maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/Target/PowerPC/PPCISelLowering.cpp

16418 lines
641 KiB
C++
Raw Blame History

//===-- PPCISelLowering.cpp - PPC DAG Lowering Implementation -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements the PPCISelLowering class.
//
//===----------------------------------------------------------------------===//
#include "PPCISelLowering.h"
#include "MCTargetDesc/PPCPredicates.h"
#include "PPC.h"
#include "PPCCCState.h"
#include "PPCCallingConv.h"
#include "PPCFrameLowering.h"
#include "PPCInstrInfo.h"
#include "PPCMachineFunctionInfo.h"
#include "PPCPerfectShuffle.h"
#include "PPCRegisterInfo.h"
#include "PPCSubtarget.h"
#include "PPCTargetMachine.h"
#include "llvm/ADT/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringSwitch.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RuntimeLibcalls.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/TargetLoweringObjectFileImpl.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalValue.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/IntrinsicsPowerPC.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/Use.h"
#include "llvm/IR/Value.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCRegisterInfo.h"
#include "llvm/MC/MCSectionXCOFF.h"
#include "llvm/MC/MCSymbolXCOFF.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/Support/BranchProbability.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <iterator>
#include <list>
#include <utility>
#include <vector>
using namespace llvm;
#define DEBUG_TYPE "ppc-lowering"
static cl::opt<bool> DisablePPCPreinc("disable-ppc-preinc",
cl::desc("disable preincrement load/store generation on PPC"), cl::Hidden);
static cl::opt<bool> DisableILPPref("disable-ppc-ilp-pref",
cl::desc("disable setting the node scheduling preference to ILP on PPC"), cl::Hidden);
static cl::opt<bool> DisablePPCUnaligned("disable-ppc-unaligned",
cl::desc("disable unaligned load/store generation on PPC"), cl::Hidden);
static cl::opt<bool> DisableSCO("disable-ppc-sco",
cl::desc("disable sibling call optimization on ppc"), cl::Hidden);
static cl::opt<bool> DisableInnermostLoopAlign32("disable-ppc-innermost-loop-align32",
cl::desc("don't always align innermost loop to 32 bytes on ppc"), cl::Hidden);
static cl::opt<bool> UseAbsoluteJumpTables("ppc-use-absolute-jumptables",
cl::desc("use absolute jump tables on ppc"), cl::Hidden);
// TODO - Remove this option if soft fp128 has been fully supported .
static cl::opt<bool>
EnableSoftFP128("enable-soft-fp128",
cl::desc("temp option to enable soft fp128"), cl::Hidden);
STATISTIC(NumTailCalls, "Number of tail calls");
STATISTIC(NumSiblingCalls, "Number of sibling calls");
STATISTIC(ShufflesHandledWithVPERM, "Number of shuffles lowered to a VPERM");
STATISTIC(NumDynamicAllocaProbed, "Number of dynamic stack allocation probed");
static bool isNByteElemShuffleMask(ShuffleVectorSDNode *, unsigned, int);
static SDValue widenVec(SelectionDAG &DAG, SDValue Vec, const SDLoc &dl);
// FIXME: Remove this once the bug has been fixed!
extern cl::opt<bool> ANDIGlueBug;
PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
const PPCSubtarget &STI)
: TargetLowering(TM), Subtarget(STI) {
// On PPC32/64, arguments smaller than 4/8 bytes are extended, so all
// arguments are at least 4/8 bytes aligned.
bool isPPC64 = Subtarget.isPPC64();
setMinStackArgumentAlignment(isPPC64 ? Align(8) : Align(4));
// Set up the register classes.
addRegisterClass(MVT::i32, &PPC::GPRCRegClass);
if (!useSoftFloat()) {
if (hasSPE()) {
addRegisterClass(MVT::f32, &PPC::GPRCRegClass);
// EFPU2 APU only supports f32
if (!Subtarget.hasEFPU2())
addRegisterClass(MVT::f64, &PPC::SPERCRegClass);
} else {
addRegisterClass(MVT::f32, &PPC::F4RCRegClass);
addRegisterClass(MVT::f64, &PPC::F8RCRegClass);
}
}
// Match BITREVERSE to customized fast code sequence in the td file.
setOperationAction(ISD::BITREVERSE, MVT::i32, Legal);
setOperationAction(ISD::BITREVERSE, MVT::i64, Legal);
// Sub-word ATOMIC_CMP_SWAP need to ensure that the input is zero-extended.
setOperationAction(ISD::ATOMIC_CMP_SWAP, MVT::i32, Custom);
// PowerPC has an i16 but no i8 (or i1) SEXTLOAD.
for (MVT VT : MVT::integer_valuetypes()) {
setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Expand);
}
if (Subtarget.isISA3_0()) {
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Legal);
setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Legal);
setTruncStoreAction(MVT::f64, MVT::f16, Legal);
setTruncStoreAction(MVT::f32, MVT::f16, Legal);
} else {
// No extending loads from f16 or HW conversions back and forth.
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
setTruncStoreAction(MVT::f64, MVT::f16, Expand);
setTruncStoreAction(MVT::f32, MVT::f16, Expand);
}
setTruncStoreAction(MVT::f64, MVT::f32, Expand);
// PowerPC has pre-inc load and store's.
setIndexedLoadAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i8, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i16, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i64, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i8, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i16, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i32, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::i64, Legal);
if (!Subtarget.hasSPE()) {
setIndexedLoadAction(ISD::PRE_INC, MVT::f32, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::f64, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::f32, Legal);
setIndexedStoreAction(ISD::PRE_INC, MVT::f64, Legal);
}
// PowerPC uses ADDC/ADDE/SUBC/SUBE to propagate carry.
const MVT ScalarIntVTs[] = { MVT::i32, MVT::i64 };
for (MVT VT : ScalarIntVTs) {
setOperationAction(ISD::ADDC, VT, Legal);
setOperationAction(ISD::ADDE, VT, Legal);
setOperationAction(ISD::SUBC, VT, Legal);
setOperationAction(ISD::SUBE, VT, Legal);
}
if (Subtarget.useCRBits()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
if (isPPC64 || Subtarget.hasFPCVT()) {
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i1, Promote);
AddPromotedToType(ISD::STRICT_SINT_TO_FP, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i1, Promote);
AddPromotedToType(ISD::STRICT_UINT_TO_FP, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::SINT_TO_FP, MVT::i1, Promote);
AddPromotedToType (ISD::SINT_TO_FP, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::UINT_TO_FP, MVT::i1, Promote);
AddPromotedToType(ISD::UINT_TO_FP, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i1, Promote);
AddPromotedToType(ISD::STRICT_FP_TO_SINT, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i1, Promote);
AddPromotedToType(ISD::STRICT_FP_TO_UINT, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::FP_TO_SINT, MVT::i1, Promote);
AddPromotedToType(ISD::FP_TO_SINT, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
setOperationAction(ISD::FP_TO_UINT, MVT::i1, Promote);
AddPromotedToType(ISD::FP_TO_UINT, MVT::i1,
isPPC64 ? MVT::i64 : MVT::i32);
} else {
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i1, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i1, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::i1, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i1, Custom);
}
// PowerPC does not support direct load/store of condition registers.
setOperationAction(ISD::LOAD, MVT::i1, Custom);
setOperationAction(ISD::STORE, MVT::i1, Custom);
// FIXME: Remove this once the ANDI glue bug is fixed:
if (ANDIGlueBug)
setOperationAction(ISD::TRUNCATE, MVT::i1, Custom);
for (MVT VT : MVT::integer_valuetypes()) {
setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
setTruncStoreAction(VT, MVT::i1, Expand);
}
addRegisterClass(MVT::i1, &PPC::CRBITRCRegClass);
}
// Expand ppcf128 to i32 by hand for the benefit of llvm-gcc bootstrap on
// PPC (the libcall is not available).
setOperationAction(ISD::FP_TO_SINT, MVT::ppcf128, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::ppcf128, Custom);
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::ppcf128, Custom);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::ppcf128, Custom);
// We do not currently implement these libm ops for PowerPC.
setOperationAction(ISD::FFLOOR, MVT::ppcf128, Expand);
setOperationAction(ISD::FCEIL, MVT::ppcf128, Expand);
setOperationAction(ISD::FTRUNC, MVT::ppcf128, Expand);
setOperationAction(ISD::FRINT, MVT::ppcf128, Expand);
setOperationAction(ISD::FNEARBYINT, MVT::ppcf128, Expand);
setOperationAction(ISD::FREM, MVT::ppcf128, Expand);
// PowerPC has no SREM/UREM instructions unless we are on P9
// On P9 we may use a hardware instruction to compute the remainder.
// When the result of both the remainder and the division is required it is
// more efficient to compute the remainder from the result of the division
// rather than use the remainder instruction. The instructions are legalized
// directly because the DivRemPairsPass performs the transformation at the IR
// level.
if (Subtarget.isISA3_0()) {
setOperationAction(ISD::SREM, MVT::i32, Legal);
setOperationAction(ISD::UREM, MVT::i32, Legal);
setOperationAction(ISD::SREM, MVT::i64, Legal);
setOperationAction(ISD::UREM, MVT::i64, Legal);
} else {
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::UREM, MVT::i32, Expand);
setOperationAction(ISD::SREM, MVT::i64, Expand);
setOperationAction(ISD::UREM, MVT::i64, Expand);
}
// Don't use SMUL_LOHI/UMUL_LOHI or SDIVREM/UDIVREM to lower SREM/UREM.
setOperationAction(ISD::UMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i32, Expand);
setOperationAction(ISD::UMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::SMUL_LOHI, MVT::i64, Expand);
setOperationAction(ISD::UDIVREM, MVT::i32, Expand);
setOperationAction(ISD::SDIVREM, MVT::i32, Expand);
setOperationAction(ISD::UDIVREM, MVT::i64, Expand);
setOperationAction(ISD::SDIVREM, MVT::i64, Expand);
// Handle constrained floating-point operations of scalar.
// TODO: Handle SPE specific operation.
setOperationAction(ISD::STRICT_FADD, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FMUL, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FDIV, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FMA, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FADD, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FMUL, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FDIV, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FMA, MVT::f64, Legal);
if (Subtarget.hasVSX()) {
setOperationAction(ISD::STRICT_FRINT, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FRINT, MVT::f64, Legal);
}
if (Subtarget.hasFSQRT()) {
setOperationAction(ISD::STRICT_FSQRT, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FSQRT, MVT::f64, Legal);
}
if (Subtarget.hasFPRND()) {
setOperationAction(ISD::STRICT_FFLOOR, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FCEIL, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FTRUNC, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FROUND, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FFLOOR, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FCEIL, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FTRUNC, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FROUND, MVT::f64, Legal);
}
// We don't support sin/cos/sqrt/fmod/pow
setOperationAction(ISD::FSIN , MVT::f64, Expand);
setOperationAction(ISD::FCOS , MVT::f64, Expand);
setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
setOperationAction(ISD::FREM , MVT::f64, Expand);
setOperationAction(ISD::FPOW , MVT::f64, Expand);
setOperationAction(ISD::FSIN , MVT::f32, Expand);
setOperationAction(ISD::FCOS , MVT::f32, Expand);
setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
setOperationAction(ISD::FREM , MVT::f32, Expand);
setOperationAction(ISD::FPOW , MVT::f32, Expand);
if (Subtarget.hasSPE()) {
setOperationAction(ISD::FMA , MVT::f64, Expand);
setOperationAction(ISD::FMA , MVT::f32, Expand);
} else {
setOperationAction(ISD::FMA , MVT::f64, Legal);
setOperationAction(ISD::FMA , MVT::f32, Legal);
}
if (Subtarget.hasSPE())
setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
setOperationAction(ISD::FLT_ROUNDS_, MVT::i32, Custom);
// If we're enabling GP optimizations, use hardware square root
if (!Subtarget.hasFSQRT() &&
!(TM.Options.UnsafeFPMath && Subtarget.hasFRSQRTE() &&
Subtarget.hasFRE()))
setOperationAction(ISD::FSQRT, MVT::f64, Expand);
if (!Subtarget.hasFSQRT() &&
!(TM.Options.UnsafeFPMath && Subtarget.hasFRSQRTES() &&
Subtarget.hasFRES()))
setOperationAction(ISD::FSQRT, MVT::f32, Expand);
if (Subtarget.hasFCPSGN()) {
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Legal);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Legal);
} else {
setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
}
if (Subtarget.hasFPRND()) {
setOperationAction(ISD::FFLOOR, MVT::f64, Legal);
setOperationAction(ISD::FCEIL, MVT::f64, Legal);
setOperationAction(ISD::FTRUNC, MVT::f64, Legal);
setOperationAction(ISD::FROUND, MVT::f64, Legal);
setOperationAction(ISD::FFLOOR, MVT::f32, Legal);
setOperationAction(ISD::FCEIL, MVT::f32, Legal);
setOperationAction(ISD::FTRUNC, MVT::f32, Legal);
setOperationAction(ISD::FROUND, MVT::f32, Legal);
}
// PowerPC does not have BSWAP, but we can use vector BSWAP instruction xxbrd
// to speed up scalar BSWAP64.
// CTPOP or CTTZ were introduced in P8/P9 respectively
setOperationAction(ISD::BSWAP, MVT::i32 , Expand);
if (Subtarget.hasP9Vector())
setOperationAction(ISD::BSWAP, MVT::i64 , Custom);
else
setOperationAction(ISD::BSWAP, MVT::i64 , Expand);
if (Subtarget.isISA3_0()) {
setOperationAction(ISD::CTTZ , MVT::i32 , Legal);
setOperationAction(ISD::CTTZ , MVT::i64 , Legal);
} else {
setOperationAction(ISD::CTTZ , MVT::i32 , Expand);
setOperationAction(ISD::CTTZ , MVT::i64 , Expand);
}
if (Subtarget.hasPOPCNTD() == PPCSubtarget::POPCNTD_Fast) {
setOperationAction(ISD::CTPOP, MVT::i32 , Legal);
setOperationAction(ISD::CTPOP, MVT::i64 , Legal);
} else {
setOperationAction(ISD::CTPOP, MVT::i32 , Expand);
setOperationAction(ISD::CTPOP, MVT::i64 , Expand);
}
// PowerPC does not have ROTR
setOperationAction(ISD::ROTR, MVT::i32 , Expand);
setOperationAction(ISD::ROTR, MVT::i64 , Expand);
if (!Subtarget.useCRBits()) {
// PowerPC does not have Select
setOperationAction(ISD::SELECT, MVT::i32, Expand);
setOperationAction(ISD::SELECT, MVT::i64, Expand);
setOperationAction(ISD::SELECT, MVT::f32, Expand);
setOperationAction(ISD::SELECT, MVT::f64, Expand);
}
// PowerPC wants to turn select_cc of FP into fsel when possible.
setOperationAction(ISD::SELECT_CC, MVT::f32, Custom);
setOperationAction(ISD::SELECT_CC, MVT::f64, Custom);
// PowerPC wants to optimize integer setcc a bit
if (!Subtarget.useCRBits())
setOperationAction(ISD::SETCC, MVT::i32, Custom);
if (Subtarget.hasFPU()) {
setOperationAction(ISD::STRICT_FSETCC, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FSETCC, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FSETCC, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FSETCCS, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FSETCCS, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FSETCCS, MVT::f128, Legal);
}
// PowerPC does not have BRCOND which requires SetCC
if (!Subtarget.useCRBits())
setOperationAction(ISD::BRCOND, MVT::Other, Expand);
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
if (Subtarget.hasSPE()) {
// SPE has built-in conversions
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i32, Legal);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i32, Legal);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i32, Legal);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Legal);
} else {
// PowerPC turns FP_TO_SINT into FCTIWZ and some load/stores.
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
// PowerPC does not have [U|S]INT_TO_FP
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Expand);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Expand);
}
if (Subtarget.hasDirectMove() && isPPC64) {
setOperationAction(ISD::BITCAST, MVT::f32, Legal);
setOperationAction(ISD::BITCAST, MVT::i32, Legal);
setOperationAction(ISD::BITCAST, MVT::i64, Legal);
setOperationAction(ISD::BITCAST, MVT::f64, Legal);
if (TM.Options.UnsafeFPMath) {
setOperationAction(ISD::LRINT, MVT::f64, Legal);
setOperationAction(ISD::LRINT, MVT::f32, Legal);
setOperationAction(ISD::LLRINT, MVT::f64, Legal);
setOperationAction(ISD::LLRINT, MVT::f32, Legal);
setOperationAction(ISD::LROUND, MVT::f64, Legal);
setOperationAction(ISD::LROUND, MVT::f32, Legal);
setOperationAction(ISD::LLROUND, MVT::f64, Legal);
setOperationAction(ISD::LLROUND, MVT::f32, Legal);
}
} else {
setOperationAction(ISD::BITCAST, MVT::f32, Expand);
setOperationAction(ISD::BITCAST, MVT::i32, Expand);
setOperationAction(ISD::BITCAST, MVT::i64, Expand);
setOperationAction(ISD::BITCAST, MVT::f64, Expand);
}
// We cannot sextinreg(i1). Expand to shifts.
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
// NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intended to support
// SjLj exception handling but a light-weight setjmp/longjmp replacement to
// support continuation, user-level threading, and etc.. As a result, no
// other SjLj exception interfaces are implemented and please don't build
// your own exception handling based on them.
// LLVM/Clang supports zero-cost DWARF exception handling.
setOperationAction(ISD::EH_SJLJ_SETJMP, MVT::i32, Custom);
setOperationAction(ISD::EH_SJLJ_LONGJMP, MVT::Other, Custom);
// We want to legalize GlobalAddress and ConstantPool nodes into the
// appropriate instructions to materialize the address.
setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i32, Custom);
setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
setOperationAction(ISD::ConstantPool, MVT::i32, Custom);
setOperationAction(ISD::JumpTable, MVT::i32, Custom);
setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
setOperationAction(ISD::BlockAddress, MVT::i64, Custom);
setOperationAction(ISD::ConstantPool, MVT::i64, Custom);
setOperationAction(ISD::JumpTable, MVT::i64, Custom);
// TRAP is legal.
setOperationAction(ISD::TRAP, MVT::Other, Legal);
// TRAMPOLINE is custom lowered.
setOperationAction(ISD::INIT_TRAMPOLINE, MVT::Other, Custom);
setOperationAction(ISD::ADJUST_TRAMPOLINE, MVT::Other, Custom);
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
if (Subtarget.is64BitELFABI()) {
// VAARG always uses double-word chunks, so promote anything smaller.
setOperationAction(ISD::VAARG, MVT::i1, Promote);
AddPromotedToType(ISD::VAARG, MVT::i1, MVT::i64);
setOperationAction(ISD::VAARG, MVT::i8, Promote);
AddPromotedToType(ISD::VAARG, MVT::i8, MVT::i64);
setOperationAction(ISD::VAARG, MVT::i16, Promote);
AddPromotedToType(ISD::VAARG, MVT::i16, MVT::i64);
setOperationAction(ISD::VAARG, MVT::i32, Promote);
AddPromotedToType(ISD::VAARG, MVT::i32, MVT::i64);
setOperationAction(ISD::VAARG, MVT::Other, Expand);
} else if (Subtarget.is32BitELFABI()) {
// VAARG is custom lowered with the 32-bit SVR4 ABI.
setOperationAction(ISD::VAARG, MVT::Other, Custom);
setOperationAction(ISD::VAARG, MVT::i64, Custom);
} else
setOperationAction(ISD::VAARG, MVT::Other, Expand);
// VACOPY is custom lowered with the 32-bit SVR4 ABI.
if (Subtarget.is32BitELFABI())
setOperationAction(ISD::VACOPY , MVT::Other, Custom);
else
setOperationAction(ISD::VACOPY , MVT::Other, Expand);
// Use the default implementation.
setOperationAction(ISD::VAEND , MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE , MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE , MVT::Other, Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Custom);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Custom);
setOperationAction(ISD::GET_DYNAMIC_AREA_OFFSET, MVT::i32, Custom);
setOperationAction(ISD::GET_DYNAMIC_AREA_OFFSET, MVT::i64, Custom);
setOperationAction(ISD::EH_DWARF_CFA, MVT::i32, Custom);
setOperationAction(ISD::EH_DWARF_CFA, MVT::i64, Custom);
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
// To handle counter-based loop conditions.
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i1, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::i8, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::i16, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::i32, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
// Comparisons that require checking two conditions.
if (Subtarget.hasSPE()) {
setCondCodeAction(ISD::SETO, MVT::f32, Expand);
setCondCodeAction(ISD::SETO, MVT::f64, Expand);
setCondCodeAction(ISD::SETUO, MVT::f32, Expand);
setCondCodeAction(ISD::SETUO, MVT::f64, Expand);
}
setCondCodeAction(ISD::SETULT, MVT::f32, Expand);
setCondCodeAction(ISD::SETULT, MVT::f64, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f32, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f64, Expand);
setCondCodeAction(ISD::SETUEQ, MVT::f32, Expand);
setCondCodeAction(ISD::SETUEQ, MVT::f64, Expand);
setCondCodeAction(ISD::SETOGE, MVT::f32, Expand);
setCondCodeAction(ISD::SETOGE, MVT::f64, Expand);
setCondCodeAction(ISD::SETOLE, MVT::f32, Expand);
setCondCodeAction(ISD::SETOLE, MVT::f64, Expand);
setCondCodeAction(ISD::SETONE, MVT::f32, Expand);
setCondCodeAction(ISD::SETONE, MVT::f64, Expand);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f64, Legal);
if (Subtarget.has64BitSupport()) {
// They also have instructions for converting between i64 and fp.
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i64, Expand);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i64, Expand);
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i64, Expand);
setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Expand);
// This is just the low 32 bits of a (signed) fp->i64 conversion.
// We cannot do this with Promote because i64 is not a legal type.
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
if (Subtarget.hasLFIWAX() || Subtarget.isPPC64()) {
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i32, Custom);
}
} else {
// PowerPC does not have FP_TO_UINT on 32-bit implementations.
if (Subtarget.hasSPE()) {
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i32, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Legal);
} else {
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i32, Expand);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Expand);
}
}
// With the instructions enabled under FPCVT, we can do everything.
if (Subtarget.hasFPCVT()) {
if (Subtarget.has64BitSupport()) {
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i64, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i64, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
}
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::i32, Custom);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::i32, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Custom);
}
if (Subtarget.use64BitRegs()) {
// 64-bit PowerPC implementations can support i64 types directly
addRegisterClass(MVT::i64, &PPC::G8RCRegClass);
// BUILD_PAIR can't be handled natively, and should be expanded to shl/or
setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
// 64-bit PowerPC wants to expand i128 shifts itself.
setOperationAction(ISD::SHL_PARTS, MVT::i64, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i64, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i64, Custom);
} else {
// 32-bit PowerPC wants to expand i64 shifts itself.
setOperationAction(ISD::SHL_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRA_PARTS, MVT::i32, Custom);
setOperationAction(ISD::SRL_PARTS, MVT::i32, Custom);
}
// PowerPC has better expansions for funnel shifts than the generic
// TargetLowering::expandFunnelShift.
if (Subtarget.has64BitSupport()) {
setOperationAction(ISD::FSHL, MVT::i64, Custom);
setOperationAction(ISD::FSHR, MVT::i64, Custom);
}
setOperationAction(ISD::FSHL, MVT::i32, Custom);
setOperationAction(ISD::FSHR, MVT::i32, Custom);
if (Subtarget.hasVSX()) {
setOperationAction(ISD::FMAXNUM_IEEE, MVT::f64, Legal);
setOperationAction(ISD::FMAXNUM_IEEE, MVT::f32, Legal);
setOperationAction(ISD::FMINNUM_IEEE, MVT::f64, Legal);
setOperationAction(ISD::FMINNUM_IEEE, MVT::f32, Legal);
}
if (Subtarget.hasAltivec()) {
for (MVT VT : { MVT::v16i8, MVT::v8i16, MVT::v4i32 }) {
setOperationAction(ISD::SADDSAT, VT, Legal);
setOperationAction(ISD::SSUBSAT, VT, Legal);
setOperationAction(ISD::UADDSAT, VT, Legal);
setOperationAction(ISD::USUBSAT, VT, Legal);
}
// First set operation action for all vector types to expand. Then we
// will selectively turn on ones that can be effectively codegen'd.
for (MVT VT : MVT::fixedlen_vector_valuetypes()) {
// add/sub are legal for all supported vector VT's.
setOperationAction(ISD::ADD, VT, Legal);
setOperationAction(ISD::SUB, VT, Legal);
// For v2i64, these are only valid with P8Vector. This is corrected after
// the loop.
if (VT.getSizeInBits() <= 128 && VT.getScalarSizeInBits() <= 64) {
setOperationAction(ISD::SMAX, VT, Legal);
setOperationAction(ISD::SMIN, VT, Legal);
setOperationAction(ISD::UMAX, VT, Legal);
setOperationAction(ISD::UMIN, VT, Legal);
}
else {
setOperationAction(ISD::SMAX, VT, Expand);
setOperationAction(ISD::SMIN, VT, Expand);
setOperationAction(ISD::UMAX, VT, Expand);
setOperationAction(ISD::UMIN, VT, Expand);
}
if (Subtarget.hasVSX()) {
setOperationAction(ISD::FMAXNUM, VT, Legal);
setOperationAction(ISD::FMINNUM, VT, Legal);
}
// Vector instructions introduced in P8
if (Subtarget.hasP8Altivec() && (VT.SimpleTy != MVT::v1i128)) {
setOperationAction(ISD::CTPOP, VT, Legal);
setOperationAction(ISD::CTLZ, VT, Legal);
}
else {
setOperationAction(ISD::CTPOP, VT, Expand);
setOperationAction(ISD::CTLZ, VT, Expand);
}
// Vector instructions introduced in P9
if (Subtarget.hasP9Altivec() && (VT.SimpleTy != MVT::v1i128))
setOperationAction(ISD::CTTZ, VT, Legal);
else
setOperationAction(ISD::CTTZ, VT, Expand);
// We promote all shuffles to v16i8.
setOperationAction(ISD::VECTOR_SHUFFLE, VT, Promote);
AddPromotedToType (ISD::VECTOR_SHUFFLE, VT, MVT::v16i8);
// We promote all non-typed operations to v4i32.
setOperationAction(ISD::AND , VT, Promote);
AddPromotedToType (ISD::AND , VT, MVT::v4i32);
setOperationAction(ISD::OR , VT, Promote);
AddPromotedToType (ISD::OR , VT, MVT::v4i32);
setOperationAction(ISD::XOR , VT, Promote);
AddPromotedToType (ISD::XOR , VT, MVT::v4i32);
setOperationAction(ISD::LOAD , VT, Promote);
AddPromotedToType (ISD::LOAD , VT, MVT::v4i32);
setOperationAction(ISD::SELECT, VT, Promote);
AddPromotedToType (ISD::SELECT, VT, MVT::v4i32);
setOperationAction(ISD::VSELECT, VT, Legal);
setOperationAction(ISD::SELECT_CC, VT, Promote);
AddPromotedToType (ISD::SELECT_CC, VT, MVT::v4i32);
setOperationAction(ISD::STORE, VT, Promote);
AddPromotedToType (ISD::STORE, VT, MVT::v4i32);
// No other operations are legal.
setOperationAction(ISD::MUL , VT, Expand);
setOperationAction(ISD::SDIV, VT, Expand);
setOperationAction(ISD::SREM, VT, Expand);
setOperationAction(ISD::UDIV, VT, Expand);
setOperationAction(ISD::UREM, VT, Expand);
setOperationAction(ISD::FDIV, VT, Expand);
setOperationAction(ISD::FREM, VT, Expand);
setOperationAction(ISD::FNEG, VT, Expand);
setOperationAction(ISD::FSQRT, VT, Expand);
setOperationAction(ISD::FLOG, VT, Expand);
setOperationAction(ISD::FLOG10, VT, Expand);
setOperationAction(ISD::FLOG2, VT, Expand);
setOperationAction(ISD::FEXP, VT, Expand);
setOperationAction(ISD::FEXP2, VT, Expand);
setOperationAction(ISD::FSIN, VT, Expand);
setOperationAction(ISD::FCOS, VT, Expand);
setOperationAction(ISD::FABS, VT, Expand);
setOperationAction(ISD::FFLOOR, VT, Expand);
setOperationAction(ISD::FCEIL, VT, Expand);
setOperationAction(ISD::FTRUNC, VT, Expand);
setOperationAction(ISD::FRINT, VT, Expand);
setOperationAction(ISD::FNEARBYINT, VT, Expand);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Expand);
setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Expand);
setOperationAction(ISD::BUILD_VECTOR, VT, Expand);
setOperationAction(ISD::MULHU, VT, Expand);
setOperationAction(ISD::MULHS, VT, Expand);
setOperationAction(ISD::UMUL_LOHI, VT, Expand);
setOperationAction(ISD::SMUL_LOHI, VT, Expand);
setOperationAction(ISD::UDIVREM, VT, Expand);
setOperationAction(ISD::SDIVREM, VT, Expand);
setOperationAction(ISD::SCALAR_TO_VECTOR, VT, Expand);
setOperationAction(ISD::FPOW, VT, Expand);
setOperationAction(ISD::BSWAP, VT, Expand);
setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
setOperationAction(ISD::ROTL, VT, Expand);
setOperationAction(ISD::ROTR, VT, Expand);
for (MVT InnerVT : MVT::fixedlen_vector_valuetypes()) {
setTruncStoreAction(VT, InnerVT, Expand);
setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
}
}
setOperationAction(ISD::SELECT_CC, MVT::v4i32, Expand);
if (!Subtarget.hasP8Vector()) {
setOperationAction(ISD::SMAX, MVT::v2i64, Expand);
setOperationAction(ISD::SMIN, MVT::v2i64, Expand);
setOperationAction(ISD::UMAX, MVT::v2i64, Expand);
setOperationAction(ISD::UMIN, MVT::v2i64, Expand);
}
// We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
// with merges, splats, etc.
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i8, Custom);
// Vector truncates to sub-word integer that fit in an Altivec/VSX register
// are cheap, so handle them before they get expanded to scalar.
setOperationAction(ISD::TRUNCATE, MVT::v8i8, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v4i8, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v2i8, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v4i16, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v2i16, Custom);
setOperationAction(ISD::AND , MVT::v4i32, Legal);
setOperationAction(ISD::OR , MVT::v4i32, Legal);
setOperationAction(ISD::XOR , MVT::v4i32, Legal);
setOperationAction(ISD::LOAD , MVT::v4i32, Legal);
setOperationAction(ISD::SELECT, MVT::v4i32,
Subtarget.useCRBits() ? Legal : Expand);
setOperationAction(ISD::STORE , MVT::v4i32, Legal);
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::v4i32, Legal);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::v4i32, Legal);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::FP_TO_SINT, MVT::v4i32, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::FFLOOR, MVT::v4f32, Legal);
setOperationAction(ISD::FCEIL, MVT::v4f32, Legal);
setOperationAction(ISD::FTRUNC, MVT::v4f32, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Legal);
// Custom lowering ROTL v1i128 to VECTOR_SHUFFLE v16i8.
setOperationAction(ISD::ROTL, MVT::v1i128, Custom);
// With hasAltivec set, we can lower ISD::ROTL to vrl(b|h|w).
if (Subtarget.hasAltivec())
for (auto VT : {MVT::v4i32, MVT::v8i16, MVT::v16i8})
setOperationAction(ISD::ROTL, VT, Legal);
// With hasP8Altivec set, we can lower ISD::ROTL to vrld.
if (Subtarget.hasP8Altivec())
setOperationAction(ISD::ROTL, MVT::v2i64, Legal);
addRegisterClass(MVT::v4f32, &PPC::VRRCRegClass);
addRegisterClass(MVT::v4i32, &PPC::VRRCRegClass);
addRegisterClass(MVT::v8i16, &PPC::VRRCRegClass);
addRegisterClass(MVT::v16i8, &PPC::VRRCRegClass);
setOperationAction(ISD::MUL, MVT::v4f32, Legal);
setOperationAction(ISD::FMA, MVT::v4f32, Legal);
if (Subtarget.hasVSX()) {
setOperationAction(ISD::FDIV, MVT::v4f32, Legal);
setOperationAction(ISD::FSQRT, MVT::v4f32, Legal);
}
if (Subtarget.hasP8Altivec())
setOperationAction(ISD::MUL, MVT::v4i32, Legal);
else
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
if (Subtarget.isISA3_1()) {
setOperationAction(ISD::MUL, MVT::v2i64, Legal);
setOperationAction(ISD::MULHS, MVT::v2i64, Legal);
setOperationAction(ISD::MULHU, MVT::v2i64, Legal);
setOperationAction(ISD::MULHS, MVT::v4i32, Legal);
setOperationAction(ISD::MULHU, MVT::v4i32, Legal);
setOperationAction(ISD::UDIV, MVT::v2i64, Legal);
setOperationAction(ISD::SDIV, MVT::v2i64, Legal);
setOperationAction(ISD::UDIV, MVT::v4i32, Legal);
setOperationAction(ISD::SDIV, MVT::v4i32, Legal);
setOperationAction(ISD::UREM, MVT::v2i64, Legal);
setOperationAction(ISD::SREM, MVT::v2i64, Legal);
setOperationAction(ISD::UREM, MVT::v4i32, Legal);
setOperationAction(ISD::SREM, MVT::v4i32, Legal);
setOperationAction(ISD::UREM, MVT::v1i128, Legal);
setOperationAction(ISD::SREM, MVT::v1i128, Legal);
setOperationAction(ISD::UDIV, MVT::v1i128, Legal);
setOperationAction(ISD::SDIV, MVT::v1i128, Legal);
setOperationAction(ISD::ROTL, MVT::v1i128, Legal);
}
setOperationAction(ISD::MUL, MVT::v8i16, Legal);
setOperationAction(ISD::MUL, MVT::v16i8, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Custom);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v16i8, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v8i16, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4i32, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v4f32, Custom);
// Altivec does not contain unordered floating-point compare instructions
setCondCodeAction(ISD::SETUO, MVT::v4f32, Expand);
setCondCodeAction(ISD::SETUEQ, MVT::v4f32, Expand);
setCondCodeAction(ISD::SETO, MVT::v4f32, Expand);
setCondCodeAction(ISD::SETONE, MVT::v4f32, Expand);
if (Subtarget.hasVSX()) {
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v2f64, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f64, Legal);
if (Subtarget.hasP8Vector()) {
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4f32, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4f32, Legal);
}
if (Subtarget.hasDirectMove() && isPPC64) {
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v16i8, Legal);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v8i16, Legal);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v4i32, Legal);
setOperationAction(ISD::SCALAR_TO_VECTOR, MVT::v2i64, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v16i8, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v8i16, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4i32, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2i64, Legal);
}
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f64, Legal);
// The nearbyint variants are not allowed to raise the inexact exception
// so we can only code-gen them with unsafe math.
if (TM.Options.UnsafeFPMath) {
setOperationAction(ISD::FNEARBYINT, MVT::f64, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::f32, Legal);
}
setOperationAction(ISD::FFLOOR, MVT::v2f64, Legal);
setOperationAction(ISD::FCEIL, MVT::v2f64, Legal);
setOperationAction(ISD::FTRUNC, MVT::v2f64, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::v2f64, Legal);
setOperationAction(ISD::FRINT, MVT::v2f64, Legal);
setOperationAction(ISD::FROUND, MVT::v2f64, Legal);
setOperationAction(ISD::FROUND, MVT::f64, Legal);
setOperationAction(ISD::FRINT, MVT::f64, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::v4f32, Legal);
setOperationAction(ISD::FRINT, MVT::v4f32, Legal);
setOperationAction(ISD::FROUND, MVT::v4f32, Legal);
setOperationAction(ISD::FROUND, MVT::f32, Legal);
setOperationAction(ISD::FRINT, MVT::f32, Legal);
setOperationAction(ISD::MUL, MVT::v2f64, Legal);
setOperationAction(ISD::FMA, MVT::v2f64, Legal);
setOperationAction(ISD::FDIV, MVT::v2f64, Legal);
setOperationAction(ISD::FSQRT, MVT::v2f64, Legal);
// Share the Altivec comparison restrictions.
setCondCodeAction(ISD::SETUO, MVT::v2f64, Expand);
setCondCodeAction(ISD::SETUEQ, MVT::v2f64, Expand);
setCondCodeAction(ISD::SETO, MVT::v2f64, Expand);
setCondCodeAction(ISD::SETONE, MVT::v2f64, Expand);
setOperationAction(ISD::LOAD, MVT::v2f64, Legal);
setOperationAction(ISD::STORE, MVT::v2f64, Legal);
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v2f64, Legal);
if (Subtarget.hasP8Vector())
addRegisterClass(MVT::f32, &PPC::VSSRCRegClass);
addRegisterClass(MVT::f64, &PPC::VSFRCRegClass);
addRegisterClass(MVT::v4i32, &PPC::VSRCRegClass);
addRegisterClass(MVT::v4f32, &PPC::VSRCRegClass);
addRegisterClass(MVT::v2f64, &PPC::VSRCRegClass);
if (Subtarget.hasP8Altivec()) {
setOperationAction(ISD::SHL, MVT::v2i64, Legal);
setOperationAction(ISD::SRA, MVT::v2i64, Legal);
setOperationAction(ISD::SRL, MVT::v2i64, Legal);
// 128 bit shifts can be accomplished via 3 instructions for SHL and
// SRL, but not for SRA because of the instructions available:
// VS{RL} and VS{RL}O. However due to direct move costs, it's not worth
// doing
setOperationAction(ISD::SHL, MVT::v1i128, Expand);
setOperationAction(ISD::SRL, MVT::v1i128, Expand);
setOperationAction(ISD::SRA, MVT::v1i128, Expand);
setOperationAction(ISD::SETCC, MVT::v2i64, Legal);
}
else {
setOperationAction(ISD::SHL, MVT::v2i64, Expand);
setOperationAction(ISD::SRA, MVT::v2i64, Expand);
setOperationAction(ISD::SRL, MVT::v2i64, Expand);
setOperationAction(ISD::SETCC, MVT::v2i64, Custom);
// VSX v2i64 only supports non-arithmetic operations.
setOperationAction(ISD::ADD, MVT::v2i64, Expand);
setOperationAction(ISD::SUB, MVT::v2i64, Expand);
}
if (Subtarget.isISA3_1())
setOperationAction(ISD::SETCC, MVT::v1i128, Legal);
else
setOperationAction(ISD::SETCC, MVT::v1i128, Expand);
setOperationAction(ISD::LOAD, MVT::v2i64, Promote);
AddPromotedToType (ISD::LOAD, MVT::v2i64, MVT::v2f64);
setOperationAction(ISD::STORE, MVT::v2i64, Promote);
AddPromotedToType (ISD::STORE, MVT::v2i64, MVT::v2f64);
setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v2i64, Legal);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v2i64, Legal);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v2i64, Legal);
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::v2i64, Legal);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::v2i64, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::v2i64, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::v2i64, Legal);
setOperationAction(ISD::FP_TO_SINT, MVT::v2i64, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::v2i64, Legal);
// Custom handling for partial vectors of integers converted to
// floating point. We already have optimal handling for v2i32 through
// the DAG combine, so those aren't necessary.
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v2i8, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v4i8, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v2i16, Custom);
setOperationAction(ISD::STRICT_UINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v2i8, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v4i8, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v2i16, Custom);
setOperationAction(ISD::STRICT_SINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v2i8, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i8, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v2i16, Custom);
setOperationAction(ISD::UINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::v2i8, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::v4i8, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::v2i16, Custom);
setOperationAction(ISD::SINT_TO_FP, MVT::v4i16, Custom);
setOperationAction(ISD::FNEG, MVT::v4f32, Legal);
setOperationAction(ISD::FNEG, MVT::v2f64, Legal);
setOperationAction(ISD::FABS, MVT::v4f32, Legal);
setOperationAction(ISD::FABS, MVT::v2f64, Legal);
setOperationAction(ISD::FCOPYSIGN, MVT::v4f32, Legal);
setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Legal);
if (Subtarget.hasDirectMove())
setOperationAction(ISD::BUILD_VECTOR, MVT::v2i64, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v2f64, Custom);
// Handle constrained floating-point operations of vector.
// The predictor is `hasVSX` because altivec instruction has
// no exception but VSX vector instruction has.
setOperationAction(ISD::STRICT_FADD, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FMUL, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FDIV, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FMA, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FSQRT, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FMAXNUM, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FMINNUM, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FRINT, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FFLOOR, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FCEIL, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FTRUNC, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FROUND, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FADD, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FMUL, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FDIV, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FMA, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FSQRT, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FMAXNUM, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FMINNUM, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FRINT, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FFLOOR, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FCEIL, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FTRUNC, MVT::v2f64, Legal);
setOperationAction(ISD::STRICT_FROUND, MVT::v2f64, Legal);
addRegisterClass(MVT::v2i64, &PPC::VSRCRegClass);
}
if (Subtarget.hasP8Altivec()) {
addRegisterClass(MVT::v2i64, &PPC::VRRCRegClass);
addRegisterClass(MVT::v1i128, &PPC::VRRCRegClass);
}
if (Subtarget.hasP9Vector()) {
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Custom);
// 128 bit shifts can be accomplished via 3 instructions for SHL and
// SRL, but not for SRA because of the instructions available:
// VS{RL} and VS{RL}O.
setOperationAction(ISD::SHL, MVT::v1i128, Legal);
setOperationAction(ISD::SRL, MVT::v1i128, Legal);
setOperationAction(ISD::SRA, MVT::v1i128, Expand);
addRegisterClass(MVT::f128, &PPC::VRRCRegClass);
setOperationAction(ISD::FADD, MVT::f128, Legal);
setOperationAction(ISD::FSUB, MVT::f128, Legal);
setOperationAction(ISD::FDIV, MVT::f128, Legal);
setOperationAction(ISD::FMUL, MVT::f128, Legal);
setOperationAction(ISD::FP_EXTEND, MVT::f128, Legal);
// No extending loads to f128 on PPC.
for (MVT FPT : MVT::fp_valuetypes())
setLoadExtAction(ISD::EXTLOAD, MVT::f128, FPT, Expand);
setOperationAction(ISD::FMA, MVT::f128, Legal);
setCondCodeAction(ISD::SETULT, MVT::f128, Expand);
setCondCodeAction(ISD::SETUGT, MVT::f128, Expand);
setCondCodeAction(ISD::SETUEQ, MVT::f128, Expand);
setCondCodeAction(ISD::SETOGE, MVT::f128, Expand);
setCondCodeAction(ISD::SETOLE, MVT::f128, Expand);
setCondCodeAction(ISD::SETONE, MVT::f128, Expand);
setOperationAction(ISD::FTRUNC, MVT::f128, Legal);
setOperationAction(ISD::FRINT, MVT::f128, Legal);
setOperationAction(ISD::FFLOOR, MVT::f128, Legal);
setOperationAction(ISD::FCEIL, MVT::f128, Legal);
setOperationAction(ISD::FNEARBYINT, MVT::f128, Legal);
setOperationAction(ISD::FROUND, MVT::f128, Legal);
setOperationAction(ISD::SELECT, MVT::f128, Expand);
setOperationAction(ISD::FP_ROUND, MVT::f64, Legal);
setOperationAction(ISD::FP_ROUND, MVT::f32, Legal);
setTruncStoreAction(MVT::f128, MVT::f64, Expand);
setTruncStoreAction(MVT::f128, MVT::f32, Expand);
setOperationAction(ISD::BITCAST, MVT::i128, Custom);
// No implementation for these ops for PowerPC.
setOperationAction(ISD::FSIN, MVT::f128, Expand);
setOperationAction(ISD::FCOS, MVT::f128, Expand);
setOperationAction(ISD::FPOW, MVT::f128, Expand);
setOperationAction(ISD::FPOWI, MVT::f128, Expand);
setOperationAction(ISD::FREM, MVT::f128, Expand);
// Handle constrained floating-point operations of fp128
setOperationAction(ISD::STRICT_FADD, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FMUL, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FDIV, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FMA, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FSQRT, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f64, Legal);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f32, Legal);
setOperationAction(ISD::STRICT_FRINT, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FNEARBYINT, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FFLOOR, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FCEIL, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FTRUNC, MVT::f128, Legal);
setOperationAction(ISD::STRICT_FROUND, MVT::f128, Legal);
setOperationAction(ISD::FP_EXTEND, MVT::v2f32, Custom);
setOperationAction(ISD::BSWAP, MVT::v8i16, Legal);
setOperationAction(ISD::BSWAP, MVT::v4i32, Legal);
setOperationAction(ISD::BSWAP, MVT::v2i64, Legal);
setOperationAction(ISD::BSWAP, MVT::v1i128, Legal);
} else if (Subtarget.hasAltivec() && EnableSoftFP128) {
addRegisterClass(MVT::f128, &PPC::VRRCRegClass);
for (MVT FPT : MVT::fp_valuetypes())
setLoadExtAction(ISD::EXTLOAD, MVT::f128, FPT, Expand);
setOperationAction(ISD::LOAD, MVT::f128, Promote);
setOperationAction(ISD::STORE, MVT::f128, Promote);
AddPromotedToType(ISD::LOAD, MVT::f128, MVT::v4i32);
AddPromotedToType(ISD::STORE, MVT::f128, MVT::v4i32);
// Set FADD/FSUB as libcall to avoid the legalizer to expand the
// fp_to_uint and int_to_fp.
setOperationAction(ISD::FADD, MVT::f128, LibCall);
setOperationAction(ISD::FSUB, MVT::f128, LibCall);
setOperationAction(ISD::FMUL, MVT::f128, Expand);
setOperationAction(ISD::FDIV, MVT::f128, Expand);
setOperationAction(ISD::FNEG, MVT::f128, Expand);
setOperationAction(ISD::FABS, MVT::f128, Expand);
setOperationAction(ISD::FSIN, MVT::f128, Expand);
setOperationAction(ISD::FCOS, MVT::f128, Expand);
setOperationAction(ISD::FPOW, MVT::f128, Expand);
setOperationAction(ISD::FPOWI, MVT::f128, Expand);
setOperationAction(ISD::FREM, MVT::f128, Expand);
setOperationAction(ISD::FSQRT, MVT::f128, Expand);
setOperationAction(ISD::FMA, MVT::f128, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::f128, Expand);
setTruncStoreAction(MVT::f128, MVT::f64, Expand);
setTruncStoreAction(MVT::f128, MVT::f32, Expand);
// Expand the fp_extend if the target type is fp128.
setOperationAction(ISD::FP_EXTEND, MVT::f128, Expand);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f128, Expand);
// Expand the fp_round if the source type is fp128.
for (MVT VT : {MVT::f32, MVT::f64}) {
setOperationAction(ISD::FP_ROUND, VT, Custom);
setOperationAction(ISD::STRICT_FP_ROUND, VT, Custom);
}
}
if (Subtarget.hasP9Altivec()) {
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v8i16, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v16i8, Custom);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i8, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i16, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v4i32, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i8, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i16, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i32, Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::v2i64, Legal);
}
}
if (Subtarget.pairedVectorMemops()) {
addRegisterClass(MVT::v256i1, &PPC::VSRpRCRegClass);
setOperationAction(ISD::LOAD, MVT::v256i1, Custom);
setOperationAction(ISD::STORE, MVT::v256i1, Custom);
}
if (Subtarget.hasMMA()) {
addRegisterClass(MVT::v512i1, &PPC::UACCRCRegClass);
setOperationAction(ISD::LOAD, MVT::v512i1, Custom);
setOperationAction(ISD::STORE, MVT::v512i1, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v512i1, Custom);
}
if (Subtarget.has64BitSupport())
setOperationAction(ISD::PREFETCH, MVT::Other, Legal);
if (Subtarget.isISA3_1())
setOperationAction(ISD::SRA, MVT::v1i128, Legal);
setOperationAction(ISD::READCYCLECOUNTER, MVT::i64, isPPC64 ? Legal : Custom);
if (!isPPC64) {
setOperationAction(ISD::ATOMIC_LOAD, MVT::i64, Expand);
setOperationAction(ISD::ATOMIC_STORE, MVT::i64, Expand);
}
setBooleanContents(ZeroOrOneBooleanContent);
if (Subtarget.hasAltivec()) {
// Altivec instructions set fields to all zeros or all ones.
setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
}
if (!isPPC64) {
// These libcalls are not available in 32-bit.
setLibcallName(RTLIB::SHL_I128, nullptr);
setLibcallName(RTLIB::SRL_I128, nullptr);
setLibcallName(RTLIB::SRA_I128, nullptr);
}
if (!isPPC64)
setMaxAtomicSizeInBitsSupported(32);
setStackPointerRegisterToSaveRestore(isPPC64 ? PPC::X1 : PPC::R1);
// We have target-specific dag combine patterns for the following nodes:
setTargetDAGCombine(ISD::ADD);
setTargetDAGCombine(ISD::SHL);
setTargetDAGCombine(ISD::SRA);
setTargetDAGCombine(ISD::SRL);
setTargetDAGCombine(ISD::MUL);
setTargetDAGCombine(ISD::FMA);
setTargetDAGCombine(ISD::SINT_TO_FP);
setTargetDAGCombine(ISD::BUILD_VECTOR);
if (Subtarget.hasFPCVT())
setTargetDAGCombine(ISD::UINT_TO_FP);
setTargetDAGCombine(ISD::LOAD);
setTargetDAGCombine(ISD::STORE);
setTargetDAGCombine(ISD::BR_CC);
if (Subtarget.useCRBits())
setTargetDAGCombine(ISD::BRCOND);
setTargetDAGCombine(ISD::BSWAP);
setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
setTargetDAGCombine(ISD::INTRINSIC_W_CHAIN);
setTargetDAGCombine(ISD::INTRINSIC_VOID);
setTargetDAGCombine(ISD::SIGN_EXTEND);
setTargetDAGCombine(ISD::ZERO_EXTEND);
setTargetDAGCombine(ISD::ANY_EXTEND);
setTargetDAGCombine(ISD::TRUNCATE);
setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
if (Subtarget.useCRBits()) {
setTargetDAGCombine(ISD::TRUNCATE);
setTargetDAGCombine(ISD::SETCC);
setTargetDAGCombine(ISD::SELECT_CC);
}
if (Subtarget.hasP9Altivec()) {
setTargetDAGCombine(ISD::ABS);
setTargetDAGCombine(ISD::VSELECT);
}
setLibcallName(RTLIB::LOG_F128, "logf128");
setLibcallName(RTLIB::LOG2_F128, "log2f128");
setLibcallName(RTLIB::LOG10_F128, "log10f128");
setLibcallName(RTLIB::EXP_F128, "expf128");
setLibcallName(RTLIB::EXP2_F128, "exp2f128");
setLibcallName(RTLIB::SIN_F128, "sinf128");
setLibcallName(RTLIB::COS_F128, "cosf128");
setLibcallName(RTLIB::POW_F128, "powf128");
setLibcallName(RTLIB::FMIN_F128, "fminf128");
setLibcallName(RTLIB::FMAX_F128, "fmaxf128");
setLibcallName(RTLIB::REM_F128, "fmodf128");
setLibcallName(RTLIB::SQRT_F128, "sqrtf128");
setLibcallName(RTLIB::CEIL_F128, "ceilf128");
setLibcallName(RTLIB::FLOOR_F128, "floorf128");
setLibcallName(RTLIB::TRUNC_F128, "truncf128");
setLibcallName(RTLIB::ROUND_F128, "roundf128");
setLibcallName(RTLIB::LROUND_F128, "lroundf128");
setLibcallName(RTLIB::LLROUND_F128, "llroundf128");
setLibcallName(RTLIB::RINT_F128, "rintf128");
setLibcallName(RTLIB::LRINT_F128, "lrintf128");
setLibcallName(RTLIB::LLRINT_F128, "llrintf128");
setLibcallName(RTLIB::NEARBYINT_F128, "nearbyintf128");
setLibcallName(RTLIB::FMA_F128, "fmaf128");
// With 32 condition bits, we don't need to sink (and duplicate) compares
// aggressively in CodeGenPrep.
if (Subtarget.useCRBits()) {
setHasMultipleConditionRegisters();
setJumpIsExpensive();
}
setMinFunctionAlignment(Align(4));
switch (Subtarget.getCPUDirective()) {
default: break;
case PPC::DIR_970:
case PPC::DIR_A2:
case PPC::DIR_E500:
case PPC::DIR_E500mc:
case PPC::DIR_E5500:
case PPC::DIR_PWR4:
case PPC::DIR_PWR5:
case PPC::DIR_PWR5X:
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
case PPC::DIR_PWR8:
case PPC::DIR_PWR9:
case PPC::DIR_PWR10:
case PPC::DIR_PWR_FUTURE:
setPrefLoopAlignment(Align(16));
setPrefFunctionAlignment(Align(16));
break;
}
if (Subtarget.enableMachineScheduler())
setSchedulingPreference(Sched::Source);
else
setSchedulingPreference(Sched::Hybrid);
computeRegisterProperties(STI.getRegisterInfo());
// The Freescale cores do better with aggressive inlining of memcpy and
// friends. GCC uses same threshold of 128 bytes (= 32 word stores).
if (Subtarget.getCPUDirective() == PPC::DIR_E500mc ||
Subtarget.getCPUDirective() == PPC::DIR_E5500) {
MaxStoresPerMemset = 32;
MaxStoresPerMemsetOptSize = 16;
MaxStoresPerMemcpy = 32;
MaxStoresPerMemcpyOptSize = 8;
MaxStoresPerMemmove = 32;
MaxStoresPerMemmoveOptSize = 8;
} else if (Subtarget.getCPUDirective() == PPC::DIR_A2) {
// The A2 also benefits from (very) aggressive inlining of memcpy and
// friends. The overhead of a the function call, even when warm, can be
// over one hundred cycles.
MaxStoresPerMemset = 128;
MaxStoresPerMemcpy = 128;
MaxStoresPerMemmove = 128;
MaxLoadsPerMemcmp = 128;
} else {
MaxLoadsPerMemcmp = 8;
MaxLoadsPerMemcmpOptSize = 4;
}
IsStrictFPEnabled = true;
// Let the subtarget (CPU) decide if a predictable select is more expensive
// than the corresponding branch. This information is used in CGP to decide
// when to convert selects into branches.
PredictableSelectIsExpensive = Subtarget.isPredictableSelectIsExpensive();
}
/// getMaxByValAlign - Helper for getByValTypeAlignment to determine
/// the desired ByVal argument alignment.
static void getMaxByValAlign(Type *Ty, Align &MaxAlign, Align MaxMaxAlign) {
if (MaxAlign == MaxMaxAlign)
return;
if (VectorType *VTy = dyn_cast<VectorType>(Ty)) {
if (MaxMaxAlign >= 32 &&
VTy->getPrimitiveSizeInBits().getFixedSize() >= 256)
MaxAlign = Align(32);
else if (VTy->getPrimitiveSizeInBits().getFixedSize() >= 128 &&
MaxAlign < 16)
MaxAlign = Align(16);
} else if (ArrayType *ATy = dyn_cast<ArrayType>(Ty)) {
Align EltAlign;
getMaxByValAlign(ATy->getElementType(), EltAlign, MaxMaxAlign);
if (EltAlign > MaxAlign)
MaxAlign = EltAlign;
} else if (StructType *STy = dyn_cast<StructType>(Ty)) {
for (auto *EltTy : STy->elements()) {
Align EltAlign;
getMaxByValAlign(EltTy, EltAlign, MaxMaxAlign);
if (EltAlign > MaxAlign)
MaxAlign = EltAlign;
if (MaxAlign == MaxMaxAlign)
break;
}
}
}
/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area.
unsigned PPCTargetLowering::getByValTypeAlignment(Type *Ty,
const DataLayout &DL) const {
// 16byte and wider vectors are passed on 16byte boundary.
// The rest is 8 on PPC64 and 4 on PPC32 boundary.
Align Alignment = Subtarget.isPPC64() ? Align(8) : Align(4);
if (Subtarget.hasAltivec())
getMaxByValAlign(Ty, Alignment, Align(16));
return Alignment.value();
}
bool PPCTargetLowering::useSoftFloat() const {
return Subtarget.useSoftFloat();
}
bool PPCTargetLowering::hasSPE() const {
return Subtarget.hasSPE();
}
bool PPCTargetLowering::preferIncOfAddToSubOfNot(EVT VT) const {
return VT.isScalarInteger();
}
const char *PPCTargetLowering::getTargetNodeName(unsigned Opcode) const {
switch ((PPCISD::NodeType)Opcode) {
case PPCISD::FIRST_NUMBER: break;
case PPCISD::FSEL: return "PPCISD::FSEL";
case PPCISD::XSMAXCDP: return "PPCISD::XSMAXCDP";
case PPCISD::XSMINCDP: return "PPCISD::XSMINCDP";
case PPCISD::FCFID: return "PPCISD::FCFID";
case PPCISD::FCFIDU: return "PPCISD::FCFIDU";
case PPCISD::FCFIDS: return "PPCISD::FCFIDS";
case PPCISD::FCFIDUS: return "PPCISD::FCFIDUS";
case PPCISD::FCTIDZ: return "PPCISD::FCTIDZ";
case PPCISD::FCTIWZ: return "PPCISD::FCTIWZ";
case PPCISD::FCTIDUZ: return "PPCISD::FCTIDUZ";
case PPCISD::FCTIWUZ: return "PPCISD::FCTIWUZ";
case PPCISD::FP_TO_UINT_IN_VSR:
return "PPCISD::FP_TO_UINT_IN_VSR,";
case PPCISD::FP_TO_SINT_IN_VSR:
return "PPCISD::FP_TO_SINT_IN_VSR";
case PPCISD::FRE: return "PPCISD::FRE";
case PPCISD::FRSQRTE: return "PPCISD::FRSQRTE";
case PPCISD::FTSQRT:
return "PPCISD::FTSQRT";
case PPCISD::FSQRT:
return "PPCISD::FSQRT";
case PPCISD::STFIWX: return "PPCISD::STFIWX";
case PPCISD::VPERM: return "PPCISD::VPERM";
case PPCISD::XXSPLT: return "PPCISD::XXSPLT";
case PPCISD::XXSPLTI_SP_TO_DP:
return "PPCISD::XXSPLTI_SP_TO_DP";
case PPCISD::XXSPLTI32DX:
return "PPCISD::XXSPLTI32DX";
case PPCISD::VECINSERT: return "PPCISD::VECINSERT";
case PPCISD::XXPERMDI: return "PPCISD::XXPERMDI";
case PPCISD::VECSHL: return "PPCISD::VECSHL";
case PPCISD::CMPB: return "PPCISD::CMPB";
case PPCISD::Hi: return "PPCISD::Hi";
case PPCISD::Lo: return "PPCISD::Lo";
case PPCISD::TOC_ENTRY: return "PPCISD::TOC_ENTRY";
case PPCISD::ATOMIC_CMP_SWAP_8: return "PPCISD::ATOMIC_CMP_SWAP_8";
case PPCISD::ATOMIC_CMP_SWAP_16: return "PPCISD::ATOMIC_CMP_SWAP_16";
case PPCISD::DYNALLOC: return "PPCISD::DYNALLOC";
case PPCISD::DYNAREAOFFSET: return "PPCISD::DYNAREAOFFSET";
case PPCISD::PROBED_ALLOCA: return "PPCISD::PROBED_ALLOCA";
case PPCISD::GlobalBaseReg: return "PPCISD::GlobalBaseReg";
case PPCISD::SRL: return "PPCISD::SRL";
case PPCISD::SRA: return "PPCISD::SRA";
case PPCISD::SHL: return "PPCISD::SHL";
case PPCISD::SRA_ADDZE: return "PPCISD::SRA_ADDZE";
case PPCISD::CALL: return "PPCISD::CALL";
case PPCISD::CALL_NOP: return "PPCISD::CALL_NOP";
case PPCISD::CALL_NOTOC: return "PPCISD::CALL_NOTOC";
case PPCISD::MTCTR: return "PPCISD::MTCTR";
case PPCISD::BCTRL: return "PPCISD::BCTRL";
case PPCISD::BCTRL_LOAD_TOC: return "PPCISD::BCTRL_LOAD_TOC";
case PPCISD::RET_FLAG: return "PPCISD::RET_FLAG";
case PPCISD::READ_TIME_BASE: return "PPCISD::READ_TIME_BASE";
case PPCISD::EH_SJLJ_SETJMP: return "PPCISD::EH_SJLJ_SETJMP";
case PPCISD::EH_SJLJ_LONGJMP: return "PPCISD::EH_SJLJ_LONGJMP";
case PPCISD::MFOCRF: return "PPCISD::MFOCRF";
case PPCISD::MFVSR: return "PPCISD::MFVSR";
case PPCISD::MTVSRA: return "PPCISD::MTVSRA";
case PPCISD::MTVSRZ: return "PPCISD::MTVSRZ";
case PPCISD::SINT_VEC_TO_FP: return "PPCISD::SINT_VEC_TO_FP";
case PPCISD::UINT_VEC_TO_FP: return "PPCISD::UINT_VEC_TO_FP";
case PPCISD::SCALAR_TO_VECTOR_PERMUTED:
return "PPCISD::SCALAR_TO_VECTOR_PERMUTED";
case PPCISD::ANDI_rec_1_EQ_BIT:
return "PPCISD::ANDI_rec_1_EQ_BIT";
case PPCISD::ANDI_rec_1_GT_BIT:
return "PPCISD::ANDI_rec_1_GT_BIT";
case PPCISD::VCMP: return "PPCISD::VCMP";
case PPCISD::VCMP_rec: return "PPCISD::VCMP_rec";
case PPCISD::LBRX: return "PPCISD::LBRX";
case PPCISD::STBRX: return "PPCISD::STBRX";
case PPCISD::LFIWAX: return "PPCISD::LFIWAX";
case PPCISD::LFIWZX: return "PPCISD::LFIWZX";
case PPCISD::LXSIZX: return "PPCISD::LXSIZX";
case PPCISD::STXSIX: return "PPCISD::STXSIX";
case PPCISD::VEXTS: return "PPCISD::VEXTS";
case PPCISD::LXVD2X: return "PPCISD::LXVD2X";
case PPCISD::STXVD2X: return "PPCISD::STXVD2X";
case PPCISD::LOAD_VEC_BE: return "PPCISD::LOAD_VEC_BE";
case PPCISD::STORE_VEC_BE: return "PPCISD::STORE_VEC_BE";
case PPCISD::ST_VSR_SCAL_INT:
return "PPCISD::ST_VSR_SCAL_INT";
case PPCISD::COND_BRANCH: return "PPCISD::COND_BRANCH";
case PPCISD::BDNZ: return "PPCISD::BDNZ";
case PPCISD::BDZ: return "PPCISD::BDZ";
case PPCISD::MFFS: return "PPCISD::MFFS";
case PPCISD::FADDRTZ: return "PPCISD::FADDRTZ";
case PPCISD::TC_RETURN: return "PPCISD::TC_RETURN";
case PPCISD::CR6SET: return "PPCISD::CR6SET";
case PPCISD::CR6UNSET: return "PPCISD::CR6UNSET";
case PPCISD::PPC32_GOT: return "PPCISD::PPC32_GOT";
case PPCISD::PPC32_PICGOT: return "PPCISD::PPC32_PICGOT";
case PPCISD::ADDIS_GOT_TPREL_HA: return "PPCISD::ADDIS_GOT_TPREL_HA";
case PPCISD::LD_GOT_TPREL_L: return "PPCISD::LD_GOT_TPREL_L";
case PPCISD::ADD_TLS: return "PPCISD::ADD_TLS";
case PPCISD::ADDIS_TLSGD_HA: return "PPCISD::ADDIS_TLSGD_HA";
case PPCISD::ADDI_TLSGD_L: return "PPCISD::ADDI_TLSGD_L";
case PPCISD::GET_TLS_ADDR: return "PPCISD::GET_TLS_ADDR";
case PPCISD::ADDI_TLSGD_L_ADDR: return "PPCISD::ADDI_TLSGD_L_ADDR";
case PPCISD::ADDIS_TLSLD_HA: return "PPCISD::ADDIS_TLSLD_HA";
case PPCISD::ADDI_TLSLD_L: return "PPCISD::ADDI_TLSLD_L";
case PPCISD::GET_TLSLD_ADDR: return "PPCISD::GET_TLSLD_ADDR";
case PPCISD::ADDI_TLSLD_L_ADDR: return "PPCISD::ADDI_TLSLD_L_ADDR";
case PPCISD::ADDIS_DTPREL_HA: return "PPCISD::ADDIS_DTPREL_HA";
case PPCISD::ADDI_DTPREL_L: return "PPCISD::ADDI_DTPREL_L";
case PPCISD::PADDI_DTPREL:
return "PPCISD::PADDI_DTPREL";
case PPCISD::VADD_SPLAT: return "PPCISD::VADD_SPLAT";
case PPCISD::SC: return "PPCISD::SC";
case PPCISD::CLRBHRB: return "PPCISD::CLRBHRB";
case PPCISD::MFBHRBE: return "PPCISD::MFBHRBE";
case PPCISD::RFEBB: return "PPCISD::RFEBB";
case PPCISD::XXSWAPD: return "PPCISD::XXSWAPD";
case PPCISD::SWAP_NO_CHAIN: return "PPCISD::SWAP_NO_CHAIN";
case PPCISD::VABSD: return "PPCISD::VABSD";
case PPCISD::BUILD_FP128: return "PPCISD::BUILD_FP128";
case PPCISD::BUILD_SPE64: return "PPCISD::BUILD_SPE64";
case PPCISD::EXTRACT_SPE: return "PPCISD::EXTRACT_SPE";
case PPCISD::EXTSWSLI: return "PPCISD::EXTSWSLI";
case PPCISD::LD_VSX_LH: return "PPCISD::LD_VSX_LH";
case PPCISD::FP_EXTEND_HALF: return "PPCISD::FP_EXTEND_HALF";
case PPCISD::MAT_PCREL_ADDR: return "PPCISD::MAT_PCREL_ADDR";
case PPCISD::TLS_DYNAMIC_MAT_PCREL_ADDR:
return "PPCISD::TLS_DYNAMIC_MAT_PCREL_ADDR";
case PPCISD::TLS_LOCAL_EXEC_MAT_ADDR:
return "PPCISD::TLS_LOCAL_EXEC_MAT_ADDR";
case PPCISD::ACC_BUILD: return "PPCISD::ACC_BUILD";
case PPCISD::PAIR_BUILD: return "PPCISD::PAIR_BUILD";
case PPCISD::EXTRACT_VSX_REG: return "PPCISD::EXTRACT_VSX_REG";
case PPCISD::XXMFACC: return "PPCISD::XXMFACC";
case PPCISD::LD_SPLAT: return "PPCISD::LD_SPLAT";
case PPCISD::FNMSUB: return "PPCISD::FNMSUB";
case PPCISD::STRICT_FADDRTZ:
return "PPCISD::STRICT_FADDRTZ";
case PPCISD::STRICT_FCTIDZ:
return "PPCISD::STRICT_FCTIDZ";
case PPCISD::STRICT_FCTIWZ:
return "PPCISD::STRICT_FCTIWZ";
case PPCISD::STRICT_FCTIDUZ:
return "PPCISD::STRICT_FCTIDUZ";
case PPCISD::STRICT_FCTIWUZ:
return "PPCISD::STRICT_FCTIWUZ";
case PPCISD::STRICT_FCFID:
return "PPCISD::STRICT_FCFID";
case PPCISD::STRICT_FCFIDU:
return "PPCISD::STRICT_FCFIDU";
case PPCISD::STRICT_FCFIDS:
return "PPCISD::STRICT_FCFIDS";
case PPCISD::STRICT_FCFIDUS:
return "PPCISD::STRICT_FCFIDUS";
case PPCISD::LXVRZX: return "PPCISD::LXVRZX";
}
return nullptr;
}
EVT PPCTargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &C,
EVT VT) const {
if (!VT.isVector())
return Subtarget.useCRBits() ? MVT::i1 : MVT::i32;
return VT.changeVectorElementTypeToInteger();
}
bool PPCTargetLowering::enableAggressiveFMAFusion(EVT VT) const {
assert(VT.isFloatingPoint() && "Non-floating-point FMA?");
return true;
}
//===----------------------------------------------------------------------===//
// Node matching predicates, for use by the tblgen matching code.
//===----------------------------------------------------------------------===//
/// isFloatingPointZero - Return true if this is 0.0 or -0.0.
static bool isFloatingPointZero(SDValue Op) {
if (ConstantFPSDNode *CFP = dyn_cast<ConstantFPSDNode>(Op))
return CFP->getValueAPF().isZero();
else if (ISD::isEXTLoad(Op.getNode()) || ISD::isNON_EXTLoad(Op.getNode())) {
// Maybe this has already been legalized into the constant pool?
if (ConstantPoolSDNode *CP = dyn_cast<ConstantPoolSDNode>(Op.getOperand(1)))
if (const ConstantFP *CFP = dyn_cast<ConstantFP>(CP->getConstVal()))
return CFP->getValueAPF().isZero();
}
return false;
}
/// isConstantOrUndef - Op is either an undef node or a ConstantSDNode. Return
/// true if Op is undef or if it matches the specified value.
static bool isConstantOrUndef(int Op, int Val) {
return Op < 0 || Op == Val;
}
/// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUHUM instruction.
/// The ShuffleKind distinguishes between big-endian operations with
/// two different inputs (0), either-endian operations with two identical
/// inputs (1), and little-endian operations with two different inputs (2).
/// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
bool PPC::isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
SelectionDAG &DAG) {
bool IsLE = DAG.getDataLayout().isLittleEndian();
if (ShuffleKind == 0) {
if (IsLE)
return false;
for (unsigned i = 0; i != 16; ++i)
if (!isConstantOrUndef(N->getMaskElt(i), i*2+1))
return false;
} else if (ShuffleKind == 2) {
if (!IsLE)
return false;
for (unsigned i = 0; i != 16; ++i)
if (!isConstantOrUndef(N->getMaskElt(i), i*2))
return false;
} else if (ShuffleKind == 1) {
unsigned j = IsLE ? 0 : 1;
for (unsigned i = 0; i != 8; ++i)
if (!isConstantOrUndef(N->getMaskElt(i), i*2+j) ||
!isConstantOrUndef(N->getMaskElt(i+8), i*2+j))
return false;
}
return true;
}
/// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUWUM instruction.
/// The ShuffleKind distinguishes between big-endian operations with
/// two different inputs (0), either-endian operations with two identical
/// inputs (1), and little-endian operations with two different inputs (2).
/// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
bool PPC::isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
SelectionDAG &DAG) {
bool IsLE = DAG.getDataLayout().isLittleEndian();
if (ShuffleKind == 0) {
if (IsLE)
return false;
for (unsigned i = 0; i != 16; i += 2)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2+2) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+3))
return false;
} else if (ShuffleKind == 2) {
if (!IsLE)
return false;
for (unsigned i = 0; i != 16; i += 2)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+1))
return false;
} else if (ShuffleKind == 1) {
unsigned j = IsLE ? 0 : 2;
for (unsigned i = 0; i != 8; i += 2)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2+j) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+j+1) ||
!isConstantOrUndef(N->getMaskElt(i+8), i*2+j) ||
!isConstantOrUndef(N->getMaskElt(i+9), i*2+j+1))
return false;
}
return true;
}
/// isVPKUDUMShuffleMask - Return true if this is the shuffle mask for a
/// VPKUDUM instruction, AND the VPKUDUM instruction exists for the
/// current subtarget.
///
/// The ShuffleKind distinguishes between big-endian operations with
/// two different inputs (0), either-endian operations with two identical
/// inputs (1), and little-endian operations with two different inputs (2).
/// For the latter, the input operands are swapped (see PPCInstrAltivec.td).
bool PPC::isVPKUDUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
SelectionDAG &DAG) {
const PPCSubtarget& Subtarget =
static_cast<const PPCSubtarget&>(DAG.getSubtarget());
if (!Subtarget.hasP8Vector())
return false;
bool IsLE = DAG.getDataLayout().isLittleEndian();
if (ShuffleKind == 0) {
if (IsLE)
return false;
for (unsigned i = 0; i != 16; i += 4)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2+4) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+5) ||
!isConstantOrUndef(N->getMaskElt(i+2), i*2+6) ||
!isConstantOrUndef(N->getMaskElt(i+3), i*2+7))
return false;
} else if (ShuffleKind == 2) {
if (!IsLE)
return false;
for (unsigned i = 0; i != 16; i += 4)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+1) ||
!isConstantOrUndef(N->getMaskElt(i+2), i*2+2) ||
!isConstantOrUndef(N->getMaskElt(i+3), i*2+3))
return false;
} else if (ShuffleKind == 1) {
unsigned j = IsLE ? 0 : 4;
for (unsigned i = 0; i != 8; i += 4)
if (!isConstantOrUndef(N->getMaskElt(i ), i*2+j) ||
!isConstantOrUndef(N->getMaskElt(i+1), i*2+j+1) ||
!isConstantOrUndef(N->getMaskElt(i+2), i*2+j+2) ||
!isConstantOrUndef(N->getMaskElt(i+3), i*2+j+3) ||
!isConstantOrUndef(N->getMaskElt(i+8), i*2+j) ||
!isConstantOrUndef(N->getMaskElt(i+9), i*2+j+1) ||
!isConstantOrUndef(N->getMaskElt(i+10), i*2+j+2) ||
!isConstantOrUndef(N->getMaskElt(i+11), i*2+j+3))
return false;
}
return true;
}
/// isVMerge - Common function, used to match vmrg* shuffles.
///
static bool isVMerge(ShuffleVectorSDNode *N, unsigned UnitSize,
unsigned LHSStart, unsigned RHSStart) {
if (N->getValueType(0) != MVT::v16i8)
return false;
assert((UnitSize == 1 || UnitSize == 2 || UnitSize == 4) &&
"Unsupported merge size!");
for (unsigned i = 0; i != 8/UnitSize; ++i) // Step over units
for (unsigned j = 0; j != UnitSize; ++j) { // Step over bytes within unit
if (!isConstantOrUndef(N->getMaskElt(i*UnitSize*2+j),
LHSStart+j+i*UnitSize) ||
!isConstantOrUndef(N->getMaskElt(i*UnitSize*2+UnitSize+j),
RHSStart+j+i*UnitSize))
return false;
}
return true;
}
/// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
/// a VMRGL* instruction with the specified unit size (1,2 or 4 bytes).
/// The ShuffleKind distinguishes between big-endian merges with two
/// different inputs (0), either-endian merges with two identical inputs (1),
/// and little-endian merges with two different inputs (2). For the latter,
/// the input operands are swapped (see PPCInstrAltivec.td).
bool PPC::isVMRGLShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
unsigned ShuffleKind, SelectionDAG &DAG) {
if (DAG.getDataLayout().isLittleEndian()) {
if (ShuffleKind == 1) // unary
return isVMerge(N, UnitSize, 0, 0);
else if (ShuffleKind == 2) // swapped
return isVMerge(N, UnitSize, 0, 16);
else
return false;
} else {
if (ShuffleKind == 1) // unary
return isVMerge(N, UnitSize, 8, 8);
else if (ShuffleKind == 0) // normal
return isVMerge(N, UnitSize, 8, 24);
else
return false;
}
}
/// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
/// a VMRGH* instruction with the specified unit size (1,2 or 4 bytes).
/// The ShuffleKind distinguishes between big-endian merges with two
/// different inputs (0), either-endian merges with two identical inputs (1),
/// and little-endian merges with two different inputs (2). For the latter,
/// the input operands are swapped (see PPCInstrAltivec.td).
bool PPC::isVMRGHShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
unsigned ShuffleKind, SelectionDAG &DAG) {
if (DAG.getDataLayout().isLittleEndian()) {
if (ShuffleKind == 1) // unary
return isVMerge(N, UnitSize, 8, 8);
else if (ShuffleKind == 2) // swapped
return isVMerge(N, UnitSize, 8, 24);
else
return false;
} else {
if (ShuffleKind == 1) // unary
return isVMerge(N, UnitSize, 0, 0);
else if (ShuffleKind == 0) // normal
return isVMerge(N, UnitSize, 0, 16);
else
return false;
}
}
/**
* Common function used to match vmrgew and vmrgow shuffles
*
* The indexOffset determines whether to look for even or odd words in
* the shuffle mask. This is based on the of the endianness of the target
* machine.
* - Little Endian:
* - Use offset of 0 to check for odd elements
* - Use offset of 4 to check for even elements
* - Big Endian:
* - Use offset of 0 to check for even elements
* - Use offset of 4 to check for odd elements
* A detailed description of the vector element ordering for little endian and
* big endian can be found at
* http://www.ibm.com/developerworks/library/l-ibm-xl-c-cpp-compiler/index.html
* Targeting your applications - what little endian and big endian IBM XL C/C++
* compiler differences mean to you
*
* The mask to the shuffle vector instruction specifies the indices of the
* elements from the two input vectors to place in the result. The elements are
* numbered in array-access order, starting with the first vector. These vectors
* are always of type v16i8, thus each vector will contain 16 elements of size
* 8. More info on the shuffle vector can be found in the
* http://llvm.org/docs/LangRef.html#shufflevector-instruction
* Language Reference.
*
* The RHSStartValue indicates whether the same input vectors are used (unary)
* or two different input vectors are used, based on the following:
* - If the instruction uses the same vector for both inputs, the range of the
* indices will be 0 to 15. In this case, the RHSStart value passed should
* be 0.
* - If the instruction has two different vectors then the range of the
* indices will be 0 to 31. In this case, the RHSStart value passed should
* be 16 (indices 0-15 specify elements in the first vector while indices 16
* to 31 specify elements in the second vector).
*
* \param[in] N The shuffle vector SD Node to analyze
* \param[in] IndexOffset Specifies whether to look for even or odd elements
* \param[in] RHSStartValue Specifies the starting index for the righthand input
* vector to the shuffle_vector instruction
* \return true iff this shuffle vector represents an even or odd word merge
*/
static bool isVMerge(ShuffleVectorSDNode *N, unsigned IndexOffset,
unsigned RHSStartValue) {
if (N->getValueType(0) != MVT::v16i8)
return false;
for (unsigned i = 0; i < 2; ++i)
for (unsigned j = 0; j < 4; ++j)
if (!isConstantOrUndef(N->getMaskElt(i*4+j),
i*RHSStartValue+j+IndexOffset) ||
!isConstantOrUndef(N->getMaskElt(i*4+j+8),
i*RHSStartValue+j+IndexOffset+8))
return false;
return true;
}
/**
* Determine if the specified shuffle mask is suitable for the vmrgew or
* vmrgow instructions.
*
* \param[in] N The shuffle vector SD Node to analyze
* \param[in] CheckEven Check for an even merge (true) or an odd merge (false)
* \param[in] ShuffleKind Identify the type of merge:
* - 0 = big-endian merge with two different inputs;
* - 1 = either-endian merge with two identical inputs;
* - 2 = little-endian merge with two different inputs (inputs are swapped for
* little-endian merges).
* \param[in] DAG The current SelectionDAG
* \return true iff this shuffle mask
*/
bool PPC::isVMRGEOShuffleMask(ShuffleVectorSDNode *N, bool CheckEven,
unsigned ShuffleKind, SelectionDAG &DAG) {
if (DAG.getDataLayout().isLittleEndian()) {
unsigned indexOffset = CheckEven ? 4 : 0;
if (ShuffleKind == 1) // Unary
return isVMerge(N, indexOffset, 0);
else if (ShuffleKind == 2) // swapped
return isVMerge(N, indexOffset, 16);
else
return false;
}
else {
unsigned indexOffset = CheckEven ? 0 : 4;
if (ShuffleKind == 1) // Unary
return isVMerge(N, indexOffset, 0);
else if (ShuffleKind == 0) // Normal
return isVMerge(N, indexOffset, 16);
else
return false;
}
return false;
}
/// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
/// amount, otherwise return -1.
/// The ShuffleKind distinguishes between big-endian operations with two
/// different inputs (0), either-endian operations with two identical inputs
/// (1), and little-endian operations with two different inputs (2). For the
/// latter, the input operands are swapped (see PPCInstrAltivec.td).
int PPC::isVSLDOIShuffleMask(SDNode *N, unsigned ShuffleKind,
SelectionDAG &DAG) {
if (N->getValueType(0) != MVT::v16i8)
return -1;
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(N);
// Find the first non-undef value in the shuffle mask.
unsigned i;
for (i = 0; i != 16 && SVOp->getMaskElt(i) < 0; ++i)
/*search*/;
if (i == 16) return -1; // all undef.
// Otherwise, check to see if the rest of the elements are consecutively
// numbered from this value.
unsigned ShiftAmt = SVOp->getMaskElt(i);
if (ShiftAmt < i) return -1;
ShiftAmt -= i;
bool isLE = DAG.getDataLayout().isLittleEndian();
if ((ShuffleKind == 0 && !isLE) || (ShuffleKind == 2 && isLE)) {
// Check the rest of the elements to see if they are consecutive.
for (++i; i != 16; ++i)
if (!isConstantOrUndef(SVOp->getMaskElt(i), ShiftAmt+i))
return -1;
} else if (ShuffleKind == 1) {
// Check the rest of the elements to see if they are consecutive.
for (++i; i != 16; ++i)
if (!isConstantOrUndef(SVOp->getMaskElt(i), (ShiftAmt+i) & 15))
return -1;
} else
return -1;
if (isLE)
ShiftAmt = 16 - ShiftAmt;
return ShiftAmt;
}
/// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
/// specifies a splat of a single element that is suitable for input to
/// one of the splat operations (VSPLTB/VSPLTH/VSPLTW/XXSPLTW/LXVDSX/etc.).
bool PPC::isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize) {
assert(N->getValueType(0) == MVT::v16i8 && isPowerOf2_32(EltSize) &&
EltSize <= 8 && "Can only handle 1,2,4,8 byte element sizes");
// The consecutive indices need to specify an element, not part of two
// different elements. So abandon ship early if this isn't the case.
if (N->getMaskElt(0) % EltSize != 0)
return false;
// This is a splat operation if each element of the permute is the same, and
// if the value doesn't reference the second vector.
unsigned ElementBase = N->getMaskElt(0);
// FIXME: Handle UNDEF elements too!
if (ElementBase >= 16)
return false;
// Check that the indices are consecutive, in the case of a multi-byte element
// splatted with a v16i8 mask.
for (unsigned i = 1; i != EltSize; ++i)
if (N->getMaskElt(i) < 0 || N->getMaskElt(i) != (int)(i+ElementBase))
return false;
for (unsigned i = EltSize, e = 16; i != e; i += EltSize) {
if (N->getMaskElt(i) < 0) continue;
for (unsigned j = 0; j != EltSize; ++j)
if (N->getMaskElt(i+j) != N->getMaskElt(j))
return false;
}
return true;
}
/// Check that the mask is shuffling N byte elements. Within each N byte
/// element of the mask, the indices could be either in increasing or
/// decreasing order as long as they are consecutive.
/// \param[in] N the shuffle vector SD Node to analyze
/// \param[in] Width the element width in bytes, could be 2/4/8/16 (HalfWord/
/// Word/DoubleWord/QuadWord).
/// \param[in] StepLen the delta indices number among the N byte element, if
/// the mask is in increasing/decreasing order then it is 1/-1.
/// \return true iff the mask is shuffling N byte elements.
static bool isNByteElemShuffleMask(ShuffleVectorSDNode *N, unsigned Width,
int StepLen) {
assert((Width == 2 || Width == 4 || Width == 8 || Width == 16) &&
"Unexpected element width.");
assert((StepLen == 1 || StepLen == -1) && "Unexpected element width.");
unsigned NumOfElem = 16 / Width;
unsigned MaskVal[16]; // Width is never greater than 16
for (unsigned i = 0; i < NumOfElem; ++i) {
MaskVal[0] = N->getMaskElt(i * Width);
if ((StepLen == 1) && (MaskVal[0] % Width)) {
return false;
} else if ((StepLen == -1) && ((MaskVal[0] + 1) % Width)) {
return false;
}
for (unsigned int j = 1; j < Width; ++j) {
MaskVal[j] = N->getMaskElt(i * Width + j);
if (MaskVal[j] != MaskVal[j-1] + StepLen) {
return false;
}
}
}
return true;
}
bool PPC::isXXINSERTWMask(ShuffleVectorSDNode *N, unsigned &ShiftElts,
unsigned &InsertAtByte, bool &Swap, bool IsLE) {
if (!isNByteElemShuffleMask(N, 4, 1))
return false;
// Now we look at mask elements 0,4,8,12
unsigned M0 = N->getMaskElt(0) / 4;
unsigned M1 = N->getMaskElt(4) / 4;
unsigned M2 = N->getMaskElt(8) / 4;
unsigned M3 = N->getMaskElt(12) / 4;
unsigned LittleEndianShifts[] = { 2, 1, 0, 3 };
unsigned BigEndianShifts[] = { 3, 0, 1, 2 };
// Below, let H and L be arbitrary elements of the shuffle mask
// where H is in the range [4,7] and L is in the range [0,3].
// H, 1, 2, 3 or L, 5, 6, 7
if ((M0 > 3 && M1 == 1 && M2 == 2 && M3 == 3) ||
(M0 < 4 && M1 == 5 && M2 == 6 && M3 == 7)) {
ShiftElts = IsLE ? LittleEndianShifts[M0 & 0x3] : BigEndianShifts[M0 & 0x3];
InsertAtByte = IsLE ? 12 : 0;
Swap = M0 < 4;
return true;
}
// 0, H, 2, 3 or 4, L, 6, 7
if ((M1 > 3 && M0 == 0 && M2 == 2 && M3 == 3) ||
(M1 < 4 && M0 == 4 && M2 == 6 && M3 == 7)) {
ShiftElts = IsLE ? LittleEndianShifts[M1 & 0x3] : BigEndianShifts[M1 & 0x3];
InsertAtByte = IsLE ? 8 : 4;
Swap = M1 < 4;
return true;
}
// 0, 1, H, 3 or 4, 5, L, 7
if ((M2 > 3 && M0 == 0 && M1 == 1 && M3 == 3) ||
(M2 < 4 && M0 == 4 && M1 == 5 && M3 == 7)) {
ShiftElts = IsLE ? LittleEndianShifts[M2 & 0x3] : BigEndianShifts[M2 & 0x3];
InsertAtByte = IsLE ? 4 : 8;
Swap = M2 < 4;
return true;
}
// 0, 1, 2, H or 4, 5, 6, L
if ((M3 > 3 && M0 == 0 && M1 == 1 && M2 == 2) ||
(M3 < 4 && M0 == 4 && M1 == 5 && M2 == 6)) {
ShiftElts = IsLE ? LittleEndianShifts[M3 & 0x3] : BigEndianShifts[M3 & 0x3];
InsertAtByte = IsLE ? 0 : 12;
Swap = M3 < 4;
return true;
}
// If both vector operands for the shuffle are the same vector, the mask will
// contain only elements from the first one and the second one will be undef.
if (N->getOperand(1).isUndef()) {
ShiftElts = 0;
Swap = true;
unsigned XXINSERTWSrcElem = IsLE ? 2 : 1;
if (M0 == XXINSERTWSrcElem && M1 == 1 && M2 == 2 && M3 == 3) {
InsertAtByte = IsLE ? 12 : 0;
return true;
}
if (M0 == 0 && M1 == XXINSERTWSrcElem && M2 == 2 && M3 == 3) {
InsertAtByte = IsLE ? 8 : 4;
return true;
}
if (M0 == 0 && M1 == 1 && M2 == XXINSERTWSrcElem && M3 == 3) {
InsertAtByte = IsLE ? 4 : 8;
return true;
}
if (M0 == 0 && M1 == 1 && M2 == 2 && M3 == XXINSERTWSrcElem) {
InsertAtByte = IsLE ? 0 : 12;
return true;
}
}
return false;
}
bool PPC::isXXSLDWIShuffleMask(ShuffleVectorSDNode *N, unsigned &ShiftElts,
bool &Swap, bool IsLE) {
assert(N->getValueType(0) == MVT::v16i8 && "Shuffle vector expects v16i8");
// Ensure each byte index of the word is consecutive.
if (!isNByteElemShuffleMask(N, 4, 1))
return false;
// Now we look at mask elements 0,4,8,12, which are the beginning of words.
unsigned M0 = N->getMaskElt(0) / 4;
unsigned M1 = N->getMaskElt(4) / 4;
unsigned M2 = N->getMaskElt(8) / 4;
unsigned M3 = N->getMaskElt(12) / 4;
// If both vector operands for the shuffle are the same vector, the mask will
// contain only elements from the first one and the second one will be undef.
if (N->getOperand(1).isUndef()) {
assert(M0 < 4 && "Indexing into an undef vector?");
if (M1 != (M0 + 1) % 4 || M2 != (M1 + 1) % 4 || M3 != (M2 + 1) % 4)
return false;
ShiftElts = IsLE ? (4 - M0) % 4 : M0;
Swap = false;
return true;
}
// Ensure each word index of the ShuffleVector Mask is consecutive.
if (M1 != (M0 + 1) % 8 || M2 != (M1 + 1) % 8 || M3 != (M2 + 1) % 8)
return false;
if (IsLE) {
if (M0 == 0 || M0 == 7 || M0 == 6 || M0 == 5) {
// Input vectors don't need to be swapped if the leading element
// of the result is one of the 3 left elements of the second vector
// (or if there is no shift to be done at all).
Swap = false;
ShiftElts = (8 - M0) % 8;
} else if (M0 == 4 || M0 == 3 || M0 == 2 || M0 == 1) {
// Input vectors need to be swapped if the leading element
// of the result is one of the 3 left elements of the first vector
// (or if we're shifting by 4 - thereby simply swapping the vectors).
Swap = true;
ShiftElts = (4 - M0) % 4;
}
return true;
} else { // BE
if (M0 == 0 || M0 == 1 || M0 == 2 || M0 == 3) {
// Input vectors don't need to be swapped if the leading element
// of the result is one of the 4 elements of the first vector.
Swap = false;
ShiftElts = M0;
} else if (M0 == 4 || M0 == 5 || M0 == 6 || M0 == 7) {
// Input vectors need to be swapped if the leading element
// of the result is one of the 4 elements of the right vector.
Swap = true;
ShiftElts = M0 - 4;
}
return true;
}
}
bool static isXXBRShuffleMaskHelper(ShuffleVectorSDNode *N, int Width) {
assert(N->getValueType(0) == MVT::v16i8 && "Shuffle vector expects v16i8");
if (!isNByteElemShuffleMask(N, Width, -1))
return false;
for (int i = 0; i < 16; i += Width)
if (N->getMaskElt(i) != i + Width - 1)
return false;
return true;
}
bool PPC::isXXBRHShuffleMask(ShuffleVectorSDNode *N) {
return isXXBRShuffleMaskHelper(N, 2);
}
bool PPC::isXXBRWShuffleMask(ShuffleVectorSDNode *N) {
return isXXBRShuffleMaskHelper(N, 4);
}
bool PPC::isXXBRDShuffleMask(ShuffleVectorSDNode *N) {
return isXXBRShuffleMaskHelper(N, 8);
}
bool PPC::isXXBRQShuffleMask(ShuffleVectorSDNode *N) {
return isXXBRShuffleMaskHelper(N, 16);
}
/// Can node \p N be lowered to an XXPERMDI instruction? If so, set \p Swap
/// if the inputs to the instruction should be swapped and set \p DM to the
/// value for the immediate.
/// Specifically, set \p Swap to true only if \p N can be lowered to XXPERMDI
/// AND element 0 of the result comes from the first input (LE) or second input
/// (BE). Set \p DM to the calculated result (0-3) only if \p N can be lowered.
/// \return true iff the given mask of shuffle node \p N is a XXPERMDI shuffle
/// mask.
bool PPC::isXXPERMDIShuffleMask(ShuffleVectorSDNode *N, unsigned &DM,
bool &Swap, bool IsLE) {
assert(N->getValueType(0) == MVT::v16i8 && "Shuffle vector expects v16i8");
// Ensure each byte index of the double word is consecutive.
if (!isNByteElemShuffleMask(N, 8, 1))
return false;
unsigned M0 = N->getMaskElt(0) / 8;
unsigned M1 = N->getMaskElt(8) / 8;
assert(((M0 | M1) < 4) && "A mask element out of bounds?");
// If both vector operands for the shuffle are the same vector, the mask will
// contain only elements from the first one and the second one will be undef.
if (N->getOperand(1).isUndef()) {
if ((M0 | M1) < 2) {
DM = IsLE ? (((~M1) & 1) << 1) + ((~M0) & 1) : (M0 << 1) + (M1 & 1);
Swap = false;
return true;
} else
return false;
}
if (IsLE) {
if (M0 > 1 && M1 < 2) {
Swap = false;
} else if (M0 < 2 && M1 > 1) {
M0 = (M0 + 2) % 4;
M1 = (M1 + 2) % 4;
Swap = true;
} else
return false;
// Note: if control flow comes here that means Swap is already set above
DM = (((~M1) & 1) << 1) + ((~M0) & 1);
return true;
} else { // BE
if (M0 < 2 && M1 > 1) {
Swap = false;
} else if (M0 > 1 && M1 < 2) {
M0 = (M0 + 2) % 4;
M1 = (M1 + 2) % 4;
Swap = true;
} else
return false;
// Note: if control flow comes here that means Swap is already set above
DM = (M0 << 1) + (M1 & 1);
return true;
}
}
/// getSplatIdxForPPCMnemonics - Return the splat index as a value that is
/// appropriate for PPC mnemonics (which have a big endian bias - namely
/// elements are counted from the left of the vector register).
unsigned PPC::getSplatIdxForPPCMnemonics(SDNode *N, unsigned EltSize,
SelectionDAG &DAG) {
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(N);
assert(isSplatShuffleMask(SVOp, EltSize));
if (DAG.getDataLayout().isLittleEndian())
return (16 / EltSize) - 1 - (SVOp->getMaskElt(0) / EltSize);
else
return SVOp->getMaskElt(0) / EltSize;
}
/// get_VSPLTI_elt - If this is a build_vector of constants which can be formed
/// by using a vspltis[bhw] instruction of the specified element size, return
/// the constant being splatted. The ByteSize field indicates the number of
/// bytes of each element [124] -> [bhw].
SDValue PPC::get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG) {
SDValue OpVal(nullptr, 0);
// If ByteSize of the splat is bigger than the element size of the
// build_vector, then we have a case where we are checking for a splat where
// multiple elements of the buildvector are folded together into a single
// logical element of the splat (e.g. "vsplish 1" to splat {0,1}*8).
unsigned EltSize = 16/N->getNumOperands();
if (EltSize < ByteSize) {
unsigned Multiple = ByteSize/EltSize; // Number of BV entries per spltval.
SDValue UniquedVals[4];
assert(Multiple > 1 && Multiple <= 4 && "How can this happen?");
// See if all of the elements in the buildvector agree across.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).isUndef()) continue;
// If the element isn't a constant, bail fully out.
if (!isa<ConstantSDNode>(N->getOperand(i))) return SDValue();
if (!UniquedVals[i&(Multiple-1)].getNode())
UniquedVals[i&(Multiple-1)] = N->getOperand(i);
else if (UniquedVals[i&(Multiple-1)] != N->getOperand(i))
return SDValue(); // no match.
}
// Okay, if we reached this point, UniquedVals[0..Multiple-1] contains
// either constant or undef values that are identical for each chunk. See
// if these chunks can form into a larger vspltis*.
// Check to see if all of the leading entries are either 0 or -1. If
// neither, then this won't fit into the immediate field.
bool LeadingZero = true;
bool LeadingOnes = true;
for (unsigned i = 0; i != Multiple-1; ++i) {
if (!UniquedVals[i].getNode()) continue; // Must have been undefs.
LeadingZero &= isNullConstant(UniquedVals[i]);
LeadingOnes &= isAllOnesConstant(UniquedVals[i]);
}
// Finally, check the least significant entry.
if (LeadingZero) {
if (!UniquedVals[Multiple-1].getNode())
return DAG.getTargetConstant(0, SDLoc(N), MVT::i32); // 0,0,0,undef
int Val = cast<ConstantSDNode>(UniquedVals[Multiple-1])->getZExtValue();
if (Val < 16) // 0,0,0,4 -> vspltisw(4)
return DAG.getTargetConstant(Val, SDLoc(N), MVT::i32);
}
if (LeadingOnes) {
if (!UniquedVals[Multiple-1].getNode())
return DAG.getTargetConstant(~0U, SDLoc(N), MVT::i32); // -1,-1,-1,undef
int Val =cast<ConstantSDNode>(UniquedVals[Multiple-1])->getSExtValue();
if (Val >= -16) // -1,-1,-1,-2 -> vspltisw(-2)
return DAG.getTargetConstant(Val, SDLoc(N), MVT::i32);
}
return SDValue();
}
// Check to see if this buildvec has a single non-undef value in its elements.
for (unsigned i = 0, e = N->getNumOperands(); i != e; ++i) {
if (N->getOperand(i).isUndef()) continue;
if (!OpVal.getNode())
OpVal = N->getOperand(i);
else if (OpVal != N->getOperand(i))
return SDValue();
}
if (!OpVal.getNode()) return SDValue(); // All UNDEF: use implicit def.
unsigned ValSizeInBytes = EltSize;
uint64_t Value = 0;
if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(OpVal)) {
Value = CN->getZExtValue();
} else if (ConstantFPSDNode *CN = dyn_cast<ConstantFPSDNode>(OpVal)) {
assert(CN->getValueType(0) == MVT::f32 && "Only one legal FP vector type!");
Value = FloatToBits(CN->getValueAPF().convertToFloat());
}
// If the splat value is larger than the element value, then we can never do
// this splat. The only case that we could fit the replicated bits into our
// immediate field for would be zero, and we prefer to use vxor for it.
if (ValSizeInBytes < ByteSize) return SDValue();
// If the element value is larger than the splat value, check if it consists
// of a repeated bit pattern of size ByteSize.
if (!APInt(ValSizeInBytes * 8, Value).isSplat(ByteSize * 8))
return SDValue();
// Properly sign extend the value.
int MaskVal = SignExtend32(Value, ByteSize * 8);
// If this is zero, don't match, zero matches ISD::isBuildVectorAllZeros.
if (MaskVal == 0) return SDValue();
// Finally, if this value fits in a 5 bit sext field, return it
if (SignExtend32<5>(MaskVal) == MaskVal)
return DAG.getTargetConstant(MaskVal, SDLoc(N), MVT::i32);
return SDValue();
}
//===----------------------------------------------------------------------===//
// Addressing Mode Selection
//===----------------------------------------------------------------------===//
/// isIntS16Immediate - This method tests to see if the node is either a 32-bit
/// or 64-bit immediate, and if the value can be accurately represented as a
/// sign extension from a 16-bit value. If so, this returns true and the
/// immediate.
bool llvm::isIntS16Immediate(SDNode *N, int16_t &Imm) {
if (!isa<ConstantSDNode>(N))
return false;
Imm = (int16_t)cast<ConstantSDNode>(N)->getZExtValue();
if (N->getValueType(0) == MVT::i32)
return Imm == (int32_t)cast<ConstantSDNode>(N)->getZExtValue();
else
return Imm == (int64_t)cast<ConstantSDNode>(N)->getZExtValue();
}
bool llvm::isIntS16Immediate(SDValue Op, int16_t &Imm) {
return isIntS16Immediate(Op.getNode(), Imm);
}
/// SelectAddressEVXRegReg - Given the specified address, check to see if it can
/// be represented as an indexed [r+r] operation.
bool PPCTargetLowering::SelectAddressEVXRegReg(SDValue N, SDValue &Base,
SDValue &Index,
SelectionDAG &DAG) const {
for (SDNode::use_iterator UI = N->use_begin(), E = N->use_end();
UI != E; ++UI) {
if (MemSDNode *Memop = dyn_cast<MemSDNode>(*UI)) {
if (Memop->getMemoryVT() == MVT::f64) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
}
}
}
return false;
}
/// isIntS34Immediate - This method tests if value of node given can be
/// accurately represented as a sign extension from a 34-bit value. If so,
/// this returns true and the immediate.
bool llvm::isIntS34Immediate(SDNode *N, int64_t &Imm) {
if (!isa<ConstantSDNode>(N))
return false;
Imm = (int64_t)cast<ConstantSDNode>(N)->getZExtValue();
return isInt<34>(Imm);
}
bool llvm::isIntS34Immediate(SDValue Op, int64_t &Imm) {
return isIntS34Immediate(Op.getNode(), Imm);
}
/// SelectAddressRegReg - Given the specified addressed, check to see if it
/// can be represented as an indexed [r+r] operation. Returns false if it
/// can be more efficiently represented as [r+imm]. If \p EncodingAlignment is
/// non-zero and N can be represented by a base register plus a signed 16-bit
/// displacement, make a more precise judgement by checking (displacement % \p
/// EncodingAlignment).
bool PPCTargetLowering::SelectAddressRegReg(
SDValue N, SDValue &Base, SDValue &Index, SelectionDAG &DAG,
MaybeAlign EncodingAlignment) const {
// If we have a PC Relative target flag don't select as [reg+reg]. It will be
// a [pc+imm].
if (SelectAddressPCRel(N, Base))
return false;
int16_t Imm = 0;
if (N.getOpcode() == ISD::ADD) {
// Is there any SPE load/store (f64), which can't handle 16bit offset?
// SPE load/store can only handle 8-bit offsets.
if (hasSPE() && SelectAddressEVXRegReg(N, Base, Index, DAG))
return true;
if (isIntS16Immediate(N.getOperand(1), Imm) &&
(!EncodingAlignment || isAligned(*EncodingAlignment, Imm)))
return false; // r+i
if (N.getOperand(1).getOpcode() == PPCISD::Lo)
return false; // r+i
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
} else if (N.getOpcode() == ISD::OR) {
if (isIntS16Immediate(N.getOperand(1), Imm) &&
(!EncodingAlignment || isAligned(*EncodingAlignment, Imm)))
return false; // r+i can fold it if we can.
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are provably
// disjoint.
KnownBits LHSKnown = DAG.computeKnownBits(N.getOperand(0));
if (LHSKnown.Zero.getBoolValue()) {
KnownBits RHSKnown = DAG.computeKnownBits(N.getOperand(1));
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
if (~(LHSKnown.Zero | RHSKnown.Zero) == 0) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
}
}
}
return false;
}
// If we happen to be doing an i64 load or store into a stack slot that has
// less than a 4-byte alignment, then the frame-index elimination may need to
// use an indexed load or store instruction (because the offset may not be a
// multiple of 4). The extra register needed to hold the offset comes from the
// register scavenger, and it is possible that the scavenger will need to use
// an emergency spill slot. As a result, we need to make sure that a spill slot
// is allocated when doing an i64 load/store into a less-than-4-byte-aligned
// stack slot.
static void fixupFuncForFI(SelectionDAG &DAG, int FrameIdx, EVT VT) {
// FIXME: This does not handle the LWA case.
if (VT != MVT::i64)
return;
// NOTE: We'll exclude negative FIs here, which come from argument
// lowering, because there are no known test cases triggering this problem
// using packed structures (or similar). We can remove this exclusion if
// we find such a test case. The reason why this is so test-case driven is
// because this entire 'fixup' is only to prevent crashes (from the
// register scavenger) on not-really-valid inputs. For example, if we have:
// %a = alloca i1
// %b = bitcast i1* %a to i64*
// store i64* a, i64 b
// then the store should really be marked as 'align 1', but is not. If it
// were marked as 'align 1' then the indexed form would have been
// instruction-selected initially, and the problem this 'fixup' is preventing
// won't happen regardless.
if (FrameIdx < 0)
return;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
if (MFI.getObjectAlign(FrameIdx) >= Align(4))
return;
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setHasNonRISpills();
}
/// Returns true if the address N can be represented by a base register plus
/// a signed 16-bit displacement [r+imm], and if it is not better
/// represented as reg+reg. If \p EncodingAlignment is non-zero, only accept
/// displacements that are multiples of that value.
bool PPCTargetLowering::SelectAddressRegImm(
SDValue N, SDValue &Disp, SDValue &Base, SelectionDAG &DAG,
MaybeAlign EncodingAlignment) const {
// FIXME dl should come from parent load or store, not from address
SDLoc dl(N);
// If we have a PC Relative target flag don't select as [reg+imm]. It will be
// a [pc+imm].
if (SelectAddressPCRel(N, Base))
return false;
// If this can be more profitably realized as r+r, fail.
if (SelectAddressRegReg(N, Disp, Base, DAG, EncodingAlignment))
return false;
if (N.getOpcode() == ISD::ADD) {
int16_t imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) &&
(!EncodingAlignment || isAligned(*EncodingAlignment, imm))) {
Disp = DAG.getTargetConstant(imm, dl, N.getValueType());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0))) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
fixupFuncForFI(DAG, FI->getIndex(), N.getValueType());
} else {
Base = N.getOperand(0);
}
return true; // [r+i]
} else if (N.getOperand(1).getOpcode() == PPCISD::Lo) {
// Match LOAD (ADD (X, Lo(G))).
assert(!cast<ConstantSDNode>(N.getOperand(1).getOperand(1))->getZExtValue()
&& "Cannot handle constant offsets yet!");
Disp = N.getOperand(1).getOperand(0); // The global address.
assert(Disp.getOpcode() == ISD::TargetGlobalAddress ||
Disp.getOpcode() == ISD::TargetGlobalTLSAddress ||
Disp.getOpcode() == ISD::TargetConstantPool ||
Disp.getOpcode() == ISD::TargetJumpTable);
Base = N.getOperand(0);
return true; // [&g+r]
}
} else if (N.getOpcode() == ISD::OR) {
int16_t imm = 0;
if (isIntS16Immediate(N.getOperand(1), imm) &&
(!EncodingAlignment || isAligned(*EncodingAlignment, imm))) {
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
KnownBits LHSKnown = DAG.computeKnownBits(N.getOperand(0));
if ((LHSKnown.Zero.getZExtValue()|~(uint64_t)imm) == ~0ULL) {
// If all of the bits are known zero on the LHS or RHS, the add won't
// carry.
if (FrameIndexSDNode *FI =
dyn_cast<FrameIndexSDNode>(N.getOperand(0))) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
fixupFuncForFI(DAG, FI->getIndex(), N.getValueType());
} else {
Base = N.getOperand(0);
}
Disp = DAG.getTargetConstant(imm, dl, N.getValueType());
return true;
}
}
} else if (ConstantSDNode *CN = dyn_cast<ConstantSDNode>(N)) {
// Loading from a constant address.
// If this address fits entirely in a 16-bit sext immediate field, codegen
// this as "d, 0"
int16_t Imm;
if (isIntS16Immediate(CN, Imm) &&
(!EncodingAlignment || isAligned(*EncodingAlignment, Imm))) {
Disp = DAG.getTargetConstant(Imm, dl, CN->getValueType(0));
Base = DAG.getRegister(Subtarget.isPPC64() ? PPC::ZERO8 : PPC::ZERO,
CN->getValueType(0));
return true;
}
// Handle 32-bit sext immediates with LIS + addr mode.
if ((CN->getValueType(0) == MVT::i32 ||
(int64_t)CN->getZExtValue() == (int)CN->getZExtValue()) &&
(!EncodingAlignment ||
isAligned(*EncodingAlignment, CN->getZExtValue()))) {
int Addr = (int)CN->getZExtValue();
// Otherwise, break this down into an LIS + disp.
Disp = DAG.getTargetConstant((short)Addr, dl, MVT::i32);
Base = DAG.getTargetConstant((Addr - (signed short)Addr) >> 16, dl,
MVT::i32);
unsigned Opc = CN->getValueType(0) == MVT::i32 ? PPC::LIS : PPC::LIS8;
Base = SDValue(DAG.getMachineNode(Opc, dl, CN->getValueType(0), Base), 0);
return true;
}
}
Disp = DAG.getTargetConstant(0, dl, getPointerTy(DAG.getDataLayout()));
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N)) {
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
fixupFuncForFI(DAG, FI->getIndex(), N.getValueType());
} else
Base = N;
return true; // [r+0]
}
/// Similar to the 16-bit case but for instructions that take a 34-bit
/// displacement field (prefixed loads/stores).
bool PPCTargetLowering::SelectAddressRegImm34(SDValue N, SDValue &Disp,
SDValue &Base,
SelectionDAG &DAG) const {
// Only on 64-bit targets.
if (N.getValueType() != MVT::i64)
return false;
SDLoc dl(N);
int64_t Imm = 0;
if (N.getOpcode() == ISD::ADD) {
if (!isIntS34Immediate(N.getOperand(1), Imm))
return false;
Disp = DAG.getTargetConstant(Imm, dl, N.getValueType());
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0)))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
else
Base = N.getOperand(0);
return true;
}
if (N.getOpcode() == ISD::OR) {
if (!isIntS34Immediate(N.getOperand(1), Imm))
return false;
// If this is an or of disjoint bitfields, we can codegen this as an add
// (for better address arithmetic) if the LHS and RHS of the OR are
// provably disjoint.
KnownBits LHSKnown = DAG.computeKnownBits(N.getOperand(0));
if ((LHSKnown.Zero.getZExtValue() | ~(uint64_t)Imm) != ~0ULL)
return false;
if (FrameIndexSDNode *FI = dyn_cast<FrameIndexSDNode>(N.getOperand(0)))
Base = DAG.getTargetFrameIndex(FI->getIndex(), N.getValueType());
else
Base = N.getOperand(0);
Disp = DAG.getTargetConstant(Imm, dl, N.getValueType());
return true;
}
if (isIntS34Immediate(N, Imm)) { // If the address is a 34-bit const.
Disp = DAG.getTargetConstant(Imm, dl, N.getValueType());
Base = DAG.getRegister(PPC::ZERO8, N.getValueType());
return true;
}
return false;
}
/// SelectAddressRegRegOnly - Given the specified addressed, force it to be
/// represented as an indexed [r+r] operation.
bool PPCTargetLowering::SelectAddressRegRegOnly(SDValue N, SDValue &Base,
SDValue &Index,
SelectionDAG &DAG) const {
// Check to see if we can easily represent this as an [r+r] address. This
// will fail if it thinks that the address is more profitably represented as
// reg+imm, e.g. where imm = 0.
if (SelectAddressRegReg(N, Base, Index, DAG))
return true;
// If the address is the result of an add, we will utilize the fact that the
// address calculation includes an implicit add. However, we can reduce
// register pressure if we do not materialize a constant just for use as the
// index register. We only get rid of the add if it is not an add of a
// value and a 16-bit signed constant and both have a single use.
int16_t imm = 0;
if (N.getOpcode() == ISD::ADD &&
(!isIntS16Immediate(N.getOperand(1), imm) ||
!N.getOperand(1).hasOneUse() || !N.getOperand(0).hasOneUse())) {
Base = N.getOperand(0);
Index = N.getOperand(1);
return true;
}
// Otherwise, do it the hard way, using R0 as the base register.
Base = DAG.getRegister(Subtarget.isPPC64() ? PPC::ZERO8 : PPC::ZERO,
N.getValueType());
Index = N;
return true;
}
template <typename Ty> static bool isValidPCRelNode(SDValue N) {
Ty *PCRelCand = dyn_cast<Ty>(N);
return PCRelCand && (PCRelCand->getTargetFlags() & PPCII::MO_PCREL_FLAG);
}
/// Returns true if this address is a PC Relative address.
/// PC Relative addresses are marked with the flag PPCII::MO_PCREL_FLAG
/// or if the node opcode is PPCISD::MAT_PCREL_ADDR.
bool PPCTargetLowering::SelectAddressPCRel(SDValue N, SDValue &Base) const {
// This is a materialize PC Relative node. Always select this as PC Relative.
Base = N;
if (N.getOpcode() == PPCISD::MAT_PCREL_ADDR)
return true;
if (isValidPCRelNode<ConstantPoolSDNode>(N) ||
isValidPCRelNode<GlobalAddressSDNode>(N) ||
isValidPCRelNode<JumpTableSDNode>(N) ||
isValidPCRelNode<BlockAddressSDNode>(N))
return true;
return false;
}
/// Returns true if we should use a direct load into vector instruction
/// (such as lxsd or lfd), instead of a load into gpr + direct move sequence.
static bool usePartialVectorLoads(SDNode *N, const PPCSubtarget& ST) {
// If there are any other uses other than scalar to vector, then we should
// keep it as a scalar load -> direct move pattern to prevent multiple
// loads.
LoadSDNode *LD = dyn_cast<LoadSDNode>(N);
if (!LD)
return false;
EVT MemVT = LD->getMemoryVT();
if (!MemVT.isSimple())
return false;
switch(MemVT.getSimpleVT().SimpleTy) {
case MVT::i64:
break;
case MVT::i32:
if (!ST.hasP8Vector())
return false;
break;
case MVT::i16:
case MVT::i8:
if (!ST.hasP9Vector())
return false;
break;
default:
return false;
}
SDValue LoadedVal(N, 0);
if (!LoadedVal.hasOneUse())
return false;
for (SDNode::use_iterator UI = LD->use_begin(), UE = LD->use_end();
UI != UE; ++UI)
if (UI.getUse().get().getResNo() == 0 &&
UI->getOpcode() != ISD::SCALAR_TO_VECTOR &&
UI->getOpcode() != PPCISD::SCALAR_TO_VECTOR_PERMUTED)
return false;
return true;
}
/// getPreIndexedAddressParts - returns true by value, base pointer and
/// offset pointer and addressing mode by reference if the node's address
/// can be legally represented as pre-indexed load / store address.
bool PPCTargetLowering::getPreIndexedAddressParts(SDNode *N, SDValue &Base,
SDValue &Offset,
ISD::MemIndexedMode &AM,
SelectionDAG &DAG) const {
if (DisablePPCPreinc) return false;
bool isLoad = true;
SDValue Ptr;
EVT VT;
unsigned Alignment;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
Ptr = LD->getBasePtr();
VT = LD->getMemoryVT();
Alignment = LD->getAlignment();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
Ptr = ST->getBasePtr();
VT = ST->getMemoryVT();
Alignment = ST->getAlignment();
isLoad = false;
} else
return false;
// Do not generate pre-inc forms for specific loads that feed scalar_to_vector
// instructions because we can fold these into a more efficient instruction
// instead, (such as LXSD).
if (isLoad && usePartialVectorLoads(N, Subtarget)) {
return false;
}
// PowerPC doesn't have preinc load/store instructions for vectors
if (VT.isVector())
return false;
if (SelectAddressRegReg(Ptr, Base, Offset, DAG)) {
// Common code will reject creating a pre-inc form if the base pointer
// is a frame index, or if N is a store and the base pointer is either
// the same as or a predecessor of the value being stored. Check for
// those situations here, and try with swapped Base/Offset instead.
bool Swap = false;
if (isa<FrameIndexSDNode>(Base) || isa<RegisterSDNode>(Base))
Swap = true;
else if (!isLoad) {
SDValue Val = cast<StoreSDNode>(N)->getValue();
if (Val == Base || Base.getNode()->isPredecessorOf(Val.getNode()))
Swap = true;
}
if (Swap)
std::swap(Base, Offset);
AM = ISD::PRE_INC;
return true;
}
// LDU/STU can only handle immediates that are a multiple of 4.
if (VT != MVT::i64) {
if (!SelectAddressRegImm(Ptr, Offset, Base, DAG, None))
return false;
} else {
// LDU/STU need an address with at least 4-byte alignment.
if (Alignment < 4)
return false;
if (!SelectAddressRegImm(Ptr, Offset, Base, DAG, Align(4)))
return false;
}
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
// PPC64 doesn't have lwau, but it does have lwaux. Reject preinc load of
// sext i32 to i64 when addr mode is r+i.
if (LD->getValueType(0) == MVT::i64 && LD->getMemoryVT() == MVT::i32 &&
LD->getExtensionType() == ISD::SEXTLOAD &&
isa<ConstantSDNode>(Offset))
return false;
}
AM = ISD::PRE_INC;
return true;
}
//===----------------------------------------------------------------------===//
// LowerOperation implementation
//===----------------------------------------------------------------------===//
/// Return true if we should reference labels using a PICBase, set the HiOpFlags
/// and LoOpFlags to the target MO flags.
static void getLabelAccessInfo(bool IsPIC, const PPCSubtarget &Subtarget,
unsigned &HiOpFlags, unsigned &LoOpFlags,
const GlobalValue *GV = nullptr) {
HiOpFlags = PPCII::MO_HA;
LoOpFlags = PPCII::MO_LO;
// Don't use the pic base if not in PIC relocation model.
if (IsPIC) {
HiOpFlags |= PPCII::MO_PIC_FLAG;
LoOpFlags |= PPCII::MO_PIC_FLAG;
}
}
static SDValue LowerLabelRef(SDValue HiPart, SDValue LoPart, bool isPIC,
SelectionDAG &DAG) {
SDLoc DL(HiPart);
EVT PtrVT = HiPart.getValueType();
SDValue Zero = DAG.getConstant(0, DL, PtrVT);
SDValue Hi = DAG.getNode(PPCISD::Hi, DL, PtrVT, HiPart, Zero);
SDValue Lo = DAG.getNode(PPCISD::Lo, DL, PtrVT, LoPart, Zero);
// With PIC, the first instruction is actually "GR+hi(&G)".
if (isPIC)
Hi = DAG.getNode(ISD::ADD, DL, PtrVT,
DAG.getNode(PPCISD::GlobalBaseReg, DL, PtrVT), Hi);
// Generate non-pic code that has direct accesses to the constant pool.
// The address of the global is just (hi(&g)+lo(&g)).
return DAG.getNode(ISD::ADD, DL, PtrVT, Hi, Lo);
}
static void setUsesTOCBasePtr(MachineFunction &MF) {
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setUsesTOCBasePtr();
}
static void setUsesTOCBasePtr(SelectionDAG &DAG) {
setUsesTOCBasePtr(DAG.getMachineFunction());
}
SDValue PPCTargetLowering::getTOCEntry(SelectionDAG &DAG, const SDLoc &dl,
SDValue GA) const {
const bool Is64Bit = Subtarget.isPPC64();
EVT VT = Is64Bit ? MVT::i64 : MVT::i32;
SDValue Reg = Is64Bit ? DAG.getRegister(PPC::X2, VT)
: Subtarget.isAIXABI()
? DAG.getRegister(PPC::R2, VT)
: DAG.getNode(PPCISD::GlobalBaseReg, dl, VT);
SDValue Ops[] = { GA, Reg };
return DAG.getMemIntrinsicNode(
PPCISD::TOC_ENTRY, dl, DAG.getVTList(VT, MVT::Other), Ops, VT,
MachinePointerInfo::getGOT(DAG.getMachineFunction()), None,
MachineMemOperand::MOLoad);
}
SDValue PPCTargetLowering::LowerConstantPool(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
const Constant *C = CP->getConstVal();
// 64-bit SVR4 ABI and AIX ABI code are always position-independent.
// The actual address of the GlobalValue is stored in the TOC.
if (Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) {
if (Subtarget.isUsingPCRelativeCalls()) {
SDLoc DL(CP);
EVT Ty = getPointerTy(DAG.getDataLayout());
SDValue ConstPool = DAG.getTargetConstantPool(
C, Ty, CP->getAlign(), CP->getOffset(), PPCII::MO_PCREL_FLAG);
return DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, ConstPool);
}
setUsesTOCBasePtr(DAG);
SDValue GA = DAG.getTargetConstantPool(C, PtrVT, CP->getAlign(), 0);
return getTOCEntry(DAG, SDLoc(CP), GA);
}
unsigned MOHiFlag, MOLoFlag;
bool IsPIC = isPositionIndependent();
getLabelAccessInfo(IsPIC, Subtarget, MOHiFlag, MOLoFlag);
if (IsPIC && Subtarget.isSVR4ABI()) {
SDValue GA =
DAG.getTargetConstantPool(C, PtrVT, CP->getAlign(), PPCII::MO_PIC_FLAG);
return getTOCEntry(DAG, SDLoc(CP), GA);
}
SDValue CPIHi =
DAG.getTargetConstantPool(C, PtrVT, CP->getAlign(), 0, MOHiFlag);
SDValue CPILo =
DAG.getTargetConstantPool(C, PtrVT, CP->getAlign(), 0, MOLoFlag);
return LowerLabelRef(CPIHi, CPILo, IsPIC, DAG);
}
// For 64-bit PowerPC, prefer the more compact relative encodings.
// This trades 32 bits per jump table entry for one or two instructions
// on the jump site.
unsigned PPCTargetLowering::getJumpTableEncoding() const {
if (isJumpTableRelative())
return MachineJumpTableInfo::EK_LabelDifference32;
return TargetLowering::getJumpTableEncoding();
}
bool PPCTargetLowering::isJumpTableRelative() const {
if (UseAbsoluteJumpTables)
return false;
if (Subtarget.isPPC64() || Subtarget.isAIXABI())
return true;
return TargetLowering::isJumpTableRelative();
}
SDValue PPCTargetLowering::getPICJumpTableRelocBase(SDValue Table,
SelectionDAG &DAG) const {
if (!Subtarget.isPPC64() || Subtarget.isAIXABI())
return TargetLowering::getPICJumpTableRelocBase(Table, DAG);
switch (getTargetMachine().getCodeModel()) {
case CodeModel::Small:
case CodeModel::Medium:
return TargetLowering::getPICJumpTableRelocBase(Table, DAG);
default:
return DAG.getNode(PPCISD::GlobalBaseReg, SDLoc(),
getPointerTy(DAG.getDataLayout()));
}
}
const MCExpr *
PPCTargetLowering::getPICJumpTableRelocBaseExpr(const MachineFunction *MF,
unsigned JTI,
MCContext &Ctx) const {
if (!Subtarget.isPPC64() || Subtarget.isAIXABI())
return TargetLowering::getPICJumpTableRelocBaseExpr(MF, JTI, Ctx);
switch (getTargetMachine().getCodeModel()) {
case CodeModel::Small:
case CodeModel::Medium:
return TargetLowering::getPICJumpTableRelocBaseExpr(MF, JTI, Ctx);
default:
return MCSymbolRefExpr::create(MF->getPICBaseSymbol(), Ctx);
}
}
SDValue PPCTargetLowering::LowerJumpTable(SDValue Op, SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
JumpTableSDNode *JT = cast<JumpTableSDNode>(Op);
// isUsingPCRelativeCalls() returns true when PCRelative is enabled
if (Subtarget.isUsingPCRelativeCalls()) {
SDLoc DL(JT);
EVT Ty = getPointerTy(DAG.getDataLayout());
SDValue GA =
DAG.getTargetJumpTable(JT->getIndex(), Ty, PPCII::MO_PCREL_FLAG);
SDValue MatAddr = DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, GA);
return MatAddr;
}
// 64-bit SVR4 ABI and AIX ABI code are always position-independent.
// The actual address of the GlobalValue is stored in the TOC.
if (Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) {
setUsesTOCBasePtr(DAG);
SDValue GA = DAG.getTargetJumpTable(JT->getIndex(), PtrVT);
return getTOCEntry(DAG, SDLoc(JT), GA);
}
unsigned MOHiFlag, MOLoFlag;
bool IsPIC = isPositionIndependent();
getLabelAccessInfo(IsPIC, Subtarget, MOHiFlag, MOLoFlag);
if (IsPIC && Subtarget.isSVR4ABI()) {
SDValue GA = DAG.getTargetJumpTable(JT->getIndex(), PtrVT,
PPCII::MO_PIC_FLAG);
return getTOCEntry(DAG, SDLoc(GA), GA);
}
SDValue JTIHi = DAG.getTargetJumpTable(JT->getIndex(), PtrVT, MOHiFlag);
SDValue JTILo = DAG.getTargetJumpTable(JT->getIndex(), PtrVT, MOLoFlag);
return LowerLabelRef(JTIHi, JTILo, IsPIC, DAG);
}
SDValue PPCTargetLowering::LowerBlockAddress(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
BlockAddressSDNode *BASDN = cast<BlockAddressSDNode>(Op);
const BlockAddress *BA = BASDN->getBlockAddress();
// isUsingPCRelativeCalls() returns true when PCRelative is enabled
if (Subtarget.isUsingPCRelativeCalls()) {
SDLoc DL(BASDN);
EVT Ty = getPointerTy(DAG.getDataLayout());
SDValue GA = DAG.getTargetBlockAddress(BA, Ty, BASDN->getOffset(),
PPCII::MO_PCREL_FLAG);
SDValue MatAddr = DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, GA);
return MatAddr;
}
// 64-bit SVR4 ABI and AIX ABI code are always position-independent.
// The actual BlockAddress is stored in the TOC.
if (Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) {
setUsesTOCBasePtr(DAG);
SDValue GA = DAG.getTargetBlockAddress(BA, PtrVT, BASDN->getOffset());
return getTOCEntry(DAG, SDLoc(BASDN), GA);
}
// 32-bit position-independent ELF stores the BlockAddress in the .got.
if (Subtarget.is32BitELFABI() && isPositionIndependent())
return getTOCEntry(
DAG, SDLoc(BASDN),
DAG.getTargetBlockAddress(BA, PtrVT, BASDN->getOffset()));
unsigned MOHiFlag, MOLoFlag;
bool IsPIC = isPositionIndependent();
getLabelAccessInfo(IsPIC, Subtarget, MOHiFlag, MOLoFlag);
SDValue TgtBAHi = DAG.getTargetBlockAddress(BA, PtrVT, 0, MOHiFlag);
SDValue TgtBALo = DAG.getTargetBlockAddress(BA, PtrVT, 0, MOLoFlag);
return LowerLabelRef(TgtBAHi, TgtBALo, IsPIC, DAG);
}
SDValue PPCTargetLowering::LowerGlobalTLSAddress(SDValue Op,
SelectionDAG &DAG) const {
// FIXME: TLS addresses currently use medium model code sequences,
// which is the most useful form. Eventually support for small and
// large models could be added if users need it, at the cost of
// additional complexity.
GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
if (DAG.getTarget().useEmulatedTLS())
return LowerToTLSEmulatedModel(GA, DAG);
SDLoc dl(GA);
const GlobalValue *GV = GA->getGlobal();
EVT PtrVT = getPointerTy(DAG.getDataLayout());
bool is64bit = Subtarget.isPPC64();
const Module *M = DAG.getMachineFunction().getFunction().getParent();
PICLevel::Level picLevel = M->getPICLevel();
const TargetMachine &TM = getTargetMachine();
TLSModel::Model Model = TM.getTLSModel(GV);
if (Model == TLSModel::LocalExec) {
if (Subtarget.isUsingPCRelativeCalls()) {
SDValue TLSReg = DAG.getRegister(PPC::X13, MVT::i64);
SDValue TGA = DAG.getTargetGlobalAddress(
GV, dl, PtrVT, 0, (PPCII::MO_PCREL_FLAG | PPCII::MO_TPREL_FLAG));
SDValue MatAddr =
DAG.getNode(PPCISD::TLS_LOCAL_EXEC_MAT_ADDR, dl, PtrVT, TGA);
return DAG.getNode(PPCISD::ADD_TLS, dl, PtrVT, TLSReg, MatAddr);
}
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
PPCII::MO_TPREL_HA);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
PPCII::MO_TPREL_LO);
SDValue TLSReg = is64bit ? DAG.getRegister(PPC::X13, MVT::i64)
: DAG.getRegister(PPC::R2, MVT::i32);
SDValue Hi = DAG.getNode(PPCISD::Hi, dl, PtrVT, TGAHi, TLSReg);
return DAG.getNode(PPCISD::Lo, dl, PtrVT, TGALo, Hi);
}
if (Model == TLSModel::InitialExec) {
bool IsPCRel = Subtarget.isUsingPCRelativeCalls();
SDValue TGA = DAG.getTargetGlobalAddress(
GV, dl, PtrVT, 0, IsPCRel ? PPCII::MO_GOT_TPREL_PCREL_FLAG : 0);
SDValue TGATLS = DAG.getTargetGlobalAddress(
GV, dl, PtrVT, 0,
IsPCRel ? (PPCII::MO_TLS | PPCII::MO_PCREL_FLAG) : PPCII::MO_TLS);
SDValue TPOffset;
if (IsPCRel) {
SDValue MatPCRel = DAG.getNode(PPCISD::MAT_PCREL_ADDR, dl, PtrVT, TGA);
TPOffset = DAG.getLoad(MVT::i64, dl, DAG.getEntryNode(), MatPCRel,
MachinePointerInfo());
} else {
SDValue GOTPtr;
if (is64bit) {
setUsesTOCBasePtr(DAG);
SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
GOTPtr =
DAG.getNode(PPCISD::ADDIS_GOT_TPREL_HA, dl, PtrVT, GOTReg, TGA);
} else {
if (!TM.isPositionIndependent())
GOTPtr = DAG.getNode(PPCISD::PPC32_GOT, dl, PtrVT);
else if (picLevel == PICLevel::SmallPIC)
GOTPtr = DAG.getNode(PPCISD::GlobalBaseReg, dl, PtrVT);
else
GOTPtr = DAG.getNode(PPCISD::PPC32_PICGOT, dl, PtrVT);
}
TPOffset = DAG.getNode(PPCISD::LD_GOT_TPREL_L, dl, PtrVT, TGA, GOTPtr);
}
return DAG.getNode(PPCISD::ADD_TLS, dl, PtrVT, TPOffset, TGATLS);
}
if (Model == TLSModel::GeneralDynamic) {
if (Subtarget.isUsingPCRelativeCalls()) {
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
PPCII::MO_GOT_TLSGD_PCREL_FLAG);
return DAG.getNode(PPCISD::TLS_DYNAMIC_MAT_PCREL_ADDR, dl, PtrVT, TGA);
}
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
SDValue GOTPtr;
if (is64bit) {
setUsesTOCBasePtr(DAG);
SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
GOTPtr = DAG.getNode(PPCISD::ADDIS_TLSGD_HA, dl, PtrVT,
GOTReg, TGA);
} else {
if (picLevel == PICLevel::SmallPIC)
GOTPtr = DAG.getNode(PPCISD::GlobalBaseReg, dl, PtrVT);
else
GOTPtr = DAG.getNode(PPCISD::PPC32_PICGOT, dl, PtrVT);
}
return DAG.getNode(PPCISD::ADDI_TLSGD_L_ADDR, dl, PtrVT,
GOTPtr, TGA, TGA);
}
if (Model == TLSModel::LocalDynamic) {
if (Subtarget.isUsingPCRelativeCalls()) {
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
PPCII::MO_GOT_TLSLD_PCREL_FLAG);
SDValue MatPCRel =
DAG.getNode(PPCISD::TLS_DYNAMIC_MAT_PCREL_ADDR, dl, PtrVT, TGA);
return DAG.getNode(PPCISD::PADDI_DTPREL, dl, PtrVT, MatPCRel, TGA);
}
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, 0);
SDValue GOTPtr;
if (is64bit) {
setUsesTOCBasePtr(DAG);
SDValue GOTReg = DAG.getRegister(PPC::X2, MVT::i64);
GOTPtr = DAG.getNode(PPCISD::ADDIS_TLSLD_HA, dl, PtrVT,
GOTReg, TGA);
} else {
if (picLevel == PICLevel::SmallPIC)
GOTPtr = DAG.getNode(PPCISD::GlobalBaseReg, dl, PtrVT);
else
GOTPtr = DAG.getNode(PPCISD::PPC32_PICGOT, dl, PtrVT);
}
SDValue TLSAddr = DAG.getNode(PPCISD::ADDI_TLSLD_L_ADDR, dl,
PtrVT, GOTPtr, TGA, TGA);
SDValue DtvOffsetHi = DAG.getNode(PPCISD::ADDIS_DTPREL_HA, dl,
PtrVT, TLSAddr, TGA);
return DAG.getNode(PPCISD::ADDI_DTPREL_L, dl, PtrVT, DtvOffsetHi, TGA);
}
llvm_unreachable("Unknown TLS model!");
}
SDValue PPCTargetLowering::LowerGlobalAddress(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = Op.getValueType();
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
SDLoc DL(GSDN);
const GlobalValue *GV = GSDN->getGlobal();
// 64-bit SVR4 ABI & AIX ABI code is always position-independent.
// The actual address of the GlobalValue is stored in the TOC.
if (Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) {
if (Subtarget.isUsingPCRelativeCalls()) {
EVT Ty = getPointerTy(DAG.getDataLayout());
if (isAccessedAsGotIndirect(Op)) {
SDValue GA = DAG.getTargetGlobalAddress(GV, DL, Ty, GSDN->getOffset(),
PPCII::MO_PCREL_FLAG |
PPCII::MO_GOT_FLAG);
SDValue MatPCRel = DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, GA);
SDValue Load = DAG.getLoad(MVT::i64, DL, DAG.getEntryNode(), MatPCRel,
MachinePointerInfo());
return Load;
} else {
SDValue GA = DAG.getTargetGlobalAddress(GV, DL, Ty, GSDN->getOffset(),
PPCII::MO_PCREL_FLAG);
return DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, Ty, GA);
}
}
setUsesTOCBasePtr(DAG);
SDValue GA = DAG.getTargetGlobalAddress(GV, DL, PtrVT, GSDN->getOffset());
return getTOCEntry(DAG, DL, GA);
}
unsigned MOHiFlag, MOLoFlag;
bool IsPIC = isPositionIndependent();
getLabelAccessInfo(IsPIC, Subtarget, MOHiFlag, MOLoFlag, GV);
if (IsPIC && Subtarget.isSVR4ABI()) {
SDValue GA = DAG.getTargetGlobalAddress(GV, DL, PtrVT,
GSDN->getOffset(),
PPCII::MO_PIC_FLAG);
return getTOCEntry(DAG, DL, GA);
}
SDValue GAHi =
DAG.getTargetGlobalAddress(GV, DL, PtrVT, GSDN->getOffset(), MOHiFlag);
SDValue GALo =
DAG.getTargetGlobalAddress(GV, DL, PtrVT, GSDN->getOffset(), MOLoFlag);
return LowerLabelRef(GAHi, GALo, IsPIC, DAG);
}
SDValue PPCTargetLowering::LowerSETCC(SDValue Op, SelectionDAG &DAG) const {
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(2))->get();
SDLoc dl(Op);
if (Op.getValueType() == MVT::v2i64) {
// When the operands themselves are v2i64 values, we need to do something
// special because VSX has no underlying comparison operations for these.
if (Op.getOperand(0).getValueType() == MVT::v2i64) {
// Equality can be handled by casting to the legal type for Altivec
// comparisons, everything else needs to be expanded.
if (CC == ISD::SETEQ || CC == ISD::SETNE) {
return DAG.getNode(ISD::BITCAST, dl, MVT::v2i64,
DAG.getSetCC(dl, MVT::v4i32,
DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op.getOperand(0)),
DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, Op.getOperand(1)),
CC));
}
return SDValue();
}
// We handle most of these in the usual way.
return Op;
}
// If we're comparing for equality to zero, expose the fact that this is
// implemented as a ctlz/srl pair on ppc, so that the dag combiner can
// fold the new nodes.
if (SDValue V = lowerCmpEqZeroToCtlzSrl(Op, DAG))
return V;
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1))) {
// Leave comparisons against 0 and -1 alone for now, since they're usually
// optimized. FIXME: revisit this when we can custom lower all setcc
// optimizations.
if (C->isAllOnesValue() || C->isNullValue())
return SDValue();
}
// If we have an integer seteq/setne, turn it into a compare against zero
// by xor'ing the rhs with the lhs, which is faster than setting a
// condition register, reading it back out, and masking the correct bit. The
// normal approach here uses sub to do this instead of xor. Using xor exposes
// the result to other bit-twiddling opportunities.
EVT LHSVT = Op.getOperand(0).getValueType();
if (LHSVT.isInteger() && (CC == ISD::SETEQ || CC == ISD::SETNE)) {
EVT VT = Op.getValueType();
SDValue Sub = DAG.getNode(ISD::XOR, dl, LHSVT, Op.getOperand(0),
Op.getOperand(1));
return DAG.getSetCC(dl, VT, Sub, DAG.getConstant(0, dl, LHSVT), CC);
}
return SDValue();
}
SDValue PPCTargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
SDNode *Node = Op.getNode();
EVT VT = Node->getValueType(0);
EVT PtrVT = getPointerTy(DAG.getDataLayout());
SDValue InChain = Node->getOperand(0);
SDValue VAListPtr = Node->getOperand(1);
const Value *SV = cast<SrcValueSDNode>(Node->getOperand(2))->getValue();
SDLoc dl(Node);
assert(!Subtarget.isPPC64() && "LowerVAARG is PPC32 only");
// gpr_index
SDValue GprIndex = DAG.getExtLoad(ISD::ZEXTLOAD, dl, MVT::i32, InChain,
VAListPtr, MachinePointerInfo(SV), MVT::i8);
InChain = GprIndex.getValue(1);
if (VT == MVT::i64) {
// Check if GprIndex is even
SDValue GprAnd = DAG.getNode(ISD::AND, dl, MVT::i32, GprIndex,
DAG.getConstant(1, dl, MVT::i32));
SDValue CC64 = DAG.getSetCC(dl, MVT::i32, GprAnd,
DAG.getConstant(0, dl, MVT::i32), ISD::SETNE);
SDValue GprIndexPlusOne = DAG.getNode(ISD::ADD, dl, MVT::i32, GprIndex,
DAG.getConstant(1, dl, MVT::i32));
// Align GprIndex to be even if it isn't
GprIndex = DAG.getNode(ISD::SELECT, dl, MVT::i32, CC64, GprIndexPlusOne,
GprIndex);
}
// fpr index is 1 byte after gpr
SDValue FprPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
DAG.getConstant(1, dl, MVT::i32));
// fpr
SDValue FprIndex = DAG.getExtLoad(ISD::ZEXTLOAD, dl, MVT::i32, InChain,
FprPtr, MachinePointerInfo(SV), MVT::i8);
InChain = FprIndex.getValue(1);
SDValue RegSaveAreaPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
DAG.getConstant(8, dl, MVT::i32));
SDValue OverflowAreaPtr = DAG.getNode(ISD::ADD, dl, PtrVT, VAListPtr,
DAG.getConstant(4, dl, MVT::i32));
// areas
SDValue OverflowArea =
DAG.getLoad(MVT::i32, dl, InChain, OverflowAreaPtr, MachinePointerInfo());
InChain = OverflowArea.getValue(1);
SDValue RegSaveArea =
DAG.getLoad(MVT::i32, dl, InChain, RegSaveAreaPtr, MachinePointerInfo());
InChain = RegSaveArea.getValue(1);
// select overflow_area if index > 8
SDValue CC = DAG.getSetCC(dl, MVT::i32, VT.isInteger() ? GprIndex : FprIndex,
DAG.getConstant(8, dl, MVT::i32), ISD::SETLT);
// adjustment constant gpr_index * 4/8
SDValue RegConstant = DAG.getNode(ISD::MUL, dl, MVT::i32,
VT.isInteger() ? GprIndex : FprIndex,
DAG.getConstant(VT.isInteger() ? 4 : 8, dl,
MVT::i32));
// OurReg = RegSaveArea + RegConstant
SDValue OurReg = DAG.getNode(ISD::ADD, dl, PtrVT, RegSaveArea,
RegConstant);
// Floating types are 32 bytes into RegSaveArea
if (VT.isFloatingPoint())
OurReg = DAG.getNode(ISD::ADD, dl, PtrVT, OurReg,
DAG.getConstant(32, dl, MVT::i32));
// increase {f,g}pr_index by 1 (or 2 if VT is i64)
SDValue IndexPlus1 = DAG.getNode(ISD::ADD, dl, MVT::i32,
VT.isInteger() ? GprIndex : FprIndex,
DAG.getConstant(VT == MVT::i64 ? 2 : 1, dl,
MVT::i32));
InChain = DAG.getTruncStore(InChain, dl, IndexPlus1,
VT.isInteger() ? VAListPtr : FprPtr,
MachinePointerInfo(SV), MVT::i8);
// determine if we should load from reg_save_area or overflow_area
SDValue Result = DAG.getNode(ISD::SELECT, dl, PtrVT, CC, OurReg, OverflowArea);
// increase overflow_area by 4/8 if gpr/fpr > 8
SDValue OverflowAreaPlusN = DAG.getNode(ISD::ADD, dl, PtrVT, OverflowArea,
DAG.getConstant(VT.isInteger() ? 4 : 8,
dl, MVT::i32));
OverflowArea = DAG.getNode(ISD::SELECT, dl, MVT::i32, CC, OverflowArea,
OverflowAreaPlusN);
InChain = DAG.getTruncStore(InChain, dl, OverflowArea, OverflowAreaPtr,
MachinePointerInfo(), MVT::i32);
return DAG.getLoad(VT, dl, InChain, Result, MachinePointerInfo());
}
SDValue PPCTargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
assert(!Subtarget.isPPC64() && "LowerVACOPY is PPC32 only");
// We have to copy the entire va_list struct:
// 2*sizeof(char) + 2 Byte alignment + 2*sizeof(char*) = 12 Byte
return DAG.getMemcpy(Op.getOperand(0), Op, Op.getOperand(1), Op.getOperand(2),
DAG.getConstant(12, SDLoc(Op), MVT::i32), Align(8),
false, true, false, MachinePointerInfo(),
MachinePointerInfo());
}
SDValue PPCTargetLowering::LowerADJUST_TRAMPOLINE(SDValue Op,
SelectionDAG &DAG) const {
if (Subtarget.isAIXABI())
report_fatal_error("ADJUST_TRAMPOLINE operation is not supported on AIX.");
return Op.getOperand(0);
}
SDValue PPCTargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
SelectionDAG &DAG) const {
if (Subtarget.isAIXABI())
report_fatal_error("INIT_TRAMPOLINE operation is not supported on AIX.");
SDValue Chain = Op.getOperand(0);
SDValue Trmp = Op.getOperand(1); // trampoline
SDValue FPtr = Op.getOperand(2); // nested function
SDValue Nest = Op.getOperand(3); // 'nest' parameter value
SDLoc dl(Op);
EVT PtrVT = getPointerTy(DAG.getDataLayout());
bool isPPC64 = (PtrVT == MVT::i64);
Type *IntPtrTy = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = IntPtrTy;
Entry.Node = Trmp; Args.push_back(Entry);
// TrampSize == (isPPC64 ? 48 : 40);
Entry.Node = DAG.getConstant(isPPC64 ? 48 : 40, dl,
isPPC64 ? MVT::i64 : MVT::i32);
Args.push_back(Entry);
Entry.Node = FPtr; Args.push_back(Entry);
Entry.Node = Nest; Args.push_back(Entry);
// Lower to a call to __trampoline_setup(Trmp, TrampSize, FPtr, ctx_reg)
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain).setLibCallee(
CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol("__trampoline_setup", PtrVT), std::move(Args));
std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
return CallResult.second;
}
SDValue PPCTargetLowering::LowerVASTART(SDValue Op, SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
EVT PtrVT = getPointerTy(MF.getDataLayout());
SDLoc dl(Op);
if (Subtarget.isPPC64() || Subtarget.isAIXABI()) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
return DAG.getStore(Op.getOperand(0), dl, FR, Op.getOperand(1),
MachinePointerInfo(SV));
}
// For the 32-bit SVR4 ABI we follow the layout of the va_list struct.
// We suppose the given va_list is already allocated.
//
// typedef struct {
// char gpr; /* index into the array of 8 GPRs
// * stored in the register save area
// * gpr=0 corresponds to r3,
// * gpr=1 to r4, etc.
// */
// char fpr; /* index into the array of 8 FPRs
// * stored in the register save area
// * fpr=0 corresponds to f1,
// * fpr=1 to f2, etc.
// */
// char *overflow_arg_area;
// /* location on stack that holds
// * the next overflow argument
// */
// char *reg_save_area;
// /* where r3:r10 and f1:f8 (if saved)
// * are stored
// */
// } va_list[1];
SDValue ArgGPR = DAG.getConstant(FuncInfo->getVarArgsNumGPR(), dl, MVT::i32);
SDValue ArgFPR = DAG.getConstant(FuncInfo->getVarArgsNumFPR(), dl, MVT::i32);
SDValue StackOffsetFI = DAG.getFrameIndex(FuncInfo->getVarArgsStackOffset(),
PtrVT);
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
PtrVT);
uint64_t FrameOffset = PtrVT.getSizeInBits()/8;
SDValue ConstFrameOffset = DAG.getConstant(FrameOffset, dl, PtrVT);
uint64_t StackOffset = PtrVT.getSizeInBits()/8 - 1;
SDValue ConstStackOffset = DAG.getConstant(StackOffset, dl, PtrVT);
uint64_t FPROffset = 1;
SDValue ConstFPROffset = DAG.getConstant(FPROffset, dl, PtrVT);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// Store first byte : number of int regs
SDValue firstStore =
DAG.getTruncStore(Op.getOperand(0), dl, ArgGPR, Op.getOperand(1),
MachinePointerInfo(SV), MVT::i8);
uint64_t nextOffset = FPROffset;
SDValue nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, Op.getOperand(1),
ConstFPROffset);
// Store second byte : number of float regs
SDValue secondStore =
DAG.getTruncStore(firstStore, dl, ArgFPR, nextPtr,
MachinePointerInfo(SV, nextOffset), MVT::i8);
nextOffset += StackOffset;
nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, nextPtr, ConstStackOffset);
// Store second word : arguments given on stack
SDValue thirdStore = DAG.getStore(secondStore, dl, StackOffsetFI, nextPtr,
MachinePointerInfo(SV, nextOffset));
nextOffset += FrameOffset;
nextPtr = DAG.getNode(ISD::ADD, dl, PtrVT, nextPtr, ConstFrameOffset);
// Store third word : arguments given in registers
return DAG.getStore(thirdStore, dl, FR, nextPtr,
MachinePointerInfo(SV, nextOffset));
}
/// FPR - The set of FP registers that should be allocated for arguments
/// on Darwin and AIX.
static const MCPhysReg FPR[] = {PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5,
PPC::F6, PPC::F7, PPC::F8, PPC::F9, PPC::F10,
PPC::F11, PPC::F12, PPC::F13};
/// CalculateStackSlotSize - Calculates the size reserved for this argument on
/// the stack.
static unsigned CalculateStackSlotSize(EVT ArgVT, ISD::ArgFlagsTy Flags,
unsigned PtrByteSize) {
unsigned ArgSize = ArgVT.getStoreSize();
if (Flags.isByVal())
ArgSize = Flags.getByValSize();
// Round up to multiples of the pointer size, except for array members,
// which are always packed.
if (!Flags.isInConsecutiveRegs())
ArgSize = ((ArgSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
return ArgSize;
}
/// CalculateStackSlotAlignment - Calculates the alignment of this argument
/// on the stack.
static Align CalculateStackSlotAlignment(EVT ArgVT, EVT OrigVT,
ISD::ArgFlagsTy Flags,
unsigned PtrByteSize) {
Align Alignment(PtrByteSize);
// Altivec parameters are padded to a 16 byte boundary.
if (ArgVT == MVT::v4f32 || ArgVT == MVT::v4i32 ||
ArgVT == MVT::v8i16 || ArgVT == MVT::v16i8 ||
ArgVT == MVT::v2f64 || ArgVT == MVT::v2i64 ||
ArgVT == MVT::v1i128 || ArgVT == MVT::f128)
Alignment = Align(16);
// ByVal parameters are aligned as requested.
if (Flags.isByVal()) {
auto BVAlign = Flags.getNonZeroByValAlign();
if (BVAlign > PtrByteSize) {
if (BVAlign.value() % PtrByteSize != 0)
llvm_unreachable(
"ByVal alignment is not a multiple of the pointer size");
Alignment = BVAlign;
}
}
// Array members are always packed to their original alignment.
if (Flags.isInConsecutiveRegs()) {
// If the array member was split into multiple registers, the first
// needs to be aligned to the size of the full type. (Except for
// ppcf128, which is only aligned as its f64 components.)
if (Flags.isSplit() && OrigVT != MVT::ppcf128)
Alignment = Align(OrigVT.getStoreSize());
else
Alignment = Align(ArgVT.getStoreSize());
}
return Alignment;
}
/// CalculateStackSlotUsed - Return whether this argument will use its
/// stack slot (instead of being passed in registers). ArgOffset,
/// AvailableFPRs, and AvailableVRs must hold the current argument
/// position, and will be updated to account for this argument.
static bool CalculateStackSlotUsed(EVT ArgVT, EVT OrigVT, ISD::ArgFlagsTy Flags,
unsigned PtrByteSize, unsigned LinkageSize,
unsigned ParamAreaSize, unsigned &ArgOffset,
unsigned &AvailableFPRs,
unsigned &AvailableVRs) {
bool UseMemory = false;
// Respect alignment of argument on the stack.
Align Alignment =
CalculateStackSlotAlignment(ArgVT, OrigVT, Flags, PtrByteSize);
ArgOffset = alignTo(ArgOffset, Alignment);
// If there's no space left in the argument save area, we must
// use memory (this check also catches zero-sized arguments).
if (ArgOffset >= LinkageSize + ParamAreaSize)
UseMemory = true;
// Allocate argument on the stack.
ArgOffset += CalculateStackSlotSize(ArgVT, Flags, PtrByteSize);
if (Flags.isInConsecutiveRegsLast())
ArgOffset = ((ArgOffset + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
// If we overran the argument save area, we must use memory
// (this check catches arguments passed partially in memory)
if (ArgOffset > LinkageSize + ParamAreaSize)
UseMemory = true;
// However, if the argument is actually passed in an FPR or a VR,
// we don't use memory after all.
if (!Flags.isByVal()) {
if (ArgVT == MVT::f32 || ArgVT == MVT::f64)
if (AvailableFPRs > 0) {
--AvailableFPRs;
return false;
}
if (ArgVT == MVT::v4f32 || ArgVT == MVT::v4i32 ||
ArgVT == MVT::v8i16 || ArgVT == MVT::v16i8 ||
ArgVT == MVT::v2f64 || ArgVT == MVT::v2i64 ||
ArgVT == MVT::v1i128 || ArgVT == MVT::f128)
if (AvailableVRs > 0) {
--AvailableVRs;
return false;
}
}
return UseMemory;
}
/// EnsureStackAlignment - Round stack frame size up from NumBytes to
/// ensure minimum alignment required for target.
static unsigned EnsureStackAlignment(const PPCFrameLowering *Lowering,
unsigned NumBytes) {
return alignTo(NumBytes, Lowering->getStackAlign());
}
SDValue PPCTargetLowering::LowerFormalArguments(
SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
if (Subtarget.isAIXABI())
return LowerFormalArguments_AIX(Chain, CallConv, isVarArg, Ins, dl, DAG,
InVals);
if (Subtarget.is64BitELFABI())
return LowerFormalArguments_64SVR4(Chain, CallConv, isVarArg, Ins, dl, DAG,
InVals);
assert(Subtarget.is32BitELFABI());
return LowerFormalArguments_32SVR4(Chain, CallConv, isVarArg, Ins, dl, DAG,
InVals);
}
SDValue PPCTargetLowering::LowerFormalArguments_32SVR4(
SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
// 32-bit SVR4 ABI Stack Frame Layout:
// +-----------------------------------+
// +--> | Back chain |
// | +-----------------------------------+
// | | Floating-point register save area |
// | +-----------------------------------+
// | | General register save area |
// | +-----------------------------------+
// | | CR save word |
// | +-----------------------------------+
// | | VRSAVE save word |
// | +-----------------------------------+
// | | Alignment padding |
// | +-----------------------------------+
// | | Vector register save area |
// | +-----------------------------------+
// | | Local variable space |
// | +-----------------------------------+
// | | Parameter list area |
// | +-----------------------------------+
// | | LR save word |
// | +-----------------------------------+
// SP--> +--- | Back chain |
// +-----------------------------------+
//
// Specifications:
// System V Application Binary Interface PowerPC Processor Supplement
// AltiVec Technology Programming Interface Manual
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
EVT PtrVT = getPointerTy(MF.getDataLayout());
// Potential tail calls could cause overwriting of argument stack slots.
bool isImmutable = !(getTargetMachine().Options.GuaranteedTailCallOpt &&
(CallConv == CallingConv::Fast));
const Align PtrAlign(4);
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
PPCCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
*DAG.getContext());
// Reserve space for the linkage area on the stack.
unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
CCInfo.AllocateStack(LinkageSize, PtrAlign);
if (useSoftFloat())
CCInfo.PreAnalyzeFormalArguments(Ins);
CCInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4);
CCInfo.clearWasPPCF128();
for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
CCValAssign &VA = ArgLocs[i];
// Arguments stored in registers.
if (VA.isRegLoc()) {
const TargetRegisterClass *RC;
EVT ValVT = VA.getValVT();
switch (ValVT.getSimpleVT().SimpleTy) {
default:
llvm_unreachable("ValVT not supported by formal arguments Lowering");
case MVT::i1:
case MVT::i32:
RC = &PPC::GPRCRegClass;
break;
case MVT::f32:
if (Subtarget.hasP8Vector())
RC = &PPC::VSSRCRegClass;
else if (Subtarget.hasSPE())
RC = &PPC::GPRCRegClass;
else
RC = &PPC::F4RCRegClass;
break;
case MVT::f64:
if (Subtarget.hasVSX())
RC = &PPC::VSFRCRegClass;
else if (Subtarget.hasSPE())
// SPE passes doubles in GPR pairs.
RC = &PPC::GPRCRegClass;
else
RC = &PPC::F8RCRegClass;
break;
case MVT::v16i8:
case MVT::v8i16:
case MVT::v4i32:
RC = &PPC::VRRCRegClass;
break;
case MVT::v4f32:
RC = &PPC::VRRCRegClass;
break;
case MVT::v2f64:
case MVT::v2i64:
RC = &PPC::VRRCRegClass;
break;
}
SDValue ArgValue;
// Transform the arguments stored in physical registers into
// virtual ones.
if (VA.getLocVT() == MVT::f64 && Subtarget.hasSPE()) {
assert(i + 1 < e && "No second half of double precision argument");
unsigned RegLo = MF.addLiveIn(VA.getLocReg(), RC);
unsigned RegHi = MF.addLiveIn(ArgLocs[++i].getLocReg(), RC);
SDValue ArgValueLo = DAG.getCopyFromReg(Chain, dl, RegLo, MVT::i32);
SDValue ArgValueHi = DAG.getCopyFromReg(Chain, dl, RegHi, MVT::i32);
if (!Subtarget.isLittleEndian())
std::swap (ArgValueLo, ArgValueHi);
ArgValue = DAG.getNode(PPCISD::BUILD_SPE64, dl, MVT::f64, ArgValueLo,
ArgValueHi);
} else {
unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
ArgValue = DAG.getCopyFromReg(Chain, dl, Reg,
ValVT == MVT::i1 ? MVT::i32 : ValVT);
if (ValVT == MVT::i1)
ArgValue = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, ArgValue);
}
InVals.push_back(ArgValue);
} else {
// Argument stored in memory.
assert(VA.isMemLoc());
// Get the extended size of the argument type in stack
unsigned ArgSize = VA.getLocVT().getStoreSize();
// Get the actual size of the argument type
unsigned ObjSize = VA.getValVT().getStoreSize();
unsigned ArgOffset = VA.getLocMemOffset();
// Stack objects in PPC32 are right justified.
ArgOffset += ArgSize - ObjSize;
int FI = MFI.CreateFixedObject(ArgSize, ArgOffset, isImmutable);
// Create load nodes to retrieve arguments from the stack.
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
InVals.push_back(
DAG.getLoad(VA.getValVT(), dl, Chain, FIN, MachinePointerInfo()));
}
}
// Assign locations to all of the incoming aggregate by value arguments.
// Aggregates passed by value are stored in the local variable space of the
// caller's stack frame, right above the parameter list area.
SmallVector<CCValAssign, 16> ByValArgLocs;
CCState CCByValInfo(CallConv, isVarArg, DAG.getMachineFunction(),
ByValArgLocs, *DAG.getContext());
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrAlign);
CCByValInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4_ByVal);
// Area that is at least reserved in the caller of this function.
unsigned MinReservedArea = CCByValInfo.getNextStackOffset();
MinReservedArea = std::max(MinReservedArea, LinkageSize);
// Set the size that is at least reserved in caller of this function. Tail
// call optimized function's reserved stack space needs to be aligned so that
// taking the difference between two stack areas will result in an aligned
// stack.
MinReservedArea =
EnsureStackAlignment(Subtarget.getFrameLowering(), MinReservedArea);
FuncInfo->setMinReservedArea(MinReservedArea);
SmallVector<SDValue, 8> MemOps;
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
if (isVarArg) {
static const MCPhysReg GPArgRegs[] = {
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
};
const unsigned NumGPArgRegs = array_lengthof(GPArgRegs);
static const MCPhysReg FPArgRegs[] = {
PPC::F1, PPC::F2, PPC::F3, PPC::F4, PPC::F5, PPC::F6, PPC::F7,
PPC::F8
};
unsigned NumFPArgRegs = array_lengthof(FPArgRegs);
if (useSoftFloat() || hasSPE())
NumFPArgRegs = 0;
FuncInfo->setVarArgsNumGPR(CCInfo.getFirstUnallocated(GPArgRegs));
FuncInfo->setVarArgsNumFPR(CCInfo.getFirstUnallocated(FPArgRegs));
// Make room for NumGPArgRegs and NumFPArgRegs.
int Depth = NumGPArgRegs * PtrVT.getSizeInBits()/8 +
NumFPArgRegs * MVT(MVT::f64).getSizeInBits()/8;
FuncInfo->setVarArgsStackOffset(
MFI.CreateFixedObject(PtrVT.getSizeInBits()/8,
CCInfo.getNextStackOffset(), true));
FuncInfo->setVarArgsFrameIndex(
MFI.CreateStackObject(Depth, Align(8), false));
SDValue FIN = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
// The fixed integer arguments of a variadic function are stored to the
// VarArgsFrameIndex on the stack so that they may be loaded by
// dereferencing the result of va_next.
for (unsigned GPRIndex = 0; GPRIndex != NumGPArgRegs; ++GPRIndex) {
// Get an existing live-in vreg, or add a new one.
unsigned VReg = MF.getRegInfo().getLiveInVirtReg(GPArgRegs[GPRIndex]);
if (!VReg)
VReg = MF.addLiveIn(GPArgRegs[GPRIndex], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN, MachinePointerInfo());
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDValue PtrOff = DAG.getConstant(PtrVT.getSizeInBits()/8, dl, PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
// FIXME 32-bit SVR4: We only need to save FP argument registers if CR bit 6
// is set.
// The double arguments are stored to the VarArgsFrameIndex
// on the stack.
for (unsigned FPRIndex = 0; FPRIndex != NumFPArgRegs; ++FPRIndex) {
// Get an existing live-in vreg, or add a new one.
unsigned VReg = MF.getRegInfo().getLiveInVirtReg(FPArgRegs[FPRIndex]);
if (!VReg)
VReg = MF.addLiveIn(FPArgRegs[FPRIndex], &PPC::F8RCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::f64);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN, MachinePointerInfo());
MemOps.push_back(Store);
// Increment the address by eight for the next argument to store
SDValue PtrOff = DAG.getConstant(MVT(MVT::f64).getSizeInBits()/8, dl,
PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
}
if (!MemOps.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
return Chain;
}
// PPC64 passes i8, i16, and i32 values in i64 registers. Promote
// value to MVT::i64 and then truncate to the correct register size.
SDValue PPCTargetLowering::extendArgForPPC64(ISD::ArgFlagsTy Flags,
EVT ObjectVT, SelectionDAG &DAG,
SDValue ArgVal,
const SDLoc &dl) const {
if (Flags.isSExt())
ArgVal = DAG.getNode(ISD::AssertSext, dl, MVT::i64, ArgVal,
DAG.getValueType(ObjectVT));
else if (Flags.isZExt())
ArgVal = DAG.getNode(ISD::AssertZext, dl, MVT::i64, ArgVal,
DAG.getValueType(ObjectVT));
return DAG.getNode(ISD::TRUNCATE, dl, ObjectVT, ArgVal);
}
SDValue PPCTargetLowering::LowerFormalArguments_64SVR4(
SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
// TODO: add description of PPC stack frame format, or at least some docs.
//
bool isELFv2ABI = Subtarget.isELFv2ABI();
bool isLittleEndian = Subtarget.isLittleEndian();
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
assert(!(CallConv == CallingConv::Fast && isVarArg) &&
"fastcc not supported on varargs functions");
EVT PtrVT = getPointerTy(MF.getDataLayout());
// Potential tail calls could cause overwriting of argument stack slots.
bool isImmutable = !(getTargetMachine().Options.GuaranteedTailCallOpt &&
(CallConv == CallingConv::Fast));
unsigned PtrByteSize = 8;
unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
static const MCPhysReg GPR[] = {
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const MCPhysReg VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned Num_GPR_Regs = array_lengthof(GPR);
const unsigned Num_FPR_Regs = useSoftFloat() ? 0 : 13;
const unsigned Num_VR_Regs = array_lengthof(VR);
// Do a first pass over the arguments to determine whether the ABI
// guarantees that our caller has allocated the parameter save area
// on its stack frame. In the ELFv1 ABI, this is always the case;
// in the ELFv2 ABI, it is true if this is a vararg function or if
// any parameter is located in a stack slot.
bool HasParameterArea = !isELFv2ABI || isVarArg;
unsigned ParamAreaSize = Num_GPR_Regs * PtrByteSize;
unsigned NumBytes = LinkageSize;
unsigned AvailableFPRs = Num_FPR_Regs;
unsigned AvailableVRs = Num_VR_Regs;
for (unsigned i = 0, e = Ins.size(); i != e; ++i) {
if (Ins[i].Flags.isNest())
continue;
if (CalculateStackSlotUsed(Ins[i].VT, Ins[i].ArgVT, Ins[i].Flags,
PtrByteSize, LinkageSize, ParamAreaSize,
NumBytes, AvailableFPRs, AvailableVRs))
HasParameterArea = true;
}
// Add DAG nodes to load the arguments or copy them out of registers. On
// entry to a function on PPC, the arguments start after the linkage area,
// although the first ones are often in registers.
unsigned ArgOffset = LinkageSize;
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
SmallVector<SDValue, 8> MemOps;
Function::const_arg_iterator FuncArg = MF.getFunction().arg_begin();
unsigned CurArgIdx = 0;
for (unsigned ArgNo = 0, e = Ins.size(); ArgNo != e; ++ArgNo) {
SDValue ArgVal;
bool needsLoad = false;
EVT ObjectVT = Ins[ArgNo].VT;
EVT OrigVT = Ins[ArgNo].ArgVT;
unsigned ObjSize = ObjectVT.getStoreSize();
unsigned ArgSize = ObjSize;
ISD::ArgFlagsTy Flags = Ins[ArgNo].Flags;
if (Ins[ArgNo].isOrigArg()) {
std::advance(FuncArg, Ins[ArgNo].getOrigArgIndex() - CurArgIdx);
CurArgIdx = Ins[ArgNo].getOrigArgIndex();
}
// We re-align the argument offset for each argument, except when using the
// fast calling convention, when we need to make sure we do that only when
// we'll actually use a stack slot.
unsigned CurArgOffset;
Align Alignment;
auto ComputeArgOffset = [&]() {
/* Respect alignment of argument on the stack. */
Alignment =
CalculateStackSlotAlignment(ObjectVT, OrigVT, Flags, PtrByteSize);
ArgOffset = alignTo(ArgOffset, Alignment);
CurArgOffset = ArgOffset;
};
if (CallConv != CallingConv::Fast) {
ComputeArgOffset();
/* Compute GPR index associated with argument offset. */
GPR_idx = (ArgOffset - LinkageSize) / PtrByteSize;
GPR_idx = std::min(GPR_idx, Num_GPR_Regs);
}
// FIXME the codegen can be much improved in some cases.
// We do not have to keep everything in memory.
if (Flags.isByVal()) {
assert(Ins[ArgNo].isOrigArg() && "Byval arguments cannot be implicit");
if (CallConv == CallingConv::Fast)
ComputeArgOffset();
// ObjSize is the true size, ArgSize rounded up to multiple of registers.
ObjSize = Flags.getByValSize();
ArgSize = ((ObjSize + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
// Empty aggregate parameters do not take up registers. Examples:
// struct { } a;
// union { } b;
// int c[0];
// etc. However, we have to provide a place-holder in InVals, so
// pretend we have an 8-byte item at the current address for that
// purpose.
if (!ObjSize) {
int FI = MFI.CreateFixedObject(PtrByteSize, ArgOffset, true);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
InVals.push_back(FIN);
continue;
}
// Create a stack object covering all stack doublewords occupied
// by the argument. If the argument is (fully or partially) on
// the stack, or if the argument is fully in registers but the
// caller has allocated the parameter save anyway, we can refer
// directly to the caller's stack frame. Otherwise, create a
// local copy in our own frame.
int FI;
if (HasParameterArea ||
ArgSize + ArgOffset > LinkageSize + Num_GPR_Regs * PtrByteSize)
FI = MFI.CreateFixedObject(ArgSize, ArgOffset, false, true);
else
FI = MFI.CreateStackObject(ArgSize, Alignment, false);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
// Handle aggregates smaller than 8 bytes.
if (ObjSize < PtrByteSize) {
// The value of the object is its address, which differs from the
// address of the enclosing doubleword on big-endian systems.
SDValue Arg = FIN;
if (!isLittleEndian) {
SDValue ArgOff = DAG.getConstant(PtrByteSize - ObjSize, dl, PtrVT);
Arg = DAG.getNode(ISD::ADD, dl, ArgOff.getValueType(), Arg, ArgOff);
}
InVals.push_back(Arg);
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx++], &PPC::G8RCRegClass);
FuncInfo->addLiveInAttr(VReg, Flags);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store;
if (ObjSize==1 || ObjSize==2 || ObjSize==4) {
EVT ObjType = (ObjSize == 1 ? MVT::i8 :
(ObjSize == 2 ? MVT::i16 : MVT::i32));
Store = DAG.getTruncStore(Val.getValue(1), dl, Val, Arg,
MachinePointerInfo(&*FuncArg), ObjType);
} else {
// For sizes that don't fit a truncating store (3, 5, 6, 7),
// store the whole register as-is to the parameter save area
// slot.
Store = DAG.getStore(Val.getValue(1), dl, Val, FIN,
MachinePointerInfo(&*FuncArg));
}
MemOps.push_back(Store);
}
// Whether we copied from a register or not, advance the offset
// into the parameter save area by a full doubleword.
ArgOffset += PtrByteSize;
continue;
}
// The value of the object is its address, which is the address of
// its first stack doubleword.
InVals.push_back(FIN);
// Store whatever pieces of the object are in registers to memory.
for (unsigned j = 0; j < ArgSize; j += PtrByteSize) {
if (GPR_idx == Num_GPR_Regs)
break;
unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::G8RCRegClass);
FuncInfo->addLiveInAttr(VReg, Flags);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Addr = FIN;
if (j) {
SDValue Off = DAG.getConstant(j, dl, PtrVT);
Addr = DAG.getNode(ISD::ADD, dl, Off.getValueType(), Addr, Off);
}
SDValue Store = DAG.getStore(Val.getValue(1), dl, Val, Addr,
MachinePointerInfo(&*FuncArg, j));
MemOps.push_back(Store);
++GPR_idx;
}
ArgOffset += ArgSize;
continue;
}
switch (ObjectVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unhandled argument type!");
case MVT::i1:
case MVT::i32:
case MVT::i64:
if (Flags.isNest()) {
// The 'nest' parameter, if any, is passed in R11.
unsigned VReg = MF.addLiveIn(PPC::X11, &PPC::G8RCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i64);
if (ObjectVT == MVT::i32 || ObjectVT == MVT::i1)
ArgVal = extendArgForPPC64(Flags, ObjectVT, DAG, ArgVal, dl);
break;
}
// These can be scalar arguments or elements of an integer array type
// passed directly. Clang may use those instead of "byval" aggregate
// types to avoid forcing arguments to memory unnecessarily.
if (GPR_idx != Num_GPR_Regs) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx++], &PPC::G8RCRegClass);
FuncInfo->addLiveInAttr(VReg, Flags);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i64);
if (ObjectVT == MVT::i32 || ObjectVT == MVT::i1)
// PPC64 passes i8, i16, and i32 values in i64 registers. Promote
// value to MVT::i64 and then truncate to the correct register size.
ArgVal = extendArgForPPC64(Flags, ObjectVT, DAG, ArgVal, dl);
} else {
if (CallConv == CallingConv::Fast)
ComputeArgOffset();
needsLoad = true;
ArgSize = PtrByteSize;
}
if (CallConv != CallingConv::Fast || needsLoad)
ArgOffset += 8;
break;
case MVT::f32:
case MVT::f64:
// These can be scalar arguments or elements of a float array type
// passed directly. The latter are used to implement ELFv2 homogenous
// float aggregates.
if (FPR_idx != Num_FPR_Regs) {
unsigned VReg;
if (ObjectVT == MVT::f32)
VReg = MF.addLiveIn(FPR[FPR_idx],
Subtarget.hasP8Vector()
? &PPC::VSSRCRegClass
: &PPC::F4RCRegClass);
else
VReg = MF.addLiveIn(FPR[FPR_idx], Subtarget.hasVSX()
? &PPC::VSFRCRegClass
: &PPC::F8RCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, ObjectVT);
++FPR_idx;
} else if (GPR_idx != Num_GPR_Regs && CallConv != CallingConv::Fast) {
// FIXME: We may want to re-enable this for CallingConv::Fast on the P8
// once we support fp <-> gpr moves.
// This can only ever happen in the presence of f32 array types,
// since otherwise we never run out of FPRs before running out
// of GPRs.
unsigned VReg = MF.addLiveIn(GPR[GPR_idx++], &PPC::G8RCRegClass);
FuncInfo->addLiveInAttr(VReg, Flags);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i64);
if (ObjectVT == MVT::f32) {
if ((ArgOffset % PtrByteSize) == (isLittleEndian ? 4 : 0))
ArgVal = DAG.getNode(ISD::SRL, dl, MVT::i64, ArgVal,
DAG.getConstant(32, dl, MVT::i32));
ArgVal = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, ArgVal);
}
ArgVal = DAG.getNode(ISD::BITCAST, dl, ObjectVT, ArgVal);
} else {
if (CallConv == CallingConv::Fast)
ComputeArgOffset();
needsLoad = true;
}
// When passing an array of floats, the array occupies consecutive
// space in the argument area; only round up to the next doubleword
// at the end of the array. Otherwise, each float takes 8 bytes.
if (CallConv != CallingConv::Fast || needsLoad) {
ArgSize = Flags.isInConsecutiveRegs() ? ObjSize : PtrByteSize;
ArgOffset += ArgSize;
if (Flags.isInConsecutiveRegsLast())
ArgOffset = ((ArgOffset + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
}
break;
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
case MVT::v2f64:
case MVT::v2i64:
case MVT::v1i128:
case MVT::f128:
// These can be scalar arguments or elements of a vector array type
// passed directly. The latter are used to implement ELFv2 homogenous
// vector aggregates.
if (VR_idx != Num_VR_Regs) {
unsigned VReg = MF.addLiveIn(VR[VR_idx], &PPC::VRRCRegClass);
ArgVal = DAG.getCopyFromReg(Chain, dl, VReg, ObjectVT);
++VR_idx;
} else {
if (CallConv == CallingConv::Fast)
ComputeArgOffset();
needsLoad = true;
}
if (CallConv != CallingConv::Fast || needsLoad)
ArgOffset += 16;
break;
}
// We need to load the argument to a virtual register if we determined
// above that we ran out of physical registers of the appropriate type.
if (needsLoad) {
if (ObjSize < ArgSize && !isLittleEndian)
CurArgOffset += ArgSize - ObjSize;
int FI = MFI.CreateFixedObject(ObjSize, CurArgOffset, isImmutable);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
ArgVal = DAG.getLoad(ObjectVT, dl, Chain, FIN, MachinePointerInfo());
}
InVals.push_back(ArgVal);
}
// Area that is at least reserved in the caller of this function.
unsigned MinReservedArea;
if (HasParameterArea)
MinReservedArea = std::max(ArgOffset, LinkageSize + 8 * PtrByteSize);
else
MinReservedArea = LinkageSize;
// Set the size that is at least reserved in caller of this function. Tail
// call optimized functions' reserved stack space needs to be aligned so that
// taking the difference between two stack areas will result in an aligned
// stack.
MinReservedArea =
EnsureStackAlignment(Subtarget.getFrameLowering(), MinReservedArea);
FuncInfo->setMinReservedArea(MinReservedArea);
// If the function takes variable number of arguments, make a frame index for
// the start of the first vararg value... for expansion of llvm.va_start.
// On ELFv2ABI spec, it writes:
// C programs that are intended to be *portable* across different compilers
// and architectures must use the header file <stdarg.h> to deal with variable
// argument lists.
if (isVarArg && MFI.hasVAStart()) {
int Depth = ArgOffset;
FuncInfo->setVarArgsFrameIndex(
MFI.CreateFixedObject(PtrByteSize, Depth, true));
SDValue FIN = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
// If this function is vararg, store any remaining integer argument regs
// to their spots on the stack so that they may be loaded by dereferencing
// the result of va_next.
for (GPR_idx = (ArgOffset - LinkageSize) / PtrByteSize;
GPR_idx < Num_GPR_Regs; ++GPR_idx) {
unsigned VReg = MF.addLiveIn(GPR[GPR_idx], &PPC::G8RCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN, MachinePointerInfo());
MemOps.push_back(Store);
// Increment the address by four for the next argument to store
SDValue PtrOff = DAG.getConstant(PtrByteSize, dl, PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
}
if (!MemOps.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
return Chain;
}
/// CalculateTailCallSPDiff - Get the amount the stack pointer has to be
/// adjusted to accommodate the arguments for the tailcall.
static int CalculateTailCallSPDiff(SelectionDAG& DAG, bool isTailCall,
unsigned ParamSize) {
if (!isTailCall) return 0;
PPCFunctionInfo *FI = DAG.getMachineFunction().getInfo<PPCFunctionInfo>();
unsigned CallerMinReservedArea = FI->getMinReservedArea();
int SPDiff = (int)CallerMinReservedArea - (int)ParamSize;
// Remember only if the new adjustment is bigger.
if (SPDiff < FI->getTailCallSPDelta())
FI->setTailCallSPDelta(SPDiff);
return SPDiff;
}
static bool isFunctionGlobalAddress(SDValue Callee);
static bool callsShareTOCBase(const Function *Caller, SDValue Callee,
const TargetMachine &TM) {
// It does not make sense to call callsShareTOCBase() with a caller that
// is PC Relative since PC Relative callers do not have a TOC.
#ifndef NDEBUG
const PPCSubtarget *STICaller = &TM.getSubtarget<PPCSubtarget>(*Caller);
assert(!STICaller->isUsingPCRelativeCalls() &&
"PC Relative callers do not have a TOC and cannot share a TOC Base");
#endif
// Callee is either a GlobalAddress or an ExternalSymbol. ExternalSymbols
// don't have enough information to determine if the caller and callee share
// the same TOC base, so we have to pessimistically assume they don't for
// correctness.
GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
if (!G)
return false;
const GlobalValue *GV = G->getGlobal();
// If the callee is preemptable, then the static linker will use a plt-stub
// which saves the toc to the stack, and needs a nop after the call
// instruction to convert to a toc-restore.
if (!TM.shouldAssumeDSOLocal(*Caller->getParent(), GV))
return false;
// Functions with PC Relative enabled may clobber the TOC in the same DSO.
// We may need a TOC restore in the situation where the caller requires a
// valid TOC but the callee is PC Relative and does not.
const Function *F = dyn_cast<Function>(GV);
const GlobalAlias *Alias = dyn_cast<GlobalAlias>(GV);
// If we have an Alias we can try to get the function from there.
if (Alias) {
const GlobalObject *GlobalObj = Alias->getBaseObject();
F = dyn_cast<Function>(GlobalObj);
}
// If we still have no valid function pointer we do not have enough
// information to determine if the callee uses PC Relative calls so we must
// assume that it does.
if (!F)
return false;
// If the callee uses PC Relative we cannot guarantee that the callee won't
// clobber the TOC of the caller and so we must assume that the two
// functions do not share a TOC base.
const PPCSubtarget *STICallee = &TM.getSubtarget<PPCSubtarget>(*F);
if (STICallee->isUsingPCRelativeCalls())
return false;
// If the GV is not a strong definition then we need to assume it can be
// replaced by another function at link time. The function that replaces
// it may not share the same TOC as the caller since the callee may be
// replaced by a PC Relative version of the same function.
if (!GV->isStrongDefinitionForLinker())
return false;
// The medium and large code models are expected to provide a sufficiently
// large TOC to provide all data addressing needs of a module with a
// single TOC.
if (CodeModel::Medium == TM.getCodeModel() ||
CodeModel::Large == TM.getCodeModel())
return true;
// Any explicitly-specified sections and section prefixes must also match.
// Also, if we're using -ffunction-sections, then each function is always in
// a different section (the same is true for COMDAT functions).
if (TM.getFunctionSections() || GV->hasComdat() || Caller->hasComdat() ||
GV->getSection() != Caller->getSection())
return false;
if (const auto *F = dyn_cast<Function>(GV)) {
if (F->getSectionPrefix() != Caller->getSectionPrefix())
return false;
}
return true;
}
static bool
needStackSlotPassParameters(const PPCSubtarget &Subtarget,
const SmallVectorImpl<ISD::OutputArg> &Outs) {
assert(Subtarget.is64BitELFABI());
const unsigned PtrByteSize = 8;
const unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
static const MCPhysReg GPR[] = {
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const MCPhysReg VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned NumGPRs = array_lengthof(GPR);
const unsigned NumFPRs = 13;
const unsigned NumVRs = array_lengthof(VR);
const unsigned ParamAreaSize = NumGPRs * PtrByteSize;
unsigned NumBytes = LinkageSize;
unsigned AvailableFPRs = NumFPRs;
unsigned AvailableVRs = NumVRs;
for (const ISD::OutputArg& Param : Outs) {
if (Param.Flags.isNest()) continue;
if (CalculateStackSlotUsed(Param.VT, Param.ArgVT, Param.Flags, PtrByteSize,
LinkageSize, ParamAreaSize, NumBytes,
AvailableFPRs, AvailableVRs))
return true;
}
return false;
}
static bool hasSameArgumentList(const Function *CallerFn, const CallBase &CB) {
if (CB.arg_size() != CallerFn->arg_size())
return false;
auto CalleeArgIter = CB.arg_begin();
auto CalleeArgEnd = CB.arg_end();
Function::const_arg_iterator CallerArgIter = CallerFn->arg_begin();
for (; CalleeArgIter != CalleeArgEnd; ++CalleeArgIter, ++CallerArgIter) {
const Value* CalleeArg = *CalleeArgIter;
const Value* CallerArg = &(*CallerArgIter);
if (CalleeArg == CallerArg)
continue;
// e.g. @caller([4 x i64] %a, [4 x i64] %b) {
// tail call @callee([4 x i64] undef, [4 x i64] %b)
// }
// 1st argument of callee is undef and has the same type as caller.
if (CalleeArg->getType() == CallerArg->getType() &&
isa<UndefValue>(CalleeArg))
continue;
return false;
}
return true;
}
// Returns true if TCO is possible between the callers and callees
// calling conventions.
static bool
areCallingConvEligibleForTCO_64SVR4(CallingConv::ID CallerCC,
CallingConv::ID CalleeCC) {
// Tail calls are possible with fastcc and ccc.
auto isTailCallableCC = [] (CallingConv::ID CC){
return CC == CallingConv::C || CC == CallingConv::Fast;
};
if (!isTailCallableCC(CallerCC) || !isTailCallableCC(CalleeCC))
return false;
// We can safely tail call both fastcc and ccc callees from a c calling
// convention caller. If the caller is fastcc, we may have less stack space
// than a non-fastcc caller with the same signature so disable tail-calls in
// that case.
return CallerCC == CallingConv::C || CallerCC == CalleeCC;
}
bool PPCTargetLowering::IsEligibleForTailCallOptimization_64SVR4(
SDValue Callee, CallingConv::ID CalleeCC, const CallBase *CB, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<ISD::InputArg> &Ins, SelectionDAG &DAG) const {
bool TailCallOpt = getTargetMachine().Options.GuaranteedTailCallOpt;
if (DisableSCO && !TailCallOpt) return false;
// Variadic argument functions are not supported.
if (isVarArg) return false;
auto &Caller = DAG.getMachineFunction().getFunction();
// Check that the calling conventions are compatible for tco.
if (!areCallingConvEligibleForTCO_64SVR4(Caller.getCallingConv(), CalleeCC))
return false;
// Caller contains any byval parameter is not supported.
if (any_of(Ins, [](const ISD::InputArg &IA) { return IA.Flags.isByVal(); }))
return false;
// Callee contains any byval parameter is not supported, too.
// Note: This is a quick work around, because in some cases, e.g.
// caller's stack size > callee's stack size, we are still able to apply
// sibling call optimization. For example, gcc is able to do SCO for caller1
// in the following example, but not for caller2.
// struct test {
// long int a;
// char ary[56];
// } gTest;
// __attribute__((noinline)) int callee(struct test v, struct test *b) {
// b->a = v.a;
// return 0;
// }
// void caller1(struct test a, struct test c, struct test *b) {
// callee(gTest, b); }
// void caller2(struct test *b) { callee(gTest, b); }
if (any_of(Outs, [](const ISD::OutputArg& OA) { return OA.Flags.isByVal(); }))
return false;
// If callee and caller use different calling conventions, we cannot pass
// parameters on stack since offsets for the parameter area may be different.
if (Caller.getCallingConv() != CalleeCC &&
needStackSlotPassParameters(Subtarget, Outs))
return false;
// All variants of 64-bit ELF ABIs without PC-Relative addressing require that
// the caller and callee share the same TOC for TCO/SCO. If the caller and
// callee potentially have different TOC bases then we cannot tail call since
// we need to restore the TOC pointer after the call.
// ref: https://bugzilla.mozilla.org/show_bug.cgi?id=973977
// We cannot guarantee this for indirect calls or calls to external functions.
// When PC-Relative addressing is used, the concept of the TOC is no longer
// applicable so this check is not required.
// Check first for indirect calls.
if (!Subtarget.isUsingPCRelativeCalls() &&
!isFunctionGlobalAddress(Callee) && !isa<ExternalSymbolSDNode>(Callee))
return false;
// Check if we share the TOC base.
if (!Subtarget.isUsingPCRelativeCalls() &&
!callsShareTOCBase(&Caller, Callee, getTargetMachine()))
return false;
// TCO allows altering callee ABI, so we don't have to check further.
if (CalleeCC == CallingConv::Fast && TailCallOpt)
return true;
if (DisableSCO) return false;
// If callee use the same argument list that caller is using, then we can
// apply SCO on this case. If it is not, then we need to check if callee needs
// stack for passing arguments.
// PC Relative tail calls may not have a CallBase.
// If there is no CallBase we cannot verify if we have the same argument
// list so assume that we don't have the same argument list.
if (CB && !hasSameArgumentList(&Caller, *CB) &&
needStackSlotPassParameters(Subtarget, Outs))
return false;
else if (!CB && needStackSlotPassParameters(Subtarget, Outs))
return false;
return true;
}
/// IsEligibleForTailCallOptimization - Check whether the call is eligible
/// for tail call optimization. Targets which want to do tail call
/// optimization should implement this function.
bool
PPCTargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
CallingConv::ID CalleeCC,
bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins,
SelectionDAG& DAG) const {
if (!getTargetMachine().Options.GuaranteedTailCallOpt)
return false;
// Variable argument functions are not supported.
if (isVarArg)
return false;
MachineFunction &MF = DAG.getMachineFunction();
CallingConv::ID CallerCC = MF.getFunction().getCallingConv();
if (CalleeCC == CallingConv::Fast && CallerCC == CalleeCC) {
// Functions containing by val parameters are not supported.
for (unsigned i = 0; i != Ins.size(); i++) {
ISD::ArgFlagsTy Flags = Ins[i].Flags;
if (Flags.isByVal()) return false;
}
// Non-PIC/GOT tail calls are supported.
if (getTargetMachine().getRelocationModel() != Reloc::PIC_)
return true;
// At the moment we can only do local tail calls (in same module, hidden
// or protected) if we are generating PIC.
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
return G->getGlobal()->hasHiddenVisibility()
|| G->getGlobal()->hasProtectedVisibility();
}
return false;
}
/// isCallCompatibleAddress - Return the immediate to use if the specified
/// 32-bit value is representable in the immediate field of a BxA instruction.
static SDNode *isBLACompatibleAddress(SDValue Op, SelectionDAG &DAG) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op);
if (!C) return nullptr;
int Addr = C->getZExtValue();
if ((Addr & 3) != 0 || // Low 2 bits are implicitly zero.
SignExtend32<26>(Addr) != Addr)
return nullptr; // Top 6 bits have to be sext of immediate.
return DAG
.getConstant(
(int)C->getZExtValue() >> 2, SDLoc(Op),
DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout()))
.getNode();
}
namespace {
struct TailCallArgumentInfo {
SDValue Arg;
SDValue FrameIdxOp;
int FrameIdx = 0;
TailCallArgumentInfo() = default;
};
} // end anonymous namespace
/// StoreTailCallArgumentsToStackSlot - Stores arguments to their stack slot.
static void StoreTailCallArgumentsToStackSlot(
SelectionDAG &DAG, SDValue Chain,
const SmallVectorImpl<TailCallArgumentInfo> &TailCallArgs,
SmallVectorImpl<SDValue> &MemOpChains, const SDLoc &dl) {
for (unsigned i = 0, e = TailCallArgs.size(); i != e; ++i) {
SDValue Arg = TailCallArgs[i].Arg;
SDValue FIN = TailCallArgs[i].FrameIdxOp;
int FI = TailCallArgs[i].FrameIdx;
// Store relative to framepointer.
MemOpChains.push_back(DAG.getStore(
Chain, dl, Arg, FIN,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI)));
}
}
/// EmitTailCallStoreFPAndRetAddr - Move the frame pointer and return address to
/// the appropriate stack slot for the tail call optimized function call.
static SDValue EmitTailCallStoreFPAndRetAddr(SelectionDAG &DAG, SDValue Chain,
SDValue OldRetAddr, SDValue OldFP,
int SPDiff, const SDLoc &dl) {
if (SPDiff) {
// Calculate the new stack slot for the return address.
MachineFunction &MF = DAG.getMachineFunction();
const PPCSubtarget &Subtarget = MF.getSubtarget<PPCSubtarget>();
const PPCFrameLowering *FL = Subtarget.getFrameLowering();
bool isPPC64 = Subtarget.isPPC64();
int SlotSize = isPPC64 ? 8 : 4;
int NewRetAddrLoc = SPDiff + FL->getReturnSaveOffset();
int NewRetAddr = MF.getFrameInfo().CreateFixedObject(SlotSize,
NewRetAddrLoc, true);
EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
SDValue NewRetAddrFrIdx = DAG.getFrameIndex(NewRetAddr, VT);
Chain = DAG.getStore(Chain, dl, OldRetAddr, NewRetAddrFrIdx,
MachinePointerInfo::getFixedStack(MF, NewRetAddr));
}
return Chain;
}
/// CalculateTailCallArgDest - Remember Argument for later processing. Calculate
/// the position of the argument.
static void
CalculateTailCallArgDest(SelectionDAG &DAG, MachineFunction &MF, bool isPPC64,
SDValue Arg, int SPDiff, unsigned ArgOffset,
SmallVectorImpl<TailCallArgumentInfo>& TailCallArguments) {
int Offset = ArgOffset + SPDiff;
uint32_t OpSize = (Arg.getValueSizeInBits() + 7) / 8;
int FI = MF.getFrameInfo().CreateFixedObject(OpSize, Offset, true);
EVT VT = isPPC64 ? MVT::i64 : MVT::i32;
SDValue FIN = DAG.getFrameIndex(FI, VT);
TailCallArgumentInfo Info;
Info.Arg = Arg;
Info.FrameIdxOp = FIN;
Info.FrameIdx = FI;
TailCallArguments.push_back(Info);
}
/// EmitTCFPAndRetAddrLoad - Emit load from frame pointer and return address
/// stack slot. Returns the chain as result and the loaded frame pointers in
/// LROpOut/FPOpout. Used when tail calling.
SDValue PPCTargetLowering::EmitTailCallLoadFPAndRetAddr(
SelectionDAG &DAG, int SPDiff, SDValue Chain, SDValue &LROpOut,
SDValue &FPOpOut, const SDLoc &dl) const {
if (SPDiff) {
// Load the LR and FP stack slot for later adjusting.
EVT VT = Subtarget.isPPC64() ? MVT::i64 : MVT::i32;
LROpOut = getReturnAddrFrameIndex(DAG);
LROpOut = DAG.getLoad(VT, dl, Chain, LROpOut, MachinePointerInfo());
Chain = SDValue(LROpOut.getNode(), 1);
}
return Chain;
}
/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
/// by "Src" to address "Dst" of size "Size". Alignment information is
/// specified by the specific parameter attribute. The copy will be passed as
/// a byval function parameter.
/// Sometimes what we are copying is the end of a larger object, the part that
/// does not fit in registers.
static SDValue CreateCopyOfByValArgument(SDValue Src, SDValue Dst,
SDValue Chain, ISD::ArgFlagsTy Flags,
SelectionDAG &DAG, const SDLoc &dl) {
SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), dl, MVT::i32);
return DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode,
Flags.getNonZeroByValAlign(), false, false, false,
MachinePointerInfo(), MachinePointerInfo());
}
/// LowerMemOpCallTo - Store the argument to the stack or remember it in case of
/// tail calls.
static void LowerMemOpCallTo(
SelectionDAG &DAG, MachineFunction &MF, SDValue Chain, SDValue Arg,
SDValue PtrOff, int SPDiff, unsigned ArgOffset, bool isPPC64,
bool isTailCall, bool isVector, SmallVectorImpl<SDValue> &MemOpChains,
SmallVectorImpl<TailCallArgumentInfo> &TailCallArguments, const SDLoc &dl) {
EVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
if (!isTailCall) {
if (isVector) {
SDValue StackPtr;
if (isPPC64)
StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
else
StackPtr = DAG.getRegister(PPC::R1, MVT::i32);
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
DAG.getConstant(ArgOffset, dl, PtrVT));
}
MemOpChains.push_back(
DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo()));
// Calculate and remember argument location.
} else CalculateTailCallArgDest(DAG, MF, isPPC64, Arg, SPDiff, ArgOffset,
TailCallArguments);
}
static void
PrepareTailCall(SelectionDAG &DAG, SDValue &InFlag, SDValue &Chain,
const SDLoc &dl, int SPDiff, unsigned NumBytes, SDValue LROp,
SDValue FPOp,
SmallVectorImpl<TailCallArgumentInfo> &TailCallArguments) {
// Emit a sequence of copyto/copyfrom virtual registers for arguments that
// might overwrite each other in case of tail call optimization.
SmallVector<SDValue, 8> MemOpChains2;
// Do not flag preceding copytoreg stuff together with the following stuff.
InFlag = SDValue();
StoreTailCallArgumentsToStackSlot(DAG, Chain, TailCallArguments,
MemOpChains2, dl);
if (!MemOpChains2.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains2);
// Store the return address to the appropriate stack slot.
Chain = EmitTailCallStoreFPAndRetAddr(DAG, Chain, LROp, FPOp, SPDiff, dl);
// Emit callseq_end just before tailcall node.
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, dl, true),
DAG.getIntPtrConstant(0, dl, true), InFlag, dl);
InFlag = Chain.getValue(1);
}
// Is this global address that of a function that can be called by name? (as
// opposed to something that must hold a descriptor for an indirect call).
static bool isFunctionGlobalAddress(SDValue Callee) {
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
if (Callee.getOpcode() == ISD::GlobalTLSAddress ||
Callee.getOpcode() == ISD::TargetGlobalTLSAddress)
return false;
return G->getGlobal()->getValueType()->isFunctionTy();
}
return false;
}
SDValue PPCTargetLowering::LowerCallResult(
SDValue Chain, SDValue InFlag, CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCRetInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext());
CCRetInfo.AnalyzeCallResult(
Ins, (Subtarget.isSVR4ABI() && CallConv == CallingConv::Cold)
? RetCC_PPC_Cold
: RetCC_PPC);
// Copy all of the result registers out of their specified physreg.
for (unsigned i = 0, e = RVLocs.size(); i != e; ++i) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
SDValue Val;
if (Subtarget.hasSPE() && VA.getLocVT() == MVT::f64) {
SDValue Lo = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
InFlag);
Chain = Lo.getValue(1);
InFlag = Lo.getValue(2);
VA = RVLocs[++i]; // skip ahead to next loc
SDValue Hi = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(), MVT::i32,
InFlag);
Chain = Hi.getValue(1);
InFlag = Hi.getValue(2);
if (!Subtarget.isLittleEndian())
std::swap (Lo, Hi);
Val = DAG.getNode(PPCISD::BUILD_SPE64, dl, MVT::f64, Lo, Hi);
} else {
Val = DAG.getCopyFromReg(Chain, dl,
VA.getLocReg(), VA.getLocVT(), InFlag);
Chain = Val.getValue(1);
InFlag = Val.getValue(2);
}
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::AExt:
Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
break;
case CCValAssign::ZExt:
Val = DAG.getNode(ISD::AssertZext, dl, VA.getLocVT(), Val,
DAG.getValueType(VA.getValVT()));
Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
break;
case CCValAssign::SExt:
Val = DAG.getNode(ISD::AssertSext, dl, VA.getLocVT(), Val,
DAG.getValueType(VA.getValVT()));
Val = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val);
break;
}
InVals.push_back(Val);
}
return Chain;
}
static bool isIndirectCall(const SDValue &Callee, SelectionDAG &DAG,
const PPCSubtarget &Subtarget, bool isPatchPoint) {
// PatchPoint calls are not indirect.
if (isPatchPoint)
return false;
if (isFunctionGlobalAddress(Callee) || isa<ExternalSymbolSDNode>(Callee))
return false;
// Darwin, and 32-bit ELF can use a BLA. The descriptor based ABIs can not
// becuase the immediate function pointer points to a descriptor instead of
// a function entry point. The ELFv2 ABI cannot use a BLA because the function
// pointer immediate points to the global entry point, while the BLA would
// need to jump to the local entry point (see rL211174).
if (!Subtarget.usesFunctionDescriptors() && !Subtarget.isELFv2ABI() &&
isBLACompatibleAddress(Callee, DAG))
return false;
return true;
}
// AIX and 64-bit ELF ABIs w/o PCRel require a TOC save/restore around calls.
static inline bool isTOCSaveRestoreRequired(const PPCSubtarget &Subtarget) {
return Subtarget.isAIXABI() ||
(Subtarget.is64BitELFABI() && !Subtarget.isUsingPCRelativeCalls());
}
static unsigned getCallOpcode(PPCTargetLowering::CallFlags CFlags,
const Function &Caller,
const SDValue &Callee,
const PPCSubtarget &Subtarget,
const TargetMachine &TM) {
if (CFlags.IsTailCall)
return PPCISD::TC_RETURN;
// This is a call through a function pointer.
if (CFlags.IsIndirect) {
// AIX and the 64-bit ELF ABIs need to maintain the TOC pointer accross
// indirect calls. The save of the caller's TOC pointer to the stack will be
// inserted into the DAG as part of call lowering. The restore of the TOC
// pointer is modeled by using a pseudo instruction for the call opcode that
// represents the 2 instruction sequence of an indirect branch and link,
// immediately followed by a load of the TOC pointer from the the stack save
// slot into gpr2. For 64-bit ELFv2 ABI with PCRel, do not restore the TOC
// as it is not saved or used.
return isTOCSaveRestoreRequired(Subtarget) ? PPCISD::BCTRL_LOAD_TOC
: PPCISD::BCTRL;
}
if (Subtarget.isUsingPCRelativeCalls()) {
assert(Subtarget.is64BitELFABI() && "PC Relative is only on ELF ABI.");
return PPCISD::CALL_NOTOC;
}
// The ABIs that maintain a TOC pointer accross calls need to have a nop
// immediately following the call instruction if the caller and callee may
// have different TOC bases. At link time if the linker determines the calls
// may not share a TOC base, the call is redirected to a trampoline inserted
// by the linker. The trampoline will (among other things) save the callers
// TOC pointer at an ABI designated offset in the linkage area and the linker
// will rewrite the nop to be a load of the TOC pointer from the linkage area
// into gpr2.
if (Subtarget.isAIXABI() || Subtarget.is64BitELFABI())
return callsShareTOCBase(&Caller, Callee, TM) ? PPCISD::CALL
: PPCISD::CALL_NOP;
return PPCISD::CALL;
}
static SDValue transformCallee(const SDValue &Callee, SelectionDAG &DAG,
const SDLoc &dl, const PPCSubtarget &Subtarget) {
if (!Subtarget.usesFunctionDescriptors() && !Subtarget.isELFv2ABI())
if (SDNode *Dest = isBLACompatibleAddress(Callee, DAG))
return SDValue(Dest, 0);
// Returns true if the callee is local, and false otherwise.
auto isLocalCallee = [&]() {
const GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
const Module *Mod = DAG.getMachineFunction().getFunction().getParent();
const GlobalValue *GV = G ? G->getGlobal() : nullptr;
return DAG.getTarget().shouldAssumeDSOLocal(*Mod, GV) &&
!dyn_cast_or_null<GlobalIFunc>(GV);
};
// The PLT is only used in 32-bit ELF PIC mode. Attempting to use the PLT in
// a static relocation model causes some versions of GNU LD (2.17.50, at
// least) to force BSS-PLT, instead of secure-PLT, even if all objects are
// built with secure-PLT.
bool UsePlt =
Subtarget.is32BitELFABI() && !isLocalCallee() &&
Subtarget.getTargetMachine().getRelocationModel() == Reloc::PIC_;
const auto getAIXFuncEntryPointSymbolSDNode = [&](const GlobalValue *GV) {
const TargetMachine &TM = Subtarget.getTargetMachine();
const TargetLoweringObjectFile *TLOF = TM.getObjFileLowering();
MCSymbolXCOFF *S =
cast<MCSymbolXCOFF>(TLOF->getFunctionEntryPointSymbol(GV, TM));
MVT PtrVT = DAG.getTargetLoweringInfo().getPointerTy(DAG.getDataLayout());
return DAG.getMCSymbol(S, PtrVT);
};
if (isFunctionGlobalAddress(Callee)) {
const GlobalValue *GV = cast<GlobalAddressSDNode>(Callee)->getGlobal();
if (Subtarget.isAIXABI()) {
assert(!isa<GlobalIFunc>(GV) && "IFunc is not supported on AIX.");
return getAIXFuncEntryPointSymbolSDNode(GV);
}
return DAG.getTargetGlobalAddress(GV, dl, Callee.getValueType(), 0,
UsePlt ? PPCII::MO_PLT : 0);
}
if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
const char *SymName = S->getSymbol();
if (Subtarget.isAIXABI()) {
// If there exists a user-declared function whose name is the same as the
// ExternalSymbol's, then we pick up the user-declared version.
const Module *Mod = DAG.getMachineFunction().getFunction().getParent();
if (const Function *F =
dyn_cast_or_null<Function>(Mod->getNamedValue(SymName)))
return getAIXFuncEntryPointSymbolSDNode(F);
// On AIX, direct function calls reference the symbol for the function's
// entry point, which is named by prepending a "." before the function's
// C-linkage name. A Qualname is returned here because an external
// function entry point is a csect with XTY_ER property.
const auto getExternalFunctionEntryPointSymbol = [&](StringRef SymName) {
auto &Context = DAG.getMachineFunction().getMMI().getContext();
MCSectionXCOFF *Sec = Context.getXCOFFSection(
(Twine(".") + Twine(SymName)).str(), XCOFF::XMC_PR, XCOFF::XTY_ER,
SectionKind::getMetadata());
return Sec->getQualNameSymbol();
};
SymName = getExternalFunctionEntryPointSymbol(SymName)->getName().data();
}
return DAG.getTargetExternalSymbol(SymName, Callee.getValueType(),
UsePlt ? PPCII::MO_PLT : 0);
}
// No transformation needed.
assert(Callee.getNode() && "What no callee?");
return Callee;
}
static SDValue getOutputChainFromCallSeq(SDValue CallSeqStart) {
assert(CallSeqStart.getOpcode() == ISD::CALLSEQ_START &&
"Expected a CALLSEQ_STARTSDNode.");
// The last operand is the chain, except when the node has glue. If the node
// has glue, then the last operand is the glue, and the chain is the second
// last operand.
SDValue LastValue = CallSeqStart.getValue(CallSeqStart->getNumValues() - 1);
if (LastValue.getValueType() != MVT::Glue)
return LastValue;
return CallSeqStart.getValue(CallSeqStart->getNumValues() - 2);
}
// Creates the node that moves a functions address into the count register
// to prepare for an indirect call instruction.
static void prepareIndirectCall(SelectionDAG &DAG, SDValue &Callee,
SDValue &Glue, SDValue &Chain,
const SDLoc &dl) {
SDValue MTCTROps[] = {Chain, Callee, Glue};
EVT ReturnTypes[] = {MVT::Other, MVT::Glue};
Chain = DAG.getNode(PPCISD::MTCTR, dl, makeArrayRef(ReturnTypes, 2),
makeArrayRef(MTCTROps, Glue.getNode() ? 3 : 2));
// The glue is the second value produced.
Glue = Chain.getValue(1);
}
static void prepareDescriptorIndirectCall(SelectionDAG &DAG, SDValue &Callee,
SDValue &Glue, SDValue &Chain,
SDValue CallSeqStart,
const CallBase *CB, const SDLoc &dl,
bool hasNest,
const PPCSubtarget &Subtarget) {
// Function pointers in the 64-bit SVR4 ABI do not point to the function
// entry point, but to the function descriptor (the function entry point
// address is part of the function descriptor though).
// The function descriptor is a three doubleword structure with the
// following fields: function entry point, TOC base address and
// environment pointer.
// Thus for a call through a function pointer, the following actions need
// to be performed:
// 1. Save the TOC of the caller in the TOC save area of its stack
// frame (this is done in LowerCall_Darwin() or LowerCall_64SVR4()).
// 2. Load the address of the function entry point from the function
// descriptor.
// 3. Load the TOC of the callee from the function descriptor into r2.
// 4. Load the environment pointer from the function descriptor into
// r11.
// 5. Branch to the function entry point address.
// 6. On return of the callee, the TOC of the caller needs to be
// restored (this is done in FinishCall()).
//
// The loads are scheduled at the beginning of the call sequence, and the
// register copies are flagged together to ensure that no other
// operations can be scheduled in between. E.g. without flagging the
// copies together, a TOC access in the caller could be scheduled between
// the assignment of the callee TOC and the branch to the callee, which leads
// to incorrect code.
// Start by loading the function address from the descriptor.
SDValue LDChain = getOutputChainFromCallSeq(CallSeqStart);
auto MMOFlags = Subtarget.hasInvariantFunctionDescriptors()
? (MachineMemOperand::MODereferenceable |
MachineMemOperand::MOInvariant)
: MachineMemOperand::MONone;
MachinePointerInfo MPI(CB ? CB->getCalledOperand() : nullptr);
// Registers used in building the DAG.
const MCRegister EnvPtrReg = Subtarget.getEnvironmentPointerRegister();
const MCRegister TOCReg = Subtarget.getTOCPointerRegister();
// Offsets of descriptor members.
const unsigned TOCAnchorOffset = Subtarget.descriptorTOCAnchorOffset();
const unsigned EnvPtrOffset = Subtarget.descriptorEnvironmentPointerOffset();
const MVT RegVT = Subtarget.isPPC64() ? MVT::i64 : MVT::i32;
const unsigned Alignment = Subtarget.isPPC64() ? 8 : 4;
// One load for the functions entry point address.
SDValue LoadFuncPtr = DAG.getLoad(RegVT, dl, LDChain, Callee, MPI,
Alignment, MMOFlags);
// One for loading the TOC anchor for the module that contains the called
// function.
SDValue TOCOff = DAG.getIntPtrConstant(TOCAnchorOffset, dl);
SDValue AddTOC = DAG.getNode(ISD::ADD, dl, RegVT, Callee, TOCOff);
SDValue TOCPtr =
DAG.getLoad(RegVT, dl, LDChain, AddTOC,
MPI.getWithOffset(TOCAnchorOffset), Alignment, MMOFlags);
// One for loading the environment pointer.
SDValue PtrOff = DAG.getIntPtrConstant(EnvPtrOffset, dl);
SDValue AddPtr = DAG.getNode(ISD::ADD, dl, RegVT, Callee, PtrOff);
SDValue LoadEnvPtr =
DAG.getLoad(RegVT, dl, LDChain, AddPtr,
MPI.getWithOffset(EnvPtrOffset), Alignment, MMOFlags);
// Then copy the newly loaded TOC anchor to the TOC pointer.
SDValue TOCVal = DAG.getCopyToReg(Chain, dl, TOCReg, TOCPtr, Glue);
Chain = TOCVal.getValue(0);
Glue = TOCVal.getValue(1);
// If the function call has an explicit 'nest' parameter, it takes the
// place of the environment pointer.
assert((!hasNest || !Subtarget.isAIXABI()) &&
"Nest parameter is not supported on AIX.");
if (!hasNest) {
SDValue EnvVal = DAG.getCopyToReg(Chain, dl, EnvPtrReg, LoadEnvPtr, Glue);
Chain = EnvVal.getValue(0);
Glue = EnvVal.getValue(1);
}
// The rest of the indirect call sequence is the same as the non-descriptor
// DAG.
prepareIndirectCall(DAG, LoadFuncPtr, Glue, Chain, dl);
}
static void
buildCallOperands(SmallVectorImpl<SDValue> &Ops,
PPCTargetLowering::CallFlags CFlags, const SDLoc &dl,
SelectionDAG &DAG,
SmallVector<std::pair<unsigned, SDValue>, 8> &RegsToPass,
SDValue Glue, SDValue Chain, SDValue &Callee, int SPDiff,
const PPCSubtarget &Subtarget) {
const bool IsPPC64 = Subtarget.isPPC64();
// MVT for a general purpose register.
const MVT RegVT = IsPPC64 ? MVT::i64 : MVT::i32;
// First operand is always the chain.
Ops.push_back(Chain);
// If it's a direct call pass the callee as the second operand.
if (!CFlags.IsIndirect)
Ops.push_back(Callee);
else {
assert(!CFlags.IsPatchPoint && "Patch point calls are not indirect.");
// For the TOC based ABIs, we have saved the TOC pointer to the linkage area
// on the stack (this would have been done in `LowerCall_64SVR4` or
// `LowerCall_AIX`). The call instruction is a pseudo instruction that
// represents both the indirect branch and a load that restores the TOC
// pointer from the linkage area. The operand for the TOC restore is an add
// of the TOC save offset to the stack pointer. This must be the second
// operand: after the chain input but before any other variadic arguments.
// For 64-bit ELFv2 ABI with PCRel, do not restore the TOC as it is not
// saved or used.
if (isTOCSaveRestoreRequired(Subtarget)) {
const MCRegister StackPtrReg = Subtarget.getStackPointerRegister();
SDValue StackPtr = DAG.getRegister(StackPtrReg, RegVT);
unsigned TOCSaveOffset = Subtarget.getFrameLowering()->getTOCSaveOffset();
SDValue TOCOff = DAG.getIntPtrConstant(TOCSaveOffset, dl);
SDValue AddTOC = DAG.getNode(ISD::ADD, dl, RegVT, StackPtr, TOCOff);
Ops.push_back(AddTOC);
}
// Add the register used for the environment pointer.
if (Subtarget.usesFunctionDescriptors() && !CFlags.HasNest)
Ops.push_back(DAG.getRegister(Subtarget.getEnvironmentPointerRegister(),
RegVT));
// Add CTR register as callee so a bctr can be emitted later.
if (CFlags.IsTailCall)
Ops.push_back(DAG.getRegister(IsPPC64 ? PPC::CTR8 : PPC::CTR, RegVT));
}
// If this is a tail call add stack pointer delta.
if (CFlags.IsTailCall)
Ops.push_back(DAG.getConstant(SPDiff, dl, MVT::i32));
// Add argument registers to the end of the list so that they are known live
// into the call.
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
Ops.push_back(DAG.getRegister(RegsToPass[i].first,
RegsToPass[i].second.getValueType()));
// We cannot add R2/X2 as an operand here for PATCHPOINT, because there is
// no way to mark dependencies as implicit here.
// We will add the R2/X2 dependency in EmitInstrWithCustomInserter.
if ((Subtarget.is64BitELFABI() || Subtarget.isAIXABI()) &&
!CFlags.IsPatchPoint && !Subtarget.isUsingPCRelativeCalls())
Ops.push_back(DAG.getRegister(Subtarget.getTOCPointerRegister(), RegVT));
// Add implicit use of CR bit 6 for 32-bit SVR4 vararg calls
if (CFlags.IsVarArg && Subtarget.is32BitELFABI())
Ops.push_back(DAG.getRegister(PPC::CR1EQ, MVT::i32));
// Add a register mask operand representing the call-preserved registers.
const TargetRegisterInfo *TRI = Subtarget.getRegisterInfo();
const uint32_t *Mask =
TRI->getCallPreservedMask(DAG.getMachineFunction(), CFlags.CallConv);
assert(Mask && "Missing call preserved mask for calling convention");
Ops.push_back(DAG.getRegisterMask(Mask));
// If the glue is valid, it is the last operand.
if (Glue.getNode())
Ops.push_back(Glue);
}
SDValue PPCTargetLowering::FinishCall(
CallFlags CFlags, const SDLoc &dl, SelectionDAG &DAG,
SmallVector<std::pair<unsigned, SDValue>, 8> &RegsToPass, SDValue Glue,
SDValue Chain, SDValue CallSeqStart, SDValue &Callee, int SPDiff,
unsigned NumBytes, const SmallVectorImpl<ISD::InputArg> &Ins,
SmallVectorImpl<SDValue> &InVals, const CallBase *CB) const {
if ((Subtarget.is64BitELFABI() && !Subtarget.isUsingPCRelativeCalls()) ||
Subtarget.isAIXABI())
setUsesTOCBasePtr(DAG);
unsigned CallOpc =
getCallOpcode(CFlags, DAG.getMachineFunction().getFunction(), Callee,
Subtarget, DAG.getTarget());
if (!CFlags.IsIndirect)
Callee = transformCallee(Callee, DAG, dl, Subtarget);
else if (Subtarget.usesFunctionDescriptors())
prepareDescriptorIndirectCall(DAG, Callee, Glue, Chain, CallSeqStart, CB,
dl, CFlags.HasNest, Subtarget);
else
prepareIndirectCall(DAG, Callee, Glue, Chain, dl);
// Build the operand list for the call instruction.
SmallVector<SDValue, 8> Ops;
buildCallOperands(Ops, CFlags, dl, DAG, RegsToPass, Glue, Chain, Callee,
SPDiff, Subtarget);
// Emit tail call.
if (CFlags.IsTailCall) {
// Indirect tail call when using PC Relative calls do not have the same
// constraints.
assert(((Callee.getOpcode() == ISD::Register &&
cast<RegisterSDNode>(Callee)->getReg() == PPC::CTR) ||
Callee.getOpcode() == ISD::TargetExternalSymbol ||
Callee.getOpcode() == ISD::TargetGlobalAddress ||
isa<ConstantSDNode>(Callee) ||
(CFlags.IsIndirect && Subtarget.isUsingPCRelativeCalls())) &&
"Expecting a global address, external symbol, absolute value, "
"register or an indirect tail call when PC Relative calls are "
"used.");
// PC Relative calls also use TC_RETURN as the way to mark tail calls.
assert(CallOpc == PPCISD::TC_RETURN &&
"Unexpected call opcode for a tail call.");
DAG.getMachineFunction().getFrameInfo().setHasTailCall();
return DAG.getNode(CallOpc, dl, MVT::Other, Ops);
}
std::array<EVT, 2> ReturnTypes = {{MVT::Other, MVT::Glue}};
Chain = DAG.getNode(CallOpc, dl, ReturnTypes, Ops);
DAG.addNoMergeSiteInfo(Chain.getNode(), CFlags.NoMerge);
Glue = Chain.getValue(1);
// When performing tail call optimization the callee pops its arguments off
// the stack. Account for this here so these bytes can be pushed back on in
// PPCFrameLowering::eliminateCallFramePseudoInstr.
int BytesCalleePops = (CFlags.CallConv == CallingConv::Fast &&
getTargetMachine().Options.GuaranteedTailCallOpt)
? NumBytes
: 0;
Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, dl, true),
DAG.getIntPtrConstant(BytesCalleePops, dl, true),
Glue, dl);
Glue = Chain.getValue(1);
return LowerCallResult(Chain, Glue, CFlags.CallConv, CFlags.IsVarArg, Ins, dl,
DAG, InVals);
}
SDValue
PPCTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
SelectionDAG &DAG = CLI.DAG;
SDLoc &dl = CLI.DL;
SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
SmallVectorImpl<SDValue> &OutVals = CLI.OutVals;
SmallVectorImpl<ISD::InputArg> &Ins = CLI.Ins;
SDValue Chain = CLI.Chain;
SDValue Callee = CLI.Callee;
bool &isTailCall = CLI.IsTailCall;
CallingConv::ID CallConv = CLI.CallConv;
bool isVarArg = CLI.IsVarArg;
bool isPatchPoint = CLI.IsPatchPoint;
const CallBase *CB = CLI.CB;
if (isTailCall) {
if (Subtarget.useLongCalls() && !(CB && CB->isMustTailCall()))
isTailCall = false;
else if (Subtarget.isSVR4ABI() && Subtarget.isPPC64())
isTailCall = IsEligibleForTailCallOptimization_64SVR4(
Callee, CallConv, CB, isVarArg, Outs, Ins, DAG);
else
isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv, isVarArg,
Ins, DAG);
if (isTailCall) {
++NumTailCalls;
if (!getTargetMachine().Options.GuaranteedTailCallOpt)
++NumSiblingCalls;
// PC Relative calls no longer guarantee that the callee is a Global
// Address Node. The callee could be an indirect tail call in which
// case the SDValue for the callee could be a load (to load the address
// of a function pointer) or it may be a register copy (to move the
// address of the callee from a function parameter into a virtual
// register). It may also be an ExternalSymbolSDNode (ex memcopy).
assert((Subtarget.isUsingPCRelativeCalls() ||
isa<GlobalAddressSDNode>(Callee)) &&
"Callee should be an llvm::Function object.");
LLVM_DEBUG(dbgs() << "TCO caller: " << DAG.getMachineFunction().getName()
<< "\nTCO callee: ");
LLVM_DEBUG(Callee.dump());
}
}
if (!isTailCall && CB && CB->isMustTailCall())
report_fatal_error("failed to perform tail call elimination on a call "
"site marked musttail");
// When long calls (i.e. indirect calls) are always used, calls are always
// made via function pointer. If we have a function name, first translate it
// into a pointer.
if (Subtarget.useLongCalls() && isa<GlobalAddressSDNode>(Callee) &&
!isTailCall)
Callee = LowerGlobalAddress(Callee, DAG);
CallFlags CFlags(
CallConv, isTailCall, isVarArg, isPatchPoint,
isIndirectCall(Callee, DAG, Subtarget, isPatchPoint),
// hasNest
Subtarget.is64BitELFABI() &&
any_of(Outs, [](ISD::OutputArg Arg) { return Arg.Flags.isNest(); }),
CLI.NoMerge);
if (Subtarget.isAIXABI())
return LowerCall_AIX(Chain, Callee, CFlags, Outs, OutVals, Ins, dl, DAG,
InVals, CB);
assert(Subtarget.isSVR4ABI());
if (Subtarget.isPPC64())
return LowerCall_64SVR4(Chain, Callee, CFlags, Outs, OutVals, Ins, dl, DAG,
InVals, CB);
return LowerCall_32SVR4(Chain, Callee, CFlags, Outs, OutVals, Ins, dl, DAG,
InVals, CB);
}
SDValue PPCTargetLowering::LowerCall_32SVR4(
SDValue Chain, SDValue Callee, CallFlags CFlags,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals,
const CallBase *CB) const {
// See PPCTargetLowering::LowerFormalArguments_32SVR4() for a description
// of the 32-bit SVR4 ABI stack frame layout.
const CallingConv::ID CallConv = CFlags.CallConv;
const bool IsVarArg = CFlags.IsVarArg;
const bool IsTailCall = CFlags.IsTailCall;
assert((CallConv == CallingConv::C ||
CallConv == CallingConv::Cold ||
CallConv == CallingConv::Fast) && "Unknown calling convention!");
const Align PtrAlign(4);
MachineFunction &MF = DAG.getMachineFunction();
// Mark this function as potentially containing a function that contains a
// tail call. As a consequence the frame pointer will be used for dynamicalloc
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
if (getTargetMachine().Options.GuaranteedTailCallOpt &&
CallConv == CallingConv::Fast)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
// Count how many bytes are to be pushed on the stack, including the linkage
// area, parameter list area and the part of the local variable space which
// contains copies of aggregates which are passed by value.
// Assign locations to all of the outgoing arguments.
SmallVector<CCValAssign, 16> ArgLocs;
PPCCCState CCInfo(CallConv, IsVarArg, MF, ArgLocs, *DAG.getContext());
// Reserve space for the linkage area on the stack.
CCInfo.AllocateStack(Subtarget.getFrameLowering()->getLinkageSize(),
PtrAlign);
if (useSoftFloat())
CCInfo.PreAnalyzeCallOperands(Outs);
if (IsVarArg) {
// Handle fixed and variable vector arguments differently.
// Fixed vector arguments go into registers as long as registers are
// available. Variable vector arguments always go into memory.
unsigned NumArgs = Outs.size();
for (unsigned i = 0; i != NumArgs; ++i) {
MVT ArgVT = Outs[i].VT;
ISD::ArgFlagsTy ArgFlags = Outs[i].Flags;
bool Result;
if (Outs[i].IsFixed) {
Result = CC_PPC32_SVR4(i, ArgVT, ArgVT, CCValAssign::Full, ArgFlags,
CCInfo);
} else {
Result = CC_PPC32_SVR4_VarArg(i, ArgVT, ArgVT, CCValAssign::Full,
ArgFlags, CCInfo);
}
if (Result) {
#ifndef NDEBUG
errs() << "Call operand #" << i << " has unhandled type "
<< EVT(ArgVT).getEVTString() << "\n";
#endif
llvm_unreachable(nullptr);
}
}
} else {
// All arguments are treated the same.
CCInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4);
}
CCInfo.clearWasPPCF128();
// Assign locations to all of the outgoing aggregate by value arguments.
SmallVector<CCValAssign, 16> ByValArgLocs;
CCState CCByValInfo(CallConv, IsVarArg, MF, ByValArgLocs, *DAG.getContext());
// Reserve stack space for the allocations in CCInfo.
CCByValInfo.AllocateStack(CCInfo.getNextStackOffset(), PtrAlign);
CCByValInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4_ByVal);
// Size of the linkage area, parameter list area and the part of the local
// space variable where copies of aggregates which are passed by value are
// stored.
unsigned NumBytes = CCByValInfo.getNextStackOffset();
// Calculate by how many bytes the stack has to be adjusted in case of tail
// call optimization.
int SPDiff = CalculateTailCallSPDiff(DAG, IsTailCall, NumBytes);
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
Chain = DAG.getCALLSEQ_START(Chain, NumBytes, 0, dl);
SDValue CallSeqStart = Chain;
// Load the return address and frame pointer so it can be moved somewhere else
// later.
SDValue LROp, FPOp;
Chain = EmitTailCallLoadFPAndRetAddr(DAG, SPDiff, Chain, LROp, FPOp, dl);
// Set up a copy of the stack pointer for use loading and storing any
// arguments that may not fit in the registers available for argument
// passing.
SDValue StackPtr = DAG.getRegister(PPC::R1, MVT::i32);
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<TailCallArgumentInfo, 8> TailCallArguments;
SmallVector<SDValue, 8> MemOpChains;
bool seenFloatArg = false;
// Walk the register/memloc assignments, inserting copies/loads.
// i - Tracks the index into the list of registers allocated for the call
// RealArgIdx - Tracks the index into the list of actual function arguments
// j - Tracks the index into the list of byval arguments
for (unsigned i = 0, RealArgIdx = 0, j = 0, e = ArgLocs.size();
i != e;
++i, ++RealArgIdx) {
CCValAssign &VA = ArgLocs[i];
SDValue Arg = OutVals[RealArgIdx];
ISD::ArgFlagsTy Flags = Outs[RealArgIdx].Flags;
if (Flags.isByVal()) {
// Argument is an aggregate which is passed by value, thus we need to
// create a copy of it in the local variable space of the current stack
// frame (which is the stack frame of the caller) and pass the address of
// this copy to the callee.
assert((j < ByValArgLocs.size()) && "Index out of bounds!");
CCValAssign &ByValVA = ByValArgLocs[j++];
assert((VA.getValNo() == ByValVA.getValNo()) && "ValNo mismatch!");
// Memory reserved in the local variable space of the callers stack frame.
unsigned LocMemOffset = ByValVA.getLocMemOffset();
SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset, dl);
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(MF.getDataLayout()),
StackPtr, PtrOff);
// Create a copy of the argument in the local area of the current
// stack frame.
SDValue MemcpyCall =
CreateCopyOfByValArgument(Arg, PtrOff,
CallSeqStart.getNode()->getOperand(0),
Flags, DAG, dl);
// This must go outside the CALLSEQ_START..END.
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall, NumBytes, 0,
SDLoc(MemcpyCall));
DAG.ReplaceAllUsesWith(CallSeqStart.getNode(),
NewCallSeqStart.getNode());
Chain = CallSeqStart = NewCallSeqStart;
// Pass the address of the aggregate copy on the stack either in a
// physical register or in the parameter list area of the current stack
// frame to the callee.
Arg = PtrOff;
}
// When useCRBits() is true, there can be i1 arguments.
// It is because getRegisterType(MVT::i1) => MVT::i1,
// and for other integer types getRegisterType() => MVT::i32.
// Extend i1 and ensure callee will get i32.
if (Arg.getValueType() == MVT::i1)
Arg = DAG.getNode(Flags.isSExt() ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND,
dl, MVT::i32, Arg);
if (VA.isRegLoc()) {
seenFloatArg |= VA.getLocVT().isFloatingPoint();
// Put argument in a physical register.
if (Subtarget.hasSPE() && Arg.getValueType() == MVT::f64) {
bool IsLE = Subtarget.isLittleEndian();
SDValue SVal = DAG.getNode(PPCISD::EXTRACT_SPE, dl, MVT::i32, Arg,
DAG.getIntPtrConstant(IsLE ? 0 : 1, dl));
RegsToPass.push_back(std::make_pair(VA.getLocReg(), SVal.getValue(0)));
SVal = DAG.getNode(PPCISD::EXTRACT_SPE, dl, MVT::i32, Arg,
DAG.getIntPtrConstant(IsLE ? 1 : 0, dl));
RegsToPass.push_back(std::make_pair(ArgLocs[++i].getLocReg(),
SVal.getValue(0)));
} else
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
} else {
// Put argument in the parameter list area of the current stack frame.
assert(VA.isMemLoc());
unsigned LocMemOffset = VA.getLocMemOffset();
if (!IsTailCall) {
SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset, dl);
PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(MF.getDataLayout()),
StackPtr, PtrOff);
MemOpChains.push_back(
DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo()));
} else {
// Calculate and remember argument location.
CalculateTailCallArgDest(DAG, MF, false, Arg, SPDiff, LocMemOffset,
TailCallArguments);
}
}
}
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
// Set CR bit 6 to true if this is a vararg call with floating args passed in
// registers.
if (IsVarArg) {
SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, InFlag };
Chain = DAG.getNode(seenFloatArg ? PPCISD::CR6SET : PPCISD::CR6UNSET,
dl, VTs, makeArrayRef(Ops, InFlag.getNode() ? 2 : 1));
InFlag = Chain.getValue(1);
}
if (IsTailCall)
PrepareTailCall(DAG, InFlag, Chain, dl, SPDiff, NumBytes, LROp, FPOp,
TailCallArguments);
return FinishCall(CFlags, dl, DAG, RegsToPass, InFlag, Chain, CallSeqStart,
Callee, SPDiff, NumBytes, Ins, InVals, CB);
}
// Copy an argument into memory, being careful to do this outside the
// call sequence for the call to which the argument belongs.
SDValue PPCTargetLowering::createMemcpyOutsideCallSeq(
SDValue Arg, SDValue PtrOff, SDValue CallSeqStart, ISD::ArgFlagsTy Flags,
SelectionDAG &DAG, const SDLoc &dl) const {
SDValue MemcpyCall = CreateCopyOfByValArgument(Arg, PtrOff,
CallSeqStart.getNode()->getOperand(0),
Flags, DAG, dl);
// The MEMCPY must go outside the CALLSEQ_START..END.
int64_t FrameSize = CallSeqStart.getConstantOperandVal(1);
SDValue NewCallSeqStart = DAG.getCALLSEQ_START(MemcpyCall, FrameSize, 0,
SDLoc(MemcpyCall));
DAG.ReplaceAllUsesWith(CallSeqStart.getNode(),
NewCallSeqStart.getNode());
return NewCallSeqStart;
}
SDValue PPCTargetLowering::LowerCall_64SVR4(
SDValue Chain, SDValue Callee, CallFlags CFlags,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals,
const CallBase *CB) const {
bool isELFv2ABI = Subtarget.isELFv2ABI();
bool isLittleEndian = Subtarget.isLittleEndian();
unsigned NumOps = Outs.size();
bool IsSibCall = false;
bool IsFastCall = CFlags.CallConv == CallingConv::Fast;
EVT PtrVT = getPointerTy(DAG.getDataLayout());
unsigned PtrByteSize = 8;
MachineFunction &MF = DAG.getMachineFunction();
if (CFlags.IsTailCall && !getTargetMachine().Options.GuaranteedTailCallOpt)
IsSibCall = true;
// Mark this function as potentially containing a function that contains a
// tail call. As a consequence the frame pointer will be used for dynamicalloc
// and restoring the callers stack pointer in this functions epilog. This is
// done because by tail calling the called function might overwrite the value
// in this function's (MF) stack pointer stack slot 0(SP).
if (getTargetMachine().Options.GuaranteedTailCallOpt && IsFastCall)
MF.getInfo<PPCFunctionInfo>()->setHasFastCall();
assert(!(IsFastCall && CFlags.IsVarArg) &&
"fastcc not supported on varargs functions");
// Count how many bytes are to be pushed on the stack, including the linkage
// area, and parameter passing area. On ELFv1, the linkage area is 48 bytes
// reserved space for [SP][CR][LR][2 x unused][TOC]; on ELFv2, the linkage
// area is 32 bytes reserved space for [SP][CR][LR][TOC].
unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
unsigned NumBytes = LinkageSize;
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
static const MCPhysReg GPR[] = {
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10,
};
static const MCPhysReg VR[] = {
PPC::V2, PPC::V3, PPC::V4, PPC::V5, PPC::V6, PPC::V7, PPC::V8,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
const unsigned NumGPRs = array_lengthof(GPR);
const unsigned NumFPRs = useSoftFloat() ? 0 : 13;
const unsigned NumVRs = array_lengthof(VR);
// On ELFv2, we can avoid allocating the parameter area if all the arguments
// can be passed to the callee in registers.
// For the fast calling convention, there is another check below.
// Note: We should keep consistent with LowerFormalArguments_64SVR4()
bool HasParameterArea = !isELFv2ABI || CFlags.IsVarArg || IsFastCall;
if (!HasParameterArea) {
unsigned ParamAreaSize = NumGPRs * PtrByteSize;
unsigned AvailableFPRs = NumFPRs;
unsigned AvailableVRs = NumVRs;
unsigned NumBytesTmp = NumBytes;
for (unsigned i = 0; i != NumOps; ++i) {
if (Outs[i].Flags.isNest()) continue;
if (CalculateStackSlotUsed(Outs[i].VT, Outs[i].ArgVT, Outs[i].Flags,
PtrByteSize, LinkageSize, ParamAreaSize,
NumBytesTmp, AvailableFPRs, AvailableVRs))
HasParameterArea = true;
}
}
// When using the fast calling convention, we don't provide backing for
// arguments that will be in registers.
unsigned NumGPRsUsed = 0, NumFPRsUsed = 0, NumVRsUsed = 0;
// Avoid allocating parameter area for fastcc functions if all the arguments
// can be passed in the registers.
if (IsFastCall)
HasParameterArea = false;
// Add up all the space actually used.
for (unsigned i = 0; i != NumOps; ++i) {
ISD::ArgFlagsTy Flags = Outs[i].Flags;
EVT ArgVT = Outs[i].VT;
EVT OrigVT = Outs[i].ArgVT;
if (Flags.isNest())
continue;
if (IsFastCall) {
if (Flags.isByVal()) {
NumGPRsUsed += (Flags.getByValSize()+7)/8;
if (NumGPRsUsed > NumGPRs)
HasParameterArea = true;
} else {
switch (ArgVT.getSimpleVT().SimpleTy) {
default: llvm_unreachable("Unexpected ValueType for argument!");
case MVT::i1:
case MVT::i32:
case MVT::i64:
if (++NumGPRsUsed <= NumGPRs)
continue;
break;
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
case MVT::v2f64:
case MVT::v2i64:
case MVT::v1i128:
case MVT::f128:
if (++NumVRsUsed <= NumVRs)
continue;
break;
case MVT::v4f32:
if (++NumVRsUsed <= NumVRs)
continue;
break;
case MVT::f32:
case MVT::f64:
if (++NumFPRsUsed <= NumFPRs)
continue;
break;
}
HasParameterArea = true;
}
}
/* Respect alignment of argument on the stack. */
auto Alignement =
CalculateStackSlotAlignment(ArgVT, OrigVT, Flags, PtrByteSize);
NumBytes = alignTo(NumBytes, Alignement);
NumBytes += CalculateStackSlotSize(ArgVT, Flags, PtrByteSize);
if (Flags.isInConsecutiveRegsLast())
NumBytes = ((NumBytes + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
}
unsigned NumBytesActuallyUsed = NumBytes;
// In the old ELFv1 ABI,
// the prolog code of the callee may store up to 8 GPR argument registers to
// the stack, allowing va_start to index over them in memory if its varargs.
// Because we cannot tell if this is needed on the caller side, we have to
// conservatively assume that it is needed. As such, make sure we have at
// least enough stack space for the caller to store the 8 GPRs.
// In the ELFv2 ABI, we allocate the parameter area iff a callee
// really requires memory operands, e.g. a vararg function.
if (HasParameterArea)
NumBytes = std::max(NumBytes, LinkageSize + 8 * PtrByteSize);
else
NumBytes = LinkageSize;
// Tail call needs the stack to be aligned.
if (getTargetMachine().Options.GuaranteedTailCallOpt && IsFastCall)
NumBytes = EnsureStackAlignment(Subtarget.getFrameLowering(), NumBytes);
int SPDiff = 0;
// Calculate by how many bytes the stack has to be adjusted in case of tail
// call optimization.
if (!IsSibCall)
SPDiff = CalculateTailCallSPDiff(DAG, CFlags.IsTailCall, NumBytes);
// To protect arguments on the stack from being clobbered in a tail call,
// force all the loads to happen before doing any other lowering.
if (CFlags.IsTailCall)
Chain = DAG.getStackArgumentTokenFactor(Chain);
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
if (!IsSibCall)
Chain = DAG.getCALLSEQ_START(Chain, NumBytes, 0, dl);
SDValue CallSeqStart = Chain;
// Load the return address and frame pointer so it can be move somewhere else
// later.
SDValue LROp, FPOp;
Chain = EmitTailCallLoadFPAndRetAddr(DAG, SPDiff, Chain, LROp, FPOp, dl);
// Set up a copy of the stack pointer for use loading and storing any
// arguments that may not fit in the registers available for argument
// passing.
SDValue StackPtr = DAG.getRegister(PPC::X1, MVT::i64);
// Figure out which arguments are going to go in registers, and which in
// memory. Also, if this is a vararg function, floating point operations
// must be stored to our stack, and loaded into integer regs as well, if
// any integer regs are available for argument passing.
unsigned ArgOffset = LinkageSize;
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<TailCallArgumentInfo, 8> TailCallArguments;
SmallVector<SDValue, 8> MemOpChains;
for (unsigned i = 0; i != NumOps; ++i) {
SDValue Arg = OutVals[i];
ISD::ArgFlagsTy Flags = Outs[i].Flags;
EVT ArgVT = Outs[i].VT;
EVT OrigVT = Outs[i].ArgVT;
// PtrOff will be used to store the current argument to the stack if a
// register cannot be found for it.
SDValue PtrOff;
// We re-align the argument offset for each argument, except when using the
// fast calling convention, when we need to make sure we do that only when
// we'll actually use a stack slot.
auto ComputePtrOff = [&]() {
/* Respect alignment of argument on the stack. */
auto Alignment =
CalculateStackSlotAlignment(ArgVT, OrigVT, Flags, PtrByteSize);
ArgOffset = alignTo(ArgOffset, Alignment);
PtrOff = DAG.getConstant(ArgOffset, dl, StackPtr.getValueType());
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
};
if (!IsFastCall) {
ComputePtrOff();
/* Compute GPR index associated with argument offset. */
GPR_idx = (ArgOffset - LinkageSize) / PtrByteSize;
GPR_idx = std::min(GPR_idx, NumGPRs);
}
// Promote integers to 64-bit values.
if (Arg.getValueType() == MVT::i32 || Arg.getValueType() == MVT::i1) {
// FIXME: Should this use ANY_EXTEND if neither sext nor zext?
unsigned ExtOp = Flags.isSExt() ? ISD::SIGN_EXTEND : ISD::ZERO_EXTEND;
Arg = DAG.getNode(ExtOp, dl, MVT::i64, Arg);
}
// FIXME memcpy is used way more than necessary. Correctness first.
// Note: "by value" is code for passing a structure by value, not
// basic types.
if (Flags.isByVal()) {
// Note: Size includes alignment padding, so
// struct x { short a; char b; }
// will have Size = 4. With #pragma pack(1), it will have Size = 3.
// These are the proper values we need for right-justifying the
// aggregate in a parameter register.
unsigned Size = Flags.getByValSize();
// An empty aggregate parameter takes up no storage and no
// registers.
if (Size == 0)
continue;
if (IsFastCall)
ComputePtrOff();
// All aggregates smaller than 8 bytes must be passed right-justified.
if (Size==1 || Size==2 || Size==4) {
EVT VT = (Size==1) ? MVT::i8 : ((Size==2) ? MVT::i16 : MVT::i32);
if (GPR_idx != NumGPRs) {
SDValue Load = DAG.getExtLoad(ISD::EXTLOAD, dl, PtrVT, Chain, Arg,
MachinePointerInfo(), VT);
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
ArgOffset += PtrByteSize;
continue;
}
}
if (GPR_idx == NumGPRs && Size < 8) {
SDValue AddPtr = PtrOff;
if (!isLittleEndian) {
SDValue Const = DAG.getConstant(PtrByteSize - Size, dl,
PtrOff.getValueType());
AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, Const);
}
Chain = CallSeqStart = createMemcpyOutsideCallSeq(Arg, AddPtr,
CallSeqStart,
Flags, DAG, dl);
ArgOffset += PtrByteSize;
continue;
}
// Copy entire object into memory. There are cases where gcc-generated
// code assumes it is there, even if it could be put entirely into
// registers. (This is not what the doc says.)
// FIXME: The above statement is likely due to a misunderstanding of the
// documents. All arguments must be copied into the parameter area BY
// THE CALLEE in the event that the callee takes the address of any
// formal argument. That has not yet been implemented. However, it is
// reasonable to use the stack area as a staging area for the register
// load.
// Skip this for small aggregates, as we will use the same slot for a
// right-justified copy, below.
if (Size >= 8)
Chain = CallSeqStart = createMemcpyOutsideCallSeq(Arg, PtrOff,
CallSeqStart,
Flags, DAG, dl);
// When a register is available, pass a small aggregate right-justified.
if (Size < 8 && GPR_idx != NumGPRs) {
// The easiest way to get this right-justified in a register
// is to copy the structure into the rightmost portion of a
// local variable slot, then load the whole slot into the
// register.
// FIXME: The memcpy seems to produce pretty awful code for
// small aggregates, particularly for packed ones.
// FIXME: It would be preferable to use the slot in the
// parameter save area instead of a new local variable.
SDValue AddPtr = PtrOff;
if (!isLittleEndian) {
SDValue Const = DAG.getConstant(8 - Size, dl, PtrOff.getValueType());
AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, Const);
}
Chain = CallSeqStart = createMemcpyOutsideCallSeq(Arg, AddPtr,
CallSeqStart,
Flags, DAG, dl);
// Load the slot into the register.
SDValue Load =
DAG.getLoad(PtrVT, dl, Chain, PtrOff, MachinePointerInfo());
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
// Done with this argument.
ArgOffset += PtrByteSize;
continue;
}
// For aggregates larger than PtrByteSize, copy the pieces of the
// object that fit into registers from the parameter save area.
for (unsigned j=0; j<Size; j+=PtrByteSize) {
SDValue Const = DAG.getConstant(j, dl, PtrOff.getValueType());
SDValue AddArg = DAG.getNode(ISD::ADD, dl, PtrVT, Arg, Const);
if (GPR_idx != NumGPRs) {
SDValue Load =
DAG.getLoad(PtrVT, dl, Chain, AddArg, MachinePointerInfo());
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
ArgOffset += PtrByteSize;
} else {
ArgOffset += ((Size - j + PtrByteSize-1)/PtrByteSize)*PtrByteSize;
break;
}
}
continue;
}
switch (Arg.getSimpleValueType().SimpleTy) {
default: llvm_unreachable("Unexpected ValueType for argument!");
case MVT::i1:
case MVT::i32:
case MVT::i64:
if (Flags.isNest()) {
// The 'nest' parameter, if any, is passed in R11.
RegsToPass.push_back(std::make_pair(PPC::X11, Arg));
break;
}
// These can be scalar arguments or elements of an integer array type
// passed directly. Clang may use those instead of "byval" aggregate
// types to avoid forcing arguments to memory unnecessarily.
if (GPR_idx != NumGPRs) {
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Arg));
} else {
if (IsFastCall)
ComputePtrOff();
assert(HasParameterArea &&
"Parameter area must exist to pass an argument in memory.");
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
true, CFlags.IsTailCall, false, MemOpChains,
TailCallArguments, dl);
if (IsFastCall)
ArgOffset += PtrByteSize;
}
if (!IsFastCall)
ArgOffset += PtrByteSize;
break;
case MVT::f32:
case MVT::f64: {
// These can be scalar arguments or elements of a float array type
// passed directly. The latter are used to implement ELFv2 homogenous
// float aggregates.
// Named arguments go into FPRs first, and once they overflow, the
// remaining arguments go into GPRs and then the parameter save area.
// Unnamed arguments for vararg functions always go to GPRs and
// then the parameter save area. For now, put all arguments to vararg
// routines always in both locations (FPR *and* GPR or stack slot).
bool NeedGPROrStack = CFlags.IsVarArg || FPR_idx == NumFPRs;
bool NeededLoad = false;
// First load the argument into the next available FPR.
if (FPR_idx != NumFPRs)
RegsToPass.push_back(std::make_pair(FPR[FPR_idx++], Arg));
// Next, load the argument into GPR or stack slot if needed.
if (!NeedGPROrStack)
;
else if (GPR_idx != NumGPRs && !IsFastCall) {
// FIXME: We may want to re-enable this for CallingConv::Fast on the P8
// once we support fp <-> gpr moves.
// In the non-vararg case, this can only ever happen in the
// presence of f32 array types, since otherwise we never run
// out of FPRs before running out of GPRs.
SDValue ArgVal;
// Double values are always passed in a single GPR.
if (Arg.getValueType() != MVT::f32) {
ArgVal = DAG.getNode(ISD::BITCAST, dl, MVT::i64, Arg);
// Non-array float values are extended and passed in a GPR.
} else if (!Flags.isInConsecutiveRegs()) {
ArgVal = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Arg);
ArgVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i64, ArgVal);
// If we have an array of floats, we collect every odd element
// together with its predecessor into one GPR.
} else if (ArgOffset % PtrByteSize != 0) {
SDValue Lo, Hi;
Lo = DAG.getNode(ISD::BITCAST, dl, MVT::i32, OutVals[i - 1]);
Hi = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Arg);
if (!isLittleEndian)
std::swap(Lo, Hi);
ArgVal = DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Lo, Hi);
// The final element, if even, goes into the first half of a GPR.
} else if (Flags.isInConsecutiveRegsLast()) {
ArgVal = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Arg);
ArgVal = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i64, ArgVal);
if (!isLittleEndian)
ArgVal = DAG.getNode(ISD::SHL, dl, MVT::i64, ArgVal,
DAG.getConstant(32, dl, MVT::i32));
// Non-final even elements are skipped; they will be handled
// together the with subsequent argument on the next go-around.
} else
ArgVal = SDValue();
if (ArgVal.getNode())
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], ArgVal));
} else {
if (IsFastCall)
ComputePtrOff();
// Single-precision floating-point values are mapped to the
// second (rightmost) word of the stack doubleword.
if (Arg.getValueType() == MVT::f32 &&
!isLittleEndian && !Flags.isInConsecutiveRegs()) {
SDValue ConstFour = DAG.getConstant(4, dl, PtrOff.getValueType());
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff, ConstFour);
}
assert(HasParameterArea &&
"Parameter area must exist to pass an argument in memory.");
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
true, CFlags.IsTailCall, false, MemOpChains,
TailCallArguments, dl);
NeededLoad = true;
}
// When passing an array of floats, the array occupies consecutive
// space in the argument area; only round up to the next doubleword
// at the end of the array. Otherwise, each float takes 8 bytes.
if (!IsFastCall || NeededLoad) {
ArgOffset += (Arg.getValueType() == MVT::f32 &&
Flags.isInConsecutiveRegs()) ? 4 : 8;
if (Flags.isInConsecutiveRegsLast())
ArgOffset = ((ArgOffset + PtrByteSize - 1)/PtrByteSize) * PtrByteSize;
}
break;
}
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
case MVT::v2f64:
case MVT::v2i64:
case MVT::v1i128:
case MVT::f128:
// These can be scalar arguments or elements of a vector array type
// passed directly. The latter are used to implement ELFv2 homogenous
// vector aggregates.
// For a varargs call, named arguments go into VRs or on the stack as
// usual; unnamed arguments always go to the stack or the corresponding
// GPRs when within range. For now, we always put the value in both
// locations (or even all three).
if (CFlags.IsVarArg) {
assert(HasParameterArea &&
"Parameter area must exist if we have a varargs call.");
// We could elide this store in the case where the object fits
// entirely in R registers. Maybe later.
SDValue Store =
DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo());
MemOpChains.push_back(Store);
if (VR_idx != NumVRs) {
SDValue Load =
DAG.getLoad(MVT::v4f32, dl, Store, PtrOff, MachinePointerInfo());
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(VR[VR_idx++], Load));
}
ArgOffset += 16;
for (unsigned i=0; i<16; i+=PtrByteSize) {
if (GPR_idx == NumGPRs)
break;
SDValue Ix = DAG.getNode(ISD::ADD, dl, PtrVT, PtrOff,
DAG.getConstant(i, dl, PtrVT));
SDValue Load =
DAG.getLoad(PtrVT, dl, Store, Ix, MachinePointerInfo());
MemOpChains.push_back(Load.getValue(1));
RegsToPass.push_back(std::make_pair(GPR[GPR_idx++], Load));
}
break;
}
// Non-varargs Altivec params go into VRs or on the stack.
if (VR_idx != NumVRs) {
RegsToPass.push_back(std::make_pair(VR[VR_idx++], Arg));
} else {
if (IsFastCall)
ComputePtrOff();
assert(HasParameterArea &&
"Parameter area must exist to pass an argument in memory.");
LowerMemOpCallTo(DAG, MF, Chain, Arg, PtrOff, SPDiff, ArgOffset,
true, CFlags.IsTailCall, true, MemOpChains,
TailCallArguments, dl);
if (IsFastCall)
ArgOffset += 16;
}
if (!IsFastCall)
ArgOffset += 16;
break;
}
}
assert((!HasParameterArea || NumBytesActuallyUsed == ArgOffset) &&
"mismatch in size of parameter area");
(void)NumBytesActuallyUsed;
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
// Check if this is an indirect call (MTCTR/BCTRL).
// See prepareDescriptorIndirectCall and buildCallOperands for more
// information about calls through function pointers in the 64-bit SVR4 ABI.
if (CFlags.IsIndirect) {
// For 64-bit ELFv2 ABI with PCRel, do not save the TOC of the
// caller in the TOC save area.
if (isTOCSaveRestoreRequired(Subtarget)) {
assert(!CFlags.IsTailCall && "Indirect tails calls not supported");
// Load r2 into a virtual register and store it to the TOC save area.
setUsesTOCBasePtr(DAG);
SDValue Val = DAG.getCopyFromReg(Chain, dl, PPC::X2, MVT::i64);
// TOC save area offset.
unsigned TOCSaveOffset = Subtarget.getFrameLowering()->getTOCSaveOffset();
SDValue PtrOff = DAG.getIntPtrConstant(TOCSaveOffset, dl);
SDValue AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
Chain = DAG.getStore(Val.getValue(1), dl, Val, AddPtr,
MachinePointerInfo::getStack(
DAG.getMachineFunction(), TOCSaveOffset));
}
// In the ELFv2 ABI, R12 must contain the address of an indirect callee.
// This does not mean the MTCTR instruction must use R12; it's easier
// to model this as an extra parameter, so do that.
if (isELFv2ABI && !CFlags.IsPatchPoint)
RegsToPass.push_back(std::make_pair((unsigned)PPC::X12, Callee));
}
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDValue InFlag;
for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
RegsToPass[i].second, InFlag);
InFlag = Chain.getValue(1);
}
if (CFlags.IsTailCall && !IsSibCall)
PrepareTailCall(DAG, InFlag, Chain, dl, SPDiff, NumBytes, LROp, FPOp,
TailCallArguments);
return FinishCall(CFlags, dl, DAG, RegsToPass, InFlag, Chain, CallSeqStart,
Callee, SPDiff, NumBytes, Ins, InVals, CB);
}
static bool CC_AIX(unsigned ValNo, MVT ValVT, MVT LocVT,
CCValAssign::LocInfo LocInfo, ISD::ArgFlagsTy ArgFlags,
CCState &State) {
const PPCSubtarget &Subtarget = static_cast<const PPCSubtarget &>(
State.getMachineFunction().getSubtarget());
const bool IsPPC64 = Subtarget.isPPC64();
const Align PtrAlign = IsPPC64 ? Align(8) : Align(4);
const MVT RegVT = IsPPC64 ? MVT::i64 : MVT::i32;
if (ValVT.isVector() && !State.getMachineFunction()
.getTarget()
.Options.EnableAIXExtendedAltivecABI)
report_fatal_error("the default Altivec AIX ABI is not yet supported");
if (ValVT == MVT::f128)
report_fatal_error("f128 is unimplemented on AIX.");
if (ArgFlags.isNest())
report_fatal_error("Nest arguments are unimplemented.");
static const MCPhysReg GPR_32[] = {// 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10};
static const MCPhysReg GPR_64[] = {// 64-bit registers.
PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10};
static const MCPhysReg VR[] = {// Vector registers.
PPC::V2, PPC::V3, PPC::V4, PPC::V5,
PPC::V6, PPC::V7, PPC::V8, PPC::V9,
PPC::V10, PPC::V11, PPC::V12, PPC::V13};
if (ArgFlags.isByVal()) {
if (ArgFlags.getNonZeroByValAlign() > PtrAlign)
report_fatal_error("Pass-by-value arguments with alignment greater than "
"register width are not supported.");
const unsigned ByValSize = ArgFlags.getByValSize();
// An empty aggregate parameter takes up no storage and no registers,
// but needs a MemLoc for a stack slot for the formal arguments side.
if (ByValSize == 0) {
State.addLoc(CCValAssign::getMem(ValNo, MVT::INVALID_SIMPLE_VALUE_TYPE,
State.getNextStackOffset(), RegVT,
LocInfo));
return false;
}
const unsigned StackSize = alignTo(ByValSize, PtrAlign);
unsigned Offset = State.AllocateStack(StackSize, PtrAlign);
for (const unsigned E = Offset + StackSize; Offset < E;
Offset += PtrAlign.value()) {
if (unsigned Reg = State.AllocateReg(IsPPC64 ? GPR_64 : GPR_32))
State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, RegVT, LocInfo));
else {
State.addLoc(CCValAssign::getMem(ValNo, MVT::INVALID_SIMPLE_VALUE_TYPE,
Offset, MVT::INVALID_SIMPLE_VALUE_TYPE,
LocInfo));
break;
}
}
return false;
}
// Arguments always reserve parameter save area.
switch (ValVT.SimpleTy) {
default:
report_fatal_error("Unhandled value type for argument.");
case MVT::i64:
// i64 arguments should have been split to i32 for PPC32.
assert(IsPPC64 && "PPC32 should have split i64 values.");
LLVM_FALLTHROUGH;
case MVT::i1:
case MVT::i32: {
const unsigned Offset = State.AllocateStack(PtrAlign.value(), PtrAlign);
// AIX integer arguments are always passed in register width.
if (ValVT.getFixedSizeInBits() < RegVT.getFixedSizeInBits())
LocInfo = ArgFlags.isSExt() ? CCValAssign::LocInfo::SExt
: CCValAssign::LocInfo::ZExt;
if (unsigned Reg = State.AllocateReg(IsPPC64 ? GPR_64 : GPR_32))
State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, RegVT, LocInfo));
else
State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, RegVT, LocInfo));
return false;
}
case MVT::f32:
case MVT::f64: {
// Parameter save area (PSA) is reserved even if the float passes in fpr.
const unsigned StoreSize = LocVT.getStoreSize();
// Floats are always 4-byte aligned in the PSA on AIX.
// This includes f64 in 64-bit mode for ABI compatibility.
const unsigned Offset =
State.AllocateStack(IsPPC64 ? 8 : StoreSize, Align(4));
unsigned FReg = State.AllocateReg(FPR);
if (FReg)
State.addLoc(CCValAssign::getReg(ValNo, ValVT, FReg, LocVT, LocInfo));
// Reserve and initialize GPRs or initialize the PSA as required.
for (unsigned I = 0; I < StoreSize; I += PtrAlign.value()) {
if (unsigned Reg = State.AllocateReg(IsPPC64 ? GPR_64 : GPR_32)) {
assert(FReg && "An FPR should be available when a GPR is reserved.");
if (State.isVarArg()) {
// Successfully reserved GPRs are only initialized for vararg calls.
// Custom handling is required for:
// f64 in PPC32 needs to be split into 2 GPRs.
// f32 in PPC64 needs to occupy only lower 32 bits of 64-bit GPR.
State.addLoc(
CCValAssign::getCustomReg(ValNo, ValVT, Reg, RegVT, LocInfo));
}
} else {
// If there are insufficient GPRs, the PSA needs to be initialized.
// Initialization occurs even if an FPR was initialized for
// compatibility with the AIX XL compiler. The full memory for the
// argument will be initialized even if a prior word is saved in GPR.
// A custom memLoc is used when the argument also passes in FPR so
// that the callee handling can skip over it easily.
State.addLoc(
FReg ? CCValAssign::getCustomMem(ValNo, ValVT, Offset, LocVT,
LocInfo)
: CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
break;
}
}
return false;
}
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
case MVT::v2i64:
case MVT::v2f64:
case MVT::v1i128: {
if (State.isVarArg())
report_fatal_error(
"variadic arguments for vector types are unimplemented for AIX");
if (unsigned VReg = State.AllocateReg(VR))
State.addLoc(CCValAssign::getReg(ValNo, ValVT, VReg, LocVT, LocInfo));
else {
report_fatal_error(
"passing vector parameters to the stack is unimplemented for AIX");
}
return false;
}
}
return true;
}
static const TargetRegisterClass *getRegClassForSVT(MVT::SimpleValueType SVT,
bool IsPPC64) {
assert((IsPPC64 || SVT != MVT::i64) &&
"i64 should have been split for 32-bit codegen.");
switch (SVT) {
default:
report_fatal_error("Unexpected value type for formal argument");
case MVT::i1:
case MVT::i32:
case MVT::i64:
return IsPPC64 ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
case MVT::f32:
return &PPC::F4RCRegClass;
case MVT::f64:
return &PPC::F8RCRegClass;
case MVT::v4f32:
case MVT::v4i32:
case MVT::v8i16:
case MVT::v16i8:
case MVT::v2i64:
case MVT::v2f64:
case MVT::v1i128:
return &PPC::VRRCRegClass;
}
}
static SDValue truncateScalarIntegerArg(ISD::ArgFlagsTy Flags, EVT ValVT,
SelectionDAG &DAG, SDValue ArgValue,
MVT LocVT, const SDLoc &dl) {
assert(ValVT.isScalarInteger() && LocVT.isScalarInteger());
assert(ValVT.getFixedSizeInBits() < LocVT.getFixedSizeInBits());
if (Flags.isSExt())
ArgValue = DAG.getNode(ISD::AssertSext, dl, LocVT, ArgValue,
DAG.getValueType(ValVT));
else if (Flags.isZExt())
ArgValue = DAG.getNode(ISD::AssertZext, dl, LocVT, ArgValue,
DAG.getValueType(ValVT));
return DAG.getNode(ISD::TRUNCATE, dl, ValVT, ArgValue);
}
static unsigned mapArgRegToOffsetAIX(unsigned Reg, const PPCFrameLowering *FL) {
const unsigned LASize = FL->getLinkageSize();
if (PPC::GPRCRegClass.contains(Reg)) {
assert(Reg >= PPC::R3 && Reg <= PPC::R10 &&
"Reg must be a valid argument register!");
return LASize + 4 * (Reg - PPC::R3);
}
if (PPC::G8RCRegClass.contains(Reg)) {
assert(Reg >= PPC::X3 && Reg <= PPC::X10 &&
"Reg must be a valid argument register!");
return LASize + 8 * (Reg - PPC::X3);
}
llvm_unreachable("Only general purpose registers expected.");
}
// AIX ABI Stack Frame Layout:
//
// Low Memory +--------------------------------------------+
// SP +---> | Back chain | ---+
// | +--------------------------------------------+ |
// | | Saved Condition Register | |
// | +--------------------------------------------+ |
// | | Saved Linkage Register | |
// | +--------------------------------------------+ | Linkage Area
// | | Reserved for compilers | |
// | +--------------------------------------------+ |
// | | Reserved for binders | |
// | +--------------------------------------------+ |
// | | Saved TOC pointer | ---+
// | +--------------------------------------------+
// | | Parameter save area |
// | +--------------------------------------------+
// | | Alloca space |
// | +--------------------------------------------+
// | | Local variable space |
// | +--------------------------------------------+
// | | Float/int conversion temporary |
// | +--------------------------------------------+
// | | Save area for AltiVec registers |
// | +--------------------------------------------+
// | | AltiVec alignment padding |
// | +--------------------------------------------+
// | | Save area for VRSAVE register |
// | +--------------------------------------------+
// | | Save area for General Purpose registers |
// | +--------------------------------------------+
// | | Save area for Floating Point registers |
// | +--------------------------------------------+
// +---- | Back chain |
// High Memory +--------------------------------------------+
//
// Specifications:
// AIX 7.2 Assembler Language Reference
// Subroutine linkage convention
SDValue PPCTargetLowering::LowerFormalArguments_AIX(
SDValue Chain, CallingConv::ID CallConv, bool isVarArg,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
assert((CallConv == CallingConv::C || CallConv == CallingConv::Cold ||
CallConv == CallingConv::Fast) &&
"Unexpected calling convention!");
if (getTargetMachine().Options.GuaranteedTailCallOpt)
report_fatal_error("Tail call support is unimplemented on AIX.");
if (useSoftFloat())
report_fatal_error("Soft float support is unimplemented on AIX.");
const PPCSubtarget &Subtarget =
static_cast<const PPCSubtarget &>(DAG.getSubtarget());
const bool IsPPC64 = Subtarget.isPPC64();
const unsigned PtrByteSize = IsPPC64 ? 8 : 4;
// Assign locations to all of the incoming arguments.
SmallVector<CCValAssign, 16> ArgLocs;
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
CCState CCInfo(CallConv, isVarArg, MF, ArgLocs, *DAG.getContext());
const EVT PtrVT = getPointerTy(MF.getDataLayout());
// Reserve space for the linkage area on the stack.
const unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
CCInfo.AllocateStack(LinkageSize, Align(PtrByteSize));
CCInfo.AnalyzeFormalArguments(Ins, CC_AIX);
SmallVector<SDValue, 8> MemOps;
for (size_t I = 0, End = ArgLocs.size(); I != End; /* No increment here */) {
CCValAssign &VA = ArgLocs[I++];
MVT LocVT = VA.getLocVT();
ISD::ArgFlagsTy Flags = Ins[VA.getValNo()].Flags;
if (VA.isMemLoc() && VA.getValVT().isVector())
report_fatal_error(
"passing vector parameters to the stack is unimplemented for AIX");
// For compatibility with the AIX XL compiler, the float args in the
// parameter save area are initialized even if the argument is available
// in register. The caller is required to initialize both the register
// and memory, however, the callee can choose to expect it in either.
// The memloc is dismissed here because the argument is retrieved from
// the register.
if (VA.isMemLoc() && VA.needsCustom())
continue;
if (VA.isRegLoc()) {
if (VA.getValVT().isScalarInteger())
FuncInfo->appendParameterType(PPCFunctionInfo::FixedType);
else if (VA.getValVT().isFloatingPoint() && !VA.getValVT().isVector())
FuncInfo->appendParameterType(VA.getValVT().SimpleTy == MVT::f32
? PPCFunctionInfo::ShortFloatPoint
: PPCFunctionInfo::LongFloatPoint);
}
if (Flags.isByVal() && VA.isMemLoc()) {
const unsigned Size =
alignTo(Flags.getByValSize() ? Flags.getByValSize() : PtrByteSize,
PtrByteSize);
const int FI = MF.getFrameInfo().CreateFixedObject(
Size, VA.getLocMemOffset(), /* IsImmutable */ false,
/* IsAliased */ true);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
InVals.push_back(FIN);
continue;
}
if (Flags.isByVal()) {
assert(VA.isRegLoc() && "MemLocs should already be handled.");
const MCPhysReg ArgReg = VA.getLocReg();
const PPCFrameLowering *FL = Subtarget.getFrameLowering();
if (Flags.getNonZeroByValAlign() > PtrByteSize)
report_fatal_error("Over aligned byvals not supported yet.");
const unsigned StackSize = alignTo(Flags.getByValSize(), PtrByteSize);
const int FI = MF.getFrameInfo().CreateFixedObject(
StackSize, mapArgRegToOffsetAIX(ArgReg, FL), /* IsImmutable */ false,
/* IsAliased */ true);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
InVals.push_back(FIN);
// Add live ins for all the RegLocs for the same ByVal.
const TargetRegisterClass *RegClass =
IsPPC64 ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
auto HandleRegLoc = [&, RegClass, LocVT](const MCPhysReg PhysReg,
unsigned Offset) {
const unsigned VReg = MF.addLiveIn(PhysReg, RegClass);
// Since the callers side has left justified the aggregate in the
// register, we can simply store the entire register into the stack
// slot.
SDValue CopyFrom = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
// The store to the fixedstack object is needed becuase accessing a
// field of the ByVal will use a gep and load. Ideally we will optimize
// to extracting the value from the register directly, and elide the
// stores when the arguments address is not taken, but that will need to
// be future work.
SDValue Store = DAG.getStore(
CopyFrom.getValue(1), dl, CopyFrom,
DAG.getObjectPtrOffset(dl, FIN, TypeSize::Fixed(Offset)),
MachinePointerInfo::getFixedStack(MF, FI, Offset));
MemOps.push_back(Store);
};
unsigned Offset = 0;
HandleRegLoc(VA.getLocReg(), Offset);
Offset += PtrByteSize;
for (; Offset != StackSize && ArgLocs[I].isRegLoc();
Offset += PtrByteSize) {
assert(ArgLocs[I].getValNo() == VA.getValNo() &&
"RegLocs should be for ByVal argument.");
const CCValAssign RL = ArgLocs[I++];
HandleRegLoc(RL.getLocReg(), Offset);
FuncInfo->appendParameterType(PPCFunctionInfo::FixedType);
}
if (Offset != StackSize) {
assert(ArgLocs[I].getValNo() == VA.getValNo() &&
"Expected MemLoc for remaining bytes.");
assert(ArgLocs[I].isMemLoc() && "Expected MemLoc for remaining bytes.");
// Consume the MemLoc.The InVal has already been emitted, so nothing
// more needs to be done.
++I;
}
continue;
}
EVT ValVT = VA.getValVT();
if (VA.isRegLoc() && !VA.needsCustom()) {
MVT::SimpleValueType SVT = ValVT.getSimpleVT().SimpleTy;
unsigned VReg =
MF.addLiveIn(VA.getLocReg(), getRegClassForSVT(SVT, IsPPC64));
SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
if (ValVT.isScalarInteger() &&
(ValVT.getFixedSizeInBits() < LocVT.getFixedSizeInBits())) {
ArgValue =
truncateScalarIntegerArg(Flags, ValVT, DAG, ArgValue, LocVT, dl);
}
InVals.push_back(ArgValue);
continue;
}
if (VA.isMemLoc()) {
const unsigned LocSize = LocVT.getStoreSize();
const unsigned ValSize = ValVT.getStoreSize();
assert((ValSize <= LocSize) &&
"Object size is larger than size of MemLoc");
int CurArgOffset = VA.getLocMemOffset();
// Objects are right-justified because AIX is big-endian.
if (LocSize > ValSize)
CurArgOffset += LocSize - ValSize;
// Potential tail calls could cause overwriting of argument stack slots.
const bool IsImmutable =
!(getTargetMachine().Options.GuaranteedTailCallOpt &&
(CallConv == CallingConv::Fast));
int FI = MFI.CreateFixedObject(ValSize, CurArgOffset, IsImmutable);
SDValue FIN = DAG.getFrameIndex(FI, PtrVT);
SDValue ArgValue =
DAG.getLoad(ValVT, dl, Chain, FIN, MachinePointerInfo());
InVals.push_back(ArgValue);
continue;
}
}
// On AIX a minimum of 8 words is saved to the parameter save area.
const unsigned MinParameterSaveArea = 8 * PtrByteSize;
// Area that is at least reserved in the caller of this function.
unsigned CallerReservedArea =
std::max(CCInfo.getNextStackOffset(), LinkageSize + MinParameterSaveArea);
// Set the size that is at least reserved in caller of this function. Tail
// call optimized function's reserved stack space needs to be aligned so
// that taking the difference between two stack areas will result in an
// aligned stack.
CallerReservedArea =
EnsureStackAlignment(Subtarget.getFrameLowering(), CallerReservedArea);
FuncInfo->setMinReservedArea(CallerReservedArea);
if (isVarArg) {
FuncInfo->setVarArgsFrameIndex(
MFI.CreateFixedObject(PtrByteSize, CCInfo.getNextStackOffset(), true));
SDValue FIN = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
static const MCPhysReg GPR_32[] = {PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10};
static const MCPhysReg GPR_64[] = {PPC::X3, PPC::X4, PPC::X5, PPC::X6,
PPC::X7, PPC::X8, PPC::X9, PPC::X10};
const unsigned NumGPArgRegs = array_lengthof(IsPPC64 ? GPR_64 : GPR_32);
// The fixed integer arguments of a variadic function are stored to the
// VarArgsFrameIndex on the stack so that they may be loaded by
// dereferencing the result of va_next.
for (unsigned GPRIndex =
(CCInfo.getNextStackOffset() - LinkageSize) / PtrByteSize;
GPRIndex < NumGPArgRegs; ++GPRIndex) {
const unsigned VReg =
IsPPC64 ? MF.addLiveIn(GPR_64[GPRIndex], &PPC::G8RCRegClass)
: MF.addLiveIn(GPR_32[GPRIndex], &PPC::GPRCRegClass);
SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, PtrVT);
SDValue Store =
DAG.getStore(Val.getValue(1), dl, Val, FIN, MachinePointerInfo());
MemOps.push_back(Store);
// Increment the address for the next argument to store.
SDValue PtrOff = DAG.getConstant(PtrByteSize, dl, PtrVT);
FIN = DAG.getNode(ISD::ADD, dl, PtrOff.getValueType(), FIN, PtrOff);
}
}
if (!MemOps.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
return Chain;
}
SDValue PPCTargetLowering::LowerCall_AIX(
SDValue Chain, SDValue Callee, CallFlags CFlags,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals,
const CallBase *CB) const {
// See PPCTargetLowering::LowerFormalArguments_AIX() for a description of the
// AIX ABI stack frame layout.
assert((CFlags.CallConv == CallingConv::C ||
CFlags.CallConv == CallingConv::Cold ||
CFlags.CallConv == CallingConv::Fast) &&
"Unexpected calling convention!");
if (CFlags.IsPatchPoint)
report_fatal_error("This call type is unimplemented on AIX.");
const PPCSubtarget& Subtarget =
static_cast<const PPCSubtarget&>(DAG.getSubtarget());
MachineFunction &MF = DAG.getMachineFunction();
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CFlags.CallConv, CFlags.IsVarArg, MF, ArgLocs,
*DAG.getContext());
// Reserve space for the linkage save area (LSA) on the stack.
// In both PPC32 and PPC64 there are 6 reserved slots in the LSA:
// [SP][CR][LR][2 x reserved][TOC].
// The LSA is 24 bytes (6x4) in PPC32 and 48 bytes (6x8) in PPC64.
const unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
const bool IsPPC64 = Subtarget.isPPC64();
const EVT PtrVT = getPointerTy(DAG.getDataLayout());
const unsigned PtrByteSize = IsPPC64 ? 8 : 4;
CCInfo.AllocateStack(LinkageSize, Align(PtrByteSize));
CCInfo.AnalyzeCallOperands(Outs, CC_AIX);
// The prolog code of the callee may store up to 8 GPR argument registers to
// the stack, allowing va_start to index over them in memory if the callee
// is variadic.
// Because we cannot tell if this is needed on the caller side, we have to
// conservatively assume that it is needed. As such, make sure we have at
// least enough stack space for the caller to store the 8 GPRs.
const unsigned MinParameterSaveAreaSize = 8 * PtrByteSize;
const unsigned NumBytes = std::max(LinkageSize + MinParameterSaveAreaSize,
CCInfo.getNextStackOffset());
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass.
Chain = DAG.getCALLSEQ_START(Chain, NumBytes, 0, dl);
SDValue CallSeqStart = Chain;
SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
SmallVector<SDValue, 8> MemOpChains;
// Set up a copy of the stack pointer for loading and storing any
// arguments that may not fit in the registers available for argument
// passing.
const SDValue StackPtr = IsPPC64 ? DAG.getRegister(PPC::X1, MVT::i64)
: DAG.getRegister(PPC::R1, MVT::i32);
for (unsigned I = 0, E = ArgLocs.size(); I != E;) {
const unsigned ValNo = ArgLocs[I].getValNo();
SDValue Arg = OutVals[ValNo];
ISD::ArgFlagsTy Flags = Outs[ValNo].Flags;
if (Flags.isByVal()) {
const unsigned ByValSize = Flags.getByValSize();
// Nothing to do for zero-sized ByVals on the caller side.
if (!ByValSize) {
++I;
continue;
}
auto GetLoad = [&](EVT VT, unsigned LoadOffset) {
return DAG.getExtLoad(
ISD::ZEXTLOAD, dl, PtrVT, Chain,
(LoadOffset != 0)
? DAG.getObjectPtrOffset(dl, Arg, TypeSize::Fixed(LoadOffset))
: Arg,
MachinePointerInfo(), VT);
};
unsigned LoadOffset = 0;
// Initialize registers, which are fully occupied by the by-val argument.
while (LoadOffset + PtrByteSize <= ByValSize && ArgLocs[I].isRegLoc()) {
SDValue Load = GetLoad(PtrVT, LoadOffset);
MemOpChains.push_back(Load.getValue(1));
LoadOffset += PtrByteSize;
const CCValAssign &ByValVA = ArgLocs[I++];
assert(ByValVA.getValNo() == ValNo &&
"Unexpected location for pass-by-value argument.");
RegsToPass.push_back(std::make_pair(ByValVA.getLocReg(), Load));
}
if (LoadOffset == ByValSize)
continue;
// There must be one more loc to handle the remainder.
assert(ArgLocs[I].getValNo() == ValNo &&
"Expected additional location for by-value argument.");
if (ArgLocs[I].isMemLoc()) {
assert(LoadOffset < ByValSize && "Unexpected memloc for by-val arg.");
const CCValAssign &ByValVA = ArgLocs[I++];
ISD::ArgFlagsTy MemcpyFlags = Flags;
// Only memcpy the bytes that don't pass in register.
MemcpyFlags.setByValSize(ByValSize - LoadOffset);
Chain = CallSeqStart = createMemcpyOutsideCallSeq(
(LoadOffset != 0)
? DAG.getObjectPtrOffset(dl, Arg, TypeSize::Fixed(LoadOffset))
: Arg,
DAG.getObjectPtrOffset(dl, StackPtr,
TypeSize::Fixed(ByValVA.getLocMemOffset())),
CallSeqStart, MemcpyFlags, DAG, dl);
continue;
}
// Initialize the final register residue.
// Any residue that occupies the final by-val arg register must be
// left-justified on AIX. Loads must be a power-of-2 size and cannot be
// larger than the ByValSize. For example: a 7 byte by-val arg requires 4,
// 2 and 1 byte loads.
const unsigned ResidueBytes = ByValSize % PtrByteSize;
assert(ResidueBytes != 0 && LoadOffset + PtrByteSize > ByValSize &&
"Unexpected register residue for by-value argument.");
SDValue ResidueVal;
for (unsigned Bytes = 0; Bytes != ResidueBytes;) {
const unsigned N = PowerOf2Floor(ResidueBytes - Bytes);
const MVT VT =
N == 1 ? MVT::i8
: ((N == 2) ? MVT::i16 : (N == 4 ? MVT::i32 : MVT::i64));
SDValue Load = GetLoad(VT, LoadOffset);
MemOpChains.push_back(Load.getValue(1));
LoadOffset += N;
Bytes += N;
// By-val arguments are passed left-justfied in register.
// Every load here needs to be shifted, otherwise a full register load
// should have been used.
assert(PtrVT.getSimpleVT().getSizeInBits() > (Bytes * 8) &&
"Unexpected load emitted during handling of pass-by-value "
"argument.");
unsigned NumSHLBits = PtrVT.getSimpleVT().getSizeInBits() - (Bytes * 8);
EVT ShiftAmountTy =
getShiftAmountTy(Load->getValueType(0), DAG.getDataLayout());
SDValue SHLAmt = DAG.getConstant(NumSHLBits, dl, ShiftAmountTy);
SDValue ShiftedLoad =
DAG.getNode(ISD::SHL, dl, Load.getValueType(), Load, SHLAmt);
ResidueVal = ResidueVal ? DAG.getNode(ISD::OR, dl, PtrVT, ResidueVal,
ShiftedLoad)
: ShiftedLoad;
}
const CCValAssign &ByValVA = ArgLocs[I++];
RegsToPass.push_back(std::make_pair(ByValVA.getLocReg(), ResidueVal));
continue;
}
CCValAssign &VA = ArgLocs[I++];
const MVT LocVT = VA.getLocVT();
const MVT ValVT = VA.getValVT();
if (VA.isMemLoc() && VA.getValVT().isVector())
report_fatal_error(
"passing vector parameters to the stack is unimplemented for AIX");
switch (VA.getLocInfo()) {
default:
report_fatal_error("Unexpected argument extension type.");
case CCValAssign::Full:
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
}
if (VA.isRegLoc() && !VA.needsCustom()) {
RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
continue;
}
if (VA.isMemLoc()) {
SDValue PtrOff =
DAG.getConstant(VA.getLocMemOffset(), dl, StackPtr.getValueType());
PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
MemOpChains.push_back(
DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo()));
continue;
}
// Custom handling is used for GPR initializations for vararg float
// arguments.
assert(VA.isRegLoc() && VA.needsCustom() && CFlags.IsVarArg &&
ValVT.isFloatingPoint() && LocVT.isInteger() &&
"Unexpected register handling for calling convention.");
SDValue ArgAsInt =
DAG.getBitcast(MVT::getIntegerVT(ValVT.getSizeInBits()), Arg);
if (Arg.getValueType().getStoreSize() == LocVT.getStoreSize())
// f32 in 32-bit GPR
// f64 in 64-bit GPR
RegsToPass.push_back(std::make_pair(VA.getLocReg(), ArgAsInt));
else if (Arg.getValueType().getFixedSizeInBits() <
LocVT.getFixedSizeInBits())
// f32 in 64-bit GPR.
RegsToPass.push_back(std::make_pair(
VA.getLocReg(), DAG.getZExtOrTrunc(ArgAsInt, dl, LocVT)));
else {
// f64 in two 32-bit GPRs
// The 2 GPRs are marked custom and expected to be adjacent in ArgLocs.
assert(Arg.getValueType() == MVT::f64 && CFlags.IsVarArg && !IsPPC64 &&
"Unexpected custom register for argument!");
CCValAssign &GPR1 = VA;
SDValue MSWAsI64 = DAG.getNode(ISD::SRL, dl, MVT::i64, ArgAsInt,
DAG.getConstant(32, dl, MVT::i8));
RegsToPass.push_back(std::make_pair(
GPR1.getLocReg(), DAG.getZExtOrTrunc(MSWAsI64, dl, MVT::i32)));
if (I != E) {
// If only 1 GPR was available, there will only be one custom GPR and
// the argument will also pass in memory.
CCValAssign &PeekArg = ArgLocs[I];
if (PeekArg.isRegLoc() && PeekArg.getValNo() == PeekArg.getValNo()) {
assert(PeekArg.needsCustom() && "A second custom GPR is expected.");
CCValAssign &GPR2 = ArgLocs[I++];
RegsToPass.push_back(std::make_pair(
GPR2.getLocReg(), DAG.getZExtOrTrunc(ArgAsInt, dl, MVT::i32)));
}
}
}
}
if (!MemOpChains.empty())
Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
// For indirect calls, we need to save the TOC base to the stack for
// restoration after the call.
if (CFlags.IsIndirect) {
assert(!CFlags.IsTailCall && "Indirect tail-calls not supported.");
const MCRegister TOCBaseReg = Subtarget.getTOCPointerRegister();
const MCRegister StackPtrReg = Subtarget.getStackPointerRegister();
const MVT PtrVT = Subtarget.isPPC64() ? MVT::i64 : MVT::i32;
const unsigned TOCSaveOffset =
Subtarget.getFrameLowering()->getTOCSaveOffset();
setUsesTOCBasePtr(DAG);
SDValue Val = DAG.getCopyFromReg(Chain, dl, TOCBaseReg, PtrVT);
SDValue PtrOff = DAG.getIntPtrConstant(TOCSaveOffset, dl);
SDValue StackPtr = DAG.getRegister(StackPtrReg, PtrVT);
SDValue AddPtr = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
Chain = DAG.getStore(
Val.getValue(1), dl, Val, AddPtr,
MachinePointerInfo::getStack(DAG.getMachineFunction(), TOCSaveOffset));
}
// Build a sequence of copy-to-reg nodes chained together with token chain
// and flag operands which copy the outgoing args into the appropriate regs.
SDValue InFlag;
for (auto Reg : RegsToPass) {
Chain = DAG.getCopyToReg(Chain, dl, Reg.first, Reg.second, InFlag);
InFlag = Chain.getValue(1);
}
const int SPDiff = 0;
return FinishCall(CFlags, dl, DAG, RegsToPass, InFlag, Chain, CallSeqStart,
Callee, SPDiff, NumBytes, Ins, InVals, CB);
}
bool
PPCTargetLowering::CanLowerReturn(CallingConv::ID CallConv,
MachineFunction &MF, bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
LLVMContext &Context) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, MF, RVLocs, Context);
return CCInfo.CheckReturn(
Outs, (Subtarget.isSVR4ABI() && CallConv == CallingConv::Cold)
? RetCC_PPC_Cold
: RetCC_PPC);
}
SDValue
PPCTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
bool isVarArg,
const SmallVectorImpl<ISD::OutputArg> &Outs,
const SmallVectorImpl<SDValue> &OutVals,
const SDLoc &dl, SelectionDAG &DAG) const {
SmallVector<CCValAssign, 16> RVLocs;
CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), RVLocs,
*DAG.getContext());
CCInfo.AnalyzeReturn(Outs,
(Subtarget.isSVR4ABI() && CallConv == CallingConv::Cold)
? RetCC_PPC_Cold
: RetCC_PPC);
SDValue Flag;
SmallVector<SDValue, 4> RetOps(1, Chain);
// Copy the result values into the output registers.
for (unsigned i = 0, RealResIdx = 0; i != RVLocs.size(); ++i, ++RealResIdx) {
CCValAssign &VA = RVLocs[i];
assert(VA.isRegLoc() && "Can only return in registers!");
SDValue Arg = OutVals[RealResIdx];
switch (VA.getLocInfo()) {
default: llvm_unreachable("Unknown loc info!");
case CCValAssign::Full: break;
case CCValAssign::AExt:
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::ZExt:
Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
break;
case CCValAssign::SExt:
Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
break;
}
if (Subtarget.hasSPE() && VA.getLocVT() == MVT::f64) {
bool isLittleEndian = Subtarget.isLittleEndian();
// Legalize ret f64 -> ret 2 x i32.
SDValue SVal =
DAG.getNode(PPCISD::EXTRACT_SPE, dl, MVT::i32, Arg,
DAG.getIntPtrConstant(isLittleEndian ? 0 : 1, dl));
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), SVal, Flag);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
SVal = DAG.getNode(PPCISD::EXTRACT_SPE, dl, MVT::i32, Arg,
DAG.getIntPtrConstant(isLittleEndian ? 1 : 0, dl));
Flag = Chain.getValue(1);
VA = RVLocs[++i]; // skip ahead to next loc
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), SVal, Flag);
} else
Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), Arg, Flag);
Flag = Chain.getValue(1);
RetOps.push_back(DAG.getRegister(VA.getLocReg(), VA.getLocVT()));
}
RetOps[0] = Chain; // Update chain.
// Add the flag if we have it.
if (Flag.getNode())
RetOps.push_back(Flag);
return DAG.getNode(PPCISD::RET_FLAG, dl, MVT::Other, RetOps);
}
SDValue
PPCTargetLowering::LowerGET_DYNAMIC_AREA_OFFSET(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
// Get the correct type for integers.
EVT IntVT = Op.getValueType();
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue FPSIdx = getFramePointerFrameIndex(DAG);
// Build a DYNAREAOFFSET node.
SDValue Ops[2] = {Chain, FPSIdx};
SDVTList VTs = DAG.getVTList(IntVT);
return DAG.getNode(PPCISD::DYNAREAOFFSET, dl, VTs, Ops);
}
SDValue PPCTargetLowering::LowerSTACKRESTORE(SDValue Op,
SelectionDAG &DAG) const {
// When we pop the dynamic allocation we need to restore the SP link.
SDLoc dl(Op);
// Get the correct type for pointers.
EVT PtrVT = getPointerTy(DAG.getDataLayout());
// Construct the stack pointer operand.
bool isPPC64 = Subtarget.isPPC64();
unsigned SP = isPPC64 ? PPC::X1 : PPC::R1;
SDValue StackPtr = DAG.getRegister(SP, PtrVT);
// Get the operands for the STACKRESTORE.
SDValue Chain = Op.getOperand(0);
SDValue SaveSP = Op.getOperand(1);
// Load the old link SP.
SDValue LoadLinkSP =
DAG.getLoad(PtrVT, dl, Chain, StackPtr, MachinePointerInfo());
// Restore the stack pointer.
Chain = DAG.getCopyToReg(LoadLinkSP.getValue(1), dl, SP, SaveSP);
// Store the old link SP.
return DAG.getStore(Chain, dl, LoadLinkSP, StackPtr, MachinePointerInfo());
}
SDValue PPCTargetLowering::getReturnAddrFrameIndex(SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
bool isPPC64 = Subtarget.isPPC64();
EVT PtrVT = getPointerTy(MF.getDataLayout());
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
int RASI = FI->getReturnAddrSaveIndex();
// If the frame pointer save index hasn't been defined yet.
if (!RASI) {
// Find out what the fix offset of the frame pointer save area.
int LROffset = Subtarget.getFrameLowering()->getReturnSaveOffset();
// Allocate the frame index for frame pointer save area.
RASI = MF.getFrameInfo().CreateFixedObject(isPPC64? 8 : 4, LROffset, false);
// Save the result.
FI->setReturnAddrSaveIndex(RASI);
}
return DAG.getFrameIndex(RASI, PtrVT);
}
SDValue
PPCTargetLowering::getFramePointerFrameIndex(SelectionDAG & DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
bool isPPC64 = Subtarget.isPPC64();
EVT PtrVT = getPointerTy(MF.getDataLayout());
// Get current frame pointer save index. The users of this index will be
// primarily DYNALLOC instructions.
PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
int FPSI = FI->getFramePointerSaveIndex();
// If the frame pointer save index hasn't been defined yet.
if (!FPSI) {
// Find out what the fix offset of the frame pointer save area.
int FPOffset = Subtarget.getFrameLowering()->getFramePointerSaveOffset();
// Allocate the frame index for frame pointer save area.
FPSI = MF.getFrameInfo().CreateFixedObject(isPPC64? 8 : 4, FPOffset, true);
// Save the result.
FI->setFramePointerSaveIndex(FPSI);
}
return DAG.getFrameIndex(FPSI, PtrVT);
}
SDValue PPCTargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
SDLoc dl(Op);
// Get the correct type for pointers.
EVT PtrVT = getPointerTy(DAG.getDataLayout());
// Negate the size.
SDValue NegSize = DAG.getNode(ISD::SUB, dl, PtrVT,
DAG.getConstant(0, dl, PtrVT), Size);
// Construct a node for the frame pointer save index.
SDValue FPSIdx = getFramePointerFrameIndex(DAG);
SDValue Ops[3] = { Chain, NegSize, FPSIdx };
SDVTList VTs = DAG.getVTList(PtrVT, MVT::Other);
if (hasInlineStackProbe(MF))
return DAG.getNode(PPCISD::PROBED_ALLOCA, dl, VTs, Ops);
return DAG.getNode(PPCISD::DYNALLOC, dl, VTs, Ops);
}
SDValue PPCTargetLowering::LowerEH_DWARF_CFA(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
bool isPPC64 = Subtarget.isPPC64();
EVT PtrVT = getPointerTy(DAG.getDataLayout());
int FI = MF.getFrameInfo().CreateFixedObject(isPPC64 ? 8 : 4, 0, false);
return DAG.getFrameIndex(FI, PtrVT);
}
SDValue PPCTargetLowering::lowerEH_SJLJ_SETJMP(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
return DAG.getNode(PPCISD::EH_SJLJ_SETJMP, DL,
DAG.getVTList(MVT::i32, MVT::Other),
Op.getOperand(0), Op.getOperand(1));
}
SDValue PPCTargetLowering::lowerEH_SJLJ_LONGJMP(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
return DAG.getNode(PPCISD::EH_SJLJ_LONGJMP, DL, MVT::Other,
Op.getOperand(0), Op.getOperand(1));
}
SDValue PPCTargetLowering::LowerLOAD(SDValue Op, SelectionDAG &DAG) const {
if (Op.getValueType().isVector())
return LowerVectorLoad(Op, DAG);
assert(Op.getValueType() == MVT::i1 &&
"Custom lowering only for i1 loads");
// First, load 8 bits into 32 bits, then truncate to 1 bit.
SDLoc dl(Op);
LoadSDNode *LD = cast<LoadSDNode>(Op);
SDValue Chain = LD->getChain();
SDValue BasePtr = LD->getBasePtr();
MachineMemOperand *MMO = LD->getMemOperand();
SDValue NewLD =
DAG.getExtLoad(ISD::EXTLOAD, dl, getPointerTy(DAG.getDataLayout()), Chain,
BasePtr, MVT::i8, MMO);
SDValue Result = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, NewLD);
SDValue Ops[] = { Result, SDValue(NewLD.getNode(), 1) };
return DAG.getMergeValues(Ops, dl);
}
SDValue PPCTargetLowering::LowerSTORE(SDValue Op, SelectionDAG &DAG) const {
if (Op.getOperand(1).getValueType().isVector())
return LowerVectorStore(Op, DAG);
assert(Op.getOperand(1).getValueType() == MVT::i1 &&
"Custom lowering only for i1 stores");
// First, zero extend to 32 bits, then use a truncating store to 8 bits.
SDLoc dl(Op);
StoreSDNode *ST = cast<StoreSDNode>(Op);
SDValue Chain = ST->getChain();
SDValue BasePtr = ST->getBasePtr();
SDValue Value = ST->getValue();
MachineMemOperand *MMO = ST->getMemOperand();
Value = DAG.getNode(ISD::ZERO_EXTEND, dl, getPointerTy(DAG.getDataLayout()),
Value);
return DAG.getTruncStore(Chain, dl, Value, BasePtr, MVT::i8, MMO);
}
// FIXME: Remove this once the ANDI glue bug is fixed:
SDValue PPCTargetLowering::LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const {
assert(Op.getValueType() == MVT::i1 &&
"Custom lowering only for i1 results");
SDLoc DL(Op);
return DAG.getNode(PPCISD::ANDI_rec_1_GT_BIT, DL, MVT::i1, Op.getOperand(0));
}
SDValue PPCTargetLowering::LowerTRUNCATEVector(SDValue Op,
SelectionDAG &DAG) const {
// Implements a vector truncate that fits in a vector register as a shuffle.
// We want to legalize vector truncates down to where the source fits in
// a vector register (and target is therefore smaller than vector register
// size). At that point legalization will try to custom lower the sub-legal
// result and get here - where we can contain the truncate as a single target
// operation.
// For example a trunc <2 x i16> to <2 x i8> could be visualized as follows:
// <MSB1|LSB1, MSB2|LSB2> to <LSB1, LSB2>
//
// We will implement it for big-endian ordering as this (where x denotes
// undefined):
// < MSB1|LSB1, MSB2|LSB2, uu, uu, uu, uu, uu, uu> to
// < LSB1, LSB2, u, u, u, u, u, u, u, u, u, u, u, u, u, u>
//
// The same operation in little-endian ordering will be:
// <uu, uu, uu, uu, uu, uu, LSB2|MSB2, LSB1|MSB1> to
// <u, u, u, u, u, u, u, u, u, u, u, u, u, u, LSB2, LSB1>
EVT TrgVT = Op.getValueType();
assert(TrgVT.isVector() && "Vector type expected.");
unsigned TrgNumElts = TrgVT.getVectorNumElements();
EVT EltVT = TrgVT.getVectorElementType();
if (!isOperationCustom(Op.getOpcode(), TrgVT) ||
TrgVT.getSizeInBits() > 128 || !isPowerOf2_32(TrgNumElts) ||
!isPowerOf2_32(EltVT.getSizeInBits()))
return SDValue();
SDValue N1 = Op.getOperand(0);
EVT SrcVT = N1.getValueType();
unsigned SrcSize = SrcVT.getSizeInBits();
if (SrcSize > 256 ||
!isPowerOf2_32(SrcVT.getVectorNumElements()) ||
!isPowerOf2_32(SrcVT.getVectorElementType().getSizeInBits()))
return SDValue();
if (SrcSize == 256 && SrcVT.getVectorNumElements() < 2)
return SDValue();
unsigned WideNumElts = 128 / EltVT.getSizeInBits();
EVT WideVT = EVT::getVectorVT(*DAG.getContext(), EltVT, WideNumElts);
SDLoc DL(Op);
SDValue Op1, Op2;
if (SrcSize == 256) {
EVT VecIdxTy = getVectorIdxTy(DAG.getDataLayout());
EVT SplitVT =
N1.getValueType().getHalfNumVectorElementsVT(*DAG.getContext());
unsigned SplitNumElts = SplitVT.getVectorNumElements();
Op1 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, SplitVT, N1,
DAG.getConstant(0, DL, VecIdxTy));
Op2 = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, SplitVT, N1,
DAG.getConstant(SplitNumElts, DL, VecIdxTy));
}
else {
Op1 = SrcSize == 128 ? N1 : widenVec(DAG, N1, DL);
Op2 = DAG.getUNDEF(WideVT);
}
// First list the elements we want to keep.
unsigned SizeMult = SrcSize / TrgVT.getSizeInBits();
SmallVector<int, 16> ShuffV;
if (Subtarget.isLittleEndian())
for (unsigned i = 0; i < TrgNumElts; ++i)
ShuffV.push_back(i * SizeMult);
else
for (unsigned i = 1; i <= TrgNumElts; ++i)
ShuffV.push_back(i * SizeMult - 1);
// Populate the remaining elements with undefs.
for (unsigned i = TrgNumElts; i < WideNumElts; ++i)
// ShuffV.push_back(i + WideNumElts);
ShuffV.push_back(WideNumElts + 1);
Op1 = DAG.getNode(ISD::BITCAST, DL, WideVT, Op1);
Op2 = DAG.getNode(ISD::BITCAST, DL, WideVT, Op2);
return DAG.getVectorShuffle(WideVT, DL, Op1, Op2, ShuffV);
}
/// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when
/// possible.
SDValue PPCTargetLowering::LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const {
// Not FP, or using SPE? Not a fsel.
if (!Op.getOperand(0).getValueType().isFloatingPoint() ||
!Op.getOperand(2).getValueType().isFloatingPoint() || Subtarget.hasSPE())
return Op;
ISD::CondCode CC = cast<CondCodeSDNode>(Op.getOperand(4))->get();
EVT ResVT = Op.getValueType();
EVT CmpVT = Op.getOperand(0).getValueType();
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDValue TV = Op.getOperand(2), FV = Op.getOperand(3);
SDLoc dl(Op);
SDNodeFlags Flags = Op.getNode()->getFlags();
// We have xsmaxcdp/xsmincdp which are OK to emit even in the
// presence of infinities.
if (Subtarget.hasP9Vector() && LHS == TV && RHS == FV) {
switch (CC) {
default:
break;
case ISD::SETOGT:
case ISD::SETGT:
return DAG.getNode(PPCISD::XSMAXCDP, dl, Op.getValueType(), LHS, RHS);
case ISD::SETOLT:
case ISD::SETLT:
return DAG.getNode(PPCISD::XSMINCDP, dl, Op.getValueType(), LHS, RHS);
}
}
// We might be able to do better than this under some circumstances, but in
// general, fsel-based lowering of select is a finite-math-only optimization.
// For more information, see section F.3 of the 2.06 ISA specification.
// With ISA 3.0
if ((!DAG.getTarget().Options.NoInfsFPMath && !Flags.hasNoInfs()) ||
(!DAG.getTarget().Options.NoNaNsFPMath && !Flags.hasNoNaNs()))
return Op;
// If the RHS of the comparison is a 0.0, we don't need to do the
// subtraction at all.
SDValue Sel1;
if (isFloatingPointZero(RHS))
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
case ISD::SETNE:
std::swap(TV, FV);
LLVM_FALLTHROUGH;
case ISD::SETEQ:
if (LHS.getValueType() == MVT::f32) // Comparison is always 64-bits
LHS = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, LHS);
Sel1 = DAG.getNode(PPCISD::FSEL, dl, ResVT, LHS, TV, FV);
if (Sel1.getValueType() == MVT::f32) // Comparison is always 64-bits
Sel1 = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Sel1);
return DAG.getNode(PPCISD::FSEL, dl, ResVT,
DAG.getNode(ISD::FNEG, dl, MVT::f64, LHS), Sel1, FV);
case ISD::SETULT:
case ISD::SETLT:
std::swap(TV, FV); // fsel is natively setge, swap operands for setlt
LLVM_FALLTHROUGH;
case ISD::SETOGE:
case ISD::SETGE:
if (LHS.getValueType() == MVT::f32) // Comparison is always 64-bits
LHS = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, LHS);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, LHS, TV, FV);
case ISD::SETUGT:
case ISD::SETGT:
std::swap(TV, FV); // fsel is natively setge, swap operands for setlt
LLVM_FALLTHROUGH;
case ISD::SETOLE:
case ISD::SETLE:
if (LHS.getValueType() == MVT::f32) // Comparison is always 64-bits
LHS = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, LHS);
return DAG.getNode(PPCISD::FSEL, dl, ResVT,
DAG.getNode(ISD::FNEG, dl, MVT::f64, LHS), TV, FV);
}
SDValue Cmp;
switch (CC) {
default: break; // SETUO etc aren't handled by fsel.
case ISD::SETNE:
std::swap(TV, FV);
LLVM_FALLTHROUGH;
case ISD::SETEQ:
Cmp = DAG.getNode(ISD::FSUB, dl, CmpVT, LHS, RHS, Flags);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
Sel1 = DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, TV, FV);
if (Sel1.getValueType() == MVT::f32) // Comparison is always 64-bits
Sel1 = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Sel1);
return DAG.getNode(PPCISD::FSEL, dl, ResVT,
DAG.getNode(ISD::FNEG, dl, MVT::f64, Cmp), Sel1, FV);
case ISD::SETULT:
case ISD::SETLT:
Cmp = DAG.getNode(ISD::FSUB, dl, CmpVT, LHS, RHS, Flags);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, FV, TV);
case ISD::SETOGE:
case ISD::SETGE:
Cmp = DAG.getNode(ISD::FSUB, dl, CmpVT, LHS, RHS, Flags);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, TV, FV);
case ISD::SETUGT:
case ISD::SETGT:
Cmp = DAG.getNode(ISD::FSUB, dl, CmpVT, RHS, LHS, Flags);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, FV, TV);
case ISD::SETOLE:
case ISD::SETLE:
Cmp = DAG.getNode(ISD::FSUB, dl, CmpVT, RHS, LHS, Flags);
if (Cmp.getValueType() == MVT::f32) // Comparison is always 64-bits
Cmp = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Cmp);
return DAG.getNode(PPCISD::FSEL, dl, ResVT, Cmp, TV, FV);
}
return Op;
}
static unsigned getPPCStrictOpcode(unsigned Opc) {
switch (Opc) {
default:
llvm_unreachable("No strict version of this opcode!");
case PPCISD::FCTIDZ:
return PPCISD::STRICT_FCTIDZ;
case PPCISD::FCTIWZ:
return PPCISD::STRICT_FCTIWZ;
case PPCISD::FCTIDUZ:
return PPCISD::STRICT_FCTIDUZ;
case PPCISD::FCTIWUZ:
return PPCISD::STRICT_FCTIWUZ;
case PPCISD::FCFID:
return PPCISD::STRICT_FCFID;
case PPCISD::FCFIDU:
return PPCISD::STRICT_FCFIDU;
case PPCISD::FCFIDS:
return PPCISD::STRICT_FCFIDS;
case PPCISD::FCFIDUS:
return PPCISD::STRICT_FCFIDUS;
}
}
static SDValue convertFPToInt(SDValue Op, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
SDLoc dl(Op);
bool IsStrict = Op->isStrictFPOpcode();
bool IsSigned = Op.getOpcode() == ISD::FP_TO_SINT ||
Op.getOpcode() == ISD::STRICT_FP_TO_SINT;
// TODO: Any other flags to propagate?
SDNodeFlags Flags;
Flags.setNoFPExcept(Op->getFlags().hasNoFPExcept());
// For strict nodes, source is the second operand.
SDValue Src = Op.getOperand(IsStrict ? 1 : 0);
SDValue Chain = IsStrict ? Op.getOperand(0) : SDValue();
assert(Src.getValueType().isFloatingPoint());
if (Src.getValueType() == MVT::f32) {
if (IsStrict) {
Src =
DAG.getNode(ISD::STRICT_FP_EXTEND, dl,
DAG.getVTList(MVT::f64, MVT::Other), {Chain, Src}, Flags);
Chain = Src.getValue(1);
} else
Src = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Src);
}
SDValue Conv;
unsigned Opc = ISD::DELETED_NODE;
switch (Op.getSimpleValueType().SimpleTy) {
default: llvm_unreachable("Unhandled FP_TO_INT type in custom expander!");
case MVT::i32:
Opc = IsSigned ? PPCISD::FCTIWZ
: (Subtarget.hasFPCVT() ? PPCISD::FCTIWUZ : PPCISD::FCTIDZ);
break;
case MVT::i64:
assert((IsSigned || Subtarget.hasFPCVT()) &&
"i64 FP_TO_UINT is supported only with FPCVT");
Opc = IsSigned ? PPCISD::FCTIDZ : PPCISD::FCTIDUZ;
}
if (IsStrict) {
Opc = getPPCStrictOpcode(Opc);
Conv = DAG.getNode(Opc, dl, DAG.getVTList(MVT::f64, MVT::Other),
{Chain, Src}, Flags);
} else {
Conv = DAG.getNode(Opc, dl, MVT::f64, Src);
}
return Conv;
}
void PPCTargetLowering::LowerFP_TO_INTForReuse(SDValue Op, ReuseLoadInfo &RLI,
SelectionDAG &DAG,
const SDLoc &dl) const {
SDValue Tmp = convertFPToInt(Op, DAG, Subtarget);
bool IsSigned = Op.getOpcode() == ISD::FP_TO_SINT ||
Op.getOpcode() == ISD::STRICT_FP_TO_SINT;
bool IsStrict = Op->isStrictFPOpcode();
// Convert the FP value to an int value through memory.
bool i32Stack = Op.getValueType() == MVT::i32 && Subtarget.hasSTFIWX() &&
(IsSigned || Subtarget.hasFPCVT());
SDValue FIPtr = DAG.CreateStackTemporary(i32Stack ? MVT::i32 : MVT::f64);
int FI = cast<FrameIndexSDNode>(FIPtr)->getIndex();
MachinePointerInfo MPI =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI);
// Emit a store to the stack slot.
SDValue Chain = IsStrict ? Tmp.getValue(1) : DAG.getEntryNode();
Align Alignment(DAG.getEVTAlign(Tmp.getValueType()));
if (i32Stack) {
MachineFunction &MF = DAG.getMachineFunction();
Alignment = Align(4);
MachineMemOperand *MMO =
MF.getMachineMemOperand(MPI, MachineMemOperand::MOStore, 4, Alignment);
SDValue Ops[] = { Chain, Tmp, FIPtr };
Chain = DAG.getMemIntrinsicNode(PPCISD::STFIWX, dl,
DAG.getVTList(MVT::Other), Ops, MVT::i32, MMO);
} else
Chain = DAG.getStore(Chain, dl, Tmp, FIPtr, MPI, Alignment);
// Result is a load from the stack slot. If loading 4 bytes, make sure to
// add in a bias on big endian.
if (Op.getValueType() == MVT::i32 && !i32Stack) {
FIPtr = DAG.getNode(ISD::ADD, dl, FIPtr.getValueType(), FIPtr,
DAG.getConstant(4, dl, FIPtr.getValueType()));
MPI = MPI.getWithOffset(Subtarget.isLittleEndian() ? 0 : 4);
}
RLI.Chain = Chain;
RLI.Ptr = FIPtr;
RLI.MPI = MPI;
RLI.Alignment = Alignment;
}
/// Custom lowers floating point to integer conversions to use
/// the direct move instructions available in ISA 2.07 to avoid the
/// need for load/store combinations.
SDValue PPCTargetLowering::LowerFP_TO_INTDirectMove(SDValue Op,
SelectionDAG &DAG,
const SDLoc &dl) const {
SDValue Conv = convertFPToInt(Op, DAG, Subtarget);
SDValue Mov = DAG.getNode(PPCISD::MFVSR, dl, Op.getValueType(), Conv);
if (Op->isStrictFPOpcode())
return DAG.getMergeValues({Mov, Conv.getValue(1)}, dl);
else
return Mov;
}
SDValue PPCTargetLowering::LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG,
const SDLoc &dl) const {
bool IsStrict = Op->isStrictFPOpcode();
bool IsSigned = Op.getOpcode() == ISD::FP_TO_SINT ||
Op.getOpcode() == ISD::STRICT_FP_TO_SINT;
SDValue Src = Op.getOperand(IsStrict ? 1 : 0);
EVT SrcVT = Src.getValueType();
EVT DstVT = Op.getValueType();
// FP to INT conversions are legal for f128.
if (SrcVT == MVT::f128)
return Subtarget.hasP9Vector() ? Op : SDValue();
// Expand ppcf128 to i32 by hand for the benefit of llvm-gcc bootstrap on
// PPC (the libcall is not available).
if (SrcVT == MVT::ppcf128) {
if (DstVT == MVT::i32) {
// TODO: Conservatively pass only nofpexcept flag here. Need to check and
// set other fast-math flags to FP operations in both strict and
// non-strict cases. (FP_TO_SINT, FSUB)
SDNodeFlags Flags;
Flags.setNoFPExcept(Op->getFlags().hasNoFPExcept());
if (IsSigned) {
SDValue Lo = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::f64, Src,
DAG.getIntPtrConstant(0, dl));
SDValue Hi = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, MVT::f64, Src,
DAG.getIntPtrConstant(1, dl));
// Add the two halves of the long double in round-to-zero mode, and use
// a smaller FP_TO_SINT.
if (IsStrict) {
SDValue Res = DAG.getNode(PPCISD::STRICT_FADDRTZ, dl,
DAG.getVTList(MVT::f64, MVT::Other),
{Op.getOperand(0), Lo, Hi}, Flags);
return DAG.getNode(ISD::STRICT_FP_TO_SINT, dl,
DAG.getVTList(MVT::i32, MVT::Other),
{Res.getValue(1), Res}, Flags);
} else {
SDValue Res = DAG.getNode(PPCISD::FADDRTZ, dl, MVT::f64, Lo, Hi);
return DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Res);
}
} else {
const uint64_t TwoE31[] = {0x41e0000000000000LL, 0};
APFloat APF = APFloat(APFloat::PPCDoubleDouble(), APInt(128, TwoE31));
SDValue Cst = DAG.getConstantFP(APF, dl, SrcVT);
SDValue SignMask = DAG.getConstant(0x80000000, dl, DstVT);
if (IsStrict) {
// Sel = Src < 0x80000000
// FltOfs = select Sel, 0.0, 0x80000000
// IntOfs = select Sel, 0, 0x80000000
// Result = fp_to_sint(Src - FltOfs) ^ IntOfs
SDValue Chain = Op.getOperand(0);
EVT SetCCVT =
getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), SrcVT);
EVT DstSetCCVT =
getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), DstVT);
SDValue Sel = DAG.getSetCC(dl, SetCCVT, Src, Cst, ISD::SETLT,
Chain, true);
Chain = Sel.getValue(1);
SDValue FltOfs = DAG.getSelect(
dl, SrcVT, Sel, DAG.getConstantFP(0.0, dl, SrcVT), Cst);
Sel = DAG.getBoolExtOrTrunc(Sel, dl, DstSetCCVT, DstVT);
SDValue Val = DAG.getNode(ISD::STRICT_FSUB, dl,
DAG.getVTList(SrcVT, MVT::Other),
{Chain, Src, FltOfs}, Flags);
Chain = Val.getValue(1);
SDValue SInt = DAG.getNode(ISD::STRICT_FP_TO_SINT, dl,
DAG.getVTList(DstVT, MVT::Other),
{Chain, Val}, Flags);
Chain = SInt.getValue(1);
SDValue IntOfs = DAG.getSelect(
dl, DstVT, Sel, DAG.getConstant(0, dl, DstVT), SignMask);
SDValue Result = DAG.getNode(ISD::XOR, dl, DstVT, SInt, IntOfs);
return DAG.getMergeValues({Result, Chain}, dl);
} else {
// X>=2^31 ? (int)(X-2^31)+0x80000000 : (int)X
// FIXME: generated code sucks.
SDValue True = DAG.getNode(ISD::FSUB, dl, MVT::ppcf128, Src, Cst);
True = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, True);
True = DAG.getNode(ISD::ADD, dl, MVT::i32, True, SignMask);
SDValue False = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, Src);
return DAG.getSelectCC(dl, Src, Cst, True, False, ISD::SETGE);
}
}
}
return SDValue();
}
if (Subtarget.hasDirectMove() && Subtarget.isPPC64())
return LowerFP_TO_INTDirectMove(Op, DAG, dl);
ReuseLoadInfo RLI;
LowerFP_TO_INTForReuse(Op, RLI, DAG, dl);
return DAG.getLoad(Op.getValueType(), dl, RLI.Chain, RLI.Ptr, RLI.MPI,
RLI.Alignment, RLI.MMOFlags(), RLI.AAInfo, RLI.Ranges);
}
// We're trying to insert a regular store, S, and then a load, L. If the
// incoming value, O, is a load, we might just be able to have our load use the
// address used by O. However, we don't know if anything else will store to
// that address before we can load from it. To prevent this situation, we need
// to insert our load, L, into the chain as a peer of O. To do this, we give L
// the same chain operand as O, we create a token factor from the chain results
// of O and L, and we replace all uses of O's chain result with that token
// factor (see spliceIntoChain below for this last part).
bool PPCTargetLowering::canReuseLoadAddress(SDValue Op, EVT MemVT,
ReuseLoadInfo &RLI,
SelectionDAG &DAG,
ISD::LoadExtType ET) const {
// Conservatively skip reusing for constrained FP nodes.
if (Op->isStrictFPOpcode())
return false;
SDLoc dl(Op);
bool ValidFPToUint = Op.getOpcode() == ISD::FP_TO_UINT &&
(Subtarget.hasFPCVT() || Op.getValueType() == MVT::i32);
if (ET == ISD::NON_EXTLOAD &&
(ValidFPToUint || Op.getOpcode() == ISD::FP_TO_SINT) &&
isOperationLegalOrCustom(Op.getOpcode(),
Op.getOperand(0).getValueType())) {
LowerFP_TO_INTForReuse(Op, RLI, DAG, dl);
return true;
}
LoadSDNode *LD = dyn_cast<LoadSDNode>(Op);
if (!LD || LD->getExtensionType() != ET || LD->isVolatile() ||
LD->isNonTemporal())
return false;
if (LD->getMemoryVT() != MemVT)
return false;
// If the result of the load is an illegal type, then we can't build a
// valid chain for reuse since the legalised loads and token factor node that
// ties the legalised loads together uses a different output chain then the
// illegal load.
if (!isTypeLegal(LD->getValueType(0)))
return false;
RLI.Ptr = LD->getBasePtr();
if (LD->isIndexed() && !LD->getOffset().isUndef()) {
assert(LD->getAddressingMode() == ISD::PRE_INC &&
"Non-pre-inc AM on PPC?");
RLI.Ptr = DAG.getNode(ISD::ADD, dl, RLI.Ptr.getValueType(), RLI.Ptr,
LD->getOffset());
}
RLI.Chain = LD->getChain();
RLI.MPI = LD->getPointerInfo();
RLI.IsDereferenceable = LD->isDereferenceable();
RLI.IsInvariant = LD->isInvariant();
RLI.Alignment = LD->getAlign();
RLI.AAInfo = LD->getAAInfo();
RLI.Ranges = LD->getRanges();
RLI.ResChain = SDValue(LD, LD->isIndexed() ? 2 : 1);
return true;
}
// Given the head of the old chain, ResChain, insert a token factor containing
// it and NewResChain, and make users of ResChain now be users of that token
// factor.
// TODO: Remove and use DAG::makeEquivalentMemoryOrdering() instead.
void PPCTargetLowering::spliceIntoChain(SDValue ResChain,
SDValue NewResChain,
SelectionDAG &DAG) const {
if (!ResChain)
return;
SDLoc dl(NewResChain);
SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
NewResChain, DAG.getUNDEF(MVT::Other));
assert(TF.getNode() != NewResChain.getNode() &&
"A new TF really is required here");
DAG.ReplaceAllUsesOfValueWith(ResChain, TF);
DAG.UpdateNodeOperands(TF.getNode(), ResChain, NewResChain);
}
/// Analyze profitability of direct move
/// prefer float load to int load plus direct move
/// when there is no integer use of int load
bool PPCTargetLowering::directMoveIsProfitable(const SDValue &Op) const {
SDNode *Origin = Op.getOperand(0).getNode();
if (Origin->getOpcode() != ISD::LOAD)
return true;
// If there is no LXSIBZX/LXSIHZX, like Power8,
// prefer direct move if the memory size is 1 or 2 bytes.
MachineMemOperand *MMO = cast<LoadSDNode>(Origin)->getMemOperand();
if (!Subtarget.hasP9Vector() && MMO->getSize() <= 2)
return true;
for (SDNode::use_iterator UI = Origin->use_begin(),
UE = Origin->use_end();
UI != UE; ++UI) {
// Only look at the users of the loaded value.
if (UI.getUse().get().getResNo() != 0)
continue;
if (UI->getOpcode() != ISD::SINT_TO_FP &&
UI->getOpcode() != ISD::UINT_TO_FP &&
UI->getOpcode() != ISD::STRICT_SINT_TO_FP &&
UI->getOpcode() != ISD::STRICT_UINT_TO_FP)
return true;
}
return false;
}
static SDValue convertIntToFP(SDValue Op, SDValue Src, SelectionDAG &DAG,
const PPCSubtarget &Subtarget,
SDValue Chain = SDValue()) {
bool IsSigned = Op.getOpcode() == ISD::SINT_TO_FP ||
Op.getOpcode() == ISD::STRICT_SINT_TO_FP;
SDLoc dl(Op);
// TODO: Any other flags to propagate?
SDNodeFlags Flags;
Flags.setNoFPExcept(Op->getFlags().hasNoFPExcept());
// If we have FCFIDS, then use it when converting to single-precision.
// Otherwise, convert to double-precision and then round.
bool IsSingle = Op.getValueType() == MVT::f32 && Subtarget.hasFPCVT();
unsigned ConvOpc = IsSingle ? (IsSigned ? PPCISD::FCFIDS : PPCISD::FCFIDUS)
: (IsSigned ? PPCISD::FCFID : PPCISD::FCFIDU);
EVT ConvTy = IsSingle ? MVT::f32 : MVT::f64;
if (Op->isStrictFPOpcode()) {
if (!Chain)
Chain = Op.getOperand(0);
return DAG.getNode(getPPCStrictOpcode(ConvOpc), dl,
DAG.getVTList(ConvTy, MVT::Other), {Chain, Src}, Flags);
} else
return DAG.getNode(ConvOpc, dl, ConvTy, Src);
}
/// Custom lowers integer to floating point conversions to use
/// the direct move instructions available in ISA 2.07 to avoid the
/// need for load/store combinations.
SDValue PPCTargetLowering::LowerINT_TO_FPDirectMove(SDValue Op,
SelectionDAG &DAG,
const SDLoc &dl) const {
assert((Op.getValueType() == MVT::f32 ||
Op.getValueType() == MVT::f64) &&
"Invalid floating point type as target of conversion");
assert(Subtarget.hasFPCVT() &&
"Int to FP conversions with direct moves require FPCVT");
SDValue Src = Op.getOperand(Op->isStrictFPOpcode() ? 1 : 0);
bool WordInt = Src.getSimpleValueType().SimpleTy == MVT::i32;
bool Signed = Op.getOpcode() == ISD::SINT_TO_FP ||
Op.getOpcode() == ISD::STRICT_SINT_TO_FP;
unsigned MovOpc = (WordInt && !Signed) ? PPCISD::MTVSRZ : PPCISD::MTVSRA;
SDValue Mov = DAG.getNode(MovOpc, dl, MVT::f64, Src);
return convertIntToFP(Op, Mov, DAG, Subtarget);
}
static SDValue widenVec(SelectionDAG &DAG, SDValue Vec, const SDLoc &dl) {
EVT VecVT = Vec.getValueType();
assert(VecVT.isVector() && "Expected a vector type.");
assert(VecVT.getSizeInBits() < 128 && "Vector is already full width.");
EVT EltVT = VecVT.getVectorElementType();
unsigned WideNumElts = 128 / EltVT.getSizeInBits();
EVT WideVT = EVT::getVectorVT(*DAG.getContext(), EltVT, WideNumElts);
unsigned NumConcat = WideNumElts / VecVT.getVectorNumElements();
SmallVector<SDValue, 16> Ops(NumConcat);
Ops[0] = Vec;
SDValue UndefVec = DAG.getUNDEF(VecVT);
for (unsigned i = 1; i < NumConcat; ++i)
Ops[i] = UndefVec;
return DAG.getNode(ISD::CONCAT_VECTORS, dl, WideVT, Ops);
}
SDValue PPCTargetLowering::LowerINT_TO_FPVector(SDValue Op, SelectionDAG &DAG,
const SDLoc &dl) const {
bool IsStrict = Op->isStrictFPOpcode();
unsigned Opc = Op.getOpcode();
SDValue Src = Op.getOperand(IsStrict ? 1 : 0);
assert((Opc == ISD::UINT_TO_FP || Opc == ISD::SINT_TO_FP ||
Opc == ISD::STRICT_UINT_TO_FP || Opc == ISD::STRICT_SINT_TO_FP) &&
"Unexpected conversion type");
assert((Op.getValueType() == MVT::v2f64 || Op.getValueType() == MVT::v4f32) &&
"Supports conversions to v2f64/v4f32 only.");
// TODO: Any other flags to propagate?
SDNodeFlags Flags;
Flags.setNoFPExcept(Op->getFlags().hasNoFPExcept());
bool SignedConv = Opc == ISD::SINT_TO_FP || Opc == ISD::STRICT_SINT_TO_FP;
bool FourEltRes = Op.getValueType() == MVT::v4f32;
SDValue Wide = widenVec(DAG, Src, dl);
EVT WideVT = Wide.getValueType();
unsigned WideNumElts = WideVT.getVectorNumElements();
MVT IntermediateVT = FourEltRes ? MVT::v4i32 : MVT::v2i64;
SmallVector<int, 16> ShuffV;
for (unsigned i = 0; i < WideNumElts; ++i)
ShuffV.push_back(i + WideNumElts);
int Stride = FourEltRes ? WideNumElts / 4 : WideNumElts / 2;
int SaveElts = FourEltRes ? 4 : 2;
if (Subtarget.isLittleEndian())
for (int i = 0; i < SaveElts; i++)
ShuffV[i * Stride] = i;
else
for (int i = 1; i <= SaveElts; i++)
ShuffV[i * Stride - 1] = i - 1;
SDValue ShuffleSrc2 =
SignedConv ? DAG.getUNDEF(WideVT) : DAG.getConstant(0, dl, WideVT);
SDValue Arrange = DAG.getVectorShuffle(WideVT, dl, Wide, ShuffleSrc2, ShuffV);
SDValue Extend;
if (SignedConv) {
Arrange = DAG.getBitcast(IntermediateVT, Arrange);
EVT ExtVT = Src.getValueType();
if (Subtarget.hasP9Altivec())
ExtVT = EVT::getVectorVT(*DAG.getContext(), WideVT.getVectorElementType(),
IntermediateVT.getVectorNumElements());
Extend = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, IntermediateVT, Arrange,
DAG.getValueType(ExtVT));
} else
Extend = DAG.getNode(ISD::BITCAST, dl, IntermediateVT, Arrange);
if (IsStrict)
return DAG.getNode(Opc, dl, DAG.getVTList(Op.getValueType(), MVT::Other),
{Op.getOperand(0), Extend}, Flags);
return DAG.getNode(Opc, dl, Op.getValueType(), Extend);
}
SDValue PPCTargetLowering::LowerINT_TO_FP(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
bool IsSigned = Op.getOpcode() == ISD::SINT_TO_FP ||
Op.getOpcode() == ISD::STRICT_SINT_TO_FP;
bool IsStrict = Op->isStrictFPOpcode();
SDValue Src = Op.getOperand(IsStrict ? 1 : 0);
SDValue Chain = IsStrict ? Op.getOperand(0) : DAG.getEntryNode();
// TODO: Any other flags to propagate?
SDNodeFlags Flags;
Flags.setNoFPExcept(Op->getFlags().hasNoFPExcept());
EVT InVT = Src.getValueType();
EVT OutVT = Op.getValueType();
if (OutVT.isVector() && OutVT.isFloatingPoint() &&
isOperationCustom(Op.getOpcode(), InVT))
return LowerINT_TO_FPVector(Op, DAG, dl);
// Conversions to f128 are legal.
if (Op.getValueType() == MVT::f128)
return Subtarget.hasP9Vector() ? Op : SDValue();
// Don't handle ppc_fp128 here; let it be lowered to a libcall.
if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
return SDValue();
if (Src.getValueType() == MVT::i1) {
SDValue Sel = DAG.getNode(ISD::SELECT, dl, Op.getValueType(), Src,
DAG.getConstantFP(1.0, dl, Op.getValueType()),
DAG.getConstantFP(0.0, dl, Op.getValueType()));
if (IsStrict)
return DAG.getMergeValues({Sel, Chain}, dl);
else
return Sel;
}
// If we have direct moves, we can do all the conversion, skip the store/load
// however, without FPCVT we can't do most conversions.
if (Subtarget.hasDirectMove() && directMoveIsProfitable(Op) &&
Subtarget.isPPC64() && Subtarget.hasFPCVT())
return LowerINT_TO_FPDirectMove(Op, DAG, dl);
assert((IsSigned || Subtarget.hasFPCVT()) &&
"UINT_TO_FP is supported only with FPCVT");
if (Src.getValueType() == MVT::i64) {
SDValue SINT = Src;
// When converting to single-precision, we actually need to convert
// to double-precision first and then round to single-precision.
// To avoid double-rounding effects during that operation, we have
// to prepare the input operand. Bits that might be truncated when
// converting to double-precision are replaced by a bit that won't
// be lost at this stage, but is below the single-precision rounding
// position.
//
// However, if -enable-unsafe-fp-math is in effect, accept double
// rounding to avoid the extra overhead.
if (Op.getValueType() == MVT::f32 &&
!Subtarget.hasFPCVT() &&
!DAG.getTarget().Options.UnsafeFPMath) {
// Twiddle input to make sure the low 11 bits are zero. (If this
// is the case, we are guaranteed the value will fit into the 53 bit
// mantissa of an IEEE double-precision value without rounding.)
// If any of those low 11 bits were not zero originally, make sure
// bit 12 (value 2048) is set instead, so that the final rounding
// to single-precision gets the correct result.
SDValue Round = DAG.getNode(ISD::AND, dl, MVT::i64,
SINT, DAG.getConstant(2047, dl, MVT::i64));
Round = DAG.getNode(ISD::ADD, dl, MVT::i64,
Round, DAG.getConstant(2047, dl, MVT::i64));
Round = DAG.getNode(ISD::OR, dl, MVT::i64, Round, SINT);
Round = DAG.getNode(ISD::AND, dl, MVT::i64,
Round, DAG.getConstant(-2048, dl, MVT::i64));
// However, we cannot use that value unconditionally: if the magnitude
// of the input value is small, the bit-twiddling we did above might
// end up visibly changing the output. Fortunately, in that case, we
// don't need to twiddle bits since the original input will convert
// exactly to double-precision floating-point already. Therefore,
// construct a conditional to use the original value if the top 11
// bits are all sign-bit copies, and use the rounded value computed
// above otherwise.
SDValue Cond = DAG.getNode(ISD::SRA, dl, MVT::i64,
SINT, DAG.getConstant(53, dl, MVT::i32));
Cond = DAG.getNode(ISD::ADD, dl, MVT::i64,
Cond, DAG.getConstant(1, dl, MVT::i64));
Cond = DAG.getSetCC(
dl,
getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), MVT::i64),
Cond, DAG.getConstant(1, dl, MVT::i64), ISD::SETUGT);
SINT = DAG.getNode(ISD::SELECT, dl, MVT::i64, Cond, Round, SINT);
}
ReuseLoadInfo RLI;
SDValue Bits;
MachineFunction &MF = DAG.getMachineFunction();
if (canReuseLoadAddress(SINT, MVT::i64, RLI, DAG)) {
Bits = DAG.getLoad(MVT::f64, dl, RLI.Chain, RLI.Ptr, RLI.MPI,
RLI.Alignment, RLI.MMOFlags(), RLI.AAInfo, RLI.Ranges);
spliceIntoChain(RLI.ResChain, Bits.getValue(1), DAG);
} else if (Subtarget.hasLFIWAX() &&
canReuseLoadAddress(SINT, MVT::i32, RLI, DAG, ISD::SEXTLOAD)) {
MachineMemOperand *MMO =
MF.getMachineMemOperand(RLI.MPI, MachineMemOperand::MOLoad, 4,
RLI.Alignment, RLI.AAInfo, RLI.Ranges);
SDValue Ops[] = { RLI.Chain, RLI.Ptr };
Bits = DAG.getMemIntrinsicNode(PPCISD::LFIWAX, dl,
DAG.getVTList(MVT::f64, MVT::Other),
Ops, MVT::i32, MMO);
spliceIntoChain(RLI.ResChain, Bits.getValue(1), DAG);
} else if (Subtarget.hasFPCVT() &&
canReuseLoadAddress(SINT, MVT::i32, RLI, DAG, ISD::ZEXTLOAD)) {
MachineMemOperand *MMO =
MF.getMachineMemOperand(RLI.MPI, MachineMemOperand::MOLoad, 4,
RLI.Alignment, RLI.AAInfo, RLI.Ranges);
SDValue Ops[] = { RLI.Chain, RLI.Ptr };
Bits = DAG.getMemIntrinsicNode(PPCISD::LFIWZX, dl,
DAG.getVTList(MVT::f64, MVT::Other),
Ops, MVT::i32, MMO);
spliceIntoChain(RLI.ResChain, Bits.getValue(1), DAG);
} else if (((Subtarget.hasLFIWAX() &&
SINT.getOpcode() == ISD::SIGN_EXTEND) ||
(Subtarget.hasFPCVT() &&
SINT.getOpcode() == ISD::ZERO_EXTEND)) &&
SINT.getOperand(0).getValueType() == MVT::i32) {
MachineFrameInfo &MFI = MF.getFrameInfo();
EVT PtrVT = getPointerTy(DAG.getDataLayout());
int FrameIdx = MFI.CreateStackObject(4, Align(4), false);
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
SDValue Store = DAG.getStore(Chain, dl, SINT.getOperand(0), FIdx,
MachinePointerInfo::getFixedStack(
DAG.getMachineFunction(), FrameIdx));
Chain = Store;
assert(cast<StoreSDNode>(Store)->getMemoryVT() == MVT::i32 &&
"Expected an i32 store");
RLI.Ptr = FIdx;
RLI.Chain = Chain;
RLI.MPI =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FrameIdx);
RLI.Alignment = Align(4);
MachineMemOperand *MMO =
MF.getMachineMemOperand(RLI.MPI, MachineMemOperand::MOLoad, 4,
RLI.Alignment, RLI.AAInfo, RLI.Ranges);
SDValue Ops[] = { RLI.Chain, RLI.Ptr };
Bits = DAG.getMemIntrinsicNode(SINT.getOpcode() == ISD::ZERO_EXTEND ?
PPCISD::LFIWZX : PPCISD::LFIWAX,
dl, DAG.getVTList(MVT::f64, MVT::Other),
Ops, MVT::i32, MMO);
Chain = Bits.getValue(1);
} else
Bits = DAG.getNode(ISD::BITCAST, dl, MVT::f64, SINT);
SDValue FP = convertIntToFP(Op, Bits, DAG, Subtarget, Chain);
if (IsStrict)
Chain = FP.getValue(1);
if (Op.getValueType() == MVT::f32 && !Subtarget.hasFPCVT()) {
if (IsStrict)
FP = DAG.getNode(ISD::STRICT_FP_ROUND, dl,
DAG.getVTList(MVT::f32, MVT::Other),
{Chain, FP, DAG.getIntPtrConstant(0, dl)}, Flags);
else
FP = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, FP,
DAG.getIntPtrConstant(0, dl));
}
return FP;
}
assert(Src.getValueType() == MVT::i32 &&
"Unhandled INT_TO_FP type in custom expander!");
// Since we only generate this in 64-bit mode, we can take advantage of
// 64-bit registers. In particular, sign extend the input value into the
// 64-bit register with extsw, store the WHOLE 64-bit value into the stack
// then lfd it and fcfid it.
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
EVT PtrVT = getPointerTy(MF.getDataLayout());
SDValue Ld;
if (Subtarget.hasLFIWAX() || Subtarget.hasFPCVT()) {
ReuseLoadInfo RLI;
bool ReusingLoad;
if (!(ReusingLoad = canReuseLoadAddress(Src, MVT::i32, RLI, DAG))) {
int FrameIdx = MFI.CreateStackObject(4, Align(4), false);
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
SDValue Store = DAG.getStore(Chain, dl, Src, FIdx,
MachinePointerInfo::getFixedStack(
DAG.getMachineFunction(), FrameIdx));
Chain = Store;
assert(cast<StoreSDNode>(Store)->getMemoryVT() == MVT::i32 &&
"Expected an i32 store");
RLI.Ptr = FIdx;
RLI.Chain = Chain;
RLI.MPI =
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FrameIdx);
RLI.Alignment = Align(4);
}
MachineMemOperand *MMO =
MF.getMachineMemOperand(RLI.MPI, MachineMemOperand::MOLoad, 4,
RLI.Alignment, RLI.AAInfo, RLI.Ranges);
SDValue Ops[] = { RLI.Chain, RLI.Ptr };
Ld = DAG.getMemIntrinsicNode(IsSigned ? PPCISD::LFIWAX : PPCISD::LFIWZX, dl,
DAG.getVTList(MVT::f64, MVT::Other), Ops,
MVT::i32, MMO);
Chain = Ld.getValue(1);
if (ReusingLoad)
spliceIntoChain(RLI.ResChain, Ld.getValue(1), DAG);
} else {
assert(Subtarget.isPPC64() &&
"i32->FP without LFIWAX supported only on PPC64");
int FrameIdx = MFI.CreateStackObject(8, Align(8), false);
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
SDValue Ext64 = DAG.getNode(ISD::SIGN_EXTEND, dl, MVT::i64, Src);
// STD the extended value into the stack slot.
SDValue Store = DAG.getStore(
Chain, dl, Ext64, FIdx,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FrameIdx));
Chain = Store;
// Load the value as a double.
Ld = DAG.getLoad(
MVT::f64, dl, Chain, FIdx,
MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FrameIdx));
Chain = Ld.getValue(1);
}
// FCFID it and return it.
SDValue FP = convertIntToFP(Op, Ld, DAG, Subtarget, Chain);
if (IsStrict)
Chain = FP.getValue(1);
if (Op.getValueType() == MVT::f32 && !Subtarget.hasFPCVT()) {
if (IsStrict)
FP = DAG.getNode(ISD::STRICT_FP_ROUND, dl,
DAG.getVTList(MVT::f32, MVT::Other),
{Chain, FP, DAG.getIntPtrConstant(0, dl)}, Flags);
else
FP = DAG.getNode(ISD::FP_ROUND, dl, MVT::f32, FP,
DAG.getIntPtrConstant(0, dl));
}
return FP;
}
SDValue PPCTargetLowering::LowerFLT_ROUNDS_(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
/*
The rounding mode is in bits 30:31 of FPSR, and has the following
settings:
00 Round to nearest
01 Round to 0
10 Round to +inf
11 Round to -inf
FLT_ROUNDS, on the other hand, expects the following:
-1 Undefined
0 Round to 0
1 Round to nearest
2 Round to +inf
3 Round to -inf
To perform the conversion, we do:
((FPSCR & 0x3) ^ ((~FPSCR & 0x3) >> 1))
*/
MachineFunction &MF = DAG.getMachineFunction();
EVT VT = Op.getValueType();
EVT PtrVT = getPointerTy(MF.getDataLayout());
// Save FP Control Word to register
SDValue Chain = Op.getOperand(0);
SDValue MFFS = DAG.getNode(PPCISD::MFFS, dl, {MVT::f64, MVT::Other}, Chain);
Chain = MFFS.getValue(1);
SDValue CWD;
if (isTypeLegal(MVT::i64)) {
CWD = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
DAG.getNode(ISD::BITCAST, dl, MVT::i64, MFFS));
} else {
// Save FP register to stack slot
int SSFI = MF.getFrameInfo().CreateStackObject(8, Align(8), false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
Chain = DAG.getStore(Chain, dl, MFFS, StackSlot, MachinePointerInfo());
// Load FP Control Word from low 32 bits of stack slot.
assert(hasBigEndianPartOrdering(MVT::i64, MF.getDataLayout()) &&
"Stack slot adjustment is valid only on big endian subtargets!");
SDValue Four = DAG.getConstant(4, dl, PtrVT);
SDValue Addr = DAG.getNode(ISD::ADD, dl, PtrVT, StackSlot, Four);
CWD = DAG.getLoad(MVT::i32, dl, Chain, Addr, MachinePointerInfo());
Chain = CWD.getValue(1);
}
// Transform as necessary
SDValue CWD1 =
DAG.getNode(ISD::AND, dl, MVT::i32,
CWD, DAG.getConstant(3, dl, MVT::i32));
SDValue CWD2 =
DAG.getNode(ISD::SRL, dl, MVT::i32,
DAG.getNode(ISD::AND, dl, MVT::i32,
DAG.getNode(ISD::XOR, dl, MVT::i32,
CWD, DAG.getConstant(3, dl, MVT::i32)),
DAG.getConstant(3, dl, MVT::i32)),
DAG.getConstant(1, dl, MVT::i32));
SDValue RetVal =
DAG.getNode(ISD::XOR, dl, MVT::i32, CWD1, CWD2);
RetVal =
DAG.getNode((VT.getSizeInBits() < 16 ? ISD::TRUNCATE : ISD::ZERO_EXTEND),
dl, VT, RetVal);
return DAG.getMergeValues({RetVal, Chain}, dl);
}
SDValue PPCTargetLowering::LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
SDLoc dl(Op);
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
"Unexpected SHL!");
// Expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, dl, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SHL, dl, VT, Hi, Amt);
SDValue Tmp3 = DAG.getNode(PPCISD::SRL, dl, VT, Lo, Tmp1);
SDValue Tmp4 = DAG.getNode(ISD::OR , dl, VT, Tmp2, Tmp3);
SDValue Tmp5 = DAG.getNode(ISD::ADD, dl, AmtVT, Amt,
DAG.getConstant(-BitWidth, dl, AmtVT));
SDValue Tmp6 = DAG.getNode(PPCISD::SHL, dl, VT, Lo, Tmp5);
SDValue OutHi = DAG.getNode(ISD::OR, dl, VT, Tmp4, Tmp6);
SDValue OutLo = DAG.getNode(PPCISD::SHL, dl, VT, Lo, Amt);
SDValue OutOps[] = { OutLo, OutHi };
return DAG.getMergeValues(OutOps, dl);
}
SDValue PPCTargetLowering::LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) const {
EVT VT = Op.getValueType();
SDLoc dl(Op);
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
"Unexpected SRL!");
// Expand into a bunch of logical ops. Note that these ops
// depend on the PPC behavior for oversized shift amounts.
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, dl, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SRL, dl, VT, Lo, Amt);
SDValue Tmp3 = DAG.getNode(PPCISD::SHL, dl, VT, Hi, Tmp1);
SDValue Tmp4 = DAG.getNode(ISD::OR, dl, VT, Tmp2, Tmp3);
SDValue Tmp5 = DAG.getNode(ISD::ADD, dl, AmtVT, Amt,
DAG.getConstant(-BitWidth, dl, AmtVT));
SDValue Tmp6 = DAG.getNode(PPCISD::SRL, dl, VT, Hi, Tmp5);
SDValue OutLo = DAG.getNode(ISD::OR, dl, VT, Tmp4, Tmp6);
SDValue OutHi = DAG.getNode(PPCISD::SRL, dl, VT, Hi, Amt);
SDValue OutOps[] = { OutLo, OutHi };
return DAG.getMergeValues(OutOps, dl);
}
SDValue PPCTargetLowering::LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
assert(Op.getNumOperands() == 3 &&
VT == Op.getOperand(1).getValueType() &&
"Unexpected SRA!");
// Expand into a bunch of logical ops, followed by a select_cc.
SDValue Lo = Op.getOperand(0);
SDValue Hi = Op.getOperand(1);
SDValue Amt = Op.getOperand(2);
EVT AmtVT = Amt.getValueType();
SDValue Tmp1 = DAG.getNode(ISD::SUB, dl, AmtVT,
DAG.getConstant(BitWidth, dl, AmtVT), Amt);
SDValue Tmp2 = DAG.getNode(PPCISD::SRL, dl, VT, Lo, Amt);
SDValue Tmp3 = DAG.getNode(PPCISD::SHL, dl, VT, Hi, Tmp1);
SDValue Tmp4 = DAG.getNode(ISD::OR, dl, VT, Tmp2, Tmp3);
SDValue Tmp5 = DAG.getNode(ISD::ADD, dl, AmtVT, Amt,
DAG.getConstant(-BitWidth, dl, AmtVT));
SDValue Tmp6 = DAG.getNode(PPCISD::SRA, dl, VT, Hi, Tmp5);
SDValue OutHi = DAG.getNode(PPCISD::SRA, dl, VT, Hi, Amt);
SDValue OutLo = DAG.getSelectCC(dl, Tmp5, DAG.getConstant(0, dl, AmtVT),
Tmp4, Tmp6, ISD::SETLE);
SDValue OutOps[] = { OutLo, OutHi };
return DAG.getMergeValues(OutOps, dl);
}
SDValue PPCTargetLowering::LowerFunnelShift(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
EVT VT = Op.getValueType();
unsigned BitWidth = VT.getSizeInBits();
bool IsFSHL = Op.getOpcode() == ISD::FSHL;
SDValue X = Op.getOperand(0);
SDValue Y = Op.getOperand(1);
SDValue Z = Op.getOperand(2);
EVT AmtVT = Z.getValueType();
// fshl: (X << (Z % BW)) | (Y >> (BW - (Z % BW)))
// fshr: (X << (BW - (Z % BW))) | (Y >> (Z % BW))
// This is simpler than TargetLowering::expandFunnelShift because we can rely
// on PowerPC shift by BW being well defined.
Z = DAG.getNode(ISD::AND, dl, AmtVT, Z,
DAG.getConstant(BitWidth - 1, dl, AmtVT));
SDValue SubZ =
DAG.getNode(ISD::SUB, dl, AmtVT, DAG.getConstant(BitWidth, dl, AmtVT), Z);
X = DAG.getNode(PPCISD::SHL, dl, VT, X, IsFSHL ? Z : SubZ);
Y = DAG.getNode(PPCISD::SRL, dl, VT, Y, IsFSHL ? SubZ : Z);
return DAG.getNode(ISD::OR, dl, VT, X, Y);
}
//===----------------------------------------------------------------------===//
// Vector related lowering.
//
/// getCanonicalConstSplat - Build a canonical splat immediate of Val with an
/// element size of SplatSize. Cast the result to VT.
static SDValue getCanonicalConstSplat(uint64_t Val, unsigned SplatSize, EVT VT,
SelectionDAG &DAG, const SDLoc &dl) {
static const MVT VTys[] = { // canonical VT to use for each size.
MVT::v16i8, MVT::v8i16, MVT::Other, MVT::v4i32
};
EVT ReqVT = VT != MVT::Other ? VT : VTys[SplatSize-1];
// For a splat with all ones, turn it to vspltisb 0xFF to canonicalize.
if (Val == ((1LLU << (SplatSize * 8)) - 1)) {
SplatSize = 1;
Val = 0xFF;
}
EVT CanonicalVT = VTys[SplatSize-1];
// Build a canonical splat for this value.
return DAG.getBitcast(ReqVT, DAG.getConstant(Val, dl, CanonicalVT));
}
/// BuildIntrinsicOp - Return a unary operator intrinsic node with the
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op, SelectionDAG &DAG,
const SDLoc &dl, EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, dl, MVT::i32), Op);
}
/// BuildIntrinsicOp - Return a binary operator intrinsic node with the
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue LHS, SDValue RHS,
SelectionDAG &DAG, const SDLoc &dl,
EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = LHS.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, dl, MVT::i32), LHS, RHS);
}
/// BuildIntrinsicOp - Return a ternary operator intrinsic node with the
/// specified intrinsic ID.
static SDValue BuildIntrinsicOp(unsigned IID, SDValue Op0, SDValue Op1,
SDValue Op2, SelectionDAG &DAG, const SDLoc &dl,
EVT DestVT = MVT::Other) {
if (DestVT == MVT::Other) DestVT = Op0.getValueType();
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, DestVT,
DAG.getConstant(IID, dl, MVT::i32), Op0, Op1, Op2);
}
/// BuildVSLDOI - Return a VECTOR_SHUFFLE that is a vsldoi of the specified
/// amount. The result has the specified value type.
static SDValue BuildVSLDOI(SDValue LHS, SDValue RHS, unsigned Amt, EVT VT,
SelectionDAG &DAG, const SDLoc &dl) {
// Force LHS/RHS to be the right type.
LHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, LHS);
RHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, RHS);
int Ops[16];
for (unsigned i = 0; i != 16; ++i)
Ops[i] = i + Amt;
SDValue T = DAG.getVectorShuffle(MVT::v16i8, dl, LHS, RHS, Ops);
return DAG.getNode(ISD::BITCAST, dl, VT, T);
}
/// Do we have an efficient pattern in a .td file for this node?
///
/// \param V - pointer to the BuildVectorSDNode being matched
/// \param HasDirectMove - does this subtarget have VSR <-> GPR direct moves?
///
/// There are some patterns where it is beneficial to keep a BUILD_VECTOR
/// node as a BUILD_VECTOR node rather than expanding it. The patterns where
/// the opposite is true (expansion is beneficial) are:
/// - The node builds a vector out of integers that are not 32 or 64-bits
/// - The node builds a vector out of constants
/// - The node is a "load-and-splat"
/// In all other cases, we will choose to keep the BUILD_VECTOR.
static bool haveEfficientBuildVectorPattern(BuildVectorSDNode *V,
bool HasDirectMove,
bool HasP8Vector) {
EVT VecVT = V->getValueType(0);
bool RightType = VecVT == MVT::v2f64 ||
(HasP8Vector && VecVT == MVT::v4f32) ||
(HasDirectMove && (VecVT == MVT::v2i64 || VecVT == MVT::v4i32));
if (!RightType)
return false;
bool IsSplat = true;
bool IsLoad = false;
SDValue Op0 = V->getOperand(0);
// This function is called in a block that confirms the node is not a constant
// splat. So a constant BUILD_VECTOR here means the vector is built out of
// different constants.
if (V->isConstant())
return false;
for (int i = 0, e = V->getNumOperands(); i < e; ++i) {
if (V->getOperand(i).isUndef())
return false;
// We want to expand nodes that represent load-and-splat even if the
// loaded value is a floating point truncation or conversion to int.
if (V->getOperand(i).getOpcode() == ISD::LOAD ||
(V->getOperand(i).getOpcode() == ISD::FP_ROUND &&
V->getOperand(i).getOperand(0).getOpcode() == ISD::LOAD) ||
(V->getOperand(i).getOpcode() == ISD::FP_TO_SINT &&
V->getOperand(i).getOperand(0).getOpcode() == ISD::LOAD) ||
(V->getOperand(i).getOpcode() == ISD::FP_TO_UINT &&
V->getOperand(i).getOperand(0).getOpcode() == ISD::LOAD))
IsLoad = true;
// If the operands are different or the input is not a load and has more
// uses than just this BV node, then it isn't a splat.
if (V->getOperand(i) != Op0 ||
(!IsLoad && !V->isOnlyUserOf(V->getOperand(i).getNode())))
IsSplat = false;
}
return !(IsSplat && IsLoad);
}
// Lower BITCAST(f128, (build_pair i64, i64)) to BUILD_FP128.
SDValue PPCTargetLowering::LowerBITCAST(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
SDValue Op0 = Op->getOperand(0);
if ((Op.getValueType() != MVT::f128) ||
(Op0.getOpcode() != ISD::BUILD_PAIR) ||
(Op0.getOperand(0).getValueType() != MVT::i64) ||
(Op0.getOperand(1).getValueType() != MVT::i64))
return SDValue();
return DAG.getNode(PPCISD::BUILD_FP128, dl, MVT::f128, Op0.getOperand(0),
Op0.getOperand(1));
}
static const SDValue *getNormalLoadInput(const SDValue &Op, bool &IsPermuted) {
const SDValue *InputLoad = &Op;
if (InputLoad->getOpcode() == ISD::BITCAST)
InputLoad = &InputLoad->getOperand(0);
if (InputLoad->getOpcode() == ISD::SCALAR_TO_VECTOR ||
InputLoad->getOpcode() == PPCISD::SCALAR_TO_VECTOR_PERMUTED) {
IsPermuted = InputLoad->getOpcode() == PPCISD::SCALAR_TO_VECTOR_PERMUTED;
InputLoad = &InputLoad->getOperand(0);
}
if (InputLoad->getOpcode() != ISD::LOAD)
return nullptr;
LoadSDNode *LD = cast<LoadSDNode>(*InputLoad);
return ISD::isNormalLoad(LD) ? InputLoad : nullptr;
}
// Convert the argument APFloat to a single precision APFloat if there is no
// loss in information during the conversion to single precision APFloat and the
// resulting number is not a denormal number. Return true if successful.
bool llvm::convertToNonDenormSingle(APFloat &ArgAPFloat) {
APFloat APFloatToConvert = ArgAPFloat;
bool LosesInfo = true;
APFloatToConvert.convert(APFloat::IEEEsingle(), APFloat::rmNearestTiesToEven,
&LosesInfo);
bool Success = (!LosesInfo && !APFloatToConvert.isDenormal());
if (Success)
ArgAPFloat = APFloatToConvert;
return Success;
}
// Bitcast the argument APInt to a double and convert it to a single precision
// APFloat, bitcast the APFloat to an APInt and assign it to the original
// argument if there is no loss in information during the conversion from
// double to single precision APFloat and the resulting number is not a denormal
// number. Return true if successful.
bool llvm::convertToNonDenormSingle(APInt &ArgAPInt) {
double DpValue = ArgAPInt.bitsToDouble();
APFloat APFloatDp(DpValue);
bool Success = convertToNonDenormSingle(APFloatDp);
if (Success)
ArgAPInt = APFloatDp.bitcastToAPInt();
return Success;
}
// If this is a case we can't handle, return null and let the default
// expansion code take care of it. If we CAN select this case, and if it
// selects to a single instruction, return Op. Otherwise, if we can codegen
// this case more efficiently than a constant pool load, lower it to the
// sequence of ops that should be used.
SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(Op.getNode());
assert(BVN && "Expected a BuildVectorSDNode in LowerBUILD_VECTOR");
// Check if this is a splat of a constant value.
APInt APSplatBits, APSplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
bool BVNIsConstantSplat =
BVN->isConstantSplat(APSplatBits, APSplatUndef, SplatBitSize,
HasAnyUndefs, 0, !Subtarget.isLittleEndian());
// If it is a splat of a double, check if we can shrink it to a 32 bit
// non-denormal float which when converted back to double gives us the same
// double. This is to exploit the XXSPLTIDP instruction.
// If we lose precision, we use XXSPLTI32DX.
if (BVNIsConstantSplat && (SplatBitSize == 64) &&
Subtarget.hasPrefixInstrs()) {
// Check the type first to short-circuit so we don't modify APSplatBits if
// this block isn't executed.
if ((Op->getValueType(0) == MVT::v2f64) &&
convertToNonDenormSingle(APSplatBits)) {
SDValue SplatNode = DAG.getNode(
PPCISD::XXSPLTI_SP_TO_DP, dl, MVT::v2f64,
DAG.getTargetConstant(APSplatBits.getZExtValue(), dl, MVT::i32));
return DAG.getBitcast(Op.getValueType(), SplatNode);
} else {
// We may lose precision, so we have to use XXSPLTI32DX.
uint32_t Hi =
(uint32_t)((APSplatBits.getZExtValue() & 0xFFFFFFFF00000000LL) >> 32);
uint32_t Lo =
(uint32_t)(APSplatBits.getZExtValue() & 0xFFFFFFFF);
SDValue SplatNode = DAG.getUNDEF(MVT::v2i64);
if (!Hi || !Lo)
// If either load is 0, then we should generate XXLXOR to set to 0.
SplatNode = DAG.getTargetConstant(0, dl, MVT::v2i64);
if (Hi)
SplatNode = DAG.getNode(
PPCISD::XXSPLTI32DX, dl, MVT::v2i64, SplatNode,
DAG.getTargetConstant(0, dl, MVT::i32),
DAG.getTargetConstant(Hi, dl, MVT::i32));
if (Lo)
SplatNode =
DAG.getNode(PPCISD::XXSPLTI32DX, dl, MVT::v2i64, SplatNode,
DAG.getTargetConstant(1, dl, MVT::i32),
DAG.getTargetConstant(Lo, dl, MVT::i32));
return DAG.getBitcast(Op.getValueType(), SplatNode);
}
}
if (!BVNIsConstantSplat || SplatBitSize > 32) {
bool IsPermutedLoad = false;
const SDValue *InputLoad =
getNormalLoadInput(Op.getOperand(0), IsPermutedLoad);
// Handle load-and-splat patterns as we have instructions that will do this
// in one go.
if (InputLoad && DAG.isSplatValue(Op, true)) {
LoadSDNode *LD = cast<LoadSDNode>(*InputLoad);
// We have handling for 4 and 8 byte elements.
unsigned ElementSize = LD->getMemoryVT().getScalarSizeInBits();
// Checking for a single use of this load, we have to check for vector
// width (128 bits) / ElementSize uses (since each operand of the
// BUILD_VECTOR is a separate use of the value.
unsigned NumUsesOfInputLD = 128 / ElementSize;
for (SDValue BVInOp : Op->ops())
if (BVInOp.isUndef())
NumUsesOfInputLD--;
assert(NumUsesOfInputLD > 0 && "No uses of input LD of a build_vector?");
if (InputLoad->getNode()->hasNUsesOfValue(NumUsesOfInputLD, 0) &&
((Subtarget.hasVSX() && ElementSize == 64) ||
(Subtarget.hasP9Vector() && ElementSize == 32))) {
SDValue Ops[] = {
LD->getChain(), // Chain
LD->getBasePtr(), // Ptr
DAG.getValueType(Op.getValueType()) // VT
};
SDValue LdSplt = DAG.getMemIntrinsicNode(
PPCISD::LD_SPLAT, dl, DAG.getVTList(Op.getValueType(), MVT::Other),
Ops, LD->getMemoryVT(), LD->getMemOperand());
// Replace all uses of the output chain of the original load with the
// output chain of the new load.
DAG.ReplaceAllUsesOfValueWith(InputLoad->getValue(1),
LdSplt.getValue(1));
return LdSplt;
}
}
// In 64BIT mode BUILD_VECTOR nodes that are not constant splats of up to
// 32-bits can be lowered to VSX instructions under certain conditions.
// Without VSX, there is no pattern more efficient than expanding the node.
if (Subtarget.hasVSX() && Subtarget.isPPC64() &&
haveEfficientBuildVectorPattern(BVN, Subtarget.hasDirectMove(),
Subtarget.hasP8Vector()))
return Op;
return SDValue();
}
uint64_t SplatBits = APSplatBits.getZExtValue();
uint64_t SplatUndef = APSplatUndef.getZExtValue();
unsigned SplatSize = SplatBitSize / 8;
// First, handle single instruction cases.
// All zeros?
if (SplatBits == 0) {
// Canonicalize all zero vectors to be v4i32.
if (Op.getValueType() != MVT::v4i32 || HasAnyUndefs) {
SDValue Z = DAG.getConstant(0, dl, MVT::v4i32);
Op = DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Z);
}
return Op;
}
// We have XXSPLTIW for constant splats four bytes wide.
// Given vector length is a multiple of 4, 2-byte splats can be replaced
// with 4-byte splats. We replicate the SplatBits in case of 2-byte splat to
// make a 4-byte splat element. For example: 2-byte splat of 0xABAB can be
// turned into a 4-byte splat of 0xABABABAB.
if (Subtarget.hasPrefixInstrs() && SplatSize == 2)
return getCanonicalConstSplat(SplatBits | (SplatBits << 16), SplatSize * 2,
Op.getValueType(), DAG, dl);
if (Subtarget.hasPrefixInstrs() && SplatSize == 4)
return getCanonicalConstSplat(SplatBits, SplatSize, Op.getValueType(), DAG,
dl);
// We have XXSPLTIB for constant splats one byte wide.
if (Subtarget.hasP9Vector() && SplatSize == 1)
return getCanonicalConstSplat(SplatBits, SplatSize, Op.getValueType(), DAG,
dl);
// If the sign extended value is in the range [-16,15], use VSPLTI[bhw].
int32_t SextVal= (int32_t(SplatBits << (32-SplatBitSize)) >>
(32-SplatBitSize));
if (SextVal >= -16 && SextVal <= 15)
return getCanonicalConstSplat(SextVal, SplatSize, Op.getValueType(), DAG,
dl);
// Two instruction sequences.
// If this value is in the range [-32,30] and is even, use:
// VSPLTI[bhw](val/2) + VSPLTI[bhw](val/2)
// If this value is in the range [17,31] and is odd, use:
// VSPLTI[bhw](val-16) - VSPLTI[bhw](-16)
// If this value is in the range [-31,-17] and is odd, use:
// VSPLTI[bhw](val+16) + VSPLTI[bhw](-16)
// Note the last two are three-instruction sequences.
if (SextVal >= -32 && SextVal <= 31) {
// To avoid having these optimizations undone by constant folding,
// we convert to a pseudo that will be expanded later into one of
// the above forms.
SDValue Elt = DAG.getConstant(SextVal, dl, MVT::i32);
EVT VT = (SplatSize == 1 ? MVT::v16i8 :
(SplatSize == 2 ? MVT::v8i16 : MVT::v4i32));
SDValue EltSize = DAG.getConstant(SplatSize, dl, MVT::i32);
SDValue RetVal = DAG.getNode(PPCISD::VADD_SPLAT, dl, VT, Elt, EltSize);
if (VT == Op.getValueType())
return RetVal;
else
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), RetVal);
}
// If this is 0x8000_0000 x 4, turn into vspltisw + vslw. If it is
// 0x7FFF_FFFF x 4, turn it into not(0x8000_0000). This is important
// for fneg/fabs.
if (SplatSize == 4 && SplatBits == (0x7FFFFFFF&~SplatUndef)) {
// Make -1 and vspltisw -1:
SDValue OnesV = getCanonicalConstSplat(-1, 4, MVT::v4i32, DAG, dl);
// Make the VSLW intrinsic, computing 0x8000_0000.
SDValue Res = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, OnesV,
OnesV, DAG, dl);
// xor by OnesV to invert it.
Res = DAG.getNode(ISD::XOR, dl, MVT::v4i32, Res, OnesV);
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
}
// Check to see if this is a wide variety of vsplti*, binop self cases.
static const signed char SplatCsts[] = {
-1, 1, -2, 2, -3, 3, -4, 4, -5, 5, -6, 6, -7, 7,
-8, 8, -9, 9, -10, 10, -11, 11, -12, 12, -13, 13, 14, -14, 15, -15, -16
};
for (unsigned idx = 0; idx < array_lengthof(SplatCsts); ++idx) {
// Indirect through the SplatCsts array so that we favor 'vsplti -1' for
// cases which are ambiguous (e.g. formation of 0x8000_0000). 'vsplti -1'
int i = SplatCsts[idx];
// Figure out what shift amount will be used by altivec if shifted by i in
// this splat size.
unsigned TypeShiftAmt = i & (SplatBitSize-1);
// vsplti + shl self.
if (SextVal == (int)((unsigned)i << TypeShiftAmt)) {
SDValue Res = getCanonicalConstSplat(i, SplatSize, MVT::Other, DAG, dl);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vslb, Intrinsic::ppc_altivec_vslh, 0,
Intrinsic::ppc_altivec_vslw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
}
// vsplti + srl self.
if (SextVal == (int)((unsigned)i >> TypeShiftAmt)) {
SDValue Res = getCanonicalConstSplat(i, SplatSize, MVT::Other, DAG, dl);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vsrb, Intrinsic::ppc_altivec_vsrh, 0,
Intrinsic::ppc_altivec_vsrw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
}
// vsplti + rol self.
if (SextVal == (int)(((unsigned)i << TypeShiftAmt) |
((unsigned)i >> (SplatBitSize-TypeShiftAmt)))) {
SDValue Res = getCanonicalConstSplat(i, SplatSize, MVT::Other, DAG, dl);
static const unsigned IIDs[] = { // Intrinsic to use for each size.
Intrinsic::ppc_altivec_vrlb, Intrinsic::ppc_altivec_vrlh, 0,
Intrinsic::ppc_altivec_vrlw
};
Res = BuildIntrinsicOp(IIDs[SplatSize-1], Res, Res, DAG, dl);
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Res);
}
// t = vsplti c, result = vsldoi t, t, 1
if (SextVal == (int)(((unsigned)i << 8) | (i < 0 ? 0xFF : 0))) {
SDValue T = getCanonicalConstSplat(i, SplatSize, MVT::v16i8, DAG, dl);
unsigned Amt = Subtarget.isLittleEndian() ? 15 : 1;
return BuildVSLDOI(T, T, Amt, Op.getValueType(), DAG, dl);
}
// t = vsplti c, result = vsldoi t, t, 2
if (SextVal == (int)(((unsigned)i << 16) | (i < 0 ? 0xFFFF : 0))) {
SDValue T = getCanonicalConstSplat(i, SplatSize, MVT::v16i8, DAG, dl);
unsigned Amt = Subtarget.isLittleEndian() ? 14 : 2;
return BuildVSLDOI(T, T, Amt, Op.getValueType(), DAG, dl);
}
// t = vsplti c, result = vsldoi t, t, 3
if (SextVal == (int)(((unsigned)i << 24) | (i < 0 ? 0xFFFFFF : 0))) {
SDValue T = getCanonicalConstSplat(i, SplatSize, MVT::v16i8, DAG, dl);
unsigned Amt = Subtarget.isLittleEndian() ? 13 : 3;
return BuildVSLDOI(T, T, Amt, Op.getValueType(), DAG, dl);
}
}
return SDValue();
}
/// GeneratePerfectShuffle - Given an entry in the perfect-shuffle table, emit
/// the specified operations to build the shuffle.
static SDValue GeneratePerfectShuffle(unsigned PFEntry, SDValue LHS,
SDValue RHS, SelectionDAG &DAG,
const SDLoc &dl) {
unsigned OpNum = (PFEntry >> 26) & 0x0F;
unsigned LHSID = (PFEntry >> 13) & ((1 << 13)-1);
unsigned RHSID = (PFEntry >> 0) & ((1 << 13)-1);
enum {
OP_COPY = 0, // Copy, used for things like <u,u,u,3> to say it is <0,1,2,3>
OP_VMRGHW,
OP_VMRGLW,
OP_VSPLTISW0,
OP_VSPLTISW1,
OP_VSPLTISW2,
OP_VSPLTISW3,
OP_VSLDOI4,
OP_VSLDOI8,
OP_VSLDOI12
};
if (OpNum == OP_COPY) {
if (LHSID == (1*9+2)*9+3) return LHS;
assert(LHSID == ((4*9+5)*9+6)*9+7 && "Illegal OP_COPY!");
return RHS;
}
SDValue OpLHS, OpRHS;
OpLHS = GeneratePerfectShuffle(PerfectShuffleTable[LHSID], LHS, RHS, DAG, dl);
OpRHS = GeneratePerfectShuffle(PerfectShuffleTable[RHSID], LHS, RHS, DAG, dl);
int ShufIdxs[16];
switch (OpNum) {
default: llvm_unreachable("Unknown i32 permute!");
case OP_VMRGHW:
ShufIdxs[ 0] = 0; ShufIdxs[ 1] = 1; ShufIdxs[ 2] = 2; ShufIdxs[ 3] = 3;
ShufIdxs[ 4] = 16; ShufIdxs[ 5] = 17; ShufIdxs[ 6] = 18; ShufIdxs[ 7] = 19;
ShufIdxs[ 8] = 4; ShufIdxs[ 9] = 5; ShufIdxs[10] = 6; ShufIdxs[11] = 7;
ShufIdxs[12] = 20; ShufIdxs[13] = 21; ShufIdxs[14] = 22; ShufIdxs[15] = 23;
break;
case OP_VMRGLW:
ShufIdxs[ 0] = 8; ShufIdxs[ 1] = 9; ShufIdxs[ 2] = 10; ShufIdxs[ 3] = 11;
ShufIdxs[ 4] = 24; ShufIdxs[ 5] = 25; ShufIdxs[ 6] = 26; ShufIdxs[ 7] = 27;
ShufIdxs[ 8] = 12; ShufIdxs[ 9] = 13; ShufIdxs[10] = 14; ShufIdxs[11] = 15;
ShufIdxs[12] = 28; ShufIdxs[13] = 29; ShufIdxs[14] = 30; ShufIdxs[15] = 31;
break;
case OP_VSPLTISW0:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+0;
break;
case OP_VSPLTISW1:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+4;
break;
case OP_VSPLTISW2:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+8;
break;
case OP_VSPLTISW3:
for (unsigned i = 0; i != 16; ++i)
ShufIdxs[i] = (i&3)+12;
break;
case OP_VSLDOI4:
return BuildVSLDOI(OpLHS, OpRHS, 4, OpLHS.getValueType(), DAG, dl);
case OP_VSLDOI8:
return BuildVSLDOI(OpLHS, OpRHS, 8, OpLHS.getValueType(), DAG, dl);
case OP_VSLDOI12:
return BuildVSLDOI(OpLHS, OpRHS, 12, OpLHS.getValueType(), DAG, dl);
}
EVT VT = OpLHS.getValueType();
OpLHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OpLHS);
OpRHS = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OpRHS);
SDValue T = DAG.getVectorShuffle(MVT::v16i8, dl, OpLHS, OpRHS, ShufIdxs);
return DAG.getNode(ISD::BITCAST, dl, VT, T);
}
/// lowerToVINSERTB - Return the SDValue if this VECTOR_SHUFFLE can be handled
/// by the VINSERTB instruction introduced in ISA 3.0, else just return default
/// SDValue.
SDValue PPCTargetLowering::lowerToVINSERTB(ShuffleVectorSDNode *N,
SelectionDAG &DAG) const {
const unsigned BytesInVector = 16;
bool IsLE = Subtarget.isLittleEndian();
SDLoc dl(N);
SDValue V1 = N->getOperand(0);
SDValue V2 = N->getOperand(1);
unsigned ShiftElts = 0, InsertAtByte = 0;
bool Swap = false;
// Shifts required to get the byte we want at element 7.
unsigned LittleEndianShifts[] = {8, 7, 6, 5, 4, 3, 2, 1,
0, 15, 14, 13, 12, 11, 10, 9};
unsigned BigEndianShifts[] = {9, 10, 11, 12, 13, 14, 15, 0,
1, 2, 3, 4, 5, 6, 7, 8};
ArrayRef<int> Mask = N->getMask();
int OriginalOrder[] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
// For each mask element, find out if we're just inserting something
// from V2 into V1 or vice versa.
// Possible permutations inserting an element from V2 into V1:
// X, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
// 0, X, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
// ...
// 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, X
// Inserting from V1 into V2 will be similar, except mask range will be
// [16,31].
bool FoundCandidate = false;
// If both vector operands for the shuffle are the same vector, the mask
// will contain only elements from the first one and the second one will be
// undef.
unsigned VINSERTBSrcElem = IsLE ? 8 : 7;
// Go through the mask of half-words to find an element that's being moved
// from one vector to the other.
for (unsigned i = 0; i < BytesInVector; ++i) {
unsigned CurrentElement = Mask[i];
// If 2nd operand is undefined, we should only look for element 7 in the
// Mask.
if (V2.isUndef() && CurrentElement != VINSERTBSrcElem)
continue;
bool OtherElementsInOrder = true;
// Examine the other elements in the Mask to see if they're in original
// order.
for (unsigned j = 0; j < BytesInVector; ++j) {
if (j == i)
continue;
// If CurrentElement is from V1 [0,15], then we the rest of the Mask to be
// from V2 [16,31] and vice versa. Unless the 2nd operand is undefined,
// in which we always assume we're always picking from the 1st operand.
int MaskOffset =
(!V2.isUndef() && CurrentElement < BytesInVector) ? BytesInVector : 0;
if (Mask[j] != OriginalOrder[j] + MaskOffset) {
OtherElementsInOrder = false;
break;
}
}
// If other elements are in original order, we record the number of shifts
// we need to get the element we want into element 7. Also record which byte
// in the vector we should insert into.
if (OtherElementsInOrder) {
// If 2nd operand is undefined, we assume no shifts and no swapping.
if (V2.isUndef()) {
ShiftElts = 0;
Swap = false;
} else {
// Only need the last 4-bits for shifts because operands will be swapped if CurrentElement is >= 2^4.
ShiftElts = IsLE ? LittleEndianShifts[CurrentElement & 0xF]
: BigEndianShifts[CurrentElement & 0xF];
Swap = CurrentElement < BytesInVector;
}
InsertAtByte = IsLE ? BytesInVector - (i + 1) : i;
FoundCandidate = true;
break;
}
}
if (!FoundCandidate)
return SDValue();
// Candidate found, construct the proper SDAG sequence with VINSERTB,
// optionally with VECSHL if shift is required.
if (Swap)
std::swap(V1, V2);
if (V2.isUndef())
V2 = V1;
if (ShiftElts) {
SDValue Shl = DAG.getNode(PPCISD::VECSHL, dl, MVT::v16i8, V2, V2,
DAG.getConstant(ShiftElts, dl, MVT::i32));
return DAG.getNode(PPCISD::VECINSERT, dl, MVT::v16i8, V1, Shl,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
}
return DAG.getNode(PPCISD::VECINSERT, dl, MVT::v16i8, V1, V2,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
}
/// lowerToVINSERTH - Return the SDValue if this VECTOR_SHUFFLE can be handled
/// by the VINSERTH instruction introduced in ISA 3.0, else just return default
/// SDValue.
SDValue PPCTargetLowering::lowerToVINSERTH(ShuffleVectorSDNode *N,
SelectionDAG &DAG) const {
const unsigned NumHalfWords = 8;
const unsigned BytesInVector = NumHalfWords * 2;
// Check that the shuffle is on half-words.
if (!isNByteElemShuffleMask(N, 2, 1))
return SDValue();
bool IsLE = Subtarget.isLittleEndian();
SDLoc dl(N);
SDValue V1 = N->getOperand(0);
SDValue V2 = N->getOperand(1);
unsigned ShiftElts = 0, InsertAtByte = 0;
bool Swap = false;
// Shifts required to get the half-word we want at element 3.
unsigned LittleEndianShifts[] = {4, 3, 2, 1, 0, 7, 6, 5};
unsigned BigEndianShifts[] = {5, 6, 7, 0, 1, 2, 3, 4};
uint32_t Mask = 0;
uint32_t OriginalOrderLow = 0x1234567;
uint32_t OriginalOrderHigh = 0x89ABCDEF;
// Now we look at mask elements 0,2,4,6,8,10,12,14. Pack the mask into a
// 32-bit space, only need 4-bit nibbles per element.
for (unsigned i = 0; i < NumHalfWords; ++i) {
unsigned MaskShift = (NumHalfWords - 1 - i) * 4;
Mask |= ((uint32_t)(N->getMaskElt(i * 2) / 2) << MaskShift);
}
// For each mask element, find out if we're just inserting something
// from V2 into V1 or vice versa. Possible permutations inserting an element
// from V2 into V1:
// X, 1, 2, 3, 4, 5, 6, 7
// 0, X, 2, 3, 4, 5, 6, 7
// 0, 1, X, 3, 4, 5, 6, 7
// 0, 1, 2, X, 4, 5, 6, 7
// 0, 1, 2, 3, X, 5, 6, 7
// 0, 1, 2, 3, 4, X, 6, 7
// 0, 1, 2, 3, 4, 5, X, 7
// 0, 1, 2, 3, 4, 5, 6, X
// Inserting from V1 into V2 will be similar, except mask range will be [8,15].
bool FoundCandidate = false;
// Go through the mask of half-words to find an element that's being moved
// from one vector to the other.
for (unsigned i = 0; i < NumHalfWords; ++i) {
unsigned MaskShift = (NumHalfWords - 1 - i) * 4;
uint32_t MaskOneElt = (Mask >> MaskShift) & 0xF;
uint32_t MaskOtherElts = ~(0xF << MaskShift);
uint32_t TargetOrder = 0x0;
// If both vector operands for the shuffle are the same vector, the mask
// will contain only elements from the first one and the second one will be
// undef.
if (V2.isUndef()) {
ShiftElts = 0;
unsigned VINSERTHSrcElem = IsLE ? 4 : 3;
TargetOrder = OriginalOrderLow;
Swap = false;
// Skip if not the correct element or mask of other elements don't equal
// to our expected order.
if (MaskOneElt == VINSERTHSrcElem &&
(Mask & MaskOtherElts) == (TargetOrder & MaskOtherElts)) {
InsertAtByte = IsLE ? BytesInVector - (i + 1) * 2 : i * 2;
FoundCandidate = true;
break;
}
} else { // If both operands are defined.
// Target order is [8,15] if the current mask is between [0,7].
TargetOrder =
(MaskOneElt < NumHalfWords) ? OriginalOrderHigh : OriginalOrderLow;
// Skip if mask of other elements don't equal our expected order.
if ((Mask & MaskOtherElts) == (TargetOrder & MaskOtherElts)) {
// We only need the last 3 bits for the number of shifts.
ShiftElts = IsLE ? LittleEndianShifts[MaskOneElt & 0x7]
: BigEndianShifts[MaskOneElt & 0x7];
InsertAtByte = IsLE ? BytesInVector - (i + 1) * 2 : i * 2;
Swap = MaskOneElt < NumHalfWords;
FoundCandidate = true;
break;
}
}
}
if (!FoundCandidate)
return SDValue();
// Candidate found, construct the proper SDAG sequence with VINSERTH,
// optionally with VECSHL if shift is required.
if (Swap)
std::swap(V1, V2);
if (V2.isUndef())
V2 = V1;
SDValue Conv1 = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
if (ShiftElts) {
// Double ShiftElts because we're left shifting on v16i8 type.
SDValue Shl = DAG.getNode(PPCISD::VECSHL, dl, MVT::v16i8, V2, V2,
DAG.getConstant(2 * ShiftElts, dl, MVT::i32));
SDValue Conv2 = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, Shl);
SDValue Ins = DAG.getNode(PPCISD::VECINSERT, dl, MVT::v8i16, Conv1, Conv2,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Ins);
}
SDValue Conv2 = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V2);
SDValue Ins = DAG.getNode(PPCISD::VECINSERT, dl, MVT::v8i16, Conv1, Conv2,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Ins);
}
/// lowerToXXSPLTI32DX - Return the SDValue if this VECTOR_SHUFFLE can be
/// handled by the XXSPLTI32DX instruction introduced in ISA 3.1, otherwise
/// return the default SDValue.
SDValue PPCTargetLowering::lowerToXXSPLTI32DX(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG) const {
// The LHS and RHS may be bitcasts to v16i8 as we canonicalize shuffles
// to v16i8. Peek through the bitcasts to get the actual operands.
SDValue LHS = peekThroughBitcasts(SVN->getOperand(0));
SDValue RHS = peekThroughBitcasts(SVN->getOperand(1));
auto ShuffleMask = SVN->getMask();
SDValue VecShuffle(SVN, 0);
SDLoc DL(SVN);
// Check that we have a four byte shuffle.
if (!isNByteElemShuffleMask(SVN, 4, 1))
return SDValue();
// Canonicalize the RHS being a BUILD_VECTOR when lowering to xxsplti32dx.
if (RHS->getOpcode() != ISD::BUILD_VECTOR) {
std::swap(LHS, RHS);
VecShuffle = DAG.getCommutedVectorShuffle(*SVN);
ShuffleMask = cast<ShuffleVectorSDNode>(VecShuffle)->getMask();
}
// Ensure that the RHS is a vector of constants.
BuildVectorSDNode *BVN = dyn_cast<BuildVectorSDNode>(RHS.getNode());
if (!BVN)
return SDValue();
// Check if RHS is a splat of 4-bytes (or smaller).
APInt APSplatValue, APSplatUndef;
unsigned SplatBitSize;
bool HasAnyUndefs;
if (!BVN->isConstantSplat(APSplatValue, APSplatUndef, SplatBitSize,
HasAnyUndefs, 0, !Subtarget.isLittleEndian()) ||
SplatBitSize > 32)
return SDValue();
// Check that the shuffle mask matches the semantics of XXSPLTI32DX.
// The instruction splats a constant C into two words of the source vector
// producing { C, Unchanged, C, Unchanged } or { Unchanged, C, Unchanged, C }.
// Thus we check that the shuffle mask is the equivalent of
// <0, [4-7], 2, [4-7]> or <[4-7], 1, [4-7], 3> respectively.
// Note: the check above of isNByteElemShuffleMask() ensures that the bytes
// within each word are consecutive, so we only need to check the first byte.
SDValue Index;
bool IsLE = Subtarget.isLittleEndian();
if ((ShuffleMask[0] == 0 && ShuffleMask[8] == 8) &&
(ShuffleMask[4] % 4 == 0 && ShuffleMask[12] % 4 == 0 &&
ShuffleMask[4] > 15 && ShuffleMask[12] > 15))
Index = DAG.getTargetConstant(IsLE ? 0 : 1, DL, MVT::i32);
else if ((ShuffleMask[4] == 4 && ShuffleMask[12] == 12) &&
(ShuffleMask[0] % 4 == 0 && ShuffleMask[8] % 4 == 0 &&
ShuffleMask[0] > 15 && ShuffleMask[8] > 15))
Index = DAG.getTargetConstant(IsLE ? 1 : 0, DL, MVT::i32);
else
return SDValue();
// If the splat is narrower than 32-bits, we need to get the 32-bit value
// for XXSPLTI32DX.
unsigned SplatVal = APSplatValue.getZExtValue();
for (; SplatBitSize < 32; SplatBitSize <<= 1)
SplatVal |= (SplatVal << SplatBitSize);
SDValue SplatNode = DAG.getNode(
PPCISD::XXSPLTI32DX, DL, MVT::v2i64, DAG.getBitcast(MVT::v2i64, LHS),
Index, DAG.getTargetConstant(SplatVal, DL, MVT::i32));
return DAG.getNode(ISD::BITCAST, DL, MVT::v16i8, SplatNode);
}
/// LowerROTL - Custom lowering for ROTL(v1i128) to vector_shuffle(v16i8).
/// We lower ROTL(v1i128) to vector_shuffle(v16i8) only if shift amount is
/// a multiple of 8. Otherwise convert it to a scalar rotation(i128)
/// i.e (or (shl x, C1), (srl x, 128-C1)).
SDValue PPCTargetLowering::LowerROTL(SDValue Op, SelectionDAG &DAG) const {
assert(Op.getOpcode() == ISD::ROTL && "Should only be called for ISD::ROTL");
assert(Op.getValueType() == MVT::v1i128 &&
"Only set v1i128 as custom, other type shouldn't reach here!");
SDLoc dl(Op);
SDValue N0 = peekThroughBitcasts(Op.getOperand(0));
SDValue N1 = peekThroughBitcasts(Op.getOperand(1));
unsigned SHLAmt = N1.getConstantOperandVal(0);
if (SHLAmt % 8 == 0) {
SmallVector<int, 16> Mask(16, 0);
std::iota(Mask.begin(), Mask.end(), 0);
std::rotate(Mask.begin(), Mask.begin() + SHLAmt / 8, Mask.end());
if (SDValue Shuffle =
DAG.getVectorShuffle(MVT::v16i8, dl, DAG.getBitcast(MVT::v16i8, N0),
DAG.getUNDEF(MVT::v16i8), Mask))
return DAG.getNode(ISD::BITCAST, dl, MVT::v1i128, Shuffle);
}
SDValue ArgVal = DAG.getBitcast(MVT::i128, N0);
SDValue SHLOp = DAG.getNode(ISD::SHL, dl, MVT::i128, ArgVal,
DAG.getConstant(SHLAmt, dl, MVT::i32));
SDValue SRLOp = DAG.getNode(ISD::SRL, dl, MVT::i128, ArgVal,
DAG.getConstant(128 - SHLAmt, dl, MVT::i32));
SDValue OROp = DAG.getNode(ISD::OR, dl, MVT::i128, SHLOp, SRLOp);
return DAG.getNode(ISD::BITCAST, dl, MVT::v1i128, OROp);
}
/// LowerVECTOR_SHUFFLE - Return the code we lower for VECTOR_SHUFFLE. If this
/// is a shuffle we can handle in a single instruction, return it. Otherwise,
/// return the code it can be lowered into. Worst case, it can always be
/// lowered into a vperm.
SDValue PPCTargetLowering::LowerVECTOR_SHUFFLE(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
// Any nodes that were combined in the target-independent combiner prior
// to vector legalization will not be sent to the target combine. Try to
// combine it here.
if (SDValue NewShuffle = combineVectorShuffle(SVOp, DAG)) {
if (!isa<ShuffleVectorSDNode>(NewShuffle))
return NewShuffle;
Op = NewShuffle;
SVOp = cast<ShuffleVectorSDNode>(Op);
V1 = Op.getOperand(0);
V2 = Op.getOperand(1);
}
EVT VT = Op.getValueType();
bool isLittleEndian = Subtarget.isLittleEndian();
unsigned ShiftElts, InsertAtByte;
bool Swap = false;
// If this is a load-and-splat, we can do that with a single instruction
// in some cases. However if the load has multiple uses, we don't want to
// combine it because that will just produce multiple loads.
bool IsPermutedLoad = false;
const SDValue *InputLoad = getNormalLoadInput(V1, IsPermutedLoad);
if (InputLoad && Subtarget.hasVSX() && V2.isUndef() &&
(PPC::isSplatShuffleMask(SVOp, 4) || PPC::isSplatShuffleMask(SVOp, 8)) &&
InputLoad->hasOneUse()) {
bool IsFourByte = PPC::isSplatShuffleMask(SVOp, 4);
int SplatIdx =
PPC::getSplatIdxForPPCMnemonics(SVOp, IsFourByte ? 4 : 8, DAG);
// The splat index for permuted loads will be in the left half of the vector
// which is strictly wider than the loaded value by 8 bytes. So we need to
// adjust the splat index to point to the correct address in memory.
if (IsPermutedLoad) {
assert(isLittleEndian && "Unexpected permuted load on big endian target");
SplatIdx += IsFourByte ? 2 : 1;
assert((SplatIdx < (IsFourByte ? 4 : 2)) &&
"Splat of a value outside of the loaded memory");
}
LoadSDNode *LD = cast<LoadSDNode>(*InputLoad);
// For 4-byte load-and-splat, we need Power9.
if ((IsFourByte && Subtarget.hasP9Vector()) || !IsFourByte) {
uint64_t Offset = 0;
if (IsFourByte)
Offset = isLittleEndian ? (3 - SplatIdx) * 4 : SplatIdx * 4;
else
Offset = isLittleEndian ? (1 - SplatIdx) * 8 : SplatIdx * 8;
SDValue BasePtr = LD->getBasePtr();
if (Offset != 0)
BasePtr = DAG.getNode(ISD::ADD, dl, getPointerTy(DAG.getDataLayout()),
BasePtr, DAG.getIntPtrConstant(Offset, dl));
SDValue Ops[] = {
LD->getChain(), // Chain
BasePtr, // BasePtr
DAG.getValueType(Op.getValueType()) // VT
};
SDVTList VTL =
DAG.getVTList(IsFourByte ? MVT::v4i32 : MVT::v2i64, MVT::Other);
SDValue LdSplt =
DAG.getMemIntrinsicNode(PPCISD::LD_SPLAT, dl, VTL,
Ops, LD->getMemoryVT(), LD->getMemOperand());
DAG.ReplaceAllUsesOfValueWith(InputLoad->getValue(1), LdSplt.getValue(1));
if (LdSplt.getValueType() != SVOp->getValueType(0))
LdSplt = DAG.getBitcast(SVOp->getValueType(0), LdSplt);
return LdSplt;
}
}
if (Subtarget.hasP9Vector() &&
PPC::isXXINSERTWMask(SVOp, ShiftElts, InsertAtByte, Swap,
isLittleEndian)) {
if (Swap)
std::swap(V1, V2);
SDValue Conv1 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V1);
SDValue Conv2 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V2);
if (ShiftElts) {
SDValue Shl = DAG.getNode(PPCISD::VECSHL, dl, MVT::v4i32, Conv2, Conv2,
DAG.getConstant(ShiftElts, dl, MVT::i32));
SDValue Ins = DAG.getNode(PPCISD::VECINSERT, dl, MVT::v4i32, Conv1, Shl,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Ins);
}
SDValue Ins = DAG.getNode(PPCISD::VECINSERT, dl, MVT::v4i32, Conv1, Conv2,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Ins);
}
if (Subtarget.hasPrefixInstrs()) {
SDValue SplatInsertNode;
if ((SplatInsertNode = lowerToXXSPLTI32DX(SVOp, DAG)))
return SplatInsertNode;
}
if (Subtarget.hasP9Altivec()) {
SDValue NewISDNode;
if ((NewISDNode = lowerToVINSERTH(SVOp, DAG)))
return NewISDNode;
if ((NewISDNode = lowerToVINSERTB(SVOp, DAG)))
return NewISDNode;
}
if (Subtarget.hasVSX() &&
PPC::isXXSLDWIShuffleMask(SVOp, ShiftElts, Swap, isLittleEndian)) {
if (Swap)
std::swap(V1, V2);
SDValue Conv1 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V1);
SDValue Conv2 =
DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V2.isUndef() ? V1 : V2);
SDValue Shl = DAG.getNode(PPCISD::VECSHL, dl, MVT::v4i32, Conv1, Conv2,
DAG.getConstant(ShiftElts, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Shl);
}
if (Subtarget.hasVSX() &&
PPC::isXXPERMDIShuffleMask(SVOp, ShiftElts, Swap, isLittleEndian)) {
if (Swap)
std::swap(V1, V2);
SDValue Conv1 = DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, V1);
SDValue Conv2 =
DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, V2.isUndef() ? V1 : V2);
SDValue PermDI = DAG.getNode(PPCISD::XXPERMDI, dl, MVT::v2i64, Conv1, Conv2,
DAG.getConstant(ShiftElts, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, PermDI);
}
if (Subtarget.hasP9Vector()) {
if (PPC::isXXBRHShuffleMask(SVOp)) {
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
SDValue ReveHWord = DAG.getNode(ISD::BSWAP, dl, MVT::v8i16, Conv);
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, ReveHWord);
} else if (PPC::isXXBRWShuffleMask(SVOp)) {
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V1);
SDValue ReveWord = DAG.getNode(ISD::BSWAP, dl, MVT::v4i32, Conv);
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, ReveWord);
} else if (PPC::isXXBRDShuffleMask(SVOp)) {
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, V1);
SDValue ReveDWord = DAG.getNode(ISD::BSWAP, dl, MVT::v2i64, Conv);
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, ReveDWord);
} else if (PPC::isXXBRQShuffleMask(SVOp)) {
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v1i128, V1);
SDValue ReveQWord = DAG.getNode(ISD::BSWAP, dl, MVT::v1i128, Conv);
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, ReveQWord);
}
}
if (Subtarget.hasVSX()) {
if (V2.isUndef() && PPC::isSplatShuffleMask(SVOp, 4)) {
int SplatIdx = PPC::getSplatIdxForPPCMnemonics(SVOp, 4, DAG);
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V1);
SDValue Splat = DAG.getNode(PPCISD::XXSPLT, dl, MVT::v4i32, Conv,
DAG.getConstant(SplatIdx, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Splat);
}
// Left shifts of 8 bytes are actually swaps. Convert accordingly.
if (V2.isUndef() && PPC::isVSLDOIShuffleMask(SVOp, 1, DAG) == 8) {
SDValue Conv = DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, V1);
SDValue Swap = DAG.getNode(PPCISD::SWAP_NO_CHAIN, dl, MVT::v2f64, Conv);
return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, Swap);
}
}
// Cases that are handled by instructions that take permute immediates
// (such as vsplt*) should be left as VECTOR_SHUFFLE nodes so they can be
// selected by the instruction selector.
if (V2.isUndef()) {
if (PPC::isSplatShuffleMask(SVOp, 1) ||
PPC::isSplatShuffleMask(SVOp, 2) ||
PPC::isSplatShuffleMask(SVOp, 4) ||
PPC::isVPKUWUMShuffleMask(SVOp, 1, DAG) ||
PPC::isVPKUHUMShuffleMask(SVOp, 1, DAG) ||
PPC::isVSLDOIShuffleMask(SVOp, 1, DAG) != -1 ||
PPC::isVMRGLShuffleMask(SVOp, 1, 1, DAG) ||
PPC::isVMRGLShuffleMask(SVOp, 2, 1, DAG) ||
PPC::isVMRGLShuffleMask(SVOp, 4, 1, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 1, 1, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 2, 1, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 4, 1, DAG) ||
(Subtarget.hasP8Altivec() && (
PPC::isVPKUDUMShuffleMask(SVOp, 1, DAG) ||
PPC::isVMRGEOShuffleMask(SVOp, true, 1, DAG) ||
PPC::isVMRGEOShuffleMask(SVOp, false, 1, DAG)))) {
return Op;
}
}
// Altivec has a variety of "shuffle immediates" that take two vector inputs
// and produce a fixed permutation. If any of these match, do not lower to
// VPERM.
unsigned int ShuffleKind = isLittleEndian ? 2 : 0;
if (PPC::isVPKUWUMShuffleMask(SVOp, ShuffleKind, DAG) ||
PPC::isVPKUHUMShuffleMask(SVOp, ShuffleKind, DAG) ||
PPC::isVSLDOIShuffleMask(SVOp, ShuffleKind, DAG) != -1 ||
PPC::isVMRGLShuffleMask(SVOp, 1, ShuffleKind, DAG) ||
PPC::isVMRGLShuffleMask(SVOp, 2, ShuffleKind, DAG) ||
PPC::isVMRGLShuffleMask(SVOp, 4, ShuffleKind, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 1, ShuffleKind, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 2, ShuffleKind, DAG) ||
PPC::isVMRGHShuffleMask(SVOp, 4, ShuffleKind, DAG) ||
(Subtarget.hasP8Altivec() && (
PPC::isVPKUDUMShuffleMask(SVOp, ShuffleKind, DAG) ||
PPC::isVMRGEOShuffleMask(SVOp, true, ShuffleKind, DAG) ||
PPC::isVMRGEOShuffleMask(SVOp, false, ShuffleKind, DAG))))
return Op;
// Check to see if this is a shuffle of 4-byte values. If so, we can use our
// perfect shuffle table to emit an optimal matching sequence.
ArrayRef<int> PermMask = SVOp->getMask();
unsigned PFIndexes[4];
bool isFourElementShuffle = true;
for (unsigned i = 0; i != 4 && isFourElementShuffle; ++i) { // Element number
unsigned EltNo = 8; // Start out undef.
for (unsigned j = 0; j != 4; ++j) { // Intra-element byte.
if (PermMask[i*4+j] < 0)
continue; // Undef, ignore it.
unsigned ByteSource = PermMask[i*4+j];
if ((ByteSource & 3) != j) {
isFourElementShuffle = false;
break;
}
if (EltNo == 8) {
EltNo = ByteSource/4;
} else if (EltNo != ByteSource/4) {
isFourElementShuffle = false;
break;
}
}
PFIndexes[i] = EltNo;
}
// If this shuffle can be expressed as a shuffle of 4-byte elements, use the
// perfect shuffle vector to determine if it is cost effective to do this as
// discrete instructions, or whether we should use a vperm.
// For now, we skip this for little endian until such time as we have a
// little-endian perfect shuffle table.
if (isFourElementShuffle && !isLittleEndian) {
// Compute the index in the perfect shuffle table.
unsigned PFTableIndex =
PFIndexes[0]*9*9*9+PFIndexes[1]*9*9+PFIndexes[2]*9+PFIndexes[3];
unsigned PFEntry = PerfectShuffleTable[PFTableIndex];
unsigned Cost = (PFEntry >> 30);
// Determining when to avoid vperm is tricky. Many things affect the cost
// of vperm, particularly how many times the perm mask needs to be computed.
// For example, if the perm mask can be hoisted out of a loop or is already
// used (perhaps because there are multiple permutes with the same shuffle
// mask?) the vperm has a cost of 1. OTOH, hoisting the permute mask out of
// the loop requires an extra register.
//
// As a compromise, we only emit discrete instructions if the shuffle can be
// generated in 3 or fewer operations. When we have loop information
// available, if this block is within a loop, we should avoid using vperm
// for 3-operation perms and use a constant pool load instead.
if (Cost < 3)
return GeneratePerfectShuffle(PFEntry, V1, V2, DAG, dl);
}
// Lower this to a VPERM(V1, V2, V3) expression, where V3 is a constant
// vector that will get spilled to the constant pool.
if (V2.isUndef()) V2 = V1;
// The SHUFFLE_VECTOR mask is almost exactly what we want for vperm, except
// that it is in input element units, not in bytes. Convert now.
// For little endian, the order of the input vectors is reversed, and
// the permutation mask is complemented with respect to 31. This is
// necessary to produce proper semantics with the big-endian-biased vperm
// instruction.
EVT EltVT = V1.getValueType().getVectorElementType();
unsigned BytesPerElement = EltVT.getSizeInBits()/8;
SmallVector<SDValue, 16> ResultMask;
for (unsigned i = 0, e = VT.getVectorNumElements(); i != e; ++i) {
unsigned SrcElt = PermMask[i] < 0 ? 0 : PermMask[i];
for (unsigned j = 0; j != BytesPerElement; ++j)
if (isLittleEndian)
ResultMask.push_back(DAG.getConstant(31 - (SrcElt*BytesPerElement + j),
dl, MVT::i32));
else
ResultMask.push_back(DAG.getConstant(SrcElt*BytesPerElement + j, dl,
MVT::i32));
}
ShufflesHandledWithVPERM++;
SDValue VPermMask = DAG.getBuildVector(MVT::v16i8, dl, ResultMask);
LLVM_DEBUG(dbgs() << "Emitting a VPERM for the following shuffle:\n");
LLVM_DEBUG(SVOp->dump());
LLVM_DEBUG(dbgs() << "With the following permute control vector:\n");
LLVM_DEBUG(VPermMask.dump());
if (isLittleEndian)
return DAG.getNode(PPCISD::VPERM, dl, V1.getValueType(),
V2, V1, VPermMask);
else
return DAG.getNode(PPCISD::VPERM, dl, V1.getValueType(),
V1, V2, VPermMask);
}
/// getVectorCompareInfo - Given an intrinsic, return false if it is not a
/// vector comparison. If it is, return true and fill in Opc/isDot with
/// information about the intrinsic.
static bool getVectorCompareInfo(SDValue Intrin, int &CompareOpc,
bool &isDot, const PPCSubtarget &Subtarget) {
unsigned IntrinsicID =
cast<ConstantSDNode>(Intrin.getOperand(0))->getZExtValue();
CompareOpc = -1;
isDot = false;
switch (IntrinsicID) {
default:
return false;
// Comparison predicates.
case Intrinsic::ppc_altivec_vcmpbfp_p:
CompareOpc = 966;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpeqfp_p:
CompareOpc = 198;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpequb_p:
CompareOpc = 6;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpequh_p:
CompareOpc = 70;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpequw_p:
CompareOpc = 134;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpequd_p:
if (Subtarget.hasP8Altivec()) {
CompareOpc = 199;
isDot = true;
} else
return false;
break;
case Intrinsic::ppc_altivec_vcmpneb_p:
case Intrinsic::ppc_altivec_vcmpneh_p:
case Intrinsic::ppc_altivec_vcmpnew_p:
case Intrinsic::ppc_altivec_vcmpnezb_p:
case Intrinsic::ppc_altivec_vcmpnezh_p:
case Intrinsic::ppc_altivec_vcmpnezw_p:
if (Subtarget.hasP9Altivec()) {
switch (IntrinsicID) {
default:
llvm_unreachable("Unknown comparison intrinsic.");
case Intrinsic::ppc_altivec_vcmpneb_p:
CompareOpc = 7;
break;
case Intrinsic::ppc_altivec_vcmpneh_p:
CompareOpc = 71;
break;
case Intrinsic::ppc_altivec_vcmpnew_p:
CompareOpc = 135;
break;
case Intrinsic::ppc_altivec_vcmpnezb_p:
CompareOpc = 263;
break;
case Intrinsic::ppc_altivec_vcmpnezh_p:
CompareOpc = 327;
break;
case Intrinsic::ppc_altivec_vcmpnezw_p:
CompareOpc = 391;
break;
}
isDot = true;
} else
return false;
break;
case Intrinsic::ppc_altivec_vcmpgefp_p:
CompareOpc = 454;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtfp_p:
CompareOpc = 710;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtsb_p:
CompareOpc = 774;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtsh_p:
CompareOpc = 838;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtsw_p:
CompareOpc = 902;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtsd_p:
if (Subtarget.hasP8Altivec()) {
CompareOpc = 967;
isDot = true;
} else
return false;
break;
case Intrinsic::ppc_altivec_vcmpgtub_p:
CompareOpc = 518;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtuh_p:
CompareOpc = 582;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtuw_p:
CompareOpc = 646;
isDot = true;
break;
case Intrinsic::ppc_altivec_vcmpgtud_p:
if (Subtarget.hasP8Altivec()) {
CompareOpc = 711;
isDot = true;
} else
return false;
break;
case Intrinsic::ppc_altivec_vcmpequq:
case Intrinsic::ppc_altivec_vcmpgtsq:
case Intrinsic::ppc_altivec_vcmpgtuq:
if (!Subtarget.isISA3_1())
return false;
switch (IntrinsicID) {
default:
llvm_unreachable("Unknown comparison intrinsic.");
case Intrinsic::ppc_altivec_vcmpequq:
CompareOpc = 455;
break;
case Intrinsic::ppc_altivec_vcmpgtsq:
CompareOpc = 903;
break;
case Intrinsic::ppc_altivec_vcmpgtuq:
CompareOpc = 647;
break;
}
break;
// VSX predicate comparisons use the same infrastructure
case Intrinsic::ppc_vsx_xvcmpeqdp_p:
case Intrinsic::ppc_vsx_xvcmpgedp_p:
case Intrinsic::ppc_vsx_xvcmpgtdp_p:
case Intrinsic::ppc_vsx_xvcmpeqsp_p:
case Intrinsic::ppc_vsx_xvcmpgesp_p:
case Intrinsic::ppc_vsx_xvcmpgtsp_p:
if (Subtarget.hasVSX()) {
switch (IntrinsicID) {
case Intrinsic::ppc_vsx_xvcmpeqdp_p:
CompareOpc = 99;
break;
case Intrinsic::ppc_vsx_xvcmpgedp_p:
CompareOpc = 115;
break;
case Intrinsic::ppc_vsx_xvcmpgtdp_p:
CompareOpc = 107;
break;
case Intrinsic::ppc_vsx_xvcmpeqsp_p:
CompareOpc = 67;
break;
case Intrinsic::ppc_vsx_xvcmpgesp_p:
CompareOpc = 83;
break;
case Intrinsic::ppc_vsx_xvcmpgtsp_p:
CompareOpc = 75;
break;
}
isDot = true;
} else
return false;
break;
// Normal Comparisons.
case Intrinsic::ppc_altivec_vcmpbfp:
CompareOpc = 966;
break;
case Intrinsic::ppc_altivec_vcmpeqfp:
CompareOpc = 198;
break;
case Intrinsic::ppc_altivec_vcmpequb:
CompareOpc = 6;
break;
case Intrinsic::ppc_altivec_vcmpequh:
CompareOpc = 70;
break;
case Intrinsic::ppc_altivec_vcmpequw:
CompareOpc = 134;
break;
case Intrinsic::ppc_altivec_vcmpequd:
if (Subtarget.hasP8Altivec())
CompareOpc = 199;
else
return false;
break;
case Intrinsic::ppc_altivec_vcmpneb:
case Intrinsic::ppc_altivec_vcmpneh:
case Intrinsic::ppc_altivec_vcmpnew:
case Intrinsic::ppc_altivec_vcmpnezb:
case Intrinsic::ppc_altivec_vcmpnezh:
case Intrinsic::ppc_altivec_vcmpnezw:
if (Subtarget.hasP9Altivec())
switch (IntrinsicID) {
default:
llvm_unreachable("Unknown comparison intrinsic.");
case Intrinsic::ppc_altivec_vcmpneb:
CompareOpc = 7;
break;
case Intrinsic::ppc_altivec_vcmpneh:
CompareOpc = 71;
break;
case Intrinsic::ppc_altivec_vcmpnew:
CompareOpc = 135;
break;
case Intrinsic::ppc_altivec_vcmpnezb:
CompareOpc = 263;
break;
case Intrinsic::ppc_altivec_vcmpnezh:
CompareOpc = 327;
break;
case Intrinsic::ppc_altivec_vcmpnezw:
CompareOpc = 391;
break;
}
else
return false;
break;
case Intrinsic::ppc_altivec_vcmpgefp:
CompareOpc = 454;
break;
case Intrinsic::ppc_altivec_vcmpgtfp:
CompareOpc = 710;
break;
case Intrinsic::ppc_altivec_vcmpgtsb:
CompareOpc = 774;
break;
case Intrinsic::ppc_altivec_vcmpgtsh:
CompareOpc = 838;
break;
case Intrinsic::ppc_altivec_vcmpgtsw:
CompareOpc = 902;
break;
case Intrinsic::ppc_altivec_vcmpgtsd:
if (Subtarget.hasP8Altivec())
CompareOpc = 967;
else
return false;
break;
case Intrinsic::ppc_altivec_vcmpgtub:
CompareOpc = 518;
break;
case Intrinsic::ppc_altivec_vcmpgtuh:
CompareOpc = 582;
break;
case Intrinsic::ppc_altivec_vcmpgtuw:
CompareOpc = 646;
break;
case Intrinsic::ppc_altivec_vcmpgtud:
if (Subtarget.hasP8Altivec())
CompareOpc = 711;
else
return false;
break;
case Intrinsic::ppc_altivec_vcmpequq_p:
case Intrinsic::ppc_altivec_vcmpgtsq_p:
case Intrinsic::ppc_altivec_vcmpgtuq_p:
if (!Subtarget.isISA3_1())
return false;
switch (IntrinsicID) {
default:
llvm_unreachable("Unknown comparison intrinsic.");
case Intrinsic::ppc_altivec_vcmpequq_p:
CompareOpc = 455;
break;
case Intrinsic::ppc_altivec_vcmpgtsq_p:
CompareOpc = 903;
break;
case Intrinsic::ppc_altivec_vcmpgtuq_p:
CompareOpc = 647;
break;
}
isDot = true;
break;
}
return true;
}
/// LowerINTRINSIC_WO_CHAIN - If this is an intrinsic that we want to custom
/// lower, do it, otherwise return null.
SDValue PPCTargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
SelectionDAG &DAG) const {
unsigned IntrinsicID =
cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
SDLoc dl(Op);
switch (IntrinsicID) {
case Intrinsic::thread_pointer:
// Reads the thread pointer register, used for __builtin_thread_pointer.
if (Subtarget.isPPC64())
return DAG.getRegister(PPC::X13, MVT::i64);
return DAG.getRegister(PPC::R2, MVT::i32);
case Intrinsic::ppc_mma_disassemble_acc:
case Intrinsic::ppc_vsx_disassemble_pair: {
int NumVecs = 2;
SDValue WideVec = Op.getOperand(1);
if (IntrinsicID == Intrinsic::ppc_mma_disassemble_acc) {
NumVecs = 4;
WideVec = DAG.getNode(PPCISD::XXMFACC, dl, MVT::v512i1, WideVec);
}
SmallVector<SDValue, 4> RetOps;
for (int VecNo = 0; VecNo < NumVecs; VecNo++) {
SDValue Extract = DAG.getNode(
PPCISD::EXTRACT_VSX_REG, dl, MVT::v16i8, WideVec,
DAG.getConstant(Subtarget.isLittleEndian() ? NumVecs - 1 - VecNo
: VecNo,
dl, MVT::i64));
RetOps.push_back(Extract);
}
return DAG.getMergeValues(RetOps, dl);
}
}
// If this is a lowered altivec predicate compare, CompareOpc is set to the
// opcode number of the comparison.
int CompareOpc;
bool isDot;
if (!getVectorCompareInfo(Op, CompareOpc, isDot, Subtarget))
return SDValue(); // Don't custom lower most intrinsics.
// If this is a non-dot comparison, make the VCMP node and we are done.
if (!isDot) {
SDValue Tmp = DAG.getNode(PPCISD::VCMP, dl, Op.getOperand(2).getValueType(),
Op.getOperand(1), Op.getOperand(2),
DAG.getConstant(CompareOpc, dl, MVT::i32));
return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Tmp);
}
// Create the PPCISD altivec 'dot' comparison node.
SDValue Ops[] = {
Op.getOperand(2), // LHS
Op.getOperand(3), // RHS
DAG.getConstant(CompareOpc, dl, MVT::i32)
};
EVT VTs[] = { Op.getOperand(2).getValueType(), MVT::Glue };
SDValue CompNode = DAG.getNode(PPCISD::VCMP_rec, dl, VTs, Ops);
// Now that we have the comparison, emit a copy from the CR to a GPR.
// This is flagged to the above dot comparison.
SDValue Flags = DAG.getNode(PPCISD::MFOCRF, dl, MVT::i32,
DAG.getRegister(PPC::CR6, MVT::i32),
CompNode.getValue(1));
// Unpack the result based on how the target uses it.
unsigned BitNo; // Bit # of CR6.
bool InvertBit; // Invert result?
switch (cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue()) {
default: // Can't happen, don't crash on invalid number though.
case 0: // Return the value of the EQ bit of CR6.
BitNo = 0; InvertBit = false;
break;
case 1: // Return the inverted value of the EQ bit of CR6.
BitNo = 0; InvertBit = true;
break;
case 2: // Return the value of the LT bit of CR6.
BitNo = 2; InvertBit = false;
break;
case 3: // Return the inverted value of the LT bit of CR6.
BitNo = 2; InvertBit = true;
break;
}
// Shift the bit into the low position.
Flags = DAG.getNode(ISD::SRL, dl, MVT::i32, Flags,
DAG.getConstant(8 - (3 - BitNo), dl, MVT::i32));
// Isolate the bit.
Flags = DAG.getNode(ISD::AND, dl, MVT::i32, Flags,
DAG.getConstant(1, dl, MVT::i32));
// If we are supposed to, toggle the bit.
if (InvertBit)
Flags = DAG.getNode(ISD::XOR, dl, MVT::i32, Flags,
DAG.getConstant(1, dl, MVT::i32));
return Flags;
}
SDValue PPCTargetLowering::LowerINTRINSIC_VOID(SDValue Op,
SelectionDAG &DAG) const {
// SelectionDAGBuilder::visitTargetIntrinsic may insert one extra chain to
// the beginning of the argument list.
int ArgStart = isa<ConstantSDNode>(Op.getOperand(0)) ? 0 : 1;
SDLoc DL(Op);
switch (cast<ConstantSDNode>(Op.getOperand(ArgStart))->getZExtValue()) {
case Intrinsic::ppc_cfence: {
assert(ArgStart == 1 && "llvm.ppc.cfence must carry a chain argument.");
assert(Subtarget.isPPC64() && "Only 64-bit is supported for now.");
return SDValue(DAG.getMachineNode(PPC::CFENCE8, DL, MVT::Other,
DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64,
Op.getOperand(ArgStart + 1)),
Op.getOperand(0)),
0);
}
default:
break;
}
return SDValue();
}
// Lower scalar BSWAP64 to xxbrd.
SDValue PPCTargetLowering::LowerBSWAP(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
// MTVSRDD
Op = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, Op.getOperand(0),
Op.getOperand(0));
// XXBRD
Op = DAG.getNode(ISD::BSWAP, dl, MVT::v2i64, Op);
// MFVSRD
int VectorIndex = 0;
if (Subtarget.isLittleEndian())
VectorIndex = 1;
Op = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64, Op,
DAG.getTargetConstant(VectorIndex, dl, MVT::i32));
return Op;
}
// ATOMIC_CMP_SWAP for i8/i16 needs to zero-extend its input since it will be
// compared to a value that is atomically loaded (atomic loads zero-extend).
SDValue PPCTargetLowering::LowerATOMIC_CMP_SWAP(SDValue Op,
SelectionDAG &DAG) const {
assert(Op.getOpcode() == ISD::ATOMIC_CMP_SWAP &&
"Expecting an atomic compare-and-swap here.");
SDLoc dl(Op);
auto *AtomicNode = cast<AtomicSDNode>(Op.getNode());
EVT MemVT = AtomicNode->getMemoryVT();
if (MemVT.getSizeInBits() >= 32)
return Op;
SDValue CmpOp = Op.getOperand(2);
// If this is already correctly zero-extended, leave it alone.
auto HighBits = APInt::getHighBitsSet(32, 32 - MemVT.getSizeInBits());
if (DAG.MaskedValueIsZero(CmpOp, HighBits))
return Op;
// Clear the high bits of the compare operand.
unsigned MaskVal = (1 << MemVT.getSizeInBits()) - 1;
SDValue NewCmpOp =
DAG.getNode(ISD::AND, dl, MVT::i32, CmpOp,
DAG.getConstant(MaskVal, dl, MVT::i32));
// Replace the existing compare operand with the properly zero-extended one.
SmallVector<SDValue, 4> Ops;
for (int i = 0, e = AtomicNode->getNumOperands(); i < e; i++)
Ops.push_back(AtomicNode->getOperand(i));
Ops[2] = NewCmpOp;
MachineMemOperand *MMO = AtomicNode->getMemOperand();
SDVTList Tys = DAG.getVTList(MVT::i32, MVT::Other);
auto NodeTy =
(MemVT == MVT::i8) ? PPCISD::ATOMIC_CMP_SWAP_8 : PPCISD::ATOMIC_CMP_SWAP_16;
return DAG.getMemIntrinsicNode(NodeTy, dl, Tys, Ops, MemVT, MMO);
}
SDValue PPCTargetLowering::LowerSCALAR_TO_VECTOR(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
// Create a stack slot that is 16-byte aligned.
MachineFrameInfo &MFI = DAG.getMachineFunction().getFrameInfo();
int FrameIdx = MFI.CreateStackObject(16, Align(16), false);
EVT PtrVT = getPointerTy(DAG.getDataLayout());
SDValue FIdx = DAG.getFrameIndex(FrameIdx, PtrVT);
// Store the input value into Value#0 of the stack slot.
SDValue Store = DAG.getStore(DAG.getEntryNode(), dl, Op.getOperand(0), FIdx,
MachinePointerInfo());
// Load it out.
return DAG.getLoad(Op.getValueType(), dl, Store, FIdx, MachinePointerInfo());
}
SDValue PPCTargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
SelectionDAG &DAG) const {
assert(Op.getOpcode() == ISD::INSERT_VECTOR_ELT &&
"Should only be called for ISD::INSERT_VECTOR_ELT");
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(2));
// We have legal lowering for constant indices but not for variable ones.
if (!C)
return SDValue();
EVT VT = Op.getValueType();
SDLoc dl(Op);
SDValue V1 = Op.getOperand(0);
SDValue V2 = Op.getOperand(1);
// We can use MTVSRZ + VECINSERT for v8i16 and v16i8 types.
if (VT == MVT::v8i16 || VT == MVT::v16i8) {
SDValue Mtvsrz = DAG.getNode(PPCISD::MTVSRZ, dl, VT, V2);
unsigned BytesInEachElement = VT.getVectorElementType().getSizeInBits() / 8;
unsigned InsertAtElement = C->getZExtValue();
unsigned InsertAtByte = InsertAtElement * BytesInEachElement;
if (Subtarget.isLittleEndian()) {
InsertAtByte = (16 - BytesInEachElement) - InsertAtByte;
}
return DAG.getNode(PPCISD::VECINSERT, dl, VT, V1, Mtvsrz,
DAG.getConstant(InsertAtByte, dl, MVT::i32));
}
return Op;
}
SDValue PPCTargetLowering::LowerVectorLoad(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
LoadSDNode *LN = cast<LoadSDNode>(Op.getNode());
SDValue LoadChain = LN->getChain();
SDValue BasePtr = LN->getBasePtr();
EVT VT = Op.getValueType();
if (VT != MVT::v256i1 && VT != MVT::v512i1)
return Op;
// Type v256i1 is used for pairs and v512i1 is used for accumulators.
// Here we create 2 or 4 v16i8 loads to load the pair or accumulator value in
// 2 or 4 vsx registers.
assert((VT != MVT::v512i1 || Subtarget.hasMMA()) &&
"Type unsupported without MMA");
assert((VT != MVT::v256i1 || Subtarget.pairedVectorMemops()) &&
"Type unsupported without paired vector support");
Align Alignment = LN->getAlign();
SmallVector<SDValue, 4> Loads;
SmallVector<SDValue, 4> LoadChains;
unsigned NumVecs = VT.getSizeInBits() / 128;
for (unsigned Idx = 0; Idx < NumVecs; ++Idx) {
SDValue Load =
DAG.getLoad(MVT::v16i8, dl, LoadChain, BasePtr,
LN->getPointerInfo().getWithOffset(Idx * 16),
commonAlignment(Alignment, Idx * 16),
LN->getMemOperand()->getFlags(), LN->getAAInfo());
BasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
DAG.getConstant(16, dl, BasePtr.getValueType()));
Loads.push_back(Load);
LoadChains.push_back(Load.getValue(1));
}
if (Subtarget.isLittleEndian()) {
std::reverse(Loads.begin(), Loads.end());
std::reverse(LoadChains.begin(), LoadChains.end());
}
SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, LoadChains);
SDValue Value =
DAG.getNode(VT == MVT::v512i1 ? PPCISD::ACC_BUILD : PPCISD::PAIR_BUILD,
dl, VT, Loads);
SDValue RetOps[] = {Value, TF};
return DAG.getMergeValues(RetOps, dl);
}
SDValue PPCTargetLowering::LowerVectorStore(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
StoreSDNode *SN = cast<StoreSDNode>(Op.getNode());
SDValue StoreChain = SN->getChain();
SDValue BasePtr = SN->getBasePtr();
SDValue Value = SN->getValue();
EVT StoreVT = Value.getValueType();
if (StoreVT != MVT::v256i1 && StoreVT != MVT::v512i1)
return Op;
// Type v256i1 is used for pairs and v512i1 is used for accumulators.
// Here we create 2 or 4 v16i8 stores to store the pair or accumulator
// underlying registers individually.
assert((StoreVT != MVT::v512i1 || Subtarget.hasMMA()) &&
"Type unsupported without MMA");
assert((StoreVT != MVT::v256i1 || Subtarget.pairedVectorMemops()) &&
"Type unsupported without paired vector support");
Align Alignment = SN->getAlign();
SmallVector<SDValue, 4> Stores;
unsigned NumVecs = 2;
if (StoreVT == MVT::v512i1) {
Value = DAG.getNode(PPCISD::XXMFACC, dl, MVT::v512i1, Value);
NumVecs = 4;
}
for (unsigned Idx = 0; Idx < NumVecs; ++Idx) {
unsigned VecNum = Subtarget.isLittleEndian() ? NumVecs - 1 - Idx : Idx;
SDValue Elt = DAG.getNode(PPCISD::EXTRACT_VSX_REG, dl, MVT::v16i8, Value,
DAG.getConstant(VecNum, dl, MVT::i64));
SDValue Store =
DAG.getStore(StoreChain, dl, Elt, BasePtr,
SN->getPointerInfo().getWithOffset(Idx * 16),
commonAlignment(Alignment, Idx * 16),
SN->getMemOperand()->getFlags(), SN->getAAInfo());
BasePtr = DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
DAG.getConstant(16, dl, BasePtr.getValueType()));
Stores.push_back(Store);
}
SDValue TF = DAG.getTokenFactor(dl, Stores);
return TF;
}
SDValue PPCTargetLowering::LowerMUL(SDValue Op, SelectionDAG &DAG) const {
SDLoc dl(Op);
if (Op.getValueType() == MVT::v4i32) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
SDValue Zero = getCanonicalConstSplat(0, 1, MVT::v4i32, DAG, dl);
// +16 as shift amt.
SDValue Neg16 = getCanonicalConstSplat(-16, 4, MVT::v4i32, DAG, dl);
SDValue RHSSwap = // = vrlw RHS, 16
BuildIntrinsicOp(Intrinsic::ppc_altivec_vrlw, RHS, Neg16, DAG, dl);
// Shrinkify inputs to v8i16.
LHS = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, LHS);
RHS = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, RHS);
RHSSwap = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, RHSSwap);
// Low parts multiplied together, generating 32-bit results (we ignore the
// top parts).
SDValue LoProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmulouh,
LHS, RHS, DAG, dl, MVT::v4i32);
SDValue HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmsumuhm,
LHS, RHSSwap, Zero, DAG, dl, MVT::v4i32);
// Shift the high parts up 16 bits.
HiProd = BuildIntrinsicOp(Intrinsic::ppc_altivec_vslw, HiProd,
Neg16, DAG, dl);
return DAG.getNode(ISD::ADD, dl, MVT::v4i32, LoProd, HiProd);
} else if (Op.getValueType() == MVT::v16i8) {
SDValue LHS = Op.getOperand(0), RHS = Op.getOperand(1);
bool isLittleEndian = Subtarget.isLittleEndian();
// Multiply the even 8-bit parts, producing 16-bit sums.
SDValue EvenParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuleub,
LHS, RHS, DAG, dl, MVT::v8i16);
EvenParts = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, EvenParts);
// Multiply the odd 8-bit parts, producing 16-bit sums.
SDValue OddParts = BuildIntrinsicOp(Intrinsic::ppc_altivec_vmuloub,
LHS, RHS, DAG, dl, MVT::v8i16);
OddParts = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, OddParts);
// Merge the results together. Because vmuleub and vmuloub are
// instructions with a big-endian bias, we must reverse the
// element numbering and reverse the meaning of "odd" and "even"
// when generating little endian code.
int Ops[16];
for (unsigned i = 0; i != 8; ++i) {
if (isLittleEndian) {
Ops[i*2 ] = 2*i;
Ops[i*2+1] = 2*i+16;
} else {
Ops[i*2 ] = 2*i+1;
Ops[i*2+1] = 2*i+1+16;
}
}
if (isLittleEndian)
return DAG.getVectorShuffle(MVT::v16i8, dl, OddParts, EvenParts, Ops);
else
return DAG.getVectorShuffle(MVT::v16i8, dl, EvenParts, OddParts, Ops);
} else {
llvm_unreachable("Unknown mul to lower!");
}
}
SDValue PPCTargetLowering::LowerFP_ROUND(SDValue Op, SelectionDAG &DAG) const {
bool IsStrict = Op->isStrictFPOpcode();
if (Op.getOperand(IsStrict ? 1 : 0).getValueType() == MVT::f128 &&
!Subtarget.hasP9Vector())
return SDValue();
return Op;
}
// Custom lowering for fpext vf32 to v2f64
SDValue PPCTargetLowering::LowerFP_EXTEND(SDValue Op, SelectionDAG &DAG) const {
assert(Op.getOpcode() == ISD::FP_EXTEND &&
"Should only be called for ISD::FP_EXTEND");
// FIXME: handle extends from half precision float vectors on P9.
// We only want to custom lower an extend from v2f32 to v2f64.
if (Op.getValueType() != MVT::v2f64 ||
Op.getOperand(0).getValueType() != MVT::v2f32)
return SDValue();
SDLoc dl(Op);
SDValue Op0 = Op.getOperand(0);
switch (Op0.getOpcode()) {
default:
return SDValue();
case ISD::EXTRACT_SUBVECTOR: {
assert(Op0.getNumOperands() == 2 &&
isa<ConstantSDNode>(Op0->getOperand(1)) &&
"Node should have 2 operands with second one being a constant!");
if (Op0.getOperand(0).getValueType() != MVT::v4f32)
return SDValue();
// Custom lower is only done for high or low doubleword.
int Idx = cast<ConstantSDNode>(Op0.getOperand(1))->getZExtValue();
if (Idx % 2 != 0)
return SDValue();
// Since input is v4f32, at this point Idx is either 0 or 2.
// Shift to get the doubleword position we want.
int DWord = Idx >> 1;
// High and low word positions are different on little endian.
if (Subtarget.isLittleEndian())
DWord ^= 0x1;
return DAG.getNode(PPCISD::FP_EXTEND_HALF, dl, MVT::v2f64,
Op0.getOperand(0), DAG.getConstant(DWord, dl, MVT::i32));
}
case ISD::FADD:
case ISD::FMUL:
case ISD::FSUB: {
SDValue NewLoad[2];
for (unsigned i = 0, ie = Op0.getNumOperands(); i != ie; ++i) {
// Ensure both input are loads.
SDValue LdOp = Op0.getOperand(i);
if (LdOp.getOpcode() != ISD::LOAD)
return SDValue();
// Generate new load node.
LoadSDNode *LD = cast<LoadSDNode>(LdOp);
SDValue LoadOps[] = {LD->getChain(), LD->getBasePtr()};
NewLoad[i] = DAG.getMemIntrinsicNode(
PPCISD::LD_VSX_LH, dl, DAG.getVTList(MVT::v4f32, MVT::Other), LoadOps,
LD->getMemoryVT(), LD->getMemOperand());
}
SDValue NewOp =
DAG.getNode(Op0.getOpcode(), SDLoc(Op0), MVT::v4f32, NewLoad[0],
NewLoad[1], Op0.getNode()->getFlags());
return DAG.getNode(PPCISD::FP_EXTEND_HALF, dl, MVT::v2f64, NewOp,
DAG.getConstant(0, dl, MVT::i32));
}
case ISD::LOAD: {
LoadSDNode *LD = cast<LoadSDNode>(Op0);
SDValue LoadOps[] = {LD->getChain(), LD->getBasePtr()};
SDValue NewLd = DAG.getMemIntrinsicNode(
PPCISD::LD_VSX_LH, dl, DAG.getVTList(MVT::v4f32, MVT::Other), LoadOps,
LD->getMemoryVT(), LD->getMemOperand());
return DAG.getNode(PPCISD::FP_EXTEND_HALF, dl, MVT::v2f64, NewLd,
DAG.getConstant(0, dl, MVT::i32));
}
}
llvm_unreachable("ERROR:Should return for all cases within swtich.");
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
switch (Op.getOpcode()) {
default: llvm_unreachable("Wasn't expecting to be able to lower this!");
case ISD::ConstantPool: return LowerConstantPool(Op, DAG);
case ISD::BlockAddress: return LowerBlockAddress(Op, DAG);
case ISD::GlobalAddress: return LowerGlobalAddress(Op, DAG);
case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SETCC: return LowerSETCC(Op, DAG);
case ISD::INIT_TRAMPOLINE: return LowerINIT_TRAMPOLINE(Op, DAG);
case ISD::ADJUST_TRAMPOLINE: return LowerADJUST_TRAMPOLINE(Op, DAG);
// Variable argument lowering.
case ISD::VASTART: return LowerVASTART(Op, DAG);
case ISD::VAARG: return LowerVAARG(Op, DAG);
case ISD::VACOPY: return LowerVACOPY(Op, DAG);
case ISD::STACKRESTORE: return LowerSTACKRESTORE(Op, DAG);
case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG);
case ISD::GET_DYNAMIC_AREA_OFFSET:
return LowerGET_DYNAMIC_AREA_OFFSET(Op, DAG);
// Exception handling lowering.
case ISD::EH_DWARF_CFA: return LowerEH_DWARF_CFA(Op, DAG);
case ISD::EH_SJLJ_SETJMP: return lowerEH_SJLJ_SETJMP(Op, DAG);
case ISD::EH_SJLJ_LONGJMP: return lowerEH_SJLJ_LONGJMP(Op, DAG);
case ISD::LOAD: return LowerLOAD(Op, DAG);
case ISD::STORE: return LowerSTORE(Op, DAG);
case ISD::TRUNCATE: return LowerTRUNCATE(Op, DAG);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::STRICT_FP_TO_UINT:
case ISD::STRICT_FP_TO_SINT:
case ISD::FP_TO_UINT:
case ISD::FP_TO_SINT: return LowerFP_TO_INT(Op, DAG, SDLoc(Op));
case ISD::STRICT_UINT_TO_FP:
case ISD::STRICT_SINT_TO_FP:
case ISD::UINT_TO_FP:
case ISD::SINT_TO_FP: return LowerINT_TO_FP(Op, DAG);
case ISD::FLT_ROUNDS_: return LowerFLT_ROUNDS_(Op, DAG);
// Lower 64-bit shifts.
case ISD::SHL_PARTS: return LowerSHL_PARTS(Op, DAG);
case ISD::SRL_PARTS: return LowerSRL_PARTS(Op, DAG);
case ISD::SRA_PARTS: return LowerSRA_PARTS(Op, DAG);
case ISD::FSHL: return LowerFunnelShift(Op, DAG);
case ISD::FSHR: return LowerFunnelShift(Op, DAG);
// Vector-related lowering.
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case ISD::VECTOR_SHUFFLE: return LowerVECTOR_SHUFFLE(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: return LowerINTRINSIC_WO_CHAIN(Op, DAG);
case ISD::SCALAR_TO_VECTOR: return LowerSCALAR_TO_VECTOR(Op, DAG);
case ISD::INSERT_VECTOR_ELT: return LowerINSERT_VECTOR_ELT(Op, DAG);
case ISD::MUL: return LowerMUL(Op, DAG);
case ISD::FP_EXTEND: return LowerFP_EXTEND(Op, DAG);
case ISD::STRICT_FP_ROUND:
case ISD::FP_ROUND:
return LowerFP_ROUND(Op, DAG);
case ISD::ROTL: return LowerROTL(Op, DAG);
// For counter-based loop handling.
case ISD::INTRINSIC_W_CHAIN: return SDValue();
case ISD::BITCAST: return LowerBITCAST(Op, DAG);
// Frame & Return address.
case ISD::RETURNADDR: return LowerRETURNADDR(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::INTRINSIC_VOID:
return LowerINTRINSIC_VOID(Op, DAG);
case ISD::BSWAP:
return LowerBSWAP(Op, DAG);
case ISD::ATOMIC_CMP_SWAP:
return LowerATOMIC_CMP_SWAP(Op, DAG);
}
}
void PPCTargetLowering::ReplaceNodeResults(SDNode *N,
SmallVectorImpl<SDValue>&Results,
SelectionDAG &DAG) const {
SDLoc dl(N);
switch (N->getOpcode()) {
default:
llvm_unreachable("Do not know how to custom type legalize this operation!");
case ISD::READCYCLECOUNTER: {
SDVTList VTs = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
SDValue RTB = DAG.getNode(PPCISD::READ_TIME_BASE, dl, VTs, N->getOperand(0));
Results.push_back(
DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, RTB, RTB.getValue(1)));
Results.push_back(RTB.getValue(2));
break;
}
case ISD::INTRINSIC_W_CHAIN: {
if (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue() !=
Intrinsic::loop_decrement)
break;
assert(N->getValueType(0) == MVT::i1 &&
"Unexpected result type for CTR decrement intrinsic");
EVT SVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
N->getValueType(0));
SDVTList VTs = DAG.getVTList(SVT, MVT::Other);
SDValue NewInt = DAG.getNode(N->getOpcode(), dl, VTs, N->getOperand(0),
N->getOperand(1));
Results.push_back(DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, NewInt));
Results.push_back(NewInt.getValue(1));
break;
}
case ISD::VAARG: {
if (!Subtarget.isSVR4ABI() || Subtarget.isPPC64())
return;
EVT VT = N->getValueType(0);
if (VT == MVT::i64) {
SDValue NewNode = LowerVAARG(SDValue(N, 1), DAG);
Results.push_back(NewNode);
Results.push_back(NewNode.getValue(1));
}
return;
}
case ISD::STRICT_FP_TO_SINT:
case ISD::STRICT_FP_TO_UINT:
case ISD::FP_TO_SINT:
case ISD::FP_TO_UINT:
// LowerFP_TO_INT() can only handle f32 and f64.
if (N->getOperand(N->isStrictFPOpcode() ? 1 : 0).getValueType() ==
MVT::ppcf128)
return;
Results.push_back(LowerFP_TO_INT(SDValue(N, 0), DAG, dl));
return;
case ISD::TRUNCATE: {
if (!N->getValueType(0).isVector())
return;
SDValue Lowered = LowerTRUNCATEVector(SDValue(N, 0), DAG);
if (Lowered)
Results.push_back(Lowered);
return;
}
case ISD::FSHL:
case ISD::FSHR:
// Don't handle funnel shifts here.
return;
case ISD::BITCAST:
// Don't handle bitcast here.
return;
case ISD::FP_EXTEND:
SDValue Lowered = LowerFP_EXTEND(SDValue(N, 0), DAG);
if (Lowered)
Results.push_back(Lowered);
return;
}
}
//===----------------------------------------------------------------------===//
// Other Lowering Code
//===----------------------------------------------------------------------===//
static Instruction* callIntrinsic(IRBuilder<> &Builder, Intrinsic::ID Id) {
Module *M = Builder.GetInsertBlock()->getParent()->getParent();
Function *Func = Intrinsic::getDeclaration(M, Id);
return Builder.CreateCall(Func, {});
}
// The mappings for emitLeading/TrailingFence is taken from
// http://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
Instruction *PPCTargetLowering::emitLeadingFence(IRBuilder<> &Builder,
Instruction *Inst,
AtomicOrdering Ord) const {
if (Ord == AtomicOrdering::SequentiallyConsistent)
return callIntrinsic(Builder, Intrinsic::ppc_sync);
if (isReleaseOrStronger(Ord))
return callIntrinsic(Builder, Intrinsic::ppc_lwsync);
return nullptr;
}
Instruction *PPCTargetLowering::emitTrailingFence(IRBuilder<> &Builder,
Instruction *Inst,
AtomicOrdering Ord) const {
if (Inst->hasAtomicLoad() && isAcquireOrStronger(Ord)) {
// See http://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html and
// http://www.rdrop.com/users/paulmck/scalability/paper/N2745r.2011.03.04a.html
// and http://www.cl.cam.ac.uk/~pes20/cppppc/ for justification.
if (isa<LoadInst>(Inst) && Subtarget.isPPC64())
return Builder.CreateCall(
Intrinsic::getDeclaration(
Builder.GetInsertBlock()->getParent()->getParent(),
Intrinsic::ppc_cfence, {Inst->getType()}),
{Inst});
// FIXME: Can use isync for rmw operation.
return callIntrinsic(Builder, Intrinsic::ppc_lwsync);
}
return nullptr;
}
MachineBasicBlock *
PPCTargetLowering::EmitAtomicBinary(MachineInstr &MI, MachineBasicBlock *BB,
unsigned AtomicSize,
unsigned BinOpcode,
unsigned CmpOpcode,
unsigned CmpPred) const {
// This also handles ATOMIC_SWAP, indicated by BinOpcode==0.
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
auto LoadMnemonic = PPC::LDARX;
auto StoreMnemonic = PPC::STDCX;
switch (AtomicSize) {
default:
llvm_unreachable("Unexpected size of atomic entity");
case 1:
LoadMnemonic = PPC::LBARX;
StoreMnemonic = PPC::STBCX;
assert(Subtarget.hasPartwordAtomics() && "Call this only with size >=4");
break;
case 2:
LoadMnemonic = PPC::LHARX;
StoreMnemonic = PPC::STHCX;
assert(Subtarget.hasPartwordAtomics() && "Call this only with size >=4");
break;
case 4:
LoadMnemonic = PPC::LWARX;
StoreMnemonic = PPC::STWCX;
break;
case 8:
LoadMnemonic = PPC::LDARX;
StoreMnemonic = PPC::STDCX;
break;
}
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction *F = BB->getParent();
MachineFunction::iterator It = ++BB->getIterator();
Register dest = MI.getOperand(0).getReg();
Register ptrA = MI.getOperand(1).getReg();
Register ptrB = MI.getOperand(2).getReg();
Register incr = MI.getOperand(3).getReg();
DebugLoc dl = MI.getDebugLoc();
MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB =
CmpOpcode ? F->CreateMachineBasicBlock(LLVM_BB) : nullptr;
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, loopMBB);
if (CmpOpcode)
F->insert(It, loop2MBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register TmpReg = (!BinOpcode) ? incr :
RegInfo.createVirtualRegister( AtomicSize == 8 ? &PPC::G8RCRegClass
: &PPC::GPRCRegClass);
// thisMBB:
// ...
// fallthrough --> loopMBB
BB->addSuccessor(loopMBB);
// loopMBB:
// l[wd]arx dest, ptr
// add r0, dest, incr
// st[wd]cx. r0, ptr
// bne- loopMBB
// fallthrough --> exitMBB
// For max/min...
// loopMBB:
// l[wd]arx dest, ptr
// cmpl?[wd] incr, dest
// bgt exitMBB
// loop2MBB:
// st[wd]cx. dest, ptr
// bne- loopMBB
// fallthrough --> exitMBB
BB = loopMBB;
BuildMI(BB, dl, TII->get(LoadMnemonic), dest)
.addReg(ptrA).addReg(ptrB);
if (BinOpcode)
BuildMI(BB, dl, TII->get(BinOpcode), TmpReg).addReg(incr).addReg(dest);
if (CmpOpcode) {
// Signed comparisons of byte or halfword values must be sign-extended.
if (CmpOpcode == PPC::CMPW && AtomicSize < 4) {
Register ExtReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
BuildMI(BB, dl, TII->get(AtomicSize == 1 ? PPC::EXTSB : PPC::EXTSH),
ExtReg).addReg(dest);
BuildMI(BB, dl, TII->get(CmpOpcode), PPC::CR0)
.addReg(incr).addReg(ExtReg);
} else
BuildMI(BB, dl, TII->get(CmpOpcode), PPC::CR0)
.addReg(incr).addReg(dest);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(CmpPred).addReg(PPC::CR0).addMBB(exitMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(exitMBB);
BB = loop2MBB;
}
BuildMI(BB, dl, TII->get(StoreMnemonic))
.addReg(TmpReg).addReg(ptrA).addReg(ptrB);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE).addReg(PPC::CR0).addMBB(loopMBB);
BB->addSuccessor(loopMBB);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
return BB;
}
static bool isSignExtended(MachineInstr &MI, const PPCInstrInfo *TII) {
switch(MI.getOpcode()) {
default:
return false;
case PPC::COPY:
return TII->isSignExtended(MI);
case PPC::LHA:
case PPC::LHA8:
case PPC::LHAU:
case PPC::LHAU8:
case PPC::LHAUX:
case PPC::LHAUX8:
case PPC::LHAX:
case PPC::LHAX8:
case PPC::LWA:
case PPC::LWAUX:
case PPC::LWAX:
case PPC::LWAX_32:
case PPC::LWA_32:
case PPC::PLHA:
case PPC::PLHA8:
case PPC::PLHA8pc:
case PPC::PLHApc:
case PPC::PLWA:
case PPC::PLWA8:
case PPC::PLWA8pc:
case PPC::PLWApc:
case PPC::EXTSB:
case PPC::EXTSB8:
case PPC::EXTSB8_32_64:
case PPC::EXTSB8_rec:
case PPC::EXTSB_rec:
case PPC::EXTSH:
case PPC::EXTSH8:
case PPC::EXTSH8_32_64:
case PPC::EXTSH8_rec:
case PPC::EXTSH_rec:
case PPC::EXTSW:
case PPC::EXTSWSLI:
case PPC::EXTSWSLI_32_64:
case PPC::EXTSWSLI_32_64_rec:
case PPC::EXTSWSLI_rec:
case PPC::EXTSW_32:
case PPC::EXTSW_32_64:
case PPC::EXTSW_32_64_rec:
case PPC::EXTSW_rec:
case PPC::SRAW:
case PPC::SRAWI:
case PPC::SRAWI_rec:
case PPC::SRAW_rec:
return true;
}
return false;
}
MachineBasicBlock *PPCTargetLowering::EmitPartwordAtomicBinary(
MachineInstr &MI, MachineBasicBlock *BB,
bool is8bit, // operation
unsigned BinOpcode, unsigned CmpOpcode, unsigned CmpPred) const {
// This also handles ATOMIC_SWAP, indicated by BinOpcode==0.
const PPCInstrInfo *TII = Subtarget.getInstrInfo();
// If this is a signed comparison and the value being compared is not known
// to be sign extended, sign extend it here.
DebugLoc dl = MI.getDebugLoc();
MachineFunction *F = BB->getParent();
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register incr = MI.getOperand(3).getReg();
bool IsSignExtended = Register::isVirtualRegister(incr) &&
isSignExtended(*RegInfo.getVRegDef(incr), TII);
if (CmpOpcode == PPC::CMPW && !IsSignExtended) {
Register ValueReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
BuildMI(*BB, MI, dl, TII->get(is8bit ? PPC::EXTSB : PPC::EXTSH), ValueReg)
.addReg(MI.getOperand(3).getReg());
MI.getOperand(3).setReg(ValueReg);
}
// If we support part-word atomic mnemonics, just use them
if (Subtarget.hasPartwordAtomics())
return EmitAtomicBinary(MI, BB, is8bit ? 1 : 2, BinOpcode, CmpOpcode,
CmpPred);
// In 64 bit mode we have to use 64 bits for addresses, even though the
// lwarx/stwcx are 32 bits. With the 32-bit atomics we can use address
// registers without caring whether they're 32 or 64, but here we're
// doing actual arithmetic on the addresses.
bool is64bit = Subtarget.isPPC64();
bool isLittleEndian = Subtarget.isLittleEndian();
unsigned ZeroReg = is64bit ? PPC::ZERO8 : PPC::ZERO;
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = ++BB->getIterator();
Register dest = MI.getOperand(0).getReg();
Register ptrA = MI.getOperand(1).getReg();
Register ptrB = MI.getOperand(2).getReg();
MachineBasicBlock *loopMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB =
CmpOpcode ? F->CreateMachineBasicBlock(LLVM_BB) : nullptr;
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, loopMBB);
if (CmpOpcode)
F->insert(It, loop2MBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
const TargetRegisterClass *RC =
is64bit ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
Register PtrReg = RegInfo.createVirtualRegister(RC);
Register Shift1Reg = RegInfo.createVirtualRegister(GPRC);
Register ShiftReg =
isLittleEndian ? Shift1Reg : RegInfo.createVirtualRegister(GPRC);
Register Incr2Reg = RegInfo.createVirtualRegister(GPRC);
Register MaskReg = RegInfo.createVirtualRegister(GPRC);
Register Mask2Reg = RegInfo.createVirtualRegister(GPRC);
Register Mask3Reg = RegInfo.createVirtualRegister(GPRC);
Register Tmp2Reg = RegInfo.createVirtualRegister(GPRC);
Register Tmp3Reg = RegInfo.createVirtualRegister(GPRC);
Register Tmp4Reg = RegInfo.createVirtualRegister(GPRC);
Register TmpDestReg = RegInfo.createVirtualRegister(GPRC);
Register Ptr1Reg;
Register TmpReg =
(!BinOpcode) ? Incr2Reg : RegInfo.createVirtualRegister(GPRC);
// thisMBB:
// ...
// fallthrough --> loopMBB
BB->addSuccessor(loopMBB);
// The 4-byte load must be aligned, while a char or short may be
// anywhere in the word. Hence all this nasty bookkeeping code.
// add ptr1, ptrA, ptrB [copy if ptrA==0]
// rlwinm shift1, ptr1, 3, 27, 28 [3, 27, 27]
// xori shift, shift1, 24 [16]
// rlwinm ptr, ptr1, 0, 0, 29
// slw incr2, incr, shift
// li mask2, 255 [li mask3, 0; ori mask2, mask3, 65535]
// slw mask, mask2, shift
// loopMBB:
// lwarx tmpDest, ptr
// add tmp, tmpDest, incr2
// andc tmp2, tmpDest, mask
// and tmp3, tmp, mask
// or tmp4, tmp3, tmp2
// stwcx. tmp4, ptr
// bne- loopMBB
// fallthrough --> exitMBB
// srw dest, tmpDest, shift
if (ptrA != ZeroReg) {
Ptr1Reg = RegInfo.createVirtualRegister(RC);
BuildMI(BB, dl, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
.addReg(ptrA)
.addReg(ptrB);
} else {
Ptr1Reg = ptrB;
}
// We need use 32-bit subregister to avoid mismatch register class in 64-bit
// mode.
BuildMI(BB, dl, TII->get(PPC::RLWINM), Shift1Reg)
.addReg(Ptr1Reg, 0, is64bit ? PPC::sub_32 : 0)
.addImm(3)
.addImm(27)
.addImm(is8bit ? 28 : 27);
if (!isLittleEndian)
BuildMI(BB, dl, TII->get(PPC::XORI), ShiftReg)
.addReg(Shift1Reg)
.addImm(is8bit ? 24 : 16);
if (is64bit)
BuildMI(BB, dl, TII->get(PPC::RLDICR), PtrReg)
.addReg(Ptr1Reg)
.addImm(0)
.addImm(61);
else
BuildMI(BB, dl, TII->get(PPC::RLWINM), PtrReg)
.addReg(Ptr1Reg)
.addImm(0)
.addImm(0)
.addImm(29);
BuildMI(BB, dl, TII->get(PPC::SLW), Incr2Reg).addReg(incr).addReg(ShiftReg);
if (is8bit)
BuildMI(BB, dl, TII->get(PPC::LI), Mask2Reg).addImm(255);
else {
BuildMI(BB, dl, TII->get(PPC::LI), Mask3Reg).addImm(0);
BuildMI(BB, dl, TII->get(PPC::ORI), Mask2Reg)
.addReg(Mask3Reg)
.addImm(65535);
}
BuildMI(BB, dl, TII->get(PPC::SLW), MaskReg)
.addReg(Mask2Reg)
.addReg(ShiftReg);
BB = loopMBB;
BuildMI(BB, dl, TII->get(PPC::LWARX), TmpDestReg)
.addReg(ZeroReg)
.addReg(PtrReg);
if (BinOpcode)
BuildMI(BB, dl, TII->get(BinOpcode), TmpReg)
.addReg(Incr2Reg)
.addReg(TmpDestReg);
BuildMI(BB, dl, TII->get(PPC::ANDC), Tmp2Reg)
.addReg(TmpDestReg)
.addReg(MaskReg);
BuildMI(BB, dl, TII->get(PPC::AND), Tmp3Reg).addReg(TmpReg).addReg(MaskReg);
if (CmpOpcode) {
// For unsigned comparisons, we can directly compare the shifted values.
// For signed comparisons we shift and sign extend.
Register SReg = RegInfo.createVirtualRegister(GPRC);
BuildMI(BB, dl, TII->get(PPC::AND), SReg)
.addReg(TmpDestReg)
.addReg(MaskReg);
unsigned ValueReg = SReg;
unsigned CmpReg = Incr2Reg;
if (CmpOpcode == PPC::CMPW) {
ValueReg = RegInfo.createVirtualRegister(GPRC);
BuildMI(BB, dl, TII->get(PPC::SRW), ValueReg)
.addReg(SReg)
.addReg(ShiftReg);
Register ValueSReg = RegInfo.createVirtualRegister(GPRC);
BuildMI(BB, dl, TII->get(is8bit ? PPC::EXTSB : PPC::EXTSH), ValueSReg)
.addReg(ValueReg);
ValueReg = ValueSReg;
CmpReg = incr;
}
BuildMI(BB, dl, TII->get(CmpOpcode), PPC::CR0)
.addReg(CmpReg)
.addReg(ValueReg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(CmpPred)
.addReg(PPC::CR0)
.addMBB(exitMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(exitMBB);
BB = loop2MBB;
}
BuildMI(BB, dl, TII->get(PPC::OR), Tmp4Reg).addReg(Tmp3Reg).addReg(Tmp2Reg);
BuildMI(BB, dl, TII->get(PPC::STWCX))
.addReg(Tmp4Reg)
.addReg(ZeroReg)
.addReg(PtrReg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(PPC::CR0)
.addMBB(loopMBB);
BB->addSuccessor(loopMBB);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
BuildMI(*BB, BB->begin(), dl, TII->get(PPC::SRW), dest)
.addReg(TmpDestReg)
.addReg(ShiftReg);
return BB;
}
llvm::MachineBasicBlock *
PPCTargetLowering::emitEHSjLjSetJmp(MachineInstr &MI,
MachineBasicBlock *MBB) const {
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
const PPCRegisterInfo *TRI = Subtarget.getRegisterInfo();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
const BasicBlock *BB = MBB->getBasicBlock();
MachineFunction::iterator I = ++MBB->getIterator();
Register DstReg = MI.getOperand(0).getReg();
const TargetRegisterClass *RC = MRI.getRegClass(DstReg);
assert(TRI->isTypeLegalForClass(*RC, MVT::i32) && "Invalid destination!");
Register mainDstReg = MRI.createVirtualRegister(RC);
Register restoreDstReg = MRI.createVirtualRegister(RC);
MVT PVT = getPointerTy(MF->getDataLayout());
assert((PVT == MVT::i64 || PVT == MVT::i32) &&
"Invalid Pointer Size!");
// For v = setjmp(buf), we generate
//
// thisMBB:
// SjLjSetup mainMBB
// bl mainMBB
// v_restore = 1
// b sinkMBB
//
// mainMBB:
// buf[LabelOffset] = LR
// v_main = 0
//
// sinkMBB:
// v = phi(main, restore)
//
MachineBasicBlock *thisMBB = MBB;
MachineBasicBlock *mainMBB = MF->CreateMachineBasicBlock(BB);
MachineBasicBlock *sinkMBB = MF->CreateMachineBasicBlock(BB);
MF->insert(I, mainMBB);
MF->insert(I, sinkMBB);
MachineInstrBuilder MIB;
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), MBB,
std::next(MachineBasicBlock::iterator(MI)), MBB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(MBB);
// Note that the structure of the jmp_buf used here is not compatible
// with that used by libc, and is not designed to be. Specifically, it
// stores only those 'reserved' registers that LLVM does not otherwise
// understand how to spill. Also, by convention, by the time this
// intrinsic is called, Clang has already stored the frame address in the
// first slot of the buffer and stack address in the third. Following the
// X86 target code, we'll store the jump address in the second slot. We also
// need to save the TOC pointer (R2) to handle jumps between shared
// libraries, and that will be stored in the fourth slot. The thread
// identifier (R13) is not affected.
// thisMBB:
const int64_t LabelOffset = 1 * PVT.getStoreSize();
const int64_t TOCOffset = 3 * PVT.getStoreSize();
const int64_t BPOffset = 4 * PVT.getStoreSize();
// Prepare IP either in reg.
const TargetRegisterClass *PtrRC = getRegClassFor(PVT);
Register LabelReg = MRI.createVirtualRegister(PtrRC);
Register BufReg = MI.getOperand(1).getReg();
if (Subtarget.is64BitELFABI()) {
setUsesTOCBasePtr(*MBB->getParent());
MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::STD))
.addReg(PPC::X2)
.addImm(TOCOffset)
.addReg(BufReg)
.cloneMemRefs(MI);
}
// Naked functions never have a base pointer, and so we use r1. For all
// other functions, this decision must be delayed until during PEI.
unsigned BaseReg;
if (MF->getFunction().hasFnAttribute(Attribute::Naked))
BaseReg = Subtarget.isPPC64() ? PPC::X1 : PPC::R1;
else
BaseReg = Subtarget.isPPC64() ? PPC::BP8 : PPC::BP;
MIB = BuildMI(*thisMBB, MI, DL,
TII->get(Subtarget.isPPC64() ? PPC::STD : PPC::STW))
.addReg(BaseReg)
.addImm(BPOffset)
.addReg(BufReg)
.cloneMemRefs(MI);
// Setup
MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::BCLalways)).addMBB(mainMBB);
MIB.addRegMask(TRI->getNoPreservedMask());
BuildMI(*thisMBB, MI, DL, TII->get(PPC::LI), restoreDstReg).addImm(1);
MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::EH_SjLj_Setup))
.addMBB(mainMBB);
MIB = BuildMI(*thisMBB, MI, DL, TII->get(PPC::B)).addMBB(sinkMBB);
thisMBB->addSuccessor(mainMBB, BranchProbability::getZero());
thisMBB->addSuccessor(sinkMBB, BranchProbability::getOne());
// mainMBB:
// mainDstReg = 0
MIB =
BuildMI(mainMBB, DL,
TII->get(Subtarget.isPPC64() ? PPC::MFLR8 : PPC::MFLR), LabelReg);
// Store IP
if (Subtarget.isPPC64()) {
MIB = BuildMI(mainMBB, DL, TII->get(PPC::STD))
.addReg(LabelReg)
.addImm(LabelOffset)
.addReg(BufReg);
} else {
MIB = BuildMI(mainMBB, DL, TII->get(PPC::STW))
.addReg(LabelReg)
.addImm(LabelOffset)
.addReg(BufReg);
}
MIB.cloneMemRefs(MI);
BuildMI(mainMBB, DL, TII->get(PPC::LI), mainDstReg).addImm(0);
mainMBB->addSuccessor(sinkMBB);
// sinkMBB:
BuildMI(*sinkMBB, sinkMBB->begin(), DL,
TII->get(PPC::PHI), DstReg)
.addReg(mainDstReg).addMBB(mainMBB)
.addReg(restoreDstReg).addMBB(thisMBB);
MI.eraseFromParent();
return sinkMBB;
}
MachineBasicBlock *
PPCTargetLowering::emitEHSjLjLongJmp(MachineInstr &MI,
MachineBasicBlock *MBB) const {
DebugLoc DL = MI.getDebugLoc();
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
MachineFunction *MF = MBB->getParent();
MachineRegisterInfo &MRI = MF->getRegInfo();
MVT PVT = getPointerTy(MF->getDataLayout());
assert((PVT == MVT::i64 || PVT == MVT::i32) &&
"Invalid Pointer Size!");
const TargetRegisterClass *RC =
(PVT == MVT::i64) ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
Register Tmp = MRI.createVirtualRegister(RC);
// Since FP is only updated here but NOT referenced, it's treated as GPR.
unsigned FP = (PVT == MVT::i64) ? PPC::X31 : PPC::R31;
unsigned SP = (PVT == MVT::i64) ? PPC::X1 : PPC::R1;
unsigned BP =
(PVT == MVT::i64)
? PPC::X30
: (Subtarget.isSVR4ABI() && isPositionIndependent() ? PPC::R29
: PPC::R30);
MachineInstrBuilder MIB;
const int64_t LabelOffset = 1 * PVT.getStoreSize();
const int64_t SPOffset = 2 * PVT.getStoreSize();
const int64_t TOCOffset = 3 * PVT.getStoreSize();
const int64_t BPOffset = 4 * PVT.getStoreSize();
Register BufReg = MI.getOperand(0).getReg();
// Reload FP (the jumped-to function may not have had a
// frame pointer, and if so, then its r31 will be restored
// as necessary).
if (PVT == MVT::i64) {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), FP)
.addImm(0)
.addReg(BufReg);
} else {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), FP)
.addImm(0)
.addReg(BufReg);
}
MIB.cloneMemRefs(MI);
// Reload IP
if (PVT == MVT::i64) {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), Tmp)
.addImm(LabelOffset)
.addReg(BufReg);
} else {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), Tmp)
.addImm(LabelOffset)
.addReg(BufReg);
}
MIB.cloneMemRefs(MI);
// Reload SP
if (PVT == MVT::i64) {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), SP)
.addImm(SPOffset)
.addReg(BufReg);
} else {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), SP)
.addImm(SPOffset)
.addReg(BufReg);
}
MIB.cloneMemRefs(MI);
// Reload BP
if (PVT == MVT::i64) {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), BP)
.addImm(BPOffset)
.addReg(BufReg);
} else {
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LWZ), BP)
.addImm(BPOffset)
.addReg(BufReg);
}
MIB.cloneMemRefs(MI);
// Reload TOC
if (PVT == MVT::i64 && Subtarget.isSVR4ABI()) {
setUsesTOCBasePtr(*MBB->getParent());
MIB = BuildMI(*MBB, MI, DL, TII->get(PPC::LD), PPC::X2)
.addImm(TOCOffset)
.addReg(BufReg)
.cloneMemRefs(MI);
}
// Jump
BuildMI(*MBB, MI, DL,
TII->get(PVT == MVT::i64 ? PPC::MTCTR8 : PPC::MTCTR)).addReg(Tmp);
BuildMI(*MBB, MI, DL, TII->get(PVT == MVT::i64 ? PPC::BCTR8 : PPC::BCTR));
MI.eraseFromParent();
return MBB;
}
bool PPCTargetLowering::hasInlineStackProbe(MachineFunction &MF) const {
// If the function specifically requests inline stack probes, emit them.
if (MF.getFunction().hasFnAttribute("probe-stack"))
return MF.getFunction().getFnAttribute("probe-stack").getValueAsString() ==
"inline-asm";
return false;
}
unsigned PPCTargetLowering::getStackProbeSize(MachineFunction &MF) const {
const TargetFrameLowering *TFI = Subtarget.getFrameLowering();
unsigned StackAlign = TFI->getStackAlignment();
assert(StackAlign >= 1 && isPowerOf2_32(StackAlign) &&
"Unexpected stack alignment");
// The default stack probe size is 4096 if the function has no
// stack-probe-size attribute.
unsigned StackProbeSize = 4096;
const Function &Fn = MF.getFunction();
if (Fn.hasFnAttribute("stack-probe-size"))
Fn.getFnAttribute("stack-probe-size")
.getValueAsString()
.getAsInteger(0, StackProbeSize);
// Round down to the stack alignment.
StackProbeSize &= ~(StackAlign - 1);
return StackProbeSize ? StackProbeSize : StackAlign;
}
// Lower dynamic stack allocation with probing. `emitProbedAlloca` is splitted
// into three phases. In the first phase, it uses pseudo instruction
// PREPARE_PROBED_ALLOCA to get the future result of actual FramePointer and
// FinalStackPtr. In the second phase, it generates a loop for probing blocks.
// At last, it uses pseudo instruction DYNAREAOFFSET to get the future result of
// MaxCallFrameSize so that it can calculate correct data area pointer.
MachineBasicBlock *
PPCTargetLowering::emitProbedAlloca(MachineInstr &MI,
MachineBasicBlock *MBB) const {
const bool isPPC64 = Subtarget.isPPC64();
MachineFunction *MF = MBB->getParent();
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
DebugLoc DL = MI.getDebugLoc();
const unsigned ProbeSize = getStackProbeSize(*MF);
const BasicBlock *ProbedBB = MBB->getBasicBlock();
MachineRegisterInfo &MRI = MF->getRegInfo();
// The CFG of probing stack looks as
// +-----+
// | MBB |
// +--+--+
// |
// +----v----+
// +--->+ TestMBB +---+
// | +----+----+ |
// | | |
// | +-----v----+ |
// +---+ BlockMBB | |
// +----------+ |
// |
// +---------+ |
// | TailMBB +<--+
// +---------+
// In MBB, calculate previous frame pointer and final stack pointer.
// In TestMBB, test if sp is equal to final stack pointer, if so, jump to
// TailMBB. In BlockMBB, update the sp atomically and jump back to TestMBB.
// TailMBB is spliced via \p MI.
MachineBasicBlock *TestMBB = MF->CreateMachineBasicBlock(ProbedBB);
MachineBasicBlock *TailMBB = MF->CreateMachineBasicBlock(ProbedBB);
MachineBasicBlock *BlockMBB = MF->CreateMachineBasicBlock(ProbedBB);
MachineFunction::iterator MBBIter = ++MBB->getIterator();
MF->insert(MBBIter, TestMBB);
MF->insert(MBBIter, BlockMBB);
MF->insert(MBBIter, TailMBB);
const TargetRegisterClass *G8RC = &PPC::G8RCRegClass;
const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
Register DstReg = MI.getOperand(0).getReg();
Register NegSizeReg = MI.getOperand(1).getReg();
Register SPReg = isPPC64 ? PPC::X1 : PPC::R1;
Register FinalStackPtr = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
Register FramePointer = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
Register ActualNegSizeReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
// Since value of NegSizeReg might be realigned in prologepilog, insert a
// PREPARE_PROBED_ALLOCA pseudo instruction to get actual FramePointer and
// NegSize.
unsigned ProbeOpc;
if (!MRI.hasOneNonDBGUse(NegSizeReg))
ProbeOpc =
isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_64 : PPC::PREPARE_PROBED_ALLOCA_32;
else
// By introducing PREPARE_PROBED_ALLOCA_NEGSIZE_OPT, ActualNegSizeReg
// and NegSizeReg will be allocated in the same phyreg to avoid
// redundant copy when NegSizeReg has only one use which is current MI and
// will be replaced by PREPARE_PROBED_ALLOCA then.
ProbeOpc = isPPC64 ? PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_64
: PPC::PREPARE_PROBED_ALLOCA_NEGSIZE_SAME_REG_32;
BuildMI(*MBB, {MI}, DL, TII->get(ProbeOpc), FramePointer)
.addDef(ActualNegSizeReg)
.addReg(NegSizeReg)
.add(MI.getOperand(2))
.add(MI.getOperand(3));
// Calculate final stack pointer, which equals to SP + ActualNegSize.
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::ADD8 : PPC::ADD4),
FinalStackPtr)
.addReg(SPReg)
.addReg(ActualNegSizeReg);
// Materialize a scratch register for update.
int64_t NegProbeSize = -(int64_t)ProbeSize;
assert(isInt<32>(NegProbeSize) && "Unhandled probe size!");
Register ScratchReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
if (!isInt<16>(NegProbeSize)) {
Register TempReg = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::LIS8 : PPC::LIS), TempReg)
.addImm(NegProbeSize >> 16);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::ORI8 : PPC::ORI),
ScratchReg)
.addReg(TempReg)
.addImm(NegProbeSize & 0xFFFF);
} else
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::LI8 : PPC::LI), ScratchReg)
.addImm(NegProbeSize);
{
// Probing leading residual part.
Register Div = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::DIVD : PPC::DIVW), Div)
.addReg(ActualNegSizeReg)
.addReg(ScratchReg);
Register Mul = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::MULLD : PPC::MULLW), Mul)
.addReg(Div)
.addReg(ScratchReg);
Register NegMod = MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::SUBF8 : PPC::SUBF), NegMod)
.addReg(Mul)
.addReg(ActualNegSizeReg);
BuildMI(*MBB, {MI}, DL, TII->get(isPPC64 ? PPC::STDUX : PPC::STWUX), SPReg)
.addReg(FramePointer)
.addReg(SPReg)
.addReg(NegMod);
}
{
// Remaining part should be multiple of ProbeSize.
Register CmpResult = MRI.createVirtualRegister(&PPC::CRRCRegClass);
BuildMI(TestMBB, DL, TII->get(isPPC64 ? PPC::CMPD : PPC::CMPW), CmpResult)
.addReg(SPReg)
.addReg(FinalStackPtr);
BuildMI(TestMBB, DL, TII->get(PPC::BCC))
.addImm(PPC::PRED_EQ)
.addReg(CmpResult)
.addMBB(TailMBB);
TestMBB->addSuccessor(BlockMBB);
TestMBB->addSuccessor(TailMBB);
}
{
// Touch the block.
// |P...|P...|P...
BuildMI(BlockMBB, DL, TII->get(isPPC64 ? PPC::STDUX : PPC::STWUX), SPReg)
.addReg(FramePointer)
.addReg(SPReg)
.addReg(ScratchReg);
BuildMI(BlockMBB, DL, TII->get(PPC::B)).addMBB(TestMBB);
BlockMBB->addSuccessor(TestMBB);
}
// Calculation of MaxCallFrameSize is deferred to prologepilog, use
// DYNAREAOFFSET pseudo instruction to get the future result.
Register MaxCallFrameSizeReg =
MRI.createVirtualRegister(isPPC64 ? G8RC : GPRC);
BuildMI(TailMBB, DL,
TII->get(isPPC64 ? PPC::DYNAREAOFFSET8 : PPC::DYNAREAOFFSET),
MaxCallFrameSizeReg)
.add(MI.getOperand(2))
.add(MI.getOperand(3));
BuildMI(TailMBB, DL, TII->get(isPPC64 ? PPC::ADD8 : PPC::ADD4), DstReg)
.addReg(SPReg)
.addReg(MaxCallFrameSizeReg);
// Splice instructions after MI to TailMBB.
TailMBB->splice(TailMBB->end(), MBB,
std::next(MachineBasicBlock::iterator(MI)), MBB->end());
TailMBB->transferSuccessorsAndUpdatePHIs(MBB);
MBB->addSuccessor(TestMBB);
// Delete the pseudo instruction.
MI.eraseFromParent();
++NumDynamicAllocaProbed;
return TailMBB;
}
MachineBasicBlock *
PPCTargetLowering::EmitInstrWithCustomInserter(MachineInstr &MI,
MachineBasicBlock *BB) const {
if (MI.getOpcode() == TargetOpcode::STACKMAP ||
MI.getOpcode() == TargetOpcode::PATCHPOINT) {
if (Subtarget.is64BitELFABI() &&
MI.getOpcode() == TargetOpcode::PATCHPOINT &&
!Subtarget.isUsingPCRelativeCalls()) {
// Call lowering should have added an r2 operand to indicate a dependence
// on the TOC base pointer value. It can't however, because there is no
// way to mark the dependence as implicit there, and so the stackmap code
// will confuse it with a regular operand. Instead, add the dependence
// here.
MI.addOperand(MachineOperand::CreateReg(PPC::X2, false, true));
}
return emitPatchPoint(MI, BB);
}
if (MI.getOpcode() == PPC::EH_SjLj_SetJmp32 ||
MI.getOpcode() == PPC::EH_SjLj_SetJmp64) {
return emitEHSjLjSetJmp(MI, BB);
} else if (MI.getOpcode() == PPC::EH_SjLj_LongJmp32 ||
MI.getOpcode() == PPC::EH_SjLj_LongJmp64) {
return emitEHSjLjLongJmp(MI, BB);
}
const TargetInstrInfo *TII = Subtarget.getInstrInfo();
// To "insert" these instructions we actually have to insert their
// control-flow patterns.
const BasicBlock *LLVM_BB = BB->getBasicBlock();
MachineFunction::iterator It = ++BB->getIterator();
MachineFunction *F = BB->getParent();
if (MI.getOpcode() == PPC::SELECT_CC_I4 ||
MI.getOpcode() == PPC::SELECT_CC_I8 || MI.getOpcode() == PPC::SELECT_I4 ||
MI.getOpcode() == PPC::SELECT_I8) {
SmallVector<MachineOperand, 2> Cond;
if (MI.getOpcode() == PPC::SELECT_CC_I4 ||
MI.getOpcode() == PPC::SELECT_CC_I8)
Cond.push_back(MI.getOperand(4));
else
Cond.push_back(MachineOperand::CreateImm(PPC::PRED_BIT_SET));
Cond.push_back(MI.getOperand(1));
DebugLoc dl = MI.getDebugLoc();
TII->insertSelect(*BB, MI, dl, MI.getOperand(0).getReg(), Cond,
MI.getOperand(2).getReg(), MI.getOperand(3).getReg());
} else if (MI.getOpcode() == PPC::SELECT_CC_F4 ||
MI.getOpcode() == PPC::SELECT_CC_F8 ||
MI.getOpcode() == PPC::SELECT_CC_F16 ||
MI.getOpcode() == PPC::SELECT_CC_VRRC ||
MI.getOpcode() == PPC::SELECT_CC_VSFRC ||
MI.getOpcode() == PPC::SELECT_CC_VSSRC ||
MI.getOpcode() == PPC::SELECT_CC_VSRC ||
MI.getOpcode() == PPC::SELECT_CC_SPE4 ||
MI.getOpcode() == PPC::SELECT_CC_SPE ||
MI.getOpcode() == PPC::SELECT_F4 ||
MI.getOpcode() == PPC::SELECT_F8 ||
MI.getOpcode() == PPC::SELECT_F16 ||
MI.getOpcode() == PPC::SELECT_SPE ||
MI.getOpcode() == PPC::SELECT_SPE4 ||
MI.getOpcode() == PPC::SELECT_VRRC ||
MI.getOpcode() == PPC::SELECT_VSFRC ||
MI.getOpcode() == PPC::SELECT_VSSRC ||
MI.getOpcode() == PPC::SELECT_VSRC) {
// The incoming instruction knows the destination vreg to set, the
// condition code register to branch on, the true/false values to
// select between, and a branch opcode to use.
// thisMBB:
// ...
// TrueVal = ...
// cmpTY ccX, r1, r2
// bCC copy1MBB
// fallthrough --> copy0MBB
MachineBasicBlock *thisMBB = BB;
MachineBasicBlock *copy0MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
DebugLoc dl = MI.getDebugLoc();
F->insert(It, copy0MBB);
F->insert(It, sinkMBB);
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
// Next, add the true and fallthrough blocks as its successors.
BB->addSuccessor(copy0MBB);
BB->addSuccessor(sinkMBB);
if (MI.getOpcode() == PPC::SELECT_I4 || MI.getOpcode() == PPC::SELECT_I8 ||
MI.getOpcode() == PPC::SELECT_F4 || MI.getOpcode() == PPC::SELECT_F8 ||
MI.getOpcode() == PPC::SELECT_F16 ||
MI.getOpcode() == PPC::SELECT_SPE4 ||
MI.getOpcode() == PPC::SELECT_SPE ||
MI.getOpcode() == PPC::SELECT_VRRC ||
MI.getOpcode() == PPC::SELECT_VSFRC ||
MI.getOpcode() == PPC::SELECT_VSSRC ||
MI.getOpcode() == PPC::SELECT_VSRC) {
BuildMI(BB, dl, TII->get(PPC::BC))
.addReg(MI.getOperand(1).getReg())
.addMBB(sinkMBB);
} else {
unsigned SelectPred = MI.getOperand(4).getImm();
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(SelectPred)
.addReg(MI.getOperand(1).getReg())
.addMBB(sinkMBB);
}
// copy0MBB:
// %FalseValue = ...
// # fallthrough to sinkMBB
BB = copy0MBB;
// Update machine-CFG edges
BB->addSuccessor(sinkMBB);
// sinkMBB:
// %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
// ...
BB = sinkMBB;
BuildMI(*BB, BB->begin(), dl, TII->get(PPC::PHI), MI.getOperand(0).getReg())
.addReg(MI.getOperand(3).getReg())
.addMBB(copy0MBB)
.addReg(MI.getOperand(2).getReg())
.addMBB(thisMBB);
} else if (MI.getOpcode() == PPC::ReadTB) {
// To read the 64-bit time-base register on a 32-bit target, we read the
// two halves. Should the counter have wrapped while it was being read, we
// need to try again.
// ...
// readLoop:
// mfspr Rx,TBU # load from TBU
// mfspr Ry,TB # load from TB
// mfspr Rz,TBU # load from TBU
// cmpw crX,Rx,Rz # check if 'old'='new'
// bne readLoop # branch if they're not equal
// ...
MachineBasicBlock *readMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *sinkMBB = F->CreateMachineBasicBlock(LLVM_BB);
DebugLoc dl = MI.getDebugLoc();
F->insert(It, readMBB);
F->insert(It, sinkMBB);
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
BB->addSuccessor(readMBB);
BB = readMBB;
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register ReadAgainReg = RegInfo.createVirtualRegister(&PPC::GPRCRegClass);
Register LoReg = MI.getOperand(0).getReg();
Register HiReg = MI.getOperand(1).getReg();
BuildMI(BB, dl, TII->get(PPC::MFSPR), HiReg).addImm(269);
BuildMI(BB, dl, TII->get(PPC::MFSPR), LoReg).addImm(268);
BuildMI(BB, dl, TII->get(PPC::MFSPR), ReadAgainReg).addImm(269);
Register CmpReg = RegInfo.createVirtualRegister(&PPC::CRRCRegClass);
BuildMI(BB, dl, TII->get(PPC::CMPW), CmpReg)
.addReg(HiReg)
.addReg(ReadAgainReg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(CmpReg)
.addMBB(readMBB);
BB->addSuccessor(readMBB);
BB->addSuccessor(sinkMBB);
} else if (MI.getOpcode() == PPC::ATOMIC_LOAD_ADD_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::ADD4);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_ADD_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::ADD4);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_ADD_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::ADD4);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_ADD_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::ADD8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_AND_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::AND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_AND_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::AND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_AND_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::AND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_AND_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::AND8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_OR_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::OR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_OR_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::OR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_OR_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::OR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_OR_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::OR8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_XOR_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::XOR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_XOR_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::XOR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_XOR_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::XOR);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_XOR_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::XOR8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_NAND_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::NAND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_NAND_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::NAND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_NAND_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::NAND);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_NAND_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::NAND8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_SUB_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, PPC::SUBF);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_SUB_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, PPC::SUBF);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_SUB_I32)
BB = EmitAtomicBinary(MI, BB, 4, PPC::SUBF);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_SUB_I64)
BB = EmitAtomicBinary(MI, BB, 8, PPC::SUBF8);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I32)
BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MIN_I64)
BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPD, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I32)
BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_MAX_I64)
BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPD, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPLW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPLW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I32)
BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPLW, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMIN_I64)
BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPLD, PPC::PRED_GE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, 0, PPC::CMPLW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, 0, PPC::CMPLW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I32)
BB = EmitAtomicBinary(MI, BB, 4, 0, PPC::CMPLW, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_LOAD_UMAX_I64)
BB = EmitAtomicBinary(MI, BB, 8, 0, PPC::CMPLD, PPC::PRED_LE);
else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I8)
BB = EmitPartwordAtomicBinary(MI, BB, true, 0);
else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I16)
BB = EmitPartwordAtomicBinary(MI, BB, false, 0);
else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I32)
BB = EmitAtomicBinary(MI, BB, 4, 0);
else if (MI.getOpcode() == PPC::ATOMIC_SWAP_I64)
BB = EmitAtomicBinary(MI, BB, 8, 0);
else if (MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I32 ||
MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I64 ||
(Subtarget.hasPartwordAtomics() &&
MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I8) ||
(Subtarget.hasPartwordAtomics() &&
MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I16)) {
bool is64bit = MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I64;
auto LoadMnemonic = PPC::LDARX;
auto StoreMnemonic = PPC::STDCX;
switch (MI.getOpcode()) {
default:
llvm_unreachable("Compare and swap of unknown size");
case PPC::ATOMIC_CMP_SWAP_I8:
LoadMnemonic = PPC::LBARX;
StoreMnemonic = PPC::STBCX;
assert(Subtarget.hasPartwordAtomics() && "No support partword atomics.");
break;
case PPC::ATOMIC_CMP_SWAP_I16:
LoadMnemonic = PPC::LHARX;
StoreMnemonic = PPC::STHCX;
assert(Subtarget.hasPartwordAtomics() && "No support partword atomics.");
break;
case PPC::ATOMIC_CMP_SWAP_I32:
LoadMnemonic = PPC::LWARX;
StoreMnemonic = PPC::STWCX;
break;
case PPC::ATOMIC_CMP_SWAP_I64:
LoadMnemonic = PPC::LDARX;
StoreMnemonic = PPC::STDCX;
break;
}
Register dest = MI.getOperand(0).getReg();
Register ptrA = MI.getOperand(1).getReg();
Register ptrB = MI.getOperand(2).getReg();
Register oldval = MI.getOperand(3).getReg();
Register newval = MI.getOperand(4).getReg();
DebugLoc dl = MI.getDebugLoc();
MachineBasicBlock *loop1MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *midMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, loop1MBB);
F->insert(It, loop2MBB);
F->insert(It, midMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
// thisMBB:
// ...
// fallthrough --> loopMBB
BB->addSuccessor(loop1MBB);
// loop1MBB:
// l[bhwd]arx dest, ptr
// cmp[wd] dest, oldval
// bne- midMBB
// loop2MBB:
// st[bhwd]cx. newval, ptr
// bne- loopMBB
// b exitBB
// midMBB:
// st[bhwd]cx. dest, ptr
// exitBB:
BB = loop1MBB;
BuildMI(BB, dl, TII->get(LoadMnemonic), dest).addReg(ptrA).addReg(ptrB);
BuildMI(BB, dl, TII->get(is64bit ? PPC::CMPD : PPC::CMPW), PPC::CR0)
.addReg(oldval)
.addReg(dest);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(PPC::CR0)
.addMBB(midMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(midMBB);
BB = loop2MBB;
BuildMI(BB, dl, TII->get(StoreMnemonic))
.addReg(newval)
.addReg(ptrA)
.addReg(ptrB);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(PPC::CR0)
.addMBB(loop1MBB);
BuildMI(BB, dl, TII->get(PPC::B)).addMBB(exitMBB);
BB->addSuccessor(loop1MBB);
BB->addSuccessor(exitMBB);
BB = midMBB;
BuildMI(BB, dl, TII->get(StoreMnemonic))
.addReg(dest)
.addReg(ptrA)
.addReg(ptrB);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
} else if (MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I8 ||
MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I16) {
// We must use 64-bit registers for addresses when targeting 64-bit,
// since we're actually doing arithmetic on them. Other registers
// can be 32-bit.
bool is64bit = Subtarget.isPPC64();
bool isLittleEndian = Subtarget.isLittleEndian();
bool is8bit = MI.getOpcode() == PPC::ATOMIC_CMP_SWAP_I8;
Register dest = MI.getOperand(0).getReg();
Register ptrA = MI.getOperand(1).getReg();
Register ptrB = MI.getOperand(2).getReg();
Register oldval = MI.getOperand(3).getReg();
Register newval = MI.getOperand(4).getReg();
DebugLoc dl = MI.getDebugLoc();
MachineBasicBlock *loop1MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *loop2MBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *midMBB = F->CreateMachineBasicBlock(LLVM_BB);
MachineBasicBlock *exitMBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, loop1MBB);
F->insert(It, loop2MBB);
F->insert(It, midMBB);
F->insert(It, exitMBB);
exitMBB->splice(exitMBB->begin(), BB,
std::next(MachineBasicBlock::iterator(MI)), BB->end());
exitMBB->transferSuccessorsAndUpdatePHIs(BB);
MachineRegisterInfo &RegInfo = F->getRegInfo();
const TargetRegisterClass *RC =
is64bit ? &PPC::G8RCRegClass : &PPC::GPRCRegClass;
const TargetRegisterClass *GPRC = &PPC::GPRCRegClass;
Register PtrReg = RegInfo.createVirtualRegister(RC);
Register Shift1Reg = RegInfo.createVirtualRegister(GPRC);
Register ShiftReg =
isLittleEndian ? Shift1Reg : RegInfo.createVirtualRegister(GPRC);
Register NewVal2Reg = RegInfo.createVirtualRegister(GPRC);
Register NewVal3Reg = RegInfo.createVirtualRegister(GPRC);
Register OldVal2Reg = RegInfo.createVirtualRegister(GPRC);
Register OldVal3Reg = RegInfo.createVirtualRegister(GPRC);
Register MaskReg = RegInfo.createVirtualRegister(GPRC);
Register Mask2Reg = RegInfo.createVirtualRegister(GPRC);
Register Mask3Reg = RegInfo.createVirtualRegister(GPRC);
Register Tmp2Reg = RegInfo.createVirtualRegister(GPRC);
Register Tmp4Reg = RegInfo.createVirtualRegister(GPRC);
Register TmpDestReg = RegInfo.createVirtualRegister(GPRC);
Register Ptr1Reg;
Register TmpReg = RegInfo.createVirtualRegister(GPRC);
Register ZeroReg = is64bit ? PPC::ZERO8 : PPC::ZERO;
// thisMBB:
// ...
// fallthrough --> loopMBB
BB->addSuccessor(loop1MBB);
// The 4-byte load must be aligned, while a char or short may be
// anywhere in the word. Hence all this nasty bookkeeping code.
// add ptr1, ptrA, ptrB [copy if ptrA==0]
// rlwinm shift1, ptr1, 3, 27, 28 [3, 27, 27]
// xori shift, shift1, 24 [16]
// rlwinm ptr, ptr1, 0, 0, 29
// slw newval2, newval, shift
// slw oldval2, oldval,shift
// li mask2, 255 [li mask3, 0; ori mask2, mask3, 65535]
// slw mask, mask2, shift
// and newval3, newval2, mask
// and oldval3, oldval2, mask
// loop1MBB:
// lwarx tmpDest, ptr
// and tmp, tmpDest, mask
// cmpw tmp, oldval3
// bne- midMBB
// loop2MBB:
// andc tmp2, tmpDest, mask
// or tmp4, tmp2, newval3
// stwcx. tmp4, ptr
// bne- loop1MBB
// b exitBB
// midMBB:
// stwcx. tmpDest, ptr
// exitBB:
// srw dest, tmpDest, shift
if (ptrA != ZeroReg) {
Ptr1Reg = RegInfo.createVirtualRegister(RC);
BuildMI(BB, dl, TII->get(is64bit ? PPC::ADD8 : PPC::ADD4), Ptr1Reg)
.addReg(ptrA)
.addReg(ptrB);
} else {
Ptr1Reg = ptrB;
}
// We need use 32-bit subregister to avoid mismatch register class in 64-bit
// mode.
BuildMI(BB, dl, TII->get(PPC::RLWINM), Shift1Reg)
.addReg(Ptr1Reg, 0, is64bit ? PPC::sub_32 : 0)
.addImm(3)
.addImm(27)
.addImm(is8bit ? 28 : 27);
if (!isLittleEndian)
BuildMI(BB, dl, TII->get(PPC::XORI), ShiftReg)
.addReg(Shift1Reg)
.addImm(is8bit ? 24 : 16);
if (is64bit)
BuildMI(BB, dl, TII->get(PPC::RLDICR), PtrReg)
.addReg(Ptr1Reg)
.addImm(0)
.addImm(61);
else
BuildMI(BB, dl, TII->get(PPC::RLWINM), PtrReg)
.addReg(Ptr1Reg)
.addImm(0)
.addImm(0)
.addImm(29);
BuildMI(BB, dl, TII->get(PPC::SLW), NewVal2Reg)
.addReg(newval)
.addReg(ShiftReg);
BuildMI(BB, dl, TII->get(PPC::SLW), OldVal2Reg)
.addReg(oldval)
.addReg(ShiftReg);
if (is8bit)
BuildMI(BB, dl, TII->get(PPC::LI), Mask2Reg).addImm(255);
else {
BuildMI(BB, dl, TII->get(PPC::LI), Mask3Reg).addImm(0);
BuildMI(BB, dl, TII->get(PPC::ORI), Mask2Reg)
.addReg(Mask3Reg)
.addImm(65535);
}
BuildMI(BB, dl, TII->get(PPC::SLW), MaskReg)
.addReg(Mask2Reg)
.addReg(ShiftReg);
BuildMI(BB, dl, TII->get(PPC::AND), NewVal3Reg)
.addReg(NewVal2Reg)
.addReg(MaskReg);
BuildMI(BB, dl, TII->get(PPC::AND), OldVal3Reg)
.addReg(OldVal2Reg)
.addReg(MaskReg);
BB = loop1MBB;
BuildMI(BB, dl, TII->get(PPC::LWARX), TmpDestReg)
.addReg(ZeroReg)
.addReg(PtrReg);
BuildMI(BB, dl, TII->get(PPC::AND), TmpReg)
.addReg(TmpDestReg)
.addReg(MaskReg);
BuildMI(BB, dl, TII->get(PPC::CMPW), PPC::CR0)
.addReg(TmpReg)
.addReg(OldVal3Reg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(PPC::CR0)
.addMBB(midMBB);
BB->addSuccessor(loop2MBB);
BB->addSuccessor(midMBB);
BB = loop2MBB;
BuildMI(BB, dl, TII->get(PPC::ANDC), Tmp2Reg)
.addReg(TmpDestReg)
.addReg(MaskReg);
BuildMI(BB, dl, TII->get(PPC::OR), Tmp4Reg)
.addReg(Tmp2Reg)
.addReg(NewVal3Reg);
BuildMI(BB, dl, TII->get(PPC::STWCX))
.addReg(Tmp4Reg)
.addReg(ZeroReg)
.addReg(PtrReg);
BuildMI(BB, dl, TII->get(PPC::BCC))
.addImm(PPC::PRED_NE)
.addReg(PPC::CR0)
.addMBB(loop1MBB);
BuildMI(BB, dl, TII->get(PPC::B)).addMBB(exitMBB);
BB->addSuccessor(loop1MBB);
BB->addSuccessor(exitMBB);
BB = midMBB;
BuildMI(BB, dl, TII->get(PPC::STWCX))
.addReg(TmpDestReg)
.addReg(ZeroReg)
.addReg(PtrReg);
BB->addSuccessor(exitMBB);
// exitMBB:
// ...
BB = exitMBB;
BuildMI(*BB, BB->begin(), dl, TII->get(PPC::SRW), dest)
.addReg(TmpReg)
.addReg(ShiftReg);
} else if (MI.getOpcode() == PPC::FADDrtz) {
// This pseudo performs an FADD with rounding mode temporarily forced
// to round-to-zero. We emit this via custom inserter since the FPSCR
// is not modeled at the SelectionDAG level.
Register Dest = MI.getOperand(0).getReg();
Register Src1 = MI.getOperand(1).getReg();
Register Src2 = MI.getOperand(2).getReg();
DebugLoc dl = MI.getDebugLoc();
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register MFFSReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
// Save FPSCR value.
BuildMI(*BB, MI, dl, TII->get(PPC::MFFS), MFFSReg);
// Set rounding mode to round-to-zero.
BuildMI(*BB, MI, dl, TII->get(PPC::MTFSB1))
.addImm(31)
.addReg(PPC::RM, RegState::ImplicitDefine);
BuildMI(*BB, MI, dl, TII->get(PPC::MTFSB0))
.addImm(30)
.addReg(PPC::RM, RegState::ImplicitDefine);
// Perform addition.
auto MIB = BuildMI(*BB, MI, dl, TII->get(PPC::FADD), Dest)
.addReg(Src1)
.addReg(Src2);
if (MI.getFlag(MachineInstr::NoFPExcept))
MIB.setMIFlag(MachineInstr::NoFPExcept);
// Restore FPSCR value.
BuildMI(*BB, MI, dl, TII->get(PPC::MTFSFb)).addImm(1).addReg(MFFSReg);
} else if (MI.getOpcode() == PPC::ANDI_rec_1_EQ_BIT ||
MI.getOpcode() == PPC::ANDI_rec_1_GT_BIT ||
MI.getOpcode() == PPC::ANDI_rec_1_EQ_BIT8 ||
MI.getOpcode() == PPC::ANDI_rec_1_GT_BIT8) {
unsigned Opcode = (MI.getOpcode() == PPC::ANDI_rec_1_EQ_BIT8 ||
MI.getOpcode() == PPC::ANDI_rec_1_GT_BIT8)
? PPC::ANDI8_rec
: PPC::ANDI_rec;
bool IsEQ = (MI.getOpcode() == PPC::ANDI_rec_1_EQ_BIT ||
MI.getOpcode() == PPC::ANDI_rec_1_EQ_BIT8);
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register Dest = RegInfo.createVirtualRegister(
Opcode == PPC::ANDI_rec ? &PPC::GPRCRegClass : &PPC::G8RCRegClass);
DebugLoc Dl = MI.getDebugLoc();
BuildMI(*BB, MI, Dl, TII->get(Opcode), Dest)
.addReg(MI.getOperand(1).getReg())
.addImm(1);
BuildMI(*BB, MI, Dl, TII->get(TargetOpcode::COPY),
MI.getOperand(0).getReg())
.addReg(IsEQ ? PPC::CR0EQ : PPC::CR0GT);
} else if (MI.getOpcode() == PPC::TCHECK_RET) {
DebugLoc Dl = MI.getDebugLoc();
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register CRReg = RegInfo.createVirtualRegister(&PPC::CRRCRegClass);
BuildMI(*BB, MI, Dl, TII->get(PPC::TCHECK), CRReg);
BuildMI(*BB, MI, Dl, TII->get(TargetOpcode::COPY),
MI.getOperand(0).getReg())
.addReg(CRReg);
} else if (MI.getOpcode() == PPC::TBEGIN_RET) {
DebugLoc Dl = MI.getDebugLoc();
unsigned Imm = MI.getOperand(1).getImm();
BuildMI(*BB, MI, Dl, TII->get(PPC::TBEGIN)).addImm(Imm);
BuildMI(*BB, MI, Dl, TII->get(TargetOpcode::COPY),
MI.getOperand(0).getReg())
.addReg(PPC::CR0EQ);
} else if (MI.getOpcode() == PPC::SETRNDi) {
DebugLoc dl = MI.getDebugLoc();
Register OldFPSCRReg = MI.getOperand(0).getReg();
// Save FPSCR value.
BuildMI(*BB, MI, dl, TII->get(PPC::MFFS), OldFPSCRReg);
// The floating point rounding mode is in the bits 62:63 of FPCSR, and has
// the following settings:
// 00 Round to nearest
// 01 Round to 0
// 10 Round to +inf
// 11 Round to -inf
// When the operand is immediate, using the two least significant bits of
// the immediate to set the bits 62:63 of FPSCR.
unsigned Mode = MI.getOperand(1).getImm();
BuildMI(*BB, MI, dl, TII->get((Mode & 1) ? PPC::MTFSB1 : PPC::MTFSB0))
.addImm(31)
.addReg(PPC::RM, RegState::ImplicitDefine);
BuildMI(*BB, MI, dl, TII->get((Mode & 2) ? PPC::MTFSB1 : PPC::MTFSB0))
.addImm(30)
.addReg(PPC::RM, RegState::ImplicitDefine);
} else if (MI.getOpcode() == PPC::SETRND) {
DebugLoc dl = MI.getDebugLoc();
// Copy register from F8RCRegClass::SrcReg to G8RCRegClass::DestReg
// or copy register from G8RCRegClass::SrcReg to F8RCRegClass::DestReg.
// If the target doesn't have DirectMove, we should use stack to do the
// conversion, because the target doesn't have the instructions like mtvsrd
// or mfvsrd to do this conversion directly.
auto copyRegFromG8RCOrF8RC = [&] (unsigned DestReg, unsigned SrcReg) {
if (Subtarget.hasDirectMove()) {
BuildMI(*BB, MI, dl, TII->get(TargetOpcode::COPY), DestReg)
.addReg(SrcReg);
} else {
// Use stack to do the register copy.
unsigned StoreOp = PPC::STD, LoadOp = PPC::LFD;
MachineRegisterInfo &RegInfo = F->getRegInfo();
const TargetRegisterClass *RC = RegInfo.getRegClass(SrcReg);
if (RC == &PPC::F8RCRegClass) {
// Copy register from F8RCRegClass to G8RCRegclass.
assert((RegInfo.getRegClass(DestReg) == &PPC::G8RCRegClass) &&
"Unsupported RegClass.");
StoreOp = PPC::STFD;
LoadOp = PPC::LD;
} else {
// Copy register from G8RCRegClass to F8RCRegclass.
assert((RegInfo.getRegClass(SrcReg) == &PPC::G8RCRegClass) &&
(RegInfo.getRegClass(DestReg) == &PPC::F8RCRegClass) &&
"Unsupported RegClass.");
}
MachineFrameInfo &MFI = F->getFrameInfo();
int FrameIdx = MFI.CreateStackObject(8, Align(8), false);
MachineMemOperand *MMOStore = F->getMachineMemOperand(
MachinePointerInfo::getFixedStack(*F, FrameIdx, 0),
MachineMemOperand::MOStore, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlign(FrameIdx));
// Store the SrcReg into the stack.
BuildMI(*BB, MI, dl, TII->get(StoreOp))
.addReg(SrcReg)
.addImm(0)
.addFrameIndex(FrameIdx)
.addMemOperand(MMOStore);
MachineMemOperand *MMOLoad = F->getMachineMemOperand(
MachinePointerInfo::getFixedStack(*F, FrameIdx, 0),
MachineMemOperand::MOLoad, MFI.getObjectSize(FrameIdx),
MFI.getObjectAlign(FrameIdx));
// Load from the stack where SrcReg is stored, and save to DestReg,
// so we have done the RegClass conversion from RegClass::SrcReg to
// RegClass::DestReg.
BuildMI(*BB, MI, dl, TII->get(LoadOp), DestReg)
.addImm(0)
.addFrameIndex(FrameIdx)
.addMemOperand(MMOLoad);
}
};
Register OldFPSCRReg = MI.getOperand(0).getReg();
// Save FPSCR value.
BuildMI(*BB, MI, dl, TII->get(PPC::MFFS), OldFPSCRReg);
// When the operand is gprc register, use two least significant bits of the
// register and mtfsf instruction to set the bits 62:63 of FPSCR.
//
// copy OldFPSCRTmpReg, OldFPSCRReg
// (INSERT_SUBREG ExtSrcReg, (IMPLICIT_DEF ImDefReg), SrcOp, 1)
// rldimi NewFPSCRTmpReg, ExtSrcReg, OldFPSCRReg, 0, 62
// copy NewFPSCRReg, NewFPSCRTmpReg
// mtfsf 255, NewFPSCRReg
MachineOperand SrcOp = MI.getOperand(1);
MachineRegisterInfo &RegInfo = F->getRegInfo();
Register OldFPSCRTmpReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
copyRegFromG8RCOrF8RC(OldFPSCRTmpReg, OldFPSCRReg);
Register ImDefReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
Register ExtSrcReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
// The first operand of INSERT_SUBREG should be a register which has
// subregisters, we only care about its RegClass, so we should use an
// IMPLICIT_DEF register.
BuildMI(*BB, MI, dl, TII->get(TargetOpcode::IMPLICIT_DEF), ImDefReg);
BuildMI(*BB, MI, dl, TII->get(PPC::INSERT_SUBREG), ExtSrcReg)
.addReg(ImDefReg)
.add(SrcOp)
.addImm(1);
Register NewFPSCRTmpReg = RegInfo.createVirtualRegister(&PPC::G8RCRegClass);
BuildMI(*BB, MI, dl, TII->get(PPC::RLDIMI), NewFPSCRTmpReg)
.addReg(OldFPSCRTmpReg)
.addReg(ExtSrcReg)
.addImm(0)
.addImm(62);
Register NewFPSCRReg = RegInfo.createVirtualRegister(&PPC::F8RCRegClass);
copyRegFromG8RCOrF8RC(NewFPSCRReg, NewFPSCRTmpReg);
// The mask 255 means that put the 32:63 bits of NewFPSCRReg to the 32:63
// bits of FPSCR.
BuildMI(*BB, MI, dl, TII->get(PPC::MTFSF))
.addImm(255)
.addReg(NewFPSCRReg)
.addImm(0)
.addImm(0);
} else if (MI.getOpcode() == PPC::SETFLM) {
DebugLoc Dl = MI.getDebugLoc();
// Result of setflm is previous FPSCR content, so we need to save it first.
Register OldFPSCRReg = MI.getOperand(0).getReg();
BuildMI(*BB, MI, Dl, TII->get(PPC::MFFS), OldFPSCRReg);
// Put bits in 32:63 to FPSCR.
Register NewFPSCRReg = MI.getOperand(1).getReg();
BuildMI(*BB, MI, Dl, TII->get(PPC::MTFSF))
.addImm(255)
.addReg(NewFPSCRReg)
.addImm(0)
.addImm(0);
} else if (MI.getOpcode() == PPC::PROBED_ALLOCA_32 ||
MI.getOpcode() == PPC::PROBED_ALLOCA_64) {
return emitProbedAlloca(MI, BB);
} else {
llvm_unreachable("Unexpected instr type to insert");
}
MI.eraseFromParent(); // The pseudo instruction is gone now.
return BB;
}
//===----------------------------------------------------------------------===//
// Target Optimization Hooks
//===----------------------------------------------------------------------===//
static int getEstimateRefinementSteps(EVT VT, const PPCSubtarget &Subtarget) {
// For the estimates, convergence is quadratic, so we essentially double the
// number of digits correct after every iteration. For both FRE and FRSQRTE,
// the minimum architected relative accuracy is 2^-5. When hasRecipPrec(),
// this is 2^-14. IEEE float has 23 digits and double has 52 digits.
int RefinementSteps = Subtarget.hasRecipPrec() ? 1 : 3;
if (VT.getScalarType() == MVT::f64)
RefinementSteps++;
return RefinementSteps;
}
SDValue PPCTargetLowering::getSqrtInputTest(SDValue Op, SelectionDAG &DAG,
const DenormalMode &Mode) const {
// We only have VSX Vector Test for software Square Root.
EVT VT = Op.getValueType();
if (!isTypeLegal(MVT::i1) ||
(VT != MVT::f64 &&
((VT != MVT::v2f64 && VT != MVT::v4f32) || !Subtarget.hasVSX())))
return TargetLowering::getSqrtInputTest(Op, DAG, Mode);
SDLoc DL(Op);
// The output register of FTSQRT is CR field.
SDValue FTSQRT = DAG.getNode(PPCISD::FTSQRT, DL, MVT::i32, Op);
// ftsqrt BF,FRB
// Let e_b be the unbiased exponent of the double-precision
// floating-point operand in register FRB.
// fe_flag is set to 1 if either of the following conditions occurs.
// - The double-precision floating-point operand in register FRB is a zero,
// a NaN, or an infinity, or a negative value.
// - e_b is less than or equal to -970.
// Otherwise fe_flag is set to 0.
// Both VSX and non-VSX versions would set EQ bit in the CR if the number is
// not eligible for iteration. (zero/negative/infinity/nan or unbiased
// exponent is less than -970)
SDValue SRIdxVal = DAG.getTargetConstant(PPC::sub_eq, DL, MVT::i32);
return SDValue(DAG.getMachineNode(TargetOpcode::EXTRACT_SUBREG, DL, MVT::i1,
FTSQRT, SRIdxVal),
0);
}
SDValue
PPCTargetLowering::getSqrtResultForDenormInput(SDValue Op,
SelectionDAG &DAG) const {
// We only have VSX Vector Square Root.
EVT VT = Op.getValueType();
if (VT != MVT::f64 &&
((VT != MVT::v2f64 && VT != MVT::v4f32) || !Subtarget.hasVSX()))
return TargetLowering::getSqrtResultForDenormInput(Op, DAG);
return DAG.getNode(PPCISD::FSQRT, SDLoc(Op), VT, Op);
}
SDValue PPCTargetLowering::getSqrtEstimate(SDValue Operand, SelectionDAG &DAG,
int Enabled, int &RefinementSteps,
bool &UseOneConstNR,
bool Reciprocal) const {
EVT VT = Operand.getValueType();
if ((VT == MVT::f32 && Subtarget.hasFRSQRTES()) ||
(VT == MVT::f64 && Subtarget.hasFRSQRTE()) ||
(VT == MVT::v4f32 && Subtarget.hasAltivec()) ||
(VT == MVT::v2f64 && Subtarget.hasVSX())) {
if (RefinementSteps == ReciprocalEstimate::Unspecified)
RefinementSteps = getEstimateRefinementSteps(VT, Subtarget);
// The Newton-Raphson computation with a single constant does not provide
// enough accuracy on some CPUs.
UseOneConstNR = !Subtarget.needsTwoConstNR();
return DAG.getNode(PPCISD::FRSQRTE, SDLoc(Operand), VT, Operand);
}
return SDValue();
}
SDValue PPCTargetLowering::getRecipEstimate(SDValue Operand, SelectionDAG &DAG,
int Enabled,
int &RefinementSteps) const {
EVT VT = Operand.getValueType();
if ((VT == MVT::f32 && Subtarget.hasFRES()) ||
(VT == MVT::f64 && Subtarget.hasFRE()) ||
(VT == MVT::v4f32 && Subtarget.hasAltivec()) ||
(VT == MVT::v2f64 && Subtarget.hasVSX())) {
if (RefinementSteps == ReciprocalEstimate::Unspecified)
RefinementSteps = getEstimateRefinementSteps(VT, Subtarget);
return DAG.getNode(PPCISD::FRE, SDLoc(Operand), VT, Operand);
}
return SDValue();
}
unsigned PPCTargetLowering::combineRepeatedFPDivisors() const {
// Note: This functionality is used only when unsafe-fp-math is enabled, and
// on cores with reciprocal estimates (which are used when unsafe-fp-math is
// enabled for division), this functionality is redundant with the default
// combiner logic (once the division -> reciprocal/multiply transformation
// has taken place). As a result, this matters more for older cores than for
// newer ones.
// Combine multiple FDIVs with the same divisor into multiple FMULs by the
// reciprocal if there are two or more FDIVs (for embedded cores with only
// one FP pipeline) for three or more FDIVs (for generic OOO cores).
switch (Subtarget.getCPUDirective()) {
default:
return 3;
case PPC::DIR_440:
case PPC::DIR_A2:
case PPC::DIR_E500:
case PPC::DIR_E500mc:
case PPC::DIR_E5500:
return 2;
}
}
// isConsecutiveLSLoc needs to work even if all adds have not yet been
// collapsed, and so we need to look through chains of them.
static void getBaseWithConstantOffset(SDValue Loc, SDValue &Base,
int64_t& Offset, SelectionDAG &DAG) {
if (DAG.isBaseWithConstantOffset(Loc)) {
Base = Loc.getOperand(0);
Offset += cast<ConstantSDNode>(Loc.getOperand(1))->getSExtValue();
// The base might itself be a base plus an offset, and if so, accumulate
// that as well.
getBaseWithConstantOffset(Loc.getOperand(0), Base, Offset, DAG);
}
}
static bool isConsecutiveLSLoc(SDValue Loc, EVT VT, LSBaseSDNode *Base,
unsigned Bytes, int Dist,
SelectionDAG &DAG) {
if (VT.getSizeInBits() / 8 != Bytes)
return false;
SDValue BaseLoc = Base->getBasePtr();
if (Loc.getOpcode() == ISD::FrameIndex) {
if (BaseLoc.getOpcode() != ISD::FrameIndex)
return false;
const MachineFrameInfo &MFI = DAG.getMachineFunction().getFrameInfo();
int FI = cast<FrameIndexSDNode>(Loc)->getIndex();
int BFI = cast<FrameIndexSDNode>(BaseLoc)->getIndex();
int FS = MFI.getObjectSize(FI);
int BFS = MFI.getObjectSize(BFI);
if (FS != BFS || FS != (int)Bytes) return false;
return MFI.getObjectOffset(FI) == (MFI.getObjectOffset(BFI) + Dist*Bytes);
}
SDValue Base1 = Loc, Base2 = BaseLoc;
int64_t Offset1 = 0, Offset2 = 0;
getBaseWithConstantOffset(Loc, Base1, Offset1, DAG);
getBaseWithConstantOffset(BaseLoc, Base2, Offset2, DAG);
if (Base1 == Base2 && Offset1 == (Offset2 + Dist * Bytes))
return true;
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
const GlobalValue *GV1 = nullptr;
const GlobalValue *GV2 = nullptr;
Offset1 = 0;
Offset2 = 0;
bool isGA1 = TLI.isGAPlusOffset(Loc.getNode(), GV1, Offset1);
bool isGA2 = TLI.isGAPlusOffset(BaseLoc.getNode(), GV2, Offset2);
if (isGA1 && isGA2 && GV1 == GV2)
return Offset1 == (Offset2 + Dist*Bytes);
return false;
}
// Like SelectionDAG::isConsecutiveLoad, but also works for stores, and does
// not enforce equality of the chain operands.
static bool isConsecutiveLS(SDNode *N, LSBaseSDNode *Base,
unsigned Bytes, int Dist,
SelectionDAG &DAG) {
if (LSBaseSDNode *LS = dyn_cast<LSBaseSDNode>(N)) {
EVT VT = LS->getMemoryVT();
SDValue Loc = LS->getBasePtr();
return isConsecutiveLSLoc(Loc, VT, Base, Bytes, Dist, DAG);
}
if (N->getOpcode() == ISD::INTRINSIC_W_CHAIN) {
EVT VT;
switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
default: return false;
case Intrinsic::ppc_altivec_lvx:
case Intrinsic::ppc_altivec_lvxl:
case Intrinsic::ppc_vsx_lxvw4x:
case Intrinsic::ppc_vsx_lxvw4x_be:
VT = MVT::v4i32;
break;
case Intrinsic::ppc_vsx_lxvd2x:
case Intrinsic::ppc_vsx_lxvd2x_be:
VT = MVT::v2f64;
break;
case Intrinsic::ppc_altivec_lvebx:
VT = MVT::i8;
break;
case Intrinsic::ppc_altivec_lvehx:
VT = MVT::i16;
break;
case Intrinsic::ppc_altivec_lvewx:
VT = MVT::i32;
break;
}
return isConsecutiveLSLoc(N->getOperand(2), VT, Base, Bytes, Dist, DAG);
}
if (N->getOpcode() == ISD::INTRINSIC_VOID) {
EVT VT;
switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
default: return false;
case Intrinsic::ppc_altivec_stvx:
case Intrinsic::ppc_altivec_stvxl:
case Intrinsic::ppc_vsx_stxvw4x:
VT = MVT::v4i32;
break;
case Intrinsic::ppc_vsx_stxvd2x:
VT = MVT::v2f64;
break;
case Intrinsic::ppc_vsx_stxvw4x_be:
VT = MVT::v4i32;
break;
case Intrinsic::ppc_vsx_stxvd2x_be:
VT = MVT::v2f64;
break;
case Intrinsic::ppc_altivec_stvebx:
VT = MVT::i8;
break;
case Intrinsic::ppc_altivec_stvehx:
VT = MVT::i16;
break;
case Intrinsic::ppc_altivec_stvewx:
VT = MVT::i32;
break;
}
return isConsecutiveLSLoc(N->getOperand(3), VT, Base, Bytes, Dist, DAG);
}
return false;
}
// Return true is there is a nearyby consecutive load to the one provided
// (regardless of alignment). We search up and down the chain, looking though
// token factors and other loads (but nothing else). As a result, a true result
// indicates that it is safe to create a new consecutive load adjacent to the
// load provided.
static bool findConsecutiveLoad(LoadSDNode *LD, SelectionDAG &DAG) {
SDValue Chain = LD->getChain();
EVT VT = LD->getMemoryVT();
SmallSet<SDNode *, 16> LoadRoots;
SmallVector<SDNode *, 8> Queue(1, Chain.getNode());
SmallSet<SDNode *, 16> Visited;
// First, search up the chain, branching to follow all token-factor operands.
// If we find a consecutive load, then we're done, otherwise, record all
// nodes just above the top-level loads and token factors.
while (!Queue.empty()) {
SDNode *ChainNext = Queue.pop_back_val();
if (!Visited.insert(ChainNext).second)
continue;
if (MemSDNode *ChainLD = dyn_cast<MemSDNode>(ChainNext)) {
if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
return true;
if (!Visited.count(ChainLD->getChain().getNode()))
Queue.push_back(ChainLD->getChain().getNode());
} else if (ChainNext->getOpcode() == ISD::TokenFactor) {
for (const SDUse &O : ChainNext->ops())
if (!Visited.count(O.getNode()))
Queue.push_back(O.getNode());
} else
LoadRoots.insert(ChainNext);
}
// Second, search down the chain, starting from the top-level nodes recorded
// in the first phase. These top-level nodes are the nodes just above all
// loads and token factors. Starting with their uses, recursively look though
// all loads (just the chain uses) and token factors to find a consecutive
// load.
Visited.clear();
Queue.clear();
for (SmallSet<SDNode *, 16>::iterator I = LoadRoots.begin(),
IE = LoadRoots.end(); I != IE; ++I) {
Queue.push_back(*I);
while (!Queue.empty()) {
SDNode *LoadRoot = Queue.pop_back_val();
if (!Visited.insert(LoadRoot).second)
continue;
if (MemSDNode *ChainLD = dyn_cast<MemSDNode>(LoadRoot))
if (isConsecutiveLS(ChainLD, LD, VT.getStoreSize(), 1, DAG))
return true;
for (SDNode::use_iterator UI = LoadRoot->use_begin(),
UE = LoadRoot->use_end(); UI != UE; ++UI)
if (((isa<MemSDNode>(*UI) &&
cast<MemSDNode>(*UI)->getChain().getNode() == LoadRoot) ||
UI->getOpcode() == ISD::TokenFactor) && !Visited.count(*UI))
Queue.push_back(*UI);
}
}
return false;
}
/// This function is called when we have proved that a SETCC node can be replaced
/// by subtraction (and other supporting instructions) so that the result of
/// comparison is kept in a GPR instead of CR. This function is purely for
/// codegen purposes and has some flags to guide the codegen process.
static SDValue generateEquivalentSub(SDNode *N, int Size, bool Complement,
bool Swap, SDLoc &DL, SelectionDAG &DAG) {
assert(N->getOpcode() == ISD::SETCC && "ISD::SETCC Expected.");
// Zero extend the operands to the largest legal integer. Originally, they
// must be of a strictly smaller size.
auto Op0 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N->getOperand(0),
DAG.getConstant(Size, DL, MVT::i32));
auto Op1 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, N->getOperand(1),
DAG.getConstant(Size, DL, MVT::i32));
// Swap if needed. Depends on the condition code.
if (Swap)
std::swap(Op0, Op1);
// Subtract extended integers.
auto SubNode = DAG.getNode(ISD::SUB, DL, MVT::i64, Op0, Op1);
// Move the sign bit to the least significant position and zero out the rest.
// Now the least significant bit carries the result of original comparison.
auto Shifted = DAG.getNode(ISD::SRL, DL, MVT::i64, SubNode,
DAG.getConstant(Size - 1, DL, MVT::i32));
auto Final = Shifted;
// Complement the result if needed. Based on the condition code.
if (Complement)
Final = DAG.getNode(ISD::XOR, DL, MVT::i64, Shifted,
DAG.getConstant(1, DL, MVT::i64));
return DAG.getNode(ISD::TRUNCATE, DL, MVT::i1, Final);
}
SDValue PPCTargetLowering::ConvertSETCCToSubtract(SDNode *N,
DAGCombinerInfo &DCI) const {
assert(N->getOpcode() == ISD::SETCC && "ISD::SETCC Expected.");
SelectionDAG &DAG = DCI.DAG;
SDLoc DL(N);
// Size of integers being compared has a critical role in the following
// analysis, so we prefer to do this when all types are legal.
if (!DCI.isAfterLegalizeDAG())
return SDValue();
// If all users of SETCC extend its value to a legal integer type
// then we replace SETCC with a subtraction
for (SDNode::use_iterator UI = N->use_begin(),
UE = N->use_end(); UI != UE; ++UI) {
if (UI->getOpcode() != ISD::ZERO_EXTEND)
return SDValue();
}
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get();
auto OpSize = N->getOperand(0).getValueSizeInBits();
unsigned Size = DAG.getDataLayout().getLargestLegalIntTypeSizeInBits();
if (OpSize < Size) {
switch (CC) {
default: break;
case ISD::SETULT:
return generateEquivalentSub(N, Size, false, false, DL, DAG);
case ISD::SETULE:
return generateEquivalentSub(N, Size, true, true, DL, DAG);
case ISD::SETUGT:
return generateEquivalentSub(N, Size, false, true, DL, DAG);
case ISD::SETUGE:
return generateEquivalentSub(N, Size, true, false, DL, DAG);
}
}
return SDValue();
}
SDValue PPCTargetLowering::DAGCombineTruncBoolExt(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
assert(Subtarget.useCRBits() && "Expecting to be tracking CR bits");
// If we're tracking CR bits, we need to be careful that we don't have:
// trunc(binary-ops(zext(x), zext(y)))
// or
// trunc(binary-ops(binary-ops(zext(x), zext(y)), ...)
// such that we're unnecessarily moving things into GPRs when it would be
// better to keep them in CR bits.
// Note that trunc here can be an actual i1 trunc, or can be the effective
// truncation that comes from a setcc or select_cc.
if (N->getOpcode() == ISD::TRUNCATE &&
N->getValueType(0) != MVT::i1)
return SDValue();
if (N->getOperand(0).getValueType() != MVT::i32 &&
N->getOperand(0).getValueType() != MVT::i64)
return SDValue();
if (N->getOpcode() == ISD::SETCC ||
N->getOpcode() == ISD::SELECT_CC) {
// If we're looking at a comparison, then we need to make sure that the
// high bits (all except for the first) don't matter the result.
ISD::CondCode CC =
cast<CondCodeSDNode>(N->getOperand(
N->getOpcode() == ISD::SETCC ? 2 : 4))->get();
unsigned OpBits = N->getOperand(0).getValueSizeInBits();
if (ISD::isSignedIntSetCC(CC)) {
if (DAG.ComputeNumSignBits(N->getOperand(0)) != OpBits ||
DAG.ComputeNumSignBits(N->getOperand(1)) != OpBits)
return SDValue();
} else if (ISD::isUnsignedIntSetCC(CC)) {
if (!DAG.MaskedValueIsZero(N->getOperand(0),
APInt::getHighBitsSet(OpBits, OpBits-1)) ||
!DAG.MaskedValueIsZero(N->getOperand(1),
APInt::getHighBitsSet(OpBits, OpBits-1)))
return (N->getOpcode() == ISD::SETCC ? ConvertSETCCToSubtract(N, DCI)
: SDValue());
} else {
// This is neither a signed nor an unsigned comparison, just make sure
// that the high bits are equal.
KnownBits Op1Known = DAG.computeKnownBits(N->getOperand(0));
KnownBits Op2Known = DAG.computeKnownBits(N->getOperand(1));
// We don't really care about what is known about the first bit (if
// anything), so pretend that it is known zero for both to ensure they can
// be compared as constants.
Op1Known.Zero.setBit(0); Op1Known.One.clearBit(0);
Op2Known.Zero.setBit(0); Op2Known.One.clearBit(0);
if (!Op1Known.isConstant() || !Op2Known.isConstant() ||
Op1Known.getConstant() != Op2Known.getConstant())
return SDValue();
}
}
// We now know that the higher-order bits are irrelevant, we just need to
// make sure that all of the intermediate operations are bit operations, and
// all inputs are extensions.
if (N->getOperand(0).getOpcode() != ISD::AND &&
N->getOperand(0).getOpcode() != ISD::OR &&
N->getOperand(0).getOpcode() != ISD::XOR &&
N->getOperand(0).getOpcode() != ISD::SELECT &&
N->getOperand(0).getOpcode() != ISD::SELECT_CC &&
N->getOperand(0).getOpcode() != ISD::TRUNCATE &&
N->getOperand(0).getOpcode() != ISD::SIGN_EXTEND &&
N->getOperand(0).getOpcode() != ISD::ZERO_EXTEND &&
N->getOperand(0).getOpcode() != ISD::ANY_EXTEND)
return SDValue();
if ((N->getOpcode() == ISD::SETCC || N->getOpcode() == ISD::SELECT_CC) &&
N->getOperand(1).getOpcode() != ISD::AND &&
N->getOperand(1).getOpcode() != ISD::OR &&
N->getOperand(1).getOpcode() != ISD::XOR &&
N->getOperand(1).getOpcode() != ISD::SELECT &&
N->getOperand(1).getOpcode() != ISD::SELECT_CC &&
N->getOperand(1).getOpcode() != ISD::TRUNCATE &&
N->getOperand(1).getOpcode() != ISD::SIGN_EXTEND &&
N->getOperand(1).getOpcode() != ISD::ZERO_EXTEND &&
N->getOperand(1).getOpcode() != ISD::ANY_EXTEND)
return SDValue();
SmallVector<SDValue, 4> Inputs;
SmallVector<SDValue, 8> BinOps, PromOps;
SmallPtrSet<SDNode *, 16> Visited;
for (unsigned i = 0; i < 2; ++i) {
if (((N->getOperand(i).getOpcode() == ISD::SIGN_EXTEND ||
N->getOperand(i).getOpcode() == ISD::ZERO_EXTEND ||
N->getOperand(i).getOpcode() == ISD::ANY_EXTEND) &&
N->getOperand(i).getOperand(0).getValueType() == MVT::i1) ||
isa<ConstantSDNode>(N->getOperand(i)))
Inputs.push_back(N->getOperand(i));
else
BinOps.push_back(N->getOperand(i));
if (N->getOpcode() == ISD::TRUNCATE)
break;
}
// Visit all inputs, collect all binary operations (and, or, xor and
// select) that are all fed by extensions.
while (!BinOps.empty()) {
SDValue BinOp = BinOps.pop_back_val();
if (!Visited.insert(BinOp.getNode()).second)
continue;
PromOps.push_back(BinOp);
for (unsigned i = 0, ie = BinOp.getNumOperands(); i != ie; ++i) {
// The condition of the select is not promoted.
if (BinOp.getOpcode() == ISD::SELECT && i == 0)
continue;
if (BinOp.getOpcode() == ISD::SELECT_CC && i != 2 && i != 3)
continue;
if (((BinOp.getOperand(i).getOpcode() == ISD::SIGN_EXTEND ||
BinOp.getOperand(i).getOpcode() == ISD::ZERO_EXTEND ||
BinOp.getOperand(i).getOpcode() == ISD::ANY_EXTEND) &&
BinOp.getOperand(i).getOperand(0).getValueType() == MVT::i1) ||
isa<ConstantSDNode>(BinOp.getOperand(i))) {
Inputs.push_back(BinOp.getOperand(i));
} else if (BinOp.getOperand(i).getOpcode() == ISD::AND ||
BinOp.getOperand(i).getOpcode() == ISD::OR ||
BinOp.getOperand(i).getOpcode() == ISD::XOR ||
BinOp.getOperand(i).getOpcode() == ISD::SELECT ||
BinOp.getOperand(i).getOpcode() == ISD::SELECT_CC ||
BinOp.getOperand(i).getOpcode() == ISD::TRUNCATE ||
BinOp.getOperand(i).getOpcode() == ISD::SIGN_EXTEND ||
BinOp.getOperand(i).getOpcode() == ISD::ZERO_EXTEND ||
BinOp.getOperand(i).getOpcode() == ISD::ANY_EXTEND) {
BinOps.push_back(BinOp.getOperand(i));
} else {
// We have an input that is not an extension or another binary
// operation; we'll abort this transformation.
return SDValue();
}
}
}
// Make sure that this is a self-contained cluster of operations (which
// is not quite the same thing as saying that everything has only one
// use).
for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
if (isa<ConstantSDNode>(Inputs[i]))
continue;
for (SDNode::use_iterator UI = Inputs[i].getNode()->use_begin(),
UE = Inputs[i].getNode()->use_end();
UI != UE; ++UI) {
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
// Make sure that we're not going to promote the non-output-value
// operand(s) or SELECT or SELECT_CC.
// FIXME: Although we could sometimes handle this, and it does occur in
// practice that one of the condition inputs to the select is also one of
// the outputs, we currently can't deal with this.
if (User->getOpcode() == ISD::SELECT) {
if (User->getOperand(0) == Inputs[i])
return SDValue();
} else if (User->getOpcode() == ISD::SELECT_CC) {
if (User->getOperand(0) == Inputs[i] ||
User->getOperand(1) == Inputs[i])
return SDValue();
}
}
}
for (unsigned i = 0, ie = PromOps.size(); i != ie; ++i) {
for (SDNode::use_iterator UI = PromOps[i].getNode()->use_begin(),
UE = PromOps[i].getNode()->use_end();
UI != UE; ++UI) {
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
// Make sure that we're not going to promote the non-output-value
// operand(s) or SELECT or SELECT_CC.
// FIXME: Although we could sometimes handle this, and it does occur in
// practice that one of the condition inputs to the select is also one of
// the outputs, we currently can't deal with this.
if (User->getOpcode() == ISD::SELECT) {
if (User->getOperand(0) == PromOps[i])
return SDValue();
} else if (User->getOpcode() == ISD::SELECT_CC) {
if (User->getOperand(0) == PromOps[i] ||
User->getOperand(1) == PromOps[i])
return SDValue();
}
}
}
// Replace all inputs with the extension operand.
for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
// Constants may have users outside the cluster of to-be-promoted nodes,
// and so we need to replace those as we do the promotions.
if (isa<ConstantSDNode>(Inputs[i]))
continue;
else
DAG.ReplaceAllUsesOfValueWith(Inputs[i], Inputs[i].getOperand(0));
}
std::list<HandleSDNode> PromOpHandles;
for (auto &PromOp : PromOps)
PromOpHandles.emplace_back(PromOp);
// Replace all operations (these are all the same, but have a different
// (i1) return type). DAG.getNode will validate that the types of
// a binary operator match, so go through the list in reverse so that
// we've likely promoted both operands first. Any intermediate truncations or
// extensions disappear.
while (!PromOpHandles.empty()) {
SDValue PromOp = PromOpHandles.back().getValue();
PromOpHandles.pop_back();
if (PromOp.getOpcode() == ISD::TRUNCATE ||
PromOp.getOpcode() == ISD::SIGN_EXTEND ||
PromOp.getOpcode() == ISD::ZERO_EXTEND ||
PromOp.getOpcode() == ISD::ANY_EXTEND) {
if (!isa<ConstantSDNode>(PromOp.getOperand(0)) &&
PromOp.getOperand(0).getValueType() != MVT::i1) {
// The operand is not yet ready (see comment below).
PromOpHandles.emplace_front(PromOp);
continue;
}
SDValue RepValue = PromOp.getOperand(0);
if (isa<ConstantSDNode>(RepValue))
RepValue = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, RepValue);
DAG.ReplaceAllUsesOfValueWith(PromOp, RepValue);
continue;
}
unsigned C;
switch (PromOp.getOpcode()) {
default: C = 0; break;
case ISD::SELECT: C = 1; break;
case ISD::SELECT_CC: C = 2; break;
}
if ((!isa<ConstantSDNode>(PromOp.getOperand(C)) &&
PromOp.getOperand(C).getValueType() != MVT::i1) ||
(!isa<ConstantSDNode>(PromOp.getOperand(C+1)) &&
PromOp.getOperand(C+1).getValueType() != MVT::i1)) {
// The to-be-promoted operands of this node have not yet been
// promoted (this should be rare because we're going through the
// list backward, but if one of the operands has several users in
// this cluster of to-be-promoted nodes, it is possible).
PromOpHandles.emplace_front(PromOp);
continue;
}
SmallVector<SDValue, 3> Ops(PromOp.getNode()->op_begin(),
PromOp.getNode()->op_end());
// If there are any constant inputs, make sure they're replaced now.
for (unsigned i = 0; i < 2; ++i)
if (isa<ConstantSDNode>(Ops[C+i]))
Ops[C+i] = DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, Ops[C+i]);
DAG.ReplaceAllUsesOfValueWith(PromOp,
DAG.getNode(PromOp.getOpcode(), dl, MVT::i1, Ops));
}
// Now we're left with the initial truncation itself.
if (N->getOpcode() == ISD::TRUNCATE)
return N->getOperand(0);
// Otherwise, this is a comparison. The operands to be compared have just
// changed type (to i1), but everything else is the same.
return SDValue(N, 0);
}
SDValue PPCTargetLowering::DAGCombineExtBoolTrunc(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
// If we're tracking CR bits, we need to be careful that we don't have:
// zext(binary-ops(trunc(x), trunc(y)))
// or
// zext(binary-ops(binary-ops(trunc(x), trunc(y)), ...)
// such that we're unnecessarily moving things into CR bits that can more
// efficiently stay in GPRs. Note that if we're not certain that the high
// bits are set as required by the final extension, we still may need to do
// some masking to get the proper behavior.
// This same functionality is important on PPC64 when dealing with
// 32-to-64-bit extensions; these occur often when 32-bit values are used as
// the return values of functions. Because it is so similar, it is handled
// here as well.
if (N->getValueType(0) != MVT::i32 &&
N->getValueType(0) != MVT::i64)
return SDValue();
if (!((N->getOperand(0).getValueType() == MVT::i1 && Subtarget.useCRBits()) ||
(N->getOperand(0).getValueType() == MVT::i32 && Subtarget.isPPC64())))
return SDValue();
if (N->getOperand(0).getOpcode() != ISD::AND &&
N->getOperand(0).getOpcode() != ISD::OR &&
N->getOperand(0).getOpcode() != ISD::XOR &&
N->getOperand(0).getOpcode() != ISD::SELECT &&
N->getOperand(0).getOpcode() != ISD::SELECT_CC)
return SDValue();
SmallVector<SDValue, 4> Inputs;
SmallVector<SDValue, 8> BinOps(1, N->getOperand(0)), PromOps;
SmallPtrSet<SDNode *, 16> Visited;
// Visit all inputs, collect all binary operations (and, or, xor and
// select) that are all fed by truncations.
while (!BinOps.empty()) {
SDValue BinOp = BinOps.pop_back_val();
if (!Visited.insert(BinOp.getNode()).second)
continue;
PromOps.push_back(BinOp);
for (unsigned i = 0, ie = BinOp.getNumOperands(); i != ie; ++i) {
// The condition of the select is not promoted.
if (BinOp.getOpcode() == ISD::SELECT && i == 0)
continue;
if (BinOp.getOpcode() == ISD::SELECT_CC && i != 2 && i != 3)
continue;
if (BinOp.getOperand(i).getOpcode() == ISD::TRUNCATE ||
isa<ConstantSDNode>(BinOp.getOperand(i))) {
Inputs.push_back(BinOp.getOperand(i));
} else if (BinOp.getOperand(i).getOpcode() == ISD::AND ||
BinOp.getOperand(i).getOpcode() == ISD::OR ||
BinOp.getOperand(i).getOpcode() == ISD::XOR ||
BinOp.getOperand(i).getOpcode() == ISD::SELECT ||
BinOp.getOperand(i).getOpcode() == ISD::SELECT_CC) {
BinOps.push_back(BinOp.getOperand(i));
} else {
// We have an input that is not a truncation or another binary
// operation; we'll abort this transformation.
return SDValue();
}
}
}
// The operands of a select that must be truncated when the select is
// promoted because the operand is actually part of the to-be-promoted set.
DenseMap<SDNode *, EVT> SelectTruncOp[2];
// Make sure that this is a self-contained cluster of operations (which
// is not quite the same thing as saying that everything has only one
// use).
for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
if (isa<ConstantSDNode>(Inputs[i]))
continue;
for (SDNode::use_iterator UI = Inputs[i].getNode()->use_begin(),
UE = Inputs[i].getNode()->use_end();
UI != UE; ++UI) {
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
// If we're going to promote the non-output-value operand(s) or SELECT or
// SELECT_CC, record them for truncation.
if (User->getOpcode() == ISD::SELECT) {
if (User->getOperand(0) == Inputs[i])
SelectTruncOp[0].insert(std::make_pair(User,
User->getOperand(0).getValueType()));
} else if (User->getOpcode() == ISD::SELECT_CC) {
if (User->getOperand(0) == Inputs[i])
SelectTruncOp[0].insert(std::make_pair(User,
User->getOperand(0).getValueType()));
if (User->getOperand(1) == Inputs[i])
SelectTruncOp[1].insert(std::make_pair(User,
User->getOperand(1).getValueType()));
}
}
}
for (unsigned i = 0, ie = PromOps.size(); i != ie; ++i) {
for (SDNode::use_iterator UI = PromOps[i].getNode()->use_begin(),
UE = PromOps[i].getNode()->use_end();
UI != UE; ++UI) {
SDNode *User = *UI;
if (User != N && !Visited.count(User))
return SDValue();
// If we're going to promote the non-output-value operand(s) or SELECT or
// SELECT_CC, record them for truncation.
if (User->getOpcode() == ISD::SELECT) {
if (User->getOperand(0) == PromOps[i])
SelectTruncOp[0].insert(std::make_pair(User,
User->getOperand(0).getValueType()));
} else if (User->getOpcode() == ISD::SELECT_CC) {
if (User->getOperand(0) == PromOps[i])
SelectTruncOp[0].insert(std::make_pair(User,
User->getOperand(0).getValueType()));
if (User->getOperand(1) == PromOps[i])
SelectTruncOp[1].insert(std::make_pair(User,
User->getOperand(1).getValueType()));
}
}
}
unsigned PromBits = N->getOperand(0).getValueSizeInBits();
bool ReallyNeedsExt = false;
if (N->getOpcode() != ISD::ANY_EXTEND) {
// If all of the inputs are not already sign/zero extended, then
// we'll still need to do that at the end.
for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
if (isa<ConstantSDNode>(Inputs[i]))
continue;
unsigned OpBits =
Inputs[i].getOperand(0).getValueSizeInBits();
assert(PromBits < OpBits && "Truncation not to a smaller bit count?");
if ((N->getOpcode() == ISD::ZERO_EXTEND &&
!DAG.MaskedValueIsZero(Inputs[i].getOperand(0),
APInt::getHighBitsSet(OpBits,
OpBits-PromBits))) ||
(N->getOpcode() == ISD::SIGN_EXTEND &&
DAG.ComputeNumSignBits(Inputs[i].getOperand(0)) <
(OpBits-(PromBits-1)))) {
ReallyNeedsExt = true;
break;
}
}
}
// Replace all inputs, either with the truncation operand, or a
// truncation or extension to the final output type.
for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
// Constant inputs need to be replaced with the to-be-promoted nodes that
// use them because they might have users outside of the cluster of
// promoted nodes.
if (isa<ConstantSDNode>(Inputs[i]))
continue;
SDValue InSrc = Inputs[i].getOperand(0);
if (Inputs[i].getValueType() == N->getValueType(0))
DAG.ReplaceAllUsesOfValueWith(Inputs[i], InSrc);
else if (N->getOpcode() == ISD::SIGN_EXTEND)
DAG.ReplaceAllUsesOfValueWith(Inputs[i],
DAG.getSExtOrTrunc(InSrc, dl, N->getValueType(0)));
else if (N->getOpcode() == ISD::ZERO_EXTEND)
DAG.ReplaceAllUsesOfValueWith(Inputs[i],
DAG.getZExtOrTrunc(InSrc, dl, N->getValueType(0)));
else
DAG.ReplaceAllUsesOfValueWith(Inputs[i],
DAG.getAnyExtOrTrunc(InSrc, dl, N->getValueType(0)));
}
std::list<HandleSDNode> PromOpHandles;
for (auto &PromOp : PromOps)
PromOpHandles.emplace_back(PromOp);
// Replace all operations (these are all the same, but have a different
// (promoted) return type). DAG.getNode will validate that the types of
// a binary operator match, so go through the list in reverse so that
// we've likely promoted both operands first.
while (!PromOpHandles.empty()) {
SDValue PromOp = PromOpHandles.back().getValue();
PromOpHandles.pop_back();
unsigned C;
switch (PromOp.getOpcode()) {
default: C = 0; break;
case ISD::SELECT: C = 1; break;
case ISD::SELECT_CC: C = 2; break;
}
if ((!isa<ConstantSDNode>(PromOp.getOperand(C)) &&
PromOp.getOperand(C).getValueType() != N->getValueType(0)) ||
(!isa<ConstantSDNode>(PromOp.getOperand(C+1)) &&
PromOp.getOperand(C+1).getValueType() != N->getValueType(0))) {
// The to-be-promoted operands of this node have not yet been
// promoted (this should be rare because we're going through the
// list backward, but if one of the operands has several users in
// this cluster of to-be-promoted nodes, it is possible).
PromOpHandles.emplace_front(PromOp);
continue;
}
// For SELECT and SELECT_CC nodes, we do a similar check for any
// to-be-promoted comparison inputs.
if (PromOp.getOpcode() == ISD::SELECT ||
PromOp.getOpcode() == ISD::SELECT_CC) {
if ((SelectTruncOp[0].count(PromOp.getNode()) &&
PromOp.getOperand(0).getValueType() != N->getValueType(0)) ||
(SelectTruncOp[1].count(PromOp.getNode()) &&
PromOp.getOperand(1).getValueType() != N->getValueType(0))) {
PromOpHandles.emplace_front(PromOp);
continue;
}
}
SmallVector<SDValue, 3> Ops(PromOp.getNode()->op_begin(),
PromOp.getNode()->op_end());
// If this node has constant inputs, then they'll need to be promoted here.
for (unsigned i = 0; i < 2; ++i) {
if (!isa<ConstantSDNode>(Ops[C+i]))
continue;
if (Ops[C+i].getValueType() == N->getValueType(0))
continue;
if (N->getOpcode() == ISD::SIGN_EXTEND)
Ops[C+i] = DAG.getSExtOrTrunc(Ops[C+i], dl, N->getValueType(0));
else if (N->getOpcode() == ISD::ZERO_EXTEND)
Ops[C+i] = DAG.getZExtOrTrunc(Ops[C+i], dl, N->getValueType(0));
else
Ops[C+i] = DAG.getAnyExtOrTrunc(Ops[C+i], dl, N->getValueType(0));
}
// If we've promoted the comparison inputs of a SELECT or SELECT_CC,
// truncate them again to the original value type.
if (PromOp.getOpcode() == ISD::SELECT ||
PromOp.getOpcode() == ISD::SELECT_CC) {
auto SI0 = SelectTruncOp[0].find(PromOp.getNode());
if (SI0 != SelectTruncOp[0].end())
Ops[0] = DAG.getNode(ISD::TRUNCATE, dl, SI0->second, Ops[0]);
auto SI1 = SelectTruncOp[1].find(PromOp.getNode());
if (SI1 != SelectTruncOp[1].end())
Ops[1] = DAG.getNode(ISD::TRUNCATE, dl, SI1->second, Ops[1]);
}
DAG.ReplaceAllUsesOfValueWith(PromOp,
DAG.getNode(PromOp.getOpcode(), dl, N->getValueType(0), Ops));
}
// Now we're left with the initial extension itself.
if (!ReallyNeedsExt)
return N->getOperand(0);
// To zero extend, just mask off everything except for the first bit (in the
// i1 case).
if (N->getOpcode() == ISD::ZERO_EXTEND)
return DAG.getNode(ISD::AND, dl, N->getValueType(0), N->getOperand(0),
DAG.getConstant(APInt::getLowBitsSet(
N->getValueSizeInBits(0), PromBits),
dl, N->getValueType(0)));
assert(N->getOpcode() == ISD::SIGN_EXTEND &&
"Invalid extension type");
EVT ShiftAmountTy = getShiftAmountTy(N->getValueType(0), DAG.getDataLayout());
SDValue ShiftCst =
DAG.getConstant(N->getValueSizeInBits(0) - PromBits, dl, ShiftAmountTy);
return DAG.getNode(
ISD::SRA, dl, N->getValueType(0),
DAG.getNode(ISD::SHL, dl, N->getValueType(0), N->getOperand(0), ShiftCst),
ShiftCst);
}
SDValue PPCTargetLowering::combineSetCC(SDNode *N,
DAGCombinerInfo &DCI) const {
assert(N->getOpcode() == ISD::SETCC &&
"Should be called with a SETCC node");
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get();
if (CC == ISD::SETNE || CC == ISD::SETEQ) {
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
// If there is a '0 - y' pattern, canonicalize the pattern to the RHS.
if (LHS.getOpcode() == ISD::SUB && isNullConstant(LHS.getOperand(0)) &&
LHS.hasOneUse())
std::swap(LHS, RHS);
// x == 0-y --> x+y == 0
// x != 0-y --> x+y != 0
if (RHS.getOpcode() == ISD::SUB && isNullConstant(RHS.getOperand(0)) &&
RHS.hasOneUse()) {
SDLoc DL(N);
SelectionDAG &DAG = DCI.DAG;
EVT VT = N->getValueType(0);
EVT OpVT = LHS.getValueType();
SDValue Add = DAG.getNode(ISD::ADD, DL, OpVT, LHS, RHS.getOperand(1));
return DAG.getSetCC(DL, VT, Add, DAG.getConstant(0, DL, OpVT), CC);
}
}
return DAGCombineTruncBoolExt(N, DCI);
}
// Is this an extending load from an f32 to an f64?
static bool isFPExtLoad(SDValue Op) {
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Op.getNode()))
return LD->getExtensionType() == ISD::EXTLOAD &&
Op.getValueType() == MVT::f64;
return false;
}
/// Reduces the number of fp-to-int conversion when building a vector.
///
/// If this vector is built out of floating to integer conversions,
/// transform it to a vector built out of floating point values followed by a
/// single floating to integer conversion of the vector.
/// Namely (build_vector (fptosi $A), (fptosi $B), ...)
/// becomes (fptosi (build_vector ($A, $B, ...)))
SDValue PPCTargetLowering::
combineElementTruncationToVectorTruncation(SDNode *N,
DAGCombinerInfo &DCI) const {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
"Should be called with a BUILD_VECTOR node");
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue FirstInput = N->getOperand(0);
assert(FirstInput.getOpcode() == PPCISD::MFVSR &&
"The input operand must be an fp-to-int conversion.");
// This combine happens after legalization so the fp_to_[su]i nodes are
// already converted to PPCSISD nodes.
unsigned FirstConversion = FirstInput.getOperand(0).getOpcode();
if (FirstConversion == PPCISD::FCTIDZ ||
FirstConversion == PPCISD::FCTIDUZ ||
FirstConversion == PPCISD::FCTIWZ ||
FirstConversion == PPCISD::FCTIWUZ) {
bool IsSplat = true;
bool Is32Bit = FirstConversion == PPCISD::FCTIWZ ||
FirstConversion == PPCISD::FCTIWUZ;
EVT SrcVT = FirstInput.getOperand(0).getValueType();
SmallVector<SDValue, 4> Ops;
EVT TargetVT = N->getValueType(0);
for (int i = 0, e = N->getNumOperands(); i < e; ++i) {
SDValue NextOp = N->getOperand(i);
if (NextOp.getOpcode() != PPCISD::MFVSR)
return SDValue();
unsigned NextConversion = NextOp.getOperand(0).getOpcode();
if (NextConversion != FirstConversion)
return SDValue();
// If we are converting to 32-bit integers, we need to add an FP_ROUND.
// This is not valid if the input was originally double precision. It is
// also not profitable to do unless this is an extending load in which
// case doing this combine will allow us to combine consecutive loads.
if (Is32Bit && !isFPExtLoad(NextOp.getOperand(0).getOperand(0)))
return SDValue();
if (N->getOperand(i) != FirstInput)
IsSplat = false;
}
// If this is a splat, we leave it as-is since there will be only a single
// fp-to-int conversion followed by a splat of the integer. This is better
// for 32-bit and smaller ints and neutral for 64-bit ints.
if (IsSplat)
return SDValue();
// Now that we know we have the right type of node, get its operands
for (int i = 0, e = N->getNumOperands(); i < e; ++i) {
SDValue In = N->getOperand(i).getOperand(0);
if (Is32Bit) {
// For 32-bit values, we need to add an FP_ROUND node (if we made it
// here, we know that all inputs are extending loads so this is safe).
if (In.isUndef())
Ops.push_back(DAG.getUNDEF(SrcVT));
else {
SDValue Trunc = DAG.getNode(ISD::FP_ROUND, dl,
MVT::f32, In.getOperand(0),
DAG.getIntPtrConstant(1, dl));
Ops.push_back(Trunc);
}
} else
Ops.push_back(In.isUndef() ? DAG.getUNDEF(SrcVT) : In.getOperand(0));
}
unsigned Opcode;
if (FirstConversion == PPCISD::FCTIDZ ||
FirstConversion == PPCISD::FCTIWZ)
Opcode = ISD::FP_TO_SINT;
else
Opcode = ISD::FP_TO_UINT;
EVT NewVT = TargetVT == MVT::v2i64 ? MVT::v2f64 : MVT::v4f32;
SDValue BV = DAG.getBuildVector(NewVT, dl, Ops);
return DAG.getNode(Opcode, dl, TargetVT, BV);
}
return SDValue();
}
/// Reduce the number of loads when building a vector.
///
/// Building a vector out of multiple loads can be converted to a load
/// of the vector type if the loads are consecutive. If the loads are
/// consecutive but in descending order, a shuffle is added at the end
/// to reorder the vector.
static SDValue combineBVOfConsecutiveLoads(SDNode *N, SelectionDAG &DAG) {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
"Should be called with a BUILD_VECTOR node");
SDLoc dl(N);
// Return early for non byte-sized type, as they can't be consecutive.
if (!N->getValueType(0).getVectorElementType().isByteSized())
return SDValue();
bool InputsAreConsecutiveLoads = true;
bool InputsAreReverseConsecutive = true;
unsigned ElemSize = N->getValueType(0).getScalarType().getStoreSize();
SDValue FirstInput = N->getOperand(0);
bool IsRoundOfExtLoad = false;
if (FirstInput.getOpcode() == ISD::FP_ROUND &&
FirstInput.getOperand(0).getOpcode() == ISD::LOAD) {
LoadSDNode *LD = dyn_cast<LoadSDNode>(FirstInput.getOperand(0));
IsRoundOfExtLoad = LD->getExtensionType() == ISD::EXTLOAD;
}
// Not a build vector of (possibly fp_rounded) loads.
if ((!IsRoundOfExtLoad && FirstInput.getOpcode() != ISD::LOAD) ||
N->getNumOperands() == 1)
return SDValue();
for (int i = 1, e = N->getNumOperands(); i < e; ++i) {
// If any inputs are fp_round(extload), they all must be.
if (IsRoundOfExtLoad && N->getOperand(i).getOpcode() != ISD::FP_ROUND)
return SDValue();
SDValue NextInput = IsRoundOfExtLoad ? N->getOperand(i).getOperand(0) :
N->getOperand(i);
if (NextInput.getOpcode() != ISD::LOAD)
return SDValue();
SDValue PreviousInput =
IsRoundOfExtLoad ? N->getOperand(i-1).getOperand(0) : N->getOperand(i-1);
LoadSDNode *LD1 = dyn_cast<LoadSDNode>(PreviousInput);
LoadSDNode *LD2 = dyn_cast<LoadSDNode>(NextInput);
// If any inputs are fp_round(extload), they all must be.
if (IsRoundOfExtLoad && LD2->getExtensionType() != ISD::EXTLOAD)
return SDValue();
if (!isConsecutiveLS(LD2, LD1, ElemSize, 1, DAG))
InputsAreConsecutiveLoads = false;
if (!isConsecutiveLS(LD1, LD2, ElemSize, 1, DAG))
InputsAreReverseConsecutive = false;
// Exit early if the loads are neither consecutive nor reverse consecutive.
if (!InputsAreConsecutiveLoads && !InputsAreReverseConsecutive)
return SDValue();
}
assert(!(InputsAreConsecutiveLoads && InputsAreReverseConsecutive) &&
"The loads cannot be both consecutive and reverse consecutive.");
SDValue FirstLoadOp =
IsRoundOfExtLoad ? FirstInput.getOperand(0) : FirstInput;
SDValue LastLoadOp =
IsRoundOfExtLoad ? N->getOperand(N->getNumOperands()-1).getOperand(0) :
N->getOperand(N->getNumOperands()-1);
LoadSDNode *LD1 = dyn_cast<LoadSDNode>(FirstLoadOp);
LoadSDNode *LDL = dyn_cast<LoadSDNode>(LastLoadOp);
if (InputsAreConsecutiveLoads) {
assert(LD1 && "Input needs to be a LoadSDNode.");
return DAG.getLoad(N->getValueType(0), dl, LD1->getChain(),
LD1->getBasePtr(), LD1->getPointerInfo(),
LD1->getAlignment());
}
if (InputsAreReverseConsecutive) {
assert(LDL && "Input needs to be a LoadSDNode.");
SDValue Load = DAG.getLoad(N->getValueType(0), dl, LDL->getChain(),
LDL->getBasePtr(), LDL->getPointerInfo(),
LDL->getAlignment());
SmallVector<int, 16> Ops;
for (int i = N->getNumOperands() - 1; i >= 0; i--)
Ops.push_back(i);
return DAG.getVectorShuffle(N->getValueType(0), dl, Load,
DAG.getUNDEF(N->getValueType(0)), Ops);
}
return SDValue();
}
// This function adds the required vector_shuffle needed to get
// the elements of the vector extract in the correct position
// as specified by the CorrectElems encoding.
static SDValue addShuffleForVecExtend(SDNode *N, SelectionDAG &DAG,
SDValue Input, uint64_t Elems,
uint64_t CorrectElems) {
SDLoc dl(N);
unsigned NumElems = Input.getValueType().getVectorNumElements();
SmallVector<int, 16> ShuffleMask(NumElems, -1);
// Knowing the element indices being extracted from the original
// vector and the order in which they're being inserted, just put
// them at element indices required for the instruction.
for (unsigned i = 0; i < N->getNumOperands(); i++) {
if (DAG.getDataLayout().isLittleEndian())
ShuffleMask[CorrectElems & 0xF] = Elems & 0xF;
else
ShuffleMask[(CorrectElems & 0xF0) >> 4] = (Elems & 0xF0) >> 4;
CorrectElems = CorrectElems >> 8;
Elems = Elems >> 8;
}
SDValue Shuffle =
DAG.getVectorShuffle(Input.getValueType(), dl, Input,
DAG.getUNDEF(Input.getValueType()), ShuffleMask);
EVT VT = N->getValueType(0);
SDValue Conv = DAG.getBitcast(VT, Shuffle);
EVT ExtVT = EVT::getVectorVT(*DAG.getContext(),
Input.getValueType().getVectorElementType(),
VT.getVectorNumElements());
return DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, VT, Conv,
DAG.getValueType(ExtVT));
}
// Look for build vector patterns where input operands come from sign
// extended vector_extract elements of specific indices. If the correct indices
// aren't used, add a vector shuffle to fix up the indices and create
// SIGN_EXTEND_INREG node which selects the vector sign extend instructions
// during instruction selection.
static SDValue combineBVOfVecSExt(SDNode *N, SelectionDAG &DAG) {
// This array encodes the indices that the vector sign extend instructions
// extract from when extending from one type to another for both BE and LE.
// The right nibble of each byte corresponds to the LE incides.
// and the left nibble of each byte corresponds to the BE incides.
// For example: 0x3074B8FC byte->word
// For LE: the allowed indices are: 0x0,0x4,0x8,0xC
// For BE: the allowed indices are: 0x3,0x7,0xB,0xF
// For example: 0x000070F8 byte->double word
// For LE: the allowed indices are: 0x0,0x8
// For BE: the allowed indices are: 0x7,0xF
uint64_t TargetElems[] = {
0x3074B8FC, // b->w
0x000070F8, // b->d
0x10325476, // h->w
0x00003074, // h->d
0x00001032, // w->d
};
uint64_t Elems = 0;
int Index;
SDValue Input;
auto isSExtOfVecExtract = [&](SDValue Op) -> bool {
if (!Op)
return false;
if (Op.getOpcode() != ISD::SIGN_EXTEND &&
Op.getOpcode() != ISD::SIGN_EXTEND_INREG)
return false;
// A SIGN_EXTEND_INREG might be fed by an ANY_EXTEND to produce a value
// of the right width.
SDValue Extract = Op.getOperand(0);
if (Extract.getOpcode() == ISD::ANY_EXTEND)
Extract = Extract.getOperand(0);
if (Extract.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return false;
ConstantSDNode *ExtOp = dyn_cast<ConstantSDNode>(Extract.getOperand(1));
if (!ExtOp)
return false;
Index = ExtOp->getZExtValue();
if (Input && Input != Extract.getOperand(0))
return false;
if (!Input)
Input = Extract.getOperand(0);
Elems = Elems << 8;
Index = DAG.getDataLayout().isLittleEndian() ? Index : Index << 4;
Elems |= Index;
return true;
};
// If the build vector operands aren't sign extended vector extracts,
// of the same input vector, then return.
for (unsigned i = 0; i < N->getNumOperands(); i++) {
if (!isSExtOfVecExtract(N->getOperand(i))) {
return SDValue();
}
}
// If the vector extract indicies are not correct, add the appropriate
// vector_shuffle.
int TgtElemArrayIdx;
int InputSize = Input.getValueType().getScalarSizeInBits();
int OutputSize = N->getValueType(0).getScalarSizeInBits();
if (InputSize + OutputSize == 40)
TgtElemArrayIdx = 0;
else if (InputSize + OutputSize == 72)
TgtElemArrayIdx = 1;
else if (InputSize + OutputSize == 48)
TgtElemArrayIdx = 2;
else if (InputSize + OutputSize == 80)
TgtElemArrayIdx = 3;
else if (InputSize + OutputSize == 96)
TgtElemArrayIdx = 4;
else
return SDValue();
uint64_t CorrectElems = TargetElems[TgtElemArrayIdx];
CorrectElems = DAG.getDataLayout().isLittleEndian()
? CorrectElems & 0x0F0F0F0F0F0F0F0F
: CorrectElems & 0xF0F0F0F0F0F0F0F0;
if (Elems != CorrectElems) {
return addShuffleForVecExtend(N, DAG, Input, Elems, CorrectElems);
}
// Regular lowering will catch cases where a shuffle is not needed.
return SDValue();
}
// Look for the pattern of a load from a narrow width to i128, feeding
// into a BUILD_VECTOR of v1i128. Replace this sequence with a PPCISD node
// (LXVRZX). This node represents a zero extending load that will be matched
// to the Load VSX Vector Rightmost instructions.
static SDValue combineBVZEXTLOAD(SDNode *N, SelectionDAG &DAG) {
SDLoc DL(N);
// This combine is only eligible for a BUILD_VECTOR of v1i128.
if (N->getValueType(0) != MVT::v1i128)
return SDValue();
SDValue Operand = N->getOperand(0);
// Proceed with the transformation if the operand to the BUILD_VECTOR
// is a load instruction.
if (Operand.getOpcode() != ISD::LOAD)
return SDValue();
LoadSDNode *LD = dyn_cast<LoadSDNode>(Operand);
EVT MemoryType = LD->getMemoryVT();
// This transformation is only valid if the we are loading either a byte,
// halfword, word, or doubleword.
bool ValidLDType = MemoryType == MVT::i8 || MemoryType == MVT::i16 ||
MemoryType == MVT::i32 || MemoryType == MVT::i64;
// Ensure that the load from the narrow width is being zero extended to i128.
if (!ValidLDType ||
(LD->getExtensionType() != ISD::ZEXTLOAD &&
LD->getExtensionType() != ISD::EXTLOAD))
return SDValue();
SDValue LoadOps[] = {
LD->getChain(), LD->getBasePtr(),
DAG.getIntPtrConstant(MemoryType.getScalarSizeInBits(), DL)};
return DAG.getMemIntrinsicNode(PPCISD::LXVRZX, DL,
DAG.getVTList(MVT::v1i128, MVT::Other),
LoadOps, MemoryType, LD->getMemOperand());
}
SDValue PPCTargetLowering::DAGCombineBuildVector(SDNode *N,
DAGCombinerInfo &DCI) const {
assert(N->getOpcode() == ISD::BUILD_VECTOR &&
"Should be called with a BUILD_VECTOR node");
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
if (!Subtarget.hasVSX())
return SDValue();
// The target independent DAG combiner will leave a build_vector of
// float-to-int conversions intact. We can generate MUCH better code for
// a float-to-int conversion of a vector of floats.
SDValue FirstInput = N->getOperand(0);
if (FirstInput.getOpcode() == PPCISD::MFVSR) {
SDValue Reduced = combineElementTruncationToVectorTruncation(N, DCI);
if (Reduced)
return Reduced;
}
// If we're building a vector out of consecutive loads, just load that
// vector type.
SDValue Reduced = combineBVOfConsecutiveLoads(N, DAG);
if (Reduced)
return Reduced;
// If we're building a vector out of extended elements from another vector
// we have P9 vector integer extend instructions. The code assumes legal
// input types (i.e. it can't handle things like v4i16) so do not run before
// legalization.
if (Subtarget.hasP9Altivec() && !DCI.isBeforeLegalize()) {
Reduced = combineBVOfVecSExt(N, DAG);
if (Reduced)
return Reduced;
}
// On Power10, the Load VSX Vector Rightmost instructions can be utilized
// if this is a BUILD_VECTOR of v1i128, and if the operand to the BUILD_VECTOR
// is a load from <valid narrow width> to i128.
if (Subtarget.isISA3_1()) {
SDValue BVOfZLoad = combineBVZEXTLOAD(N, DAG);
if (BVOfZLoad)
return BVOfZLoad;
}
if (N->getValueType(0) != MVT::v2f64)
return SDValue();
// Looking for:
// (build_vector ([su]int_to_fp (extractelt 0)), [su]int_to_fp (extractelt 1))
if (FirstInput.getOpcode() != ISD::SINT_TO_FP &&
FirstInput.getOpcode() != ISD::UINT_TO_FP)
return SDValue();
if (N->getOperand(1).getOpcode() != ISD::SINT_TO_FP &&
N->getOperand(1).getOpcode() != ISD::UINT_TO_FP)
return SDValue();
if (FirstInput.getOpcode() != N->getOperand(1).getOpcode())
return SDValue();
SDValue Ext1 = FirstInput.getOperand(0);
SDValue Ext2 = N->getOperand(1).getOperand(0);
if(Ext1.getOpcode() != ISD::EXTRACT_VECTOR_ELT ||
Ext2.getOpcode() != ISD::EXTRACT_VECTOR_ELT)
return SDValue();
ConstantSDNode *Ext1Op = dyn_cast<ConstantSDNode>(Ext1.getOperand(1));
ConstantSDNode *Ext2Op = dyn_cast<ConstantSDNode>(Ext2.getOperand(1));
if (!Ext1Op || !Ext2Op)
return SDValue();
if (Ext1.getOperand(0).getValueType() != MVT::v4i32 ||
Ext1.getOperand(0) != Ext2.getOperand(0))
return SDValue();
int FirstElem = Ext1Op->getZExtValue();
int SecondElem = Ext2Op->getZExtValue();
int SubvecIdx;
if (FirstElem == 0 && SecondElem == 1)
SubvecIdx = Subtarget.isLittleEndian() ? 1 : 0;
else if (FirstElem == 2 && SecondElem == 3)
SubvecIdx = Subtarget.isLittleEndian() ? 0 : 1;
else
return SDValue();
SDValue SrcVec = Ext1.getOperand(0);
auto NodeType = (N->getOperand(1).getOpcode() == ISD::SINT_TO_FP) ?
PPCISD::SINT_VEC_TO_FP : PPCISD::UINT_VEC_TO_FP;
return DAG.getNode(NodeType, dl, MVT::v2f64,
SrcVec, DAG.getIntPtrConstant(SubvecIdx, dl));
}
SDValue PPCTargetLowering::combineFPToIntToFP(SDNode *N,
DAGCombinerInfo &DCI) const {
assert((N->getOpcode() == ISD::SINT_TO_FP ||
N->getOpcode() == ISD::UINT_TO_FP) &&
"Need an int -> FP conversion node here");
if (useSoftFloat() || !Subtarget.has64BitSupport())
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue Op(N, 0);
// Don't handle ppc_fp128 here or conversions that are out-of-range capable
// from the hardware.
if (Op.getValueType() != MVT::f32 && Op.getValueType() != MVT::f64)
return SDValue();
if (!Op.getOperand(0).getValueType().isSimple())
return SDValue();
if (Op.getOperand(0).getValueType().getSimpleVT() <= MVT(MVT::i1) ||
Op.getOperand(0).getValueType().getSimpleVT() > MVT(MVT::i64))
return SDValue();
SDValue FirstOperand(Op.getOperand(0));
bool SubWordLoad = FirstOperand.getOpcode() == ISD::LOAD &&
(FirstOperand.getValueType() == MVT::i8 ||
FirstOperand.getValueType() == MVT::i16);
if (Subtarget.hasP9Vector() && Subtarget.hasP9Altivec() && SubWordLoad) {
bool Signed = N->getOpcode() == ISD::SINT_TO_FP;
bool DstDouble = Op.getValueType() == MVT::f64;
unsigned ConvOp = Signed ?
(DstDouble ? PPCISD::FCFID : PPCISD::FCFIDS) :
(DstDouble ? PPCISD::FCFIDU : PPCISD::FCFIDUS);
SDValue WidthConst =
DAG.getIntPtrConstant(FirstOperand.getValueType() == MVT::i8 ? 1 : 2,
dl, false);
LoadSDNode *LDN = cast<LoadSDNode>(FirstOperand.getNode());
SDValue Ops[] = { LDN->getChain(), LDN->getBasePtr(), WidthConst };
SDValue Ld = DAG.getMemIntrinsicNode(PPCISD::LXSIZX, dl,
DAG.getVTList(MVT::f64, MVT::Other),
Ops, MVT::i8, LDN->getMemOperand());
// For signed conversion, we need to sign-extend the value in the VSR
if (Signed) {
SDValue ExtOps[] = { Ld, WidthConst };
SDValue Ext = DAG.getNode(PPCISD::VEXTS, dl, MVT::f64, ExtOps);
return DAG.getNode(ConvOp, dl, DstDouble ? MVT::f64 : MVT::f32, Ext);
} else
return DAG.getNode(ConvOp, dl, DstDouble ? MVT::f64 : MVT::f32, Ld);
}
// For i32 intermediate values, unfortunately, the conversion functions
// leave the upper 32 bits of the value are undefined. Within the set of
// scalar instructions, we have no method for zero- or sign-extending the
// value. Thus, we cannot handle i32 intermediate values here.
if (Op.getOperand(0).getValueType() == MVT::i32)
return SDValue();
assert((Op.getOpcode() == ISD::SINT_TO_FP || Subtarget.hasFPCVT()) &&
"UINT_TO_FP is supported only with FPCVT");
// If we have FCFIDS, then use it when converting to single-precision.
// Otherwise, convert to double-precision and then round.
unsigned FCFOp = (Subtarget.hasFPCVT() && Op.getValueType() == MVT::f32)
? (Op.getOpcode() == ISD::UINT_TO_FP ? PPCISD::FCFIDUS
: PPCISD::FCFIDS)
: (Op.getOpcode() == ISD::UINT_TO_FP ? PPCISD::FCFIDU
: PPCISD::FCFID);
MVT FCFTy = (Subtarget.hasFPCVT() && Op.getValueType() == MVT::f32)
? MVT::f32
: MVT::f64;
// If we're converting from a float, to an int, and back to a float again,
// then we don't need the store/load pair at all.
if ((Op.getOperand(0).getOpcode() == ISD::FP_TO_UINT &&
Subtarget.hasFPCVT()) ||
(Op.getOperand(0).getOpcode() == ISD::FP_TO_SINT)) {
SDValue Src = Op.getOperand(0).getOperand(0);
if (Src.getValueType() == MVT::f32) {
Src = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Src);
DCI.AddToWorklist(Src.getNode());
} else if (Src.getValueType() != MVT::f64) {
// Make sure that we don't pick up a ppc_fp128 source value.
return SDValue();
}
unsigned FCTOp =
Op.getOperand(0).getOpcode() == ISD::FP_TO_SINT ? PPCISD::FCTIDZ :
PPCISD::FCTIDUZ;
SDValue Tmp = DAG.getNode(FCTOp, dl, MVT::f64, Src);
SDValue FP = DAG.getNode(FCFOp, dl, FCFTy, Tmp);
if (Op.getValueType() == MVT::f32 && !Subtarget.hasFPCVT()) {
FP = DAG.getNode(ISD::FP_ROUND, dl,
MVT::f32, FP, DAG.getIntPtrConstant(0, dl));
DCI.AddToWorklist(FP.getNode());
}
return FP;
}
return SDValue();
}
// expandVSXLoadForLE - Convert VSX loads (which may be intrinsics for
// builtins) into loads with swaps.
SDValue PPCTargetLowering::expandVSXLoadForLE(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue Chain;
SDValue Base;
MachineMemOperand *MMO;
switch (N->getOpcode()) {
default:
llvm_unreachable("Unexpected opcode for little endian VSX load");
case ISD::LOAD: {
LoadSDNode *LD = cast<LoadSDNode>(N);
Chain = LD->getChain();
Base = LD->getBasePtr();
MMO = LD->getMemOperand();
// If the MMO suggests this isn't a load of a full vector, leave
// things alone. For a built-in, we have to make the change for
// correctness, so if there is a size problem that will be a bug.
if (MMO->getSize() < 16)
return SDValue();
break;
}
case ISD::INTRINSIC_W_CHAIN: {
MemIntrinsicSDNode *Intrin = cast<MemIntrinsicSDNode>(N);
Chain = Intrin->getChain();
// Similarly to the store case below, Intrin->getBasePtr() doesn't get
// us what we want. Get operand 2 instead.
Base = Intrin->getOperand(2);
MMO = Intrin->getMemOperand();
break;
}
}
MVT VecTy = N->getValueType(0).getSimpleVT();
// Do not expand to PPCISD::LXVD2X + PPCISD::XXSWAPD when the load is
// aligned and the type is a vector with elements up to 4 bytes
if (Subtarget.needsSwapsForVSXMemOps() && MMO->getAlign() >= Align(16) &&
VecTy.getScalarSizeInBits() <= 32) {
return SDValue();
}
SDValue LoadOps[] = { Chain, Base };
SDValue Load = DAG.getMemIntrinsicNode(PPCISD::LXVD2X, dl,
DAG.getVTList(MVT::v2f64, MVT::Other),
LoadOps, MVT::v2f64, MMO);
DCI.AddToWorklist(Load.getNode());
Chain = Load.getValue(1);
SDValue Swap = DAG.getNode(
PPCISD::XXSWAPD, dl, DAG.getVTList(MVT::v2f64, MVT::Other), Chain, Load);
DCI.AddToWorklist(Swap.getNode());
// Add a bitcast if the resulting load type doesn't match v2f64.
if (VecTy != MVT::v2f64) {
SDValue N = DAG.getNode(ISD::BITCAST, dl, VecTy, Swap);
DCI.AddToWorklist(N.getNode());
// Package {bitcast value, swap's chain} to match Load's shape.
return DAG.getNode(ISD::MERGE_VALUES, dl, DAG.getVTList(VecTy, MVT::Other),
N, Swap.getValue(1));
}
return Swap;
}
// expandVSXStoreForLE - Convert VSX stores (which may be intrinsics for
// builtins) into stores with swaps.
SDValue PPCTargetLowering::expandVSXStoreForLE(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue Chain;
SDValue Base;
unsigned SrcOpnd;
MachineMemOperand *MMO;
switch (N->getOpcode()) {
default:
llvm_unreachable("Unexpected opcode for little endian VSX store");
case ISD::STORE: {
StoreSDNode *ST = cast<StoreSDNode>(N);
Chain = ST->getChain();
Base = ST->getBasePtr();
MMO = ST->getMemOperand();
SrcOpnd = 1;
// If the MMO suggests this isn't a store of a full vector, leave
// things alone. For a built-in, we have to make the change for
// correctness, so if there is a size problem that will be a bug.
if (MMO->getSize() < 16)
return SDValue();
break;
}
case ISD::INTRINSIC_VOID: {
MemIntrinsicSDNode *Intrin = cast<MemIntrinsicSDNode>(N);
Chain = Intrin->getChain();
// Intrin->getBasePtr() oddly does not get what we want.
Base = Intrin->getOperand(3);
MMO = Intrin->getMemOperand();
SrcOpnd = 2;
break;
}
}
SDValue Src = N->getOperand(SrcOpnd);
MVT VecTy = Src.getValueType().getSimpleVT();
// Do not expand to PPCISD::XXSWAPD and PPCISD::STXVD2X when the load is
// aligned and the type is a vector with elements up to 4 bytes
if (Subtarget.needsSwapsForVSXMemOps() && MMO->getAlign() >= Align(16) &&
VecTy.getScalarSizeInBits() <= 32) {
return SDValue();
}
// All stores are done as v2f64 and possible bit cast.
if (VecTy != MVT::v2f64) {
Src = DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Src);
DCI.AddToWorklist(Src.getNode());
}
SDValue Swap = DAG.getNode(PPCISD::XXSWAPD, dl,
DAG.getVTList(MVT::v2f64, MVT::Other), Chain, Src);
DCI.AddToWorklist(Swap.getNode());
Chain = Swap.getValue(1);
SDValue StoreOps[] = { Chain, Swap, Base };
SDValue Store = DAG.getMemIntrinsicNode(PPCISD::STXVD2X, dl,
DAG.getVTList(MVT::Other),
StoreOps, VecTy, MMO);
DCI.AddToWorklist(Store.getNode());
return Store;
}
// Handle DAG combine for STORE (FP_TO_INT F).
SDValue PPCTargetLowering::combineStoreFPToInt(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
unsigned Opcode = N->getOperand(1).getOpcode();
assert((Opcode == ISD::FP_TO_SINT || Opcode == ISD::FP_TO_UINT)
&& "Not a FP_TO_INT Instruction!");
SDValue Val = N->getOperand(1).getOperand(0);
EVT Op1VT = N->getOperand(1).getValueType();
EVT ResVT = Val.getValueType();
if (!isTypeLegal(ResVT))
return SDValue();
// Only perform combine for conversion to i64/i32 or power9 i16/i8.
bool ValidTypeForStoreFltAsInt =
(Op1VT == MVT::i32 || Op1VT == MVT::i64 ||
(Subtarget.hasP9Vector() && (Op1VT == MVT::i16 || Op1VT == MVT::i8)));
if (ResVT == MVT::ppcf128 || !Subtarget.hasP8Vector() ||
cast<StoreSDNode>(N)->isTruncatingStore() || !ValidTypeForStoreFltAsInt)
return SDValue();
// Extend f32 values to f64
if (ResVT.getScalarSizeInBits() == 32) {
Val = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, Val);
DCI.AddToWorklist(Val.getNode());
}
// Set signed or unsigned conversion opcode.
unsigned ConvOpcode = (Opcode == ISD::FP_TO_SINT) ?
PPCISD::FP_TO_SINT_IN_VSR :
PPCISD::FP_TO_UINT_IN_VSR;
Val = DAG.getNode(ConvOpcode,
dl, ResVT == MVT::f128 ? MVT::f128 : MVT::f64, Val);
DCI.AddToWorklist(Val.getNode());
// Set number of bytes being converted.
unsigned ByteSize = Op1VT.getScalarSizeInBits() / 8;
SDValue Ops[] = { N->getOperand(0), Val, N->getOperand(2),
DAG.getIntPtrConstant(ByteSize, dl, false),
DAG.getValueType(Op1VT) };
Val = DAG.getMemIntrinsicNode(PPCISD::ST_VSR_SCAL_INT, dl,
DAG.getVTList(MVT::Other), Ops,
cast<StoreSDNode>(N)->getMemoryVT(),
cast<StoreSDNode>(N)->getMemOperand());
DCI.AddToWorklist(Val.getNode());
return Val;
}
static bool isAlternatingShuffMask(const ArrayRef<int> &Mask, int NumElts) {
// Check that the source of the element keeps flipping
// (i.e. Mask[i] < NumElts -> Mask[i+i] >= NumElts).
bool PrevElemFromFirstVec = Mask[0] < NumElts;
for (int i = 1, e = Mask.size(); i < e; i++) {
if (PrevElemFromFirstVec && Mask[i] < NumElts)
return false;
if (!PrevElemFromFirstVec && Mask[i] >= NumElts)
return false;
PrevElemFromFirstVec = !PrevElemFromFirstVec;
}
return true;
}
static bool isSplatBV(SDValue Op) {
if (Op.getOpcode() != ISD::BUILD_VECTOR)
return false;
SDValue FirstOp;
// Find first non-undef input.
for (int i = 0, e = Op.getNumOperands(); i < e; i++) {
FirstOp = Op.getOperand(i);
if (!FirstOp.isUndef())
break;
}
// All inputs are undef or the same as the first non-undef input.
for (int i = 1, e = Op.getNumOperands(); i < e; i++)
if (Op.getOperand(i) != FirstOp && !Op.getOperand(i).isUndef())
return false;
return true;
}
static SDValue isScalarToVec(SDValue Op) {
if (Op.getOpcode() == ISD::SCALAR_TO_VECTOR)
return Op;
if (Op.getOpcode() != ISD::BITCAST)
return SDValue();
Op = Op.getOperand(0);
if (Op.getOpcode() == ISD::SCALAR_TO_VECTOR)
return Op;
return SDValue();
}
static void fixupShuffleMaskForPermutedSToV(SmallVectorImpl<int> &ShuffV,
int LHSMaxIdx, int RHSMinIdx,
int RHSMaxIdx, int HalfVec) {
for (int i = 0, e = ShuffV.size(); i < e; i++) {
int Idx = ShuffV[i];
if ((Idx >= 0 && Idx < LHSMaxIdx) || (Idx >= RHSMinIdx && Idx < RHSMaxIdx))
ShuffV[i] += HalfVec;
}
}
// Replace a SCALAR_TO_VECTOR with a SCALAR_TO_VECTOR_PERMUTED except if
// the original is:
// (<n x Ty> (scalar_to_vector (Ty (extract_elt <n x Ty> %a, C))))
// In such a case, just change the shuffle mask to extract the element
// from the permuted index.
static SDValue getSToVPermuted(SDValue OrigSToV, SelectionDAG &DAG) {
SDLoc dl(OrigSToV);
EVT VT = OrigSToV.getValueType();
assert(OrigSToV.getOpcode() == ISD::SCALAR_TO_VECTOR &&
"Expecting a SCALAR_TO_VECTOR here");
SDValue Input = OrigSToV.getOperand(0);
if (Input.getOpcode() == ISD::EXTRACT_VECTOR_ELT) {
ConstantSDNode *Idx = dyn_cast<ConstantSDNode>(Input.getOperand(1));
SDValue OrigVector = Input.getOperand(0);
// Can't handle non-const element indices or different vector types
// for the input to the extract and the output of the scalar_to_vector.
if (Idx && VT == OrigVector.getValueType()) {
SmallVector<int, 16> NewMask(VT.getVectorNumElements(), -1);
NewMask[VT.getVectorNumElements() / 2] = Idx->getZExtValue();
return DAG.getVectorShuffle(VT, dl, OrigVector, OrigVector, NewMask);
}
}
return DAG.getNode(PPCISD::SCALAR_TO_VECTOR_PERMUTED, dl, VT,
OrigSToV.getOperand(0));
}
// On little endian subtargets, combine shuffles such as:
// vector_shuffle<16,1,17,3,18,5,19,7,20,9,21,11,22,13,23,15>, <zero>, %b
// into:
// vector_shuffle<16,0,17,1,18,2,19,3,20,4,21,5,22,6,23,7>, <zero>, %b
// because the latter can be matched to a single instruction merge.
// Furthermore, SCALAR_TO_VECTOR on little endian always involves a permute
// to put the value into element zero. Adjust the shuffle mask so that the
// vector can remain in permuted form (to prevent a swap prior to a shuffle).
SDValue PPCTargetLowering::combineVectorShuffle(ShuffleVectorSDNode *SVN,
SelectionDAG &DAG) const {
SDValue LHS = SVN->getOperand(0);
SDValue RHS = SVN->getOperand(1);
auto Mask = SVN->getMask();
int NumElts = LHS.getValueType().getVectorNumElements();
SDValue Res(SVN, 0);
SDLoc dl(SVN);
// None of these combines are useful on big endian systems since the ISA
// already has a big endian bias.
if (!Subtarget.isLittleEndian() || !Subtarget.hasVSX())
return Res;
// If this is not a shuffle of a shuffle and the first element comes from
// the second vector, canonicalize to the commuted form. This will make it
// more likely to match one of the single instruction patterns.
if (Mask[0] >= NumElts && LHS.getOpcode() != ISD::VECTOR_SHUFFLE &&
RHS.getOpcode() != ISD::VECTOR_SHUFFLE) {
std::swap(LHS, RHS);
Res = DAG.getCommutedVectorShuffle(*SVN);
Mask = cast<ShuffleVectorSDNode>(Res)->getMask();
}
// Adjust the shuffle mask if either input vector comes from a
// SCALAR_TO_VECTOR and keep the respective input vector in permuted
// form (to prevent the need for a swap).
SmallVector<int, 16> ShuffV(Mask.begin(), Mask.end());
SDValue SToVLHS = isScalarToVec(LHS);
SDValue SToVRHS = isScalarToVec(RHS);
if (SToVLHS || SToVRHS) {
int NumEltsIn = SToVLHS ? SToVLHS.getValueType().getVectorNumElements()
: SToVRHS.getValueType().getVectorNumElements();
int NumEltsOut = ShuffV.size();
// Initially assume that neither input is permuted. These will be adjusted
// accordingly if either input is.
int LHSMaxIdx = -1;
int RHSMinIdx = -1;
int RHSMaxIdx = -1;
int HalfVec = LHS.getValueType().getVectorNumElements() / 2;
// Get the permuted scalar to vector nodes for the source(s) that come from
// ISD::SCALAR_TO_VECTOR.
if (SToVLHS) {
// Set up the values for the shuffle vector fixup.
LHSMaxIdx = NumEltsOut / NumEltsIn;
SToVLHS = getSToVPermuted(SToVLHS, DAG);
if (SToVLHS.getValueType() != LHS.getValueType())
SToVLHS = DAG.getBitcast(LHS.getValueType(), SToVLHS);
LHS = SToVLHS;
}
if (SToVRHS) {
RHSMinIdx = NumEltsOut;
RHSMaxIdx = NumEltsOut / NumEltsIn + RHSMinIdx;
SToVRHS = getSToVPermuted(SToVRHS, DAG);
if (SToVRHS.getValueType() != RHS.getValueType())
SToVRHS = DAG.getBitcast(RHS.getValueType(), SToVRHS);
RHS = SToVRHS;
}
// Fix up the shuffle mask to reflect where the desired element actually is.
// The minimum and maximum indices that correspond to element zero for both
// the LHS and RHS are computed and will control which shuffle mask entries
// are to be changed. For example, if the RHS is permuted, any shuffle mask
// entries in the range [RHSMinIdx,RHSMaxIdx) will be incremented by
// HalfVec to refer to the corresponding element in the permuted vector.
fixupShuffleMaskForPermutedSToV(ShuffV, LHSMaxIdx, RHSMinIdx, RHSMaxIdx,
HalfVec);
Res = DAG.getVectorShuffle(SVN->getValueType(0), dl, LHS, RHS, ShuffV);
// We may have simplified away the shuffle. We won't be able to do anything
// further with it here.
if (!isa<ShuffleVectorSDNode>(Res))
return Res;
Mask = cast<ShuffleVectorSDNode>(Res)->getMask();
}
// The common case after we commuted the shuffle is that the RHS is a splat
// and we have elements coming in from the splat at indices that are not
// conducive to using a merge.
// Example:
// vector_shuffle<0,17,1,19,2,21,3,23,4,25,5,27,6,29,7,31> t1, <zero>
if (!isSplatBV(RHS))
return Res;
// We are looking for a mask such that all even elements are from
// one vector and all odd elements from the other.
if (!isAlternatingShuffMask(Mask, NumElts))
return Res;
// Adjust the mask so we are pulling in the same index from the splat
// as the index from the interesting vector in consecutive elements.
// Example (even elements from first vector):
// vector_shuffle<0,16,1,17,2,18,3,19,4,20,5,21,6,22,7,23> t1, <zero>
if (Mask[0] < NumElts)
for (int i = 1, e = Mask.size(); i < e; i += 2)
ShuffV[i] = (ShuffV[i - 1] + NumElts);
// Example (odd elements from first vector):
// vector_shuffle<16,0,17,1,18,2,19,3,20,4,21,5,22,6,23,7> t1, <zero>
else
for (int i = 0, e = Mask.size(); i < e; i += 2)
ShuffV[i] = (ShuffV[i + 1] + NumElts);
// If the RHS has undefs, we need to remove them since we may have created
// a shuffle that adds those instead of the splat value.
SDValue SplatVal = cast<BuildVectorSDNode>(RHS.getNode())->getSplatValue();
RHS = DAG.getSplatBuildVector(RHS.getValueType(), dl, SplatVal);
Res = DAG.getVectorShuffle(SVN->getValueType(0), dl, LHS, RHS, ShuffV);
return Res;
}
SDValue PPCTargetLowering::combineVReverseMemOP(ShuffleVectorSDNode *SVN,
LSBaseSDNode *LSBase,
DAGCombinerInfo &DCI) const {
assert((ISD::isNormalLoad(LSBase) || ISD::isNormalStore(LSBase)) &&
"Not a reverse memop pattern!");
auto IsElementReverse = [](const ShuffleVectorSDNode *SVN) -> bool {
auto Mask = SVN->getMask();
int i = 0;
auto I = Mask.rbegin();
auto E = Mask.rend();
for (; I != E; ++I) {
if (*I != i)
return false;
i++;
}
return true;
};
SelectionDAG &DAG = DCI.DAG;
EVT VT = SVN->getValueType(0);
if (!isTypeLegal(VT) || !Subtarget.isLittleEndian() || !Subtarget.hasVSX())
return SDValue();
// Before P9, we have PPCVSXSwapRemoval pass to hack the element order.
// See comment in PPCVSXSwapRemoval.cpp.
// It is conflict with PPCVSXSwapRemoval opt. So we don't do it.
if (!Subtarget.hasP9Vector())
return SDValue();
if(!IsElementReverse(SVN))
return SDValue();
if (LSBase->getOpcode() == ISD::LOAD) {
SDLoc dl(SVN);
SDValue LoadOps[] = {LSBase->getChain(), LSBase->getBasePtr()};
return DAG.getMemIntrinsicNode(
PPCISD::LOAD_VEC_BE, dl, DAG.getVTList(VT, MVT::Other), LoadOps,
LSBase->getMemoryVT(), LSBase->getMemOperand());
}
if (LSBase->getOpcode() == ISD::STORE) {
SDLoc dl(LSBase);
SDValue StoreOps[] = {LSBase->getChain(), SVN->getOperand(0),
LSBase->getBasePtr()};
return DAG.getMemIntrinsicNode(
PPCISD::STORE_VEC_BE, dl, DAG.getVTList(MVT::Other), StoreOps,
LSBase->getMemoryVT(), LSBase->getMemOperand());
}
llvm_unreachable("Expected a load or store node here");
}
SDValue PPCTargetLowering::PerformDAGCombine(SDNode *N,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
switch (N->getOpcode()) {
default: break;
case ISD::ADD:
return combineADD(N, DCI);
case ISD::SHL:
return combineSHL(N, DCI);
case ISD::SRA:
return combineSRA(N, DCI);
case ISD::SRL:
return combineSRL(N, DCI);
case ISD::MUL:
return combineMUL(N, DCI);
case ISD::FMA:
case PPCISD::FNMSUB:
return combineFMALike(N, DCI);
case PPCISD::SHL:
if (isNullConstant(N->getOperand(0))) // 0 << V -> 0.
return N->getOperand(0);
break;
case PPCISD::SRL:
if (isNullConstant(N->getOperand(0))) // 0 >>u V -> 0.
return N->getOperand(0);
break;
case PPCISD::SRA:
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(0))) {
if (C->isNullValue() || // 0 >>s V -> 0.
C->isAllOnesValue()) // -1 >>s V -> -1.
return N->getOperand(0);
}
break;
case ISD::SIGN_EXTEND:
case ISD::ZERO_EXTEND:
case ISD::ANY_EXTEND:
return DAGCombineExtBoolTrunc(N, DCI);
case ISD::TRUNCATE:
return combineTRUNCATE(N, DCI);
case ISD::SETCC:
if (SDValue CSCC = combineSetCC(N, DCI))
return CSCC;
LLVM_FALLTHROUGH;
case ISD::SELECT_CC:
return DAGCombineTruncBoolExt(N, DCI);
case ISD::SINT_TO_FP:
case ISD::UINT_TO_FP:
return combineFPToIntToFP(N, DCI);
case ISD::VECTOR_SHUFFLE:
if (ISD::isNormalLoad(N->getOperand(0).getNode())) {
LSBaseSDNode* LSBase = cast<LSBaseSDNode>(N->getOperand(0));
return combineVReverseMemOP(cast<ShuffleVectorSDNode>(N), LSBase, DCI);
}
return combineVectorShuffle(cast<ShuffleVectorSDNode>(N), DCI.DAG);
case ISD::STORE: {
EVT Op1VT = N->getOperand(1).getValueType();
unsigned Opcode = N->getOperand(1).getOpcode();
if (Opcode == ISD::FP_TO_SINT || Opcode == ISD::FP_TO_UINT) {
SDValue Val= combineStoreFPToInt(N, DCI);
if (Val)
return Val;
}
if (Opcode == ISD::VECTOR_SHUFFLE && ISD::isNormalStore(N)) {
ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N->getOperand(1));
SDValue Val= combineVReverseMemOP(SVN, cast<LSBaseSDNode>(N), DCI);
if (Val)
return Val;
}
// Turn STORE (BSWAP) -> sthbrx/stwbrx.
if (cast<StoreSDNode>(N)->isUnindexed() && Opcode == ISD::BSWAP &&
N->getOperand(1).getNode()->hasOneUse() &&
(Op1VT == MVT::i32 || Op1VT == MVT::i16 ||
(Subtarget.hasLDBRX() && Subtarget.isPPC64() && Op1VT == MVT::i64))) {
// STBRX can only handle simple types and it makes no sense to store less
// two bytes in byte-reversed order.
EVT mVT = cast<StoreSDNode>(N)->getMemoryVT();
if (mVT.isExtended() || mVT.getSizeInBits() < 16)
break;
SDValue BSwapOp = N->getOperand(1).getOperand(0);
// Do an any-extend to 32-bits if this is a half-word input.
if (BSwapOp.getValueType() == MVT::i16)
BSwapOp = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, BSwapOp);
// If the type of BSWAP operand is wider than stored memory width
// it need to be shifted to the right side before STBRX.
if (Op1VT.bitsGT(mVT)) {
int Shift = Op1VT.getSizeInBits() - mVT.getSizeInBits();
BSwapOp = DAG.getNode(ISD::SRL, dl, Op1VT, BSwapOp,
DAG.getConstant(Shift, dl, MVT::i32));
// Need to truncate if this is a bswap of i64 stored as i32/i16.
if (Op1VT == MVT::i64)
BSwapOp = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, BSwapOp);
}
SDValue Ops[] = {
N->getOperand(0), BSwapOp, N->getOperand(2), DAG.getValueType(mVT)
};
return
DAG.getMemIntrinsicNode(PPCISD::STBRX, dl, DAG.getVTList(MVT::Other),
Ops, cast<StoreSDNode>(N)->getMemoryVT(),
cast<StoreSDNode>(N)->getMemOperand());
}
// STORE Constant:i32<0> -> STORE<trunc to i32> Constant:i64<0>
// So it can increase the chance of CSE constant construction.
if (Subtarget.isPPC64() && !DCI.isBeforeLegalize() &&
isa<ConstantSDNode>(N->getOperand(1)) && Op1VT == MVT::i32) {
// Need to sign-extended to 64-bits to handle negative values.
EVT MemVT = cast<StoreSDNode>(N)->getMemoryVT();
uint64_t Val64 = SignExtend64(N->getConstantOperandVal(1),
MemVT.getSizeInBits());
SDValue Const64 = DAG.getConstant(Val64, dl, MVT::i64);
// DAG.getTruncStore() can't be used here because it doesn't accept
// the general (base + offset) addressing mode.
// So we use UpdateNodeOperands and setTruncatingStore instead.
DAG.UpdateNodeOperands(N, N->getOperand(0), Const64, N->getOperand(2),
N->getOperand(3));
cast<StoreSDNode>(N)->setTruncatingStore(true);
return SDValue(N, 0);
}
// For little endian, VSX stores require generating xxswapd/lxvd2x.
// Not needed on ISA 3.0 based CPUs since we have a non-permuting store.
if (Op1VT.isSimple()) {
MVT StoreVT = Op1VT.getSimpleVT();
if (Subtarget.needsSwapsForVSXMemOps() &&
(StoreVT == MVT::v2f64 || StoreVT == MVT::v2i64 ||
StoreVT == MVT::v4f32 || StoreVT == MVT::v4i32))
return expandVSXStoreForLE(N, DCI);
}
break;
}
case ISD::LOAD: {
LoadSDNode *LD = cast<LoadSDNode>(N);
EVT VT = LD->getValueType(0);
// For little endian, VSX loads require generating lxvd2x/xxswapd.
// Not needed on ISA 3.0 based CPUs since we have a non-permuting load.
if (VT.isSimple()) {
MVT LoadVT = VT.getSimpleVT();
if (Subtarget.needsSwapsForVSXMemOps() &&
(LoadVT == MVT::v2f64 || LoadVT == MVT::v2i64 ||
LoadVT == MVT::v4f32 || LoadVT == MVT::v4i32))
return expandVSXLoadForLE(N, DCI);
}
// We sometimes end up with a 64-bit integer load, from which we extract
// two single-precision floating-point numbers. This happens with
// std::complex<float>, and other similar structures, because of the way we
// canonicalize structure copies. However, if we lack direct moves,
// then the final bitcasts from the extracted integer values to the
// floating-point numbers turn into store/load pairs. Even with direct moves,
// just loading the two floating-point numbers is likely better.
auto ReplaceTwoFloatLoad = [&]() {
if (VT != MVT::i64)
return false;
if (LD->getExtensionType() != ISD::NON_EXTLOAD ||
LD->isVolatile())
return false;
// We're looking for a sequence like this:
// t13: i64,ch = load<LD8[%ref.tmp]> t0, t6, undef:i64
// t16: i64 = srl t13, Constant:i32<32>
// t17: i32 = truncate t16
// t18: f32 = bitcast t17
// t19: i32 = truncate t13
// t20: f32 = bitcast t19
if (!LD->hasNUsesOfValue(2, 0))
return false;
auto UI = LD->use_begin();
while (UI.getUse().getResNo() != 0) ++UI;
SDNode *Trunc = *UI++;
while (UI.getUse().getResNo() != 0) ++UI;
SDNode *RightShift = *UI;
if (Trunc->getOpcode() != ISD::TRUNCATE)
std::swap(Trunc, RightShift);
if (Trunc->getOpcode() != ISD::TRUNCATE ||
Trunc->getValueType(0) != MVT::i32 ||
!Trunc->hasOneUse())
return false;
if (RightShift->getOpcode() != ISD::SRL ||
!isa<ConstantSDNode>(RightShift->getOperand(1)) ||
RightShift->getConstantOperandVal(1) != 32 ||
!RightShift->hasOneUse())
return false;
SDNode *Trunc2 = *RightShift->use_begin();
if (Trunc2->getOpcode() != ISD::TRUNCATE ||
Trunc2->getValueType(0) != MVT::i32 ||
!Trunc2->hasOneUse())
return false;
SDNode *Bitcast = *Trunc->use_begin();
SDNode *Bitcast2 = *Trunc2->use_begin();
if (Bitcast->getOpcode() != ISD::BITCAST ||
Bitcast->getValueType(0) != MVT::f32)
return false;
if (Bitcast2->getOpcode() != ISD::BITCAST ||
Bitcast2->getValueType(0) != MVT::f32)
return false;
if (Subtarget.isLittleEndian())
std::swap(Bitcast, Bitcast2);
// Bitcast has the second float (in memory-layout order) and Bitcast2
// has the first one.
SDValue BasePtr = LD->getBasePtr();
if (LD->isIndexed()) {
assert(LD->getAddressingMode() == ISD::PRE_INC &&
"Non-pre-inc AM on PPC?");
BasePtr =
DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(), BasePtr,
LD->getOffset());
}
auto MMOFlags =
LD->getMemOperand()->getFlags() & ~MachineMemOperand::MOVolatile;
SDValue FloatLoad = DAG.getLoad(MVT::f32, dl, LD->getChain(), BasePtr,
LD->getPointerInfo(), LD->getAlignment(),
MMOFlags, LD->getAAInfo());
SDValue AddPtr =
DAG.getNode(ISD::ADD, dl, BasePtr.getValueType(),
BasePtr, DAG.getIntPtrConstant(4, dl));
SDValue FloatLoad2 = DAG.getLoad(
MVT::f32, dl, SDValue(FloatLoad.getNode(), 1), AddPtr,
LD->getPointerInfo().getWithOffset(4),
MinAlign(LD->getAlignment(), 4), MMOFlags, LD->getAAInfo());
if (LD->isIndexed()) {
// Note that DAGCombine should re-form any pre-increment load(s) from
// what is produced here if that makes sense.
DAG.ReplaceAllUsesOfValueWith(SDValue(LD, 1), BasePtr);
}
DCI.CombineTo(Bitcast2, FloatLoad);
DCI.CombineTo(Bitcast, FloatLoad2);
DAG.ReplaceAllUsesOfValueWith(SDValue(LD, LD->isIndexed() ? 2 : 1),
SDValue(FloatLoad2.getNode(), 1));
return true;
};
if (ReplaceTwoFloatLoad())
return SDValue(N, 0);
EVT MemVT = LD->getMemoryVT();
Type *Ty = MemVT.getTypeForEVT(*DAG.getContext());
Align ABIAlignment = DAG.getDataLayout().getABITypeAlign(Ty);
if (LD->isUnindexed() && VT.isVector() &&
((Subtarget.hasAltivec() && ISD::isNON_EXTLoad(N) &&
// P8 and later hardware should just use LOAD.
!Subtarget.hasP8Vector() &&
(VT == MVT::v16i8 || VT == MVT::v8i16 || VT == MVT::v4i32 ||
VT == MVT::v4f32))) &&
LD->getAlign() < ABIAlignment) {
// This is a type-legal unaligned Altivec load.
SDValue Chain = LD->getChain();
SDValue Ptr = LD->getBasePtr();
bool isLittleEndian = Subtarget.isLittleEndian();
// This implements the loading of unaligned vectors as described in
// the venerable Apple Velocity Engine overview. Specifically:
// https://developer.apple.com/hardwaredrivers/ve/alignment.html
// https://developer.apple.com/hardwaredrivers/ve/code_optimization.html
//
// The general idea is to expand a sequence of one or more unaligned
// loads into an alignment-based permutation-control instruction (lvsl
// or lvsr), a series of regular vector loads (which always truncate
// their input address to an aligned address), and a series of
// permutations. The results of these permutations are the requested
// loaded values. The trick is that the last "extra" load is not taken
// from the address you might suspect (sizeof(vector) bytes after the
// last requested load), but rather sizeof(vector) - 1 bytes after the
// last requested vector. The point of this is to avoid a page fault if
// the base address happened to be aligned. This works because if the
// base address is aligned, then adding less than a full vector length
// will cause the last vector in the sequence to be (re)loaded.
// Otherwise, the next vector will be fetched as you might suspect was
// necessary.
// We might be able to reuse the permutation generation from
// a different base address offset from this one by an aligned amount.
// The INTRINSIC_WO_CHAIN DAG combine will attempt to perform this
// optimization later.
Intrinsic::ID Intr, IntrLD, IntrPerm;
MVT PermCntlTy, PermTy, LDTy;
Intr = isLittleEndian ? Intrinsic::ppc_altivec_lvsr
: Intrinsic::ppc_altivec_lvsl;
IntrLD = Intrinsic::ppc_altivec_lvx;
IntrPerm = Intrinsic::ppc_altivec_vperm;
PermCntlTy = MVT::v16i8;
PermTy = MVT::v4i32;
LDTy = MVT::v4i32;
SDValue PermCntl = BuildIntrinsicOp(Intr, Ptr, DAG, dl, PermCntlTy);
// Create the new MMO for the new base load. It is like the original MMO,
// but represents an area in memory almost twice the vector size centered
// on the original address. If the address is unaligned, we might start
// reading up to (sizeof(vector)-1) bytes below the address of the
// original unaligned load.
MachineFunction &MF = DAG.getMachineFunction();
MachineMemOperand *BaseMMO =
MF.getMachineMemOperand(LD->getMemOperand(),
-(long)MemVT.getStoreSize()+1,
2*MemVT.getStoreSize()-1);
// Create the new base load.
SDValue LDXIntID =
DAG.getTargetConstant(IntrLD, dl, getPointerTy(MF.getDataLayout()));
SDValue BaseLoadOps[] = { Chain, LDXIntID, Ptr };
SDValue BaseLoad =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, dl,
DAG.getVTList(PermTy, MVT::Other),
BaseLoadOps, LDTy, BaseMMO);
// Note that the value of IncOffset (which is provided to the next
// load's pointer info offset value, and thus used to calculate the
// alignment), and the value of IncValue (which is actually used to
// increment the pointer value) are different! This is because we
// require the next load to appear to be aligned, even though it
// is actually offset from the base pointer by a lesser amount.
int IncOffset = VT.getSizeInBits() / 8;
int IncValue = IncOffset;
// Walk (both up and down) the chain looking for another load at the real
// (aligned) offset (the alignment of the other load does not matter in
// this case). If found, then do not use the offset reduction trick, as
// that will prevent the loads from being later combined (as they would
// otherwise be duplicates).
if (!findConsecutiveLoad(LD, DAG))
--IncValue;
SDValue Increment =
DAG.getConstant(IncValue, dl, getPointerTy(MF.getDataLayout()));
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, Increment);
MachineMemOperand *ExtraMMO =
MF.getMachineMemOperand(LD->getMemOperand(),
1, 2*MemVT.getStoreSize()-1);
SDValue ExtraLoadOps[] = { Chain, LDXIntID, Ptr };
SDValue ExtraLoad =
DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, dl,
DAG.getVTList(PermTy, MVT::Other),
ExtraLoadOps, LDTy, ExtraMMO);
SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
BaseLoad.getValue(1), ExtraLoad.getValue(1));
// Because vperm has a big-endian bias, we must reverse the order
// of the input vectors and complement the permute control vector
// when generating little endian code. We have already handled the
// latter by using lvsr instead of lvsl, so just reverse BaseLoad
// and ExtraLoad here.
SDValue Perm;
if (isLittleEndian)
Perm = BuildIntrinsicOp(IntrPerm,
ExtraLoad, BaseLoad, PermCntl, DAG, dl);
else
Perm = BuildIntrinsicOp(IntrPerm,
BaseLoad, ExtraLoad, PermCntl, DAG, dl);
if (VT != PermTy)
Perm = Subtarget.hasAltivec()
? DAG.getNode(ISD::BITCAST, dl, VT, Perm)
: DAG.getNode(ISD::FP_ROUND, dl, VT, Perm,
DAG.getTargetConstant(1, dl, MVT::i64));
// second argument is 1 because this rounding
// is always exact.
// The output of the permutation is our loaded result, the TokenFactor is
// our new chain.
DCI.CombineTo(N, Perm, TF);
return SDValue(N, 0);
}
}
break;
case ISD::INTRINSIC_WO_CHAIN: {
bool isLittleEndian = Subtarget.isLittleEndian();
unsigned IID = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
Intrinsic::ID Intr = (isLittleEndian ? Intrinsic::ppc_altivec_lvsr
: Intrinsic::ppc_altivec_lvsl);
if (IID == Intr && N->getOperand(1)->getOpcode() == ISD::ADD) {
SDValue Add = N->getOperand(1);
int Bits = 4 /* 16 byte alignment */;
if (DAG.MaskedValueIsZero(Add->getOperand(1),
APInt::getAllOnesValue(Bits /* alignment */)
.zext(Add.getScalarValueSizeInBits()))) {
SDNode *BasePtr = Add->getOperand(0).getNode();
for (SDNode::use_iterator UI = BasePtr->use_begin(),
UE = BasePtr->use_end();
UI != UE; ++UI) {
if (UI->getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
cast<ConstantSDNode>(UI->getOperand(0))->getZExtValue() ==
IID) {
// We've found another LVSL/LVSR, and this address is an aligned
// multiple of that one. The results will be the same, so use the
// one we've just found instead.
return SDValue(*UI, 0);
}
}
}
if (isa<ConstantSDNode>(Add->getOperand(1))) {
SDNode *BasePtr = Add->getOperand(0).getNode();
for (SDNode::use_iterator UI = BasePtr->use_begin(),
UE = BasePtr->use_end(); UI != UE; ++UI) {
if (UI->getOpcode() == ISD::ADD &&
isa<ConstantSDNode>(UI->getOperand(1)) &&
(cast<ConstantSDNode>(Add->getOperand(1))->getZExtValue() -
cast<ConstantSDNode>(UI->getOperand(1))->getZExtValue()) %
(1ULL << Bits) == 0) {
SDNode *OtherAdd = *UI;
for (SDNode::use_iterator VI = OtherAdd->use_begin(),
VE = OtherAdd->use_end(); VI != VE; ++VI) {
if (VI->getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
cast<ConstantSDNode>(VI->getOperand(0))->getZExtValue() == IID) {
return SDValue(*VI, 0);
}
}
}
}
}
}
// Combine vmaxsw/h/b(a, a's negation) to abs(a)
// Expose the vabsduw/h/b opportunity for down stream
if (!DCI.isAfterLegalizeDAG() && Subtarget.hasP9Altivec() &&
(IID == Intrinsic::ppc_altivec_vmaxsw ||
IID == Intrinsic::ppc_altivec_vmaxsh ||
IID == Intrinsic::ppc_altivec_vmaxsb)) {
SDValue V1 = N->getOperand(1);
SDValue V2 = N->getOperand(2);
if ((V1.getSimpleValueType() == MVT::v4i32 ||
V1.getSimpleValueType() == MVT::v8i16 ||
V1.getSimpleValueType() == MVT::v16i8) &&
V1.getSimpleValueType() == V2.getSimpleValueType()) {
// (0-a, a)
if (V1.getOpcode() == ISD::SUB &&
ISD::isBuildVectorAllZeros(V1.getOperand(0).getNode()) &&
V1.getOperand(1) == V2) {
return DAG.getNode(ISD::ABS, dl, V2.getValueType(), V2);
}
// (a, 0-a)
if (V2.getOpcode() == ISD::SUB &&
ISD::isBuildVectorAllZeros(V2.getOperand(0).getNode()) &&
V2.getOperand(1) == V1) {
return DAG.getNode(ISD::ABS, dl, V1.getValueType(), V1);
}
// (x-y, y-x)
if (V1.getOpcode() == ISD::SUB && V2.getOpcode() == ISD::SUB &&
V1.getOperand(0) == V2.getOperand(1) &&
V1.getOperand(1) == V2.getOperand(0)) {
return DAG.getNode(ISD::ABS, dl, V1.getValueType(), V1);
}
}
}
}
break;
case ISD::INTRINSIC_W_CHAIN:
// For little endian, VSX loads require generating lxvd2x/xxswapd.
// Not needed on ISA 3.0 based CPUs since we have a non-permuting load.
if (Subtarget.needsSwapsForVSXMemOps()) {
switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
default:
break;
case Intrinsic::ppc_vsx_lxvw4x:
case Intrinsic::ppc_vsx_lxvd2x:
return expandVSXLoadForLE(N, DCI);
}
}
break;
case ISD::INTRINSIC_VOID:
// For little endian, VSX stores require generating xxswapd/stxvd2x.
// Not needed on ISA 3.0 based CPUs since we have a non-permuting store.
if (Subtarget.needsSwapsForVSXMemOps()) {
switch (cast<ConstantSDNode>(N->getOperand(1))->getZExtValue()) {
default:
break;
case Intrinsic::ppc_vsx_stxvw4x:
case Intrinsic::ppc_vsx_stxvd2x:
return expandVSXStoreForLE(N, DCI);
}
}
break;
case ISD::BSWAP:
// Turn BSWAP (LOAD) -> lhbrx/lwbrx.
if (ISD::isNON_EXTLoad(N->getOperand(0).getNode()) &&
N->getOperand(0).hasOneUse() &&
(N->getValueType(0) == MVT::i32 || N->getValueType(0) == MVT::i16 ||
(Subtarget.hasLDBRX() && Subtarget.isPPC64() &&
N->getValueType(0) == MVT::i64))) {
SDValue Load = N->getOperand(0);
LoadSDNode *LD = cast<LoadSDNode>(Load);
// Create the byte-swapping load.
SDValue Ops[] = {
LD->getChain(), // Chain
LD->getBasePtr(), // Ptr
DAG.getValueType(N->getValueType(0)) // VT
};
SDValue BSLoad =
DAG.getMemIntrinsicNode(PPCISD::LBRX, dl,
DAG.getVTList(N->getValueType(0) == MVT::i64 ?
MVT::i64 : MVT::i32, MVT::Other),
Ops, LD->getMemoryVT(), LD->getMemOperand());
// If this is an i16 load, insert the truncate.
SDValue ResVal = BSLoad;
if (N->getValueType(0) == MVT::i16)
ResVal = DAG.getNode(ISD::TRUNCATE, dl, MVT::i16, BSLoad);
// First, combine the bswap away. This makes the value produced by the
// load dead.
DCI.CombineTo(N, ResVal);
// Next, combine the load away, we give it a bogus result value but a real
// chain result. The result value is dead because the bswap is dead.
DCI.CombineTo(Load.getNode(), ResVal, BSLoad.getValue(1));
// Return N so it doesn't get rechecked!
return SDValue(N, 0);
}
break;
case PPCISD::VCMP:
// If a VCMP_rec node already exists with exactly the same operands as this
// node, use its result instead of this node (VCMP_rec computes both a CR6
// and a normal output).
//
if (!N->getOperand(0).hasOneUse() &&
!N->getOperand(1).hasOneUse() &&
!N->getOperand(2).hasOneUse()) {
// Scan all of the users of the LHS, looking for VCMP_rec's that match.
SDNode *VCMPrecNode = nullptr;
SDNode *LHSN = N->getOperand(0).getNode();
for (SDNode::use_iterator UI = LHSN->use_begin(), E = LHSN->use_end();
UI != E; ++UI)
if (UI->getOpcode() == PPCISD::VCMP_rec &&
UI->getOperand(1) == N->getOperand(1) &&
UI->getOperand(2) == N->getOperand(2) &&
UI->getOperand(0) == N->getOperand(0)) {
VCMPrecNode = *UI;
break;
}
// If there is no VCMP_rec node, or if the flag value has a single use,
// don't transform this.
if (!VCMPrecNode || VCMPrecNode->hasNUsesOfValue(0, 1))
break;
// Look at the (necessarily single) use of the flag value. If it has a
// chain, this transformation is more complex. Note that multiple things
// could use the value result, which we should ignore.
SDNode *FlagUser = nullptr;
for (SDNode::use_iterator UI = VCMPrecNode->use_begin();
FlagUser == nullptr; ++UI) {
assert(UI != VCMPrecNode->use_end() && "Didn't find user!");
SDNode *User = *UI;
for (unsigned i = 0, e = User->getNumOperands(); i != e; ++i) {
if (User->getOperand(i) == SDValue(VCMPrecNode, 1)) {
FlagUser = User;
break;
}
}
}
// If the user is a MFOCRF instruction, we know this is safe.
// Otherwise we give up for right now.
if (FlagUser->getOpcode() == PPCISD::MFOCRF)
return SDValue(VCMPrecNode, 0);
}
break;
case ISD::BRCOND: {
SDValue Cond = N->getOperand(1);
SDValue Target = N->getOperand(2);
if (Cond.getOpcode() == ISD::INTRINSIC_W_CHAIN &&
cast<ConstantSDNode>(Cond.getOperand(1))->getZExtValue() ==
Intrinsic::loop_decrement) {
// We now need to make the intrinsic dead (it cannot be instruction
// selected).
DAG.ReplaceAllUsesOfValueWith(Cond.getValue(1), Cond.getOperand(0));
assert(Cond.getNode()->hasOneUse() &&
"Counter decrement has more than one use");
return DAG.getNode(PPCISD::BDNZ, dl, MVT::Other,
N->getOperand(0), Target);
}
}
break;
case ISD::BR_CC: {
// If this is a branch on an altivec predicate comparison, lower this so
// that we don't have to do a MFOCRF: instead, branch directly on CR6. This
// lowering is done pre-legalize, because the legalizer lowers the predicate
// compare down to code that is difficult to reassemble.
ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(1))->get();
SDValue LHS = N->getOperand(2), RHS = N->getOperand(3);
// Sometimes the promoted value of the intrinsic is ANDed by some non-zero
// value. If so, pass-through the AND to get to the intrinsic.
if (LHS.getOpcode() == ISD::AND &&
LHS.getOperand(0).getOpcode() == ISD::INTRINSIC_W_CHAIN &&
cast<ConstantSDNode>(LHS.getOperand(0).getOperand(1))->getZExtValue() ==
Intrinsic::loop_decrement &&
isa<ConstantSDNode>(LHS.getOperand(1)) &&
!isNullConstant(LHS.getOperand(1)))
LHS = LHS.getOperand(0);
if (LHS.getOpcode() == ISD::INTRINSIC_W_CHAIN &&
cast<ConstantSDNode>(LHS.getOperand(1))->getZExtValue() ==
Intrinsic::loop_decrement &&
isa<ConstantSDNode>(RHS)) {
assert((CC == ISD::SETEQ || CC == ISD::SETNE) &&
"Counter decrement comparison is not EQ or NE");
unsigned Val = cast<ConstantSDNode>(RHS)->getZExtValue();
bool isBDNZ = (CC == ISD::SETEQ && Val) ||
(CC == ISD::SETNE && !Val);
// We now need to make the intrinsic dead (it cannot be instruction
// selected).
DAG.ReplaceAllUsesOfValueWith(LHS.getValue(1), LHS.getOperand(0));
assert(LHS.getNode()->hasOneUse() &&
"Counter decrement has more than one use");
return DAG.getNode(isBDNZ ? PPCISD::BDNZ : PPCISD::BDZ, dl, MVT::Other,
N->getOperand(0), N->getOperand(4));
}
int CompareOpc;
bool isDot;
if (LHS.getOpcode() == ISD::INTRINSIC_WO_CHAIN &&
isa<ConstantSDNode>(RHS) && (CC == ISD::SETEQ || CC == ISD::SETNE) &&
getVectorCompareInfo(LHS, CompareOpc, isDot, Subtarget)) {
assert(isDot && "Can't compare against a vector result!");
// If this is a comparison against something other than 0/1, then we know
// that the condition is never/always true.
unsigned Val = cast<ConstantSDNode>(RHS)->getZExtValue();
if (Val != 0 && Val != 1) {
if (CC == ISD::SETEQ) // Cond never true, remove branch.
return N->getOperand(0);
// Always !=, turn it into an unconditional branch.
return DAG.getNode(ISD::BR, dl, MVT::Other,
N->getOperand(0), N->getOperand(4));
}
bool BranchOnWhenPredTrue = (CC == ISD::SETEQ) ^ (Val == 0);
// Create the PPCISD altivec 'dot' comparison node.
SDValue Ops[] = {
LHS.getOperand(2), // LHS of compare
LHS.getOperand(3), // RHS of compare
DAG.getConstant(CompareOpc, dl, MVT::i32)
};
EVT VTs[] = { LHS.getOperand(2).getValueType(), MVT::Glue };
SDValue CompNode = DAG.getNode(PPCISD::VCMP_rec, dl, VTs, Ops);
// Unpack the result based on how the target uses it.
PPC::Predicate CompOpc;
switch (cast<ConstantSDNode>(LHS.getOperand(1))->getZExtValue()) {
default: // Can't happen, don't crash on invalid number though.
case 0: // Branch on the value of the EQ bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_EQ : PPC::PRED_NE;
break;
case 1: // Branch on the inverted value of the EQ bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_NE : PPC::PRED_EQ;
break;
case 2: // Branch on the value of the LT bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_LT : PPC::PRED_GE;
break;
case 3: // Branch on the inverted value of the LT bit of CR6.
CompOpc = BranchOnWhenPredTrue ? PPC::PRED_GE : PPC::PRED_LT;
break;
}
return DAG.getNode(PPCISD::COND_BRANCH, dl, MVT::Other, N->getOperand(0),
DAG.getConstant(CompOpc, dl, MVT::i32),
DAG.getRegister(PPC::CR6, MVT::i32),
N->getOperand(4), CompNode.getValue(1));
}
break;
}
case ISD::BUILD_VECTOR:
return DAGCombineBuildVector(N, DCI);
case ISD::ABS:
return combineABS(N, DCI);
case ISD::VSELECT:
return combineVSelect(N, DCI);
}
return SDValue();
}
SDValue
PPCTargetLowering::BuildSDIVPow2(SDNode *N, const APInt &Divisor,
SelectionDAG &DAG,
SmallVectorImpl<SDNode *> &Created) const {
// fold (sdiv X, pow2)
EVT VT = N->getValueType(0);
if (VT == MVT::i64 && !Subtarget.isPPC64())
return SDValue();
if ((VT != MVT::i32 && VT != MVT::i64) ||
!(Divisor.isPowerOf2() || (-Divisor).isPowerOf2()))
return SDValue();
SDLoc DL(N);
SDValue N0 = N->getOperand(0);
bool IsNegPow2 = (-Divisor).isPowerOf2();
unsigned Lg2 = (IsNegPow2 ? -Divisor : Divisor).countTrailingZeros();
SDValue ShiftAmt = DAG.getConstant(Lg2, DL, VT);
SDValue Op = DAG.getNode(PPCISD::SRA_ADDZE, DL, VT, N0, ShiftAmt);
Created.push_back(Op.getNode());
if (IsNegPow2) {
Op = DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT), Op);
Created.push_back(Op.getNode());
}
return Op;
}
//===----------------------------------------------------------------------===//
// Inline Assembly Support
//===----------------------------------------------------------------------===//
void PPCTargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
KnownBits &Known,
const APInt &DemandedElts,
const SelectionDAG &DAG,
unsigned Depth) const {
Known.resetAll();
switch (Op.getOpcode()) {
default: break;
case PPCISD::LBRX: {
// lhbrx is known to have the top bits cleared out.
if (cast<VTSDNode>(Op.getOperand(2))->getVT() == MVT::i16)
Known.Zero = 0xFFFF0000;
break;
}
case ISD::INTRINSIC_WO_CHAIN: {
switch (cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue()) {
default: break;
case Intrinsic::ppc_altivec_vcmpbfp_p:
case Intrinsic::ppc_altivec_vcmpeqfp_p:
case Intrinsic::ppc_altivec_vcmpequb_p:
case Intrinsic::ppc_altivec_vcmpequh_p:
case Intrinsic::ppc_altivec_vcmpequw_p:
case Intrinsic::ppc_altivec_vcmpequd_p:
case Intrinsic::ppc_altivec_vcmpequq_p:
case Intrinsic::ppc_altivec_vcmpgefp_p:
case Intrinsic::ppc_altivec_vcmpgtfp_p:
case Intrinsic::ppc_altivec_vcmpgtsb_p:
case Intrinsic::ppc_altivec_vcmpgtsh_p:
case Intrinsic::ppc_altivec_vcmpgtsw_p:
case Intrinsic::ppc_altivec_vcmpgtsd_p:
case Intrinsic::ppc_altivec_vcmpgtsq_p:
case Intrinsic::ppc_altivec_vcmpgtub_p:
case Intrinsic::ppc_altivec_vcmpgtuh_p:
case Intrinsic::ppc_altivec_vcmpgtuw_p:
case Intrinsic::ppc_altivec_vcmpgtud_p:
case Intrinsic::ppc_altivec_vcmpgtuq_p:
Known.Zero = ~1U; // All bits but the low one are known to be zero.
break;
}
}
}
}
Align PPCTargetLowering::getPrefLoopAlignment(MachineLoop *ML) const {
switch (Subtarget.getCPUDirective()) {
default: break;
case PPC::DIR_970:
case PPC::DIR_PWR4:
case PPC::DIR_PWR5:
case PPC::DIR_PWR5X:
case PPC::DIR_PWR6:
case PPC::DIR_PWR6X:
case PPC::DIR_PWR7:
case PPC::DIR_PWR8:
case PPC::DIR_PWR9:
case PPC::DIR_PWR10:
case PPC::DIR_PWR_FUTURE: {
if (!ML)
break;
if (!DisableInnermostLoopAlign32) {
// If the nested loop is an innermost loop, prefer to a 32-byte alignment,
// so that we can decrease cache misses and branch-prediction misses.
// Actual alignment of the loop will depend on the hotness check and other
// logic in alignBlocks.
if (ML->getLoopDepth() > 1 && ML->getSubLoops().empty())
return Align(32);
}
const PPCInstrInfo *TII = Subtarget.getInstrInfo();
// For small loops (between 5 and 8 instructions), align to a 32-byte
// boundary so that the entire loop fits in one instruction-cache line.
uint64_t LoopSize = 0;
for (auto I = ML->block_begin(), IE = ML->block_end(); I != IE; ++I)
for (auto J = (*I)->begin(), JE = (*I)->end(); J != JE; ++J) {
LoopSize += TII->getInstSizeInBytes(*J);
if (LoopSize > 32)
break;
}
if (LoopSize > 16 && LoopSize <= 32)
return Align(32);
break;
}
}
return TargetLowering::getPrefLoopAlignment(ML);
}
/// getConstraintType - Given a constraint, return the type of
/// constraint it is for this target.
PPCTargetLowering::ConstraintType
PPCTargetLowering::getConstraintType(StringRef Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
default: break;
case 'b':
case 'r':
case 'f':
case 'd':
case 'v':
case 'y':
return C_RegisterClass;
case 'Z':
// FIXME: While Z does indicate a memory constraint, it specifically
// indicates an r+r address (used in conjunction with the 'y' modifier
// in the replacement string). Currently, we're forcing the base
// register to be r0 in the asm printer (which is interpreted as zero)
// and forming the complete address in the second register. This is
// suboptimal.
return C_Memory;
}
} else if (Constraint == "wc") { // individual CR bits.
return C_RegisterClass;
} else if (Constraint == "wa" || Constraint == "wd" ||
Constraint == "wf" || Constraint == "ws" ||
Constraint == "wi" || Constraint == "ww") {
return C_RegisterClass; // VSX registers.
}
return TargetLowering::getConstraintType(Constraint);
}
/// Examine constraint type and operand type and determine a weight value.
/// This object must already have been set up with the operand type
/// and the current alternative constraint selected.
TargetLowering::ConstraintWeight
PPCTargetLowering::getSingleConstraintMatchWeight(
AsmOperandInfo &info, const char *constraint) const {
ConstraintWeight weight = CW_Invalid;
Value *CallOperandVal = info.CallOperandVal;
// If we don't have a value, we can't do a match,
// but allow it at the lowest weight.
if (!CallOperandVal)
return CW_Default;
Type *type = CallOperandVal->getType();
// Look at the constraint type.
if (StringRef(constraint) == "wc" && type->isIntegerTy(1))
return CW_Register; // an individual CR bit.
else if ((StringRef(constraint) == "wa" ||
StringRef(constraint) == "wd" ||
StringRef(constraint) == "wf") &&
type->isVectorTy())
return CW_Register;
else if (StringRef(constraint) == "wi" && type->isIntegerTy(64))
return CW_Register; // just hold 64-bit integers data.
else if (StringRef(constraint) == "ws" && type->isDoubleTy())
return CW_Register;
else if (StringRef(constraint) == "ww" && type->isFloatTy())
return CW_Register;
switch (*constraint) {
default:
weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
break;
case 'b':
if (type->isIntegerTy())
weight = CW_Register;
break;
case 'f':
if (type->isFloatTy())
weight = CW_Register;
break;
case 'd':
if (type->isDoubleTy())
weight = CW_Register;
break;
case 'v':
if (type->isVectorTy())
weight = CW_Register;
break;
case 'y':
weight = CW_Register;
break;
case 'Z':
weight = CW_Memory;
break;
}
return weight;
}
std::pair<unsigned, const TargetRegisterClass *>
PPCTargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI,
StringRef Constraint,
MVT VT) const {
if (Constraint.size() == 1) {
// GCC RS6000 Constraint Letters
switch (Constraint[0]) {
case 'b': // R1-R31
if (VT == MVT::i64 && Subtarget.isPPC64())
return std::make_pair(0U, &PPC::G8RC_NOX0RegClass);
return std::make_pair(0U, &PPC::GPRC_NOR0RegClass);
case 'r': // R0-R31
if (VT == MVT::i64 && Subtarget.isPPC64())
return std::make_pair(0U, &PPC::G8RCRegClass);
return std::make_pair(0U, &PPC::GPRCRegClass);
// 'd' and 'f' constraints are both defined to be "the floating point
// registers", where one is for 32-bit and the other for 64-bit. We don't
// really care overly much here so just give them all the same reg classes.
case 'd':
case 'f':
if (Subtarget.hasSPE()) {
if (VT == MVT::f32 || VT == MVT::i32)
return std::make_pair(0U, &PPC::GPRCRegClass);
if (VT == MVT::f64 || VT == MVT::i64)
return std::make_pair(0U, &PPC::SPERCRegClass);
} else {
if (VT == MVT::f32 || VT == MVT::i32)
return std::make_pair(0U, &PPC::F4RCRegClass);
if (VT == MVT::f64 || VT == MVT::i64)
return std::make_pair(0U, &PPC::F8RCRegClass);
}
break;
case 'v':
if (Subtarget.hasAltivec())
return std::make_pair(0U, &PPC::VRRCRegClass);
break;
case 'y': // crrc
return std::make_pair(0U, &PPC::CRRCRegClass);
}
} else if (Constraint == "wc" && Subtarget.useCRBits()) {
// An individual CR bit.
return std::make_pair(0U, &PPC::CRBITRCRegClass);
} else if ((Constraint == "wa" || Constraint == "wd" ||
Constraint == "wf" || Constraint == "wi") &&
Subtarget.hasVSX()) {
return std::make_pair(0U, &PPC::VSRCRegClass);
} else if ((Constraint == "ws" || Constraint == "ww") && Subtarget.hasVSX()) {
if (VT == MVT::f32 && Subtarget.hasP8Vector())
return std::make_pair(0U, &PPC::VSSRCRegClass);
else
return std::make_pair(0U, &PPC::VSFRCRegClass);
}
// If we name a VSX register, we can't defer to the base class because it
// will not recognize the correct register (their names will be VSL{0-31}
// and V{0-31} so they won't match). So we match them here.
if (Constraint.size() > 3 && Constraint[1] == 'v' && Constraint[2] == 's') {
int VSNum = atoi(Constraint.data() + 3);
assert(VSNum >= 0 && VSNum <= 63 &&
"Attempted to access a vsr out of range");
if (VSNum < 32)
return std::make_pair(PPC::VSL0 + VSNum, &PPC::VSRCRegClass);
return std::make_pair(PPC::V0 + VSNum - 32, &PPC::VSRCRegClass);
}
std::pair<unsigned, const TargetRegisterClass *> R =
TargetLowering::getRegForInlineAsmConstraint(TRI, Constraint, VT);
// r[0-9]+ are used, on PPC64, to refer to the corresponding 64-bit registers
// (which we call X[0-9]+). If a 64-bit value has been requested, and a
// 32-bit GPR has been selected, then 'upgrade' it to the 64-bit parent
// register.
// FIXME: If TargetLowering::getRegForInlineAsmConstraint could somehow use
// the AsmName field from *RegisterInfo.td, then this would not be necessary.
if (R.first && VT == MVT::i64 && Subtarget.isPPC64() &&
PPC::GPRCRegClass.contains(R.first))
return std::make_pair(TRI->getMatchingSuperReg(R.first,
PPC::sub_32, &PPC::G8RCRegClass),
&PPC::G8RCRegClass);
// GCC accepts 'cc' as an alias for 'cr0', and we need to do the same.
if (!R.second && StringRef("{cc}").equals_lower(Constraint)) {
R.first = PPC::CR0;
R.second = &PPC::CRRCRegClass;
}
return R;
}
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector. If it is invalid, don't add anything to Ops.
void PPCTargetLowering::LowerAsmOperandForConstraint(SDValue Op,
std::string &Constraint,
std::vector<SDValue>&Ops,
SelectionDAG &DAG) const {
SDValue Result;
// Only support length 1 constraints.
if (Constraint.length() > 1) return;
char Letter = Constraint[0];
switch (Letter) {
default: break;
case 'I':
case 'J':
case 'K':
case 'L':
case 'M':
case 'N':
case 'O':
case 'P': {
ConstantSDNode *CST = dyn_cast<ConstantSDNode>(Op);
if (!CST) return; // Must be an immediate to match.
SDLoc dl(Op);
int64_t Value = CST->getSExtValue();
EVT TCVT = MVT::i64; // All constants taken to be 64 bits so that negative
// numbers are printed as such.
switch (Letter) {
default: llvm_unreachable("Unknown constraint letter!");
case 'I': // "I" is a signed 16-bit constant.
if (isInt<16>(Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'J': // "J" is a constant with only the high-order 16 bits nonzero.
if (isShiftedUInt<16, 16>(Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'L': // "L" is a signed 16-bit constant shifted left 16 bits.
if (isShiftedInt<16, 16>(Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'K': // "K" is a constant with only the low-order 16 bits nonzero.
if (isUInt<16>(Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'M': // "M" is a constant that is greater than 31.
if (Value > 31)
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'N': // "N" is a positive constant that is an exact power of two.
if (Value > 0 && isPowerOf2_64(Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'O': // "O" is the constant zero.
if (Value == 0)
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
case 'P': // "P" is a constant whose negation is a signed 16-bit constant.
if (isInt<16>(-Value))
Result = DAG.getTargetConstant(Value, dl, TCVT);
break;
}
break;
}
}
if (Result.getNode()) {
Ops.push_back(Result);
return;
}
// Handle standard constraint letters.
TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}
// isLegalAddressingMode - Return true if the addressing mode represented
// by AM is legal for this target, for a load/store of the specified type.
bool PPCTargetLowering::isLegalAddressingMode(const DataLayout &DL,
const AddrMode &AM, Type *Ty,
unsigned AS,
Instruction *I) const {
// Vector type r+i form is supported since power9 as DQ form. We don't check
// the offset matching DQ form requirement(off % 16 == 0), because on PowerPC,
// imm form is preferred and the offset can be adjusted to use imm form later
// in pass PPCLoopInstrFormPrep. Also in LSR, for one LSRUse, it uses min and
// max offset to check legal addressing mode, we should be a little aggressive
// to contain other offsets for that LSRUse.
if (Ty->isVectorTy() && AM.BaseOffs != 0 && !Subtarget.hasP9Vector())
return false;
// PPC allows a sign-extended 16-bit immediate field.
if (AM.BaseOffs <= -(1LL << 16) || AM.BaseOffs >= (1LL << 16)-1)
return false;
// No global is ever allowed as a base.
if (AM.BaseGV)
return false;
// PPC only support r+r,
switch (AM.Scale) {
case 0: // "r+i" or just "i", depending on HasBaseReg.
break;
case 1:
if (AM.HasBaseReg && AM.BaseOffs) // "r+r+i" is not allowed.
return false;
// Otherwise we have r+r or r+i.
break;
case 2:
if (AM.HasBaseReg || AM.BaseOffs) // 2*r+r or 2*r+i is not allowed.
return false;
// Allow 2*r as r+r.
break;
default:
// No other scales are supported.
return false;
}
return true;
}
SDValue PPCTargetLowering::LowerRETURNADDR(SDValue Op,
SelectionDAG &DAG) const {
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MFI.setReturnAddressIsTaken(true);
if (verifyReturnAddressArgumentIsConstant(Op, DAG))
return SDValue();
SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
// Make sure the function does not optimize away the store of the RA to
// the stack.
PPCFunctionInfo *FuncInfo = MF.getInfo<PPCFunctionInfo>();
FuncInfo->setLRStoreRequired();
bool isPPC64 = Subtarget.isPPC64();
auto PtrVT = getPointerTy(MF.getDataLayout());
if (Depth > 0) {
SDValue FrameAddr = LowerFRAMEADDR(Op, DAG);
SDValue Offset =
DAG.getConstant(Subtarget.getFrameLowering()->getReturnSaveOffset(), dl,
isPPC64 ? MVT::i64 : MVT::i32);
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(),
DAG.getNode(ISD::ADD, dl, PtrVT, FrameAddr, Offset),
MachinePointerInfo());
}
// Just load the return address off the stack.
SDValue RetAddrFI = getReturnAddrFrameIndex(DAG);
return DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), RetAddrFI,
MachinePointerInfo());
}
SDValue PPCTargetLowering::LowerFRAMEADDR(SDValue Op,
SelectionDAG &DAG) const {
SDLoc dl(Op);
unsigned Depth = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
MachineFunction &MF = DAG.getMachineFunction();
MachineFrameInfo &MFI = MF.getFrameInfo();
MFI.setFrameAddressIsTaken(true);
EVT PtrVT = getPointerTy(MF.getDataLayout());
bool isPPC64 = PtrVT == MVT::i64;
// Naked functions never have a frame pointer, and so we use r1. For all
// other functions, this decision must be delayed until during PEI.
unsigned FrameReg;
if (MF.getFunction().hasFnAttribute(Attribute::Naked))
FrameReg = isPPC64 ? PPC::X1 : PPC::R1;
else
FrameReg = isPPC64 ? PPC::FP8 : PPC::FP;
SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), dl, FrameReg,
PtrVT);
while (Depth--)
FrameAddr = DAG.getLoad(Op.getValueType(), dl, DAG.getEntryNode(),
FrameAddr, MachinePointerInfo());
return FrameAddr;
}
// FIXME? Maybe this could be a TableGen attribute on some registers and
// this table could be generated automatically from RegInfo.
Register PPCTargetLowering::getRegisterByName(const char* RegName, LLT VT,
const MachineFunction &MF) const {
bool isPPC64 = Subtarget.isPPC64();
bool is64Bit = isPPC64 && VT == LLT::scalar(64);
if (!is64Bit && VT != LLT::scalar(32))
report_fatal_error("Invalid register global variable type");
Register Reg = StringSwitch<Register>(RegName)
.Case("r1", is64Bit ? PPC::X1 : PPC::R1)
.Case("r2", isPPC64 ? Register() : PPC::R2)
.Case("r13", (is64Bit ? PPC::X13 : PPC::R13))
.Default(Register());
if (Reg)
return Reg;
report_fatal_error("Invalid register name global variable");
}
bool PPCTargetLowering::isAccessedAsGotIndirect(SDValue GA) const {
// 32-bit SVR4 ABI access everything as got-indirect.
if (Subtarget.is32BitELFABI())
return true;
// AIX accesses everything indirectly through the TOC, which is similar to
// the GOT.
if (Subtarget.isAIXABI())
return true;
CodeModel::Model CModel = getTargetMachine().getCodeModel();
// If it is small or large code model, module locals are accessed
// indirectly by loading their address from .toc/.got.
if (CModel == CodeModel::Small || CModel == CodeModel::Large)
return true;
// JumpTable and BlockAddress are accessed as got-indirect.
if (isa<JumpTableSDNode>(GA) || isa<BlockAddressSDNode>(GA))
return true;
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(GA))
return Subtarget.isGVIndirectSymbol(G->getGlobal());
return false;
}
bool
PPCTargetLowering::isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const {
// The PowerPC target isn't yet aware of offsets.
return false;
}
bool PPCTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info,
const CallInst &I,
MachineFunction &MF,
unsigned Intrinsic) const {
switch (Intrinsic) {
case Intrinsic::ppc_altivec_lvx:
case Intrinsic::ppc_altivec_lvxl:
case Intrinsic::ppc_altivec_lvebx:
case Intrinsic::ppc_altivec_lvehx:
case Intrinsic::ppc_altivec_lvewx:
case Intrinsic::ppc_vsx_lxvd2x:
case Intrinsic::ppc_vsx_lxvw4x:
case Intrinsic::ppc_vsx_lxvd2x_be:
case Intrinsic::ppc_vsx_lxvw4x_be:
case Intrinsic::ppc_vsx_lxvl:
case Intrinsic::ppc_vsx_lxvll: {
EVT VT;
switch (Intrinsic) {
case Intrinsic::ppc_altivec_lvebx:
VT = MVT::i8;
break;
case Intrinsic::ppc_altivec_lvehx:
VT = MVT::i16;
break;
case Intrinsic::ppc_altivec_lvewx:
VT = MVT::i32;
break;
case Intrinsic::ppc_vsx_lxvd2x:
case Intrinsic::ppc_vsx_lxvd2x_be:
VT = MVT::v2f64;
break;
default:
VT = MVT::v4i32;
break;
}
Info.opc = ISD::INTRINSIC_W_CHAIN;
Info.memVT = VT;
Info.ptrVal = I.getArgOperand(0);
Info.offset = -VT.getStoreSize()+1;
Info.size = 2*VT.getStoreSize()-1;
Info.align = Align(1);
Info.flags = MachineMemOperand::MOLoad;
return true;
}
case Intrinsic::ppc_altivec_stvx:
case Intrinsic::ppc_altivec_stvxl:
case Intrinsic::ppc_altivec_stvebx:
case Intrinsic::ppc_altivec_stvehx:
case Intrinsic::ppc_altivec_stvewx:
case Intrinsic::ppc_vsx_stxvd2x:
case Intrinsic::ppc_vsx_stxvw4x:
case Intrinsic::ppc_vsx_stxvd2x_be:
case Intrinsic::ppc_vsx_stxvw4x_be:
case Intrinsic::ppc_vsx_stxvl:
case Intrinsic::ppc_vsx_stxvll: {
EVT VT;
switch (Intrinsic) {
case Intrinsic::ppc_altivec_stvebx:
VT = MVT::i8;
break;
case Intrinsic::ppc_altivec_stvehx:
VT = MVT::i16;
break;
case Intrinsic::ppc_altivec_stvewx:
VT = MVT::i32;
break;
case Intrinsic::ppc_vsx_stxvd2x:
case Intrinsic::ppc_vsx_stxvd2x_be:
VT = MVT::v2f64;
break;
default:
VT = MVT::v4i32;
break;
}
Info.opc = ISD::INTRINSIC_VOID;
Info.memVT = VT;
Info.ptrVal = I.getArgOperand(1);
Info.offset = -VT.getStoreSize()+1;
Info.size = 2*VT.getStoreSize()-1;
Info.align = Align(1);
Info.flags = MachineMemOperand::MOStore;
return true;
}
default:
break;
}
return false;
}
/// It returns EVT::Other if the type should be determined using generic
/// target-independent logic.
EVT PPCTargetLowering::getOptimalMemOpType(
const MemOp &Op, const AttributeList &FuncAttributes) const {
if (getTargetMachine().getOptLevel() != CodeGenOpt::None) {
// We should use Altivec/VSX loads and stores when available. For unaligned
// addresses, unaligned VSX loads are only fast starting with the P8.
if (Subtarget.hasAltivec() && Op.size() >= 16 &&
(Op.isAligned(Align(16)) ||
((Op.isMemset() && Subtarget.hasVSX()) || Subtarget.hasP8Vector())))
return MVT::v4i32;
}
if (Subtarget.isPPC64()) {
return MVT::i64;
}
return MVT::i32;
}
/// Returns true if it is beneficial to convert a load of a constant
/// to just the constant itself.
bool PPCTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
Type *Ty) const {
assert(Ty->isIntegerTy());
unsigned BitSize = Ty->getPrimitiveSizeInBits();
return !(BitSize == 0 || BitSize > 64);
}
bool PPCTargetLowering::isTruncateFree(Type *Ty1, Type *Ty2) const {
if (!Ty1->isIntegerTy() || !Ty2->isIntegerTy())
return false;
unsigned NumBits1 = Ty1->getPrimitiveSizeInBits();
unsigned NumBits2 = Ty2->getPrimitiveSizeInBits();
return NumBits1 == 64 && NumBits2 == 32;
}
bool PPCTargetLowering::isTruncateFree(EVT VT1, EVT VT2) const {
if (!VT1.isInteger() || !VT2.isInteger())
return false;
unsigned NumBits1 = VT1.getSizeInBits();
unsigned NumBits2 = VT2.getSizeInBits();
return NumBits1 == 64 && NumBits2 == 32;
}
bool PPCTargetLowering::isZExtFree(SDValue Val, EVT VT2) const {
// Generally speaking, zexts are not free, but they are free when they can be
// folded with other operations.
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Val)) {
EVT MemVT = LD->getMemoryVT();
if ((MemVT == MVT::i1 || MemVT == MVT::i8 || MemVT == MVT::i16 ||
(Subtarget.isPPC64() && MemVT == MVT::i32)) &&
(LD->getExtensionType() == ISD::NON_EXTLOAD ||
LD->getExtensionType() == ISD::ZEXTLOAD))
return true;
}
// FIXME: Add other cases...
// - 32-bit shifts with a zext to i64
// - zext after ctlz, bswap, etc.
// - zext after and by a constant mask
return TargetLowering::isZExtFree(Val, VT2);
}
bool PPCTargetLowering::isFPExtFree(EVT DestVT, EVT SrcVT) const {
assert(DestVT.isFloatingPoint() && SrcVT.isFloatingPoint() &&
"invalid fpext types");
// Extending to float128 is not free.
if (DestVT == MVT::f128)
return false;
return true;
}
bool PPCTargetLowering::isLegalICmpImmediate(int64_t Imm) const {
return isInt<16>(Imm) || isUInt<16>(Imm);
}
bool PPCTargetLowering::isLegalAddImmediate(int64_t Imm) const {
return isInt<16>(Imm) || isUInt<16>(Imm);
}
bool PPCTargetLowering::allowsMisalignedMemoryAccesses(EVT VT,
unsigned,
unsigned,
MachineMemOperand::Flags,
bool *Fast) const {
if (DisablePPCUnaligned)
return false;
// PowerPC supports unaligned memory access for simple non-vector types.
// Although accessing unaligned addresses is not as efficient as accessing
// aligned addresses, it is generally more efficient than manual expansion,
// and generally only traps for software emulation when crossing page
// boundaries.
if (!VT.isSimple())
return false;
if (VT.isFloatingPoint() && !VT.isVector() &&
!Subtarget.allowsUnalignedFPAccess())
return false;
if (VT.getSimpleVT().isVector()) {
if (Subtarget.hasVSX()) {
if (VT != MVT::v2f64 && VT != MVT::v2i64 &&
VT != MVT::v4f32 && VT != MVT::v4i32)
return false;
} else {
return false;
}
}
if (VT == MVT::ppcf128)
return false;
if (Fast)
*Fast = true;
return true;
}
bool PPCTargetLowering::decomposeMulByConstant(LLVMContext &Context, EVT VT,
SDValue C) const {
// Check integral scalar types.
if (!VT.isScalarInteger())
return false;
if (auto *ConstNode = dyn_cast<ConstantSDNode>(C.getNode())) {
if (!ConstNode->getAPIntValue().isSignedIntN(64))
return false;
// This transformation will generate >= 2 operations. But the following
// cases will generate <= 2 instructions during ISEL. So exclude them.
// 1. If the constant multiplier fits 16 bits, it can be handled by one
// HW instruction, ie. MULLI
// 2. If the multiplier after shifted fits 16 bits, an extra shift
// instruction is needed than case 1, ie. MULLI and RLDICR
int64_t Imm = ConstNode->getSExtValue();
unsigned Shift = countTrailingZeros<uint64_t>(Imm);
Imm >>= Shift;
if (isInt<16>(Imm))
return false;
uint64_t UImm = static_cast<uint64_t>(Imm);
if (isPowerOf2_64(UImm + 1) || isPowerOf2_64(UImm - 1) ||
isPowerOf2_64(1 - UImm) || isPowerOf2_64(-1 - UImm))
return true;
}
return false;
}
bool PPCTargetLowering::isFMAFasterThanFMulAndFAdd(const MachineFunction &MF,
EVT VT) const {
return isFMAFasterThanFMulAndFAdd(
MF.getFunction(), VT.getTypeForEVT(MF.getFunction().getContext()));
}
bool PPCTargetLowering::isFMAFasterThanFMulAndFAdd(const Function &F,
Type *Ty) const {
switch (Ty->getScalarType()->getTypeID()) {
case Type::FloatTyID:
case Type::DoubleTyID:
return true;
case Type::FP128TyID:
return Subtarget.hasP9Vector();
default:
return false;
}
}
// FIXME: add more patterns which are not profitable to hoist.
bool PPCTargetLowering::isProfitableToHoist(Instruction *I) const {
if (!I->hasOneUse())
return true;
Instruction *User = I->user_back();
assert(User && "A single use instruction with no uses.");
switch (I->getOpcode()) {
case Instruction::FMul: {
// Don't break FMA, PowerPC prefers FMA.
if (User->getOpcode() != Instruction::FSub &&
User->getOpcode() != Instruction::FAdd)
return true;
const TargetOptions &Options = getTargetMachine().Options;
const Function *F = I->getFunction();
const DataLayout &DL = F->getParent()->getDataLayout();
Type *Ty = User->getOperand(0)->getType();
return !(
isFMAFasterThanFMulAndFAdd(*F, Ty) &&
isOperationLegalOrCustom(ISD::FMA, getValueType(DL, Ty)) &&
(Options.AllowFPOpFusion == FPOpFusion::Fast || Options.UnsafeFPMath));
}
case Instruction::Load: {
// Don't break "store (load float*)" pattern, this pattern will be combined
// to "store (load int32)" in later InstCombine pass. See function
// combineLoadToOperationType. On PowerPC, loading a float point takes more
// cycles than loading a 32 bit integer.
LoadInst *LI = cast<LoadInst>(I);
// For the loads that combineLoadToOperationType does nothing, like
// ordered load, it should be profitable to hoist them.
// For swifterror load, it can only be used for pointer to pointer type, so
// later type check should get rid of this case.
if (!LI->isUnordered())
return true;
if (User->getOpcode() != Instruction::Store)
return true;
if (I->getType()->getTypeID() != Type::FloatTyID)
return true;
return false;
}
default:
return true;
}
return true;
}
const MCPhysReg *
PPCTargetLowering::getScratchRegisters(CallingConv::ID) const {
// LR is a callee-save register, but we must treat it as clobbered by any call
// site. Hence we include LR in the scratch registers, which are in turn added
// as implicit-defs for stackmaps and patchpoints. The same reasoning applies
// to CTR, which is used by any indirect call.
static const MCPhysReg ScratchRegs[] = {
PPC::X12, PPC::LR8, PPC::CTR8, 0
};
return ScratchRegs;
}
Register PPCTargetLowering::getExceptionPointerRegister(
const Constant *PersonalityFn) const {
return Subtarget.isPPC64() ? PPC::X3 : PPC::R3;
}
Register PPCTargetLowering::getExceptionSelectorRegister(
const Constant *PersonalityFn) const {
return Subtarget.isPPC64() ? PPC::X4 : PPC::R4;
}
bool
PPCTargetLowering::shouldExpandBuildVectorWithShuffles(
EVT VT , unsigned DefinedValues) const {
if (VT == MVT::v2i64)
return Subtarget.hasDirectMove(); // Don't need stack ops with direct moves
if (Subtarget.hasVSX())
return true;
return TargetLowering::shouldExpandBuildVectorWithShuffles(VT, DefinedValues);
}
Sched::Preference PPCTargetLowering::getSchedulingPreference(SDNode *N) const {
if (DisableILPPref || Subtarget.enableMachineScheduler())
return TargetLowering::getSchedulingPreference(N);
return Sched::ILP;
}
// Create a fast isel object.
FastISel *
PPCTargetLowering::createFastISel(FunctionLoweringInfo &FuncInfo,
const TargetLibraryInfo *LibInfo) const {
return PPC::createFastISel(FuncInfo, LibInfo);
}
// 'Inverted' means the FMA opcode after negating one multiplicand.
// For example, (fma -a b c) = (fnmsub a b c)
static unsigned invertFMAOpcode(unsigned Opc) {
switch (Opc) {
default:
llvm_unreachable("Invalid FMA opcode for PowerPC!");
case ISD::FMA:
return PPCISD::FNMSUB;
case PPCISD::FNMSUB:
return ISD::FMA;
}
}
SDValue PPCTargetLowering::getNegatedExpression(SDValue Op, SelectionDAG &DAG,
bool LegalOps, bool OptForSize,
NegatibleCost &Cost,
unsigned Depth) const {
if (Depth > SelectionDAG::MaxRecursionDepth)
return SDValue();
unsigned Opc = Op.getOpcode();
EVT VT = Op.getValueType();
SDNodeFlags Flags = Op.getNode()->getFlags();
switch (Opc) {
case PPCISD::FNMSUB:
if (!Op.hasOneUse() || !isTypeLegal(VT))
break;
const TargetOptions &Options = getTargetMachine().Options;
SDValue N0 = Op.getOperand(0);
SDValue N1 = Op.getOperand(1);
SDValue N2 = Op.getOperand(2);
SDLoc Loc(Op);
NegatibleCost N2Cost = NegatibleCost::Expensive;
SDValue NegN2 =
getNegatedExpression(N2, DAG, LegalOps, OptForSize, N2Cost, Depth + 1);
if (!NegN2)
return SDValue();
// (fneg (fnmsub a b c)) => (fnmsub (fneg a) b (fneg c))
// (fneg (fnmsub a b c)) => (fnmsub a (fneg b) (fneg c))
// These transformations may change sign of zeroes. For example,
// -(-ab-(-c))=-0 while -(-(ab-c))=+0 when a=b=c=1.
if (Flags.hasNoSignedZeros() || Options.NoSignedZerosFPMath) {
// Try and choose the cheaper one to negate.
NegatibleCost N0Cost = NegatibleCost::Expensive;
SDValue NegN0 = getNegatedExpression(N0, DAG, LegalOps, OptForSize,
N0Cost, Depth + 1);
NegatibleCost N1Cost = NegatibleCost::Expensive;
SDValue NegN1 = getNegatedExpression(N1, DAG, LegalOps, OptForSize,
N1Cost, Depth + 1);
if (NegN0 && N0Cost <= N1Cost) {
Cost = std::min(N0Cost, N2Cost);
return DAG.getNode(Opc, Loc, VT, NegN0, N1, NegN2, Flags);
} else if (NegN1) {
Cost = std::min(N1Cost, N2Cost);
return DAG.getNode(Opc, Loc, VT, N0, NegN1, NegN2, Flags);
}
}
// (fneg (fnmsub a b c)) => (fma a b (fneg c))
if (isOperationLegal(ISD::FMA, VT)) {
Cost = N2Cost;
return DAG.getNode(ISD::FMA, Loc, VT, N0, N1, NegN2, Flags);
}
break;
}
return TargetLowering::getNegatedExpression(Op, DAG, LegalOps, OptForSize,
Cost, Depth);
}
// Override to enable LOAD_STACK_GUARD lowering on Linux.
bool PPCTargetLowering::useLoadStackGuardNode() const {
if (!Subtarget.isTargetLinux())
return TargetLowering::useLoadStackGuardNode();
return true;
}
// Override to disable global variable loading on Linux.
void PPCTargetLowering::insertSSPDeclarations(Module &M) const {
if (!Subtarget.isTargetLinux())
return TargetLowering::insertSSPDeclarations(M);
}
bool PPCTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT,
bool ForCodeSize) const {
if (!VT.isSimple() || !Subtarget.hasVSX())
return false;
switch(VT.getSimpleVT().SimpleTy) {
default:
// For FP types that are currently not supported by PPC backend, return
// false. Examples: f16, f80.
return false;
case MVT::f32:
case MVT::f64:
if (Subtarget.hasPrefixInstrs()) {
// With prefixed instructions, we can materialize anything that can be
// represented with a 32-bit immediate, not just positive zero.
APFloat APFloatOfImm = Imm;
return convertToNonDenormSingle(APFloatOfImm);
}
LLVM_FALLTHROUGH;
case MVT::ppcf128:
return Imm.isPosZero();
}
}
// For vector shift operation op, fold
// (op x, (and y, ((1 << numbits(x)) - 1))) -> (target op x, y)
static SDValue stripModuloOnShift(const TargetLowering &TLI, SDNode *N,
SelectionDAG &DAG) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
EVT VT = N0.getValueType();
unsigned OpSizeInBits = VT.getScalarSizeInBits();
unsigned Opcode = N->getOpcode();
unsigned TargetOpcode;
switch (Opcode) {
default:
llvm_unreachable("Unexpected shift operation");
case ISD::SHL:
TargetOpcode = PPCISD::SHL;
break;
case ISD::SRL:
TargetOpcode = PPCISD::SRL;
break;
case ISD::SRA:
TargetOpcode = PPCISD::SRA;
break;
}
if (VT.isVector() && TLI.isOperationLegal(Opcode, VT) &&
N1->getOpcode() == ISD::AND)
if (ConstantSDNode *Mask = isConstOrConstSplat(N1->getOperand(1)))
if (Mask->getZExtValue() == OpSizeInBits - 1)
return DAG.getNode(TargetOpcode, SDLoc(N), VT, N0, N1->getOperand(0));
return SDValue();
}
SDValue PPCTargetLowering::combineSHL(SDNode *N, DAGCombinerInfo &DCI) const {
if (auto Value = stripModuloOnShift(*this, N, DCI.DAG))
return Value;
SDValue N0 = N->getOperand(0);
ConstantSDNode *CN1 = dyn_cast<ConstantSDNode>(N->getOperand(1));
if (!Subtarget.isISA3_0() || !Subtarget.isPPC64() ||
N0.getOpcode() != ISD::SIGN_EXTEND ||
N0.getOperand(0).getValueType() != MVT::i32 || CN1 == nullptr ||
N->getValueType(0) != MVT::i64)
return SDValue();
// We can't save an operation here if the value is already extended, and
// the existing shift is easier to combine.
SDValue ExtsSrc = N0.getOperand(0);
if (ExtsSrc.getOpcode() == ISD::TRUNCATE &&
ExtsSrc.getOperand(0).getOpcode() == ISD::AssertSext)
return SDValue();
SDLoc DL(N0);
SDValue ShiftBy = SDValue(CN1, 0);
// We want the shift amount to be i32 on the extswli, but the shift could
// have an i64.
if (ShiftBy.getValueType() == MVT::i64)
ShiftBy = DCI.DAG.getConstant(CN1->getZExtValue(), DL, MVT::i32);
return DCI.DAG.getNode(PPCISD::EXTSWSLI, DL, MVT::i64, N0->getOperand(0),
ShiftBy);
}
SDValue PPCTargetLowering::combineSRA(SDNode *N, DAGCombinerInfo &DCI) const {
if (auto Value = stripModuloOnShift(*this, N, DCI.DAG))
return Value;
return SDValue();
}
SDValue PPCTargetLowering::combineSRL(SDNode *N, DAGCombinerInfo &DCI) const {
if (auto Value = stripModuloOnShift(*this, N, DCI.DAG))
return Value;
return SDValue();
}
// Transform (add X, (zext(setne Z, C))) -> (addze X, (addic (addi Z, -C), -1))
// Transform (add X, (zext(sete Z, C))) -> (addze X, (subfic (addi Z, -C), 0))
// When C is zero, the equation (addi Z, -C) can be simplified to Z
// Requirement: -C in [-32768, 32767], X and Z are MVT::i64 types
static SDValue combineADDToADDZE(SDNode *N, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
if (!Subtarget.isPPC64())
return SDValue();
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
auto isZextOfCompareWithConstant = [](SDValue Op) {
if (Op.getOpcode() != ISD::ZERO_EXTEND || !Op.hasOneUse() ||
Op.getValueType() != MVT::i64)
return false;
SDValue Cmp = Op.getOperand(0);
if (Cmp.getOpcode() != ISD::SETCC || !Cmp.hasOneUse() ||
Cmp.getOperand(0).getValueType() != MVT::i64)
return false;
if (auto *Constant = dyn_cast<ConstantSDNode>(Cmp.getOperand(1))) {
int64_t NegConstant = 0 - Constant->getSExtValue();
// Due to the limitations of the addi instruction,
// -C is required to be [-32768, 32767].
return isInt<16>(NegConstant);
}
return false;
};
bool LHSHasPattern = isZextOfCompareWithConstant(LHS);
bool RHSHasPattern = isZextOfCompareWithConstant(RHS);
// If there is a pattern, canonicalize a zext operand to the RHS.
if (LHSHasPattern && !RHSHasPattern)
std::swap(LHS, RHS);
else if (!LHSHasPattern && !RHSHasPattern)
return SDValue();
SDLoc DL(N);
SDVTList VTs = DAG.getVTList(MVT::i64, MVT::Glue);
SDValue Cmp = RHS.getOperand(0);
SDValue Z = Cmp.getOperand(0);
auto *Constant = dyn_cast<ConstantSDNode>(Cmp.getOperand(1));
assert(Constant && "Constant Should not be a null pointer.");
int64_t NegConstant = 0 - Constant->getSExtValue();
switch(cast<CondCodeSDNode>(Cmp.getOperand(2))->get()) {
default: break;
case ISD::SETNE: {
// when C == 0
// --> addze X, (addic Z, -1).carry
// /
// add X, (zext(setne Z, C))--
// \ when -32768 <= -C <= 32767 && C != 0
// --> addze X, (addic (addi Z, -C), -1).carry
SDValue Add = DAG.getNode(ISD::ADD, DL, MVT::i64, Z,
DAG.getConstant(NegConstant, DL, MVT::i64));
SDValue AddOrZ = NegConstant != 0 ? Add : Z;
SDValue Addc = DAG.getNode(ISD::ADDC, DL, DAG.getVTList(MVT::i64, MVT::Glue),
AddOrZ, DAG.getConstant(-1ULL, DL, MVT::i64));
return DAG.getNode(ISD::ADDE, DL, VTs, LHS, DAG.getConstant(0, DL, MVT::i64),
SDValue(Addc.getNode(), 1));
}
case ISD::SETEQ: {
// when C == 0
// --> addze X, (subfic Z, 0).carry
// /
// add X, (zext(sete Z, C))--
// \ when -32768 <= -C <= 32767 && C != 0
// --> addze X, (subfic (addi Z, -C), 0).carry
SDValue Add = DAG.getNode(ISD::ADD, DL, MVT::i64, Z,
DAG.getConstant(NegConstant, DL, MVT::i64));
SDValue AddOrZ = NegConstant != 0 ? Add : Z;
SDValue Subc = DAG.getNode(ISD::SUBC, DL, DAG.getVTList(MVT::i64, MVT::Glue),
DAG.getConstant(0, DL, MVT::i64), AddOrZ);
return DAG.getNode(ISD::ADDE, DL, VTs, LHS, DAG.getConstant(0, DL, MVT::i64),
SDValue(Subc.getNode(), 1));
}
}
return SDValue();
}
// Transform
// (add C1, (MAT_PCREL_ADDR GlobalAddr+C2)) to
// (MAT_PCREL_ADDR GlobalAddr+(C1+C2))
// In this case both C1 and C2 must be known constants.
// C1+C2 must fit into a 34 bit signed integer.
static SDValue combineADDToMAT_PCREL_ADDR(SDNode *N, SelectionDAG &DAG,
const PPCSubtarget &Subtarget) {
if (!Subtarget.isUsingPCRelativeCalls())
return SDValue();
// Check both Operand 0 and Operand 1 of the ADD node for the PCRel node.
// If we find that node try to cast the Global Address and the Constant.
SDValue LHS = N->getOperand(0);
SDValue RHS = N->getOperand(1);
if (LHS.getOpcode() != PPCISD::MAT_PCREL_ADDR)
std::swap(LHS, RHS);
if (LHS.getOpcode() != PPCISD::MAT_PCREL_ADDR)
return SDValue();
// Operand zero of PPCISD::MAT_PCREL_ADDR is the GA node.
GlobalAddressSDNode *GSDN = dyn_cast<GlobalAddressSDNode>(LHS.getOperand(0));
ConstantSDNode* ConstNode = dyn_cast<ConstantSDNode>(RHS);
// Check that both casts succeeded.
if (!GSDN || !ConstNode)
return SDValue();
int64_t NewOffset = GSDN->getOffset() + ConstNode->getSExtValue();
SDLoc DL(GSDN);
// The signed int offset needs to fit in 34 bits.
if (!isInt<34>(NewOffset))
return SDValue();
// The new global address is a copy of the old global address except
// that it has the updated Offset.
SDValue GA =
DAG.getTargetGlobalAddress(GSDN->getGlobal(), DL, GSDN->getValueType(0),
NewOffset, GSDN->getTargetFlags());
SDValue MatPCRel =
DAG.getNode(PPCISD::MAT_PCREL_ADDR, DL, GSDN->getValueType(0), GA);
return MatPCRel;
}
SDValue PPCTargetLowering::combineADD(SDNode *N, DAGCombinerInfo &DCI) const {
if (auto Value = combineADDToADDZE(N, DCI.DAG, Subtarget))
return Value;
if (auto Value = combineADDToMAT_PCREL_ADDR(N, DCI.DAG, Subtarget))
return Value;
return SDValue();
}
// Detect TRUNCATE operations on bitcasts of float128 values.
// What we are looking for here is the situtation where we extract a subset
// of bits from a 128 bit float.
// This can be of two forms:
// 1) BITCAST of f128 feeding TRUNCATE
// 2) BITCAST of f128 feeding SRL (a shift) feeding TRUNCATE
// The reason this is required is because we do not have a legal i128 type
// and so we want to prevent having to store the f128 and then reload part
// of it.
SDValue PPCTargetLowering::combineTRUNCATE(SDNode *N,
DAGCombinerInfo &DCI) const {
// If we are using CRBits then try that first.
if (Subtarget.useCRBits()) {
// Check if CRBits did anything and return that if it did.
if (SDValue CRTruncValue = DAGCombineTruncBoolExt(N, DCI))
return CRTruncValue;
}
SDLoc dl(N);
SDValue Op0 = N->getOperand(0);
// fold (truncate (abs (sub (zext a), (zext b)))) -> (vabsd a, b)
if (Subtarget.hasP9Altivec() && Op0.getOpcode() == ISD::ABS) {
EVT VT = N->getValueType(0);
if (VT != MVT::v4i32 && VT != MVT::v8i16 && VT != MVT::v16i8)
return SDValue();
SDValue Sub = Op0.getOperand(0);
if (Sub.getOpcode() == ISD::SUB) {
SDValue SubOp0 = Sub.getOperand(0);
SDValue SubOp1 = Sub.getOperand(1);
if ((SubOp0.getOpcode() == ISD::ZERO_EXTEND) &&
(SubOp1.getOpcode() == ISD::ZERO_EXTEND)) {
return DCI.DAG.getNode(PPCISD::VABSD, dl, VT, SubOp0.getOperand(0),
SubOp1.getOperand(0),
DCI.DAG.getTargetConstant(0, dl, MVT::i32));
}
}
}
// Looking for a truncate of i128 to i64.
if (Op0.getValueType() != MVT::i128 || N->getValueType(0) != MVT::i64)
return SDValue();
int EltToExtract = DCI.DAG.getDataLayout().isBigEndian() ? 1 : 0;
// SRL feeding TRUNCATE.
if (Op0.getOpcode() == ISD::SRL) {
ConstantSDNode *ConstNode = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
// The right shift has to be by 64 bits.
if (!ConstNode || ConstNode->getZExtValue() != 64)
return SDValue();
// Switch the element number to extract.
EltToExtract = EltToExtract ? 0 : 1;
// Update Op0 past the SRL.
Op0 = Op0.getOperand(0);
}
// BITCAST feeding a TRUNCATE possibly via SRL.
if (Op0.getOpcode() == ISD::BITCAST &&
Op0.getValueType() == MVT::i128 &&
Op0.getOperand(0).getValueType() == MVT::f128) {
SDValue Bitcast = DCI.DAG.getBitcast(MVT::v2i64, Op0.getOperand(0));
return DCI.DAG.getNode(
ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64, Bitcast,
DCI.DAG.getTargetConstant(EltToExtract, dl, MVT::i32));
}
return SDValue();
}
SDValue PPCTargetLowering::combineMUL(SDNode *N, DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
ConstantSDNode *ConstOpOrElement = isConstOrConstSplat(N->getOperand(1));
if (!ConstOpOrElement)
return SDValue();
// An imul is usually smaller than the alternative sequence for legal type.
if (DAG.getMachineFunction().getFunction().hasMinSize() &&
isOperationLegal(ISD::MUL, N->getValueType(0)))
return SDValue();
auto IsProfitable = [this](bool IsNeg, bool IsAddOne, EVT VT) -> bool {
switch (this->Subtarget.getCPUDirective()) {
default:
// TODO: enhance the condition for subtarget before pwr8
return false;
case PPC::DIR_PWR8:
// type mul add shl
// scalar 4 1 1
// vector 7 2 2
return true;
case PPC::DIR_PWR9:
case PPC::DIR_PWR10:
case PPC::DIR_PWR_FUTURE:
// type mul add shl
// scalar 5 2 2
// vector 7 2 2
// The cycle RATIO of related operations are showed as a table above.
// Because mul is 5(scalar)/7(vector), add/sub/shl are all 2 for both
// scalar and vector type. For 2 instrs patterns, add/sub + shl
// are 4, it is always profitable; but for 3 instrs patterns
// (mul x, -(2^N + 1)) => -(add (shl x, N), x), sub + add + shl are 6.
// So we should only do it for vector type.
return IsAddOne && IsNeg ? VT.isVector() : true;
}
};
EVT VT = N->getValueType(0);
SDLoc DL(N);
const APInt &MulAmt = ConstOpOrElement->getAPIntValue();
bool IsNeg = MulAmt.isNegative();
APInt MulAmtAbs = MulAmt.abs();
if ((MulAmtAbs - 1).isPowerOf2()) {
// (mul x, 2^N + 1) => (add (shl x, N), x)
// (mul x, -(2^N + 1)) => -(add (shl x, N), x)
if (!IsProfitable(IsNeg, true, VT))
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 =
DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
DAG.getConstant((MulAmtAbs - 1).logBase2(), DL, VT));
SDValue Res = DAG.getNode(ISD::ADD, DL, VT, Op0, Op1);
if (!IsNeg)
return Res;
return DAG.getNode(ISD::SUB, DL, VT, DAG.getConstant(0, DL, VT), Res);
} else if ((MulAmtAbs + 1).isPowerOf2()) {
// (mul x, 2^N - 1) => (sub (shl x, N), x)
// (mul x, -(2^N - 1)) => (sub x, (shl x, N))
if (!IsProfitable(IsNeg, false, VT))
return SDValue();
SDValue Op0 = N->getOperand(0);
SDValue Op1 =
DAG.getNode(ISD::SHL, DL, VT, N->getOperand(0),
DAG.getConstant((MulAmtAbs + 1).logBase2(), DL, VT));
if (!IsNeg)
return DAG.getNode(ISD::SUB, DL, VT, Op1, Op0);
else
return DAG.getNode(ISD::SUB, DL, VT, Op0, Op1);
} else {
return SDValue();
}
}
// Combine fma-like op (like fnmsub) with fnegs to appropriate op. Do this
// in combiner since we need to check SD flags and other subtarget features.
SDValue PPCTargetLowering::combineFMALike(SDNode *N,
DAGCombinerInfo &DCI) const {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDValue N2 = N->getOperand(2);
SDNodeFlags Flags = N->getFlags();
EVT VT = N->getValueType(0);
SelectionDAG &DAG = DCI.DAG;
const TargetOptions &Options = getTargetMachine().Options;
unsigned Opc = N->getOpcode();
bool CodeSize = DAG.getMachineFunction().getFunction().hasOptSize();
bool LegalOps = !DCI.isBeforeLegalizeOps();
SDLoc Loc(N);
if (!isOperationLegal(ISD::FMA, VT))
return SDValue();
// Allowing transformation to FNMSUB may change sign of zeroes when ab-c=0
// since (fnmsub a b c)=-0 while c-ab=+0.
if (!Flags.hasNoSignedZeros() && !Options.NoSignedZerosFPMath)
return SDValue();
// (fma (fneg a) b c) => (fnmsub a b c)
// (fnmsub (fneg a) b c) => (fma a b c)
if (SDValue NegN0 = getCheaperNegatedExpression(N0, DAG, LegalOps, CodeSize))
return DAG.getNode(invertFMAOpcode(Opc), Loc, VT, NegN0, N1, N2, Flags);
// (fma a (fneg b) c) => (fnmsub a b c)
// (fnmsub a (fneg b) c) => (fma a b c)
if (SDValue NegN1 = getCheaperNegatedExpression(N1, DAG, LegalOps, CodeSize))
return DAG.getNode(invertFMAOpcode(Opc), Loc, VT, N0, NegN1, N2, Flags);
return SDValue();
}
bool PPCTargetLowering::mayBeEmittedAsTailCall(const CallInst *CI) const {
// Only duplicate to increase tail-calls for the 64bit SysV ABIs.
if (!Subtarget.is64BitELFABI())
return false;
// If not a tail call then no need to proceed.
if (!CI->isTailCall())
return false;
// If sibling calls have been disabled and tail-calls aren't guaranteed
// there is no reason to duplicate.
auto &TM = getTargetMachine();
if (!TM.Options.GuaranteedTailCallOpt && DisableSCO)
return false;
// Can't tail call a function called indirectly, or if it has variadic args.
const Function *Callee = CI->getCalledFunction();
if (!Callee || Callee->isVarArg())
return false;
// Make sure the callee and caller calling conventions are eligible for tco.
const Function *Caller = CI->getParent()->getParent();
if (!areCallingConvEligibleForTCO_64SVR4(Caller->getCallingConv(),
CI->getCallingConv()))
return false;
// If the function is local then we have a good chance at tail-calling it
return getTargetMachine().shouldAssumeDSOLocal(*Caller->getParent(), Callee);
}
bool PPCTargetLowering::hasBitPreservingFPLogic(EVT VT) const {
if (!Subtarget.hasVSX())
return false;
if (Subtarget.hasP9Vector() && VT == MVT::f128)
return true;
return VT == MVT::f32 || VT == MVT::f64 ||
VT == MVT::v4f32 || VT == MVT::v2f64;
}
bool PPCTargetLowering::
isMaskAndCmp0FoldingBeneficial(const Instruction &AndI) const {
const Value *Mask = AndI.getOperand(1);
// If the mask is suitable for andi. or andis. we should sink the and.
if (const ConstantInt *CI = dyn_cast<ConstantInt>(Mask)) {
// Can't handle constants wider than 64-bits.
if (CI->getBitWidth() > 64)
return false;
int64_t ConstVal = CI->getZExtValue();
return isUInt<16>(ConstVal) ||
(isUInt<16>(ConstVal >> 16) && !(ConstVal & 0xFFFF));
}
// For non-constant masks, we can always use the record-form and.
return true;
}
// Transform (abs (sub (zext a), (zext b))) to (vabsd a b 0)
// Transform (abs (sub (zext a), (zext_invec b))) to (vabsd a b 0)
// Transform (abs (sub (zext_invec a), (zext_invec b))) to (vabsd a b 0)
// Transform (abs (sub (zext_invec a), (zext b))) to (vabsd a b 0)
// Transform (abs (sub a, b) to (vabsd a b 1)) if a & b of type v4i32
SDValue PPCTargetLowering::combineABS(SDNode *N, DAGCombinerInfo &DCI) const {
assert((N->getOpcode() == ISD::ABS) && "Need ABS node here");
assert(Subtarget.hasP9Altivec() &&
"Only combine this when P9 altivec supported!");
EVT VT = N->getValueType(0);
if (VT != MVT::v4i32 && VT != MVT::v8i16 && VT != MVT::v16i8)
return SDValue();
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
if (N->getOperand(0).getOpcode() == ISD::SUB) {
// Even for signed integers, if it's known to be positive (as signed
// integer) due to zero-extended inputs.
unsigned SubOpcd0 = N->getOperand(0)->getOperand(0).getOpcode();
unsigned SubOpcd1 = N->getOperand(0)->getOperand(1).getOpcode();
if ((SubOpcd0 == ISD::ZERO_EXTEND ||
SubOpcd0 == ISD::ZERO_EXTEND_VECTOR_INREG) &&
(SubOpcd1 == ISD::ZERO_EXTEND ||
SubOpcd1 == ISD::ZERO_EXTEND_VECTOR_INREG)) {
return DAG.getNode(PPCISD::VABSD, dl, N->getOperand(0).getValueType(),
N->getOperand(0)->getOperand(0),
N->getOperand(0)->getOperand(1),
DAG.getTargetConstant(0, dl, MVT::i32));
}
// For type v4i32, it can be optimized with xvnegsp + vabsduw
if (N->getOperand(0).getValueType() == MVT::v4i32 &&
N->getOperand(0).hasOneUse()) {
return DAG.getNode(PPCISD::VABSD, dl, N->getOperand(0).getValueType(),
N->getOperand(0)->getOperand(0),
N->getOperand(0)->getOperand(1),
DAG.getTargetConstant(1, dl, MVT::i32));
}
}
return SDValue();
}
// For type v4i32/v8ii16/v16i8, transform
// from (vselect (setcc a, b, setugt), (sub a, b), (sub b, a)) to (vabsd a, b)
// from (vselect (setcc a, b, setuge), (sub a, b), (sub b, a)) to (vabsd a, b)
// from (vselect (setcc a, b, setult), (sub b, a), (sub a, b)) to (vabsd a, b)
// from (vselect (setcc a, b, setule), (sub b, a), (sub a, b)) to (vabsd a, b)
SDValue PPCTargetLowering::combineVSelect(SDNode *N,
DAGCombinerInfo &DCI) const {
assert((N->getOpcode() == ISD::VSELECT) && "Need VSELECT node here");
assert(Subtarget.hasP9Altivec() &&
"Only combine this when P9 altivec supported!");
SelectionDAG &DAG = DCI.DAG;
SDLoc dl(N);
SDValue Cond = N->getOperand(0);
SDValue TrueOpnd = N->getOperand(1);
SDValue FalseOpnd = N->getOperand(2);
EVT VT = N->getOperand(1).getValueType();
if (Cond.getOpcode() != ISD::SETCC || TrueOpnd.getOpcode() != ISD::SUB ||
FalseOpnd.getOpcode() != ISD::SUB)
return SDValue();
// ABSD only available for type v4i32/v8i16/v16i8
if (VT != MVT::v4i32 && VT != MVT::v8i16 && VT != MVT::v16i8)
return SDValue();
// At least to save one more dependent computation
if (!(Cond.hasOneUse() || TrueOpnd.hasOneUse() || FalseOpnd.hasOneUse()))
return SDValue();
ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
// Can only handle unsigned comparison here
switch (CC) {
default:
return SDValue();
case ISD::SETUGT:
case ISD::SETUGE:
break;
case ISD::SETULT:
case ISD::SETULE:
std::swap(TrueOpnd, FalseOpnd);
break;
}
SDValue CmpOpnd1 = Cond.getOperand(0);
SDValue CmpOpnd2 = Cond.getOperand(1);
// SETCC CmpOpnd1 CmpOpnd2 cond
// TrueOpnd = CmpOpnd1 - CmpOpnd2
// FalseOpnd = CmpOpnd2 - CmpOpnd1
if (TrueOpnd.getOperand(0) == CmpOpnd1 &&
TrueOpnd.getOperand(1) == CmpOpnd2 &&
FalseOpnd.getOperand(0) == CmpOpnd2 &&
FalseOpnd.getOperand(1) == CmpOpnd1) {
return DAG.getNode(PPCISD::VABSD, dl, N->getOperand(1).getValueType(),
CmpOpnd1, CmpOpnd2,
DAG.getTargetConstant(0, dl, MVT::i32));
}
return SDValue();
}
Reference in New Issue View Git Blame Copy Permalink