[C2y] Fix _Countof handling of VLAs

[AaronBallman]

It turns out that getVLASize() does not get you the size of a single dimension of the VLA, it gets you the full count of all elements. This caused _Countof to return invalid values on VLA ranks. Now switched to using getVLAElements1D() instead, which only gets a single dimension.

Fixes #141409

[llvmbot]

@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-clang

Author: Aaron Ballman (AaronBallman)

Changes

It turns out that getVLASize() does not get you the size of a single dimension of the VLA, it gets you the full count of all elements. This caused _Countof to return invalid values on VLA ranks. Now switched to using getVLAElements1D() instead, which only gets a single dimension.

Fixes #141409

---

Full diff: https://github.com/llvm/llvm-project/pull/141621.diff

2 Files Affected:

• (modified) clang/lib/CodeGen/CGExprScalar.cpp (+11-5)
• (modified) clang/test/C/C2y/n3369_3.c (+54)

diff --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp
index b9eb12efd6670..0b0428d1e916e 100644
--- a/clang/lib/CodeGen/CGExprScalar.cpp
+++ b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -3574,20 +3574,26 @@ ScalarExprEmitter::VisitUnaryExprOrTypeTraitExpr(
           CGF.EmitIgnoredExpr(E->getArgumentExpr());
         }
 
-        auto VlaSize = CGF.getVLASize(VAT);
-        llvm::Value *size = VlaSize.NumElts;
-
         // For sizeof and __datasizeof, we need to scale the number of elements
         // by the size of the array element type. For _Countof, we just want to
         // return the size directly.
+        llvm::Value *Size;
         if (Kind != UETT_CountOf) {
+          auto VlaSize = CGF.getVLASize(VAT);
+          Size = VlaSize.NumElts;
+
           // Scale the number of non-VLA elements by the non-VLA element size.
           CharUnits eltSize = CGF.getContext().getTypeSizeInChars(VlaSize.Type);
           if (!eltSize.isOne())
-            size = CGF.Builder.CreateNUWMul(CGF.CGM.getSize(eltSize), size);
+            Size = CGF.Builder.CreateNUWMul(CGF.CGM.getSize(eltSize), Size);
+        } else {
+          // For _Countof, we don't want the size of the entire VLA, just the
+          // single dimension we're currently looking at.
+          auto VlaSize = CGF.getVLAElements1D(VAT);
+          Size = VlaSize.NumElts;
         }
 
-        return size;
+        return Size;
       }
     }
   } else if (E->getKind() == UETT_OpenMPRequiredSimdAlign) {
diff --git a/clang/test/C/C2y/n3369_3.c b/clang/test/C/C2y/n3369_3.c
index d3909b2e5bc92..792c076413372 100644
--- a/clang/test/C/C2y/n3369_3.c
+++ b/clang/test/C/C2y/n3369_3.c
@@ -164,3 +164,57 @@ void test4(int n) {
   int y = _Countof(int[n++][7]);
 }
 
+// CHECK-64-LABEL: define dso_local void @test5(
+// CHECK-64-SAME: ) #[[ATTR0]] {
+// CHECK-64-NEXT:  [[ENTRY:.*:]]
+// CHECK-64-NEXT:    [[J:%.*]] = alloca i32, align
+// CHECK-64-NEXT:    [[SAVED_STACK:%.*]] = alloca ptr, align
+// CHECK-64-NEXT:    [[A:%.*]] = alloca i32, align
+// CHECK-64-NEXT:    [[B:%.*]] = alloca i32, align
+// CHECK-64-NEXT:    [[C:%.*]] = alloca i32, align
+// CHECK-64-NEXT:    [[D:%.*]] = alloca i32, align
+// CHECK-64-NEXT:    store i32 2, ptr [[J]], align
+// CHECK-64-NEXT:    [[TMP0:%.*]] = call ptr @llvm.stacksave.p0()
+// CHECK-64-NEXT:    store ptr [[TMP0]], ptr [[SAVED_STACK]], align
+// CHECK-64-NEXT:    [[VLA:%.*]] = alloca i32, i64 120, align
+// CHECK-64-NEXT:    store i32 1, ptr [[A]], align
+// CHECK-64-NEXT:    store i32 4, ptr [[B]], align
+// CHECK-64-NEXT:    store i32 5, ptr [[C]], align
+// CHECK-64-NEXT:    store i32 6, ptr [[D]], align
+// CHECK-64-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[SAVED_STACK]], align
+// CHECK-64-NEXT:    call void @llvm.stackrestore.p0(ptr [[TMP1]])
+// CHECK-64-NEXT:    ret void
+//
+// CHECK-32-LABEL: define dso_local void @test5(
+// CHECK-32-SAME: ) #[[ATTR0]] {
+// CHECK-32-NEXT:  [[ENTRY:.*:]]
+// CHECK-32-NEXT:    [[J:%.*]] = alloca i32, align
+// CHECK-32-NEXT:    [[SAVED_STACK:%.*]] = alloca ptr, align
+// CHECK-32-NEXT:    [[A:%.*]] = alloca i32, align
+// CHECK-32-NEXT:    [[B:%.*]] = alloca i32, align
+// CHECK-32-NEXT:    [[C:%.*]] = alloca i32, align
+// CHECK-32-NEXT:    [[D:%.*]] = alloca i32, align
+// CHECK-32-NEXT:    store i32 2, ptr [[J]], align
+// CHECK-32-NEXT:    [[TMP0:%.*]] = call ptr @llvm.stacksave.p0()
+// CHECK-32-NEXT:    store ptr [[TMP0]], ptr [[SAVED_STACK]], align
+// CHECK-32-NEXT:    [[VLA:%.*]] = alloca i32, i32 120, align
+// CHECK-32-NEXT:    store i32 1, ptr [[A]], align
+// CHECK-32-NEXT:    store i32 4, ptr [[B]], align
+// CHECK-32-NEXT:    store i32 5, ptr [[C]], align
+// CHECK-32-NEXT:    store i32 6, ptr [[D]], align
+// CHECK-32-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[SAVED_STACK]], align
+// CHECK-32-NEXT:    call void @llvm.stackrestore.p0(ptr [[TMP1]])
+// CHECK-32-NEXT:    ret void
+//
+void test5() {
+  // Ensure we're getting the variable-length dimensions correctly. We were
+  // previously returning the size of all VLA dimensions multiplied together
+  // to get the total element count rather than the element count for a single
+  // array rank. See GH141409
+  const int j = 2;
+  int arr[1][j + 2][j + 3][j + 4];
+  int a = _Countof arr;
+  int b = _Countof *arr;
+  int c = _Countof **arr;
+  int d = _Countof ***arr;
+}

[efriedma-quic]

LGTM