Add sqrt and powi intrinsics for long double.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@42423 91177308-0d34-0410-b5e6-96231b3b80d8

9 files changed, 95 insertions, 7 deletions

diff --git a/include/llvm/CodeGen/RuntimeLibcalls.h b/include/llvm/CodeGen/RuntimeLibcalls.h
index 0b36c6d..80cc33a 100644
--- a/include/llvm/CodeGen/RuntimeLibcalls.h
+++ b/include/llvm/CodeGen/RuntimeLibcalls.h
@@ -18,7 +18,9 @@
 namespace llvm {
 namespace RTLIB {
   /// RTLIB::Libcall enum - This enum defines all of the runtime library calls
-  /// the backend can emit.
+  /// the backend can emit.  "LD" is used for all long double types, since
+  /// these functions will have the same interface on different targets even 
+  /// though the data is not in the same format.
   ///
   enum Libcall {
     // Integer
@@ -56,8 +58,10 @@ namespace RTLIB {
     NEG_F64,
     POWI_F32,
     POWI_F64,
+    POWI_LD,
     SQRT_F32,
     SQRT_F64,
+    SQRT_LD,
     SIN_F32,
     SIN_F64,
     COS_F32,
diff --git a/include/llvm/Intrinsics.td b/include/llvm/Intrinsics.td
index 9d3b180..4b1e419 100644
--- a/include/llvm/Intrinsics.td
+++ b/include/llvm/Intrinsics.td
@@ -76,6 +76,9 @@ def llvm_i32_ty        : LLVMType<i32>;
 def llvm_i64_ty        : LLVMType<i64>;
 def llvm_float_ty      : LLVMType<f32>;
 def llvm_double_ty     : LLVMType<f64>;
+def llvm_f80_ty        : LLVMType<f80>;
+def llvm_f128_ty       : LLVMType<f128>;
+def llvm_ppcf128_ty    : LLVMType<ppcf128>;
 def llvm_ptr_ty        : LLVMPointerType<llvm_i8_ty>;             // i8*
 def llvm_ptrptr_ty     : LLVMPointerType<llvm_ptr_ty>;            // i8**
 def llvm_empty_ty      : LLVMType<OtherVT>;                       // { }
@@ -177,9 +180,16 @@ let Properties = [IntrWriteArgMem] in {
 let Properties = [IntrNoMem] in {
   def int_sqrt_f32 : Intrinsic<[llvm_float_ty, llvm_float_ty]>;
   def int_sqrt_f64 : Intrinsic<[llvm_double_ty, llvm_double_ty]>;
+  def int_sqrt_f80 : Intrinsic<[llvm_f80_ty, llvm_f80_ty]>;
+  def int_sqrt_f128 : Intrinsic<[llvm_f128_ty, llvm_f128_ty]>;
+  def int_sqrt_ppcf128 : Intrinsic<[llvm_ppcf128_ty, llvm_ppcf128_ty]>;
 
   def int_powi_f32 : Intrinsic<[llvm_float_ty, llvm_float_ty, llvm_i32_ty]>;
   def int_powi_f64 : Intrinsic<[llvm_double_ty, llvm_double_ty, llvm_i32_ty]>;
+  def int_powi_f80 : Intrinsic<[llvm_f80_ty, llvm_f80_ty, llvm_i32_ty]>;
+  def int_powi_f128 : Intrinsic<[llvm_f128_ty, llvm_f128_ty, llvm_i32_ty]>;
+  def int_powi_ppcf128 : Intrinsic<[llvm_ppcf128_ty, llvm_ppcf128_ty, 
+                                    llvm_i32_ty]>;
 }
 
 // NOTE: these are internal interfaces.
diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp
index 6c828fa..d689615 100644
--- a/lib/Analysis/ConstantFolding.cpp
+++ b/lib/Analysis/ConstantFolding.cpp
@@ -331,8 +331,14 @@ llvm::canConstantFoldCallTo(Function *F) {
   switch (F->getIntrinsicID()) {
   case Intrinsic::sqrt_f32:
   case Intrinsic::sqrt_f64:
+  case Intrinsic::sqrt_f80:
+  case Intrinsic::sqrt_f128:
+  case Intrinsic::sqrt_ppcf128:
   case Intrinsic::powi_f32:
   case Intrinsic::powi_f64:
+  case Intrinsic::powi_f80:
+  case Intrinsic::powi_f128:
+  case Intrinsic::powi_ppcf128:
   case Intrinsic::bswap:
   case Intrinsic::ctpop:
   case Intrinsic::ctlz:
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 509e225..b976195 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -3044,7 +3044,8 @@ SDOperand SelectionDAGLegalize::LegalizeOp(SDOperand Op) {
         RTLIB::Libcall LC = RTLIB::UNKNOWN_LIBCALL;
         switch(Node->getOpcode()) {
         case ISD::FSQRT:
-          LC = VT == MVT::f32 ? RTLIB::SQRT_F32 : RTLIB::SQRT_F64;
+          LC = VT == MVT::f32 ? RTLIB::SQRT_F32 : 
+               VT == MVT::f64 ? RTLIB::SQRT_F64 : RTLIB::SQRT_LD;
           break;
         case ISD::FSIN:
           LC = VT == MVT::f32 ? RTLIB::SIN_F32 : RTLIB::SIN_F64;
@@ -3065,8 +3066,10 @@ SDOperand SelectionDAGLegalize::LegalizeOp(SDOperand Op) {
     break;
   case ISD::FPOWI: {
     // We always lower FPOWI into a libcall.  No target support it yet.
-    RTLIB::Libcall LC = Node->getValueType(0) == MVT::f32
-      ? RTLIB::POWI_F32 : RTLIB::POWI_F64;
+    RTLIB::Libcall LC = 
+      Node->getValueType(0) == MVT::f32 ? RTLIB::POWI_F32 : 
+      Node->getValueType(0) == MVT::f64 ? RTLIB::POWI_F64 : 
+      RTLIB::POWI_LD;
     SDOperand Dummy;
     Result = ExpandLibCall(TLI.getLibcallName(LC), Node,
                            false/*sign irrelevant*/, Dummy);
@@ -5688,8 +5691,9 @@ void SelectionDAGLegalize::ExpandOp(SDOperand Op, SDOperand &Lo, SDOperand &Hi){
     Lo = ExpandLibCall(TLI.getLibcallName(RTLIB::FPROUND_F64_F32),Node,true,Hi);
     break;
   case ISD::FPOWI:
-    Lo = ExpandLibCall(TLI.getLibcallName((VT == MVT::f32)
-                                          ? RTLIB::POWI_F32 : RTLIB::POWI_F64),
+    Lo = ExpandLibCall(TLI.getLibcallName((VT == MVT::f32) ? RTLIB::POWI_F32 : 
+                                          (VT == MVT::f64) ? RTLIB::POWI_F64 :
+                                          RTLIB::POWI_LD),
                        Node, false, Hi);
     break;
   case ISD::FSQRT:
@@ -5698,7 +5702,8 @@ void SelectionDAGLegalize::ExpandOp(SDOperand Op, SDOperand &Lo, SDOperand &Hi){
     RTLIB::Libcall LC = RTLIB::UNKNOWN_LIBCALL;
     switch(Node->getOpcode()) {
     case ISD::FSQRT:
-      LC = (VT == MVT::f32) ? RTLIB::SQRT_F32 : RTLIB::SQRT_F64;
+      LC = (VT == MVT::f32) ? RTLIB::SQRT_F32 : 
+           (VT == MVT::f64) ? RTLIB::SQRT_F64 : RTLIB::SQRT_LD;
       break;
     case ISD::FSIN:
       LC = (VT == MVT::f32) ? RTLIB::SIN_F32 : RTLIB::SIN_F64;
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
index f582c8f..23ad28f 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp
@@ -2798,12 +2798,18 @@ SelectionDAGLowering::visitIntrinsicCall(CallInst &I, unsigned Intrinsic) {
 
   case Intrinsic::sqrt_f32:
   case Intrinsic::sqrt_f64:
+  case Intrinsic::sqrt_f80:
+  case Intrinsic::sqrt_f128:
+  case Intrinsic::sqrt_ppcf128:
     setValue(&I, DAG.getNode(ISD::FSQRT,
                              getValue(I.getOperand(1)).getValueType(),
                              getValue(I.getOperand(1))));
     return 0;
   case Intrinsic::powi_f32:
   case Intrinsic::powi_f64:
+  case Intrinsic::powi_f80:
+  case Intrinsic::powi_f128:
+  case Intrinsic::powi_ppcf128:
     setValue(&I, DAG.getNode(ISD::FPOWI,
                              getValue(I.getOperand(1)).getValueType(),
                              getValue(I.getOperand(1)),
diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 0898173..f87ebe4 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -58,8 +58,10 @@ static void InitLibcallNames(const char **Names) {
   Names[RTLIB::NEG_F64] = "__negdf2";
   Names[RTLIB::POWI_F32] = "__powisf2";
   Names[RTLIB::POWI_F64] = "__powidf2";
+  Names[RTLIB::POWI_LD] = "__powixf2";
   Names[RTLIB::SQRT_F32] = "sqrtf";
   Names[RTLIB::SQRT_F64] = "sqrt";
+  Names[RTLIB::SQRT_LD] = "sqrtl";
   Names[RTLIB::SIN_F32] = "sinf";
   Names[RTLIB::SIN_F64] = "sin";
   Names[RTLIB::COS_F32] = "cosf";
diff --git a/test/CodeGen/X86/2007-09-27-LDIntrinsics.ll b/test/CodeGen/X86/2007-09-27-LDIntrinsics.ll
new file mode 100644
index 0000000..be51c04
--- /dev/null
+++ b/test/CodeGen/X86/2007-09-27-LDIntrinsics.ll
@@ -0,0 +1,47 @@
+; RUN: llvm-as < %s | llc | grep powixf2
+; RUN: llvm-as < %s | llc | grep fsqrt
+; ModuleID = 'yyy.c'
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
+target triple = "i686-apple-darwin8"
+
+define x86_fp80 @foo(x86_fp80 %x) {
+entry:
+	%x_addr = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%retval = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%tmp = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
+	store x86_fp80 %x, x86_fp80* %x_addr
+	%tmp1 = load x86_fp80* %x_addr, align 16		; <x86_fp80> [#uses=1]
+	%tmp2 = call x86_fp80 @llvm.sqrt.f80( x86_fp80 %tmp1 )		; <x86_fp80> [#uses=1]
+	store x86_fp80 %tmp2, x86_fp80* %tmp, align 16
+	%tmp3 = load x86_fp80* %tmp, align 16		; <x86_fp80> [#uses=1]
+	store x86_fp80 %tmp3, x86_fp80* %retval, align 16
+	br label %return
+
+return:		; preds = %entry
+	%retval4 = load x86_fp80* %retval		; <x86_fp80> [#uses=1]
+	ret x86_fp80 %retval4
+}
+
+declare x86_fp80 @llvm.sqrt.f80(x86_fp80)
+
+define x86_fp80 @bar(x86_fp80 %x) {
+entry:
+	%x_addr = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%retval = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%tmp = alloca x86_fp80		; <x86_fp80*> [#uses=2]
+	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
+	store x86_fp80 %x, x86_fp80* %x_addr
+	%tmp1 = load x86_fp80* %x_addr, align 16		; <x86_fp80> [#uses=1]
+	%tmp2 = call x86_fp80 @llvm.powi.f80( x86_fp80 %tmp1, i32 3 )		; <x86_fp80> [#uses=1]
+	store x86_fp80 %tmp2, x86_fp80* %tmp, align 16
+	%tmp3 = load x86_fp80* %tmp, align 16		; <x86_fp80> [#uses=1]
+	store x86_fp80 %tmp3, x86_fp80* %retval, align 16
+	br label %return
+
+return:		; preds = %entry
+	%retval4 = load x86_fp80* %retval		; <x86_fp80> [#uses=1]
+	ret x86_fp80 %retval4
+}
+
+declare x86_fp80 @llvm.powi.f80(x86_fp80, i32)
diff --git a/utils/TableGen/CodeGenTarget.cpp b/utils/TableGen/CodeGenTarget.cpp
index c73ffb9..ea96cf4 100644
--- a/utils/TableGen/CodeGenTarget.cpp
+++ b/utils/TableGen/CodeGenTarget.cpp
@@ -49,6 +49,7 @@ std::string llvm::getName(MVT::ValueType T) {
   case MVT::f64:   return "MVT::f64";
   case MVT::f80:   return "MVT::f80";
   case MVT::f128:  return "MVT::f128";
+  case MVT::ppcf128:  return "MVT::ppcf128";
   case MVT::Flag:  return "MVT::Flag";
   case MVT::isVoid:return "MVT::void";
   case MVT::v8i8:  return "MVT::v8i8";
@@ -84,6 +85,7 @@ std::string llvm::getEnumName(MVT::ValueType T) {
   case MVT::f64:   return "MVT::f64";
   case MVT::f80:   return "MVT::f80";
   case MVT::f128:  return "MVT::f128";
+  case MVT::ppcf128:  return "MVT::ppcf128";
   case MVT::Flag:  return "MVT::Flag";
   case MVT::isVoid:return "MVT::isVoid";
   case MVT::v8i8:  return "MVT::v8i8";
diff --git a/utils/TableGen/IntrinsicEmitter.cpp b/utils/TableGen/IntrinsicEmitter.cpp
index c3db273..b7c8093 100644
--- a/utils/TableGen/IntrinsicEmitter.cpp
+++ b/utils/TableGen/IntrinsicEmitter.cpp
@@ -131,6 +131,12 @@ static void EmitTypeForValueType(std::ostream &OS, MVT::ValueType VT) {
     OS << "Type::FloatTy";
   } else if (VT == MVT::f64) {
     OS << "Type::DoubleTy";
+  } else if (VT == MVT::f80) {
+    OS << "Type::X86_FP80Ty";
+  } else if (VT == MVT::f128) {
+    OS << "Type::FP128Ty";
+  } else if (VT == MVT::ppcf128) {
+    OS << "Type::PPC_FP128Ty";
   } else if (VT == MVT::isVoid) {
     OS << "Type::VoidTy";
   } else {