From 37196de31e6eab348030ef08cf2c3a78af42ee83 Mon Sep 17 00:00:00 2001
From: Benjamin Kramer <benny.kra@googlemail.com>
Date: Sat, 17 Nov 2012 17:30:55 +0000
Subject: [PATCH] Enable inlining of 4 byte atomic ops on ppc32, 8 byte atomic
 ops on ppc64.

Also fixes a bit/byte mismatch when checking if a target supports atomic ops of a certain size.

llvm-svn: 168260
---
 clang/lib/Basic/Targets.cpp      |  7 ++++++-
 clang/lib/CodeGen/CGExpr.cpp     |  9 ++++-----
 clang/test/CodeGen/ppc-atomics.c | 35 +++++++++++++++++++++++++++++++++++
 3 files changed, 45 insertions(+), 6 deletions(-)
 create mode 100644 clang/test/CodeGen/ppc-atomics.c

diff --git a/clang/lib/Basic/Targets.cpp b/clang/lib/Basic/Targets.cpp
index 305dc7c..95f7d58 100644
--- a/clang/lib/Basic/Targets.cpp
+++ b/clang/lib/Basic/Targets.cpp
@@ -1037,6 +1037,9 @@ public:
       LongDoubleWidth = LongDoubleAlign = 64;
       LongDoubleFormat = &llvm::APFloat::IEEEdouble;
     }
+
+    // PPC32 supports atomics up to 4 bytes.
+    MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 32;
   }
 
   virtual BuiltinVaListKind getBuiltinVaListKind() const {
@@ -1065,7 +1068,9 @@ public:
       DescriptionString = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-"
                           "i64:64:64-f32:32:32-f64:64:64-f128:128:128-"
                           "v128:128:128-n32:64";
-    
+
+    // PPC64 supports atomics up to 8 bytes.
+    MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
   }
   virtual BuiltinVaListKind getBuiltinVaListKind() const {
     return TargetInfo::CharPtrBuiltinVaList;
diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
index 393d3b9..cb274fc4 100644
--- a/clang/lib/CodeGen/CGExpr.cpp
+++ b/clang/lib/CodeGen/CGExpr.cpp
@@ -3166,11 +3166,10 @@ RValue CodeGenFunction::EmitAtomicExpr(AtomicExpr *E, llvm::Value *Dest) {
   uint64_t Size = sizeChars.getQuantity();
   CharUnits alignChars = getContext().getTypeAlignInChars(AtomicTy);
   unsigned Align = alignChars.getQuantity();
-  unsigned MaxInlineWidth =
-      getContext().getTargetInfo().getMaxAtomicInlineWidth();
-  bool UseLibcall = (Size != Align || Size > MaxInlineWidth);
-
-
+  unsigned MaxInlineWidthInBits =
+    getContext().getTargetInfo().getMaxAtomicInlineWidth();
+  bool UseLibcall = (Size != Align ||
+                     getContext().toBits(sizeChars) > MaxInlineWidthInBits);
 
   llvm::Value *Ptr, *Order, *OrderFail = 0, *Val1 = 0, *Val2 = 0;
   Ptr = EmitScalarExpr(E->getPtr());
diff --git a/clang/test/CodeGen/ppc-atomics.c b/clang/test/CodeGen/ppc-atomics.c
new file mode 100644
index 0000000..3fcb0fb
--- /dev/null
+++ b/clang/test/CodeGen/ppc-atomics.c
@@ -0,0 +1,35 @@
+// RUN: %clang_cc1 -triple powerpc-linux-gnu -emit-llvm %s -o - | FileCheck %s -check-prefix=32
+// RUN: %clang_cc1 -triple powerpc64-linux-gnu -emit-llvm %s -o - | FileCheck %s -check-prefix=64
+
+unsigned char c1, c2;
+unsigned short s1, s2;
+unsigned int i1, i2;
+unsigned long long ll1, ll2;
+
+enum memory_order {
+  memory_order_relaxed,
+  memory_order_consume,
+  memory_order_acquire,
+  memory_order_release,
+  memory_order_acq_rel,
+  memory_order_seq_cst
+};
+
+void test1(void) {
+  (void)__atomic_load(&c1, &c2, memory_order_seq_cst);
+  (void)__atomic_load(&s1, &s2, memory_order_seq_cst);
+  (void)__atomic_load(&i1, &i2, memory_order_seq_cst);
+  (void)__atomic_load(&ll1, &ll2, memory_order_seq_cst);
+
+// 32: define void @test1
+// 32: load atomic i8* @c1 seq_cst
+// 32: load atomic i16* @s1 seq_cst
+// 32: load atomic i32* @i1 seq_cst
+// 32: call void @__atomic_load(i32 8, i8* bitcast (i64* @ll1 to i8*)
+
+// 64: define void @test1
+// 64: load atomic i8* @c1 seq_cst
+// 64: load atomic i16* @s1 seq_cst
+// 64: load atomic i32* @i1 seq_cst
+// 64: load atomic i64* @ll1 seq_cst
+}
-- 
2.7.4