From eaf27d276a043eabf50ec7b234bfcfd62061e7ba Mon Sep 17 00:00:00 2001
From: Anton Korobeynikov <asl@math.spbu.ru>
Date: Thu, 1 Mar 2007 16:29:22 +0000
Subject: [PATCH] Ensure that fastcall'ed function is correctly mangled & stack
 is properly aligned

llvm-svn: 34788
---
 lib/Target/X86/X86AsmPrinter.cpp              |  6 ++---
 lib/Target/X86/X86ISelLowering.cpp            | 27 ++++++++++++-------
 test/CodeGen/X86/fast-cc-callee-pops.ll       |  2 +-
 test/CodeGen/X86/fastcall-correct-mangling.ll |  7 +++++
 4 files changed, 28 insertions(+), 14 deletions(-)
 create mode 100644 test/CodeGen/X86/fastcall-correct-mangling.ll

diff --git a/lib/Target/X86/X86AsmPrinter.cpp b/lib/Target/X86/X86AsmPrinter.cpp
index 0e9171bb711..75003e6b748 100644
--- a/lib/Target/X86/X86AsmPrinter.cpp
+++ b/lib/Target/X86/X86AsmPrinter.cpp
@@ -49,10 +49,8 @@ static X86FunctionInfo calculateFunctionInfo(const Function *F,
 
   for (Function::const_arg_iterator AI = F->arg_begin(), AE = F->arg_end();
        AI != AE; ++AI)
-    Size += TD->getTypeSize(AI->getType());
-
-  // Size should be aligned to DWORD boundary
-  Size = ((Size + 3)/4)*4;
+    // Size should be aligned to DWORD boundary
+    Size += ((TD->getTypeSize(AI->getType()) + 3)/4)*4;
   
   // We're not supporting tooooo huge arguments :)
   Info.setBytesToPopOnReturn((unsigned int)Size);
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index bbabc423ab5..65593ebb9bc 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -30,11 +30,16 @@
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/SelectionDAG.h"
 #include "llvm/CodeGen/SSARegMap.h"
+#include "llvm/Support/CommandLine.h"
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/ADT/StringExtras.h"
 using namespace llvm;
 
+static cl::opt<bool> FastCallAlignStack("x86-fastcc-align-stack", cl::Hidden,
+             cl::desc("Align stack to 8-byte boundary for fastcall function"),
+                                        cl::init(false));
+
 X86TargetLowering::X86TargetLowering(TargetMachine &TM)
   : TargetLowering(TM) {
   Subtarget = &TM.getSubtarget<X86Subtarget>();
@@ -903,11 +908,13 @@ X86TargetLowering::LowerFastCCArguments(SDOperand Op, SelectionDAG &DAG) {
   ArgValues.push_back(Root);
 
   unsigned StackSize = CCInfo.getNextStackOffset();
-  
-  // Make sure the instruction takes 8n+4 bytes to make sure the start of the
-  // arguments and the arguments after the retaddr has been pushed are aligned.
-  if ((StackSize & 7) == 0)
-    StackSize += 4;
+
+  if (FastCallAlignStack) {
+    // Make sure the instruction takes 8n+4 bytes to make sure the start of the
+    // arguments and the arguments after the retaddr has been pushed are aligned.
+    if ((StackSize & 7) == 0)
+      StackSize += 4;
+  }
 
   VarArgsFrameIndex = 0xAAAAAAA;   // fastcc functions can't have varargs.
   RegSaveFrameIndex = 0xAAAAAAA;   // X86-64 only.
@@ -936,10 +943,12 @@ SDOperand X86TargetLowering::LowerFastCCCallTo(SDOperand Op, SelectionDAG &DAG,
   // Get a count of how many bytes are to be pushed on the stack.
   unsigned NumBytes = CCInfo.getNextStackOffset();
 
-  // Make sure the instruction takes 8n+4 bytes to make sure the start of the
-  // arguments and the arguments after the retaddr has been pushed are aligned.
-  if ((NumBytes & 7) == 0)
-    NumBytes += 4;
+  if (FastCallAlignStack) {
+    // Make sure the instruction takes 8n+4 bytes to make sure the start of the
+    // arguments and the arguments after the retaddr has been pushed are aligned.
+    if ((NumBytes & 7) == 0)
+      NumBytes += 4;
+  }
 
   Chain = DAG.getCALLSEQ_START(Chain,DAG.getConstant(NumBytes, getPointerTy()));
 
diff --git a/test/CodeGen/X86/fast-cc-callee-pops.ll b/test/CodeGen/X86/fast-cc-callee-pops.ll
index 73d4e7d0c8c..75e36dce338 100644
--- a/test/CodeGen/X86/fast-cc-callee-pops.ll
+++ b/test/CodeGen/X86/fast-cc-callee-pops.ll
@@ -1,4 +1,4 @@
-; RUN: llvm-upgrade < %s | llvm-as | llc -march=x86 -x86-asm-syntax=intel -mcpu=yonah | grep 'ret 28'
+; RUN: llvm-upgrade < %s | llvm-as | llc -march=x86 -x86-asm-syntax=intel -mcpu=yonah | grep 'ret 20'
 
 ; Check that a fastcc function pops its stack variables before returning.
 
diff --git a/test/CodeGen/X86/fastcall-correct-mangling.ll b/test/CodeGen/X86/fastcall-correct-mangling.ll
new file mode 100644
index 00000000000..bf8833dd2c8
--- /dev/null
+++ b/test/CodeGen/X86/fastcall-correct-mangling.ll
@@ -0,0 +1,7 @@
+; RUN: llvm-upgrade < %s | llvm-as | llc -march=x86 -mtriple=mingw32 | grep '@12'
+
+; Check that a fastcall function gets correct mangling
+
+x86_fastcallcc void %func(long %X, ubyte %Y, ubyte %G, ushort %Z) {
+	ret void
+}