[Clang][Sema] Add special handling of mfloat8 in initializer lists

[Lukacma]

This patch fixes assertion failures in clang, caused by unique properties of _mfp8 type, namely it not being either scalar or vector type and it not being either integer or float type.

[llvmbot]

@llvm/pr-subscribers-clang

Author: None (Lukacma)

Changes

This patch fixes assertion failures in clang, caused by unique properties of _mfp8 type, namely it not being either scalar or vector type and it not being either integer or float type.

---

Full diff: https://github.com/llvm/llvm-project/pull/125097.diff

3 Files Affected:

• (modified) clang/lib/AST/ExprConstant.cpp (+5)
• (modified) clang/lib/Sema/SemaInit.cpp (+1-1)
• (added) clang/test/CodeGen/AArch64/fp8-init-list.c (+59)

diff --git a/clang/lib/AST/ExprConstant.cpp b/clang/lib/AST/ExprConstant.cpp
index 0e41e3dbc8a32a..e11b92956b411d 100644
--- a/clang/lib/AST/ExprConstant.cpp
+++ b/clang/lib/AST/ExprConstant.cpp
@@ -11172,6 +11172,11 @@ VectorExprEvaluator::VisitInitListExpr(const InitListExpr *E) {
   QualType EltTy = VT->getElementType();
   SmallVector<APValue, 4> Elements;
 
+  // MFloat8 type doesn't have constants and thus constant folding 
+  // is impossible.
+  if (EltTy->isMFloat8Type())
+    return false;
+
   // The number of initializers can be less than the number of
   // vector elements. For OpenCL, this can be due to nested vector
   // initialization. For GCC compatibility, missing trailing elements
diff --git a/clang/lib/Sema/SemaInit.cpp b/clang/lib/Sema/SemaInit.cpp
index b95cbbf4222056..505d9df8d44feb 100644
--- a/clang/lib/Sema/SemaInit.cpp
+++ b/clang/lib/Sema/SemaInit.cpp
@@ -1589,7 +1589,7 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
 
   } else {
     assert((ElemType->isRecordType() || ElemType->isVectorType() ||
-            ElemType->isOpenCLSpecificType()) && "Unexpected type");
+            ElemType->isOpenCLSpecificType() || ElemType->isMFloat8Type()) && "Unexpected type");
 
     // C99 6.7.8p13:
     //
diff --git a/clang/test/CodeGen/AArch64/fp8-init-list.c b/clang/test/CodeGen/AArch64/fp8-init-list.c
new file mode 100644
index 00000000000000..8b4b31a71c46a2
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/fp8-init-list.c
@@ -0,0 +1,59 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+
+// RUN: %clang_cc1        -triple aarch64-none-linux-gnu -target-feature +neon -O2 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +neon -O2 -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
+
+// RUN: %clang_cc1        -triple aarch64-none-linux-gnu -target-feature +neon -O2 -Werror -Wall -S -o /dev/null %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_neon.h>
+
+// CHECK-LABEL: define dso_local <8 x i8> @vector_init_test(
+// CHECK-SAME: <1 x i8> [[X:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[VECINIT14:%.*]] = shufflevector <1 x i8> [[X]], <1 x i8> poison, <8 x i32> zeroinitializer
+// CHECK-NEXT:    ret <8 x i8> [[VECINIT14]]
+//
+// CHECK-CXX-LABEL: define dso_local <8 x i8> @_Z16vector_init_testu6__mfp8(
+// CHECK-CXX-SAME: <1 x i8> [[X:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-NEXT:  [[ENTRY:.*:]]
+// CHECK-CXX-NEXT:    [[VECINIT14:%.*]] = shufflevector <1 x i8> [[X]], <1 x i8> poison, <8 x i32> zeroinitializer
+// CHECK-CXX-NEXT:    ret <8 x i8> [[VECINIT14]]
+//
+mfloat8x8_t vector_init_test(__mfp8 x) {
+   return (mfloat8x8_t) {x, x, x, x, x, x, x, x};
+}
+
+struct S {
+    __mfp8 x;
+};
+
+struct S s;
+
+// CHECK-LABEL: define dso_local void @f(
+// CHECK-SAME: <1 x i8> [[X:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    store <1 x i8> [[X]], ptr @s, align 1, !tbaa [[TBAA2:![0-9]+]]
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z1fu6__mfp8(
+// CHECK-CXX-SAME: <1 x i8> [[X:%.*]]) local_unnamed_addr #[[ATTR1:[0-9]+]] {
+// CHECK-CXX-NEXT:  [[ENTRY:.*:]]
+// CHECK-CXX-NEXT:    store <1 x i8> [[X]], ptr @s, align 1, !tbaa [[TBAA2:![0-9]+]]
+// CHECK-CXX-NEXT:    ret void
+//
+void f(__mfp8 x) {
+    s = (struct S){x};
+}
+//.
+// CHECK: [[TBAA2]] = !{[[META3:![0-9]+]], [[META3]], i64 0}
+// CHECK: [[META3]] = !{!"__mfp8", [[META4:![0-9]+]], i64 0}
+// CHECK: [[META4]] = !{!"omnipotent char", [[META5:![0-9]+]], i64 0}
+// CHECK: [[META5]] = !{!"Simple C/C++ TBAA"}
+//.
+// CHECK-CXX: [[TBAA2]] = !{[[META3:![0-9]+]], [[META3]], i64 0}
+// CHECK-CXX: [[META3]] = !{!"__mfp8", [[META4:![0-9]+]], i64 0}
+// CHECK-CXX: [[META4]] = !{!"omnipotent char", [[META5:![0-9]+]], i64 0}
+// CHECK-CXX: [[META5]] = !{!"Simple C++ TBAA"}
+//.

[efriedma-quic]

Please don't mark this "NFC"; I'd consider this a functional change, even if it doesn't have any obvious effect on non-asserts builds.

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[tbaederr]

Changes seem fine once the code formatting is fixed.

[Lukacma]

Ping

[jthackray]

LGTM