Tue Jan 01, 2019 6:12 am
I have tried simple example using eigen vector functions on both the platforms : Intel and s390x.
On Intel , the code can compile with vectorization (by default -msse2 ) and generates output.
On s390x, compilation failed with an error: __builtin_s390_vfi was not declared in this scope.
Same issue is observed while building TensorFlow with vectorization on z14 vm.
I have used -march=z14 -mzvector flags on z14 s390x vm.
Reference: http://eigen.tuxfamily.org/index.php?ti ... torization
Could you confirm if Eigen supports vectorization for s390x?