Fix a fastcc + sret bug. If fastcc and sret, callee doesn't need to pop the hidden struct ptr; Re-enable fastcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@56061 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 50 insertions, 2 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index f30490a..2ddddf6 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1103,6 +1103,8 @@ CCAssignFn *X86TargetLowering::CCAssignFnForNode(SDValue Op) const {
     return CC_X86_32_FastCall;
   else if (CC == CallingConv::Fast && PerformTailCallOpt)
     return CC_X86_32_TailCall;
+  else if (CC == CallingConv::Fast)
+    return CC_X86_32_FastCC;
   else
     return CC_X86_32_C;
 }
@@ -1391,7 +1393,7 @@ X86TargetLowering::LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG) {
   } else {
     BytesToPopOnReturn  = 0; // Callee pops nothing.
     // If this is an sret function, the return should pop the hidden pointer.
-    if (!Is64Bit && ArgsAreStructReturn(Op))
+    if (!Is64Bit && CC != CallingConv::Fast && ArgsAreStructReturn(Op))
       BytesToPopOnReturn = 4;  
     BytesCallerReserves = StackSize;
   }
@@ -1773,7 +1775,7 @@ SDValue X86TargetLowering::LowerCALL(SDValue Op, SelectionDAG &DAG) {
   unsigned NumBytesForCalleeToPush;
   if (IsCalleePop(Op))
     NumBytesForCalleeToPush = NumBytes;    // Callee pops everything
-  else if (!Is64Bit && IsStructRet)
+  else if (!Is64Bit && CC != CallingConv::Fast && IsStructRet)
     // If this is is a call to a struct-return function, the callee
     // pops the hidden struct pointer, so we have to push it back.
     // This is common for Darwin/X86, Linux & Mingw32 targets.
diff --git a/test/CodeGen/X86/fastcc-sret.ll b/test/CodeGen/X86/fastcc-sret.ll
new file mode 100644
index 0000000..b5338d7
--- /dev/null
+++ b/test/CodeGen/X86/fastcc-sret.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=x86 | grep ret | not grep 4
+
+	%struct.foo = type { [4 x i32] }
+
+define fastcc void @bar(%struct.foo* noalias sret %agg.result) nounwind  {
+entry:
+	%tmp1 = getelementptr %struct.foo* %agg.result, i32 0, i32 0
+	%tmp3 = getelementptr [4 x i32]* %tmp1, i32 0, i32 0
+	store i32 1, i32* %tmp3, align 8
+        ret void
+}
+
+@dst = external global i32
+
+define void @foo() nounwind {
+	%memtmp = alloca %struct.foo, align 4
+        call fastcc void @bar( %struct.foo* sret %memtmp ) nounwind
+        %tmp4 = getelementptr %struct.foo* %memtmp, i32 0, i32 0
+	%tmp5 = getelementptr [4 x i32]* %tmp4, i32 0, i32 0
+        %tmp6 = load i32* %tmp5
+        store i32 %tmp6, i32* @dst
+        ret void
+}
diff --git a/test/CodeGen/X86/sret.ll b/test/CodeGen/X86/sret.ll
new file mode 100644
index 0000000..30e5af4
--- /dev/null
+++ b/test/CodeGen/X86/sret.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=x86 | grep ret | grep 4
+
+	%struct.foo = type { [4 x i32] }
+
+define void @bar(%struct.foo* noalias sret %agg.result) nounwind  {
+entry:
+	%tmp1 = getelementptr %struct.foo* %agg.result, i32 0, i32 0
+	%tmp3 = getelementptr [4 x i32]* %tmp1, i32 0, i32 0
+	store i32 1, i32* %tmp3, align 8
+        ret void
+}
+
+@dst = external global i32
+
+define void @foo() nounwind {
+	%memtmp = alloca %struct.foo, align 4
+        call void @bar( %struct.foo* sret %memtmp ) nounwind
+        %tmp4 = getelementptr %struct.foo* %memtmp, i32 0, i32 0
+	%tmp5 = getelementptr [4 x i32]* %tmp4, i32 0, i32 0
+        %tmp6 = load i32* %tmp5
+        store i32 %tmp6, i32* @dst
+        ret void
+}