AArch64: treat [N x Ty] as a block during procedure calls.
The AAPCS treats small structs and homogeneous floating (or vector) aggregates
specially, and guarantees they either get passed as a contiguous block of
registers, or prevent any future use of those registers and get passed on the
stack.
This concept can fit quite neatly into LLVM's own type system, mapping an HFA
to [N x float] and so on, and small structs to [N x i64]. Doing so allows
front-ends to emit AAPCS compliant code without having to duplicate the
register counting logic.
llvm-svn: 222903
diff --git a/llvm/lib/Target/AArch64/AArch64CallingConvention.h b/llvm/lib/Target/AArch64/AArch64CallingConvention.h
new file mode 100644
index 0000000..b2be99e
--- /dev/null
+++ b/llvm/lib/Target/AArch64/AArch64CallingConvention.h
@@ -0,0 +1,136 @@
+//=== AArch64CallingConv.h - Custom Calling Convention Routines -*- C++ -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the custom routines for the AArch64 Calling Convention
+// that aren't done by tablegen.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_LIB_TARGET_AARCH64_AARCH64CALLINGCONVENTION_H
+#define LLVM_LIB_TARGET_AARCH64_AARCH64CALLINGCONVENTION_H
+
+#include "AArch64.h"
+#include "AArch64InstrInfo.h"
+#include "AArch64Subtarget.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+#include "llvm/IR/CallingConv.h"
+#include "llvm/Target/TargetInstrInfo.h"
+
+namespace {
+using namespace llvm;
+
+static const uint16_t XRegList[] = {AArch64::X0, AArch64::X1, AArch64::X2,
+ AArch64::X3, AArch64::X4, AArch64::X5,
+ AArch64::X6, AArch64::X7};
+static const uint16_t SRegList[] = {AArch64::S0, AArch64::S1, AArch64::S2,
+ AArch64::S3, AArch64::S4, AArch64::S5,
+ AArch64::S6, AArch64::S7};
+static const uint16_t DRegList[] = {AArch64::D0, AArch64::D1, AArch64::D2,
+ AArch64::D3, AArch64::D4, AArch64::D5,
+ AArch64::D6, AArch64::D7};
+static const uint16_t QRegList[] = {AArch64::Q0, AArch64::Q1, AArch64::Q2,
+ AArch64::Q3, AArch64::Q4, AArch64::Q5,
+ AArch64::Q6, AArch64::Q7};
+
+static bool finishStackBlock(SmallVectorImpl<CCValAssign> &PendingMembers,
+ MVT LocVT, ISD::ArgFlagsTy &ArgFlags,
+ CCState &State, unsigned SlotAlign) {
+ unsigned Size = LocVT.getSizeInBits() / 8;
+ unsigned StackAlign = State.getMachineFunction()
+ .getSubtarget()
+ .getDataLayout()
+ ->getStackAlignment();
+ unsigned Align = std::min(ArgFlags.getOrigAlign(), StackAlign);
+
+ for (auto &It : PendingMembers) {
+ It.convertToMem(State.AllocateStack(Size, std::max(Align, SlotAlign)));
+ State.addLoc(It);
+ SlotAlign = 1;
+ }
+
+ // All pending members have now been allocated
+ PendingMembers.clear();
+ return true;
+}
+
+/// The Darwin variadic PCS places anonymous arguments in 8-byte stack slots. An
+/// [N x Ty] type must still be contiguous in memory though.
+static bool CC_AArch64_Custom_Stack_Block(
+ unsigned &ValNo, MVT &ValVT, MVT &LocVT, CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags, CCState &State) {
+ SmallVectorImpl<CCValAssign> &PendingMembers = State.getPendingLocs();
+
+ // Add the argument to the list to be allocated once we know the size of the
+ // block.
+ PendingMembers.push_back(
+ CCValAssign::getPending(ValNo, ValVT, LocVT, LocInfo));
+
+ if (!ArgFlags.isInConsecutiveRegsLast())
+ return true;
+
+ return finishStackBlock(PendingMembers, LocVT, ArgFlags, State, 8);
+}
+
+/// Given an [N x Ty] block, it should be passed in a consecutive sequence of
+/// registers. If no such sequence is available, mark the rest of the registers
+/// of that type as used and place the argument on the stack.
+static bool CC_AArch64_Custom_Block(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
+ CCValAssign::LocInfo &LocInfo,
+ ISD::ArgFlagsTy &ArgFlags, CCState &State) {
+ // Try to allocate a contiguous block of registers, each of the correct
+ // size to hold one member.
+ ArrayRef<const uint16_t> RegList;
+ if (LocVT.SimpleTy == MVT::i64)
+ RegList = XRegList;
+ else if (LocVT.SimpleTy == MVT::f32)
+ RegList = SRegList;
+ else if (LocVT.SimpleTy == MVT::f64)
+ RegList = DRegList;
+ else if (LocVT.SimpleTy == MVT::v2f64)
+ RegList = QRegList;
+ else {
+ // Not an array we want to split up after all.
+ return false;
+ }
+
+ SmallVectorImpl<CCValAssign> &PendingMembers = State.getPendingLocs();
+
+ // Add the argument to the list to be allocated once we know the size of the
+ // block.
+ PendingMembers.push_back(
+ CCValAssign::getPending(ValNo, ValVT, LocVT, LocInfo));
+
+ if (!ArgFlags.isInConsecutiveRegsLast())
+ return true;
+
+ unsigned RegResult = State.AllocateRegBlock(RegList, PendingMembers.size());
+ if (RegResult) {
+ for (auto &It : PendingMembers) {
+ It.convertToReg(RegResult);
+ State.addLoc(It);
+ ++RegResult;
+ }
+ PendingMembers.clear();
+ return true;
+ }
+
+ // Mark all regs in the class as unavailable
+ for (auto Reg : RegList)
+ State.AllocateReg(Reg);
+
+ const AArch64Subtarget &Subtarget = static_cast<const AArch64Subtarget &>(
+ State.getMachineFunction().getSubtarget());
+ unsigned SlotAlign = Subtarget.isTargetDarwin() ? 1 : 8;
+
+ return finishStackBlock(PendingMembers, LocVT, ArgFlags, State, SlotAlign);
+}
+
+}
+
+#endif
diff --git a/llvm/lib/Target/AArch64/AArch64CallingConvention.td b/llvm/lib/Target/AArch64/AArch64CallingConvention.td
index 9e707e4..1a80402 100644
--- a/llvm/lib/Target/AArch64/AArch64CallingConvention.td
+++ b/llvm/lib/Target/AArch64/AArch64CallingConvention.td
@@ -40,6 +40,8 @@
// slot is 64-bit.
CCIfByVal<CCPassByVal<8, 8>>,
+ CCIfConsecutiveRegs<CCCustom<"CC_AArch64_Custom_Block">>,
+
// Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers,
// up to eight each of GPR and FPR.
CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
@@ -119,6 +121,8 @@
// slot is 64-bit.
CCIfByVal<CCPassByVal<8, 8>>,
+ CCIfConsecutiveRegs<CCCustom<"CC_AArch64_Custom_Block">>,
+
// Handle i1, i8, i16, i32, i64, f32, f64 and v2f64 by passing in registers,
// up to eight each of GPR and FPR.
CCIfType<[i1, i8, i16], CCPromoteToType<i32>>,
@@ -159,6 +163,8 @@
CCIfType<[v2f32], CCBitConvertToType<v2i32>>,
CCIfType<[v2f64, v4f32, f128], CCBitConvertToType<v2i64>>,
+ CCIfConsecutiveRegs<CCCustom<"CC_AArch64_Custom_Stack_Block">>,
+
// Handle all scalar types as either i64 or f64.
CCIfType<[i8, i16, i32], CCPromoteToType<i64>>,
CCIfType<[f16, f32], CCPromoteToType<f64>>,
diff --git a/llvm/lib/Target/AArch64/AArch64FastISel.cpp b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
index fb0326b..d942ace 100644
--- a/llvm/lib/Target/AArch64/AArch64FastISel.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FastISel.cpp
@@ -14,6 +14,7 @@
//===----------------------------------------------------------------------===//
#include "AArch64.h"
+#include "AArch64CallingConvention.h"
#include "AArch64Subtarget.h"
#include "AArch64TargetMachine.h"
#include "MCTargetDesc/AArch64AddressingModes.h"
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 70cf759..6da468e 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -12,6 +12,7 @@
//===----------------------------------------------------------------------===//
#include "AArch64ISelLowering.h"
+#include "AArch64CallingConvention.h"
#include "AArch64MachineFunctionInfo.h"
#include "AArch64PerfectShuffle.h"
#include "AArch64Subtarget.h"
@@ -8842,3 +8843,8 @@
Val, Stxr->getFunctionType()->getParamType(0)),
Addr);
}
+
+bool AArch64TargetLowering::functionArgumentNeedsConsecutiveRegisters(
+ Type *Ty, CallingConv::ID CallConv, bool isVarArg) const {
+ return Ty->isArrayTy();
+}
diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.h b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
index c76f6a8..458489b 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -473,6 +473,10 @@
void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
SelectionDAG &DAG) const override;
+
+ bool functionArgumentNeedsConsecutiveRegisters(Type *Ty,
+ CallingConv::ID CallConv,
+ bool isVarArg) const;
};
namespace AArch64 {
diff --git a/llvm/lib/Target/ARM/ARMCallingConv.h b/llvm/lib/Target/ARM/ARMCallingConv.h
index bd07236..4567a2a 100644
--- a/llvm/lib/Target/ARM/ARMCallingConv.h
+++ b/llvm/lib/Target/ARM/ARMCallingConv.h
@@ -194,20 +194,16 @@
// Try to allocate a contiguous block of registers, each of the correct
// size to hold one member.
- const uint16_t *RegList;
- unsigned NumRegs;
+ ArrayRef<const uint16_t> RegList;
switch (LocVT.SimpleTy) {
case MVT::f32:
RegList = SRegList;
- NumRegs = 16;
break;
case MVT::f64:
RegList = DRegList;
- NumRegs = 8;
break;
case MVT::v2f64:
RegList = QRegList;
- NumRegs = 4;
break;
default:
llvm_unreachable("Unexpected member type for HA");
@@ -215,7 +211,7 @@
}
unsigned RegResult =
- State.AllocateRegBlock(RegList, NumRegs, PendingHAMembers.size());
+ State.AllocateRegBlock(RegList, PendingHAMembers.size());
if (RegResult) {
for (SmallVectorImpl<CCValAssign>::iterator It = PendingHAMembers.begin();