You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

931 lines
36 KiB

//===- AArch64LegalizerInfo.cpp ----------------------------------*- C++ -*-==//
//
// 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
//
//===----------------------------------------------------------------------===//
/// \file
/// This file implements the targeting of the Machinelegalizer class for
/// AArch64.
/// \todo This should be generated by TableGen.
//===----------------------------------------------------------------------===//
#include "AArch64LegalizerInfo.h"
#include "AArch64Subtarget.h"
#include "llvm/CodeGen/GlobalISel/LegalizerHelper.h"
#include "llvm/CodeGen/GlobalISel/LegalizerInfo.h"
#include "llvm/CodeGen/GlobalISel/MachineIRBuilder.h"
#include "llvm/CodeGen/GlobalISel/Utils.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Type.h"
#include <initializer_list>
#include "llvm/Support/MathExtras.h"
#define DEBUG_TYPE "aarch64-legalinfo"
using namespace llvm;
using namespace LegalizeActions;
using namespace LegalizeMutations;
using namespace LegalityPredicates;
AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
: ST(&ST) {
using namespace TargetOpcode;
const LLT p0 = LLT::pointer(0, 64);
const LLT s1 = LLT::scalar(1);
const LLT s8 = LLT::scalar(8);
const LLT s16 = LLT::scalar(16);
const LLT s32 = LLT::scalar(32);
const LLT s64 = LLT::scalar(64);
const LLT s128 = LLT::scalar(128);
const LLT s256 = LLT::scalar(256);
const LLT s512 = LLT::scalar(512);
const LLT v16s8 = LLT::vector(16, 8);
const LLT v8s8 = LLT::vector(8, 8);
const LLT v4s8 = LLT::vector(4, 8);
const LLT v8s16 = LLT::vector(8, 16);
const LLT v4s16 = LLT::vector(4, 16);
const LLT v2s16 = LLT::vector(2, 16);
const LLT v2s32 = LLT::vector(2, 32);
const LLT v4s32 = LLT::vector(4, 32);
const LLT v2s64 = LLT::vector(2, 64);
const LLT v2p0 = LLT::vector(2, p0);
std::initializer_list<LLT> PackedVectorAllTypeList = {/* Begin 128bit types */
v16s8, v8s16, v4s32,
v2s64, v2p0,
/* End 128bit types */
/* Begin 64bit types */
v8s8, v4s16, v2s32};
const TargetMachine &TM = ST.getTargetLowering()->getTargetMachine();
// FIXME: support subtargets which have neon/fp-armv8 disabled.
if (!ST.hasNEON() || !ST.hasFPARMv8()) {
computeTables();
return;
}
// Some instructions only support s16 if the subtarget has full 16-bit FP
// support.
const bool HasFP16 = ST.hasFullFP16();
const LLT &MinFPScalar = HasFP16 ? s16 : s32;
getActionDefinitionsBuilder({G_IMPLICIT_DEF, G_FREEZE})
.legalFor({p0, s1, s8, s16, s32, s64})
.legalFor(PackedVectorAllTypeList)
.clampScalar(0, s1, s64)
.widenScalarToNextPow2(0, 8)
.fewerElementsIf(
[=](const LegalityQuery &Query) {
return Query.Types[0].isVector() &&
(Query.Types[0].getElementType() != s64 ||
Query.Types[0].getNumElements() != 2);
},
[=](const LegalityQuery &Query) {
LLT EltTy = Query.Types[0].getElementType();
if (EltTy == s64)
return std::make_pair(0, LLT::vector(2, 64));
return std::make_pair(0, EltTy);
});
getActionDefinitionsBuilder(G_PHI).legalFor({p0, s16, s32, s64})
.legalFor(PackedVectorAllTypeList)
.clampScalar(0, s16, s64)
.widenScalarToNextPow2(0);
getActionDefinitionsBuilder(G_BSWAP)
.legalFor({s32, s64, v4s32, v2s32, v2s64})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0);
getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR})
.legalFor({s32, s64, v2s32, v4s32, v4s16, v8s16, v16s8, v8s8})
.scalarizeIf(
[=](const LegalityQuery &Query) {
return Query.Opcode == G_MUL && Query.Types[0] == v2s64;
},
0)
.legalFor({v2s64})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.clampNumElements(0, v2s32, v4s32)
.clampNumElements(0, v2s64, v2s64)
.moreElementsToNextPow2(0);
getActionDefinitionsBuilder({G_SHL, G_ASHR, G_LSHR})
.customIf([=](const LegalityQuery &Query) {
const auto &SrcTy = Query.Types[0];
const auto &AmtTy = Query.Types[1];
return !SrcTy.isVector() && SrcTy.getSizeInBits() == 32 &&
AmtTy.getSizeInBits() == 32;
})
.legalFor({
{s32, s32},
{s32, s64},
{s64, s64},
{v8s8, v8s8},
{v16s8, v16s8},
{v4s16, v4s16},
{v8s16, v8s16},
{v2s32, v2s32},
{v4s32, v4s32},
{v2s64, v2s64},
})
.clampScalar(1, s32, s64)
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.clampNumElements(0, v2s32, v4s32)
.clampNumElements(0, v2s64, v2s64)
.moreElementsToNextPow2(0)
.minScalarSameAs(1, 0);
getActionDefinitionsBuilder(G_PTR_ADD)
.legalFor({{p0, s64}, {v2p0, v2s64}})
.clampScalar(1, s64, s64);
getActionDefinitionsBuilder(G_PTRMASK).legalFor({{p0, s64}});
getActionDefinitionsBuilder({G_SDIV, G_UDIV})
.legalFor({s32, s64})
.libcallFor({s128})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.scalarize(0);
getActionDefinitionsBuilder({G_SREM, G_UREM})
.lowerFor({s1, s8, s16, s32, s64});
getActionDefinitionsBuilder({G_SMULO, G_UMULO}).lowerFor({{s64, s1}});
getActionDefinitionsBuilder({G_SMULH, G_UMULH}).legalFor({s32, s64});
getActionDefinitionsBuilder({G_UADDE, G_USUBE, G_SADDO, G_SSUBO, G_UADDO})
.legalFor({{s32, s1}, {s64, s1}})
.minScalar(0, s32);
getActionDefinitionsBuilder({G_FADD, G_FSUB, G_FMUL, G_FDIV, G_FNEG})
.legalFor({s32, s64, v2s64, v4s32, v2s32})
.clampNumElements(0, v2s32, v4s32)
.clampNumElements(0, v2s64, v2s64);
getActionDefinitionsBuilder(G_FREM).libcallFor({s32, s64});
getActionDefinitionsBuilder({G_FCEIL, G_FABS, G_FSQRT, G_FFLOOR, G_FRINT,
G_FMA, G_INTRINSIC_TRUNC, G_INTRINSIC_ROUND,
G_FNEARBYINT, G_INTRINSIC_LRINT})
// If we don't have full FP16 support, then scalarize the elements of
// vectors containing fp16 types.
.fewerElementsIf(
[=, &ST](const LegalityQuery &Query) {
const auto &Ty = Query.Types[0];
return Ty.isVector() && Ty.getElementType() == s16 &&
!ST.hasFullFP16();
},
[=](const LegalityQuery &Query) { return std::make_pair(0, s16); })
// If we don't have full FP16 support, then widen s16 to s32 if we
// encounter it.
.widenScalarIf(
[=, &ST](const LegalityQuery &Query) {
return Query.Types[0] == s16 && !ST.hasFullFP16();
},
[=](const LegalityQuery &Query) { return std::make_pair(0, s32); })
.legalFor({s16, s32, s64, v2s32, v4s32, v2s64, v2s16, v4s16, v8s16});
getActionDefinitionsBuilder(
{G_FCOS, G_FSIN, G_FLOG10, G_FLOG, G_FLOG2, G_FEXP, G_FEXP2, G_FPOW})
// We need a call for these, so we always need to scalarize.
.scalarize(0)
// Regardless of FP16 support, widen 16-bit elements to 32-bits.
.minScalar(0, s32)
.libcallFor({s32, s64, v2s32, v4s32, v2s64});
getActionDefinitionsBuilder(G_INSERT)
.unsupportedIf([=](const LegalityQuery &Query) {
return Query.Types[0].getSizeInBits() <= Query.Types[1].getSizeInBits();
})
.legalIf([=](const LegalityQuery &Query) {
const LLT &Ty0 = Query.Types[0];
const LLT &Ty1 = Query.Types[1];
if (Ty0 != s32 && Ty0 != s64 && Ty0 != p0)
return false;
return isPowerOf2_32(Ty1.getSizeInBits()) &&
(Ty1.getSizeInBits() == 1 || Ty1.getSizeInBits() >= 8);
})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.maxScalarIf(typeInSet(0, {s32}), 1, s16)
.maxScalarIf(typeInSet(0, {s64}), 1, s32)
.widenScalarToNextPow2(1);
getActionDefinitionsBuilder(G_EXTRACT)
.unsupportedIf([=](const LegalityQuery &Query) {
return Query.Types[0].getSizeInBits() >= Query.Types[1].getSizeInBits();
})
.legalIf([=](const LegalityQuery &Query) {
const LLT &Ty0 = Query.Types[0];
const LLT &Ty1 = Query.Types[1];
if (Ty1 != s32 && Ty1 != s64 && Ty1 != s128)
return false;
if (Ty1 == p0)
return true;
return isPowerOf2_32(Ty0.getSizeInBits()) &&
(Ty0.getSizeInBits() == 1 || Ty0.getSizeInBits() >= 8);
})
.clampScalar(1, s32, s128)
.widenScalarToNextPow2(1)
.maxScalarIf(typeInSet(1, {s32}), 0, s16)
.maxScalarIf(typeInSet(1, {s64}), 0, s32)
.widenScalarToNextPow2(0);
getActionDefinitionsBuilder({G_SEXTLOAD, G_ZEXTLOAD})
.legalForTypesWithMemDesc({{s32, p0, 8, 8},
{s32, p0, 16, 8},
{s32, p0, 32, 8},
{s64, p0, 8, 2},
{s64, p0, 16, 2},
{s64, p0, 32, 4},
{s64, p0, 64, 8},
{p0, p0, 64, 8},
{v2s32, p0, 64, 8}})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
// TODO: We could support sum-of-pow2's but the lowering code doesn't know
// how to do that yet.
.unsupportedIfMemSizeNotPow2()
// Lower anything left over into G_*EXT and G_LOAD
.lower();
auto IsPtrVecPred = [=](const LegalityQuery &Query) {
const LLT &ValTy = Query.Types[0];
if (!ValTy.isVector())
return false;
const LLT EltTy = ValTy.getElementType();
return EltTy.isPointer() && EltTy.getAddressSpace() == 0;
};
getActionDefinitionsBuilder(G_LOAD)
.legalForTypesWithMemDesc({{s8, p0, 8, 8},
{s16, p0, 16, 8},
{s32, p0, 32, 8},
{s64, p0, 64, 8},
{p0, p0, 64, 8},
{s128, p0, 128, 8},
{v8s8, p0, 64, 8},
{v16s8, p0, 128, 8},
{v4s16, p0, 64, 8},
{v8s16, p0, 128, 8},
{v2s32, p0, 64, 8},
{v4s32, p0, 128, 8},
{v2s64, p0, 128, 8}})
// These extends are also legal
.legalForTypesWithMemDesc({{s32, p0, 8, 8}, {s32, p0, 16, 8}})
.clampScalar(0, s8, s64)
.lowerIfMemSizeNotPow2()
// Lower any any-extending loads left into G_ANYEXT and G_LOAD
.lowerIf([=](const LegalityQuery &Query) {
return Query.Types[0].getSizeInBits() != Query.MMODescrs[0].SizeInBits;
})
.widenScalarToNextPow2(0)
.clampMaxNumElements(0, s32, 2)
.clampMaxNumElements(0, s64, 1)
.customIf(IsPtrVecPred);
getActionDefinitionsBuilder(G_STORE)
.legalForTypesWithMemDesc({{s8, p0, 8, 8},
{s16, p0, 16, 8},
{s32, p0, 8, 8},
{s32, p0, 16, 8},
{s32, p0, 32, 8},
{s64, p0, 64, 8},
{p0, p0, 64, 8},
{s128, p0, 128, 8},
{v16s8, p0, 128, 8},
{v8s8, p0, 64, 8},
{v4s16, p0, 64, 8},
{v8s16, p0, 128, 8},
{v2s32, p0, 64, 8},
{v4s32, p0, 128, 8},
{v2s64, p0, 128, 8}})
.clampScalar(0, s8, s64)
.lowerIfMemSizeNotPow2()
.lowerIf([=](const LegalityQuery &Query) {
return Query.Types[0].isScalar() &&
Query.Types[0].getSizeInBits() != Query.MMODescrs[0].SizeInBits;
})
.clampMaxNumElements(0, s32, 2)
.clampMaxNumElements(0, s64, 1)
.customIf(IsPtrVecPred);
// Constants
getActionDefinitionsBuilder(G_CONSTANT)
.legalFor({p0, s8, s16, s32, s64})
.clampScalar(0, s8, s64)
.widenScalarToNextPow2(0);
getActionDefinitionsBuilder(G_FCONSTANT)
.legalIf([=](const LegalityQuery &Query) {
const auto &Ty = Query.Types[0];
if (HasFP16 && Ty == s16)
return true;
return Ty == s32 || Ty == s64;
})
.clampScalar(0, MinFPScalar, s64);
getActionDefinitionsBuilder({G_ICMP, G_FCMP})
.legalFor({{s32, s32},
{s32, s64},
{s32, p0},
{v4s32, v4s32},
{v2s32, v2s32},
{v2s64, v2s64},
{v2s64, v2p0},
{v4s16, v4s16},
{v8s16, v8s16},
{v8s8, v8s8},
{v16s8, v16s8}})
.clampScalar(1, s32, s64)
.clampScalar(0, s32, s32)
.minScalarEltSameAsIf(
[=](const LegalityQuery &Query) {
const LLT &Ty = Query.Types[0];
const LLT &SrcTy = Query.Types[1];
return Ty.isVector() && !SrcTy.getElementType().isPointer() &&
Ty.getElementType() != SrcTy.getElementType();
},
0, 1)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) { return Query.Types[1] == v2s16; },
1, s32)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) { return Query.Types[1] == v2p0; }, 0,
s64)
.widenScalarOrEltToNextPow2(1)
.clampNumElements(0, v2s32, v4s32);
// Extensions
auto ExtLegalFunc = [=](const LegalityQuery &Query) {
unsigned DstSize = Query.Types[0].getSizeInBits();
if (DstSize == 128 && !Query.Types[0].isVector())
return false; // Extending to a scalar s128 needs narrowing.
// Make sure that we have something that will fit in a register, and
// make sure it's a power of 2.
if (DstSize < 8 || DstSize > 128 || !isPowerOf2_32(DstSize))
return false;
const LLT &SrcTy = Query.Types[1];
// Special case for s1.
if (SrcTy == s1)
return true;
// Make sure we fit in a register otherwise. Don't bother checking that
// the source type is below 128 bits. We shouldn't be allowing anything
// through which is wider than the destination in the first place.
unsigned SrcSize = SrcTy.getSizeInBits();
if (SrcSize < 8 || !isPowerOf2_32(SrcSize))
return false;
return true;
};
getActionDefinitionsBuilder({G_ZEXT, G_SEXT, G_ANYEXT})
.legalIf(ExtLegalFunc)
.clampScalar(0, s64, s64); // Just for s128, others are handled above.
getActionDefinitionsBuilder(G_TRUNC)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) { return Query.Types[0].isVector(); },
0, s8)
.customIf([=](const LegalityQuery &Query) {
LLT DstTy = Query.Types[0];
LLT SrcTy = Query.Types[1];
return DstTy == v8s8 && SrcTy.getSizeInBits() > 128;
})
.alwaysLegal();
getActionDefinitionsBuilder(G_SEXT_INREG).legalFor({s32, s64}).lower();
// FP conversions
getActionDefinitionsBuilder(G_FPTRUNC)
.legalFor(
{{s16, s32}, {s16, s64}, {s32, s64}, {v4s16, v4s32}, {v2s32, v2s64}})
.clampMaxNumElements(0, s32, 2);
getActionDefinitionsBuilder(G_FPEXT)
.legalFor(
{{s32, s16}, {s64, s16}, {s64, s32}, {v4s32, v4s16}, {v2s64, v2s32}})
.clampMaxNumElements(0, s64, 2);
// Conversions
getActionDefinitionsBuilder({G_FPTOSI, G_FPTOUI})
.legalForCartesianProduct({s32, s64, v2s64, v4s32, v2s32})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.clampScalar(1, s32, s64)
.widenScalarToNextPow2(1);
getActionDefinitionsBuilder({G_SITOFP, G_UITOFP})
.legalForCartesianProduct({s32, s64, v2s64, v4s32, v2s32})
.clampScalar(1, s32, s64)
.minScalarSameAs(1, 0)
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0);
// Control-flow
getActionDefinitionsBuilder(G_BRCOND).legalFor({s1, s8, s16, s32});
getActionDefinitionsBuilder(G_BRINDIRECT).legalFor({p0});
getActionDefinitionsBuilder(G_SELECT)
.legalFor({{s32, s1}, {s64, s1}, {p0, s1}})
.clampScalar(0, s32, s64)
.widenScalarToNextPow2(0)
.minScalarEltSameAsIf(all(isVector(0), isVector(1)), 1, 0)
.lowerIf(isVector(0));
// Pointer-handling
getActionDefinitionsBuilder(G_FRAME_INDEX).legalFor({p0});
if (TM.getCodeModel() == CodeModel::Small)
getActionDefinitionsBuilder(G_GLOBAL_VALUE).custom();
else
getActionDefinitionsBuilder(G_GLOBAL_VALUE).legalFor({p0});
getActionDefinitionsBuilder(G_PTRTOINT)
.legalForCartesianProduct({s1, s8, s16, s32, s64}, {p0})
.maxScalar(0, s64)
.widenScalarToNextPow2(0, /*Min*/ 8);
getActionDefinitionsBuilder(G_INTTOPTR)
.unsupportedIf([&](const LegalityQuery &Query) {
return Query.Types[0].getSizeInBits() != Query.Types[1].getSizeInBits();
})
.legalFor({{p0, s64}});
// Casts for 32 and 64-bit width type are just copies.
// Same for 128-bit width type, except they are on the FPR bank.
getActionDefinitionsBuilder(G_BITCAST)
// FIXME: This is wrong since G_BITCAST is not allowed to change the
// number of bits but it's what the previous code described and fixing
// it breaks tests.
.legalForCartesianProduct({s1, s8, s16, s32, s64, s128, v16s8, v8s8, v4s8,
v8s16, v4s16, v2s16, v4s32, v2s32, v2s64,
v2p0});
getActionDefinitionsBuilder(G_VASTART).legalFor({p0});
// va_list must be a pointer, but most sized types are pretty easy to handle
// as the destination.
getActionDefinitionsBuilder(G_VAARG)
.customForCartesianProduct({s8, s16, s32, s64, p0}, {p0})
.clampScalar(0, s8, s64)
.widenScalarToNextPow2(0, /*Min*/ 8);
if (ST.hasLSE()) {
getActionDefinitionsBuilder(G_ATOMIC_CMPXCHG_WITH_SUCCESS)
.lowerIf(all(
typeInSet(0, {s8, s16, s32, s64}), typeIs(1, s1), typeIs(2, p0),
atomicOrderingAtLeastOrStrongerThan(0, AtomicOrdering::Monotonic)));
getActionDefinitionsBuilder(
{G_ATOMICRMW_XCHG, G_ATOMICRMW_ADD, G_ATOMICRMW_SUB, G_ATOMICRMW_AND,
G_ATOMICRMW_OR, G_ATOMICRMW_XOR, G_ATOMICRMW_MIN, G_ATOMICRMW_MAX,
G_ATOMICRMW_UMIN, G_ATOMICRMW_UMAX, G_ATOMIC_CMPXCHG})
.legalIf(all(
typeInSet(0, {s8, s16, s32, s64}), typeIs(1, p0),
atomicOrderingAtLeastOrStrongerThan(0, AtomicOrdering::Monotonic)));
}
getActionDefinitionsBuilder(G_BLOCK_ADDR).legalFor({p0});
// Merge/Unmerge
for (unsigned Op : {G_MERGE_VALUES, G_UNMERGE_VALUES}) {
unsigned BigTyIdx = Op == G_MERGE_VALUES ? 0 : 1;
unsigned LitTyIdx = Op == G_MERGE_VALUES ? 1 : 0;
auto notValidElt = [](const LegalityQuery &Query, unsigned TypeIdx) {
const LLT &Ty = Query.Types[TypeIdx];
if (Ty.isVector()) {
const LLT &EltTy = Ty.getElementType();
if (EltTy.getSizeInBits() < 8 || EltTy.getSizeInBits() > 64)
return true;
if (!isPowerOf2_32(EltTy.getSizeInBits()))
return true;
}
return false;
};
// FIXME: This rule is horrible, but specifies the same as what we had
// before with the particularly strange definitions removed (e.g.
// s8 = G_MERGE_VALUES s32, s32).
// Part of the complexity comes from these ops being extremely flexible. For
// example, you can build/decompose vectors with it, concatenate vectors,
// etc. and in addition to this you can also bitcast with it at the same
// time. We've been considering breaking it up into multiple ops to make it
// more manageable throughout the backend.
getActionDefinitionsBuilder(Op)
// Break up vectors with weird elements into scalars
.fewerElementsIf(
[=](const LegalityQuery &Query) { return notValidElt(Query, 0); },
scalarize(0))
.fewerElementsIf(
[=](const LegalityQuery &Query) { return notValidElt(Query, 1); },
scalarize(1))
// Clamp the big scalar to s8-s512 and make it either a power of 2, 192,
// or 384.
.clampScalar(BigTyIdx, s8, s512)
.widenScalarIf(
[=](const LegalityQuery &Query) {
const LLT &Ty = Query.Types[BigTyIdx];
return !isPowerOf2_32(Ty.getSizeInBits()) &&
Ty.getSizeInBits() % 64 != 0;
},
[=](const LegalityQuery &Query) {
// Pick the next power of 2, or a multiple of 64 over 128.
// Whichever is smaller.
const LLT &Ty = Query.Types[BigTyIdx];
unsigned NewSizeInBits = 1
<< Log2_32_Ceil(Ty.getSizeInBits() + 1);
if (NewSizeInBits >= 256) {
unsigned RoundedTo = alignTo<64>(Ty.getSizeInBits() + 1);
if (RoundedTo < NewSizeInBits)
NewSizeInBits = RoundedTo;
}
return std::make_pair(BigTyIdx, LLT::scalar(NewSizeInBits));
})
// Clamp the little scalar to s8-s256 and make it a power of 2. It's not
// worth considering the multiples of 64 since 2*192 and 2*384 are not
// valid.
.clampScalar(LitTyIdx, s8, s256)
.widenScalarToNextPow2(LitTyIdx, /*Min*/ 8)
// So at this point, we have s8, s16, s32, s64, s128, s192, s256, s384,
// s512, <X x s8>, <X x s16>, <X x s32>, or <X x s64>.
// At this point it's simple enough to accept the legal types.
.legalIf([=](const LegalityQuery &Query) {
const LLT &BigTy = Query.Types[BigTyIdx];
const LLT &LitTy = Query.Types[LitTyIdx];
if (BigTy.isVector() && BigTy.getSizeInBits() < 32)
return false;
if (LitTy.isVector() && LitTy.getSizeInBits() < 32)
return false;
return BigTy.getSizeInBits() % LitTy.getSizeInBits() == 0;
})
// Any vectors left are the wrong size. Scalarize them.
.scalarize(0)
.scalarize(1);
}
getActionDefinitionsBuilder(G_EXTRACT_VECTOR_ELT)
.unsupportedIf([=](const LegalityQuery &Query) {
const LLT &EltTy = Query.Types[1].getElementType();
return Query.Types[0] != EltTy;
})
.minScalar(2, s64)
.legalIf([=](const LegalityQuery &Query) {
const LLT &VecTy = Query.Types[1];
return VecTy == v2s16 || VecTy == v4s16 || VecTy == v8s16 ||
VecTy == v4s32 || VecTy == v2s64 || VecTy == v2s32 ||
VecTy == v16s8 || VecTy == v2s32 || VecTy == v2p0;
})
.minScalarOrEltIf(
[=](const LegalityQuery &Query) {
// We want to promote to <M x s1> to <M x s64> if that wouldn't
// cause the total vec size to be > 128b.
return Query.Types[1].getNumElements() <= 2;
},
0, s64)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) {
return Query.Types[1].getNumElements() <= 4;
},
0, s32)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) {
return Query.Types[1].getNumElements() <= 8;
},
0, s16)
.minScalarOrEltIf(
[=](const LegalityQuery &Query) {
return Query.Types[1].getNumElements() <= 16;
},
0, s8)
.minScalarOrElt(0, s8); // Worst case, we need at least s8.
getActionDefinitionsBuilder(G_INSERT_VECTOR_ELT)
.legalIf(typeInSet(0, {v8s16, v2s32, v4s32, v2s64}));
getActionDefinitionsBuilder(G_BUILD_VECTOR)
.legalFor({{v8s8, s8},
{v16s8, s8},
{v4s16, s16},
{v8s16, s16},
{v2s32, s32},
{v4s32, s32},
{v2p0, p0},
{v2s64, s64}})
.clampNumElements(0, v4s32, v4s32)
.clampNumElements(0, v2s64, v2s64)
// Deal with larger scalar types, which will be implicitly truncated.
.legalIf([=](const LegalityQuery &Query) {
return Query.Types[0].getScalarSizeInBits() <
Query.Types[1].getSizeInBits();
})
.minScalarSameAs(1, 0);
getActionDefinitionsBuilder(G_CTLZ)
.legalForCartesianProduct(
{s32, s64, v8s8, v16s8, v4s16, v8s16, v2s32, v4s32})
.scalarize(1);
getActionDefinitionsBuilder(G_SHUFFLE_VECTOR)
.legalIf([=](const LegalityQuery &Query) {
const LLT &DstTy = Query.Types[0];
const LLT &SrcTy = Query.Types[1];
// For now just support the TBL2 variant which needs the source vectors
// to be the same size as the dest.
if (DstTy != SrcTy)
return false;
for (auto &Ty : {v2s32, v4s32, v2s64, v2p0, v16s8, v8s16}) {
if (DstTy == Ty)
return true;
}
return false;
})
// G_SHUFFLE_VECTOR can have scalar sources (from 1 x s vectors), we
// just want those lowered into G_BUILD_VECTOR
.lowerIf([=](const LegalityQuery &Query) {
return !Query.Types[1].isVector();
})
.clampNumElements(0, v4s32, v4s32)
.clampNumElements(0, v2s64, v2s64);
getActionDefinitionsBuilder(G_CONCAT_VECTORS)
.legalFor({{v4s32, v2s32}, {v8s16, v4s16}});
getActionDefinitionsBuilder(G_JUMP_TABLE).legalFor({{p0}, {s64}});
getActionDefinitionsBuilder(G_BRJT).legalIf([=](const LegalityQuery &Query) {
return Query.Types[0] == p0 && Query.Types[1] == s64;
});
getActionDefinitionsBuilder(G_DYN_STACKALLOC).lower();
getActionDefinitionsBuilder({G_MEMCPY, G_MEMMOVE, G_MEMSET}).libcall();
getActionDefinitionsBuilder(G_ABS).lowerIf(
[=](const LegalityQuery &Query) { return Query.Types[0].isScalar(); });
getActionDefinitionsBuilder(G_VECREDUCE_FADD)
// We only have FADDP to do reduction-like operations. Lower the rest.
.legalFor({{s32, v2s32}, {s64, v2s64}})
.lower();
getActionDefinitionsBuilder(G_VECREDUCE_ADD)
.legalFor({{s8, v16s8}, {s16, v8s16}, {s32, v4s32}, {s64, v2s64}})
.lower();
computeTables();
verify(*ST.getInstrInfo());
}
bool AArch64LegalizerInfo::legalizeCustom(LegalizerHelper &Helper,
MachineInstr &MI) const {
MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
MachineRegisterInfo &MRI = *MIRBuilder.getMRI();
GISelChangeObserver &Observer = Helper.Observer;
switch (MI.getOpcode()) {
default:
// No idea what to do.
return false;
case TargetOpcode::G_VAARG:
return legalizeVaArg(MI, MRI, MIRBuilder);
case TargetOpcode::G_LOAD:
case TargetOpcode::G_STORE:
return legalizeLoadStore(MI, MRI, MIRBuilder, Observer);
case TargetOpcode::G_SHL:
case TargetOpcode::G_ASHR:
case TargetOpcode::G_LSHR:
return legalizeShlAshrLshr(MI, MRI, MIRBuilder, Observer);
case TargetOpcode::G_GLOBAL_VALUE:
return legalizeSmallCMGlobalValue(MI, MRI, MIRBuilder, Observer);
case TargetOpcode::G_TRUNC:
return legalizeVectorTrunc(MI, Helper);
}
llvm_unreachable("expected switch to return");
}
static void extractParts(Register Reg, MachineRegisterInfo &MRI,
MachineIRBuilder &MIRBuilder, LLT Ty, int NumParts,
SmallVectorImpl<Register> &VRegs) {
for (int I = 0; I < NumParts; ++I)
VRegs.push_back(MRI.createGenericVirtualRegister(Ty));
MIRBuilder.buildUnmerge(VRegs, Reg);
}
bool AArch64LegalizerInfo::legalizeVectorTrunc(
MachineInstr &MI, LegalizerHelper &Helper) const {
MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
MachineRegisterInfo &MRI = *MIRBuilder.getMRI();
// Similar to how operand splitting is done in SelectiondDAG, we can handle
// %res(v8s8) = G_TRUNC %in(v8s32) by generating:
// %inlo(<4x s32>), %inhi(<4 x s32>) = G_UNMERGE %in(<8 x s32>)
// %lo16(<4 x s16>) = G_TRUNC %inlo
// %hi16(<4 x s16>) = G_TRUNC %inhi
// %in16(<8 x s16>) = G_CONCAT_VECTORS %lo16, %hi16
// %res(<8 x s8>) = G_TRUNC %in16
Register DstReg = MI.getOperand(0).getReg();
Register SrcReg = MI.getOperand(1).getReg();
LLT DstTy = MRI.getType(DstReg);
LLT SrcTy = MRI.getType(SrcReg);
assert(isPowerOf2_32(DstTy.getSizeInBits()) &&
isPowerOf2_32(SrcTy.getSizeInBits()));
// Split input type.
LLT SplitSrcTy = SrcTy.changeNumElements(SrcTy.getNumElements() / 2);
// First, split the source into two smaller vectors.
SmallVector<Register, 2> SplitSrcs;
extractParts(SrcReg, MRI, MIRBuilder, SplitSrcTy, 2, SplitSrcs);
// Truncate the splits into intermediate narrower elements.
LLT InterTy = SplitSrcTy.changeElementSize(DstTy.getScalarSizeInBits() * 2);
for (unsigned I = 0; I < SplitSrcs.size(); ++I)
SplitSrcs[I] = MIRBuilder.buildTrunc(InterTy, SplitSrcs[I]).getReg(0);
auto Concat = MIRBuilder.buildConcatVectors(
DstTy.changeElementSize(DstTy.getScalarSizeInBits() * 2), SplitSrcs);
Helper.Observer.changingInstr(MI);
MI.getOperand(1).setReg(Concat.getReg(0));
Helper.Observer.changedInstr(MI);
return true;
}
bool AArch64LegalizerInfo::legalizeSmallCMGlobalValue(
MachineInstr &MI, MachineRegisterInfo &MRI, MachineIRBuilder &MIRBuilder,
GISelChangeObserver &Observer) const {
assert(MI.getOpcode() == TargetOpcode::G_GLOBAL_VALUE);
// We do this custom legalization to convert G_GLOBAL_VALUE into target ADRP +
// G_ADD_LOW instructions.
// By splitting this here, we can optimize accesses in the small code model by
// folding in the G_ADD_LOW into the load/store offset.
auto GV = MI.getOperand(1).getGlobal();
if (GV->isThreadLocal())
return true; // Don't want to modify TLS vars.
auto &TM = ST->getTargetLowering()->getTargetMachine();
unsigned OpFlags = ST->ClassifyGlobalReference(GV, TM);
if (OpFlags & AArch64II::MO_GOT)
return true;
Register DstReg = MI.getOperand(0).getReg();
auto ADRP = MIRBuilder.buildInstr(AArch64::ADRP, {LLT::pointer(0, 64)}, {})
.addGlobalAddress(GV, 0, OpFlags | AArch64II::MO_PAGE);
// Set the regclass on the dest reg too.
MRI.setRegClass(ADRP.getReg(0), &AArch64::GPR64RegClass);
// MO_TAGGED on the page indicates a tagged address. Set the tag now. We do so
// by creating a MOVK that sets bits 48-63 of the register to (global address
// + 0x100000000 - PC) >> 48. The additional 0x100000000 offset here is to
// prevent an incorrect tag being generated during relocation when the the
// global appears before the code section. Without the offset, a global at
// `0x0f00'0000'0000'1000` (i.e. at `0x1000` with tag `0xf`) that's referenced
// by code at `0x2000` would result in `0x0f00'0000'0000'1000 - 0x2000 =
// 0x0eff'ffff'ffff'f000`, meaning the tag would be incorrectly set to `0xe`
// instead of `0xf`.
// This assumes that we're in the small code model so we can assume a binary
// size of <= 4GB, which makes the untagged PC relative offset positive. The
// binary must also be loaded into address range [0, 2^48). Both of these
// properties need to be ensured at runtime when using tagged addresses.
if (OpFlags & AArch64II::MO_TAGGED) {
ADRP = MIRBuilder.buildInstr(AArch64::MOVKXi, {LLT::pointer(0, 64)}, {ADRP})
.addGlobalAddress(GV, 0x100000000,
AArch64II::MO_PREL | AArch64II::MO_G3)
.addImm(48);
MRI.setRegClass(ADRP.getReg(0), &AArch64::GPR64RegClass);
}
MIRBuilder.buildInstr(AArch64::G_ADD_LOW, {DstReg}, {ADRP})
.addGlobalAddress(GV, 0,
OpFlags | AArch64II::MO_PAGEOFF | AArch64II::MO_NC);
MI.eraseFromParent();
return true;
}
bool AArch64LegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
MachineInstr &MI) const {
return true;
}
bool AArch64LegalizerInfo::legalizeShlAshrLshr(
MachineInstr &MI, MachineRegisterInfo &MRI, MachineIRBuilder &MIRBuilder,
GISelChangeObserver &Observer) const {
assert(MI.getOpcode() == TargetOpcode::G_ASHR ||
MI.getOpcode() == TargetOpcode::G_LSHR ||
MI.getOpcode() == TargetOpcode::G_SHL);
// If the shift amount is a G_CONSTANT, promote it to a 64 bit type so the
// imported patterns can select it later. Either way, it will be legal.
Register AmtReg = MI.getOperand(2).getReg();
auto VRegAndVal = getConstantVRegValWithLookThrough(AmtReg, MRI);
if (!VRegAndVal)
return true;
// Check the shift amount is in range for an immediate form.
int64_t Amount = VRegAndVal->Value;
if (Amount > 31)
return true; // This will have to remain a register variant.
auto ExtCst = MIRBuilder.buildConstant(LLT::scalar(64), Amount);
MI.getOperand(2).setReg(ExtCst.getReg(0));
return true;
}
bool AArch64LegalizerInfo::legalizeLoadStore(
MachineInstr &MI, MachineRegisterInfo &MRI, MachineIRBuilder &MIRBuilder,
GISelChangeObserver &Observer) const {
assert(MI.getOpcode() == TargetOpcode::G_STORE ||
MI.getOpcode() == TargetOpcode::G_LOAD);
// Here we just try to handle vector loads/stores where our value type might
// have pointer elements, which the SelectionDAG importer can't handle. To
// allow the existing patterns for s64 to fire for p0, we just try to bitcast
// the value to use s64 types.
// Custom legalization requires the instruction, if not deleted, must be fully
// legalized. In order to allow further legalization of the inst, we create
// a new instruction and erase the existing one.
Register ValReg = MI.getOperand(0).getReg();
const LLT ValTy = MRI.getType(ValReg);
if (!ValTy.isVector() || !ValTy.getElementType().isPointer() ||
ValTy.getElementType().getAddressSpace() != 0) {
LLVM_DEBUG(dbgs() << "Tried to do custom legalization on wrong load/store");
return false;
}
unsigned PtrSize = ValTy.getElementType().getSizeInBits();
const LLT NewTy = LLT::vector(ValTy.getNumElements(), PtrSize);
auto &MMO = **MI.memoperands_begin();
if (MI.getOpcode() == TargetOpcode::G_STORE) {
auto Bitcast = MIRBuilder.buildBitcast(NewTy, ValReg);
MIRBuilder.buildStore(Bitcast.getReg(0), MI.getOperand(1), MMO);
} else {
auto NewLoad = MIRBuilder.buildLoad(NewTy, MI.getOperand(1), MMO);
MIRBuilder.buildBitcast(ValReg, NewLoad);
}
MI.eraseFromParent();
return true;
}
bool AArch64LegalizerInfo::legalizeVaArg(MachineInstr &MI,
MachineRegisterInfo &MRI,
MachineIRBuilder &MIRBuilder) const {
MachineFunction &MF = MIRBuilder.getMF();
Align Alignment(MI.getOperand(2).getImm());
Register Dst = MI.getOperand(0).getReg();
Register ListPtr = MI.getOperand(1).getReg();
LLT PtrTy = MRI.getType(ListPtr);
LLT IntPtrTy = LLT::scalar(PtrTy.getSizeInBits());
const unsigned PtrSize = PtrTy.getSizeInBits() / 8;
const Align PtrAlign = Align(PtrSize);
auto List = MIRBuilder.buildLoad(
PtrTy, ListPtr,
*MF.getMachineMemOperand(MachinePointerInfo(), MachineMemOperand::MOLoad,
PtrSize, PtrAlign));
MachineInstrBuilder DstPtr;
if (Alignment > PtrAlign) {
// Realign the list to the actual required alignment.
auto AlignMinus1 =
MIRBuilder.buildConstant(IntPtrTy, Alignment.value() - 1);
auto ListTmp = MIRBuilder.buildPtrAdd(PtrTy, List, AlignMinus1.getReg(0));
DstPtr = MIRBuilder.buildMaskLowPtrBits(PtrTy, ListTmp, Log2(Alignment));
} else
DstPtr = List;
uint64_t ValSize = MRI.getType(Dst).getSizeInBits() / 8;
MIRBuilder.buildLoad(
Dst, DstPtr,
*MF.getMachineMemOperand(MachinePointerInfo(), MachineMemOperand::MOLoad,
ValSize, std::max(Alignment, PtrAlign)));
auto Size = MIRBuilder.buildConstant(IntPtrTy, alignTo(ValSize, PtrAlign));
auto NewList = MIRBuilder.buildPtrAdd(PtrTy, DstPtr, Size.getReg(0));
MIRBuilder.buildStore(NewList, ListPtr,
*MF.getMachineMemOperand(MachinePointerInfo(),
MachineMemOperand::MOStore,
PtrSize, PtrAlign));
MI.eraseFromParent();
return true;
}