[llvm] [AArch64][GlobalISel] Improve fixed-point int-to-fp selection (PR #224055)

[Vimal Patel via llvm-commits]

================
@@ -386,6 +386,39 @@ def : Pat<(int_aarch64_stlxp GPR64:$lo, GPR64:$hi, GPR64:$addr),
 def : Pat<(int_aarch64_stxp GPR64:$lo, GPR64:$hi, GPR64:$addr),
           (STXPX GPR64:$lo, GPR64:$hi, GPR64:$addr)>;
 
+def : Pat<(int_aarch64_neon_vcvtfxu2fp FPR32:$Rn, vecshiftR32:$imm),
----------------
pvimal816a wrote:

It could be done. But, then ordering of these patterns relative to the existing GPR source pattern matters because SelectionDAG selector ends up inserting  unwanted `fmov`s. There is no use of these patterns for SelectionDAG selector. So, I'd put them in the GISel specific file.

https://github.com/llvm/llvm-project/pull/224055