Work around a PGI floating point bug.

Executables compiled on Intel Haswell architectures sometimes return
incorrect floating point results when using fused multiply-add
instructions.
3 jobs for bugfix/workaround-pgi-fp-bug in 0 seconds (queued for 92 minutes and 54 seconds)
Status Job ID Name Coverage
  Test
failed #7182
osx
test_osx

failed #7180
gfortran ubuntu
test_ubuntu_gfortran

failed #7181
ubuntu intel
test_ubuntu_intel

 
Name Stage Failure
failed
test_osx Test There is an unknown failure, please try again
No job trace
failed
test_ubuntu_intel Test There is an unknown failure, please try again
No job trace
failed
test_ubuntu_gfortran Test There is an unknown failure, please try again
No job trace