freebsd-nq/contrib/llvm/patches/patch-08-llvm-r230348-arm-fix-bad-ha.diff

Pull in r230348 from upstream llvm trunk (by Tim Northover):

  ARM: treat [N x i32] and [N x i64] as AAPCS composite types

  The logic is almost there already, with our special homogeneous
  aggregate handling. Tweaking it like this allows front-ends to emit
  AAPCS compliant code without ever having to count registers or add
  discarded padding arguments.

  Only arrays of i32 and i64 are needed to model AAPCS rules, but I
  decided to apply the logic to all integer arrays for more consistency.

This fixes a possible "Unexpected member type for HA" error when
compiling lib/msun/bsdsrc/b_tgamma.c for armv6.

Reported by:	Jakub Palider <jpa@semihalf.com>

Introduced here: https://svnweb.freebsd.org/changeset/base/280400

Index: include/llvm/CodeGen/CallingConvLower.h
===================================================================
--- include/llvm/CodeGen/CallingConvLower.h
+++ include/llvm/CodeGen/CallingConvLower.h
@@ -122,8 +122,8 @@ class CCValAssign {
   // There is no need to differentiate between a pending CCValAssign and other
   // kinds, as they are stored in a different list.
   static CCValAssign getPending(unsigned ValNo, MVT ValVT, MVT LocVT,
-                                LocInfo HTP) {
-    return getReg(ValNo, ValVT, 0, LocVT, HTP);
+                                LocInfo HTP, unsigned ExtraInfo = 0) {
+    return getReg(ValNo, ValVT, ExtraInfo, LocVT, HTP);
   }
 
   void convertToReg(unsigned RegNo) {
@@ -146,6 +146,7 @@ class CCValAssign {
 
   unsigned getLocReg() const { assert(isRegLoc()); return Loc; }
   unsigned getLocMemOffset() const { assert(isMemLoc()); return Loc; }
+  unsigned getExtraInfo() const { return Loc; }
   MVT getLocVT() const { return LocVT; }
 
   LocInfo getLocInfo() const { return HTP; }
Index: lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
===================================================================
--- lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -7429,11 +7429,8 @@ TargetLowering::LowerCallTo(TargetLowering::CallLo
       }
       if (Args[i].isNest)
         Flags.setNest();
-      if (NeedsRegBlock) {
+      if (NeedsRegBlock)
         Flags.setInConsecutiveRegs();
-        if (Value == NumValues - 1)
-          Flags.setInConsecutiveRegsLast();
-      }
       Flags.setOrigAlign(OriginalAlignment);
 
       MVT PartVT = getRegisterType(CLI.RetTy->getContext(), VT);
@@ -7482,6 +7479,9 @@ TargetLowering::LowerCallTo(TargetLowering::CallLo
         CLI.Outs.push_back(MyFlags);
         CLI.OutVals.push_back(Parts[j]);
       }
+
+      if (NeedsRegBlock && Value == NumValues - 1)
+        CLI.Outs[CLI.Outs.size() - 1].Flags.setInConsecutiveRegsLast();
     }
   }
 
@@ -7697,11 +7697,8 @@ void SelectionDAGISel::LowerArguments(const Functi
       }
       if (F.getAttributes().hasAttribute(Idx, Attribute::Nest))
         Flags.setNest();
-      if (NeedsRegBlock) {
+      if (NeedsRegBlock)
         Flags.setInConsecutiveRegs();
-        if (Value == NumValues - 1)
-          Flags.setInConsecutiveRegsLast();
-      }
       Flags.setOrigAlign(OriginalAlignment);
 
       MVT RegisterVT = TLI->getRegisterType(*CurDAG->getContext(), VT);
@@ -7716,6 +7713,8 @@ void SelectionDAGISel::LowerArguments(const Functi
           MyFlags.Flags.setOrigAlign(1);
         Ins.push_back(MyFlags);
       }
+      if (NeedsRegBlock && Value == NumValues - 1)
+        Ins[Ins.size() - 1].Flags.setInConsecutiveRegsLast();
       PartBase += VT.getStoreSize();
     }
   }
Index: lib/Target/ARM/ARMCallingConv.h
===================================================================
--- lib/Target/ARM/ARMCallingConv.h
+++ lib/Target/ARM/ARMCallingConv.h
@@ -160,6 +160,8 @@ static bool RetCC_ARM_AAPCS_Custom_f64(unsigned &V
                                    State);
 }
 
+static const uint16_t RRegList[] = { ARM::R0,  ARM::R1,  ARM::R2,  ARM::R3 };
+
 static const uint16_t SRegList[] = { ARM::S0,  ARM::S1,  ARM::S2,  ARM::S3,
                                      ARM::S4,  ARM::S5,  ARM::S6,  ARM::S7,
                                      ARM::S8,  ARM::S9,  ARM::S10, ARM::S11,
@@ -168,81 +170,114 @@ static const uint16_t DRegList[] = { ARM::D0, ARM:
                                      ARM::D4, ARM::D5, ARM::D6, ARM::D7 };
 static const uint16_t QRegList[] = { ARM::Q0, ARM::Q1, ARM::Q2, ARM::Q3 };
 
+
 // Allocate part of an AAPCS HFA or HVA. We assume that each member of the HA
 // has InConsecutiveRegs set, and that the last member also has
 // InConsecutiveRegsLast set. We must process all members of the HA before
 // we can allocate it, as we need to know the total number of registers that
 // will be needed in order to (attempt to) allocate a contiguous block.
-static bool CC_ARM_AAPCS_Custom_HA(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
-                                   CCValAssign::LocInfo &LocInfo,
-                                   ISD::ArgFlagsTy &ArgFlags, CCState &State) {
-  SmallVectorImpl<CCValAssign> &PendingHAMembers = State.getPendingLocs();
+static bool CC_ARM_AAPCS_Custom_Aggregate(unsigned &ValNo, MVT &ValVT,
+                                          MVT &LocVT,
+                                          CCValAssign::LocInfo &LocInfo,
+                                          ISD::ArgFlagsTy &ArgFlags,
+                                          CCState &State) {
+  SmallVectorImpl<CCValAssign> &PendingMembers = State.getPendingLocs();
 
   // AAPCS HFAs must have 1-4 elements, all of the same type
-  assert(PendingHAMembers.size() < 4);
-  if (PendingHAMembers.size() > 0)
-    assert(PendingHAMembers[0].getLocVT() == LocVT);
+  if (PendingMembers.size() > 0)
+    assert(PendingMembers[0].getLocVT() == LocVT);
 
   // Add the argument to the list to be allocated once we know the size of the
-  // HA
-  PendingHAMembers.push_back(
-      CCValAssign::getPending(ValNo, ValVT, LocVT, LocInfo));
+  // aggregate. Store the type's required alignmnent as extra info for later: in
+  // the [N x i64] case all trace has been removed by the time we actually get
+  // to do allocation.
+  PendingMembers.push_back(CCValAssign::getPending(ValNo, ValVT, LocVT, LocInfo,
+                                                   ArgFlags.getOrigAlign()));
 
-  if (ArgFlags.isInConsecutiveRegsLast()) {
-    assert(PendingHAMembers.size() > 0 && PendingHAMembers.size() <= 4 &&
-           "Homogeneous aggregates must have between 1 and 4 members");
+  if (!ArgFlags.isInConsecutiveRegsLast())
+    return true;
 
-    // Try to allocate a contiguous block of registers, each of the correct
-    // size to hold one member.
-    ArrayRef<uint16_t> RegList;
-    switch (LocVT.SimpleTy) {
-    case MVT::f32:
-      RegList = SRegList;
-      break;
-    case MVT::f64:
-      RegList = DRegList;
-      break;
-    case MVT::v2f64:
-      RegList = QRegList;
-      break;
-    default:
-      llvm_unreachable("Unexpected member type for HA");
-      break;
-    }
+  // Try to allocate a contiguous block of registers, each of the correct
+  // size to hold one member.
+  unsigned Align = std::min(PendingMembers[0].getExtraInfo(), 8U);
 
-    unsigned RegResult =
-        State.AllocateRegBlock(RegList, PendingHAMembers.size());
+  ArrayRef<uint16_t> RegList;
+  switch (LocVT.SimpleTy) {
+  case MVT::i32: {
+    RegList = RRegList;
+    unsigned RegIdx = State.getFirstUnallocated(RegList.data(), RegList.size());
 
-    if (RegResult) {
-      for (SmallVectorImpl<CCValAssign>::iterator It = PendingHAMembers.begin();
-           It != PendingHAMembers.end(); ++It) {
-        It->convertToReg(RegResult);
-        State.addLoc(*It);
-        ++RegResult;
-      }
-      PendingHAMembers.clear();
-      return true;
-    }
+    // First consume all registers that would give an unaligned object. Whether
+    // we go on stack or in regs, no-one will be using them in future.
+    unsigned RegAlign = RoundUpToAlignment(Align, 4) / 4;
+    while (RegIdx % RegAlign != 0 && RegIdx < RegList.size())
+      State.AllocateReg(RegList[RegIdx++]);
 
-    // Register allocation failed, fall back to the stack
+    break;
+  }
+  case MVT::f32:
+    RegList = SRegList;
+    break;
+  case MVT::f64:
+    RegList = DRegList;
+    break;
+  case MVT::v2f64:
+    RegList = QRegList;
+    break;
+  default:
+    llvm_unreachable("Unexpected member type for block aggregate");
+    break;
+  }
 
-    // Mark all VFP regs as unavailable (AAPCS rule C.2.vfp)
-    for (unsigned regNo = 0; regNo < 16; ++regNo)
-      State.AllocateReg(SRegList[regNo]);
+  unsigned RegResult = State.AllocateRegBlock(RegList, PendingMembers.size());
+  if (RegResult) {
+    for (SmallVectorImpl<CCValAssign>::iterator It = PendingMembers.begin();
+         It != PendingMembers.end(); ++It) {
+      It->convertToReg(RegResult);
+      State.addLoc(*It);
+      ++RegResult;
+    }
+    PendingMembers.clear();
+    return true;
+  }
 
-    unsigned Size = LocVT.getSizeInBits() / 8;
-    unsigned Align = std::min(Size, 8U);
+  // Register allocation failed, we'll be needing the stack
+  unsigned Size = LocVT.getSizeInBits() / 8;
+  if (LocVT == MVT::i32 && State.getNextStackOffset() == 0) {
+    // If nothing else has used the stack until this point, a non-HFA aggregate
+    // can be split between regs and stack.
+    unsigned RegIdx = State.getFirstUnallocated(RegList.data(), RegList.size());
+    for (auto &It : PendingMembers) {
+      if (RegIdx >= RegList.size())
+        It.convertToMem(State.AllocateStack(Size, Size));
+      else
+        It.convertToReg(State.AllocateReg(RegList[RegIdx++]));
 
-    for (auto It : PendingHAMembers) {
-      It.convertToMem(State.AllocateStack(Size, Align));
       State.addLoc(It);
     }
+    PendingMembers.clear();
+    return true;
+  } else if (LocVT != MVT::i32)
+    RegList = SRegList;
 
-    // All pending members have now been allocated
-    PendingHAMembers.clear();
+  // Mark all regs as unavailable (AAPCS rule C.2.vfp for VFP, C.6 for core)
+  for (auto Reg : RegList)
+    State.AllocateReg(Reg);
+
+  for (auto &It : PendingMembers) {
+    It.convertToMem(State.AllocateStack(Size, Align));
+    State.addLoc(It);
+
+    // After the first item has been allocated, the rest are packed as tightly
+    // as possible. (E.g. an incoming i64 would have starting Align of 8, but
+    // we'll be allocating a bunch of i32 slots).
+    Align = Size;
   }
 
-  // This will be allocated by the last member of the HA
+  // All pending members have now been allocated
+  PendingMembers.clear();
+
+  // This will be allocated by the last member of the aggregate
   return true;
 }
 
Index: lib/Target/ARM/ARMCallingConv.td
===================================================================
--- lib/Target/ARM/ARMCallingConv.td
+++ lib/Target/ARM/ARMCallingConv.td
@@ -175,7 +175,7 @@ def CC_ARM_AAPCS_VFP : CallingConv<[
   CCIfType<[v2i64, v4i32, v8i16, v16i8, v4f32], CCBitConvertToType<v2f64>>,
 
   // HFAs are passed in a contiguous block of registers, or on the stack
-  CCIfConsecutiveRegs<CCCustom<"CC_ARM_AAPCS_Custom_HA">>,
+  CCIfConsecutiveRegs<CCCustom<"CC_ARM_AAPCS_Custom_Aggregate">>,
 
   CCIfType<[v2f64], CCAssignToReg<[Q0, Q1, Q2, Q3]>>,
   CCIfType<[f64], CCAssignToReg<[D0, D1, D2, D3, D4, D5, D6, D7]>>,
Index: lib/Target/ARM/ARMISelLowering.cpp
===================================================================
--- lib/Target/ARM/ARMISelLowering.cpp
+++ lib/Target/ARM/ARMISelLowering.cpp
@@ -11285,7 +11285,9 @@ static bool isHomogeneousAggregate(Type *Ty, HABas
   return (Members > 0 && Members <= 4);
 }
 
-/// \brief Return true if a type is an AAPCS-VFP homogeneous aggregate.
+/// \brief Return true if a type is an AAPCS-VFP homogeneous aggregate or one of
+/// [N x i32] or [N x i64]. This allows front-ends to skip emitting padding when
+/// passing according to AAPCS rules.
 bool ARMTargetLowering::functionArgumentNeedsConsecutiveRegisters(
     Type *Ty, CallingConv::ID CallConv, bool isVarArg) const {
   if (getEffectiveCallingConv(CallConv, isVarArg) !=
@@ -11294,7 +11296,9 @@ bool ARMTargetLowering::functionArgumentNeedsConse
 
   HABaseType Base = HA_UNKNOWN;
   uint64_t Members = 0;
-  bool result = isHomogeneousAggregate(Ty, Base, Members);
-  DEBUG(dbgs() << "isHA: " << result << " "; Ty->dump());
-  return result;
+  bool IsHA = isHomogeneousAggregate(Ty, Base, Members);
+  DEBUG(dbgs() << "isHA: " << IsHA << " "; Ty->dump());
+
+  bool IsIntArray = Ty->isArrayTy() && Ty->getArrayElementType()->isIntegerTy();
+  return IsHA || IsIntArray;
 }
Index: test/CodeGen/ARM/aggregate-padding.ll
===================================================================
--- test/CodeGen/ARM/aggregate-padding.ll
+++ test/CodeGen/ARM/aggregate-padding.ll
@@ -0,0 +1,101 @@
+; RUN: llc -mtriple=armv7-linux-gnueabihf %s -o - | FileCheck %s
+
+; [2 x i64] should be contiguous when split (e.g. we shouldn't try to align all
+; i32 components to 64 bits). Also makes sure i64 based types are properly
+; aligned on the stack.
+define i64 @test_i64_contiguous_on_stack([8 x double], float, i32 %in, [2 x i64] %arg) nounwind {
+; CHECK-LABEL: test_i64_contiguous_on_stack:
+; CHECK-DAG: ldr [[LO0:r[0-9]+]], [sp, #8]
+; CHECK-DAG: ldr [[HI0:r[0-9]+]], [sp, #12]
+; CHECK-DAG: ldr [[LO1:r[0-9]+]], [sp, #16]
+; CHECK-DAG: ldr [[HI1:r[0-9]+]], [sp, #20]
+; CHECK: adds r0, [[LO0]], [[LO1]]
+; CHECK: adc r1, [[HI0]], [[HI1]]
+
+  %val1 = extractvalue [2 x i64] %arg, 0
+  %val2 = extractvalue [2 x i64] %arg, 1
+  %sum = add i64 %val1, %val2
+  ret i64 %sum
+}
+
+; [2 x i64] should try to use looks for 4 regs, not 8 (which might happen if the
+; i64 -> i32, i32 split wasn't handled correctly).
+define i64 @test_2xi64_uses_4_regs([8 x double], float, [2 x i64] %arg) nounwind {
+; CHECK-LABEL: test_2xi64_uses_4_regs:
+; CHECK-DAG: mov r0, r2
+; CHECK-DAG: mov r1, r3
+
+  %val = extractvalue [2 x i64] %arg, 1
+  ret i64 %val
+}
+
+; An aggregate should be able to split between registers and stack if there is
+; nothing else on the stack.
+define i32 @test_aggregates_split([8 x double], i32, [4 x i32] %arg) nounwind {
+; CHECK-LABEL: test_aggregates_split:
+; CHECK: ldr [[VAL3:r[0-9]+]], [sp]
+; CHECK: add r0, r1, [[VAL3]]
+
+  %val0 = extractvalue [4 x i32] %arg, 0
+  %val3 = extractvalue [4 x i32] %arg, 3
+  %sum = add i32 %val0, %val3
+  ret i32 %sum
+}
+
+; If an aggregate has to be moved entirely onto the stack, nothing should be
+; able to use r0-r3 any more. Also checks that [2 x i64] properly aligned when
+; it uses regs.
+define i32 @test_no_int_backfilling([8 x double], float, i32, [2 x i64], i32 %arg) nounwind {
+; CHECK-LABEL: test_no_int_backfilling:
+; CHECK: ldr r0, [sp, #24]
+  ret i32 %arg
+}
+
+; Even if the argument was successfully allocated as reg block, there should be
+; no backfillig to r1.
+define i32 @test_no_int_backfilling_regsonly(i32, [1 x i64], i32 %arg) {
+; CHECK-LABEL: test_no_int_backfilling_regsonly:
+; CHECK: ldr r0, [sp]
+  ret i32 %arg
+}
+
+; If an aggregate has to be moved entirely onto the stack, nothing should be
+; able to use r0-r3 any more.
+define float @test_no_float_backfilling([7 x double], [4 x i32], i32, [4 x double], float %arg) nounwind {
+; CHECK-LABEL: test_no_float_backfilling:
+; CHECK: vldr s0, [sp, #40]
+  ret float %arg
+}
+
+; They're a bit pointless, but types like [N x i8] should work as well.
+define i8 @test_i8_in_regs(i32, [3 x i8] %arg) {
+; CHECK-LABEL: test_i8_in_regs:
+; CHECK: add r0, r1, r3
+  %val0 = extractvalue [3 x i8] %arg, 0
+  %val2 = extractvalue [3 x i8] %arg, 2
+  %sum = add i8 %val0, %val2
+  ret i8 %sum
+}
+
+define i16 @test_i16_split(i32, i32, [3 x i16] %arg) {
+; CHECK-LABEL: test_i16_split:
+; CHECK: ldrh [[VAL2:r[0-9]+]], [sp]
+; CHECK: add r0, r2, [[VAL2]]
+  %val0 = extractvalue [3 x i16] %arg, 0
+  %val2 = extractvalue [3 x i16] %arg, 2
+  %sum = add i16 %val0, %val2
+  ret i16 %sum
+}
+
+; Beware: on the stack each i16 still gets a 32-bit slot, the array is not
+; packed.
+define i16 @test_i16_forced_stack([8 x double], double, i32, i32, [3 x i16] %arg) {
+; CHECK-LABEL: test_i16_forced_stack:
+; CHECK-DAG: ldrh [[VAL0:r[0-9]+]], [sp, #8]
+; CHECK-DAG: ldrh [[VAL2:r[0-9]+]], [sp, #16]
+; CHECK: add r0, [[VAL0]], [[VAL2]]
+  %val0 = extractvalue [3 x i16] %arg, 0
+  %val2 = extractvalue [3 x i16] %arg, 2
+  %sum = add i16 %val0, %val2
+  ret i16 %sum
+}