[Mlir-commits] [mlir] [mlir][emitc] arith.cmpf to EmitC conversion (PR #93671)

[Simon Camphausen]

================
@@ -107,6 +107,234 @@ func.func @arith_select(%arg0: i1, %arg1: tensor<8xi32>, %arg2: tensor<8xi32>) -
 
 // -----
 
+func.func @arith_cmpf_false(%arg0: f32, %arg1: f32) -> i1 {
+  // CHECK-LABEL: arith_cmpf_false
+  // CHECK-SAME: ([[Arg0:[^ ]*]]: f32, [[Arg1:[^ ]*]]: f32)
+  // CHECK-DAG: [[False:[^ ]*]] = "emitc.constant"() <{value = false}> : () -> i1
+  %false = arith.cmpf false, %arg0, %arg1 : f32
+  // CHECK: return [[False]]
+  return %false: i1
+}
+
+// -----
+
+func.func @arith_cmpf_oeq(%arg0: f32, %arg1: f32) -> i1 {
+  // CHECK-LABEL: arith_cmpf_oeq
+  // CHECK-SAME: ([[Arg0:[^ ]*]]: f32, [[Arg1:[^ ]*]]: f32)
+  // CHECK-DAG: [[EQ:[^ ]*]] = emitc.cmp eq, [[Arg0]], [[Arg1]] : (f32, f32) -> i1
+  // CHECK-DAG: [[NaNArg0:[^ ]*]] = emitc.cmp eq, [[Arg0]], [[Arg0]] : (f32, f32) -> i1
+  // CHECK-DAG: [[NaNArg1:[^ ]*]] = emitc.cmp eq, [[Arg1]], [[Arg1]] : (f32, f32) -> i1
----------------
simon-camp wrote:

nit: Capture these values as NotNaNArg for all ordered tests.

https://github.com/llvm/llvm-project/pull/93671