2,k,1,2) = ti1
cc(2,k,1,4) = ti1-ti4
end do
end if
end
compiled with -Ofast -march=armv8-a+sve
ICEs with a segfault
during GIMPLE pass: vect
complex-ice.f90:1:1:
1 | subroutine c1f4kb ( ido, l1, cc, in1)
| ^~~~~~~~~~~~~~~~~
internal compiler error: Segmentation fault
This results in WRF and a few other workloads in SPECCPU 2017 failing.
