use V2F32toI64 for i64 build_pair (i32 bitcast f32, i32 bitcast f32)

Add a dag pattern to lower:
i64 = build_pair (i32 = bitcast (f32 A), i32 = bitcast (f32 B))
-> i64 = V2F32toI64 A, B

Author: Prince781

llvm/lib/Target/NVPTX/NVPTXInstrInfo.td

@@ -3441,6 +3441,9 @@ def : Pat<(v2i16 (build_vector i16:$a, i16:$b)),
           (V2I16toI32 $a, $b)>;
 def : Pat<(v2f32 (build_vector f32:$a, f32:$b)),
           (V2F32toI64 $a, $b)>;
+def : Pat<(i64 (build_pair (i32 (bitconvert f32:$a)),
+                           (i32 (bitconvert f32:$b)))),
+          (V2F32toI64 $a, $b)>;
 def : Pat<(i64 (build_pair i32:$a, i32:$b)),
           (V2I32toI64 $a, $b)>;