[llvm] [AMDGPU] Promote uniform ops to I32 in DAGISel (PR #106383)

[Matt Arsenault via llvm-commits]

================
@@ -18,189 +18,33 @@ declare hidden half @_Z4pownDhi(half, i32)
 ; --------------------------------------------------------------------
 
 define half @test_pow_fast_f16(half %x, half %y) {
-; CHECK-LABEL: test_pow_fast_f16:
-; CHECK:       ; %bb.0:
----------------
arsenm wrote:

I fixed this in 528bcf3a55ca520c31c77ed5fbacf09bff8f39ec

https://github.com/llvm/llvm-project/pull/106383