[llvm] [VPlan] Run narrowInterleaveGroups during general VPlan optimizations. (PR #149706)

[Shunsuke Watanabe via llvm-commits]

s-watanabe314 wrote:

@fhahn  Our tests started failing after this commit. These are Fortran tests, could you please check?
The environment is graviton3 (neoverse-v1), and the options specified are `-O3 -march=armv8.4-a+sve -msve-vector-bits=256`.

Failing tests:
- [Fortran/0120/0120_0040.F90](https://github.com/fujitsu/compiler-test-suite/blob/main/Fortran/0120/0120_0040.F90)
- [Fortran/0641/0641_0021.f](https://github.com/fujitsu/compiler-test-suite/blob/main/Fortran/0641/0641_0021.f)


In the Fortran sample code below, all values of `a` should be 1, but some appear to be 0.

- sample.F90
```Fortran
#define N1 4
#define N2 16

#define TYPE real(kind=4)

program main
  TYPE, dimension(N1,N2) :: a,b

  call init(a,b,N1,N2)
  print *, a
end program main

subroutine init(a,b,n1,n2)
  TYPE, dimension(n1,n2) :: a,b

   do j=1,n2
      do k=1,n1
         a(k,j) = 1.0
         b(k,j) = 1.0
       enddo
   enddo
end subroutine init
```
```
$ flang sample.F90 -O0 -o o0.out
$ flang sample.F90 -O3 -march=armv8.4-a+sve -msve-vector-bits=256 -o v256.out
$ flang sample.F90 -O3 -march=armv8.4-a+sve -msve-vector-bits=scalable -o scalable.out
$ ./o0.out
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
$ ./v256.out
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.
$ ./scalable.out
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.
 ```

https://github.com/llvm/llvm-project/pull/149706