https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87597
Bug ID: 87597
Summary: wrong result with matmul inlining
Product: gcc
Version: 8.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: gallmeister at th dot physik.uni-frankfurt.de
Target Milestone: ---
Created attachment 44828
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=44828&action=edit
minimal program
Using matmul with a matrix, which is given as a constant, yields a compiler
warning and wrong results when compiling with optimization switched on, which
uses matmul inlining.
the attached minimal program produces folowing output:
$ gfortran-8 -O0 -Wall testMatMul.f90
$ ./a.out
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (1.00000000,0.00000000)
(2.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (3.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
$ gfortran-8 -O3 -Wall testMatMul.f90
testMatMul.f90:19:0:
B = matmul(A,gamma5)
Warning: iteration 3 invokes undefined behavior
[-Waggressive-loop-optimizations]
testMatMul.f90:19:0:
B = matmul(A,gamma5)
note: within this loop
$ ./a.out
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (1.401298464E-45,0.00000000)
(2.00000000,0.00000000) (0.00000000,0.00000000)
(0.00000000,0.00000000) (8.688050479E-44,6.00000000)
(3.00000000,0.00000000) (0.00000000,0.00000000)
(2.858648867E-43,3.443410716E-41) (1.681558157E-44,0.00000000)
