[Mlir-commits] [mlir] [mlir] add some FP classification ops and their lowering to libdevice (PR #127322)

[William Moses]

================
@@ -1058,3 +1058,29 @@ gpu.module @test_module_53 {
     func.return %result32, %result64 : f32, f64
   }
 }
+
+gpu.module @test_module_54 {
+  // CHECK: llvm.func @__nv_isinff(f32) -> i1
----------------
wsmoses wrote:

Usually the declarations in libdevice itself return an i32 not an i1. Can we emit i32 to avoid a type error?

https://github.com/llvm/llvm-project/pull/127322