[llvm] [LLVM][CodeGen][SVE] Improve isel for split vector bfloat conversions. (PR #184357)

[Graham Hunter via llvm-commits]

================
@@ -85,16 +85,13 @@ define <vscale x 4 x bfloat> @fadd_nxv4bf16(<vscale x 4 x bfloat> %a, <vscale x
 define <vscale x 8 x bfloat> @fadd_nxv8bf16(<vscale x 8 x bfloat> %a, <vscale x 8 x bfloat> %b) {
 ; NOB16B16-LABEL: fadd_nxv8bf16:
 ; NOB16B16:       // %bb.0:
-; NOB16B16-NEXT:    uunpkhi z2.s, z1.h
-; NOB16B16-NEXT:    uunpkhi z3.s, z0.h
-; NOB16B16-NEXT:    uunpklo z1.s, z1.h
-; NOB16B16-NEXT:    uunpklo z0.s, z0.h
+; NOB16B16-NEXT:    movi v2.2d, #0000000000000000
----------------
huntergr-arm wrote:

Any reason to prefer the NEON instruction over dup/fmov 0? I guess it shouldn't matter too much either way...

https://github.com/llvm/llvm-project/pull/184357