[AArch64][SVE] Fold svrev(svrev(v)) to v

[k-arrows]

GCC has implemented this (see patch). Consider the following example.
https://godbolt.org/z/xj86nKn51

#include <arm_sve.h>

svint32_t f(svint32_t v)
{
  return svrev_s32 (svrev_s32 (v));
}

GCC:

f(__SVInt32_t):
        ret

Clang:

f(__SVInt32_t):
        rev     z0.s, z0.s
        rev     z0.s, z0.s
        ret

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: None (k-arrows)

GCC has implemented this (see [patch](https://github.com/gcc-mirror/gcc/commit/f0eabc52c9a2d3da0bfc201da7a5c1658b76e9a4)). Consider the following example. https://godbolt.org/z/xj86nKn51 ```cpp #include <arm_sve.h>

svint32_t f(svint32_t v)
{
return svrev_s32 (svrev_s32 (v));
}

GCC:
```asm
f(__SVInt32_t):
        ret

Clang:

f(__SVInt32_t):
        rev     z0.s, z0.s
        rev     z0.s, z0.s
        ret

[jf-botto]

I'd love to work on this if possible?

[sdesmalen-arm]

I'd love to work on this if possible?

Go for it!

For this case we can let Clang emit a llvm.vector.reverse intrinsic for which such an InstCombine is already implemented, instead of emitting the llvm.aarch64.sve.rev intrinsic that Clang currently emits.