[CIR] Add support for EmbedExpr for ScalarExpr

[AmrDeveloper]

Add support for the EmbedExpr for ScalarExpr

[llvmbot]

@llvm/pr-subscribers-clangir

@llvm/pr-subscribers-clang

Author: Amr Hesham (AmrDeveloper)

Changes

Add support for the EmbedExpr for ScalarExpr

---

Full diff: https://github.com/llvm/llvm-project/pull/172088.diff

2 Files Affected:

• (modified) clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp (+8-2)
• (added) clang/test/CIR/CodeGen/embed-expr.c (+68)

diff --git a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
index e86c62bf7793f..4577b646ea00c 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
@@ -263,8 +263,14 @@ class ScalarExprEmitter : public StmtVisitor<ScalarExprEmitter, mlir::Value> {
     return {};
   }
   mlir::Value VisitEmbedExpr(EmbedExpr *e) {
-    cgf.cgm.errorNYI(e->getSourceRange(), "ScalarExprEmitter: embed");
-    return {};
+    assert(e->getDataElementCount() == 1);
+    auto it = e->begin();
+    QualType qualTy = e->getType();
+    mlir::Type ty = cgf.convertType(qualTy);
+    llvm::APInt value = (*it)->getValue();
+    uint64_t actualValue = qualTy->isSignedIntegerType() ? value.getSExtValue()
+                                                         : value.getZExtValue();
+    return builder.getConstInt(cgf.getLoc(e->getExprLoc()), ty, actualValue);
   }
   mlir::Value VisitOpaqueValueExpr(OpaqueValueExpr *e) {
     if (e->isGLValue())
diff --git a/clang/test/CIR/CodeGen/embed-expr.c b/clang/test/CIR/CodeGen/embed-expr.c
new file mode 100644
index 0000000000000..24009e3732736
--- /dev/null
+++ b/clang/test/CIR/CodeGen/embed-expr.c
@@ -0,0 +1,68 @@
+// RUN: %clang_cc1 -std=c23 -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=c23 -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=c23 -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
+
+void embed_expr_on_scalar_with_constants() {
+  int a[3] = {
+      1,
+      2,
+#embed __FILE__
+  };
+}
+
+// CIR: %[[A_ADDR:.*]] = cir.alloca !cir.array<!s32i x 3>, !cir.ptr<!cir.array<!s32i x 3>>, ["a", init]
+// CIR: %[[ARRAY:.*]] = cir.const #cir.const_array<[#cir.int<1> : !s32i, #cir.int<2> : !s32i, #cir.int<47> : !s32i]> : !cir.array<!s32i x 3>
+// CIR: cir.store {{.*}} %[[ARRAY]], %[[A_ADDR]] : !cir.array<!s32i x 3>, !cir.ptr<!cir.array<!s32i x 3>>
+
+// LLVM: %[[A_ADDR:.*]] = alloca [3 x i32], i64 1, align 4
+// LLVM: store [3 x i32] [i32 1, i32 2, i32 47], ptr %[[A_ADDR]], align 4
+
+// OGCG: %[[A_ADDR:.*]] = alloca [3 x i32], align 4
+// OGCG: call void @llvm.memcpy.p0.p0.i64(ptr align 4 %[[A_ADDR]], ptr align 4 @__const.embed_expr_on_scalar_with_constants.a, i64 12, i1 false)
+
+void embed_expr_on_scalar_with_non_constants() {
+  int a;
+  int b[3] = {
+      a,
+      a,
+#embed __FILE__
+  };
+}
+
+// CIR: %[[A_ADDR:.*]] = cir.alloca !s32i, !cir.ptr<!s32i>, ["a"]
+// CIR: %[[B_ADDR:.*]] = cir.alloca !cir.array<!s32i x 3>, !cir.ptr<!cir.array<!s32i x 3>>, ["b", init]
+// CIR: %[[B_PTR:.*]] = cir.cast array_to_ptrdecay %[[B_ADDR]] : !cir.ptr<!cir.array<!s32i x 3>> -> !cir.ptr<!s32i>
+// CIR: %[[TMP_A:.*]] = cir.load {{.*}} %[[A_ADDR]] : !cir.ptr<!s32i>, !s32i
+// CIR: cir.store {{.*}} %[[TMP_A]], %[[B_PTR]] : !s32i, !cir.ptr<!s32i>
+// CIR: %[[CONST_1:.*]] = cir.const #cir.int<1> : !s64i
+// CIR: %[[B_ELEM_1_PTR:.*]] = cir.ptr_stride %[[B_PTR]], %[[CONST_1]] : (!cir.ptr<!s32i>, !s64i) -> !cir.ptr<!s32i>
+// CIR: %[[TMP_A:.*]] = cir.load {{.*}} %[[A_ADDR]] : !cir.ptr<!s32i>, !s32i
+// CIR: cir.store {{.*}} %[[TMP_A]], %[[B_ELEM_1_PTR]] : !s32i, !cir.ptr<!s32i>
+// CIR: %[[CONST_2:.*]] = cir.const #cir.int<2> : !s64i
+// CIR: %[[B_ELEM_2_PTR:.*]] = cir.ptr_stride %[[B_PTR]], %[[CONST_2]] : (!cir.ptr<!s32i>, !s64i) -> !cir.ptr<!s32i>
+// CIR: %[[CONST_47:.*]] = cir.const #cir.int<47> : !s32i
+// CIR: cir.store {{.*}} %[[CONST_47]], %[[B_ELEM_2_PTR]] : !s32i, !cir.ptr<!s32i>
+
+// LLVM: %[[A_ADDR:.*]] = alloca i32, i64 1, align 4
+// LLVM: %[[B_ADDR:.*]] = alloca [3 x i32], i64 1, align 4
+// LLVM: %[[B_ELEM_0_PTR:.*]] = getelementptr i32, ptr %[[B_ADDR]], i32 0
+// LLVM: %[[TMP_A:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// LLVM: store i32 %[[TMP_A]], ptr %[[B_ELEM_0_PTR]], align 4
+// LLVM: %[[B_ELEM_1_PTR:.*]] = getelementptr i32, ptr %[[B_ELEM_0_PTR]], i64 1
+// LLVM: %[[TMP_A:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// LLVM: store i32 %[[TMP_A]], ptr %[[B_ELEM_1_PTR]], align 4
+// LLVM: %[[B_ELEM_2_PTR:.*]] = getelementptr i32, ptr %[[B_ELEM_0_PTR]], i64 2
+// LLVM: store i32 47, ptr %[[B_ELEM_2_PTR]], align 4
+
+// OGCG: %[[A_ADDR:.*]] = alloca i32, align 4
+// OGCG: %[[B_ADDR:.*]] = alloca [3 x i32], align 4
+// OGCG: %[[TMP_A:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// OGCG: store i32 %[[TMP_A]], ptr %[[B_ADDR]], align 4
+// OGCG: %[[B_ELEM_1_PTR:.*]] = getelementptr inbounds i32, ptr %[[B_ADDR]], i64 1
+// OGCG: %[[TMP_A:.*]] = load i32, ptr %[[A_ADDR]], align 4
+// OGCG: store i32 %[[TMP_A]], ptr %[[B_ELEM_1_PTR]], align 4
+// OGCG: %[[B_ELEM_2_PTR:.*]] = getelementptr inbounds i32, ptr %[[B_ADDR]], i64 2
+// OGCG: store i32 47, ptr %[[B_ELEM_2_PTR]], align 4

[AmrDeveloper]

Identically, we should use getConstInt and pass the APInt, but in our builder, when we pass APInt, we just construct APSInt with unsigned = true, so maybe we should add another parameter to this helper to control singing

llvm-project/clang/lib/CIR/CodeGen/CIRGenBuilder.cpp

Lines 55 to 58 in 333ee93

	cir::ConstantOp CIRGenBuilderTy::getConstInt(mlir::Location loc,
	llvm::APInt intVal) {
	return getConstInt(loc, llvm::APSInt(intVal));
	}

[andykaylor]

Yes, I think return builder.getConstInt(cgf.getLoc(e->getExprLoc()), value, qualTy->isSignedIntegerType()); would make more sense here if the APInt overload for getConstInt took a isSigned argument. Then we could get rid of a lot of the type handling here.

[AmrDeveloper]

In getConstInt, we pass llvm::APInt always by copy, but I will keep this change to NFC PR :D

[andykaylor]

lgtm