[ARM] Use AEABI aligned function variants

AEABI defines aligned variants of memcpy etc. that can be faster than
the default version due to not having to do alignment checks. When
emitting target code for these functions make use of these aligned
variants if possible. Also convert memset to memclr if possible.

Differential Revision: http://reviews.llvm.org/D8060

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@237127 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMSelectionDAGInfo.cpp b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
index 06bde40..a59cf98 100644 (file)
--- a/lib/Target/ARM/ARMSelectionDAGInfo.cpp
+++ b/lib/Target/ARM/ARMSelectionDAGInfo.cpp
@@ -24,6 +24,114 @@ ARMSelectionDAGInfo::ARMSelectionDAGInfo(const DataLayout &DL)
 ARMSelectionDAGInfo::~ARMSelectionDAGInfo() {
 }
 
+// Emit, if possible, a specialized version of the given Libcall. Typically this
+// means selecting the appropriately aligned version, but we also convert memset
+// of 0 into memclr.
+SDValue ARMSelectionDAGInfo::
+EmitSpecializedLibcall(SelectionDAG &DAG, SDLoc dl,
+                       SDValue Chain,
+                       SDValue Dst, SDValue Src,
+                       SDValue Size, unsigned Align,
+                       RTLIB::Libcall LC) const {
+  const ARMSubtarget &Subtarget =
+      DAG.getMachineFunction().getSubtarget<ARMSubtarget>();
+  const ARMTargetLowering *TLI = Subtarget.getTargetLowering();
+
+  // Only use a specialized AEABI function if the default version of this
+  // Libcall is an AEABI function.
+  if (std::strncmp(TLI->getLibcallName(LC), "__aeabi", 7) != 0)
+    return SDValue();
+
+  // Translate RTLIB::Libcall to AEABILibcall. We only do this in order to be
+  // able to translate memset to memclr and use the value to index the function
+  // name array.
+  enum {
+    AEABI_MEMCPY = 0,
+    AEABI_MEMMOVE,
+    AEABI_MEMSET,
+    AEABI_MEMCLR
+  } AEABILibcall;
+  switch (LC) {
+  case RTLIB::MEMCPY:
+    AEABILibcall = AEABI_MEMCPY;
+    break;
+  case RTLIB::MEMMOVE:
+    AEABILibcall = AEABI_MEMMOVE;
+    break;
+  case RTLIB::MEMSET: 
+    AEABILibcall = AEABI_MEMSET;
+    if (ConstantSDNode *ConstantSrc = dyn_cast<ConstantSDNode>(Src))
+      if (ConstantSrc->getZExtValue() == 0)
+        AEABILibcall = AEABI_MEMCLR;
+    break;
+  default:
+    return SDValue();
+  }
+
+  // Choose the most-aligned libcall variant that we can
+  enum {
+    ALIGN1 = 0,
+    ALIGN4,
+    ALIGN8
+  } AlignVariant;
+  if ((Align & 7) == 0)
+    AlignVariant = ALIGN8;
+  else if ((Align & 3) == 0)
+    AlignVariant = ALIGN4;
+  else
+    AlignVariant = ALIGN1;
+
+  TargetLowering::ArgListTy Args;
+  TargetLowering::ArgListEntry Entry;
+  Entry.Ty = TLI->getDataLayout()->getIntPtrType(*DAG.getContext());
+  Entry.Node = Dst;
+  Args.push_back(Entry);
+  if (AEABILibcall == AEABI_MEMCLR) {
+    Entry.Node = Size;
+    Args.push_back(Entry);
+  } else if (AEABILibcall == AEABI_MEMSET) {
+    // Adjust parameters for memset, EABI uses format (ptr, size, value),
+    // GNU library uses (ptr, value, size)
+    // See RTABI section 4.3.4
+    Entry.Node = Size;
+    Args.push_back(Entry);
+
+    // Extend or truncate the argument to be an i32 value for the call.
+    if (Src.getValueType().bitsGT(MVT::i32))
+      Src = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
+    else if (Src.getValueType().bitsLT(MVT::i32))
+      Src = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
+
+    Entry.Node = Src; 
+    Entry.Ty = Type::getInt32Ty(*DAG.getContext());
+    Entry.isSExt = false;
+    Args.push_back(Entry);
+  } else {
+    Entry.Node = Src;
+    Args.push_back(Entry);
+    
+    Entry.Node = Size;
+    Args.push_back(Entry);
+  }
+
+  char const *FunctionNames[4][3] = {
+    { "__aeabi_memcpy",  "__aeabi_memcpy4",  "__aeabi_memcpy8"  },
+    { "__aeabi_memmove", "__aeabi_memmove4", "__aeabi_memmove8" },
+    { "__aeabi_memset",  "__aeabi_memset4",  "__aeabi_memset8"  },
+    { "__aeabi_memclr",  "__aeabi_memclr4",  "__aeabi_memclr8"  }
+  };
+  TargetLowering::CallLoweringInfo CLI(DAG);
+  CLI.setDebugLoc(dl).setChain(Chain)
+    .setCallee(TLI->getLibcallCallingConv(LC),
+               Type::getVoidTy(*DAG.getContext()),
+               DAG.getExternalSymbol(FunctionNames[AEABILibcall][AlignVariant],
+                                     TLI->getPointerTy()), std::move(Args), 0)
+    .setDiscardResult();
+  std::pair<SDValue,SDValue> CallResult = TLI->LowerCallTo(CLI);
+  
+  return CallResult.second;
+}
+
 SDValue
 ARMSelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl,
                                              SDValue Chain,
@@ -42,10 +150,12 @@ ARMSelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl,
   // within a subtarget-specific limit.
   ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
   if (!ConstantSize)
-    return SDValue();
+    return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
+                                  RTLIB::MEMCPY);
   uint64_t SizeVal = ConstantSize->getZExtValue();
   if (!AlwaysInline && SizeVal > Subtarget.getMaxInlineSizeThreshold())
-    return SDValue();
+    return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
+                                  RTLIB::MEMCPY);
 
   unsigned BytesLeft = SizeVal & 3;
   unsigned NumMemOps = SizeVal >> 2;
@@ -142,59 +252,26 @@ ARMSelectionDAGInfo::EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl,
                      makeArrayRef(TFOps, i));
 }
 
-// Adjust parameters for memset, EABI uses format (ptr, size, value),
-// GNU library uses (ptr, value, size)
-// See RTABI section 4.3.4
+
+SDValue ARMSelectionDAGInfo::
+EmitTargetCodeForMemmove(SelectionDAG &DAG, SDLoc dl,
+                         SDValue Chain,
+                         SDValue Dst, SDValue Src,
+                         SDValue Size, unsigned Align,
+                         bool isVolatile,
+                         MachinePointerInfo DstPtrInfo,
+                         MachinePointerInfo SrcPtrInfo) const {
+  return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
+                                RTLIB::MEMMOVE);
+}
+
+
 SDValue ARMSelectionDAGInfo::
 EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
                         SDValue Chain, SDValue Dst,
                         SDValue Src, SDValue Size,
                         unsigned Align, bool isVolatile,
                         MachinePointerInfo DstPtrInfo) const {
-  const ARMSubtarget &Subtarget =
-      DAG.getMachineFunction().getSubtarget<ARMSubtarget>();
-  // Use default for non-AAPCS (or MachO) subtargets
-  if (!Subtarget.isAAPCS_ABI() || Subtarget.isTargetMachO() ||
-      Subtarget.isTargetWindows())
-    return SDValue();
-
-  const ARMTargetLowering &TLI = *Subtarget.getTargetLowering();
-  TargetLowering::ArgListTy Args;
-  TargetLowering::ArgListEntry Entry;
-
-  // First argument: data pointer
-  Type *IntPtrTy = TLI.getDataLayout()->getIntPtrType(*DAG.getContext());
-  Entry.Node = Dst;
-  Entry.Ty = IntPtrTy;
-  Args.push_back(Entry);
-
-  // Second argument: buffer size
-  Entry.Node = Size;
-  Entry.Ty = IntPtrTy;
-  Entry.isSExt = false;
-  Args.push_back(Entry);
-
-  // Extend or truncate the argument to be an i32 value for the call.
-  if (Src.getValueType().bitsGT(MVT::i32))
-    Src = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
-  else
-    Src = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
-
-  // Third argument: value to fill
-  Entry.Node = Src;
-  Entry.Ty = Type::getInt32Ty(*DAG.getContext());
-  Entry.isSExt = true;
-  Args.push_back(Entry);
-
-  // Emit __eabi_memset call
-  TargetLowering::CallLoweringInfo CLI(DAG);
-  CLI.setDebugLoc(dl).setChain(Chain)
-    .setCallee(TLI.getLibcallCallingConv(RTLIB::MEMSET),
-               Type::getVoidTy(*DAG.getContext()),
-               DAG.getExternalSymbol(TLI.getLibcallName(RTLIB::MEMSET),
-                                     TLI.getPointerTy()), std::move(Args), 0)
-    .setDiscardResult();
-
-  std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
-  return CallResult.second;
+  return EmitSpecializedLibcall(DAG, dl, Chain, Dst, Src, Size, Align,
+                                RTLIB::MEMSET);
 }

diff --git a/lib/Target/ARM/ARMSelectionDAGInfo.h b/lib/Target/ARM/ARMSelectionDAGInfo.h
index 94b98e6..1db190f 100644 (file)
--- a/lib/Target/ARM/ARMSelectionDAGInfo.h
+++ b/lib/Target/ARM/ARMSelectionDAGInfo.h
@@ -48,6 +48,13 @@ public:
                                   MachinePointerInfo DstPtrInfo,
                                   MachinePointerInfo SrcPtrInfo) const override;
 
+  SDValue EmitTargetCodeForMemmove(SelectionDAG &DAG, SDLoc dl,
+                                   SDValue Chain,
+                                   SDValue Dst, SDValue Src,
+                                   SDValue Size, unsigned Align, bool isVolatile,
+                                   MachinePointerInfo DstPtrInfo,
+                                   MachinePointerInfo SrcPtrInfo) const override;
+
   // Adjust parameters for memset, see RTABI section 4.3.4
   SDValue EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
                                   SDValue Chain,
@@ -55,6 +62,12 @@ public:
                                   SDValue Op3, unsigned Align,
                                   bool isVolatile,
                                   MachinePointerInfo DstPtrInfo) const override;
+
+  SDValue EmitSpecializedLibcall(SelectionDAG &DAG, SDLoc dl,
+                                 SDValue Chain,
+                                 SDValue Dst, SDValue Src,
+                                 SDValue Size, unsigned Align,
+                                 RTLIB::Libcall LC) const;
 };
 
 }

diff --git a/test/CodeGen/ARM/memfunc.ll b/test/CodeGen/ARM/memfunc.ll
index c214336..5223983 100644 (file)
--- a/test/CodeGen/ARM/memfunc.ll
+++ b/test/CodeGen/ARM/memfunc.ll
@@ -18,13 +18,64 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 0, i1 false)
 
   ; EABI memset swaps arguments
+  ; CHECK-IOS: mov r1, #1
+  ; CHECK-IOS: memset
+  ; CHECK-DARWIN: movs r1, #1
+  ; CHECK-DARWIN: memset
+  ; CHECK-EABI: mov r2, #1
+  ; CHECK-EABI: __aeabi_memset
+  call void @llvm.memset.p0i8.i32(i8* %dest, i8 1, i32 500, i32 0, i1 false)
+
+  ; EABI uses memclr if value set to 0
   ; CHECK-IOS: mov r1, #0
   ; CHECK-IOS: memset
   ; CHECK-DARWIN: movs r1, #0
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
-  ; CHECK-EABI: __aeabi_memset
+  ; CHECK-EABI: __aeabi_memclr
   call void @llvm.memset.p0i8.i32(i8* %dest, i8 0, i32 500, i32 0, i1 false)
+  
+  ; EABI uses aligned function variants if possible
+
+  ; CHECK-IOS: memmove
+  ; CHECK-DARWIN: memmove
+  ; CHECK-EABI: __aeabi_memmove4
+  call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 4, i1 false)
+
+  ; CHECK-IOS: memcpy
+  ; CHECK-DARWIN: memcpy
+  ; CHECK-EABI: __aeabi_memcpy4
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 4, i1 false)
+
+  ; CHECK-IOS: memset
+  ; CHECK-DARWIN: memset
+  ; CHECK-EABI: __aeabi_memset4
+  call void @llvm.memset.p0i8.i32(i8* %dest, i8 1, i32 500, i32 4, i1 false)
+
+  ; CHECK-IOS: memset
+  ; CHECK-DARWIN: memset
+  ; CHECK-EABI: __aeabi_memclr4
+  call void @llvm.memset.p0i8.i32(i8* %dest, i8 0, i32 500, i32 4, i1 false)
+
+  ; CHECK-IOS: memmove
+  ; CHECK-DARWIN: memmove
+  ; CHECK-EABI: __aeabi_memmove8
+  call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 8, i1 false)
+
+  ; CHECK-IOS: memcpy
+  ; CHECK-DARWIN: memcpy
+  ; CHECK-EABI: __aeabi_memcpy8
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 500, i32 8, i1 false)
+
+  ; CHECK-IOS: memset
+  ; CHECK-DARWIN: memset
+  ; CHECK-EABI: __aeabi_memset8
+  call void @llvm.memset.p0i8.i32(i8* %dest, i8 1, i32 500, i32 8, i1 false)
+
+  ; CHECK-IOS: memset
+  ; CHECK-DARWIN: memset
+  ; CHECK-EABI: __aeabi_memclr8
+  call void @llvm.memset.p0i8.i32(i8* %dest, i8 0, i32 500, i32 8, i1 false)
+
   unreachable
 }
 
@@ -53,17 +104,17 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK-IOS: mov r0, sp
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARINW: add r0, sp, #4
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: add r0, sp, #4
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
   ; CHECK-EABI: add r0, sp, #4
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [9 x i8], align 1
   %2 = bitcast [9 x i8]* %arr2 to i8*
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -90,15 +141,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r0, sp, #3}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [7 x i8], align 1
   %2 = bitcast [7 x i8]* %arr2 to i8*
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -125,15 +176,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r., sp, #(1|5)}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [9 x i8], align 1
   %2 = getelementptr inbounds [9 x i8], [9 x i8]* %arr2, i32 0, i32 4
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -160,15 +211,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r., sp, #(1|5)}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [13 x i8], align 1
   %2 = getelementptr inbounds [13 x i8], [13 x i8]* %arr2, i32 0, i32 1
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -195,15 +246,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r., sp, #(1|5)}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [13 x i8], align 1
   %2 = getelementptr inbounds [13 x i8], [13 x i8]* %arr2, i32 0, i32 %i
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -230,15 +281,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r., sp, #(1|5)}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [13 x i8], align 1
   %2 = getelementptr [13 x i8], [13 x i8]* %arr2, i32 0, i32 4
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }
@@ -265,15 +316,15 @@ entry:
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %1, i32 %n, i32 0, i1 false)
 
   ; CHECK: {{add(.w)? r., sp, #(1|5)}}
-  ; CHECK-IOS: mov r1, #0
+  ; CHECK-IOS: mov r1, #1
   ; CHECK-IOS: memset
-  ; CHECK-DARWIN: movs r1, #0
+  ; CHECK-DARWIN: movs r1, #1
   ; CHECK-DARWIN: memset
-  ; CHECK-EABI: mov r2, #0
+  ; CHECK-EABI: mov r2, #1
   ; CHECK-EABI: __aeabi_memset
   %arr2 = alloca [13 x i8], align 1
   %2 = getelementptr inbounds [13 x i8], [13 x i8]* %arr2, i32 0, i32 16
-  call void @llvm.memset.p0i8.i32(i8* %2, i8 0, i32 %n, i32 0, i1 false)
+  call void @llvm.memset.p0i8.i32(i8* %2, i8 1, i32 %n, i32 0, i1 false)
 
   unreachable
 }