Add intrinsics for Ivy Bridge's rdrand instruction.

The rdrand/cmov sequence is the same that is emitted by both
GCC and ICC.

Fixes PR13284.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160117 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 107 insertions, 3 deletions

diff --git a/include/llvm/IntrinsicsX86.td b/include/llvm/IntrinsicsX86.td
index 14fd76d..e8039f2 100644
--- a/include/llvm/IntrinsicsX86.td
+++ b/include/llvm/IntrinsicsX86.td
@@ -2536,3 +2536,14 @@ let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
               Intrinsic<[llvm_v8i16_ty], [llvm_v8f32_ty, llvm_i32_ty],
                         [IntrNoMem]>;
 }
+
+//===----------------------------------------------------------------------===//
+// RDRAND intrinsics. Return a random value and whether it is valid.
+
+let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
+  // These are declared side-effecting so they don't get eliminated by CSE or
+  // LICM.
+  def int_x86_rdrand_16 : Intrinsic<[llvm_i16_ty, llvm_i32_ty], [], []>;
+  def int_x86_rdrand_32 : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [], []>;
+  def int_x86_rdrand_64 : Intrinsic<[llvm_i64_ty, llvm_i32_ty], [], []>;
+}
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4dccd40..c242aaa 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1176,6 +1176,7 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
 
   // We want to custom lower some of our intrinsics.
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
+  setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
 
 
   // Only custom-lower 64-bit SADDO and friends on 64-bit because we don't
@@ -9810,6 +9811,38 @@ X86TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const
   }
 }
 
+SDValue
+X86TargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const {
+  DebugLoc dl = Op.getDebugLoc();
+  unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
+  switch (IntNo) {
+  default: return SDValue();    // Don't custom lower most intrinsics.
+
+  // RDRAND intrinsics.
+  case Intrinsic::x86_rdrand_16:
+  case Intrinsic::x86_rdrand_32:
+  case Intrinsic::x86_rdrand_64: {
+    // Emit the node with the right value type.
+    SDValue Result = DAG.getNode(X86ISD::RDRAND, dl,
+                                 DAG.getVTList(Op->getValueType(0), MVT::Glue));
+
+    // If the value returned by RDRAND was valid (CF=1), return 1. Otherwise
+    // return the value from Rand, which is always 0, casted to i32.
+    SDValue Ops[] = { DAG.getZExtOrTrunc(Result, dl, Op->getValueType(1)),
+                      DAG.getConstant(1, Op->getValueType(1)),
+                      DAG.getConstant(X86::COND_B, MVT::i32),
+                      SDValue(Result.getNode(), 1) };
+    SDValue isValid = DAG.getNode(X86ISD::CMOV, dl,
+                                  DAG.getVTList(Op->getValueType(1), MVT::Glue),
+                                  Ops, 4);
+
+    // Return { result, isValid, chain }.
+    return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), Result, isValid,
+                       Op.getOperand(0));
+  }
+  }
+}
+
 SDValue X86TargetLowering::LowerRETURNADDR(SDValue Op,
                                            SelectionDAG &DAG) const {
   MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
@@ -10894,6 +10927,7 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::VAARG:              return LowerVAARG(Op, DAG);
   case ISD::VACOPY:             return LowerVACOPY(Op, DAG);
   case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
+  case ISD::INTRINSIC_W_CHAIN:  return LowerINTRINSIC_W_CHAIN(Op, DAG);
   case ISD::RETURNADDR:         return LowerRETURNADDR(Op, DAG);
   case ISD::FRAMEADDR:          return LowerFRAMEADDR(Op, DAG);
   case ISD::FRAME_TO_ARGS_OFFSET:
@@ -11228,6 +11262,7 @@ const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::SEG_ALLOCA:         return "X86ISD::SEG_ALLOCA";
   case X86ISD::WIN_FTOL:           return "X86ISD::WIN_FTOL";
   case X86ISD::SAHF:               return "X86ISD::SAHF";
+  case X86ISD::RDRAND:             return "X86ISD::RDRAND";
   }
 }
 
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 2f561a0..41bdd06 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -322,6 +322,9 @@ namespace llvm {
       // SAHF - Store contents of %ah into %eflags.
       SAHF,
 
+      // RDRAND - Get a random integer and indicate whether it is valid in CF.
+      RDRAND,
+
       // ATOMADD64_DAG, ATOMSUB64_DAG, ATOMOR64_DAG, ATOMAND64_DAG,
       // ATOMXOR64_DAG, ATOMNAND64_DAG, ATOMSWAP64_DAG -
       // Atomic 64-bit binary operations.
@@ -768,6 +771,7 @@ namespace llvm {
     SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFRAME_TO_ARGS_OFFSET(SDValue Op, SelectionDAG &DAG) const;
diff --git a/lib/Target/X86/X86InstrInfo.td b/lib/Target/X86/X86InstrInfo.td
index 0f7b787..dc26d4c 100644
--- a/lib/Target/X86/X86InstrInfo.td
+++ b/lib/Target/X86/X86InstrInfo.td
@@ -65,6 +65,8 @@ def SDTX86SetCC_C : SDTypeProfile<1, 2,
 
 def SDTX86sahf : SDTypeProfile<1, 1, [SDTCisVT<0, i32>, SDTCisVT<1, i8>]>;
 
+def SDTX86rdrand : SDTypeProfile<2, 0, [SDTCisInt<0>, SDTCisVT<1, i32>]>;
+
 def SDTX86cas : SDTypeProfile<0, 3, [SDTCisPtrTy<0>, SDTCisInt<1>,
                                      SDTCisVT<2, i8>]>;
 def SDTX86caspair : SDTypeProfile<0, 1, [SDTCisPtrTy<0>]>;
@@ -137,6 +139,8 @@ def X86setcc_c : SDNode<"X86ISD::SETCC_CARRY", SDTX86SetCC_C>;
 
 def X86sahf    : SDNode<"X86ISD::SAHF",     SDTX86sahf>;
 
+def X86rdrand  : SDNode<"X86ISD::RDRAND",     SDTX86rdrand>;
+
 def X86cas : SDNode<"X86ISD::LCMPXCHG_DAG", SDTX86cas,
                         [SDNPHasChain, SDNPInGlue, SDNPOutGlue, SDNPMayStore,
                          SDNPMayLoad, SDNPMemOperand]>;
@@ -1466,11 +1470,14 @@ let Predicates = [HasMOVBE] in {
 //
 let Predicates = [HasRDRAND], Defs = [EFLAGS] in {
   def RDRAND16r : I<0xC7, MRM6r, (outs GR16:$dst), (ins),
-                    "rdrand{w}\t$dst", []>, OpSize, TB;
+                    "rdrand{w}\t$dst",
+                    [(set GR16:$dst, EFLAGS, (X86rdrand))]>, OpSize, TB;
   def RDRAND32r : I<0xC7, MRM6r, (outs GR32:$dst), (ins),
-                    "rdrand{l}\t$dst", []>, TB;
+                    "rdrand{l}\t$dst",
+                    [(set GR32:$dst, EFLAGS, (X86rdrand))]>, TB;
   def RDRAND64r : RI<0xC7, MRM6r, (outs GR64:$dst), (ins),
-                     "rdrand{q}\t$dst", []>, TB;
+                     "rdrand{q}\t$dst",
+                     [(set GR64:$dst, EFLAGS, (X86rdrand))]>, TB;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/test/CodeGen/X86/rdrand.ll b/test/CodeGen/X86/rdrand.ll
new file mode 100644
index 0000000..7bb25b3
--- /dev/null
+++ b/test/CodeGen/X86/rdrand.ll
@@ -0,0 +1,47 @@
+; RUN: llc < rdrand.ll -march=x86-64 -mattr=+rdrand | FileCheck %s
+declare {i16, i32} @llvm.x86.rdrand.16()
+declare {i32, i32} @llvm.x86.rdrand.32()
+declare {i64, i32} @llvm.x86.rdrand.64()
+
+define i32 @_rdrand16_step(i16* %random_val) {
+  %call = call {i16, i32} @llvm.x86.rdrand.16()
+  %randval = extractvalue {i16, i32} %call, 0
+  store i16 %randval, i16* %random_val
+  %isvalid = extractvalue {i16, i32} %call, 1
+  ret i32 %isvalid
+; CHECK: _rdrand16_step:
+; CHECK: rdrandw	%ax
+; CHECK: movw	%ax, (%rdi)
+; CHECK: movzwl	%ax, %ecx
+; CHECK: movl	$1, %eax
+; CHECK: cmovael	%ecx, %eax
+; CHECK: ret
+}
+
+define i32 @_rdrand32_step(i32* %random_val) {
+  %call = call {i32, i32} @llvm.x86.rdrand.32()
+  %randval = extractvalue {i32, i32} %call, 0
+  store i32 %randval, i32* %random_val
+  %isvalid = extractvalue {i32, i32} %call, 1
+  ret i32 %isvalid
+; CHECK: _rdrand32_step:
+; CHECK: rdrandl	%ecx
+; CHECK: movl	%ecx, (%rdi)
+; CHECK: movl	$1, %eax
+; CHECK: cmovael	%ecx, %eax
+; CHECK: ret
+}
+
+define i32 @_rdrand64_step(i64* %random_val) {
+  %call = call {i64, i32} @llvm.x86.rdrand.64()
+  %randval = extractvalue {i64, i32} %call, 0
+  store i64 %randval, i64* %random_val
+  %isvalid = extractvalue {i64, i32} %call, 1
+  ret i32 %isvalid
+; CHECK: _rdrand64_step:
+; CHECK: rdrandq	%rcx
+; CHECK: movq	%rcx, (%rdi)
+; CHECK: movl	$1, %eax
+; CHECK: cmovael	%ecx, %eax
+; CHECK: ret
+}