PTX: Implement PTXSelectionDAGInfo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140549 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 148 insertions, 0 deletions

diff --git a/lib/Target/PTX/PTXSelectionDAGInfo.cpp b/lib/Target/PTX/PTXSelectionDAGInfo.cpp
new file mode 100644
index 0000000..e333183
--- /dev/null
+++ b/lib/Target/PTX/PTXSelectionDAGInfo.cpp
@@ -0,0 +1,148 @@
+//===-- PTXSelectionDAGInfo.cpp - PTX SelectionDAG Info -------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the PTXSelectionDAGInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "ptx-selectiondag-info"
+#include "PTXTargetMachine.h"
+#include "llvm/DerivedTypes.h"
+#include "llvm/CodeGen/SelectionDAG.h"
+using namespace llvm;
+
+PTXSelectionDAGInfo::PTXSelectionDAGInfo(const TargetMachine &TM)
+  : TargetSelectionDAGInfo(TM),
+    Subtarget(&TM.getSubtarget<PTXSubtarget>()) {
+}
+
+PTXSelectionDAGInfo::~PTXSelectionDAGInfo() {
+}
+
+SDValue
+PTXSelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
+                                             SDValue Chain,
+                                             SDValue Dst, SDValue Src,
+                                             SDValue Size, unsigned Align,
+                                             bool isVolatile, bool AlwaysInline,
+                                             MachinePointerInfo DstPtrInfo,
+                                          MachinePointerInfo SrcPtrInfo) const {
+  // Do repeated 4-byte loads and stores. To be improved.
+  // This requires 4-byte alignment.
+  if ((Align & 3) != 0)
+    return SDValue();
+  // This requires the copy size to be a constant, preferably
+  // within a subtarget-specific limit.
+  ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
+  if (!ConstantSize)
+    return SDValue();
+  uint64_t SizeVal = ConstantSize->getZExtValue();
+  // Always inline memcpys. In PTX, we do not have a C library that provides
+  // a memcpy function.
+  //if (!AlwaysInline)
+  //  return SDValue();
+
+  unsigned BytesLeft = SizeVal & 3;
+  unsigned NumMemOps = SizeVal >> 2;
+  unsigned EmittedNumMemOps = 0;
+  EVT VT = MVT::i32;
+  unsigned VTSize = 4;
+  unsigned i = 0;
+  const unsigned MAX_LOADS_IN_LDM = 6;
+  SDValue TFOps[MAX_LOADS_IN_LDM];
+  SDValue Loads[MAX_LOADS_IN_LDM];
+  uint64_t SrcOff = 0, DstOff = 0;
+
+  // Emit up to MAX_LOADS_IN_LDM loads, then a TokenFactor barrier, then the
+  // same number of stores.  The loads and stores will get combined into
+  // ldm/stm later on.
+  while (EmittedNumMemOps < NumMemOps) {
+    for (i = 0;
+         i < MAX_LOADS_IN_LDM && EmittedNumMemOps + i < NumMemOps; ++i) {
+      Loads[i] = DAG.getLoad(VT, dl, Chain,
+                             DAG.getNode(ISD::ADD, dl, MVT::i32, Src,
+                                         DAG.getConstant(SrcOff, MVT::i32)),
+                             SrcPtrInfo.getWithOffset(SrcOff), isVolatile,
+                             false, 0);
+      TFOps[i] = Loads[i].getValue(1);
+      SrcOff += VTSize;
+    }
+    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
+
+    for (i = 0;
+         i < MAX_LOADS_IN_LDM && EmittedNumMemOps + i < NumMemOps; ++i) {
+      TFOps[i] = DAG.getStore(Chain, dl, Loads[i],
+                              DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
+                                          DAG.getConstant(DstOff, MVT::i32)),
+                              DstPtrInfo.getWithOffset(DstOff),
+                              isVolatile, false, 0);
+      DstOff += VTSize;
+    }
+    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
+
+    EmittedNumMemOps += i;
+  }
+
+  if (BytesLeft == 0)
+    return Chain;
+
+  // Issue loads / stores for the trailing (1 - 3) bytes.
+  unsigned BytesLeftSave = BytesLeft;
+  i = 0;
+  while (BytesLeft) {
+    if (BytesLeft >= 2) {
+      VT = MVT::i16;
+      VTSize = 2;
+    } else {
+      VT = MVT::i8;
+      VTSize = 1;
+    }
+
+    Loads[i] = DAG.getLoad(VT, dl, Chain,
+                           DAG.getNode(ISD::ADD, dl, MVT::i32, Src,
+                                       DAG.getConstant(SrcOff, MVT::i32)),
+                           SrcPtrInfo.getWithOffset(SrcOff), false, false, 0);
+    TFOps[i] = Loads[i].getValue(1);
+    ++i;
+    SrcOff += VTSize;
+    BytesLeft -= VTSize;
+  }
+  Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
+
+  i = 0;
+  BytesLeft = BytesLeftSave;
+  while (BytesLeft) {
+    if (BytesLeft >= 2) {
+      VT = MVT::i16;
+      VTSize = 2;
+    } else {
+      VT = MVT::i8;
+      VTSize = 1;
+    }
+
+    TFOps[i] = DAG.getStore(Chain, dl, Loads[i],
+                            DAG.getNode(ISD::ADD, dl, MVT::i32, Dst,
+                                        DAG.getConstant(DstOff, MVT::i32)),
+                            DstPtrInfo.getWithOffset(DstOff), false, false, 0);
+    ++i;
+    DstOff += VTSize;
+    BytesLeft -= VTSize;
+  }
+  return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
+}
+
+SDValue PTXSelectionDAGInfo::
+EmitTargetCodeForMemset(SelectionDAG &DAG, DebugLoc dl,
+                        SDValue Chain, SDValue Dst,
+                        SDValue Src, SDValue Size,
+                        unsigned Align, bool isVolatile,
+                        MachinePointerInfo DstPtrInfo) const {
+  llvm_unreachable("memset lowering not implemented for PTX yet");
+}
+