[CIR] Add support for the RequiresExpr

[AmrDeveloper]

Add support for the RequiresExpr

[AmrDeveloper]

A nice pattern to fold but not sure how common it is 🤔

[andykaylor]

We should definitely fold that somewhere (though only when optimizations are enabled). I don't like that classic codegen folds this by default at -O0.

I wouldn't expect purely constant conditions like this to be common in the initially generated IR, but after other optimizations it happens fairly often.

[AmrDeveloper]

Sure, that makes sense

[llvmbot]

@llvm/pr-subscribers-clangir

@llvm/pr-subscribers-clang

Author: Amr Hesham (AmrDeveloper)

Changes

Add support for the RequiresExpr

---

Full diff: https://github.com/llvm/llvm-project/pull/171818.diff

2 Files Affected:

• (modified) clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp (+1-2)
• (added) clang/test/CIR/CodeGen/requires-expr.cpp (+51)

diff --git a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
index 3887433e5e181..60bd76f89b00b 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
@@ -850,8 +850,7 @@ class ScalarExprEmitter : public StmtVisitor<ScalarExprEmitter, mlir::Value> {
     return {};
   }
   mlir::Value VisitRequiresExpr(const RequiresExpr *e) {
-    cgf.cgm.errorNYI(e->getSourceRange(), "ScalarExprEmitter: requires");
-    return {};
+    return builder.getBool(e->isSatisfied(), cgf.getLoc(e->getExprLoc()));
   }
   mlir::Value VisitArrayTypeTraitExpr(const ArrayTypeTraitExpr *e) {
     cgf.cgm.errorNYI(e->getSourceRange(),
diff --git a/clang/test/CIR/CodeGen/requires-expr.cpp b/clang/test/CIR/CodeGen/requires-expr.cpp
new file mode 100644
index 0000000000000..6ca6da9802508
--- /dev/null
+++ b/clang/test/CIR/CodeGen/requires-expr.cpp
@@ -0,0 +1,51 @@
+// RUN: %clang_cc1 -std=c++20 -triple x86_64-unknown-linux-gnu -Wno-unused-value -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s -check-prefix=CIR
+// RUN: %clang_cc1 -std=c++20 -triple x86_64-unknown-linux-gnu -Wno-unused-value -fclangir -emit-llvm %s -o %t-cir.ll
+// RUN: FileCheck --input-file=%t-cir.ll %s -check-prefix=LLVM
+// RUN: %clang_cc1 -std=c++20 -triple x86_64-unknown-linux-gnu -Wno-unused-value -emit-llvm %s -o %t.ll
+// RUN: FileCheck --input-file=%t.ll %s -check-prefix=OGCG
+
+template <typename T> void summable(T a) {
+  if (requires { a + a; }) {
+    T b = a + a;
+  }
+}
+
+// CIR: %[[A_ADDR:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["a", init]
+// CIR: cir.store %[[ARG_A:.*]], %[[A_ADDR]] : !s32i, !cir.ptr<!s32i>
+// CIR: cir.scope {
+// CIR:   %[[CONST_TRUE:.*]] = cir.const #true
+// CIR:   cir.if %[[CONST_TRUE]] {
+// CIR:     %[[B_ADDR:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["b", init]
+// CIR:     %[[TMP_A_1:.*]] = cir.load {{.*}} %[[A_ADDR]] : !cir.ptr<!s32i>, !s32i
+// CIR:     %[[TMP_A_2:.*]] = cir.load {{.*}} %[[A_ADDR]] : !cir.ptr<!s32i>, !s32i
+// CIR:     %[[RESULT:.*]] = cir.binop(add, %[[TMP_A_1]], %[[TMP_A_2]]) nsw : !s32i
+// CIR:     cir.store {{.*}} %[[RESULT]], %[[B_ADDR]] : !s32i, !cir.ptr<!s32i>
+// CIR:   }
+// CIR: }
+
+// LLVM:   %[[B_ADDR:.*]] = alloca i32, i64 1, align 4
+// LLVM:   %[[A_ADDR:.*]] = alloca i32, i64 1, align 4
+// LLVM:   store i32 %[[ARG_A:.*]], ptr %[[A_ADDR]], align 4
+// LLVM:   br label %[[IF_COND:.*]]
+// LLVM: [[IF_COND]]:
+// LLVM:   br i1 true, label %[[IF_THEN:.*]], label %[[IF_END:.*]]
+// LLVM: [[IF_THEN]]:
+// LLVM:   %[[TMP_A_1:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// LLVM:   %[[TMP_A_2:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// LLVM:   %[[RESULT:.*]] = add nsw i32 %[[TMP_A_1]], %[[TMP_A_2]]
+// LLVM:   store i32 %[[RESULT]], ptr %[[B_ADDR]], align 4
+// LLVM:   br label %[[IF_END]]
+// LLVM: [[IF_END]]:
+// LLVM:   br label %[[RET:.*]]
+
+// OGCG: %[[A_ADDR:.*]] = alloca i32, align 4
+// OGCG: %[[B_ADDR:.*]] = alloca i32, align 4
+// OGCG: store i32 %[[ARG_A:.*]], ptr %[[A_ADDR]], align 4
+// OGCG: %[[TMP_A_1:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// OGCG: %[[TMP_A_2:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// OGCG: %[[RESULT:.*]] = add nsw i32 %[[TMP_A_1]], %[[TMP_A_2]]
+// OGCG: store i32 %[[RESULT]], ptr %[[B_ADDR]], align 4
+
+void call_function_with_requires_expr() { summable(1); }
+

[github-actions]

🐧 Linux x64 Test Results

• 112923 tests passed
• 4163 tests skipped

✅ The build succeeded and all tests passed.

[andykaylor]

lgtm

[andykaylor]

We should definitely fold that somewhere (though only when optimizations are enabled). I don't like that classic codegen folds this by default at -O0.

I wouldn't expect purely constant conditions like this to be common in the initially generated IR, but after other optimizations it happens fairly often.