Merge 98212 from mainline.

add support, testcases, and dox for the new GHC calling
convention.  Patch by David Terei!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_27@98307 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 219 insertions, 17 deletions

diff --git a/docs/CodeGenerator.html b/docs/CodeGenerator.html
index 2be31851e36a..ace0ee52cd88 100644
--- a/docs/CodeGenerator.html
+++ b/docs/CodeGenerator.html
@@ -1678,7 +1678,8 @@ $ llc -regalloc=linearscan file.bc -o ln.s;
    supported on x86/x86-64 and PowerPC. It is performed if:</p>
 
 <ul>
-  <li>Caller and callee have the calling convention <tt>fastcc</tt>.</li>
+  <li>Caller and callee have the calling convention <tt>fastcc</tt> or
+       <tt>cc 10</tt> (GHC call convention).</li>
 
   <li>The call is a tail call - in tail position (ret immediately follows call
       and ret uses value of call or is void).</li>
diff --git a/docs/LangRef.html b/docs/LangRef.html
index 7cfa05205cea..755b94498fc9 100644
--- a/docs/LangRef.html
+++ b/docs/LangRef.html
@@ -691,9 +691,9 @@ define i32 @main() {                                        <i>; i32()* </i>
       target, without having to conform to an externally specified ABI
       (Application Binary Interface).
       <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
-      when this convention is used.</a>  This calling convention does not
-      support varargs and requires the prototype of all callees to exactly match
-      the prototype of the function definition.</dd>
+      when this or the GHC convention is used.</a>  This calling convention
+      does not support varargs and requires the prototype of all callees to
+      exactly match the prototype of the function definition.</dd>
 
   <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
   <dd>This calling convention attempts to make code in the caller as efficient
@@ -703,6 +703,26 @@ define i32 @main() {                                        <i>; i32()* </i>
       does not support varargs and requires the prototype of all callees to
       exactly match the prototype of the function definition.</dd>
 
+  <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
+  <dd>This calling convention has been implemented specifically for use by the
+      <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
+      It passes everything in registers, going to extremes to achieve this by
+      disabling callee save registers. This calling convention should not be
+      used lightly but only for specific situations such as an alternative to
+      the <em>register pinning</em> performance technique often used when
+      implementing functional programming languages.At the moment only X86
+      supports this convention and it has the following limitations:
+      <ul>
+        <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
+            floating point types are supported.</li>
+        <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
+            6 floating point parameters.</li>
+      </ul>
+      This calling convention supports
+      <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
+      requires both the caller and callee are using it.
+  </dd>
+
   <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
   <dd>Any calling convention may be specified by number, allowing
       target-specific calling conventions to be used.  Target specific calling
diff --git a/include/llvm/CallingConv.h b/include/llvm/CallingConv.h
index c54527c21877..624390d1cd84 100644
--- a/include/llvm/CallingConv.h
+++ b/include/llvm/CallingConv.h
@@ -44,6 +44,9 @@ namespace CallingConv {
     // call does not break any live ranges in the caller side.
     Cold = 9,
 
+    // GHC - Calling convention used by the Glasgow Haskell Compiler (GHC).
+    GHC = 10,
+
     // Target - This is the start of the target-specific calling conventions,
     // e.g. fastcall and thiscall on X86.
     FirstTargetCC = 64,
diff --git a/lib/Target/X86/X86CallingConv.td b/lib/Target/X86/X86CallingConv.td
index 12d3d04eb6bd..fd15efd6262a 100644
--- a/lib/Target/X86/X86CallingConv.td
+++ b/lib/Target/X86/X86CallingConv.td
@@ -221,6 +221,20 @@ def CC_X86_Win64_C : CallingConv<[
   CCIfType<[v8i8, v4i16, v2i32, v1i64], CCAssignToStack<8, 8>>
 ]>;
 
+def CC_X86_64_GHC : CallingConv<[
+  // Promote i8/i16/i32 arguments to i64.
+  CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
+
+  // Pass in STG registers: Base, Sp, Hp, R1, R2, R3, R4, R5, R6, SpLim
+  CCIfType<[i64],
+            CCAssignToReg<[R13, RBP, R12, RBX, R14, RSI, RDI, R8, R9, R15]>>,
+
+  // Pass in STG registers: F1, F2, F3, F4, D1, D2
+  CCIfType<[f32, f64, v16i8, v8i16, v4i32, v2i64, v4f32, v2f64],
+            CCIfSubtarget<"hasSSE1()",
+            CCAssignToReg<[XMM1, XMM2, XMM3, XMM4, XMM5, XMM6]>>>
+]>;
+
 //===----------------------------------------------------------------------===//
 // X86 C Calling Convention
 //===----------------------------------------------------------------------===//
@@ -320,3 +334,11 @@ def CC_X86_32_FastCC : CallingConv<[
   // Otherwise, same as everything else.
   CCDelegateTo<CC_X86_32_Common>
 ]>;
+
+def CC_X86_32_GHC : CallingConv<[
+  // Promote i8/i16 arguments to i32.
+  CCIfType<[i8, i16], CCPromoteToType<i32>>,
+
+  // Pass in STG registers: Base, Sp, Hp, R1
+  CCIfType<[i32], CCAssignToReg<[EBX, EBP, EDI, ESI]>>
+]>;
diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 98e3f4efe588..27807f29024d 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -172,7 +172,9 @@ bool X86FastISel::isTypeLegal(const Type *Ty, EVT &VT, bool AllowI1) {
 CCAssignFn *X86FastISel::CCAssignFnForCall(CallingConv::ID CC,
                                            bool isTaillCall) {
   if (Subtarget->is64Bit()) {
-    if (Subtarget->isTargetWin64())
+    if (CC == CallingConv::GHC)
+      return CC_X86_64_GHC;
+    else if (Subtarget->isTargetWin64())
       return CC_X86_Win64_C;
     else
       return CC_X86_64_C;
@@ -182,6 +184,8 @@ CCAssignFn *X86FastISel::CCAssignFnForCall(CallingConv::ID CC,
     return CC_X86_32_FastCall;
   else if (CC == CallingConv::Fast)
     return CC_X86_32_FastCC;
+  else if (CC == CallingConv::GHC)
+    return CC_X86_32_GHC;
   else
     return CC_X86_32_C;
 }
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 802bedcf314b..f75df6c8f8f8 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1378,6 +1378,8 @@ bool X86TargetLowering::IsCalleePop(bool IsVarArg, CallingConv::ID CallingConv){
     return !Subtarget->is64Bit();
   case CallingConv::Fast:
     return GuaranteedTailCallOpt;
+  case CallingConv::GHC:
+    return GuaranteedTailCallOpt;
   }
 }
 
@@ -1385,7 +1387,9 @@ bool X86TargetLowering::IsCalleePop(bool IsVarArg, CallingConv::ID CallingConv){
 /// given CallingConvention value.
 CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
   if (Subtarget->is64Bit()) {
-    if (Subtarget->isTargetWin64())
+    if (CC == CallingConv::GHC)
+      return CC_X86_64_GHC;
+    else if (Subtarget->isTargetWin64())
       return CC_X86_Win64_C;
     else
       return CC_X86_64_C;
@@ -1395,6 +1399,8 @@ CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
     return CC_X86_32_FastCall;
   else if (CC == CallingConv::Fast)
     return CC_X86_32_FastCC;
+  else if (CC == CallingConv::GHC)
+    return CC_X86_32_GHC;
   else
     return CC_X86_32_C;
 }
@@ -1412,10 +1418,16 @@ CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
                        /*AlwaysInline=*/true, NULL, 0, NULL, 0);
 }
 
+/// IsTailCallConvention - Return true if the calling convention is one that
+/// supports tail call optimization.
+static bool IsTailCallConvention(CallingConv::ID CC) {
+  return (CC == CallingConv::Fast || CC == CallingConv::GHC);
+}
+
 /// FuncIsMadeTailCallSafe - Return true if the function is being made into
 /// a tailcall target by changing its ABI.
 static bool FuncIsMadeTailCallSafe(CallingConv::ID CC) {
-  return GuaranteedTailCallOpt && CC == CallingConv::Fast;
+  return GuaranteedTailCallOpt && IsTailCallConvention(CC);
 }
 
 SDValue
@@ -1479,8 +1491,8 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
   bool Is64Bit = Subtarget->is64Bit();
   bool IsWin64 = Subtarget->isTargetWin64();
 
-  assert(!(isVarArg && CallConv == CallingConv::Fast) &&
-         "Var args not supported with calling convention fastcc");
+  assert(!(isVarArg && IsTailCallConvention(CallConv)) &&
+         "Var args not supported with calling convention fastcc or ghc");
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -1683,7 +1695,7 @@ X86TargetLowering::LowerFormalArguments(SDValue Chain,
   } else {
     BytesToPopOnReturn  = 0; // Callee pops nothing.
     // If this is an sret function, the return should pop the hidden pointer.
-    if (!Is64Bit && CallConv != CallingConv::Fast && ArgsAreStructReturn(Ins))
+    if (!Is64Bit && !IsTailCallConvention(CallConv) && ArgsAreStructReturn(Ins))
       BytesToPopOnReturn = 4;
   }
 
@@ -1779,8 +1791,8 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
       ++NumTailCalls;
   }
 
-  assert(!(isVarArg && CallConv == CallingConv::Fast) &&
-         "Var args not supported with calling convention fastcc");
+  assert(!(isVarArg && IsTailCallConvention(CallConv)) &&
+         "Var args not supported with calling convention fastcc or ghc");
 
   // Analyze operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -1794,7 +1806,7 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
     // This is a sibcall. The memory operands are available in caller's
     // own caller's stack.
     NumBytes = 0;
-  else if (GuaranteedTailCallOpt && CallConv == CallingConv::Fast)
+  else if (GuaranteedTailCallOpt && IsTailCallConvention(CallConv))
     NumBytes = GetAlignedArgumentStackSize(NumBytes, DAG);
 
   int FPDiff = 0;
@@ -2150,7 +2162,7 @@ X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
   unsigned NumBytesForCalleeToPush;
   if (IsCalleePop(isVarArg, CallConv))
     NumBytesForCalleeToPush = NumBytes;    // Callee pops everything
-  else if (!Is64Bit && CallConv != CallingConv::Fast && IsStructRet)
+  else if (!Is64Bit && !IsTailCallConvention(CallConv) && IsStructRet)
     // If this 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.
@@ -2288,14 +2300,14 @@ X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
                                     const SmallVectorImpl<ISD::OutputArg> &Outs,
                                     const SmallVectorImpl<ISD::InputArg> &Ins,
                                                      SelectionDAG& DAG) const {
-  if (CalleeCC != CallingConv::Fast &&
+  if (!IsTailCallConvention(CalleeCC) &&
       CalleeCC != CallingConv::C)
     return false;
 
   // If -tailcallopt is specified, make fastcc functions tail-callable.
   const Function *CallerF = DAG.getMachineFunction().getFunction();
   if (GuaranteedTailCallOpt) {
-    if (CalleeCC == CallingConv::Fast &&
+    if (IsTailCallConvention(CalleeCC) &&
         CallerF->getCallingConv() == CalleeCC)
       return true;
     return false;
diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index 0f4ce37649a9..4aeeecb4d3c4 100644
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -294,13 +294,20 @@ X86RegisterInfo::getCrossCopyRegClass(const TargetRegisterClass *RC) const {
 const unsigned *
 X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
   bool callsEHReturn = false;
+  bool ghcCall = false;
 
   if (MF) {
     const MachineFrameInfo *MFI = MF->getFrameInfo();
     const MachineModuleInfo *MMI = MFI->getMachineModuleInfo();
     callsEHReturn = (MMI ? MMI->callsEHReturn() : false);
+    const Function *F = MF->getFunction();
+    ghcCall = (F ? F->getCallingConv() == CallingConv::GHC : false);
   }
 
+  static const unsigned GhcCalleeSavedRegs[] = {
+    0
+  };
+
   static const unsigned CalleeSavedRegs32Bit[] = {
     X86::ESI, X86::EDI, X86::EBX, X86::EBP,  0
   };
@@ -326,7 +333,9 @@ X86RegisterInfo::getCalleeSavedRegs(const MachineFunction *MF) const {
     X86::XMM14, X86::XMM15, 0
   };
 
-  if (Is64Bit) {
+  if (ghcCall) {
+    return GhcCalleeSavedRegs;
+  } else if (Is64Bit) {
     if (IsWin64)
       return CalleeSavedRegsWin64;
     else
diff --git a/test/CodeGen/X86/ghc-cc.ll b/test/CodeGen/X86/ghc-cc.ll
new file mode 100644
index 000000000000..9393cf5a7383
--- /dev/null
+++ b/test/CodeGen/X86/ghc-cc.ll
@@ -0,0 +1,45 @@
+; RUN: llc < %s -tailcallopt -mtriple=i686-linux-gnu | FileCheck %s
+
+; Test the GHC call convention works (x86-32)
+
+@base = external global i32 ; assigned to register: EBX
+@sp   = external global i32 ; assigned to register: EBP
+@hp   = external global i32 ; assigned to register: EDI
+@r1   = external global i32 ; assigned to register: ESI
+
+define void @zap(i32 %a, i32 %b) nounwind {
+entry:
+  ; CHECK: movl {{[0-9]*}}(%esp), %ebx
+  ; CHECK-NEXT: movl {{[0-9]*}}(%esp), %ebp
+  ; CHECK-NEXT: call addtwo
+  %0 = call cc 10 i32 @addtwo(i32 %a, i32 %b)
+  ; CHECK: call foo
+  call void @foo() nounwind
+  ret void
+}
+
+define cc 10 i32 @addtwo(i32 %x, i32 %y) nounwind {
+entry:
+  ; CHECK: leal (%ebx,%ebp), %eax
+  %0 = add i32 %x, %y
+  ; CHECK-NEXT: ret
+  ret i32 %0
+}
+
+define cc 10 void @foo() nounwind {
+entry:
+  ; CHECK: movl base, %ebx
+  ; CHECK-NEXT: movl sp, %ebp
+  ; CHECK-NEXT: movl hp, %edi
+  ; CHECK-NEXT: movl r1, %esi
+  %0 = load i32* @r1
+  %1 = load i32* @hp
+  %2 = load i32* @sp
+  %3 = load i32* @base
+  ; CHECK: jmp bar
+  tail call cc 10 void @bar( i32 %3, i32 %2, i32 %1, i32 %0 ) nounwind
+  ret void
+}
+
+declare cc 10 void @bar(i32, i32, i32, i32)
+
diff --git a/test/CodeGen/X86/ghc-cc64.ll b/test/CodeGen/X86/ghc-cc64.ll
new file mode 100644
index 000000000000..fcf7e1797ad8
--- /dev/null
+++ b/test/CodeGen/X86/ghc-cc64.ll
@@ -0,0 +1,86 @@
+; RUN: llc < %s -tailcallopt -mtriple=x86_64-linux-gnu | FileCheck %s
+
+; Check the GHC call convention works (x86-64)
+
+@base  = external global i64 ; assigned to register: R13
+@sp    = external global i64 ; assigned to register: RBP
+@hp    = external global i64 ; assigned to register: R12
+@r1    = external global i64 ; assigned to register: RBX
+@r2    = external global i64 ; assigned to register: R14
+@r3    = external global i64 ; assigned to register: RSI
+@r4    = external global i64 ; assigned to register: RDI
+@r5    = external global i64 ; assigned to register: R8
+@r6    = external global i64 ; assigned to register: R9
+@splim = external global i64 ; assigned to register: R15
+
+@f1 = external global float  ; assigned to register: XMM1
+@f2 = external global float  ; assigned to register: XMM2
+@f3 = external global float  ; assigned to register: XMM3
+@f4 = external global float  ; assigned to register: XMM4
+@d1 = external global double ; assigned to register: XMM5
+@d2 = external global double ; assigned to register: XMM6
+
+define void @zap(i64 %a, i64 %b) nounwind {
+entry:
+  ; CHECK:      movq %rdi, %r13
+  ; CHECK-NEXT: movq %rsi, %rbp
+  ; CHECK-NEXT: callq addtwo
+  %0 = call cc 10 i64 @addtwo(i64 %a, i64 %b)
+  ; CHECK:      callq foo
+  call void @foo() nounwind
+  ret void
+}
+
+define cc 10 i64 @addtwo(i64 %x, i64 %y) nounwind {
+entry:
+  ; CHECK:      leaq (%r13,%rbp), %rax
+  %0 = add i64 %x, %y
+  ; CHECK-NEXT: ret
+  ret i64 %0
+}
+
+define cc 10 void @foo() nounwind {
+entry:
+  ; CHECK: movq base(%rip), %r13
+  ; CHECK-NEXT: movq sp(%rip), %rbp
+  ; CHECK-NEXT: movq hp(%rip), %r12
+  ; CHECK-NEXT: movq r1(%rip), %rbx
+  ; CHECK-NEXT: movq r2(%rip), %r14
+  ; CHECK-NEXT: movq r3(%rip), %rsi
+  ; CHECK-NEXT: movq r4(%rip), %rdi
+  ; CHECK-NEXT: movq r5(%rip), %r8
+  ; CHECK-NEXT: movq r6(%rip), %r9
+  ; CHECK-NEXT: movq splim(%rip), %r15
+  ; CHECK-NEXT: movss f1(%rip), %xmm1
+  ; CHECK-NEXT: movss f2(%rip), %xmm2
+  ; CHECK-NEXT: movss f3(%rip), %xmm3
+  ; CHECK-NEXT: movss f4(%rip), %xmm4
+  ; CHECK-NEXT: movsd d1(%rip), %xmm5
+  ; CHECK-NEXT: movsd d2(%rip), %xmm6
+  %0 = load double* @d2
+  %1 = load double* @d1
+  %2 = load float* @f4
+  %3 = load float* @f3
+  %4 = load float* @f2
+  %5 = load float* @f1
+  %6 = load i64* @splim
+  %7 = load i64* @r6
+  %8 = load i64* @r5
+  %9 = load i64* @r4
+  %10 = load i64* @r3
+  %11 = load i64* @r2
+  %12 = load i64* @r1
+  %13 = load i64* @hp
+  %14 = load i64* @sp
+  %15 = load i64* @base
+  ; CHECK: jmp bar
+  tail call cc 10 void @bar( i64 %15, i64 %14, i64 %13, i64 %12, i64 %11,
+                             i64 %10, i64 %9, i64 %8, i64 %7, i64 %6,
+                             float %5, float %4, float %3, float %2, double %1,
+                             double %0 ) nounwind
+  ret void
+}
+
+declare cc 10 void @bar(i64, i64, i64, i64, i64, i64, i64, i64, i64, i64,
+                        float, float, float, float, double, double)
+