[AArch64] Use ptrmask for vaarg stack alignment

[JonChesterfield]

No description provided.

[llvmbot]

@llvm/pr-subscribers-clang
@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-backend-aarch64

Author: Jon Chesterfield (JonChesterfield)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/92836.diff

2 Files Affected:

• (modified) clang/lib/CodeGen/Targets/AArch64.cpp (+1-12)
• (modified) clang/test/CodeGen/aarch64-varargs.c (+10-20)

diff --git a/clang/lib/CodeGen/Targets/AArch64.cpp b/clang/lib/CodeGen/Targets/AArch64.cpp
index 0a4711fb2170d..9aa3ea75681b3 100644
--- a/clang/lib/CodeGen/Targets/AArch64.cpp
+++ b/clang/lib/CodeGen/Targets/AArch64.cpp
@@ -750,18 +750,7 @@ Address AArch64ABIInfo::EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
   // Again, stack arguments may need realignment. In this case both integer and
   // floating-point ones might be affected.
   if (!IsIndirect && TyAlign.getQuantity() > 8) {
-    int Align = TyAlign.getQuantity();
-
-    OnStackPtr = CGF.Builder.CreatePtrToInt(OnStackPtr, CGF.Int64Ty);
-
-    OnStackPtr = CGF.Builder.CreateAdd(
-        OnStackPtr, llvm::ConstantInt::get(CGF.Int64Ty, Align - 1),
-        "align_stack");
-    OnStackPtr = CGF.Builder.CreateAnd(
-        OnStackPtr, llvm::ConstantInt::get(CGF.Int64Ty, -Align),
-        "align_stack");
-
-    OnStackPtr = CGF.Builder.CreateIntToPtr(OnStackPtr, CGF.Int8PtrTy);
+    OnStackPtr = emitRoundPointerUpToAlignment(CGF, OnStackPtr, TyAlign);
   }
   Address OnStackAddr = Address(OnStackPtr, CGF.Int8Ty,
                                 std::max(CharUnits::fromQuantity(8), TyAlign));
diff --git a/clang/test/CodeGen/aarch64-varargs.c b/clang/test/CodeGen/aarch64-varargs.c
index ee4e88eda4ef4..8952d6980a8d3 100644
--- a/clang/test/CodeGen/aarch64-varargs.c
+++ b/clang/test/CodeGen/aarch64-varargs.c
@@ -63,10 +63,8 @@ __int128 aligned_int(void) {
 
 // CHECK: [[VAARG_ON_STACK]]
 // CHECK: [[STACK:%[a-z_0-9]+]] = load ptr, ptr @the_list
-// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint ptr [[STACK]] to i64
-// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
-// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
-// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to ptr
+// CHECK: [[STACKINC:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[STACK]], i32 15
+// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9.]+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[STACKINC]], i64 -16)
 // CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[ALIGNED_STACK_PTR]], i64 16
 // CHECK: store ptr [[NEW_STACK]], ptr @the_list
 // CHECK: br label %[[VAARG_END]]
@@ -377,10 +375,8 @@ underaligned_int128 underaligned_int128_test(void) {
 
 // CHECK: [[VAARG_ON_STACK]]
 // CHECK: [[STACK:%[a-z_0-9]+]] = load ptr, ptr @the_list
-// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint ptr [[STACK]] to i64
-// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
-// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
-// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to ptr
+// CHECK: [[STACKINC:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[STACK]], i32 15
+// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9.]+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[STACKINC]], i64 -16)
 // CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[ALIGNED_STACK_PTR]], i64 16
 // CHECK: store ptr [[NEW_STACK]], ptr @the_list
 // CHECK: br label %[[VAARG_END]]
@@ -414,10 +410,8 @@ overaligned_int128 overaligned_int128_test(void) {
 
 // CHECK: [[VAARG_ON_STACK]]
 // CHECK: [[STACK:%[a-z_0-9]+]] = load ptr, ptr @the_list
-// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint ptr [[STACK]] to i64
-// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
-// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
-// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to ptr
+// CHECK: [[STACKINC:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[STACK]], i32 15
+// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9.]+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[STACKINC]], i64 -16)
 // CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[ALIGNED_STACK_PTR]], i64 16
 // CHECK: store ptr [[NEW_STACK]], ptr @the_list
 // CHECK: br label %[[VAARG_END]]
@@ -688,10 +682,8 @@ overaligned_int_struct_member overaligned_int_struct_member_test(void) {
 
 // CHECK: [[VAARG_ON_STACK]]
 // CHECK: [[STACK:%[a-z_0-9]+]] = load ptr, ptr @the_list
-// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint ptr [[STACK]] to i64
-// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
-// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
-// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to ptr
+// CHECK: [[STACKINC:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[STACK]], i32 15
+// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9.]+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[STACKINC]], i64 -16)
 // CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[ALIGNED_STACK_PTR]], i64 16
 // CHECK: store ptr [[NEW_STACK]], ptr @the_list
 // CHECK: br label %[[VAARG_END]]
@@ -756,10 +748,8 @@ overaligned_long_long_struct_member overaligned_long_long_struct_member_test(voi
 
 // CHECK: [[VAARG_ON_STACK]]
 // CHECK: [[STACK:%[a-z_0-9]+]] = load ptr, ptr @the_list
-// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint ptr [[STACK]] to i64
-// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
-// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
-// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to ptr
+// CHECK: [[STACKINC:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[STACK]], i32 15
+// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9.]+]] = call ptr @llvm.ptrmask.p0.i64(ptr [[STACKINC]], i64 -16)
 // CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, ptr [[ALIGNED_STACK_PTR]], i64 16
 // CHECK: store ptr [[NEW_STACK]], ptr @the_list
 // CHECK: br label %[[VAARG_END]]

[efriedma-quic]

Please don't use "NFC" markings for patches that change the emitted IR, even if it doesn't significantly impact the resulting assembly in most cases.

That said, LGTM

[JonChesterfield]

Dropped the NFC tag from the commit message. The machine code is expected to be the same or negligibly better, this patch makes some tests more legible for a different patch. Thanks for the quick review!