From cfbb32346a7eca19d4dc00fd2c1a0248537b3497 Mon Sep 17 00:00:00 2001
From: Dan Gohman <gohman@apple.com>
Date: Tue, 11 May 2010 17:31:57 +0000
Subject: Implement a bunch more TargetSelectionDAGInfo infrastructure.

Move EmitTargetCodeForMemcpy, EmitTargetCodeForMemset, and
EmitTargetCodeForMemmove out of TargetLowering and into
SelectionDAGInfo to exercise this.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103481 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/ARM/ARMSelectionDAGInfo.cpp | 116 ++++++++++++++++++++++++++++++++-
 1 file changed, 114 insertions(+), 2 deletions(-)

(limited to 'lib/Target/ARM/ARMSelectionDAGInfo.cpp')

diff --git a/lib/Target/ARM/ARMSelectionDAGInfo.cpp b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
index c04ee38..a289407 100644
--- a/lib/Target/ARM/ARMSelectionDAGInfo.cpp
+++ b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
@@ -12,11 +12,123 @@
 //===----------------------------------------------------------------------===//
 
 #define DEBUG_TYPE "arm-selectiondag-info"
-#include "ARMSelectionDAGInfo.h"
+#include "ARMTargetMachine.h"
 using namespace llvm;
 
-ARMSelectionDAGInfo::ARMSelectionDAGInfo() {
+ARMSelectionDAGInfo::ARMSelectionDAGInfo(const TargetMachine &TM)
+  : TargetSelectionDAGInfo(TM),
+    Subtarget(&TM.getSubtarget<ARMSubtarget>()) {
 }
 
 ARMSelectionDAGInfo::~ARMSelectionDAGInfo() {
 }
+
+SDValue
+ARMSelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
+                                             SDValue Chain,
+                                             SDValue Dst, SDValue Src,
+                                             SDValue Size, unsigned Align,
+                                             bool isVolatile, bool AlwaysInline,
+                                             const Value *DstSV,
+                                             uint64_t DstSVOff,
+                                             const Value *SrcSV,
+                                             uint64_t SrcSVOff) 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, preferrably
+  // within a subtarget-specific limit.
+  ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
+  if (!ConstantSize)
+    return SDValue();
+  uint64_t SizeVal = ConstantSize->getZExtValue();
+  if (!AlwaysInline && SizeVal > Subtarget->getMaxInlineSizeThreshold())
+    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)),
+                             SrcSV, SrcSVOff + 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)),
+                              DstSV, DstSVOff + 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)),
+                           SrcSV, SrcSVOff + 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)),
+                            DstSV, DstSVOff + DstOff, false, false, 0);
+    ++i;
+    DstOff += VTSize;
+    BytesLeft -= VTSize;
+  }
+  return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &TFOps[0], i);
+}
-- 
cgit v1.1