Merge commit '74570a7b19b8' from apple/stable/20200714 into swift/main

Author: git apple-llvm automerger

llvm/lib/Support/Windows/Path.inc

@@ -19,7 +19,6 @@
 #include "llvm/Support/ConvertUTF.h"
 #include "llvm/Support/WindowsError.h"
 #include <fcntl.h>
-#include <io.h>
 #include <sys/stat.h>
 #include <sys/types.h>
 
@@ -352,13 +351,13 @@ std::error_code is_local(const Twine &path, bool &result) {
 static std::error_code realPathFromHandle(HANDLE H,
                                           SmallVectorImpl<wchar_t> &Buffer) {
   DWORD CountChars = ::GetFinalPathNameByHandleW(
-      H, Buffer.begin(), Buffer.capacity() - 1, FILE_NAME_NORMALIZED);
-  if (CountChars > Buffer.capacity()) {
+      H, Buffer.begin(), Buffer.capacity(), FILE_NAME_NORMALIZED);
+  if (CountChars && CountChars >= Buffer.capacity()) {
     // The buffer wasn't big enough, try again.  In this case the return value
     // *does* indicate the size of the null terminator.
     Buffer.reserve(CountChars);
     CountChars = ::GetFinalPathNameByHandleW(
-        H, Buffer.data(), Buffer.capacity() - 1, FILE_NAME_NORMALIZED);
+        H, Buffer.begin(), Buffer.capacity(), FILE_NAME_NORMALIZED);
   }
   if (CountChars == 0)
     return mapWindowsError(GetLastError());
@@ -403,6 +402,20 @@ std::error_code is_local(int FD, bool &Result) {
 }
 
 static std::error_code setDeleteDisposition(HANDLE Handle, bool Delete) {
+  // First, check if the file is on a network (non-local) drive. If so, don't
+  // set DeleteFile to true, since it prevents opening the file for writes.
+  SmallVector<wchar_t, 128> FinalPath;
+  if (std::error_code EC = realPathFromHandle(Handle, FinalPath))
+    return EC;
+
+  bool IsLocal;
+  if (std::error_code EC = is_local_internal(FinalPath, IsLocal))
+    return EC;
+
+  if (!IsLocal)
+    return std::error_code();
+
+  // The file is on a local drive, set the DeleteFile to true.
   FILE_DISPOSITION_INFO Disposition;
   Disposition.DeleteFile = Delete;
   if (!SetFileInformationByHandle(Handle, FileDispositionInfo, &Disposition,

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

@@ -289,14 +289,15 @@ class AArch64InstructionSelector : public InstructionSelector {
   getExtendTypeForInst(MachineInstr &MI, MachineRegisterInfo &MRI,
                        bool IsLoadStore = false) const;
 
-  /// Instructions that accept extend modifiers like UXTW expect the register
-  /// being extended to be a GPR32. Narrow ExtReg to a 32-bit register using a
-  /// subregister copy if necessary. Return either ExtReg, or the result of the
-  /// new copy.
-  Register narrowExtendRegIfNeeded(Register ExtReg,
-                                             MachineIRBuilder &MIB) const;
-  Register widenGPRBankRegIfNeeded(Register Reg, unsigned Size,
-                                   MachineIRBuilder &MIB) const;
+  /// Move \p Reg to \p RC if \p Reg is not already on \p RC.
+  ///
+  /// \returns Either \p Reg if no change was necessary, or the new register
+  /// created by moving \p Reg.
+  ///
+  /// Note: This uses emitCopy right now.
+  Register moveScalarRegClass(Register Reg, const TargetRegisterClass &RC,
+                              MachineIRBuilder &MIB) const;
+
   ComplexRendererFns selectArithExtendedRegister(MachineOperand &Root) const;
 
   void renderTruncImm(MachineInstrBuilder &MIB, const MachineInstr &MI,
@@ -1195,10 +1196,10 @@ MachineInstr *AArch64InstructionSelector::emitTestBit(
   // TBNZW work.
   bool UseWReg = Bit < 32;
   unsigned NecessarySize = UseWReg ? 32 : 64;
-  if (Size < NecessarySize)
-    TestReg = widenGPRBankRegIfNeeded(TestReg, NecessarySize, MIB);
-  else if (Size > NecessarySize)
-    TestReg = narrowExtendRegIfNeeded(TestReg, MIB);
+  if (Size != NecessarySize)
+    TestReg = moveScalarRegClass(
+        TestReg, UseWReg ? AArch64::GPR32RegClass : AArch64::GPR64RegClass,
+        MIB);
 
   static const unsigned OpcTable[2][2] = {{AArch64::TBZX, AArch64::TBNZX},
                                           {AArch64::TBZW, AArch64::TBNZW}};
@@ -4907,9 +4908,19 @@ AArch64InstructionSelector::selectExtendedSHL(
     return None;
 
   unsigned OffsetOpc = OffsetInst->getOpcode();
-  if (OffsetOpc != TargetOpcode::G_SHL && OffsetOpc != TargetOpcode::G_MUL)
-    return None;
+  bool LookedThroughZExt = false;
+  if (OffsetOpc != TargetOpcode::G_SHL && OffsetOpc != TargetOpcode::G_MUL) {
+    // Try to look through a ZEXT.
+    if (OffsetOpc != TargetOpcode::G_ZEXT || !WantsExt)
+      return None;
+
+    OffsetInst = MRI.getVRegDef(OffsetInst->getOperand(1).getReg());
+    OffsetOpc = OffsetInst->getOpcode();
+    LookedThroughZExt = true;
 
+    if (OffsetOpc != TargetOpcode::G_SHL && OffsetOpc != TargetOpcode::G_MUL)
+      return None;
+  }
   // Make sure that the memory op is a valid size.
   int64_t LegalShiftVal = Log2_32(SizeInBytes);
   if (LegalShiftVal == 0)
@@ -4960,21 +4971,24 @@ AArch64InstructionSelector::selectExtendedSHL(
 
   unsigned SignExtend = 0;
   if (WantsExt) {
-    // Check if the offset is defined by an extend.
-    MachineInstr *ExtInst = getDefIgnoringCopies(OffsetReg, MRI);
-    auto Ext = getExtendTypeForInst(*ExtInst, MRI, true);
-    if (Ext == AArch64_AM::InvalidShiftExtend)
-      return None;
+    // Check if the offset is defined by an extend, unless we looked through a
+    // G_ZEXT earlier.
+    if (!LookedThroughZExt) {
+      MachineInstr *ExtInst = getDefIgnoringCopies(OffsetReg, MRI);
+      auto Ext = getExtendTypeForInst(*ExtInst, MRI, true);
+      if (Ext == AArch64_AM::InvalidShiftExtend)
+        return None;
 
-    SignExtend = isSignExtendShiftType(Ext) ? 1 : 0;
-    // We only support SXTW for signed extension here.
-    if (SignExtend && Ext != AArch64_AM::SXTW)
-      return None;
+      SignExtend = isSignExtendShiftType(Ext) ? 1 : 0;
+      // We only support SXTW for signed extension here.
+      if (SignExtend && Ext != AArch64_AM::SXTW)
+        return None;
+      OffsetReg = ExtInst->getOperand(1).getReg();
+    }
 
     // Need a 32-bit wide register here.
     MachineIRBuilder MIB(*MRI.getVRegDef(Root.getReg()));
-    OffsetReg = ExtInst->getOperand(1).getReg();
-    OffsetReg = narrowExtendRegIfNeeded(OffsetReg, MIB);
+    OffsetReg = moveScalarRegClass(OffsetReg, AArch64::GPR32RegClass, MIB);
   }
 
   // We can use the LHS of the GEP as the base, and the LHS of the shift as an
@@ -5146,8 +5160,8 @@ AArch64InstructionSelector::selectAddrModeWRO(MachineOperand &Root,
 
   // Need a 32-bit wide register.
   MachineIRBuilder MIB(*PtrAdd);
-  Register ExtReg =
-      narrowExtendRegIfNeeded(OffsetInst->getOperand(1).getReg(), MIB);
+  Register ExtReg = moveScalarRegClass(OffsetInst->getOperand(1).getReg(),
+                                       AArch64::GPR32RegClass, MIB);
   unsigned SignExtend = Ext == AArch64_AM::SXTW;
 
   // Base is LHS, offset is ExtReg.
@@ -5421,67 +5435,21 @@ AArch64_AM::ShiftExtendType AArch64InstructionSelector::getExtendTypeForInst(
   }
 }
 
-Register AArch64InstructionSelector::narrowExtendRegIfNeeded(
-    Register ExtReg, MachineIRBuilder &MIB) const {
+Register AArch64InstructionSelector::moveScalarRegClass(
+    Register Reg, const TargetRegisterClass &RC, MachineIRBuilder &MIB) const {
   MachineRegisterInfo &MRI = *MIB.getMRI();
-  if (MRI.getType(ExtReg).getSizeInBits() == 32)
-    return ExtReg;
-
-  // Insert a copy to move ExtReg to GPR32.
-  Register NarrowReg = MRI.createVirtualRegister(&AArch64::GPR32RegClass);
-  auto Copy = MIB.buildCopy({NarrowReg}, {ExtReg});
+  auto Ty = MRI.getType(Reg);
+  assert(!Ty.isVector() && "Expected scalars only!");
+  if (Ty.getSizeInBits() == TRI.getRegSizeInBits(RC))
+    return Reg;
 
-  // Select the copy into a subregister copy.
+  // Create a copy and immediately select it.
+  // FIXME: We should have an emitCopy function?
+  auto Copy = MIB.buildCopy({&RC}, {Reg});
   selectCopy(*Copy, TII, MRI, TRI, RBI);
   return Copy.getReg(0);
 }
 
-Register AArch64InstructionSelector::widenGPRBankRegIfNeeded(
-    Register Reg, unsigned WideSize, MachineIRBuilder &MIB) const {
-  assert(WideSize >= 8 && "WideSize is smaller than all possible registers?");
-  MachineRegisterInfo &MRI = *MIB.getMRI();
-  unsigned NarrowSize = MRI.getType(Reg).getSizeInBits();
-  assert(WideSize >= NarrowSize &&
-         "WideSize cannot be smaller than NarrowSize!");
-
-  // If the sizes match, just return the register.
-  //
-  // If NarrowSize is an s1, then we can select it to any size, so we'll treat
-  // it as a don't care.
-  if (NarrowSize == WideSize || NarrowSize == 1)
-    return Reg;
-
-  // Now check the register classes.
-  const RegisterBank *RB = RBI.getRegBank(Reg, MRI, TRI);
-  const TargetRegisterClass *OrigRC = getMinClassForRegBank(*RB, NarrowSize);
-  const TargetRegisterClass *WideRC = getMinClassForRegBank(*RB, WideSize);
-  assert(OrigRC && "Could not determine narrow RC?");
-  assert(WideRC && "Could not determine wide RC?");
-
-  // If the sizes differ, but the register classes are the same, there is no
-  // need to insert a SUBREG_TO_REG.
-  //
-  // For example, an s8 that's supposed to be a GPR will be selected to either
-  // a GPR32 or a GPR64 register. Note that this assumes that the s8 will
-  // always end up on a GPR32.
-  if (OrigRC == WideRC)
-    return Reg;
-
-  // We have two different register classes. Insert a SUBREG_TO_REG.
-  unsigned SubReg = 0;
-  getSubRegForClass(OrigRC, TRI, SubReg);
-  assert(SubReg && "Couldn't determine subregister?");
-
-  // Build the SUBREG_TO_REG and return the new, widened register.
-  auto SubRegToReg =
-      MIB.buildInstr(AArch64::SUBREG_TO_REG, {WideRC}, {})
-          .addImm(0)
-          .addUse(Reg)
-          .addImm(SubReg);
-  constrainSelectedInstRegOperands(*SubRegToReg, TII, TRI, RBI);
-  return SubRegToReg.getReg(0);
-}
-
 /// Select an "extended register" operand. This operand folds in an extend
 /// followed by an optional left shift.
 InstructionSelector::ComplexRendererFns
@@ -5542,7 +5510,7 @@ AArch64InstructionSelector::selectArithExtendedRegister(
   // We require a GPR32 here. Narrow the ExtReg if needed using a subregister
   // copy.
   MachineIRBuilder MIB(*RootDef);
-  ExtReg = narrowExtendRegIfNeeded(ExtReg, MIB);
+  ExtReg = moveScalarRegClass(ExtReg, AArch64::GPR32RegClass, MIB);
 
   return {{[=](MachineInstrBuilder &MIB) { MIB.addUse(ExtReg); },
            [=](MachineInstrBuilder &MIB) {

llvm/test/CodeGen/AArch64/GlobalISel/load-wro-addressing-modes.mir

@@ -428,3 +428,39 @@ body:             |
     $x1 = COPY %load(s64)
     RET_ReallyLR implicit $x1
 ...
+---
+name:            zext_shl_LDRWroW
+alignment:       4
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+liveins:
+  - { reg: '$w0' }
+  - { reg: '$x1' }
+body:             |
+  bb.1:
+    liveins: $w0, $x1
+
+    ; We try to look through the G_ZEXT of the SHL here.
+
+    ; CHECK-LABEL: name: zext_shl_LDRWroW
+    ; CHECK: liveins: $w0, $x1
+    ; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
+    ; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
+    ; CHECK: [[ANDWri:%[0-9]+]]:gpr32common = ANDWri [[COPY]], 7
+    ; CHECK: [[LDRWroW:%[0-9]+]]:gpr32 = LDRWroW [[COPY1]], [[ANDWri]], 0, 1 :: (load 4)
+    ; CHECK: $w0 = COPY [[LDRWroW]]
+    ; CHECK: RET_ReallyLR implicit $w0
+    %0:gpr(s32) = COPY $w0
+    %1:gpr(p0) = COPY $x1
+    %2:gpr(s32) = G_CONSTANT i32 255
+    %3:gpr(s32) = G_AND %0, %2
+    %13:gpr(s64) = G_CONSTANT i64 2
+    %12:gpr(s32) = G_SHL %3, %13(s64)
+    %6:gpr(s64) = G_ZEXT %12(s32)
+    %7:gpr(p0) = G_PTR_ADD %1, %6(s64)
+    %9:gpr(s32) = G_LOAD %7(p0) :: (load 4)
+    $w0 = COPY %9(s32)
+    RET_ReallyLR implicit $w0
+
+...

llvm/test/CodeGen/AArch64/GlobalISel/opt-fold-ext-tbz-tbnz.mir

@@ -78,8 +78,9 @@ body:             |
   ; CHECK:   successors: %bb.0(0x40000000), %bb.1(0x40000000)
   ; CHECK:   liveins: $h0
   ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:fpr32 = SUBREG_TO_REG 0, $h0, %subreg.hsub
-  ; CHECK:   %copy:gpr32 = COPY [[SUBREG_TO_REG]]
-  ; CHECK:   TBNZW %copy, 3, %bb.1
+  ; CHECK:   %copy:gpr32all = COPY [[SUBREG_TO_REG]]
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr32 = COPY %copy
+  ; CHECK:   TBNZW [[COPY]], 3, %bb.1
   ; CHECK:   B %bb.0
   ; CHECK: bb.1:
   ; CHECK:   RET_ReallyLR

llvm/test/CodeGen/AArch64/GlobalISel/subreg-copy.mir

@@ -34,3 +34,35 @@ body: |
   bb.1:
     RET_ReallyLR
 ...
+---
+name:              no_trunc
+alignment:         4
+legalized:         true
+regBankSelected:   true
+tracksRegLiveness: true
+body: |
+  ; CHECK-LABEL: name: no_trunc
+  ; CHECK: bb.0:
+  ; CHECK:   successors: %bb.1(0x80000000)
+  ; CHECK:   liveins: $x0
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
+  ; CHECK:   [[LDRQui:%[0-9]+]]:fpr128 = LDRQui [[COPY]], 0 :: (load 16)
+  ; CHECK:   [[COPY1:%[0-9]+]]:gpr64all = COPY [[LDRQui]].dsub
+  ; CHECK:   [[COPY2:%[0-9]+]]:gpr64 = COPY [[COPY1]]
+  ; CHECK:   TBNZX [[COPY2]], 33, %bb.1
+  ; CHECK: bb.1:
+  ; CHECK:   RET_ReallyLR
+  bb.0:
+    liveins: $x0
+    %1:gpr(p0) = COPY $x0
+    %3:gpr(s64) = G_CONSTANT i64 8589934592
+    %5:gpr(s64) = G_CONSTANT i64 0
+    %0:fpr(s128) = G_LOAD %1:gpr(p0) :: (load 16)
+    %2:fpr(s64) = G_TRUNC %0:fpr(s128)
+    %8:gpr(s64) = COPY %2:fpr(s64)
+    %4:gpr(s64) = G_AND %8:gpr, %3:gpr
+    %7:gpr(s32) = G_ICMP intpred(ne), %4:gpr(s64), %5:gpr
+    %6:gpr(s1) = G_TRUNC %7:gpr(s32)
+    G_BRCOND %6:gpr(s1), %bb.1
+  bb.1:
+    RET_ReallyLR

llvm/test/CodeGen/AArch64/GlobalISel/widen-narrow-tbz-tbnz.mir

@@ -106,8 +106,9 @@ body:             |
   ; CHECK:   successors: %bb.0(0x40000000), %bb.1(0x40000000)
   ; CHECK:   liveins: $w0
   ; CHECK:   %reg:gpr32all = COPY $w0
-  ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64 = SUBREG_TO_REG 0, %reg, %subreg.sub_32
-  ; CHECK:   TBZX [[SUBREG_TO_REG]], 33, %bb.1
+  ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, %reg, %subreg.sub_32
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr64 = COPY [[SUBREG_TO_REG]]
+  ; CHECK:   TBZX [[COPY]], 33, %bb.1
   ; CHECK:   B %bb.0
   ; CHECK: bb.1:
   ; CHECK:   RET_ReallyLR
@@ -140,8 +141,9 @@ body:             |
   ; CHECK: bb.0:
   ; CHECK:   successors: %bb.0(0x40000000), %bb.1(0x40000000)
   ; CHECK:   %reg:gpr32 = IMPLICIT_DEF
-  ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64 = SUBREG_TO_REG 0, %reg, %subreg.sub_32
-  ; CHECK:   TBZX [[SUBREG_TO_REG]], 33, %bb.1
+  ; CHECK:   [[SUBREG_TO_REG:%[0-9]+]]:gpr64all = SUBREG_TO_REG 0, %reg, %subreg.sub_32
+  ; CHECK:   [[COPY:%[0-9]+]]:gpr64 = COPY [[SUBREG_TO_REG]]
+  ; CHECK:   TBZX [[COPY]], 33, %bb.1
   ; CHECK:   B %bb.0
   ; CHECK: bb.1:
   ; CHECK:   RET_ReallyLR