[Mlir-commits] [mlir] [mlir][emitc] Add 'emitc.switch' op to the dialect (PR #102331)

[Matthias Gehre]

================
@@ -0,0 +1,129 @@
+// RUN: mlir-translate -mlir-to-cpp %s | FileCheck %s
+
+// CHECK-LABEL:   int32_t v1 = 1;
+// CHECK-LABEL:   switch(v1) {
+// CHECK:         case (2): {
+// CHECK:           int32_t v2 = func_b();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         case (5): {
+// CHECK:           int32_t v3 = func_a();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         default: {
+// CHECK:           float v4 = 4.200000000e+01f;
+// CHECK:           float v5 = 4.200000000e+01f;
+// CHECK:           func2(v4);
+// CHECK:           func3(v5, v4);
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         return;
+// CHECK:       }
+func.func @emitc_switch_default() {
+  %0 = "emitc.variable"(){value = 1 : ui32} : () -> ui32
+
+  emitc.switch %0 : ui32
+  case 2: {
+    %1 = emitc.call_opaque "func_b" () : () -> i32
+    emitc.yield
+  }
+  case 5: {
+    %2 = emitc.call_opaque "func_a" () : () -> i32
+    emitc.yield
+  }
+  default: {
+    %3 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+    %4 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+
+    emitc.call_opaque "func2" (%3) : (f32) -> ()
+    emitc.call_opaque "func3" (%3, %4) { args = [1 : index, 0 : index] } : (f32, f32) -> ()
+    emitc.yield
+  }
+  return
+}
+
+// CHECK-LABEL: void emitc_switch_i16() {
+// CHECK:         int16_t v1 = 1;
+// CHECK:         switch(v1) {
+// CHECK:         case (2): {
+// CHECK:           int32_t v2 = func_b();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         case (5): {
+// CHECK:           int32_t v3 = func_a();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         default: {
+// CHECK:           float v4 = 4.200000000e+01f;
+// CHECK:           float v5 = 4.200000000e+01f;
+// CHECK:           func2(v4);
+// CHECK:           func3(v5, v4);
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         return;
+// CHECK:       }
+func.func @emitc_switch_i16() {
+  %0 = "emitc.variable"(){value = 1 : i16} : () -> i16
+
+  emitc.switch %0 : i16
+  case 2: {
+    %1 = emitc.call_opaque "func_b" () : () -> i32
+    emitc.yield
+  }
+  case 5: {
+    %2 = emitc.call_opaque "func_a" () : () -> i32
+    emitc.yield
+  }
+  default: {
+    %3 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+    %4 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+
+    emitc.call_opaque "func2" (%3) : (f32) -> ()
+    emitc.call_opaque "func3" (%3, %4) { args = [1 : index, 0 : index] } : (f32, f32) -> ()
+    emitc.yield
+  }
+  return
+}
+
+// CHECK-LABEL: void emitc_switch_ui16() {
+// CHECK:         uint16_t v1 = 1;
+// CHECK:         switch(v1) {
+// CHECK:         case (2): {
+// CHECK:           int32_t v2 = func_b();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         case (5): {
+// CHECK:           int32_t v3 = func_a();
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         default: {
+// CHECK:           float v4 = 4.200000000e+01f;
+// CHECK:           float v5 = 4.200000000e+01f;
+// CHECK:           func2(v4);
+// CHECK:           func3(v5, v4);
+// CHECK:           break;
+// CHECK:         }
+// CHECK:         return;
+// CHECK:       }
+func.func @emitc_switch_ui16() {
+  %0 = "emitc.variable"(){value = 1 : ui16} : () -> ui16
+
+  emitc.switch %0 : ui16
+  case 2: {
+    %1 = emitc.call_opaque "func_b" () : () -> i32
+    emitc.yield
+  }
+  case 5: {
+    %2 = emitc.call_opaque "func_a" () : () -> i32
+    emitc.yield
+  }
+  default: {
+    %3 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+    %4 = "emitc.variable"(){value = 42.0 : f32} : () -> f32
+
+    emitc.call_opaque "func2" (%3) : (f32) -> ()
+    emitc.call_opaque "func3" (%3, %4) { args = [1 : index, 0 : index] } : (f32, f32) -> ()
+    emitc.yield
+  }
+  return
+}
----------------
mgehre-amd wrote:

Ok, then I think it's better to continue requiring explicit `yields` for now so users don't get confused about a missing yield and think that this is doing fall-through.

https://github.com/llvm/llvm-project/pull/102331