[mips] Correct alignment of vectors passed in varargs for the O32 ABI.

Summary:
Vectors are normally 16-byte aligned, however the O32 ABI enforces a
maximum alignment of 8-bytes since the base of the stack is 8-byte aligned.
Previously, this was enforced on the caller side, but not on the callee side.

This fixes the output of OpenCL's printf when given vectors.

Reviewers: atanasyan

Reviewed By: atanasyan

Subscribers: llvm-commits, pekka.jaaskelainen

Differential Revision: http://reviews.llvm.org/D5433

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@218248 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/TargetInfo.cpp b/lib/CodeGen/TargetInfo.cpp
index cca825c1c564f5c583404beeed6f86442a76bf7c..17552a506920ab6a42da2f564b1ad03c9904990e 100644 (file)
--- a/lib/CodeGen/TargetInfo.cpp
+++ b/lib/CodeGen/TargetInfo.cpp
@@ -5691,7 +5691,8 @@ llvm::Value* MipsABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   CGBuilderTy &Builder = CGF.Builder;
   llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP, "ap");
   llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
-  int64_t TypeAlign = getContext().getTypeAlign(Ty) / 8;
+  int64_t TypeAlign =
+      std::min(getContext().getTypeAlign(Ty) / 8, StackAlignInBytes);
   llvm::Type *PTy = llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
   llvm::Value *AddrTyped;
   unsigned PtrWidth = getTarget().getPointerWidth(0);

diff --git a/test/CodeGen/mips-varargs.c b/test/CodeGen/mips-varargs.c
new file mode 100644 (file)
index 0000000..c7a66be
--- /dev/null
+++ b/test/CodeGen/mips-varargs.c
@@ -0,0 +1,53 @@
+// RUN: %clang -target mips-unknown-linux -O3 -S -o - -emit-llvm %s | FileCheck %s -check-prefix=ALL -check-prefix=O32
+// RUN: %clang -target mips64-unknown-linux -O3 -S -o - -emit-llvm -mabi=n32 %s | FileCheck %s -check-prefix=ALL -check-prefix=N32
+// RUN: %clang -target mips64-unknown-linux -O3 -S -o - -emit-llvm %s | FileCheck %s -check-prefix=ALL -check-prefix=N64
+
+#include <stdarg.h>
+
+typedef int v4i32 __attribute__ ((__vector_size__ (16)));
+
+int test_v4i32(char *fmt, ...) {
+  va_list va;
+
+  va_start(va, fmt);
+  v4i32 v = va_arg(va, v4i32);
+  va_end(va);
+
+  return v[0];
+}
+
+// ALL: define i32 @test_v4i32(i8*{{.*}} %fmt, ...)
+// ALL: entry:
+//
+// O32:   %va = alloca i8*, align [[PTRALIGN:4]]
+// N32:   %va = alloca i8*, align [[PTRALIGN:4]]
+// N64:   %va = alloca i8*, align [[PTRALIGN:8]]
+//
+// ALL:   %va1 = bitcast i8** %va to i8*
+// ALL:   call void @llvm.va_start(i8* %va1)
+// ALL:   %ap.cur = load i8** %va, align [[PTRALIGN]]
+//
+// O32:   %0 = ptrtoint i8* %ap.cur to [[PTRTYPE:i32]]
+// N32:   %0 = ptrtoint i8* %ap.cur to [[PTRTYPE:i32]]
+// N64:   %0 = ptrtoint i8* %ap.cur to [[PTRTYPE:i64]]
+//
+// Vectors are 16-byte aligned, however the O32 ABI has a maximum alignment of
+// 8-bytes since the base of the stack is 8-byte aligned.
+// O32:   %1 = add i32 %0, 7
+// O32:   %2 = and i32 %1, -8
+//
+// N32:   %1 = add i32 %0, 15
+// N32:   %2 = and i32 %1, -16
+//
+// N64:   %1 = add i64 %0, 15
+// N64:   %2 = and i64 %1, -16
+//
+// ALL:   %3 = inttoptr [[PTRTYPE]] %2 to <4 x i32>*
+// ALL:   %4 = inttoptr [[PTRTYPE]] %2 to i8*
+// ALL:   %ap.next = getelementptr i8* %4, [[PTRTYPE]] 16
+// ALL:   store i8* %ap.next, i8** %va, align [[PTRALIGN]]
+// ALL:   %5 = load <4 x i32>* %3, align 16
+// ALL:   call void @llvm.va_end(i8* %va1)
+// ALL:   %vecext = extractelement <4 x i32> %5, i32 0
+// ALL:   ret i32 %vecext
+// ALL: }