Enables vararg functions that pass all arguments via registers to be optimized into tail-calls when possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131560 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 105 insertions, 7 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index ff4a283..9b985f9 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -2525,16 +2525,29 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
   if (RegInfo->needsStackRealignment(MF))
     return false;
 
-  // Do not sibcall optimize vararg calls unless the call site is not passing
-  // any arguments.
-  if (isVarArg && !Outs.empty())
-    return false;
-
   // Also avoid sibcall optimization if either caller or callee uses struct
   // return semantics.
   if (isCalleeStructRet || isCallerStructRet)
     return false;
 
+  // Do not sibcall optimize vararg calls unless all arguments are passed via
+  // registers
+  if (isVarArg && !Outs.empty()) {
+    SmallVector<CCValAssign, 16> ArgLocs;
+    CCState CCInfo(CalleeCC, isVarArg, getTargetMachine(),
+                   ArgLocs, *DAG.getContext());
+
+    // Allocate shadow area for Win64
+    if (Subtarget->isTargetWin64()) {
+      CCInfo.AllocateStack(32, 8);
+    }
+
+    CCInfo.AnalyzeCallOperands(Outs, CC_X86);
+    for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i)
+      if (!ArgLocs[i].isRegLoc())
+        return false;
+  }
+
   // If the call result is in ST0 / ST1, it needs to be popped off the x87 stack.
   // Therefore if it's not used by the call it is not safe to optimize this into
   // a sibcall.
diff --git a/test/CodeGen/X86/bool-zext.ll b/test/CodeGen/X86/bool-zext.ll
index d2c30c6..e04770e 100644
--- a/test/CodeGen/X86/bool-zext.ll
+++ b/test/CodeGen/X86/bool-zext.ll
@@ -2,7 +2,7 @@
 
 ; CHECK: @bar1
 ; CHECK: movzbl
-; CHECK: callq
+; CHECK: jmp
 define void @bar1(i1 zeroext %v1) nounwind ssp {
 entry:
   %conv = zext i1 %v1 to i32
@@ -12,7 +12,7 @@ entry:
 
 ; CHECK: @bar2
 ; CHECK-NOT: movzbl
-; CHECK: callq
+; CHECK: jmp
 define void @bar2(i8 zeroext %v1) nounwind ssp {
 entry:
   %conv = zext i8 %v1 to i32
diff --git a/test/CodeGen/X86/vararg_tailcall.ll b/test/CodeGen/X86/vararg_tailcall.ll
new file mode 100644
index 0000000..1187afc
--- /dev/null
+++ b/test/CodeGen/X86/vararg_tailcall.ll
@@ -0,0 +1,85 @@
+; RUN: llc < %s -march=x86-64 | FileCheck %s
+
+@.str = private unnamed_addr constant [5 x i8] c"%ld\0A\00"
+@sel = external global i8*
+@sel3 = external global i8*
+@sel4 = external global i8*
+@sel5 = external global i8*
+@sel6 = external global i8*
+@sel7 = external global i8*
+
+; CHECK: @foo
+; CHECK: jmp
+define void @foo(i64 %arg) nounwind optsize ssp noredzone {
+entry:
+  %call = tail call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([5 x i8]* @.str, i64 0, i64 0), i64 %arg) nounwind optsize noredzone
+  ret void
+}
+
+declare i32 @printf(i8*, ...) optsize noredzone
+
+; CHECK: @bar
+; CHECK: jmp
+define void @bar(i64 %arg) nounwind optsize ssp noredzone {
+entry:
+  tail call void @bar2(i8* getelementptr inbounds ([5 x i8]* @.str, i64 0, i64 0), i64 %arg) nounwind optsize noredzone
+  ret void
+}
+
+declare void @bar2(i8*, i64) optsize noredzone
+
+; CHECK: @foo2
+; CHECK: jmp
+define i8* @foo2(i8* %arg) nounwind optsize ssp noredzone {
+entry:
+  %tmp1 = load i8** @sel, align 8, !tbaa !0
+  %call = tail call i8* (i8*, i8*, ...)* @x2(i8* %arg, i8* %tmp1) nounwind optsize noredzone
+  ret i8* %call
+}
+
+declare i8* @x2(i8*, i8*, ...) optsize noredzone
+
+; CHECK: @foo6
+; CHECK: jmp
+define i8* @foo6(i8* %arg1, i8* %arg2) nounwind optsize ssp noredzone {
+entry:
+  %tmp2 = load i8** @sel3, align 8, !tbaa !0
+  %tmp3 = load i8** @sel4, align 8, !tbaa !0
+  %tmp4 = load i8** @sel5, align 8, !tbaa !0
+  %tmp5 = load i8** @sel6, align 8, !tbaa !0
+  %call = tail call i8* (i8*, i8*, i8*, ...)* @x3(i8* %arg1, i8* %arg2, i8* %tmp2, i8* %tmp3, i8* %tmp4, i8* %tmp5) nounwind optsize noredzone
+  ret i8* %call
+}
+
+declare i8* @x3(i8*, i8*, i8*, ...) optsize noredzone
+
+; CHECK: @foo7
+; CHECK: callq
+define i8* @foo7(i8* %arg1, i8* %arg2) nounwind optsize ssp noredzone {
+entry:
+  %tmp2 = load i8** @sel3, align 8, !tbaa !0
+  %tmp3 = load i8** @sel4, align 8, !tbaa !0
+  %tmp4 = load i8** @sel5, align 8, !tbaa !0
+  %tmp5 = load i8** @sel6, align 8, !tbaa !0
+  %tmp6 = load i8** @sel7, align 8, !tbaa !0
+  %call = tail call i8* (i8*, i8*, i8*, i8*, i8*, i8*, i8*, ...)* @x7(i8* %arg1, i8* %arg2, i8* %tmp2, i8* %tmp3, i8* %tmp4, i8* %tmp5, i8* %tmp6) nounwind optsize noredzone
+  ret i8* %call
+}
+
+declare i8* @x7(i8*, i8*, i8*, i8*, i8*, i8*, i8*, ...) optsize noredzone
+
+; CHECK: @foo8
+; CHECK: callq
+define i8* @foo8(i8* %arg1, i8* %arg2) nounwind optsize ssp noredzone {
+entry:
+  %tmp2 = load i8** @sel3, align 8, !tbaa !0
+  %tmp3 = load i8** @sel4, align 8, !tbaa !0
+  %tmp4 = load i8** @sel5, align 8, !tbaa !0
+  %tmp5 = load i8** @sel6, align 8, !tbaa !0
+  %call = tail call i8* (i8*, i8*, i8*, ...)* @x3(i8* %arg1, i8* %arg2, i8* %tmp2, i8* %tmp3, i8* %tmp4, i8* %tmp5, i32 48879, i32 48879) nounwind optsize noredzone
+  ret i8* %call
+}
+
+!0 = metadata !{metadata !"any pointer", metadata !1}
+!1 = metadata !{metadata !"omnipotent char", metadata !2}
+!2 = metadata !{metadata !"Simple C/C++ TBAA", null}