[llvm] [AArch64] SVE Shuffleopt: merge reduction reverse into tbl (PR #206047)

[Paul Walker via llvm-commits]

================
@@ -638,5 +638,463 @@ exit:
   ret void
 }
 
+define void @uitofp_nxv8i16_to_nxv8f64_deinterleave_fma_with_reverse_double(ptr %src, ptr %src2, ptr %dst, <vscale x 8 x i1> %mask) #0 {
+; CHECK-LABEL: uitofp_nxv8i16_to_nxv8f64_deinterleave_fma_with_reverse_double:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    cnth x9
+; CHECK-NEXT:    index z7.d, #0, #-4
+; CHECK-NEXT:    movi v0.2d, #0000000000000000
----------------
paulwalker-arm wrote:

Can the tests be IR tests? That would make it easier to see the transformation.

https://github.com/llvm/llvm-project/pull/206047