[clang] [Clang] Fix incorrect type for `__mfp8` in `extractelement` codegen (PR #192977)

[via cfe-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-clang-codegen

Author: Momchil Velikov (momchil-velikov)

<details>
<summary>Changes</summary>

The codegen for extracting an element from an FP8 vector was emitting a simple `extractelement` with `i8` type for the extracted element. The `__mfp8` type is represented as `<1 x i8>` in LLVM IR. This codegen created inconsistency in Clang - some `__mfp8` expressions would correspond to LLVM IR values with `<1 x i8>` type and some to `i8` type.

It also caused an assertion failure when the extracted element was passed as a function argument.

This patch fixes the issue by bitcasting the extracted element to `<1 x i8>`.

---
Full diff: https://github.com/llvm/llvm-project/pull/192977.diff


3 Files Affected:

- (modified) clang/lib/CodeGen/CGExprScalar.cpp (+8-1) 
- (added) clang/test/CodeGen/AArch64/fp8-extract.c (+33) 
- (modified) clang/test/CodeGen/arm-mfp8.c (+4-8) 


``````````diff
diff --git a/clang/lib/CodeGen/CGExprScalar.cpp b/clang/lib/CodeGen/CGExprScalar.cpp
index a8dcf22992983..f8997f0503491 100644
--- a/clang/lib/CodeGen/CGExprScalar.cpp
+++ b/clang/lib/CodeGen/CGExprScalar.cpp
@@ -2213,7 +2213,14 @@ Value *ScalarExprEmitter::VisitArraySubscriptExpr(ArraySubscriptExpr *E) {
   if (CGF.SanOpts.has(SanitizerKind::ArrayBounds))
     CGF.EmitBoundsCheck(E, E->getBase(), Idx, IdxTy, /*Accessed*/true);
 
-  return Builder.CreateExtractElement(Base, Idx, "vecext");
+  Value *Ret = Builder.CreateExtractElement(Base, Idx, "vecext");
+
+  // Even being a scalar the `__mfp8` type corresponds to `<1 x i8>` in LLVM IR.
+  // Cast the extracted element to the vector type to keep it consistent in
+  // Clang.
+  if (E->getType()->isMFloat8Type())
+    Ret = Builder.CreateBitCast(Ret, ConvertType(E->getType()), "mfp8ext");
+  return Ret;
 }
 
 Value *ScalarExprEmitter::VisitMatrixSingleSubscriptExpr(
diff --git a/clang/test/CodeGen/AArch64/fp8-extract.c b/clang/test/CodeGen/AArch64/fp8-extract.c
new file mode 100644
index 0000000000000..c0c113600de63
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/fp8-extract.c
@@ -0,0 +1,33 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 6
+// RUN: %clang_cc1        -triple aarch64-none-linux-gnu -target-feature +neon -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg | FileCheck %s
+// RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +neon -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg | FileCheck %s -check-prefix CHECK-CXX
+
+// RUN: %clang_cc1        -triple aarch64-none-linux-gnu -target-feature +neon -disable-O0-optnone -Werror -Wall -S -o /dev/null %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_neon.h>
+
+// Test for a n internal compiler error when extracting an element from an FP8
+// vector and passing it to a function.
+
+// CHECK-LABEL: define dso_local void @f(
+// CHECK-SAME: <16 x i8> [[V:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[VECEXT:%.*]] = extractelement <16 x i8> [[V]], i32 0
+// CHECK-NEXT:    [[MFP8EXT:%.*]] = bitcast i8 [[VECEXT]] to <1 x i8>
+// CHECK-NEXT:    call void @g(<1 x i8> [[MFP8EXT]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z1f14__Mfloat8x16_t(
+// CHECK-CXX-SAME: <16 x i8> [[V:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-NEXT:  [[ENTRY:.*:]]
+// CHECK-CXX-NEXT:    [[VECEXT:%.*]] = extractelement <16 x i8> [[V]], i32 0
+// CHECK-CXX-NEXT:    [[MFP8EXT:%.*]] = bitcast i8 [[VECEXT]] to <1 x i8>
+// CHECK-CXX-NEXT:    call void @_Z1gu6__mfp8(<1 x i8> [[MFP8EXT]])
+// CHECK-CXX-NEXT:    ret void
+//
+void f(__Mfloat8x16_t v) {
+  void g(__mfp8);
+  g(v[0]);
+}
diff --git a/clang/test/CodeGen/arm-mfp8.c b/clang/test/CodeGen/arm-mfp8.c
index 9385b537f18b3..f99c865e99a95 100644
--- a/clang/test/CodeGen/arm-mfp8.c
+++ b/clang/test/CodeGen/arm-mfp8.c
@@ -64,20 +64,16 @@ __mfp8 func1n(__mfp8 mfp8) {
 // CHECK-C-LABEL: define dso_local <1 x i8> @test_extract_element(
 // CHECK-C-SAME: <16 x i8> [[X:%.*]], i32 noundef [[I:%.*]]) #[[ATTR0]] {
 // CHECK-C-NEXT:  [[ENTRY:.*:]]
-// CHECK-C-NEXT:    [[RETVAL:%.*]] = alloca <1 x i8>, align 1
 // CHECK-C-NEXT:    [[VECEXT:%.*]] = extractelement <16 x i8> [[X]], i32 [[I]]
-// CHECK-C-NEXT:    store i8 [[VECEXT]], ptr [[RETVAL]], align 1
-// CHECK-C-NEXT:    [[TMP0:%.*]] = load <1 x i8>, ptr [[RETVAL]], align 1
-// CHECK-C-NEXT:    ret <1 x i8> [[TMP0]]
+// CHECK-C-NEXT:    [[MFP8EXT:%.*]] = bitcast i8 [[VECEXT]] to <1 x i8>
+// CHECK-C-NEXT:    ret <1 x i8> [[MFP8EXT]]
 //
 // CHECK-CXX-LABEL: define dso_local <1 x i8> @_Z20test_extract_element14__Mfloat8x16_ti(
 // CHECK-CXX-SAME: <16 x i8> [[X:%.*]], i32 noundef [[I:%.*]]) #[[ATTR0]] {
 // CHECK-CXX-NEXT:  [[ENTRY:.*:]]
-// CHECK-CXX-NEXT:    [[RETVAL:%.*]] = alloca <1 x i8>, align 1
 // CHECK-CXX-NEXT:    [[VECEXT:%.*]] = extractelement <16 x i8> [[X]], i32 [[I]]
-// CHECK-CXX-NEXT:    store i8 [[VECEXT]], ptr [[RETVAL]], align 1
-// CHECK-CXX-NEXT:    [[TMP0:%.*]] = load <1 x i8>, ptr [[RETVAL]], align 1
-// CHECK-CXX-NEXT:    ret <1 x i8> [[TMP0]]
+// CHECK-CXX-NEXT:    [[MFP8EXT:%.*]] = bitcast i8 [[VECEXT]] to <1 x i8>
+// CHECK-CXX-NEXT:    ret <1 x i8> [[MFP8EXT]]
 //
 mfloat8_t test_extract_element(mfloat8x16_t x, int i) {
   return x[i];

``````````

</details>


https://github.com/llvm/llvm-project/pull/192977