[llvm] [DAGCombiner] Narrow the integer source of uint_to_fp (PR #222899)

[Dmitry Sidorov via llvm-commits]

================
@@ -912,6 +912,311 @@ define double @v_uitofp_i8_to_f64(i8 %arg0) nounwind {
   ret double %cvt
 }
 
+define float @v_uitofp_i64_to_f32_mask255(i64 %arg0) nounwind {
----------------
MrSidims wrote:

IMHO it's better to pre-commit these tests (along with their signed counterpart).

https://github.com/llvm/llvm-project/pull/222899