Google Git
Sign in
cobalt / cobalt / a3dd835b53c74ec0f3e791c01fbcde33bbf26c6d / . / src / v8 / src / arm64 / assembler-arm64.cc
blob: a031884e1f70f2a9235a2c423dd854bc89d62000 [file] [log] [blame]
// Copyright 2013 the V8 project authors. All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
// * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#if V8_TARGET_ARCH_ARM64
#include "src/arm64/assembler-arm64.h"
#include "src/arm64/assembler-arm64-inl.h"
#include "src/base/bits.h"
#include "src/base/cpu.h"
#include "src/code-stubs.h"
#include "src/frame-constants.h"
#include "src/register-configuration.h"
namespace v8 {
namespace internal {
// -----------------------------------------------------------------------------
// CpuFeatures implementation.
void CpuFeatures::ProbeImpl(bool cross_compile) {
// AArch64 has no configuration options, no further probing is required.
supported_ = 0;
// Only use statically determined features for cross compile (snapshot).
if (cross_compile) return;
// We used to probe for coherent cache support, but on older CPUs it
// causes crashes (crbug.com/524337), and newer CPUs don't even have
// the feature any more.
}
void CpuFeatures::PrintTarget() { }
void CpuFeatures::PrintFeatures() {}
// -----------------------------------------------------------------------------
// CPURegList utilities.
CPURegister CPURegList::PopLowestIndex() {
DCHECK(IsValid());
if (IsEmpty()) {
return NoCPUReg;
}
int index = CountTrailingZeros(list_, kRegListSizeInBits);
DCHECK((1 << index) & list_);
Remove(index);
return CPURegister::Create(index, size_, type_);
}
CPURegister CPURegList::PopHighestIndex() {
DCHECK(IsValid());
if (IsEmpty()) {
return NoCPUReg;
}
int index = CountLeadingZeros(list_, kRegListSizeInBits);
index = kRegListSizeInBits - 1 - index;
DCHECK((1 << index) & list_);
Remove(index);
return CPURegister::Create(index, size_, type_);
}
void CPURegList::RemoveCalleeSaved() {
if (type() == CPURegister::kRegister) {
Remove(GetCalleeSaved(RegisterSizeInBits()));
} else if (type() == CPURegister::kVRegister) {
Remove(GetCalleeSavedV(RegisterSizeInBits()));
} else {
DCHECK_EQ(type(), CPURegister::kNoRegister);
DCHECK(IsEmpty());
// The list must already be empty, so do nothing.
}
}
CPURegList CPURegList::GetCalleeSaved(int size) {
return CPURegList(CPURegister::kRegister, size, 19, 29);
}
CPURegList CPURegList::GetCalleeSavedV(int size) {
return CPURegList(CPURegister::kVRegister, size, 8, 15);
}
CPURegList CPURegList::GetCallerSaved(int size) {
// Registers x0-x18 and lr (x30) are caller-saved.
CPURegList list = CPURegList(CPURegister::kRegister, size, 0, 18);
list.Combine(lr);
return list;
}
CPURegList CPURegList::GetCallerSavedV(int size) {
// Registers d0-d7 and d16-d31 are caller-saved.
CPURegList list = CPURegList(CPURegister::kVRegister, size, 0, 7);
list.Combine(CPURegList(CPURegister::kVRegister, size, 16, 31));
return list;
}
// This function defines the list of registers which are associated with a
// safepoint slot. Safepoint register slots are saved contiguously on the stack.
// MacroAssembler::SafepointRegisterStackIndex handles mapping from register
// code to index in the safepoint register slots. Any change here can affect
// this mapping.
CPURegList CPURegList::GetSafepointSavedRegisters() {
CPURegList list = CPURegList::GetCalleeSaved();
list.Combine(
CPURegList(CPURegister::kRegister, kXRegSizeInBits, kJSCallerSaved));
// Note that unfortunately we can't use symbolic names for registers and have
// to directly use register codes. This is because this function is used to
// initialize some static variables and we can't rely on register variables
// to be initialized due to static initialization order issues in C++.
// Drop ip0 and ip1 (i.e. x16 and x17), as they should not be expected to be
// preserved outside of the macro assembler.
list.Remove(16);
list.Remove(17);
// Add x18 to the safepoint list, as although it's not in kJSCallerSaved, it
// is a caller-saved register according to the procedure call standard.
list.Combine(18);
// Add the link register (x30) to the safepoint list.
list.Combine(30);
return list;
}
// -----------------------------------------------------------------------------
// Implementation of RelocInfo
const int RelocInfo::kApplyMask = 1 << RelocInfo::INTERNAL_REFERENCE;
bool RelocInfo::IsCodedSpecially() {
// The deserializer needs to know whether a pointer is specially coded. Being
// specially coded on ARM64 means that it is a movz/movk sequence. We don't
// generate those for relocatable pointers.
return false;
}
bool RelocInfo::IsInConstantPool() {
Instruction* instr = reinterpret_cast<Instruction*>(pc_);
return instr->IsLdrLiteralX();
}
Address RelocInfo::embedded_address() const {
return Memory::Address_at(Assembler::target_pointer_address_at(pc_));
}
uint32_t RelocInfo::embedded_size() const {
return Memory::uint32_at(Assembler::target_pointer_address_at(pc_));
}
void RelocInfo::set_embedded_address(Isolate* isolate, Address address,
ICacheFlushMode flush_mode) {
Assembler::set_target_address_at(isolate, pc_, constant_pool_, address,
flush_mode);
}
void RelocInfo::set_embedded_size(Isolate* isolate, uint32_t size,
ICacheFlushMode flush_mode) {
Memory::uint32_at(Assembler::target_pointer_address_at(pc_)) = size;
// No icache flushing needed, see comment in set_target_address_at.
}
void RelocInfo::set_js_to_wasm_address(Isolate* isolate, Address address,
ICacheFlushMode icache_flush_mode) {
DCHECK_EQ(rmode_, JS_TO_WASM_CALL);
set_embedded_address(isolate, address, icache_flush_mode);
}
Address RelocInfo::js_to_wasm_address() const {
DCHECK_EQ(rmode_, JS_TO_WASM_CALL);
return embedded_address();
}
bool AreAliased(const CPURegister& reg1, const CPURegister& reg2,
const CPURegister& reg3, const CPURegister& reg4,
const CPURegister& reg5, const CPURegister& reg6,
const CPURegister& reg7, const CPURegister& reg8) {
int number_of_valid_regs = 0;
int number_of_valid_fpregs = 0;
RegList unique_regs = 0;
RegList unique_fpregs = 0;
const CPURegister regs[] = {reg1, reg2, reg3, reg4, reg5, reg6, reg7, reg8};
for (unsigned i = 0; i < arraysize(regs); i++) {
if (regs[i].IsRegister()) {
number_of_valid_regs++;
unique_regs |= regs[i].bit();
} else if (regs[i].IsVRegister()) {
number_of_valid_fpregs++;
unique_fpregs |= regs[i].bit();
} else {
DCHECK(!regs[i].IsValid());
}
}
int number_of_unique_regs =
CountSetBits(unique_regs, sizeof(unique_regs) * kBitsPerByte);
int number_of_unique_fpregs =
CountSetBits(unique_fpregs, sizeof(unique_fpregs) * kBitsPerByte);
DCHECK(number_of_valid_regs >= number_of_unique_regs);
DCHECK(number_of_valid_fpregs >= number_of_unique_fpregs);
return (number_of_valid_regs != number_of_unique_regs) ||
(number_of_valid_fpregs != number_of_unique_fpregs);
}
bool AreSameSizeAndType(const CPURegister& reg1, const CPURegister& reg2,
const CPURegister& reg3, const CPURegister& reg4,
const CPURegister& reg5, const CPURegister& reg6,
const CPURegister& reg7, const CPURegister& reg8) {
DCHECK(reg1.IsValid());
bool match = true;
match &= !reg2.IsValid() || reg2.IsSameSizeAndType(reg1);
match &= !reg3.IsValid() || reg3.IsSameSizeAndType(reg1);
match &= !reg4.IsValid() || reg4.IsSameSizeAndType(reg1);
match &= !reg5.IsValid() || reg5.IsSameSizeAndType(reg1);
match &= !reg6.IsValid() || reg6.IsSameSizeAndType(reg1);
match &= !reg7.IsValid() || reg7.IsSameSizeAndType(reg1);
match &= !reg8.IsValid() || reg8.IsSameSizeAndType(reg1);
return match;
}
bool AreSameFormat(const VRegister& reg1, const VRegister& reg2,
const VRegister& reg3, const VRegister& reg4) {
DCHECK(reg1.IsValid());
return (!reg2.IsValid() || reg2.IsSameFormat(reg1)) &&
(!reg3.IsValid() || reg3.IsSameFormat(reg1)) &&
(!reg4.IsValid() || reg4.IsSameFormat(reg1));
}
bool AreConsecutive(const VRegister& reg1, const VRegister& reg2,
const VRegister& reg3, const VRegister& reg4) {
DCHECK(reg1.IsValid());
if (!reg2.IsValid()) {
DCHECK(!reg3.IsValid() && !reg4.IsValid());
return true;
} else if (reg2.code() != ((reg1.code() + 1) % kNumberOfVRegisters)) {
return false;
}
if (!reg3.IsValid()) {
DCHECK(!reg4.IsValid());
return true;
} else if (reg3.code() != ((reg2.code() + 1) % kNumberOfVRegisters)) {
return false;
}
if (!reg4.IsValid()) {
return true;
} else if (reg4.code() != ((reg3.code() + 1) % kNumberOfVRegisters)) {
return false;
}
return true;
}
void Immediate::InitializeHandle(Handle<HeapObject> handle) {
value_ = reinterpret_cast<intptr_t>(handle.address());
rmode_ = RelocInfo::EMBEDDED_OBJECT;
}
bool Operand::NeedsRelocation(const Assembler* assembler) const {
RelocInfo::Mode rmode = immediate_.rmode();
if (rmode == RelocInfo::EXTERNAL_REFERENCE) {
return assembler->serializer_enabled();
}
return !RelocInfo::IsNone(rmode);
}
bool ConstPool::AddSharedEntry(SharedEntryMap& entry_map, uint64_t data,
int offset) {
auto existing = entry_map.find(data);
if (existing == entry_map.end()) {
entry_map[data] = static_cast<int>(entries_.size());
entries_.push_back(std::make_pair(data, std::vector<int>(1, offset)));
return true;
}
int index = existing->second;
entries_[index].second.push_back(offset);
return false;
}
// Constant Pool.
bool ConstPool::RecordEntry(intptr_t data, RelocInfo::Mode mode) {
DCHECK(mode != RelocInfo::COMMENT && mode != RelocInfo::CONST_POOL &&
mode != RelocInfo::VENEER_POOL &&
mode != RelocInfo::DEOPT_SCRIPT_OFFSET &&
mode != RelocInfo::DEOPT_INLINING_ID &&
mode != RelocInfo::DEOPT_REASON && mode != RelocInfo::DEOPT_ID);
bool write_reloc_info = true;
uint64_t raw_data = static_cast<uint64_t>(data);
int offset = assm_->pc_offset();
if (IsEmpty()) {
first_use_ = offset;
}
if (CanBeShared(mode)) {
write_reloc_info = AddSharedEntry(shared_entries_, raw_data, offset);
} else if (mode == RelocInfo::CODE_TARGET &&
assm_->IsCodeTargetSharingAllowed() && raw_data != 0) {
// A zero data value is a placeholder and must not be shared.
write_reloc_info = AddSharedEntry(handle_to_index_map_, raw_data, offset);
} else {
entries_.push_back(std::make_pair(raw_data, std::vector<int>(1, offset)));
}
if (EntryCount() > Assembler::kApproxMaxPoolEntryCount) {
// Request constant pool emission after the next instruction.
assm_->SetNextConstPoolCheckIn(1);
}
return write_reloc_info;
}
int ConstPool::DistanceToFirstUse() {
DCHECK_GE(first_use_, 0);
return assm_->pc_offset() - first_use_;
}
int ConstPool::MaxPcOffset() {
// There are no pending entries in the pool so we can never get out of
// range.
if (IsEmpty()) return kMaxInt;
// Entries are not necessarily emitted in the order they are added so in the
// worst case the first constant pool use will be accessing the last entry.
return first_use_ + kMaxLoadLiteralRange - WorstCaseSize();
}
int ConstPool::WorstCaseSize() {
if (IsEmpty()) return 0;
// Max size prologue:
// b over
// ldr xzr, #pool_size
// blr xzr
// nop
// All entries are 64-bit for now.
return 4 * kInstructionSize + EntryCount() * kPointerSize;
}
int ConstPool::SizeIfEmittedAtCurrentPc(bool require_jump) {
if (IsEmpty()) return 0;
// Prologue is:
// b over ;; if require_jump
// ldr xzr, #pool_size
// blr xzr
// nop ;; if not 64-bit aligned
int prologue_size = require_jump ? kInstructionSize : 0;
prologue_size += 2 * kInstructionSize;
prologue_size += IsAligned(assm_->pc_offset() + prologue_size, 8) ?
0 : kInstructionSize;
// All entries are 64-bit for now.
return prologue_size + EntryCount() * kPointerSize;
}
void ConstPool::Emit(bool require_jump) {
DCHECK(!assm_->is_const_pool_blocked());
// Prevent recursive pool emission and protect from veneer pools.
Assembler::BlockPoolsScope block_pools(assm_);
int size = SizeIfEmittedAtCurrentPc(require_jump);
Label size_check;
assm_->bind(&size_check);
assm_->RecordConstPool(size);
// Emit the constant pool. It is preceded by an optional branch if
// require_jump and a header which will:
// 1) Encode the size of the constant pool, for use by the disassembler.
// 2) Terminate the program, to try to prevent execution from accidentally
// flowing into the constant pool.
// 3) align the pool entries to 64-bit.
// The header is therefore made of up to three arm64 instructions:
// ldr xzr, #<size of the constant pool in 32-bit words>
// blr xzr
// nop
//
// If executed, the header will likely segfault and lr will point to the
// instruction following the offending blr.
// TODO(all): Make the alignment part less fragile. Currently code is
// allocated as a byte array so there are no guarantees the alignment will
// be preserved on compaction. Currently it works as allocation seems to be
// 64-bit aligned.
// Emit branch if required
Label after_pool;
if (require_jump) {
assm_->b(&after_pool);
}
// Emit the header.
assm_->RecordComment("[ Constant Pool");
EmitMarker();
EmitGuard();
assm_->Align(8);
// Emit constant pool entries.
// TODO(all): currently each relocated constant is 64 bits, consider adding
// support for 32-bit entries.
EmitEntries();
assm_->RecordComment("]");
if (after_pool.is_linked()) {
assm_->bind(&after_pool);
}
DCHECK(assm_->SizeOfCodeGeneratedSince(&size_check) ==
static_cast<unsigned>(size));
}
void ConstPool::Clear() {
shared_entries_.clear();
handle_to_index_map_.clear();
entries_.clear();
first_use_ = -1;
}
bool ConstPool::CanBeShared(RelocInfo::Mode mode) {
// Constant pool currently does not support 32-bit entries.
DCHECK(mode != RelocInfo::NONE32);
return RelocInfo::IsNone(mode) ||
(mode >= RelocInfo::FIRST_SHAREABLE_RELOC_MODE);
}
void ConstPool::EmitMarker() {
// A constant pool size is expressed in number of 32-bits words.
// Currently all entries are 64-bit.
// + 1 is for the crash guard.
// + 0/1 for alignment.
int word_count = EntryCount() * 2 + 1 +
(IsAligned(assm_->pc_offset(), 8) ? 0 : 1);
assm_->Emit(LDR_x_lit |
Assembler::ImmLLiteral(word_count) |
Assembler::Rt(xzr));
}
MemOperand::PairResult MemOperand::AreConsistentForPair(
const MemOperand& operandA,
const MemOperand& operandB,
int access_size_log2) {
DCHECK_GE(access_size_log2, 0);
DCHECK_LE(access_size_log2, 3);
// Step one: check that they share the same base, that the mode is Offset
// and that the offset is a multiple of access size.
if (!operandA.base().Is(operandB.base()) ||
(operandA.addrmode() != Offset) ||
(operandB.addrmode() != Offset) ||
((operandA.offset() & ((1 << access_size_log2) - 1)) != 0)) {
return kNotPair;
}
// Step two: check that the offsets are contiguous and that the range
// is OK for ldp/stp.
if ((operandB.offset() == operandA.offset() + (1 << access_size_log2)) &&
is_int7(operandA.offset() >> access_size_log2)) {
return kPairAB;
}
if ((operandA.offset() == operandB.offset() + (1 << access_size_log2)) &&
is_int7(operandB.offset() >> access_size_log2)) {
return kPairBA;
}
return kNotPair;
}
void ConstPool::EmitGuard() {
#ifdef DEBUG
Instruction* instr = reinterpret_cast<Instruction*>(assm_->pc());
DCHECK(instr->preceding()->IsLdrLiteralX() &&
instr->preceding()->Rt() == xzr.code());
#endif
assm_->EmitPoolGuard();
}
void ConstPool::EmitEntries() {
DCHECK(IsAligned(assm_->pc_offset(), 8));
// Emit entries.
for (const auto& entry : entries_) {
for (const auto& pc : entry.second) {
Instruction* instr = assm_->InstructionAt(pc);
// Instruction to patch must be 'ldr rd, [pc, #offset]' with offset == 0.
DCHECK(instr->IsLdrLiteral() && instr->ImmLLiteral() == 0);
instr->SetImmPCOffsetTarget(assm_->isolate_data(), assm_->pc());
}
assm_->dc64(entry.first);
}
Clear();
}
// Assembler
Assembler::Assembler(IsolateData isolate_data, void* buffer, int buffer_size)
: AssemblerBase(isolate_data, buffer, buffer_size),
constpool_(this),
unresolved_branches_() {
const_pool_blocked_nesting_ = 0;
veneer_pool_blocked_nesting_ = 0;
code_target_sharing_blocked_nesting_ = 0;
Reset();
}
Assembler::~Assembler() {
DCHECK(constpool_.IsEmpty());
DCHECK_EQ(const_pool_blocked_nesting_, 0);
DCHECK_EQ(veneer_pool_blocked_nesting_, 0);
DCHECK_EQ(code_target_sharing_blocked_nesting_, 0);
}
void Assembler::Reset() {
#ifdef DEBUG
DCHECK((pc_ >= buffer_) && (pc_ < buffer_ + buffer_size_));
DCHECK_EQ(const_pool_blocked_nesting_, 0);
DCHECK_EQ(veneer_pool_blocked_nesting_, 0);
DCHECK_EQ(code_target_sharing_blocked_nesting_, 0);
DCHECK(unresolved_branches_.empty());
memset(buffer_, 0, pc_ - buffer_);
#endif
pc_ = buffer_;
reloc_info_writer.Reposition(reinterpret_cast<byte*>(buffer_ + buffer_size_),
reinterpret_cast<byte*>(pc_));
constpool_.Clear();
next_constant_pool_check_ = 0;
next_veneer_pool_check_ = kMaxInt;
no_const_pool_before_ = 0;
}
void Assembler::AllocateAndInstallRequestedHeapObjects(Isolate* isolate) {
for (auto& request : heap_object_requests_) {
Handle<HeapObject> object;
switch (request.kind()) {
case HeapObjectRequest::kHeapNumber:
object = isolate->factory()->NewHeapNumber(request.heap_number(),
IMMUTABLE, TENURED);
break;
case HeapObjectRequest::kCodeStub:
request.code_stub()->set_isolate(isolate);
object = request.code_stub()->GetCode();
break;
}
Address pc = buffer_ + request.offset();
Memory::Address_at(target_pointer_address_at(pc)) = object.address();
}
}
void Assembler::GetCode(Isolate* isolate, CodeDesc* desc) {
// Emit constant pool if necessary.
CheckConstPool(true, false);
DCHECK(constpool_.IsEmpty());
AllocateAndInstallRequestedHeapObjects(isolate);
// Set up code descriptor.
if (desc) {
desc->buffer = reinterpret_cast<byte*>(buffer_);
desc->buffer_size = buffer_size_;
desc->instr_size = pc_offset();
desc->reloc_size =
static_cast<int>((reinterpret_cast<byte*>(buffer_) + buffer_size_) -
reloc_info_writer.pos());
desc->origin = this;
desc->constant_pool_size = 0;
desc->unwinding_info_size = 0;
desc->unwinding_info = nullptr;
}
}
void Assembler::Align(int m) {
DCHECK(m >= 4 && base::bits::IsPowerOfTwo(m));
while ((pc_offset() & (m - 1)) != 0) {
nop();
}
}
void Assembler::CheckLabelLinkChain(Label const * label) {
#ifdef DEBUG
if (label->is_linked()) {
static const int kMaxLinksToCheck = 64; // Avoid O(n2) behaviour.
int links_checked = 0;
int64_t linkoffset = label->pos();
bool end_of_chain = false;
while (!end_of_chain) {
if (++links_checked > kMaxLinksToCheck) break;
Instruction * link = InstructionAt(linkoffset);
int64_t linkpcoffset = link->ImmPCOffset();
int64_t prevlinkoffset = linkoffset + linkpcoffset;
end_of_chain = (linkoffset == prevlinkoffset);
linkoffset = linkoffset + linkpcoffset;
}
}
#endif
}
void Assembler::RemoveBranchFromLabelLinkChain(Instruction* branch,
Label* label,
Instruction* label_veneer) {
DCHECK(label->is_linked());
CheckLabelLinkChain(label);
Instruction* link = InstructionAt(label->pos());
Instruction* prev_link = link;
Instruction* next_link;
bool end_of_chain = false;
while (link != branch && !end_of_chain) {
next_link = link->ImmPCOffsetTarget();
end_of_chain = (link == next_link);
prev_link = link;
link = next_link;
}
DCHECK(branch == link);
next_link = branch->ImmPCOffsetTarget();
if (branch == prev_link) {
// The branch is the first instruction in the chain.
if (branch == next_link) {
// It is also the last instruction in the chain, so it is the only branch
// currently referring to this label.
label->Unuse();
} else {
label->link_to(
static_cast<int>(reinterpret_cast<byte*>(next_link) - buffer_));
}
} else if (branch == next_link) {
// The branch is the last (but not also the first) instruction in the chain.
prev_link->SetImmPCOffsetTarget(isolate_data(), prev_link);
} else {
// The branch is in the middle of the chain.
if (prev_link->IsTargetInImmPCOffsetRange(next_link)) {
prev_link->SetImmPCOffsetTarget(isolate_data(), next_link);
} else if (label_veneer != nullptr) {
// Use the veneer for all previous links in the chain.
prev_link->SetImmPCOffsetTarget(isolate_data(), prev_link);
end_of_chain = false;
link = next_link;
while (!end_of_chain) {
next_link = link->ImmPCOffsetTarget();
end_of_chain = (link == next_link);
link->SetImmPCOffsetTarget(isolate_data(), label_veneer);
link = next_link;
}
} else {
// The assert below will fire.
// Some other work could be attempted to fix up the chain, but it would be
// rather complicated. If we crash here, we may want to consider using an
// other mechanism than a chain of branches.
//
// Note that this situation currently should not happen, as we always call
// this function with a veneer to the target label.
// However this could happen with a MacroAssembler in the following state:
// [previous code]
// B(label);
// [20KB code]
// Tbz(label); // First tbz. Pointing to unconditional branch.
// [20KB code]
// Tbz(label); // Second tbz. Pointing to the first tbz.
// [more code]
// and this function is called to remove the first tbz from the label link
// chain. Since tbz has a range of +-32KB, the second tbz cannot point to
// the unconditional branch.
CHECK(prev_link->IsTargetInImmPCOffsetRange(next_link));
UNREACHABLE();
}
}
CheckLabelLinkChain(label);
}
void Assembler::bind(Label* label) {
// Bind label to the address at pc_. All instructions (most likely branches)
// that are linked to this label will be updated to point to the newly-bound
// label.
DCHECK(!label->is_near_linked());
DCHECK(!label->is_bound());
DeleteUnresolvedBranchInfoForLabel(label);
// If the label is linked, the link chain looks something like this:
//
// |--I----I-------I-------L
// |---------------------->| pc_offset
// |-------------->| linkoffset = label->pos()
// |<------| link->ImmPCOffset()
// |------>| prevlinkoffset = linkoffset + link->ImmPCOffset()
//
// On each iteration, the last link is updated and then removed from the
// chain until only one remains. At that point, the label is bound.
//
// If the label is not linked, no preparation is required before binding.
while (label->is_linked()) {
int linkoffset = label->pos();
Instruction* link = InstructionAt(linkoffset);
int prevlinkoffset = linkoffset + static_cast<int>(link->ImmPCOffset());
CheckLabelLinkChain(label);
DCHECK_GE(linkoffset, 0);
DCHECK(linkoffset < pc_offset());
DCHECK((linkoffset > prevlinkoffset) ||
(linkoffset - prevlinkoffset == kStartOfLabelLinkChain));
DCHECK_GE(prevlinkoffset, 0);
// Update the link to point to the label.
if (link->IsUnresolvedInternalReference()) {
// Internal references do not get patched to an instruction but directly
// to an address.
internal_reference_positions_.push_back(linkoffset);
PatchingAssembler patcher(isolate_data(), reinterpret_cast<byte*>(link),
2);
patcher.dc64(reinterpret_cast<uintptr_t>(pc_));
} else {
link->SetImmPCOffsetTarget(isolate_data(),
reinterpret_cast<Instruction*>(pc_));
}
// Link the label to the previous link in the chain.
if (linkoffset - prevlinkoffset == kStartOfLabelLinkChain) {
// We hit kStartOfLabelLinkChain, so the chain is fully processed.
label->Unuse();
} else {
// Update the label for the next iteration.
label->link_to(prevlinkoffset);
}
}
label->bind_to(pc_offset());
DCHECK(label->is_bound());
DCHECK(!label->is_linked());
}
int Assembler::LinkAndGetByteOffsetTo(Label* label) {
DCHECK_EQ(sizeof(*pc_), 1);
CheckLabelLinkChain(label);
int offset;
if (label->is_bound()) {
// The label is bound, so it does not need to be updated. Referring
// instructions must link directly to the label as they will not be
// updated.
//
// In this case, label->pos() returns the offset of the label from the
// start of the buffer.
//
// Note that offset can be zero for self-referential instructions. (This
// could be useful for ADR, for example.)
offset = label->pos() - pc_offset();
DCHECK_LE(offset, 0);
} else {
if (label->is_linked()) {
// The label is linked, so the referring instruction should be added onto
// the end of the label's link chain.
//
// In this case, label->pos() returns the offset of the last linked
// instruction from the start of the buffer.
offset = label->pos() - pc_offset();
DCHECK_NE(offset, kStartOfLabelLinkChain);
// Note that the offset here needs to be PC-relative only so that the
// first instruction in a buffer can link to an unbound label. Otherwise,
// the offset would be 0 for this case, and 0 is reserved for
// kStartOfLabelLinkChain.
} else {
// The label is unused, so it now becomes linked and the referring
// instruction is at the start of the new link chain.
offset = kStartOfLabelLinkChain;
}
// The instruction at pc is now the last link in the label's chain.
label->link_to(pc_offset());
}
return offset;
}
void Assembler::DeleteUnresolvedBranchInfoForLabelTraverse(Label* label) {
DCHECK(label->is_linked());
CheckLabelLinkChain(label);
int link_offset = label->pos();
int link_pcoffset;
bool end_of_chain = false;
while (!end_of_chain) {
Instruction * link = InstructionAt(link_offset);
link_pcoffset = static_cast<int>(link->ImmPCOffset());
// ADR instructions are not handled by veneers.
if (link->IsImmBranch()) {
int max_reachable_pc =
static_cast<int>(InstructionOffset(link) +
Instruction::ImmBranchRange(link->BranchType()));
typedef std::multimap<int, FarBranchInfo>::iterator unresolved_info_it;
std::pair<unresolved_info_it, unresolved_info_it> range;
range = unresolved_branches_.equal_range(max_reachable_pc);
unresolved_info_it it;
for (it = range.first; it != range.second; ++it) {
if (it->second.pc_offset_ == link_offset) {
unresolved_branches_.erase(it);
break;
}
}
}
end_of_chain = (link_pcoffset == 0);
link_offset = link_offset + link_pcoffset;
}
}
void Assembler::DeleteUnresolvedBranchInfoForLabel(Label* label) {
if (unresolved_branches_.empty()) {
DCHECK_EQ(next_veneer_pool_check_, kMaxInt);
return;
}
if (label->is_linked()) {
// Branches to this label will be resolved when the label is bound, normally
// just after all the associated info has been deleted.
DeleteUnresolvedBranchInfoForLabelTraverse(label);
}
if (unresolved_branches_.empty()) {
next_veneer_pool_check_ = kMaxInt;
} else {
next_veneer_pool_check_ =
unresolved_branches_first_limit() - kVeneerDistanceCheckMargin;
}
}
void Assembler::StartBlockConstPool() {
if (const_pool_blocked_nesting_++ == 0) {
// Prevent constant pool checks happening by setting the next check to
// the biggest possible offset.
next_constant_pool_check_ = kMaxInt;
}
}
void Assembler::EndBlockConstPool() {
if (--const_pool_blocked_nesting_ == 0) {
// Check the constant pool hasn't been blocked for too long.
DCHECK(pc_offset() < constpool_.MaxPcOffset());
// Two cases:
// * no_const_pool_before_ >= next_constant_pool_check_ and the emission is
// still blocked
// * no_const_pool_before_ < next_constant_pool_check_ and the next emit
// will trigger a check.
next_constant_pool_check_ = no_const_pool_before_;
}
}
bool Assembler::is_const_pool_blocked() const {
return (const_pool_blocked_nesting_ > 0) ||
(pc_offset() < no_const_pool_before_);
}
bool Assembler::IsConstantPoolAt(Instruction* instr) {
// The constant pool marker is made of two instructions. These instructions
// will never be emitted by the JIT, so checking for the first one is enough:
// 0: ldr xzr, #<size of pool>
bool result = instr->IsLdrLiteralX() && (instr->Rt() == kZeroRegCode);
// It is still worth asserting the marker is complete.
// 4: blr xzr
DCHECK(!result || (instr->following()->IsBranchAndLinkToRegister() &&
instr->following()->Rn() == kZeroRegCode));
return result;
}
int Assembler::ConstantPoolSizeAt(Instruction* instr) {
#ifdef USE_SIMULATOR
// Assembler::debug() embeds constants directly into the instruction stream.
// Although this is not a genuine constant pool, treat it like one to avoid
// disassembling the constants.
if ((instr->Mask(ExceptionMask) == HLT) &&
(instr->ImmException() == kImmExceptionIsDebug)) {
const char* message =
reinterpret_cast<const char*>(
instr->InstructionAtOffset(kDebugMessageOffset));
int size = static_cast<int>(kDebugMessageOffset + strlen(message) + 1);
return RoundUp(size, kInstructionSize) / kInstructionSize;
}
// Same for printf support, see MacroAssembler::CallPrintf().
if ((instr->Mask(ExceptionMask) == HLT) &&
(instr->ImmException() == kImmExceptionIsPrintf)) {
return kPrintfLength / kInstructionSize;
}
#endif
if (IsConstantPoolAt(instr)) {
return instr->ImmLLiteral();
} else {
return -1;
}
}
void Assembler::EmitPoolGuard() {
// We must generate only one instruction as this is used in scopes that
// control the size of the code generated.
Emit(BLR | Rn(xzr));
}
void Assembler::StartBlockVeneerPool() {
++veneer_pool_blocked_nesting_;
}
void Assembler::EndBlockVeneerPool() {
if (--veneer_pool_blocked_nesting_ == 0) {
// Check the veneer pool hasn't been blocked for too long.
DCHECK(unresolved_branches_.empty() ||
(pc_offset() < unresolved_branches_first_limit()));
}
}
void Assembler::br(const Register& xn) {
DCHECK(xn.Is64Bits());
Emit(BR | Rn(xn));
}
void Assembler::blr(const Register& xn) {
DCHECK(xn.Is64Bits());
// The pattern 'blr xzr' is used as a guard to detect when execution falls
// through the constant pool. It should not be emitted.
DCHECK(!xn.Is(xzr));
Emit(BLR | Rn(xn));
}
void Assembler::ret(const Register& xn) {
DCHECK(xn.Is64Bits());
Emit(RET | Rn(xn));
}
void Assembler::b(int imm26) {
Emit(B | ImmUncondBranch(imm26));
}
void Assembler::b(Label* label) {
b(LinkAndGetInstructionOffsetTo(label));
}
void Assembler::b(int imm19, Condition cond) {
Emit(B_cond | ImmCondBranch(imm19) | cond);
}
void Assembler::b(Label* label, Condition cond) {
b(LinkAndGetInstructionOffsetTo(label), cond);
}
void Assembler::bl(int imm26) {
Emit(BL | ImmUncondBranch(imm26));
}
void Assembler::bl(Label* label) {
bl(LinkAndGetInstructionOffsetTo(label));
}
void Assembler::cbz(const Register& rt,
int imm19) {
Emit(SF(rt) | CBZ | ImmCmpBranch(imm19) | Rt(rt));
}
void Assembler::cbz(const Register& rt,
Label* label) {
cbz(rt, LinkAndGetInstructionOffsetTo(label));
}
void Assembler::cbnz(const Register& rt,
int imm19) {
Emit(SF(rt) | CBNZ | ImmCmpBranch(imm19) | Rt(rt));
}
void Assembler::cbnz(const Register& rt,
Label* label) {
cbnz(rt, LinkAndGetInstructionOffsetTo(label));
}
void Assembler::tbz(const Register& rt,
unsigned bit_pos,
int imm14) {
DCHECK(rt.Is64Bits() || (rt.Is32Bits() && (bit_pos < kWRegSizeInBits)));
Emit(TBZ | ImmTestBranchBit(bit_pos) | ImmTestBranch(imm14) | Rt(rt));
}
void Assembler::tbz(const Register& rt,
unsigned bit_pos,
Label* label) {
tbz(rt, bit_pos, LinkAndGetInstructionOffsetTo(label));
}
void Assembler::tbnz(const Register& rt,
unsigned bit_pos,
int imm14) {
DCHECK(rt.Is64Bits() || (rt.Is32Bits() && (bit_pos < kWRegSizeInBits)));
Emit(TBNZ | ImmTestBranchBit(bit_pos) | ImmTestBranch(imm14) | Rt(rt));
}
void Assembler::tbnz(const Register& rt,
unsigned bit_pos,
Label* label) {
tbnz(rt, bit_pos, LinkAndGetInstructionOffsetTo(label));
}
void Assembler::adr(const Register& rd, int imm21) {
DCHECK(rd.Is64Bits());
Emit(ADR | ImmPCRelAddress(imm21) | Rd(rd));
}
void Assembler::adr(const Register& rd, Label* label) {
adr(rd, LinkAndGetByteOffsetTo(label));
}
void Assembler::add(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSub(rd, rn, operand, LeaveFlags, ADD);
}
void Assembler::adds(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSub(rd, rn, operand, SetFlags, ADD);
}
void Assembler::cmn(const Register& rn,
const Operand& operand) {
Register zr = AppropriateZeroRegFor(rn);
adds(zr, rn, operand);
}
void Assembler::sub(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSub(rd, rn, operand, LeaveFlags, SUB);
}
void Assembler::subs(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSub(rd, rn, operand, SetFlags, SUB);
}
void Assembler::cmp(const Register& rn, const Operand& operand) {
Register zr = AppropriateZeroRegFor(rn);
subs(zr, rn, operand);
}
void Assembler::neg(const Register& rd, const Operand& operand) {
Register zr = AppropriateZeroRegFor(rd);
sub(rd, zr, operand);
}
void Assembler::negs(const Register& rd, const Operand& operand) {
Register zr = AppropriateZeroRegFor(rd);
subs(rd, zr, operand);
}
void Assembler::adc(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSubWithCarry(rd, rn, operand, LeaveFlags, ADC);
}
void Assembler::adcs(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSubWithCarry(rd, rn, operand, SetFlags, ADC);
}
void Assembler::sbc(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSubWithCarry(rd, rn, operand, LeaveFlags, SBC);
}
void Assembler::sbcs(const Register& rd,
const Register& rn,
const Operand& operand) {
AddSubWithCarry(rd, rn, operand, SetFlags, SBC);
}
void Assembler::ngc(const Register& rd, const Operand& operand) {
Register zr = AppropriateZeroRegFor(rd);
sbc(rd, zr, operand);
}
void Assembler::ngcs(const Register& rd, const Operand& operand) {
Register zr = AppropriateZeroRegFor(rd);
sbcs(rd, zr, operand);
}
// Logical instructions.
void Assembler::and_(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, AND);
}
void Assembler::ands(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, ANDS);
}
void Assembler::tst(const Register& rn,
const Operand& operand) {
ands(AppropriateZeroRegFor(rn), rn, operand);
}
void Assembler::bic(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, BIC);
}
void Assembler::bics(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, BICS);
}
void Assembler::orr(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, ORR);
}
void Assembler::orn(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, ORN);
}
void Assembler::eor(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, EOR);
}
void Assembler::eon(const Register& rd,
const Register& rn,
const Operand& operand) {
Logical(rd, rn, operand, EON);
}
void Assembler::lslv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | LSLV | Rm(rm) | Rn(rn) | Rd(rd));
}
void Assembler::lsrv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | LSRV | Rm(rm) | Rn(rn) | Rd(rd));
}
void Assembler::asrv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | ASRV | Rm(rm) | Rn(rn) | Rd(rd));
}
void Assembler::rorv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | RORV | Rm(rm) | Rn(rn) | Rd(rd));
}
// Bitfield operations.
void Assembler::bfm(const Register& rd, const Register& rn, int immr,
int imms) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
Instr N = SF(rd) >> (kSFOffset - kBitfieldNOffset);
Emit(SF(rd) | BFM | N |
ImmR(immr, rd.SizeInBits()) |
ImmS(imms, rn.SizeInBits()) |
Rn(rn) | Rd(rd));
}
void Assembler::sbfm(const Register& rd, const Register& rn, int immr,
int imms) {
DCHECK(rd.Is64Bits() || rn.Is32Bits());
Instr N = SF(rd) >> (kSFOffset - kBitfieldNOffset);
Emit(SF(rd) | SBFM | N |
ImmR(immr, rd.SizeInBits()) |
ImmS(imms, rn.SizeInBits()) |
Rn(rn) | Rd(rd));
}
void Assembler::ubfm(const Register& rd, const Register& rn, int immr,
int imms) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
Instr N = SF(rd) >> (kSFOffset - kBitfieldNOffset);
Emit(SF(rd) | UBFM | N |
ImmR(immr, rd.SizeInBits()) |
ImmS(imms, rn.SizeInBits()) |
Rn(rn) | Rd(rd));
}
void Assembler::extr(const Register& rd, const Register& rn, const Register& rm,
int lsb) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Instr N = SF(rd) >> (kSFOffset - kBitfieldNOffset);
Emit(SF(rd) | EXTR | N | Rm(rm) |
ImmS(lsb, rn.SizeInBits()) | Rn(rn) | Rd(rd));
}
void Assembler::csel(const Register& rd,
const Register& rn,
const Register& rm,
Condition cond) {
ConditionalSelect(rd, rn, rm, cond, CSEL);
}
void Assembler::csinc(const Register& rd,
const Register& rn,
const Register& rm,
Condition cond) {
ConditionalSelect(rd, rn, rm, cond, CSINC);
}
void Assembler::csinv(const Register& rd,
const Register& rn,
const Register& rm,
Condition cond) {
ConditionalSelect(rd, rn, rm, cond, CSINV);
}
void Assembler::csneg(const Register& rd,
const Register& rn,
const Register& rm,
Condition cond) {
ConditionalSelect(rd, rn, rm, cond, CSNEG);
}
void Assembler::cset(const Register &rd, Condition cond) {
DCHECK((cond != al) && (cond != nv));
Register zr = AppropriateZeroRegFor(rd);
csinc(rd, zr, zr, NegateCondition(cond));
}
void Assembler::csetm(const Register &rd, Condition cond) {
DCHECK((cond != al) && (cond != nv));
Register zr = AppropriateZeroRegFor(rd);
csinv(rd, zr, zr, NegateCondition(cond));
}
void Assembler::cinc(const Register &rd, const Register &rn, Condition cond) {
DCHECK((cond != al) && (cond != nv));
csinc(rd, rn, rn, NegateCondition(cond));
}
void Assembler::cinv(const Register &rd, const Register &rn, Condition cond) {
DCHECK((cond != al) && (cond != nv));
csinv(rd, rn, rn, NegateCondition(cond));
}
void Assembler::cneg(const Register &rd, const Register &rn, Condition cond) {
DCHECK((cond != al) && (cond != nv));
csneg(rd, rn, rn, NegateCondition(cond));
}
void Assembler::ConditionalSelect(const Register& rd,
const Register& rn,
const Register& rm,
Condition cond,
ConditionalSelectOp op) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | op | Rm(rm) | Cond(cond) | Rn(rn) | Rd(rd));
}
void Assembler::ccmn(const Register& rn,
const Operand& operand,
StatusFlags nzcv,
Condition cond) {
ConditionalCompare(rn, operand, nzcv, cond, CCMN);
}
void Assembler::ccmp(const Register& rn,
const Operand& operand,
StatusFlags nzcv,
Condition cond) {
ConditionalCompare(rn, operand, nzcv, cond, CCMP);
}
void Assembler::DataProcessing3Source(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra,
DataProcessing3SourceOp op) {
Emit(SF(rd) | op | Rm(rm) | Ra(ra) | Rn(rn) | Rd(rd));
}
void Assembler::mul(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(AreSameSizeAndType(rd, rn, rm));
Register zr = AppropriateZeroRegFor(rn);
DataProcessing3Source(rd, rn, rm, zr, MADD);
}
void Assembler::madd(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(AreSameSizeAndType(rd, rn, rm, ra));
DataProcessing3Source(rd, rn, rm, ra, MADD);
}
void Assembler::mneg(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(AreSameSizeAndType(rd, rn, rm));
Register zr = AppropriateZeroRegFor(rn);
DataProcessing3Source(rd, rn, rm, zr, MSUB);
}
void Assembler::msub(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(AreSameSizeAndType(rd, rn, rm, ra));
DataProcessing3Source(rd, rn, rm, ra, MSUB);
}
void Assembler::smaddl(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(rd.Is64Bits() && ra.Is64Bits());
DCHECK(rn.Is32Bits() && rm.Is32Bits());
DataProcessing3Source(rd, rn, rm, ra, SMADDL_x);
}
void Assembler::smsubl(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(rd.Is64Bits() && ra.Is64Bits());
DCHECK(rn.Is32Bits() && rm.Is32Bits());
DataProcessing3Source(rd, rn, rm, ra, SMSUBL_x);
}
void Assembler::umaddl(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(rd.Is64Bits() && ra.Is64Bits());
DCHECK(rn.Is32Bits() && rm.Is32Bits());
DataProcessing3Source(rd, rn, rm, ra, UMADDL_x);
}
void Assembler::umsubl(const Register& rd,
const Register& rn,
const Register& rm,
const Register& ra) {
DCHECK(rd.Is64Bits() && ra.Is64Bits());
DCHECK(rn.Is32Bits() && rm.Is32Bits());
DataProcessing3Source(rd, rn, rm, ra, UMSUBL_x);
}
void Assembler::smull(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.Is64Bits());
DCHECK(rn.Is32Bits() && rm.Is32Bits());
DataProcessing3Source(rd, rn, rm, xzr, SMADDL_x);
}
void Assembler::smulh(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(AreSameSizeAndType(rd, rn, rm));
DataProcessing3Source(rd, rn, rm, xzr, SMULH_x);
}
void Assembler::sdiv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | SDIV | Rm(rm) | Rn(rn) | Rd(rd));
}
void Assembler::udiv(const Register& rd,
const Register& rn,
const Register& rm) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(rd.SizeInBits() == rm.SizeInBits());
Emit(SF(rd) | UDIV | Rm(rm) | Rn(rn) | Rd(rd));
}
void Assembler::rbit(const Register& rd,
const Register& rn) {
DataProcessing1Source(rd, rn, RBIT);
}
void Assembler::rev16(const Register& rd,
const Register& rn) {
DataProcessing1Source(rd, rn, REV16);
}
void Assembler::rev32(const Register& rd,
const Register& rn) {
DCHECK(rd.Is64Bits());
DataProcessing1Source(rd, rn, REV);
}
void Assembler::rev(const Register& rd,
const Register& rn) {
DataProcessing1Source(rd, rn, rd.Is64Bits() ? REV_x : REV_w);
}
void Assembler::clz(const Register& rd,
const Register& rn) {
DataProcessing1Source(rd, rn, CLZ);
}
void Assembler::cls(const Register& rd,
const Register& rn) {
DataProcessing1Source(rd, rn, CLS);
}
void Assembler::ldp(const CPURegister& rt,
const CPURegister& rt2,
const MemOperand& src) {
LoadStorePair(rt, rt2, src, LoadPairOpFor(rt, rt2));
}
void Assembler::stp(const CPURegister& rt,
const CPURegister& rt2,
const MemOperand& dst) {
LoadStorePair(rt, rt2, dst, StorePairOpFor(rt, rt2));
}
void Assembler::ldpsw(const Register& rt,
const Register& rt2,
const MemOperand& src) {
DCHECK(rt.Is64Bits());
LoadStorePair(rt, rt2, src, LDPSW_x);
}
void Assembler::LoadStorePair(const CPURegister& rt,
const CPURegister& rt2,
const MemOperand& addr,
LoadStorePairOp op) {
// 'rt' and 'rt2' can only be aliased for stores.
DCHECK(((op & LoadStorePairLBit) == 0) || !rt.Is(rt2));
DCHECK(AreSameSizeAndType(rt, rt2));
DCHECK(IsImmLSPair(addr.offset(), CalcLSPairDataSize(op)));
int offset = static_cast<int>(addr.offset());
Instr memop = op | Rt(rt) | Rt2(rt2) | RnSP(addr.base()) |
ImmLSPair(offset, CalcLSPairDataSize(op));
Instr addrmodeop;
if (addr.IsImmediateOffset()) {
addrmodeop = LoadStorePairOffsetFixed;
} else {
// Pre-index and post-index modes.
DCHECK(!rt.Is(addr.base()));
DCHECK(!rt2.Is(addr.base()));
DCHECK_NE(addr.offset(), 0);
if (addr.IsPreIndex()) {
addrmodeop = LoadStorePairPreIndexFixed;
} else {
DCHECK(addr.IsPostIndex());
addrmodeop = LoadStorePairPostIndexFixed;
}
}
Emit(addrmodeop | memop);
}
// Memory instructions.
void Assembler::ldrb(const Register& rt, const MemOperand& src) {
LoadStore(rt, src, LDRB_w);
}
void Assembler::strb(const Register& rt, const MemOperand& dst) {
LoadStore(rt, dst, STRB_w);
}
void Assembler::ldrsb(const Register& rt, const MemOperand& src) {
LoadStore(rt, src, rt.Is64Bits() ? LDRSB_x : LDRSB_w);
}
void Assembler::ldrh(const Register& rt, const MemOperand& src) {
LoadStore(rt, src, LDRH_w);
}
void Assembler::strh(const Register& rt, const MemOperand& dst) {
LoadStore(rt, dst, STRH_w);
}
void Assembler::ldrsh(const Register& rt, const MemOperand& src) {
LoadStore(rt, src, rt.Is64Bits() ? LDRSH_x : LDRSH_w);
}
void Assembler::ldr(const CPURegister& rt, const MemOperand& src) {
LoadStore(rt, src, LoadOpFor(rt));
}
void Assembler::str(const CPURegister& rt, const MemOperand& src) {
LoadStore(rt, src, StoreOpFor(rt));
}
void Assembler::ldrsw(const Register& rt, const MemOperand& src) {
DCHECK(rt.Is64Bits());
LoadStore(rt, src, LDRSW_x);
}
void Assembler::ldr_pcrel(const CPURegister& rt, int imm19) {
// The pattern 'ldr xzr, #offset' is used to indicate the beginning of a
// constant pool. It should not be emitted.
DCHECK(!rt.IsZero());
Emit(LoadLiteralOpFor(rt) | ImmLLiteral(imm19) | Rt(rt));
}
Operand Operand::EmbeddedNumber(double number) {
int32_t smi;
if (DoubleToSmiInteger(number, &smi)) {
return Operand(Immediate(Smi::FromInt(smi)));
}
Operand result(0, RelocInfo::EMBEDDED_OBJECT);
result.heap_object_request_.emplace(number);
DCHECK(result.IsHeapObjectRequest());
return result;
}
Operand Operand::EmbeddedCode(CodeStub* stub) {
Operand result(0, RelocInfo::CODE_TARGET);
result.heap_object_request_.emplace(stub);
DCHECK(result.IsHeapObjectRequest());
return result;
}
void Assembler::ldr(const CPURegister& rt, const Operand& operand) {
if (operand.IsHeapObjectRequest()) {
RequestHeapObject(operand.heap_object_request());
ldr(rt, operand.immediate_for_heap_object_request());
} else {
ldr(rt, operand.immediate());
}
}
void Assembler::ldr(const CPURegister& rt, const Immediate& imm) {
// Currently we only support 64-bit literals.
DCHECK(rt.Is64Bits());
RecordRelocInfo(imm.rmode(), imm.value());
BlockConstPoolFor(1);
// The load will be patched when the constpool is emitted, patching code
// expect a load literal with offset 0.
ldr_pcrel(rt, 0);
}
void Assembler::ldar(const Register& rt, const Register& rn) {
DCHECK(rn.Is64Bits());
LoadStoreAcquireReleaseOp op = rt.Is32Bits() ? LDAR_w : LDAR_x;
Emit(op | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::ldaxr(const Register& rt, const Register& rn) {
DCHECK(rn.Is64Bits());
LoadStoreAcquireReleaseOp op = rt.Is32Bits() ? LDAXR_w : LDAXR_x;
Emit(op | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlr(const Register& rt, const Register& rn) {
DCHECK(rn.Is64Bits());
LoadStoreAcquireReleaseOp op = rt.Is32Bits() ? STLR_w : STLR_x;
Emit(op | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlxr(const Register& rs, const Register& rt,
const Register& rn) {
DCHECK(rs.Is32Bits());
DCHECK(rn.Is64Bits());
DCHECK(!rs.Is(rt) && !rs.Is(rn));
LoadStoreAcquireReleaseOp op = rt.Is32Bits() ? STLXR_w : STLXR_x;
Emit(op | Rs(rs) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::ldarb(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(LDAR_b | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::ldaxrb(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(LDAXR_b | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlrb(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(STLR_b | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlxrb(const Register& rs, const Register& rt,
const Register& rn) {
DCHECK(rs.Is32Bits());
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
DCHECK(!rs.Is(rt) && !rs.Is(rn));
Emit(STLXR_b | Rs(rs) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::ldarh(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(LDAR_h | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::ldaxrh(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(LDAXR_h | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlrh(const Register& rt, const Register& rn) {
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
Emit(STLR_h | Rs(x31) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::stlxrh(const Register& rs, const Register& rt,
const Register& rn) {
DCHECK(rs.Is32Bits());
DCHECK(rt.Is32Bits());
DCHECK(rn.Is64Bits());
DCHECK(!rs.Is(rt) && !rs.Is(rn));
Emit(STLXR_h | Rs(rs) | Rt2(x31) | RnSP(rn) | Rt(rt));
}
void Assembler::NEON3DifferentL(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEON3DifferentOp vop) {
DCHECK(AreSameFormat(vn, vm));
DCHECK((vn.Is1H() && vd.Is1S()) || (vn.Is1S() && vd.Is1D()) ||
(vn.Is8B() && vd.Is8H()) || (vn.Is4H() && vd.Is4S()) ||
(vn.Is2S() && vd.Is2D()) || (vn.Is16B() && vd.Is8H()) ||
(vn.Is8H() && vd.Is4S()) || (vn.Is4S() && vd.Is2D()));
Instr format, op = vop;
if (vd.IsScalar()) {
op |= NEON_Q | NEONScalar;
format = SFormat(vn);
} else {
format = VFormat(vn);
}
Emit(format | op | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::NEON3DifferentW(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEON3DifferentOp vop) {
DCHECK(AreSameFormat(vd, vn));
DCHECK((vm.Is8B() && vd.Is8H()) || (vm.Is4H() && vd.Is4S()) ||
(vm.Is2S() && vd.Is2D()) || (vm.Is16B() && vd.Is8H()) ||
(vm.Is8H() && vd.Is4S()) || (vm.Is4S() && vd.Is2D()));
Emit(VFormat(vm) | vop | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::NEON3DifferentHN(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEON3DifferentOp vop) {
DCHECK(AreSameFormat(vm, vn));
DCHECK((vd.Is8B() && vn.Is8H()) || (vd.Is4H() && vn.Is4S()) ||
(vd.Is2S() && vn.Is2D()) || (vd.Is16B() && vn.Is8H()) ||
(vd.Is8H() && vn.Is4S()) || (vd.Is4S() && vn.Is2D()));
Emit(VFormat(vd) | vop | Rm(vm) | Rn(vn) | Rd(vd));
}
#define NEON_3DIFF_LONG_LIST(V) \
V(pmull, NEON_PMULL, vn.IsVector() && vn.Is8B()) \
V(pmull2, NEON_PMULL2, vn.IsVector() && vn.Is16B()) \
V(saddl, NEON_SADDL, vn.IsVector() && vn.IsD()) \
V(saddl2, NEON_SADDL2, vn.IsVector() && vn.IsQ()) \
V(sabal, NEON_SABAL, vn.IsVector() && vn.IsD()) \
V(sabal2, NEON_SABAL2, vn.IsVector() && vn.IsQ()) \
V(uabal, NEON_UABAL, vn.IsVector() && vn.IsD()) \
V(uabal2, NEON_UABAL2, vn.IsVector() && vn.IsQ()) \
V(sabdl, NEON_SABDL, vn.IsVector() && vn.IsD()) \
V(sabdl2, NEON_SABDL2, vn.IsVector() && vn.IsQ()) \
V(uabdl, NEON_UABDL, vn.IsVector() && vn.IsD()) \
V(uabdl2, NEON_UABDL2, vn.IsVector() && vn.IsQ()) \
V(smlal, NEON_SMLAL, vn.IsVector() && vn.IsD()) \
V(smlal2, NEON_SMLAL2, vn.IsVector() && vn.IsQ()) \
V(umlal, NEON_UMLAL, vn.IsVector() && vn.IsD()) \
V(umlal2, NEON_UMLAL2, vn.IsVector() && vn.IsQ()) \
V(smlsl, NEON_SMLSL, vn.IsVector() && vn.IsD()) \
V(smlsl2, NEON_SMLSL2, vn.IsVector() && vn.IsQ()) \
V(umlsl, NEON_UMLSL, vn.IsVector() && vn.IsD()) \
V(umlsl2, NEON_UMLSL2, vn.IsVector() && vn.IsQ()) \
V(smull, NEON_SMULL, vn.IsVector() && vn.IsD()) \
V(smull2, NEON_SMULL2, vn.IsVector() && vn.IsQ()) \
V(umull, NEON_UMULL, vn.IsVector() && vn.IsD()) \
V(umull2, NEON_UMULL2, vn.IsVector() && vn.IsQ()) \
V(ssubl, NEON_SSUBL, vn.IsVector() && vn.IsD()) \
V(ssubl2, NEON_SSUBL2, vn.IsVector() && vn.IsQ()) \
V(uaddl, NEON_UADDL, vn.IsVector() && vn.IsD()) \
V(uaddl2, NEON_UADDL2, vn.IsVector() && vn.IsQ()) \
V(usubl, NEON_USUBL, vn.IsVector() && vn.IsD()) \
V(usubl2, NEON_USUBL2, vn.IsVector() && vn.IsQ()) \
V(sqdmlal, NEON_SQDMLAL, vn.Is1H() || vn.Is1S() || vn.Is4H() || vn.Is2S()) \
V(sqdmlal2, NEON_SQDMLAL2, vn.Is1H() || vn.Is1S() || vn.Is8H() || vn.Is4S()) \
V(sqdmlsl, NEON_SQDMLSL, vn.Is1H() || vn.Is1S() || vn.Is4H() || vn.Is2S()) \
V(sqdmlsl2, NEON_SQDMLSL2, vn.Is1H() || vn.Is1S() || vn.Is8H() || vn.Is4S()) \
V(sqdmull, NEON_SQDMULL, vn.Is1H() || vn.Is1S() || vn.Is4H() || vn.Is2S()) \
V(sqdmull2, NEON_SQDMULL2, vn.Is1H() || vn.Is1S() || vn.Is8H() || vn.Is4S())
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm) { \
DCHECK(AS); \
NEON3DifferentL(vd, vn, vm, OP); \
}
NEON_3DIFF_LONG_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
#define NEON_3DIFF_HN_LIST(V) \
V(addhn, NEON_ADDHN, vd.IsD()) \
V(addhn2, NEON_ADDHN2, vd.IsQ()) \
V(raddhn, NEON_RADDHN, vd.IsD()) \
V(raddhn2, NEON_RADDHN2, vd.IsQ()) \
V(subhn, NEON_SUBHN, vd.IsD()) \
V(subhn2, NEON_SUBHN2, vd.IsQ()) \
V(rsubhn, NEON_RSUBHN, vd.IsD()) \
V(rsubhn2, NEON_RSUBHN2, vd.IsQ())
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm) { \
DCHECK(AS); \
NEON3DifferentHN(vd, vn, vm, OP); \
}
NEON_3DIFF_HN_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
void Assembler::NEONPerm(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEONPermOp op) {
DCHECK(AreSameFormat(vd, vn, vm));
DCHECK(!vd.Is1D());
Emit(VFormat(vd) | op | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::trn1(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_TRN1);
}
void Assembler::trn2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_TRN2);
}
void Assembler::uzp1(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_UZP1);
}
void Assembler::uzp2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_UZP2);
}
void Assembler::zip1(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_ZIP1);
}
void Assembler::zip2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONPerm(vd, vn, vm, NEON_ZIP2);
}
void Assembler::NEONShiftImmediate(const VRegister& vd, const VRegister& vn,
NEONShiftImmediateOp op, int immh_immb) {
DCHECK(AreSameFormat(vd, vn));
Instr q, scalar;
if (vn.IsScalar()) {
q = NEON_Q;
scalar = NEONScalar;
} else {
q = vd.IsD() ? 0 : NEON_Q;
scalar = 0;
}
Emit(q | op | scalar | immh_immb | Rn(vn) | Rd(vd));
}
void Assembler::NEONShiftLeftImmediate(const VRegister& vd, const VRegister& vn,
int shift, NEONShiftImmediateOp op) {
int laneSizeInBits = vn.LaneSizeInBits();
DCHECK((shift >= 0) && (shift < laneSizeInBits));
NEONShiftImmediate(vd, vn, op, (laneSizeInBits + shift) << 16);
}
void Assembler::NEONShiftRightImmediate(const VRegister& vd,
const VRegister& vn, int shift,
NEONShiftImmediateOp op) {
int laneSizeInBits = vn.LaneSizeInBits();
DCHECK((shift >= 1) && (shift <= laneSizeInBits));
NEONShiftImmediate(vd, vn, op, ((2 * laneSizeInBits) - shift) << 16);
}
void Assembler::NEONShiftImmediateL(const VRegister& vd, const VRegister& vn,
int shift, NEONShiftImmediateOp op) {
int laneSizeInBits = vn.LaneSizeInBits();
DCHECK((shift >= 0) && (shift < laneSizeInBits));
int immh_immb = (laneSizeInBits + shift) << 16;
DCHECK((vn.Is8B() && vd.Is8H()) || (vn.Is4H() && vd.Is4S()) ||
(vn.Is2S() && vd.Is2D()) || (vn.Is16B() && vd.Is8H()) ||
(vn.Is8H() && vd.Is4S()) || (vn.Is4S() && vd.Is2D()));
Instr q;
q = vn.IsD() ? 0 : NEON_Q;
Emit(q | op | immh_immb | Rn(vn) | Rd(vd));
}
void Assembler::NEONShiftImmediateN(const VRegister& vd, const VRegister& vn,
int shift, NEONShiftImmediateOp op) {
Instr q, scalar;
int laneSizeInBits = vd.LaneSizeInBits();
DCHECK((shift >= 1) && (shift <= laneSizeInBits));
int immh_immb = (2 * laneSizeInBits - shift) << 16;
if (vn.IsScalar()) {
DCHECK((vd.Is1B() && vn.Is1H()) || (vd.Is1H() && vn.Is1S()) ||
(vd.Is1S() && vn.Is1D()));
q = NEON_Q;
scalar = NEONScalar;
} else {
DCHECK((vd.Is8B() && vn.Is8H()) || (vd.Is4H() && vn.Is4S()) ||
(vd.Is2S() && vn.Is2D()) || (vd.Is16B() && vn.Is8H()) ||
(vd.Is8H() && vn.Is4S()) || (vd.Is4S() && vn.Is2D()));
scalar = 0;
q = vd.IsD() ? 0 : NEON_Q;
}
Emit(q | op | scalar | immh_immb | Rn(vn) | Rd(vd));
}
void Assembler::shl(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftLeftImmediate(vd, vn, shift, NEON_SHL);
}
void Assembler::sli(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftLeftImmediate(vd, vn, shift, NEON_SLI);
}
void Assembler::sqshl(const VRegister& vd, const VRegister& vn, int shift) {
NEONShiftLeftImmediate(vd, vn, shift, NEON_SQSHL_imm);
}
void Assembler::sqshlu(const VRegister& vd, const VRegister& vn, int shift) {
NEONShiftLeftImmediate(vd, vn, shift, NEON_SQSHLU);
}
void Assembler::uqshl(const VRegister& vd, const VRegister& vn, int shift) {
NEONShiftLeftImmediate(vd, vn, shift, NEON_UQSHL_imm);
}
void Assembler::sshll(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsD());
NEONShiftImmediateL(vd, vn, shift, NEON_SSHLL);
}
void Assembler::sshll2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsQ());
NEONShiftImmediateL(vd, vn, shift, NEON_SSHLL);
}
void Assembler::sxtl(const VRegister& vd, const VRegister& vn) {
sshll(vd, vn, 0);
}
void Assembler::sxtl2(const VRegister& vd, const VRegister& vn) {
sshll2(vd, vn, 0);
}
void Assembler::ushll(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsD());
NEONShiftImmediateL(vd, vn, shift, NEON_USHLL);
}
void Assembler::ushll2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsQ());
NEONShiftImmediateL(vd, vn, shift, NEON_USHLL);
}
void Assembler::uxtl(const VRegister& vd, const VRegister& vn) {
ushll(vd, vn, 0);
}
void Assembler::uxtl2(const VRegister& vd, const VRegister& vn) {
ushll2(vd, vn, 0);
}
void Assembler::sri(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_SRI);
}
void Assembler::sshr(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_SSHR);
}
void Assembler::ushr(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_USHR);
}
void Assembler::srshr(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_SRSHR);
}
void Assembler::urshr(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_URSHR);
}
void Assembler::ssra(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_SSRA);
}
void Assembler::usra(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_USRA);
}
void Assembler::srsra(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_SRSRA);
}
void Assembler::ursra(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsVector() || vd.Is1D());
NEONShiftRightImmediate(vd, vn, shift, NEON_URSRA);
}
void Assembler::shrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsD());
NEONShiftImmediateN(vd, vn, shift, NEON_SHRN);
}
void Assembler::shrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_SHRN);
}
void Assembler::rshrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsD());
NEONShiftImmediateN(vd, vn, shift, NEON_RSHRN);
}
void Assembler::rshrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_RSHRN);
}
void Assembler::sqshrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_SQSHRN);
}
void Assembler::sqshrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_SQSHRN);
}
void Assembler::sqrshrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_SQRSHRN);
}
void Assembler::sqrshrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_SQRSHRN);
}
void Assembler::sqshrun(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_SQSHRUN);
}
void Assembler::sqshrun2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_SQSHRUN);
}
void Assembler::sqrshrun(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_SQRSHRUN);
}
void Assembler::sqrshrun2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_SQRSHRUN);
}
void Assembler::uqshrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_UQSHRN);
}
void Assembler::uqshrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_UQSHRN);
}
void Assembler::uqrshrn(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vd.IsD() || (vn.IsScalar() && vd.IsScalar()));
NEONShiftImmediateN(vd, vn, shift, NEON_UQRSHRN);
}
void Assembler::uqrshrn2(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK(vn.IsVector() && vd.IsQ());
NEONShiftImmediateN(vd, vn, shift, NEON_UQRSHRN);
}
void Assembler::uaddw(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsD());
NEON3DifferentW(vd, vn, vm, NEON_UADDW);
}
void Assembler::uaddw2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsQ());
NEON3DifferentW(vd, vn, vm, NEON_UADDW2);
}
void Assembler::saddw(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsD());
NEON3DifferentW(vd, vn, vm, NEON_SADDW);
}
void Assembler::saddw2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsQ());
NEON3DifferentW(vd, vn, vm, NEON_SADDW2);
}
void Assembler::usubw(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsD());
NEON3DifferentW(vd, vn, vm, NEON_USUBW);
}
void Assembler::usubw2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsQ());
NEON3DifferentW(vd, vn, vm, NEON_USUBW2);
}
void Assembler::ssubw(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsD());
NEON3DifferentW(vd, vn, vm, NEON_SSUBW);
}
void Assembler::ssubw2(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(vm.IsQ());
NEON3DifferentW(vd, vn, vm, NEON_SSUBW2);
}
void Assembler::mov(const Register& rd, const Register& rm) {
// Moves involving the stack pointer are encoded as add immediate with
// second operand of zero. Otherwise, orr with first operand zr is
// used.
if (rd.IsSP() || rm.IsSP()) {
add(rd, rm, 0);
} else {
orr(rd, AppropriateZeroRegFor(rd), rm);
}
}
void Assembler::ins(const VRegister& vd, int vd_index, const Register& rn) {
// We support vd arguments of the form vd.VxT() or vd.T(), where x is the
// number of lanes, and T is b, h, s or d.
int lane_size = vd.LaneSizeInBytes();
NEONFormatField format;
switch (lane_size) {
case 1:
format = NEON_16B;
DCHECK(rn.IsW());
break;
case 2:
format = NEON_8H;
DCHECK(rn.IsW());
break;
case 4:
format = NEON_4S;
DCHECK(rn.IsW());
break;
default:
DCHECK_EQ(lane_size, 8);
DCHECK(rn.IsX());
format = NEON_2D;
break;
}
DCHECK((0 <= vd_index) &&
(vd_index < LaneCountFromFormat(static_cast<VectorFormat>(format))));
Emit(NEON_INS_GENERAL | ImmNEON5(format, vd_index) | Rn(rn) | Rd(vd));
}
void Assembler::mov(const Register& rd, const VRegister& vn, int vn_index) {
DCHECK_GE(vn.SizeInBytes(), 4);
umov(rd, vn, vn_index);
}
void Assembler::smov(const Register& rd, const VRegister& vn, int vn_index) {
// We support vn arguments of the form vn.VxT() or vn.T(), where x is the
// number of lanes, and T is b, h, s.
int lane_size = vn.LaneSizeInBytes();
NEONFormatField format;
Instr q = 0;
switch (lane_size) {
case 1:
format = NEON_16B;
break;
case 2:
format = NEON_8H;
break;
default:
DCHECK_EQ(lane_size, 4);
DCHECK(rd.IsX());
format = NEON_4S;
break;
}
q = rd.IsW() ? 0 : NEON_Q;
DCHECK((0 <= vn_index) &&
(vn_index < LaneCountFromFormat(static_cast<VectorFormat>(format))));
Emit(q | NEON_SMOV | ImmNEON5(format, vn_index) | Rn(vn) | Rd(rd));
}
void Assembler::cls(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(!vd.Is1D() && !vd.Is2D());
Emit(VFormat(vn) | NEON_CLS | Rn(vn) | Rd(vd));
}
void Assembler::clz(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(!vd.Is1D() && !vd.Is2D());
Emit(VFormat(vn) | NEON_CLZ | Rn(vn) | Rd(vd));
}
void Assembler::cnt(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is8B() || vd.Is16B());
Emit(VFormat(vn) | NEON_CNT | Rn(vn) | Rd(vd));
}
void Assembler::rev16(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is8B() || vd.Is16B());
Emit(VFormat(vn) | NEON_REV16 | Rn(vn) | Rd(vd));
}
void Assembler::rev32(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is8B() || vd.Is16B() || vd.Is4H() || vd.Is8H());
Emit(VFormat(vn) | NEON_REV32 | Rn(vn) | Rd(vd));
}
void Assembler::rev64(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(!vd.Is1D() && !vd.Is2D());
Emit(VFormat(vn) | NEON_REV64 | Rn(vn) | Rd(vd));
}
void Assembler::ursqrte(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is2S() || vd.Is4S());
Emit(VFormat(vn) | NEON_URSQRTE | Rn(vn) | Rd(vd));
}
void Assembler::urecpe(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is2S() || vd.Is4S());
Emit(VFormat(vn) | NEON_URECPE | Rn(vn) | Rd(vd));
}
void Assembler::NEONAddlp(const VRegister& vd, const VRegister& vn,
NEON2RegMiscOp op) {
DCHECK((op == NEON_SADDLP) || (op == NEON_UADDLP) || (op == NEON_SADALP) ||
(op == NEON_UADALP));
DCHECK((vn.Is8B() && vd.Is4H()) || (vn.Is4H() && vd.Is2S()) ||
(vn.Is2S() && vd.Is1D()) || (vn.Is16B() && vd.Is8H()) ||
(vn.Is8H() && vd.Is4S()) || (vn.Is4S() && vd.Is2D()));
Emit(VFormat(vn) | op | Rn(vn) | Rd(vd));
}
void Assembler::saddlp(const VRegister& vd, const VRegister& vn) {
NEONAddlp(vd, vn, NEON_SADDLP);
}
void Assembler::uaddlp(const VRegister& vd, const VRegister& vn) {
NEONAddlp(vd, vn, NEON_UADDLP);
}
void Assembler::sadalp(const VRegister& vd, const VRegister& vn) {
NEONAddlp(vd, vn, NEON_SADALP);
}
void Assembler::uadalp(const VRegister& vd, const VRegister& vn) {
NEONAddlp(vd, vn, NEON_UADALP);
}
void Assembler::NEONAcrossLanesL(const VRegister& vd, const VRegister& vn,
NEONAcrossLanesOp op) {
DCHECK((vn.Is8B() && vd.Is1H()) || (vn.Is16B() && vd.Is1H()) ||
(vn.Is4H() && vd.Is1S()) || (vn.Is8H() && vd.Is1S()) ||
(vn.Is4S() && vd.Is1D()));
Emit(VFormat(vn) | op | Rn(vn) | Rd(vd));
}
void Assembler::saddlv(const VRegister& vd, const VRegister& vn) {
NEONAcrossLanesL(vd, vn, NEON_SADDLV);
}
void Assembler::uaddlv(const VRegister& vd, const VRegister& vn) {
NEONAcrossLanesL(vd, vn, NEON_UADDLV);
}
void Assembler::NEONAcrossLanes(const VRegister& vd, const VRegister& vn,
NEONAcrossLanesOp op) {
DCHECK((vn.Is8B() && vd.Is1B()) || (vn.Is16B() && vd.Is1B()) ||
(vn.Is4H() && vd.Is1H()) || (vn.Is8H() && vd.Is1H()) ||
(vn.Is4S() && vd.Is1S()));
if ((op & NEONAcrossLanesFPFMask) == NEONAcrossLanesFPFixed) {
Emit(FPFormat(vn) | op | Rn(vn) | Rd(vd));
} else {
Emit(VFormat(vn) | op | Rn(vn) | Rd(vd));
}
}
#define NEON_ACROSSLANES_LIST(V) \
V(fmaxv, NEON_FMAXV, vd.Is1S()) \
V(fminv, NEON_FMINV, vd.Is1S()) \
V(fmaxnmv, NEON_FMAXNMV, vd.Is1S()) \
V(fminnmv, NEON_FMINNMV, vd.Is1S()) \
V(addv, NEON_ADDV, true) \
V(smaxv, NEON_SMAXV, true) \
V(sminv, NEON_SMINV, true) \
V(umaxv, NEON_UMAXV, true) \
V(uminv, NEON_UMINV, true)
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn) { \
DCHECK(AS); \
NEONAcrossLanes(vd, vn, OP); \
}
NEON_ACROSSLANES_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
void Assembler::mov(const VRegister& vd, int vd_index, const Register& rn) {
ins(vd, vd_index, rn);
}
void Assembler::umov(const Register& rd, const VRegister& vn, int vn_index) {
// We support vn arguments of the form vn.VxT() or vn.T(), where x is the
// number of lanes, and T is b, h, s or d.
int lane_size = vn.LaneSizeInBytes();
NEONFormatField format;
Instr q = 0;
switch (lane_size) {
case 1:
format = NEON_16B;
DCHECK(rd.IsW());
break;
case 2:
format = NEON_8H;
DCHECK(rd.IsW());
break;
case 4:
format = NEON_4S;
DCHECK(rd.IsW());
break;
default:
DCHECK_EQ(lane_size, 8);
DCHECK(rd.IsX());
format = NEON_2D;
q = NEON_Q;
break;
}
DCHECK((0 <= vn_index) &&
(vn_index < LaneCountFromFormat(static_cast<VectorFormat>(format))));
Emit(q | NEON_UMOV | ImmNEON5(format, vn_index) | Rn(vn) | Rd(rd));
}
void Assembler::mov(const VRegister& vd, const VRegister& vn, int vn_index) {
DCHECK(vd.IsScalar());
dup(vd, vn, vn_index);
}
void Assembler::dup(const VRegister& vd, const Register& rn) {
DCHECK(!vd.Is1D());
DCHECK_EQ(vd.Is2D(), rn.IsX());
Instr q = vd.IsD() ? 0 : NEON_Q;
Emit(q | NEON_DUP_GENERAL | ImmNEON5(VFormat(vd), 0) | Rn(rn) | Rd(vd));
}
void Assembler::ins(const VRegister& vd, int vd_index, const VRegister& vn,
int vn_index) {
DCHECK(AreSameFormat(vd, vn));
// We support vd arguments of the form vd.VxT() or vd.T(), where x is the
// number of lanes, and T is b, h, s or d.
int lane_size = vd.LaneSizeInBytes();
NEONFormatField format;
switch (lane_size) {
case 1:
format = NEON_16B;
break;
case 2:
format = NEON_8H;
break;
case 4:
format = NEON_4S;
break;
default:
DCHECK_EQ(lane_size, 8);
format = NEON_2D;
break;
}
DCHECK((0 <= vd_index) &&
(vd_index < LaneCountFromFormat(static_cast<VectorFormat>(format))));
DCHECK((0 <= vn_index) &&
(vn_index < LaneCountFromFormat(static_cast<VectorFormat>(format))));
Emit(NEON_INS_ELEMENT | ImmNEON5(format, vd_index) |
ImmNEON4(format, vn_index) | Rn(vn) | Rd(vd));
}
void Assembler::NEONTable(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEONTableOp op) {
DCHECK(vd.Is16B() || vd.Is8B());
DCHECK(vn.Is16B());
DCHECK(AreSameFormat(vd, vm));
Emit(op | (vd.IsQ() ? NEON_Q : 0) | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::tbl(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONTable(vd, vn, vm, NEON_TBL_1v);
}
void Assembler::tbl(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vm) {
USE(vn2);
DCHECK(AreSameFormat(vn, vn2));
DCHECK(AreConsecutive(vn, vn2));
NEONTable(vd, vn, vm, NEON_TBL_2v);
}
void Assembler::tbl(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vn3,
const VRegister& vm) {
USE(vn2);
USE(vn3);
DCHECK(AreSameFormat(vn, vn2, vn3));
DCHECK(AreConsecutive(vn, vn2, vn3));
NEONTable(vd, vn, vm, NEON_TBL_3v);
}
void Assembler::tbl(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vn3,
const VRegister& vn4, const VRegister& vm) {
USE(vn2);
USE(vn3);
USE(vn4);
DCHECK(AreSameFormat(vn, vn2, vn3, vn4));
DCHECK(AreConsecutive(vn, vn2, vn3, vn4));
NEONTable(vd, vn, vm, NEON_TBL_4v);
}
void Assembler::tbx(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
NEONTable(vd, vn, vm, NEON_TBX_1v);
}
void Assembler::tbx(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vm) {
USE(vn2);
DCHECK(AreSameFormat(vn, vn2));
DCHECK(AreConsecutive(vn, vn2));
NEONTable(vd, vn, vm, NEON_TBX_2v);
}
void Assembler::tbx(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vn3,
const VRegister& vm) {
USE(vn2);
USE(vn3);
DCHECK(AreSameFormat(vn, vn2, vn3));
DCHECK(AreConsecutive(vn, vn2, vn3));
NEONTable(vd, vn, vm, NEON_TBX_3v);
}
void Assembler::tbx(const VRegister& vd, const VRegister& vn,
const VRegister& vn2, const VRegister& vn3,
const VRegister& vn4, const VRegister& vm) {
USE(vn2);
USE(vn3);
USE(vn4);
DCHECK(AreSameFormat(vn, vn2, vn3, vn4));
DCHECK(AreConsecutive(vn, vn2, vn3, vn4));
NEONTable(vd, vn, vm, NEON_TBX_4v);
}
void Assembler::mov(const VRegister& vd, int vd_index, const VRegister& vn,
int vn_index) {
ins(vd, vd_index, vn, vn_index);
}
void Assembler::mvn(const Register& rd, const Operand& operand) {
orn(rd, AppropriateZeroRegFor(rd), operand);
}
void Assembler::mrs(const Register& rt, SystemRegister sysreg) {
DCHECK(rt.Is64Bits());
Emit(MRS | ImmSystemRegister(sysreg) | Rt(rt));
}
void Assembler::msr(SystemRegister sysreg, const Register& rt) {
DCHECK(rt.Is64Bits());
Emit(MSR | Rt(rt) | ImmSystemRegister(sysreg));
}
void Assembler::hint(SystemHint code) { Emit(HINT | ImmHint(code) | Rt(xzr)); }
// NEON structure loads and stores.
Instr Assembler::LoadStoreStructAddrModeField(const MemOperand& addr) {
Instr addr_field = RnSP(addr.base());
if (addr.IsPostIndex()) {
static_assert(NEONLoadStoreMultiStructPostIndex ==
static_cast<NEONLoadStoreMultiStructPostIndexOp>(
NEONLoadStoreSingleStructPostIndex),
"Opcodes must match for NEON post index memop.");
addr_field |= NEONLoadStoreMultiStructPostIndex;
if (addr.offset() == 0) {
addr_field |= RmNot31(addr.regoffset());
} else {
// The immediate post index addressing mode is indicated by rm = 31.
// The immediate is implied by the number of vector registers used.
addr_field |= (0x1F << Rm_offset);
}
} else {
DCHECK(addr.IsImmediateOffset() && (addr.offset() == 0));
}
return addr_field;
}
void Assembler::LoadStoreStructVerify(const VRegister& vt,
const MemOperand& addr, Instr op) {
#ifdef DEBUG
// Assert that addressing mode is either offset (with immediate 0), post
// index by immediate of the size of the register list, or post index by a
// value in a core register.
if (addr.IsImmediateOffset()) {
DCHECK_EQ(addr.offset(), 0);
} else {
int offset = vt.SizeInBytes();
switch (op) {
case NEON_LD1_1v:
case NEON_ST1_1v:
offset *= 1;
break;
case NEONLoadStoreSingleStructLoad1:
case NEONLoadStoreSingleStructStore1:
case NEON_LD1R:
offset = (offset / vt.LaneCount()) * 1;
break;
case NEON_LD1_2v:
case NEON_ST1_2v:
case NEON_LD2:
case NEON_ST2:
offset *= 2;
break;
case NEONLoadStoreSingleStructLoad2:
case NEONLoadStoreSingleStructStore2:
case NEON_LD2R:
offset = (offset / vt.LaneCount()) * 2;
break;
case NEON_LD1_3v:
case NEON_ST1_3v:
case NEON_LD3:
case NEON_ST3:
offset *= 3;
break;
case NEONLoadStoreSingleStructLoad3:
case NEONLoadStoreSingleStructStore3:
case NEON_LD3R:
offset = (offset / vt.LaneCount()) * 3;
break;
case NEON_LD1_4v:
case NEON_ST1_4v:
case NEON_LD4:
case NEON_ST4:
offset *= 4;
break;
case NEONLoadStoreSingleStructLoad4:
case NEONLoadStoreSingleStructStore4:
case NEON_LD4R:
offset = (offset / vt.LaneCount()) * 4;
break;
default:
UNREACHABLE();
}
DCHECK(!addr.regoffset().Is(NoReg) || addr.offset() == offset);
}
#else
USE(vt);
USE(addr);
USE(op);
#endif
}
void Assembler::LoadStoreStruct(const VRegister& vt, const MemOperand& addr,
NEONLoadStoreMultiStructOp op) {
LoadStoreStructVerify(vt, addr, op);
DCHECK(vt.IsVector() || vt.Is1D());
Emit(op | LoadStoreStructAddrModeField(addr) | LSVFormat(vt) | Rt(vt));
}
void Assembler::LoadStoreStructSingleAllLanes(const VRegister& vt,
const MemOperand& addr,
NEONLoadStoreSingleStructOp op) {
LoadStoreStructVerify(vt, addr, op);
Emit(op | LoadStoreStructAddrModeField(addr) | LSVFormat(vt) | Rt(vt));
}
void Assembler::ld1(const VRegister& vt, const MemOperand& src) {
LoadStoreStruct(vt, src, NEON_LD1_1v);
}
void Assembler::ld1(const VRegister& vt, const VRegister& vt2,
const MemOperand& src) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStruct(vt, src, NEON_LD1_2v);
}
void Assembler::ld1(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const MemOperand& src) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStruct(vt, src, NEON_LD1_3v);
}
void Assembler::ld1(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4,
const MemOperand& src) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStruct(vt, src, NEON_LD1_4v);
}
void Assembler::ld2(const VRegister& vt, const VRegister& vt2,
const MemOperand& src) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStruct(vt, src, NEON_LD2);
}
void Assembler::ld2(const VRegister& vt, const VRegister& vt2, int lane,
const MemOperand& src) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStructSingle(vt, lane, src, NEONLoadStoreSingleStructLoad2);
}
void Assembler::ld2r(const VRegister& vt, const VRegister& vt2,
const MemOperand& src) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStructSingleAllLanes(vt, src, NEON_LD2R);
}
void Assembler::ld3(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const MemOperand& src) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStruct(vt, src, NEON_LD3);
}
void Assembler::ld3(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, int lane, const MemOperand& src) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStructSingle(vt, lane, src, NEONLoadStoreSingleStructLoad3);
}
void Assembler::ld3r(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const MemOperand& src) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStructSingleAllLanes(vt, src, NEON_LD3R);
}
void Assembler::ld4(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4,
const MemOperand& src) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStruct(vt, src, NEON_LD4);
}
void Assembler::ld4(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4, int lane,
const MemOperand& src) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStructSingle(vt, lane, src, NEONLoadStoreSingleStructLoad4);
}
void Assembler::ld4r(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4,
const MemOperand& src) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStructSingleAllLanes(vt, src, NEON_LD4R);
}
void Assembler::st1(const VRegister& vt, const MemOperand& src) {
LoadStoreStruct(vt, src, NEON_ST1_1v);
}
void Assembler::st1(const VRegister& vt, const VRegister& vt2,
const MemOperand& src) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStruct(vt, src, NEON_ST1_2v);
}
void Assembler::st1(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const MemOperand& src) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStruct(vt, src, NEON_ST1_3v);
}
void Assembler::st1(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4,
const MemOperand& src) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStruct(vt, src, NEON_ST1_4v);
}
void Assembler::st2(const VRegister& vt, const VRegister& vt2,
const MemOperand& dst) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStruct(vt, dst, NEON_ST2);
}
void Assembler::st2(const VRegister& vt, const VRegister& vt2, int lane,
const MemOperand& dst) {
USE(vt2);
DCHECK(AreSameFormat(vt, vt2));
DCHECK(AreConsecutive(vt, vt2));
LoadStoreStructSingle(vt, lane, dst, NEONLoadStoreSingleStructStore2);
}
void Assembler::st3(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const MemOperand& dst) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStruct(vt, dst, NEON_ST3);
}
void Assembler::st3(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, int lane, const MemOperand& dst) {
USE(vt2);
USE(vt3);
DCHECK(AreSameFormat(vt, vt2, vt3));
DCHECK(AreConsecutive(vt, vt2, vt3));
LoadStoreStructSingle(vt, lane, dst, NEONLoadStoreSingleStructStore3);
}
void Assembler::st4(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4,
const MemOperand& dst) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStruct(vt, dst, NEON_ST4);
}
void Assembler::st4(const VRegister& vt, const VRegister& vt2,
const VRegister& vt3, const VRegister& vt4, int lane,
const MemOperand& dst) {
USE(vt2);
USE(vt3);
USE(vt4);
DCHECK(AreSameFormat(vt, vt2, vt3, vt4));
DCHECK(AreConsecutive(vt, vt2, vt3, vt4));
LoadStoreStructSingle(vt, lane, dst, NEONLoadStoreSingleStructStore4);
}
void Assembler::LoadStoreStructSingle(const VRegister& vt, uint32_t lane,
const MemOperand& addr,
NEONLoadStoreSingleStructOp op) {
LoadStoreStructVerify(vt, addr, op);
// We support vt arguments of the form vt.VxT() or vt.T(), where x is the
// number of lanes, and T is b, h, s or d.
unsigned lane_size = vt.LaneSizeInBytes();
DCHECK_LT(lane, kQRegSize / lane_size);
// Lane size is encoded in the opcode field. Lane index is encoded in the Q,
// S and size fields.
lane *= lane_size;
// Encodings for S[0]/D[0] and S[2]/D[1] are distinguished using the least-
// significant bit of the size field, so we increment lane here to account for
// that.
if (lane_size == 8) lane++;
Instr size = (lane << NEONLSSize_offset) & NEONLSSize_mask;
Instr s = (lane << (NEONS_offset - 2)) & NEONS_mask;
Instr q = (lane << (NEONQ_offset - 3)) & NEONQ_mask;
Instr instr = op;
switch (lane_size) {
case 1:
instr |= NEONLoadStoreSingle_b;
break;
case 2:
instr |= NEONLoadStoreSingle_h;
break;
case 4:
instr |= NEONLoadStoreSingle_s;
break;
default:
DCHECK_EQ(lane_size, 8U);
instr |= NEONLoadStoreSingle_d;
}
Emit(instr | LoadStoreStructAddrModeField(addr) | q | size | s | Rt(vt));
}
void Assembler::ld1(const VRegister& vt, int lane, const MemOperand& src) {
LoadStoreStructSingle(vt, lane, src, NEONLoadStoreSingleStructLoad1);
}
void Assembler::ld1r(const VRegister& vt, const MemOperand& src) {
LoadStoreStructSingleAllLanes(vt, src, NEON_LD1R);
}
void Assembler::st1(const VRegister& vt, int lane, const MemOperand& dst) {
LoadStoreStructSingle(vt, lane, dst, NEONLoadStoreSingleStructStore1);
}
void Assembler::dmb(BarrierDomain domain, BarrierType type) {
Emit(DMB | ImmBarrierDomain(domain) | ImmBarrierType(type));
}
void Assembler::dsb(BarrierDomain domain, BarrierType type) {
Emit(DSB | ImmBarrierDomain(domain) | ImmBarrierType(type));
}
void Assembler::isb() {
Emit(ISB | ImmBarrierDomain(FullSystem) | ImmBarrierType(BarrierAll));
}
void Assembler::fmov(const VRegister& vd, double imm) {
if (vd.IsScalar()) {
DCHECK(vd.Is1D());
Emit(FMOV_d_imm | Rd(vd) | ImmFP(imm));
} else {
DCHECK(vd.Is2D());
Instr op = NEONModifiedImmediate_MOVI | NEONModifiedImmediateOpBit;
Emit(NEON_Q | op | ImmNEONFP(imm) | NEONCmode(0xF) | Rd(vd));
}
}
void Assembler::fmov(const VRegister& vd, float imm) {
if (vd.IsScalar()) {
DCHECK(vd.Is1S());
Emit(FMOV_s_imm | Rd(vd) | ImmFP(imm));
} else {
DCHECK(vd.Is2S() | vd.Is4S());
Instr op = NEONModifiedImmediate_MOVI;
Instr q = vd.Is4S() ? NEON_Q : 0;
Emit(q | op | ImmNEONFP(imm) | NEONCmode(0xF) | Rd(vd));
}
}
void Assembler::fmov(const Register& rd, const VRegister& fn) {
DCHECK_EQ(rd.SizeInBits(), fn.SizeInBits());
FPIntegerConvertOp op = rd.Is32Bits() ? FMOV_ws : FMOV_xd;
Emit(op | Rd(rd) | Rn(fn));
}
void Assembler::fmov(const VRegister& vd, const Register& rn) {
DCHECK_EQ(vd.SizeInBits(), rn.SizeInBits());
FPIntegerConvertOp op = vd.Is32Bits() ? FMOV_sw : FMOV_dx;
Emit(op | Rd(vd) | Rn(rn));
}
void Assembler::fmov(const VRegister& vd, const VRegister& vn) {
DCHECK_EQ(vd.SizeInBits(), vn.SizeInBits());
Emit(FPType(vd) | FMOV | Rd(vd) | Rn(vn));
}
void Assembler::fmov(const VRegister& vd, int index, const Register& rn) {
DCHECK((index == 1) && vd.Is1D() && rn.IsX());
USE(index);
Emit(FMOV_d1_x | Rd(vd) | Rn(rn));
}
void Assembler::fmov(const Register& rd, const VRegister& vn, int index) {
DCHECK((index == 1) && vn.Is1D() && rd.IsX());
USE(index);
Emit(FMOV_x_d1 | Rd(rd) | Rn(vn));
}
void Assembler::fmadd(const VRegister& fd, const VRegister& fn,
const VRegister& fm, const VRegister& fa) {
FPDataProcessing3Source(fd, fn, fm, fa, fd.Is32Bits() ? FMADD_s : FMADD_d);
}
void Assembler::fmsub(const VRegister& fd, const VRegister& fn,
const VRegister& fm, const VRegister& fa) {
FPDataProcessing3Source(fd, fn, fm, fa, fd.Is32Bits() ? FMSUB_s : FMSUB_d);
}
void Assembler::fnmadd(const VRegister& fd, const VRegister& fn,
const VRegister& fm, const VRegister& fa) {
FPDataProcessing3Source(fd, fn, fm, fa, fd.Is32Bits() ? FNMADD_s : FNMADD_d);
}
void Assembler::fnmsub(const VRegister& fd, const VRegister& fn,
const VRegister& fm, const VRegister& fa) {
FPDataProcessing3Source(fd, fn, fm, fa, fd.Is32Bits() ? FNMSUB_s : FNMSUB_d);
}
void Assembler::fnmul(const VRegister& vd, const VRegister& vn,
const VRegister& vm) {
DCHECK(AreSameSizeAndType(vd, vn, vm));
Instr op = vd.Is1S() ? FNMUL_s : FNMUL_d;
Emit(FPType(vd) | op | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::fcmp(const VRegister& fn, const VRegister& fm) {
DCHECK_EQ(fn.SizeInBits(), fm.SizeInBits());
Emit(FPType(fn) | FCMP | Rm(fm) | Rn(fn));
}
void Assembler::fcmp(const VRegister& fn, double value) {
USE(value);
// Although the fcmp instruction can strictly only take an immediate value of
// +0.0, we don't need to check for -0.0 because the sign of 0.0 doesn't
// affect the result of the comparison.
DCHECK_EQ(value, 0.0);
Emit(FPType(fn) | FCMP_zero | Rn(fn));
}
void Assembler::fccmp(const VRegister& fn, const VRegister& fm,
StatusFlags nzcv, Condition cond) {
DCHECK_EQ(fn.SizeInBits(), fm.SizeInBits());
Emit(FPType(fn) | FCCMP | Rm(fm) | Cond(cond) | Rn(fn) | Nzcv(nzcv));
}
void Assembler::fcsel(const VRegister& fd, const VRegister& fn,
const VRegister& fm, Condition cond) {
DCHECK_EQ(fd.SizeInBits(), fn.SizeInBits());
DCHECK_EQ(fd.SizeInBits(), fm.SizeInBits());
Emit(FPType(fd) | FCSEL | Rm(fm) | Cond(cond) | Rn(fn) | Rd(fd));
}
void Assembler::NEONFPConvertToInt(const Register& rd, const VRegister& vn,
Instr op) {
Emit(SF(rd) | FPType(vn) | op | Rn(vn) | Rd(rd));
}
void Assembler::NEONFPConvertToInt(const VRegister& vd, const VRegister& vn,
Instr op) {
if (vn.IsScalar()) {
DCHECK((vd.Is1S() && vn.Is1S()) || (vd.Is1D() && vn.Is1D()));
op |= NEON_Q | NEONScalar;
}
Emit(FPFormat(vn) | op | Rn(vn) | Rd(vd));
}
void Assembler::fcvt(const VRegister& vd, const VRegister& vn) {
FPDataProcessing1SourceOp op;
if (vd.Is1D()) {
DCHECK(vn.Is1S() || vn.Is1H());
op = vn.Is1S() ? FCVT_ds : FCVT_dh;
} else if (vd.Is1S()) {
DCHECK(vn.Is1D() || vn.Is1H());
op = vn.Is1D() ? FCVT_sd : FCVT_sh;
} else {
DCHECK(vd.Is1H());
DCHECK(vn.Is1D() || vn.Is1S());
op = vn.Is1D() ? FCVT_hd : FCVT_hs;
}
FPDataProcessing1Source(vd, vn, op);
}
void Assembler::fcvtl(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is4S() && vn.Is4H()) || (vd.Is2D() && vn.Is2S()));
Instr format = vd.Is2D() ? (1 << NEONSize_offset) : 0;
Emit(format | NEON_FCVTL | Rn(vn) | Rd(vd));
}
void Assembler::fcvtl2(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is4S() && vn.Is8H()) || (vd.Is2D() && vn.Is4S()));
Instr format = vd.Is2D() ? (1 << NEONSize_offset) : 0;
Emit(NEON_Q | format | NEON_FCVTL | Rn(vn) | Rd(vd));
}
void Assembler::fcvtn(const VRegister& vd, const VRegister& vn) {
DCHECK((vn.Is4S() && vd.Is4H()) || (vn.Is2D() && vd.Is2S()));
Instr format = vn.Is2D() ? (1 << NEONSize_offset) : 0;
Emit(format | NEON_FCVTN | Rn(vn) | Rd(vd));
}
void Assembler::fcvtn2(const VRegister& vd, const VRegister& vn) {
DCHECK((vn.Is4S() && vd.Is8H()) || (vn.Is2D() && vd.Is4S()));
Instr format = vn.Is2D() ? (1 << NEONSize_offset) : 0;
Emit(NEON_Q | format | NEON_FCVTN | Rn(vn) | Rd(vd));
}
void Assembler::fcvtxn(const VRegister& vd, const VRegister& vn) {
Instr format = 1 << NEONSize_offset;
if (vd.IsScalar()) {
DCHECK(vd.Is1S() && vn.Is1D());
Emit(format | NEON_FCVTXN_scalar | Rn(vn) | Rd(vd));
} else {
DCHECK(vd.Is2S() && vn.Is2D());
Emit(format | NEON_FCVTXN | Rn(vn) | Rd(vd));
}
}
void Assembler::fcvtxn2(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.Is4S() && vn.Is2D());
Instr format = 1 << NEONSize_offset;
Emit(NEON_Q | format | NEON_FCVTXN | Rn(vn) | Rd(vd));
}
#define NEON_FP2REGMISC_FCVT_LIST(V) \
V(fcvtnu, NEON_FCVTNU, FCVTNU) \
V(fcvtns, NEON_FCVTNS, FCVTNS) \
V(fcvtpu, NEON_FCVTPU, FCVTPU) \
V(fcvtps, NEON_FCVTPS, FCVTPS) \
V(fcvtmu, NEON_FCVTMU, FCVTMU) \
V(fcvtms, NEON_FCVTMS, FCVTMS) \
V(fcvtau, NEON_FCVTAU, FCVTAU) \
V(fcvtas, NEON_FCVTAS, FCVTAS)
#define DEFINE_ASM_FUNCS(FN, VEC_OP, SCA_OP) \
void Assembler::FN(const Register& rd, const VRegister& vn) { \
NEONFPConvertToInt(rd, vn, SCA_OP); \
} \
void Assembler::FN(const VRegister& vd, const VRegister& vn) { \
NEONFPConvertToInt(vd, vn, VEC_OP); \
}
NEON_FP2REGMISC_FCVT_LIST(DEFINE_ASM_FUNCS)
#undef DEFINE_ASM_FUNCS
void Assembler::scvtf(const VRegister& vd, const VRegister& vn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
NEONFP2RegMisc(vd, vn, NEON_SCVTF);
} else {
DCHECK(vd.Is1D() || vd.Is1S() || vd.Is2D() || vd.Is2S() || vd.Is4S());
NEONShiftRightImmediate(vd, vn, fbits, NEON_SCVTF_imm);
}
}
void Assembler::ucvtf(const VRegister& vd, const VRegister& vn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
NEONFP2RegMisc(vd, vn, NEON_UCVTF);
} else {
DCHECK(vd.Is1D() || vd.Is1S() || vd.Is2D() || vd.Is2S() || vd.Is4S());
NEONShiftRightImmediate(vd, vn, fbits, NEON_UCVTF_imm);
}
}
void Assembler::scvtf(const VRegister& vd, const Register& rn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
Emit(SF(rn) | FPType(vd) | SCVTF | Rn(rn) | Rd(vd));
} else {
Emit(SF(rn) | FPType(vd) | SCVTF_fixed | FPScale(64 - fbits) | Rn(rn) |
Rd(vd));
}
}
void Assembler::ucvtf(const VRegister& fd, const Register& rn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
Emit(SF(rn) | FPType(fd) | UCVTF | Rn(rn) | Rd(fd));
} else {
Emit(SF(rn) | FPType(fd) | UCVTF_fixed | FPScale(64 - fbits) | Rn(rn) |
Rd(fd));
}
}
void Assembler::NEON3Same(const VRegister& vd, const VRegister& vn,
const VRegister& vm, NEON3SameOp vop) {
DCHECK(AreSameFormat(vd, vn, vm));
DCHECK(vd.IsVector() || !vd.IsQ());
Instr format, op = vop;
if (vd.IsScalar()) {
op |= NEON_Q | NEONScalar;
format = SFormat(vd);
} else {
format = VFormat(vd);
}
Emit(format | op | Rm(vm) | Rn(vn) | Rd(vd));
}
void Assembler::NEONFP3Same(const VRegister& vd, const VRegister& vn,
const VRegister& vm, Instr op) {
DCHECK(AreSameFormat(vd, vn, vm));
Emit(FPFormat(vd) | op | Rm(vm) | Rn(vn) | Rd(vd));
}
#define NEON_FP2REGMISC_LIST(V) \
V(fabs, NEON_FABS, FABS) \
V(fneg, NEON_FNEG, FNEG) \
V(fsqrt, NEON_FSQRT, FSQRT) \
V(frintn, NEON_FRINTN, FRINTN) \
V(frinta, NEON_FRINTA, FRINTA) \
V(frintp, NEON_FRINTP, FRINTP) \
V(frintm, NEON_FRINTM, FRINTM) \
V(frintx, NEON_FRINTX, FRINTX) \
V(frintz, NEON_FRINTZ, FRINTZ) \
V(frinti, NEON_FRINTI, FRINTI) \
V(frsqrte, NEON_FRSQRTE, NEON_FRSQRTE_scalar) \
V(frecpe, NEON_FRECPE, NEON_FRECPE_scalar)
#define DEFINE_ASM_FUNC(FN, VEC_OP, SCA_OP) \
void Assembler::FN(const VRegister& vd, const VRegister& vn) { \
Instr op; \
if (vd.IsScalar()) { \
DCHECK(vd.Is1S() || vd.Is1D()); \
op = SCA_OP; \
} else { \
DCHECK(vd.Is2S() || vd.Is2D() || vd.Is4S()); \
op = VEC_OP; \
} \
NEONFP2RegMisc(vd, vn, op); \
}
NEON_FP2REGMISC_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
void Assembler::shll(const VRegister& vd, const VRegister& vn, int shift) {
DCHECK((vd.Is8H() && vn.Is8B() && shift == 8) ||
(vd.Is4S() && vn.Is4H() && shift == 16) ||
(vd.Is2D() && vn.Is2S() && shift == 32));
USE(shift);
Emit(VFormat(vn) | NEON_SHLL | Rn(vn) | Rd(vd));
}
void Assembler::shll2(const VRegister& vd, const VRegister& vn, int shift) {
USE(shift);
DCHECK((vd.Is8H() && vn.Is16B() && shift == 8) ||
(vd.Is4S() && vn.Is8H() && shift == 16) ||
(vd.Is2D() && vn.Is4S() && shift == 32));
Emit(VFormat(vn) | NEON_SHLL | Rn(vn) | Rd(vd));
}
void Assembler::NEONFP2RegMisc(const VRegister& vd, const VRegister& vn,
NEON2RegMiscOp vop, double value) {
DCHECK(AreSameFormat(vd, vn));
DCHECK_EQ(value, 0.0);
USE(value);
Instr op = vop;
if (vd.IsScalar()) {
DCHECK(vd.Is1S() || vd.Is1D());
op |= NEON_Q | NEONScalar;
} else {
DCHECK(vd.Is2S() || vd.Is2D() || vd.Is4S());
}
Emit(FPFormat(vd) | op | Rn(vn) | Rd(vd));
}
void Assembler::fcmeq(const VRegister& vd, const VRegister& vn, double value) {
NEONFP2RegMisc(vd, vn, NEON_FCMEQ_zero, value);
}
void Assembler::fcmge(const VRegister& vd, const VRegister& vn, double value) {
NEONFP2RegMisc(vd, vn, NEON_FCMGE_zero, value);
}
void Assembler::fcmgt(const VRegister& vd, const VRegister& vn, double value) {
NEONFP2RegMisc(vd, vn, NEON_FCMGT_zero, value);
}
void Assembler::fcmle(const VRegister& vd, const VRegister& vn, double value) {
NEONFP2RegMisc(vd, vn, NEON_FCMLE_zero, value);
}
void Assembler::fcmlt(const VRegister& vd, const VRegister& vn, double value) {
NEONFP2RegMisc(vd, vn, NEON_FCMLT_zero, value);
}
void Assembler::frecpx(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsScalar());
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is1S() || vd.Is1D());
Emit(FPFormat(vd) | NEON_FRECPX_scalar | Rn(vn) | Rd(vd));
}
void Assembler::fcvtzs(const Register& rd, const VRegister& vn, int fbits) {
DCHECK(vn.Is1S() || vn.Is1D());
DCHECK((fbits >= 0) && (fbits <= rd.SizeInBits()));
if (fbits == 0) {
Emit(SF(rd) | FPType(vn) | FCVTZS | Rn(vn) | Rd(rd));
} else {
Emit(SF(rd) | FPType(vn) | FCVTZS_fixed | FPScale(64 - fbits) | Rn(vn) |
Rd(rd));
}
}
void Assembler::fcvtzs(const VRegister& vd, const VRegister& vn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
NEONFP2RegMisc(vd, vn, NEON_FCVTZS);
} else {
DCHECK(vd.Is1D() || vd.Is1S() || vd.Is2D() || vd.Is2S() || vd.Is4S());
NEONShiftRightImmediate(vd, vn, fbits, NEON_FCVTZS_imm);
}
}
void Assembler::fcvtzu(const Register& rd, const VRegister& vn, int fbits) {
DCHECK(vn.Is1S() || vn.Is1D());
DCHECK((fbits >= 0) && (fbits <= rd.SizeInBits()));
if (fbits == 0) {
Emit(SF(rd) | FPType(vn) | FCVTZU | Rn(vn) | Rd(rd));
} else {
Emit(SF(rd) | FPType(vn) | FCVTZU_fixed | FPScale(64 - fbits) | Rn(vn) |
Rd(rd));
}
}
void Assembler::fcvtzu(const VRegister& vd, const VRegister& vn, int fbits) {
DCHECK_GE(fbits, 0);
if (fbits == 0) {
NEONFP2RegMisc(vd, vn, NEON_FCVTZU);
} else {
DCHECK(vd.Is1D() || vd.Is1S() || vd.Is2D() || vd.Is2S() || vd.Is4S());
NEONShiftRightImmediate(vd, vn, fbits, NEON_FCVTZU_imm);
}
}
void Assembler::NEONFP2RegMisc(const VRegister& vd, const VRegister& vn,
Instr op) {
DCHECK(AreSameFormat(vd, vn));
Emit(FPFormat(vd) | op | Rn(vn) | Rd(vd));
}
void Assembler::NEON2RegMisc(const VRegister& vd, const VRegister& vn,
NEON2RegMiscOp vop, int value) {
DCHECK(AreSameFormat(vd, vn));
DCHECK_EQ(value, 0);
USE(value);
Instr format, op = vop;
if (vd.IsScalar()) {
op |= NEON_Q | NEONScalar;
format = SFormat(vd);
} else {
format = VFormat(vd);
}
Emit(format | op | Rn(vn) | Rd(vd));
}
void Assembler::cmeq(const VRegister& vd, const VRegister& vn, int value) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_CMEQ_zero, value);
}
void Assembler::cmge(const VRegister& vd, const VRegister& vn, int value) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_CMGE_zero, value);
}
void Assembler::cmgt(const VRegister& vd, const VRegister& vn, int value) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_CMGT_zero, value);
}
void Assembler::cmle(const VRegister& vd, const VRegister& vn, int value) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_CMLE_zero, value);
}
void Assembler::cmlt(const VRegister& vd, const VRegister& vn, int value) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_CMLT_zero, value);
}
#define NEON_3SAME_LIST(V) \
V(add, NEON_ADD, vd.IsVector() || vd.Is1D()) \
V(addp, NEON_ADDP, vd.IsVector() || vd.Is1D()) \
V(sub, NEON_SUB, vd.IsVector() || vd.Is1D()) \
V(cmeq, NEON_CMEQ, vd.IsVector() || vd.Is1D()) \
V(cmge, NEON_CMGE, vd.IsVector() || vd.Is1D()) \
V(cmgt, NEON_CMGT, vd.IsVector() || vd.Is1D()) \
V(cmhi, NEON_CMHI, vd.IsVector() || vd.Is1D()) \
V(cmhs, NEON_CMHS, vd.IsVector() || vd.Is1D()) \
V(cmtst, NEON_CMTST, vd.IsVector() || vd.Is1D()) \
V(sshl, NEON_SSHL, vd.IsVector() || vd.Is1D()) \
V(ushl, NEON_USHL, vd.IsVector() || vd.Is1D()) \
V(srshl, NEON_SRSHL, vd.IsVector() || vd.Is1D()) \
V(urshl, NEON_URSHL, vd.IsVector() || vd.Is1D()) \
V(sqdmulh, NEON_SQDMULH, vd.IsLaneSizeH() || vd.IsLaneSizeS()) \
V(sqrdmulh, NEON_SQRDMULH, vd.IsLaneSizeH() || vd.IsLaneSizeS()) \
V(shadd, NEON_SHADD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(uhadd, NEON_UHADD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(srhadd, NEON_SRHADD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(urhadd, NEON_URHADD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(shsub, NEON_SHSUB, vd.IsVector() && !vd.IsLaneSizeD()) \
V(uhsub, NEON_UHSUB, vd.IsVector() && !vd.IsLaneSizeD()) \
V(smax, NEON_SMAX, vd.IsVector() && !vd.IsLaneSizeD()) \
V(smaxp, NEON_SMAXP, vd.IsVector() && !vd.IsLaneSizeD()) \
V(smin, NEON_SMIN, vd.IsVector() && !vd.IsLaneSizeD()) \
V(sminp, NEON_SMINP, vd.IsVector() && !vd.IsLaneSizeD()) \
V(umax, NEON_UMAX, vd.IsVector() && !vd.IsLaneSizeD()) \
V(umaxp, NEON_UMAXP, vd.IsVector() && !vd.IsLaneSizeD()) \
V(umin, NEON_UMIN, vd.IsVector() && !vd.IsLaneSizeD()) \
V(uminp, NEON_UMINP, vd.IsVector() && !vd.IsLaneSizeD()) \
V(saba, NEON_SABA, vd.IsVector() && !vd.IsLaneSizeD()) \
V(sabd, NEON_SABD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(uaba, NEON_UABA, vd.IsVector() && !vd.IsLaneSizeD()) \
V(uabd, NEON_UABD, vd.IsVector() && !vd.IsLaneSizeD()) \
V(mla, NEON_MLA, vd.IsVector() && !vd.IsLaneSizeD()) \
V(mls, NEON_MLS, vd.IsVector() && !vd.IsLaneSizeD()) \
V(mul, NEON_MUL, vd.IsVector() && !vd.IsLaneSizeD()) \
V(and_, NEON_AND, vd.Is8B() || vd.Is16B()) \
V(orr, NEON_ORR, vd.Is8B() || vd.Is16B()) \
V(orn, NEON_ORN, vd.Is8B() || vd.Is16B()) \
V(eor, NEON_EOR, vd.Is8B() || vd.Is16B()) \
V(bic, NEON_BIC, vd.Is8B() || vd.Is16B()) \
V(bit, NEON_BIT, vd.Is8B() || vd.Is16B()) \
V(bif, NEON_BIF, vd.Is8B() || vd.Is16B()) \
V(bsl, NEON_BSL, vd.Is8B() || vd.Is16B()) \
V(pmul, NEON_PMUL, vd.Is8B() || vd.Is16B()) \
V(uqadd, NEON_UQADD, true) \
V(sqadd, NEON_SQADD, true) \
V(uqsub, NEON_UQSUB, true) \
V(sqsub, NEON_SQSUB, true) \
V(sqshl, NEON_SQSHL, true) \
V(uqshl, NEON_UQSHL, true) \
V(sqrshl, NEON_SQRSHL, true) \
V(uqrshl, NEON_UQRSHL, true)
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm) { \
DCHECK(AS); \
NEON3Same(vd, vn, vm, OP); \
}
NEON_3SAME_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
#define NEON_FP3SAME_LIST_V2(V) \
V(fadd, NEON_FADD, FADD) \
V(fsub, NEON_FSUB, FSUB) \
V(fmul, NEON_FMUL, FMUL) \
V(fdiv, NEON_FDIV, FDIV) \
V(fmax, NEON_FMAX, FMAX) \
V(fmaxnm, NEON_FMAXNM, FMAXNM) \
V(fmin, NEON_FMIN, FMIN) \
V(fminnm, NEON_FMINNM, FMINNM) \
V(fmulx, NEON_FMULX, NEON_FMULX_scalar) \
V(frecps, NEON_FRECPS, NEON_FRECPS_scalar) \
V(frsqrts, NEON_FRSQRTS, NEON_FRSQRTS_scalar) \
V(fabd, NEON_FABD, NEON_FABD_scalar) \
V(fmla, NEON_FMLA, 0) \
V(fmls, NEON_FMLS, 0) \
V(facge, NEON_FACGE, NEON_FACGE_scalar) \
V(facgt, NEON_FACGT, NEON_FACGT_scalar) \
V(fcmeq, NEON_FCMEQ, NEON_FCMEQ_scalar) \
V(fcmge, NEON_FCMGE, NEON_FCMGE_scalar) \
V(fcmgt, NEON_FCMGT, NEON_FCMGT_scalar) \
V(faddp, NEON_FADDP, 0) \
V(fmaxp, NEON_FMAXP, 0) \
V(fminp, NEON_FMINP, 0) \
V(fmaxnmp, NEON_FMAXNMP, 0) \
V(fminnmp, NEON_FMINNMP, 0)
#define DEFINE_ASM_FUNC(FN, VEC_OP, SCA_OP) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm) { \
Instr op; \
if ((SCA_OP != 0) && vd.IsScalar()) { \
DCHECK(vd.Is1S() || vd.Is1D()); \
op = SCA_OP; \
} else { \
DCHECK(vd.IsVector()); \
DCHECK(vd.Is2S() || vd.Is2D() || vd.Is4S()); \
op = VEC_OP; \
} \
NEONFP3Same(vd, vn, vm, op); \
}
NEON_FP3SAME_LIST_V2(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
void Assembler::addp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1D() && vn.Is2D()));
Emit(SFormat(vd) | NEON_ADDP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::faddp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1S() && vn.Is2S()) || (vd.Is1D() && vn.Is2D()));
Emit(FPFormat(vd) | NEON_FADDP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::fmaxp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1S() && vn.Is2S()) || (vd.Is1D() && vn.Is2D()));
Emit(FPFormat(vd) | NEON_FMAXP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::fminp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1S() && vn.Is2S()) || (vd.Is1D() && vn.Is2D()));
Emit(FPFormat(vd) | NEON_FMINP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::fmaxnmp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1S() && vn.Is2S()) || (vd.Is1D() && vn.Is2D()));
Emit(FPFormat(vd) | NEON_FMAXNMP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::fminnmp(const VRegister& vd, const VRegister& vn) {
DCHECK((vd.Is1S() && vn.Is2S()) || (vd.Is1D() && vn.Is2D()));
Emit(FPFormat(vd) | NEON_FMINNMP_scalar | Rn(vn) | Rd(vd));
}
void Assembler::orr(const VRegister& vd, const int imm8, const int left_shift) {
NEONModifiedImmShiftLsl(vd, imm8, left_shift, NEONModifiedImmediate_ORR);
}
void Assembler::mov(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
if (vd.IsD()) {
orr(vd.V8B(), vn.V8B(), vn.V8B());
} else {
DCHECK(vd.IsQ());
orr(vd.V16B(), vn.V16B(), vn.V16B());
}
}
void Assembler::bic(const VRegister& vd, const int imm8, const int left_shift) {
NEONModifiedImmShiftLsl(vd, imm8, left_shift, NEONModifiedImmediate_BIC);
}
void Assembler::movi(const VRegister& vd, const uint64_t imm, Shift shift,
const int shift_amount) {
DCHECK((shift == LSL) || (shift == MSL));
if (vd.Is2D() || vd.Is1D()) {
DCHECK_EQ(shift_amount, 0);
int imm8 = 0;
for (int i = 0; i < 8; ++i) {
int byte = (imm >> (i * 8)) & 0xFF;
DCHECK((byte == 0) || (byte == 0xFF));
if (byte == 0xFF) {
imm8 |= (1 << i);
}
}
Instr q = vd.Is2D() ? NEON_Q : 0;
Emit(q | NEONModImmOp(1) | NEONModifiedImmediate_MOVI |
ImmNEONabcdefgh(imm8) | NEONCmode(0xE) | Rd(vd));
} else if (shift == LSL) {
NEONModifiedImmShiftLsl(vd, static_cast<int>(imm), shift_amount,
NEONModifiedImmediate_MOVI);
} else {
NEONModifiedImmShiftMsl(vd, static_cast<int>(imm), shift_amount,
NEONModifiedImmediate_MOVI);
}
}
void Assembler::mvn(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
if (vd.IsD()) {
not_(vd.V8B(), vn.V8B());
} else {
DCHECK(vd.IsQ());
not_(vd.V16B(), vn.V16B());
}
}
void Assembler::mvni(const VRegister& vd, const int imm8, Shift shift,
const int shift_amount) {
DCHECK((shift == LSL) || (shift == MSL));
if (shift == LSL) {
NEONModifiedImmShiftLsl(vd, imm8, shift_amount, NEONModifiedImmediate_MVNI);
} else {
NEONModifiedImmShiftMsl(vd, imm8, shift_amount, NEONModifiedImmediate_MVNI);
}
}
void Assembler::NEONFPByElement(const VRegister& vd, const VRegister& vn,
const VRegister& vm, int vm_index,
NEONByIndexedElementOp vop) {
DCHECK(AreSameFormat(vd, vn));
DCHECK((vd.Is2S() && vm.Is1S()) || (vd.Is4S() && vm.Is1S()) ||
(vd.Is1S() && vm.Is1S()) || (vd.Is2D() && vm.Is1D()) ||
(vd.Is1D() && vm.Is1D()));
DCHECK((vm.Is1S() && (vm_index < 4)) || (vm.Is1D() && (vm_index < 2)));
Instr op = vop;
int index_num_bits = vm.Is1S() ? 2 : 1;
if (vd.IsScalar()) {
op |= NEON_Q | NEONScalar;
}
Emit(FPFormat(vd) | op | ImmNEONHLM(vm_index, index_num_bits) | Rm(vm) |
Rn(vn) | Rd(vd));
}
void Assembler::NEONByElement(const VRegister& vd, const VRegister& vn,
const VRegister& vm, int vm_index,
NEONByIndexedElementOp vop) {
DCHECK(AreSameFormat(vd, vn));
DCHECK((vd.Is4H() && vm.Is1H()) || (vd.Is8H() && vm.Is1H()) ||
(vd.Is1H() && vm.Is1H()) || (vd.Is2S() && vm.Is1S()) ||
(vd.Is4S() && vm.Is1S()) || (vd.Is1S() && vm.Is1S()));
DCHECK((vm.Is1H() && (vm.code() < 16) && (vm_index < 8)) ||
(vm.Is1S() && (vm_index < 4)));
Instr format, op = vop;
int index_num_bits = vm.Is1H() ? 3 : 2;
if (vd.IsScalar()) {
op |= NEONScalar | NEON_Q;
format = SFormat(vn);
} else {
format = VFormat(vn);
}
Emit(format | op | ImmNEONHLM(vm_index, index_num_bits) | Rm(vm) | Rn(vn) |
Rd(vd));
}
void Assembler::NEONByElementL(const VRegister& vd, const VRegister& vn,
const VRegister& vm, int vm_index,
NEONByIndexedElementOp vop) {
DCHECK((vd.Is4S() && vn.Is4H() && vm.Is1H()) ||
(vd.Is4S() && vn.Is8H() && vm.Is1H()) ||
(vd.Is1S() && vn.Is1H() && vm.Is1H()) ||
(vd.Is2D() && vn.Is2S() && vm.Is1S()) ||
(vd.Is2D() && vn.Is4S() && vm.Is1S()) ||
(vd.Is1D() && vn.Is1S() && vm.Is1S()));
DCHECK((vm.Is1H() && (vm.code() < 16) && (vm_index < 8)) ||
(vm.Is1S() && (vm_index < 4)));
Instr format, op = vop;
int index_num_bits = vm.Is1H() ? 3 : 2;
if (vd.IsScalar()) {
op |= NEONScalar | NEON_Q;
format = SFormat(vn);
} else {
format = VFormat(vn);
}
Emit(format | op | ImmNEONHLM(vm_index, index_num_bits) | Rm(vm) | Rn(vn) |
Rd(vd));
}
#define NEON_BYELEMENT_LIST(V) \
V(mul, NEON_MUL_byelement, vn.IsVector()) \
V(mla, NEON_MLA_byelement, vn.IsVector()) \
V(mls, NEON_MLS_byelement, vn.IsVector()) \
V(sqdmulh, NEON_SQDMULH_byelement, true) \
V(sqrdmulh, NEON_SQRDMULH_byelement, true)
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm, int vm_index) { \
DCHECK(AS); \
NEONByElement(vd, vn, vm, vm_index, OP); \
}
NEON_BYELEMENT_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
#define NEON_FPBYELEMENT_LIST(V) \
V(fmul, NEON_FMUL_byelement) \
V(fmla, NEON_FMLA_byelement) \
V(fmls, NEON_FMLS_byelement) \
V(fmulx, NEON_FMULX_byelement)
#define DEFINE_ASM_FUNC(FN, OP) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm, int vm_index) { \
NEONFPByElement(vd, vn, vm, vm_index, OP); \
}
NEON_FPBYELEMENT_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
#define NEON_BYELEMENT_LONG_LIST(V) \
V(sqdmull, NEON_SQDMULL_byelement, vn.IsScalar() || vn.IsD()) \
V(sqdmull2, NEON_SQDMULL_byelement, vn.IsVector() && vn.IsQ()) \
V(sqdmlal, NEON_SQDMLAL_byelement, vn.IsScalar() || vn.IsD()) \
V(sqdmlal2, NEON_SQDMLAL_byelement, vn.IsVector() && vn.IsQ()) \
V(sqdmlsl, NEON_SQDMLSL_byelement, vn.IsScalar() || vn.IsD()) \
V(sqdmlsl2, NEON_SQDMLSL_byelement, vn.IsVector() && vn.IsQ()) \
V(smull, NEON_SMULL_byelement, vn.IsVector() && vn.IsD()) \
V(smull2, NEON_SMULL_byelement, vn.IsVector() && vn.IsQ()) \
V(umull, NEON_UMULL_byelement, vn.IsVector() && vn.IsD()) \
V(umull2, NEON_UMULL_byelement, vn.IsVector() && vn.IsQ()) \
V(smlal, NEON_SMLAL_byelement, vn.IsVector() && vn.IsD()) \
V(smlal2, NEON_SMLAL_byelement, vn.IsVector() && vn.IsQ()) \
V(umlal, NEON_UMLAL_byelement, vn.IsVector() && vn.IsD()) \
V(umlal2, NEON_UMLAL_byelement, vn.IsVector() && vn.IsQ()) \
V(smlsl, NEON_SMLSL_byelement, vn.IsVector() && vn.IsD()) \
V(smlsl2, NEON_SMLSL_byelement, vn.IsVector() && vn.IsQ()) \
V(umlsl, NEON_UMLSL_byelement, vn.IsVector() && vn.IsD()) \
V(umlsl2, NEON_UMLSL_byelement, vn.IsVector() && vn.IsQ())
#define DEFINE_ASM_FUNC(FN, OP, AS) \
void Assembler::FN(const VRegister& vd, const VRegister& vn, \
const VRegister& vm, int vm_index) { \
DCHECK(AS); \
NEONByElementL(vd, vn, vm, vm_index, OP); \
}
NEON_BYELEMENT_LONG_LIST(DEFINE_ASM_FUNC)
#undef DEFINE_ASM_FUNC
void Assembler::suqadd(const VRegister& vd, const VRegister& vn) {
NEON2RegMisc(vd, vn, NEON_SUQADD);
}
void Assembler::usqadd(const VRegister& vd, const VRegister& vn) {
NEON2RegMisc(vd, vn, NEON_USQADD);
}
void Assembler::abs(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_ABS);
}
void Assembler::sqabs(const VRegister& vd, const VRegister& vn) {
NEON2RegMisc(vd, vn, NEON_SQABS);
}
void Assembler::neg(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() || vd.Is1D());
NEON2RegMisc(vd, vn, NEON_NEG);
}
void Assembler::sqneg(const VRegister& vd, const VRegister& vn) {
NEON2RegMisc(vd, vn, NEON_SQNEG);
}
void Assembler::NEONXtn(const VRegister& vd, const VRegister& vn,
NEON2RegMiscOp vop) {
Instr format, op = vop;
if (vd.IsScalar()) {
DCHECK((vd.Is1B() && vn.Is1H()) || (vd.Is1H() && vn.Is1S()) ||
(vd.Is1S() && vn.Is1D()));
op |= NEON_Q | NEONScalar;
format = SFormat(vd);
} else {
DCHECK((vd.Is8B() && vn.Is8H()) || (vd.Is4H() && vn.Is4S()) ||
(vd.Is2S() && vn.Is2D()) || (vd.Is16B() && vn.Is8H()) ||
(vd.Is8H() && vn.Is4S()) || (vd.Is4S() && vn.Is2D()));
format = VFormat(vd);
}
Emit(format | op | Rn(vn) | Rd(vd));
}
void Assembler::xtn(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() && vd.IsD());
NEONXtn(vd, vn, NEON_XTN);
}
void Assembler::xtn2(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() && vd.IsQ());
NEONXtn(vd, vn, NEON_XTN);
}
void Assembler::sqxtn(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsScalar() || vd.IsD());
NEONXtn(vd, vn, NEON_SQXTN);
}
void Assembler::sqxtn2(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() && vd.IsQ());
NEONXtn(vd, vn, NEON_SQXTN);
}
void Assembler::sqxtun(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsScalar() || vd.IsD());
NEONXtn(vd, vn, NEON_SQXTUN);
}
void Assembler::sqxtun2(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() && vd.IsQ());
NEONXtn(vd, vn, NEON_SQXTUN);
}
void Assembler::uqxtn(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsScalar() || vd.IsD());
NEONXtn(vd, vn, NEON_UQXTN);
}
void Assembler::uqxtn2(const VRegister& vd, const VRegister& vn) {
DCHECK(vd.IsVector() && vd.IsQ());
NEONXtn(vd, vn, NEON_UQXTN);
}
// NEON NOT and RBIT are distinguised by bit 22, the bottom bit of "size".
void Assembler::not_(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is8B() || vd.Is16B());
Emit(VFormat(vd) | NEON_RBIT_NOT | Rn(vn) | Rd(vd));
}
void Assembler::rbit(const VRegister& vd, const VRegister& vn) {
DCHECK(AreSameFormat(vd, vn));
DCHECK(vd.Is8B() || vd.Is16B());
Emit(VFormat(vn) | (1 << NEONSize_offset) | NEON_RBIT_NOT | Rn(vn) | Rd(vd));
}
void Assembler::ext(const VRegister& vd, const VRegister& vn,
const VRegister& vm, int index) {
DCHECK(AreSameFormat(vd, vn, vm));
DCHECK(vd.Is8B() || vd.Is16B());
DCHECK((0 <= index) && (index < vd.LaneCount()));
Emit(VFormat(vd) | NEON_EXT | Rm(vm) | ImmNEONExt(index) | Rn(vn) | Rd(vd));
}
void Assembler::dup(const VRegister& vd, const VRegister& vn, int vn_index) {
Instr q, scalar;
// We support vn arguments of the form vn.VxT() or vn.T(), where x is the
// number of lanes, and T is b, h, s or d.
int lane_size = vn.LaneSizeInBytes();
NEONFormatField format;
switch (lane_size) {
case 1:
format = NEON_16B;
break;
case 2:
format = NEON_8H;
break;
case 4:
format = NEON_4S;
break;
default:
DCHECK_EQ(lane_size, 8);
format = NEON_2D;
break;
}
if (vd.IsScalar()) {
q = NEON_Q;
scalar = NEONScalar;
} else {
DCHECK(!vd.Is1D());
q = vd.IsD() ? 0 : NEON_Q;
scalar = 0;
}
Emit(q | scalar | NEON_DUP_ELEMENT | ImmNEON5(format, vn_index) | Rn(vn) |
Rd(vd));
}
void Assembler::dcptr(Label* label) {
RecordRelocInfo(RelocInfo::INTERNAL_REFERENCE);
if (label->is_bound()) {
// The label is bound, so it does not need to be updated and the internal
// reference should be emitted.
//
// In this case, label->pos() returns the offset of the label from the
// start of the buffer.
internal_reference_positions_.push_back(pc_offset());
dc64(reinterpret_cast<uintptr_t>(buffer_ + label->pos()));
} else {
int32_t offset;
if (label->is_linked()) {
// The label is linked, so the internal reference should be added
// onto the end of the label's link chain.
//
// In this case, label->pos() returns the offset of the last linked
// instruction from the start of the buffer.
offset = label->pos() - pc_offset();
DCHECK_NE(offset, kStartOfLabelLinkChain);
} else {
// The label is unused, so it now becomes linked and the internal
// reference is at the start of the new link chain.
offset = kStartOfLabelLinkChain;
}
// The instruction at pc is now the last link in the label's chain.
label->link_to(pc_offset());
// Traditionally the offset to the previous instruction in the chain is
// encoded in the instruction payload (e.g. branch range) but internal
// references are not instructions so while unbound they are encoded as
// two consecutive brk instructions. The two 16-bit immediates are used
// to encode the offset.
offset >>= kInstructionSizeLog2;
DCHECK(is_int32(offset));
uint32_t high16 = unsigned_bitextract_32(31, 16, offset);
uint32_t low16 = unsigned_bitextract_32(15, 0, offset);
brk(high16);
brk(low16);
}
}
// Below, a difference in case for the same letter indicates a
// negated bit. If b is 1, then B is 0.
uint32_t Assembler::FPToImm8(double imm) {
DCHECK(IsImmFP64(imm));
// bits: aBbb.bbbb.bbcd.efgh.0000.0000.0000.0000
// 0000.0000.0000.0000.0000.0000.0000.0000
uint64_t bits = bit_cast<uint64_t>(imm);
// bit7: a000.0000
uint64_t bit7 = ((bits >> 63) & 0x1) << 7;
// bit6: 0b00.0000
uint64_t bit6 = ((bits >> 61) & 0x1) << 6;
// bit5_to_0: 00cd.efgh
uint64_t bit5_to_0 = (bits >> 48) & 0x3F;
return static_cast<uint32_t>(bit7 | bit6 | bit5_to_0);
}
Instr Assembler::ImmFP(double imm) { return FPToImm8(imm) << ImmFP_offset; }
Instr Assembler::ImmNEONFP(double imm) {
return ImmNEONabcdefgh(FPToImm8(imm));
}
// Code generation helpers.
void Assembler::MoveWide(const Register& rd, uint64_t imm, int shift,
MoveWideImmediateOp mov_op) {
// Ignore the top 32 bits of an immediate if we're moving to a W register.
if (rd.Is32Bits()) {
// Check that the top 32 bits are zero (a positive 32-bit number) or top
// 33 bits are one (a negative 32-bit number, sign extended to 64 bits).
DCHECK(((imm >> kWRegSizeInBits) == 0) ||
((imm >> (kWRegSizeInBits - 1)) == 0x1FFFFFFFF));
imm &= kWRegMask;
}
if (shift >= 0) {
// Explicit shift specified.
DCHECK((shift == 0) || (shift == 16) || (shift == 32) || (shift == 48));
DCHECK(rd.Is64Bits() || (shift == 0) || (shift == 16));
shift /= 16;
} else {
// Calculate a new immediate and shift combination to encode the immediate
// argument.
shift = 0;
if ((imm & ~0xFFFFUL) == 0) {
// Nothing to do.
} else if ((imm & ~(0xFFFFUL << 16)) == 0) {
imm >>= 16;
shift = 1;
} else if ((imm & ~(0xFFFFUL << 32)) == 0) {
DCHECK(rd.Is64Bits());
imm >>= 32;
shift = 2;
} else if ((imm & ~(0xFFFFUL << 48)) == 0) {
DCHECK(rd.Is64Bits());
imm >>= 48;
shift = 3;
}
}
DCHECK(is_uint16(imm));
Emit(SF(rd) | MoveWideImmediateFixed | mov_op | Rd(rd) |
ImmMoveWide(static_cast<int>(imm)) | ShiftMoveWide(shift));
}
void Assembler::AddSub(const Register& rd, const Register& rn,
const Operand& operand, FlagsUpdate S, AddSubOp op) {
DCHECK_EQ(rd.SizeInBits(), rn.SizeInBits());
DCHECK(!operand.NeedsRelocation(this));
if (operand.IsImmediate()) {
int64_t immediate = operand.ImmediateValue();
DCHECK(IsImmAddSub(immediate));
Instr dest_reg = (S == SetFlags) ? Rd(rd) : RdSP(rd);
Emit(SF(rd) | AddSubImmediateFixed | op | Flags(S) |
ImmAddSub(static_cast<int>(immediate)) | dest_reg | RnSP(rn));
} else if (operand.IsShiftedRegister()) {
DCHECK_EQ(operand.reg().SizeInBits(), rd.SizeInBits());
DCHECK_NE(operand.shift(), ROR);
// For instructions of the form:
// add/sub wsp, <Wn>, <Wm> [, LSL #0-3 ]
// add/sub <Wd>, wsp, <Wm> [, LSL #0-3 ]
// add/sub wsp, wsp, <Wm> [, LSL #0-3 ]
// adds/subs <Wd>, wsp, <Wm> [, LSL #0-3 ]
// or their 64-bit register equivalents, convert the operand from shifted to
// extended register mode, and emit an add/sub extended instruction.
if (rn.IsSP() || rd.IsSP()) {
DCHECK(!(rd.IsSP() && (S == SetFlags)));
DataProcExtendedRegister(rd, rn, operand.ToExtendedRegister(), S,
AddSubExtendedFixed | op);
} else {
DataProcShiftedRegister(rd, rn, operand, S, AddSubShiftedFixed | op);
}
} else {
DCHECK(operand.IsExtendedRegister());
DataProcExtendedRegister(rd, rn, operand, S, AddSubExtendedFixed | op);
}
}
void Assembler::AddSubWithCarry(const Register& rd, const Register& rn,
const Operand& operand, FlagsUpdate S,
AddSubWithCarryOp op) {
DCHECK_EQ(rd.SizeInBits(), rn.SizeInBits());
DCHECK_EQ(rd.SizeInBits(), operand.reg().SizeInBits());
DCHECK(operand.IsShiftedRegister() && (operand.shift_amount() == 0));
DCHECK(!operand.NeedsRelocation(this));
Emit(SF(rd) | op | Flags(S) | Rm(operand.reg()) | Rn(rn) | Rd(rd));
}
void Assembler::hlt(int code) {
DCHECK(is_uint16(code));
Emit(HLT | ImmException(code));
}
void Assembler::brk(int code) {
DCHECK(is_uint16(code));
Emit(BRK | ImmException(code));
}
void Assembler::EmitStringData(const char* string) {
size_t len = strlen(string) + 1;
DCHECK_LE(RoundUp(len, kInstructionSize), static_cast<size_t>(kGap));
EmitData(string, static_cast<int>(len));
// Pad with nullptr characters until pc_ is aligned.
const char pad[] = {'\0', '\0', '\0', '\0'};
static_assert(sizeof(pad) == kInstructionSize,
"Size of padding must match instruction size.");
EmitData(pad, RoundUp(pc_offset(), kInstructionSize) - pc_offset());
}
void Assembler::debug(const char* message, uint32_t code, Instr params) {
#ifdef USE_SIMULATOR
// Don't generate simulator specific code if we are building a snapshot, which
// might be run on real hardware.
if (!serializer_enabled()) {
// The arguments to the debug marker need to be contiguous in memory, so
// make sure we don't try to emit pools.
BlockPoolsScope scope(this);
Label start;
bind(&start);
// Refer to instructions-arm64.h for a description of the marker and its
// arguments.
hlt(kImmExceptionIsDebug);
DCHECK_EQ(SizeOfCodeGeneratedSince(&start), kDebugCodeOffset);
dc32(code);
DCHECK_EQ(SizeOfCodeGeneratedSince(&start), kDebugParamsOffset);
dc32(params);
DCHECK_EQ(SizeOfCodeGeneratedSince(&start), kDebugMessageOffset);
EmitStringData(message);
hlt(kImmExceptionIsUnreachable);
return;
}
// Fall through if Serializer is enabled.
#endif
if (params & BREAK) {
brk(0);
}
}
void Assembler::Logical(const Register& rd,
const Register& rn,
const Operand& operand,
LogicalOp op) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
DCHECK(!operand.NeedsRelocation(this));
if (operand.IsImmediate()) {
int64_t immediate = operand.ImmediateValue();
unsigned reg_size = rd.SizeInBits();
DCHECK_NE(immediate, 0);
DCHECK_NE(immediate, -1);
DCHECK(rd.Is64Bits() || is_uint32(immediate));
// If the operation is NOT, invert the operation and immediate.
if ((op & NOT) == NOT) {
op = static_cast<LogicalOp>(op & ~NOT);
immediate = rd.Is64Bits() ? ~immediate : (~immediate & kWRegMask);
}
unsigned n, imm_s, imm_r;
if (IsImmLogical(immediate, reg_size, &n, &imm_s, &imm_r)) {
// Immediate can be encoded in the instruction.
LogicalImmediate(rd, rn, n, imm_s, imm_r, op);
} else {
// This case is handled in the macro assembler.
UNREACHABLE();
}
} else {
DCHECK(operand.IsShiftedRegister());
DCHECK(operand.reg().SizeInBits() == rd.SizeInBits());
Instr dp_op = static_cast<Instr>(op | LogicalShiftedFixed);
DataProcShiftedRegister(rd, rn, operand, LeaveFlags, dp_op);
}
}
void Assembler::LogicalImmediate(const Register& rd,
const Register& rn,
unsigned n,
unsigned imm_s,
unsigned imm_r,
LogicalOp op) {
unsigned reg_size = rd.SizeInBits();
Instr dest_reg = (op == ANDS) ? Rd(rd) : RdSP(rd);
Emit(SF(rd) | LogicalImmediateFixed | op | BitN(n, reg_size) |
ImmSetBits(imm_s, reg_size) | ImmRotate(imm_r, reg_size) | dest_reg |
Rn(rn));
}
void Assembler::ConditionalCompare(const Register& rn,
const Operand& operand,
StatusFlags nzcv,
Condition cond,
ConditionalCompareOp op) {
Instr ccmpop;
DCHECK(!operand.NeedsRelocation(this));
if (operand.IsImmediate()) {
int64_t immediate = operand.ImmediateValue();
DCHECK(IsImmConditionalCompare(immediate));
ccmpop = ConditionalCompareImmediateFixed | op |
ImmCondCmp(static_cast<unsigned>(immediate));
} else {
DCHECK(operand.IsShiftedRegister() && (operand.shift_amount() == 0));
ccmpop = ConditionalCompareRegisterFixed | op | Rm(operand.reg());
}
Emit(SF(rn) | ccmpop | Cond(cond) | Rn(rn) | Nzcv(nzcv));
}
void Assembler::DataProcessing1Source(const Register& rd,
const Register& rn,
DataProcessing1SourceOp op) {
DCHECK(rd.SizeInBits() == rn.SizeInBits());
Emit(SF(rn) | op | Rn(rn) | Rd(rd));
}
void Assembler::FPDataProcessing1Source(const VRegister& vd,
const VRegister& vn,
FPDataProcessing1SourceOp op) {
Emit(FPType(vn) | op | Rn(vn) | Rd(vd));
}
void Assembler::FPDataProcessing2Source(const VRegister& fd,
const VRegister& fn,
const VRegister& fm,
FPDataProcessing2SourceOp op) {
DCHECK(fd.SizeInBits() == fn.SizeInBits());
DCHECK(fd.SizeInBits() == fm.SizeInBits());
Emit(FPType(fd) | op | Rm(fm) | Rn(fn) | Rd(fd));
}
void Assembler::FPDataProcessing3Source(const VRegister& fd,
const VRegister& fn,
const VRegister& fm,
const VRegister& fa,
FPDataProcessing3SourceOp op) {
DCHECK(AreSameSizeAndType(fd, fn, fm, fa));
Emit(FPType(fd) | op | Rm(fm) | Rn(fn) | Rd(fd) | Ra(fa));
}
void Assembler::NEONModifiedImmShiftLsl(const VRegister& vd, const int imm8,
const int left_shift,
NEONModifiedImmediateOp op) {
DCHECK(vd.Is8B() || vd.Is16B() || vd.Is4H() || vd.Is8H() || vd.Is2S() ||
vd.Is4S());
DCHECK((left_shift == 0) || (left_shift == 8) || (left_shift == 16) ||
(left_shift == 24));
DCHECK(is_uint8(imm8));
int cmode_1, cmode_2, cmode_3;
if (vd.Is8B() || vd.Is16B()) {
DCHECK_EQ(op, NEONModifiedImmediate_MOVI);
cmode_1 = 1;
cmode_2 = 1;
cmode_3 = 1;
} else {
cmode_1 = (left_shift >> 3) & 1;
cmode_2 = left_shift >> 4;
cmode_3 = 0;
if (vd.Is4H() || vd.Is8H()) {
DCHECK((left_shift == 0) || (left_shift == 8));
cmode_3 = 1;
}
}
int cmode = (cmode_3 << 3) | (cmode_2 << 2) | (cmode_1 << 1);
Instr q = vd.IsQ() ? NEON_Q : 0;
Emit(q | op | ImmNEONabcdefgh(imm8) | NEONCmode(cmode) | Rd(vd));
}
void Assembler::NEONModifiedImmShiftMsl(const VRegister& vd, const int imm8,
const int shift_amount,
NEONModifiedImmediateOp op) {
DCHECK(vd.Is2S() || vd.Is4S());
DCHECK((shift_amount == 8) || (shift_amount == 16));
DCHECK(is_uint8(imm8));
int cmode_0 = (shift_amount >> 4) & 1;
int cmode = 0xC | cmode_0;
Instr q = vd.IsQ() ? NEON_Q : 0;
Emit(q | op | ImmNEONabcdefgh(imm8) | NEONCmode(cmode) | Rd(vd));
}
void Assembler::EmitShift(const Register& rd,
const Register& rn,
Shift shift,
unsigned shift_amount) {
switch (shift) {
case LSL:
lsl(rd, rn, shift_amount);
break;
case LSR:
lsr(rd, rn, shift_amount);
break;
case ASR:
asr(rd, rn, shift_amount);
break;
case ROR:
ror(rd, rn, shift_amount);
break;
default:
UNREACHABLE();
}
}
void Assembler::EmitExtendShift(const Register& rd,
const Register& rn,
Extend extend,
unsigned left_shift) {
DCHECK(rd.SizeInBits() >= rn.SizeInBits());
unsigned reg_size = rd.SizeInBits();
// Use the correct size of register.
Register rn_ = Register::Create(rn.code(), rd.SizeInBits());
// Bits extracted are high_bit:0.
unsigned high_bit = (8 << (extend & 0x3)) - 1;
// Number of bits left in the result that are not introduced by the shift.
unsigned non_shift_bits = (reg_size - left_shift) & (reg_size - 1);
if ((non_shift_bits > high_bit) || (non_shift_bits == 0)) {
switch (extend) {
case UXTB:
case UXTH:
case UXTW: ubfm(rd, rn_, non_shift_bits, high_bit); break;
case SXTB:
case SXTH:
case SXTW: sbfm(rd, rn_, non_shift_bits, high_bit); break;
case UXTX:
case SXTX: {
DCHECK_EQ(rn.SizeInBits(), kXRegSizeInBits);
// Nothing to extend. Just shift.
lsl(rd, rn_, left_shift);
break;
}
default: UNREACHABLE();
}
} else {
// No need to extend as the extended bits would be shifted away.
lsl(rd, rn_, left_shift);
}
}
void Assembler::DataProcShiftedRegister(const Register& rd,
const Register& rn,
const Operand& operand,
FlagsUpdate S,
Instr op) {
DCHECK(operand.IsShiftedRegister());
DCHECK(rn.Is64Bits() || (rn.Is32Bits() && is_uint5(operand.shift_amount())));
DCHECK(!operand.NeedsRelocation(this));
Emit(SF(rd) | op | Flags(S) |
ShiftDP(operand.shift()) | ImmDPShift(operand.shift_amount()) |
Rm(operand.reg()) | Rn(rn) | Rd(rd));
}
void Assembler::DataProcExtendedRegister(const Register& rd,
const Register& rn,
const Operand& operand,
FlagsUpdate S,
Instr op) {
DCHECK(!operand.NeedsRelocation(this));
Instr dest_reg = (S == SetFlags) ? Rd(rd) : RdSP(rd);
Emit(SF(rd) | op | Flags(S) | Rm(operand.reg()) |
ExtendMode(operand.extend()) | ImmExtendShift(operand.shift_amount()) |
dest_reg | RnSP(rn));
}
bool Assembler::IsImmAddSub(int64_t immediate) {
return is_uint12(immediate) ||
(is_uint12(immediate >> 12) && ((immediate & 0xFFF) == 0));
}
void Assembler::LoadStore(const CPURegister& rt,
const MemOperand& addr,
LoadStoreOp op) {
Instr memop = op | Rt(rt) | RnSP(addr.base());
if (addr.IsImmediateOffset()) {
unsigned size = CalcLSDataSize(op);
if (IsImmLSScaled(addr.offset(), size)) {
int offset = static_cast<int>(addr.offset());
// Use the scaled addressing mode.
Emit(LoadStoreUnsignedOffsetFixed | memop |
ImmLSUnsigned(offset >> size));
} else if (IsImmLSUnscaled(addr.offset())) {
int offset = static_cast<int>(addr.offset());
// Use the unscaled addressing mode.
Emit(LoadStoreUnscaledOffsetFixed | memop | ImmLS(offset));
} else {
// This case is handled in the macro assembler.
UNREACHABLE();
}
} else if (addr.IsRegisterOffset()) {
Extend ext = addr.extend();
Shift shift = addr.shift();
unsigned shift_amount = addr.shift_amount();
// LSL is encoded in the option field as UXTX.
if (shift == LSL) {
ext = UXTX;
}
// Shifts are encoded in one bit, indicating a left shift by the memory
// access size.
DCHECK((shift_amount == 0) ||
(shift_amount == static_cast<unsigned>(CalcLSDataSize(op))));
Emit(LoadStoreRegisterOffsetFixed | memop | Rm(addr.regoffset()) |
ExtendMode(ext) | ImmShiftLS((shift_amount > 0) ? 1 : 0));
} else {
// Pre-index and post-index modes.
DCHECK(!rt.Is(addr.base()));
if (IsImmLSUnscaled(addr.offset())) {
int offset = static_cast<int>(addr.offset());
if (addr.IsPreIndex()) {
Emit(LoadStorePreIndexFixed | memop | ImmLS(offset));
} else {
DCHECK(addr.IsPostIndex());
Emit(LoadStorePostIndexFixed | memop | ImmLS(offset));
}
} else {
// This case is handled in the macro assembler.
UNREACHABLE();
}
}
}
bool Assembler::IsImmLSUnscaled(int64_t offset) {
return is_int9(offset);
}
bool Assembler::IsImmLSScaled(int64_t offset, unsigned size) {
bool offset_is_size_multiple = (((offset >> size) << size) == offset);
return offset_is_size_multiple && is_uint12(offset >> size);
}
bool Assembler::IsImmLSPair(int64_t offset, unsigned size) {
bool offset_is_size_multiple = (((offset >> size) << size) == offset);
return offset_is_size_multiple && is_int7(offset >> size);
}
bool Assembler::IsImmLLiteral(int64_t offset) {
int inst_size = static_cast<int>(kInstructionSizeLog2);
bool offset_is_inst_multiple =
(((offset >> inst_size) << inst_size) == offset);
DCHECK_GT(offset, 0);
offset >>= kLoadLiteralScaleLog2;
return offset_is_inst_multiple && is_intn(offset, ImmLLiteral_width);
}
// Test if a given value can be encoded in the immediate field of a logical
// instruction.
// If it can be encoded, the function returns true, and values pointed to by n,
// imm_s and imm_r are updated with immediates encoded in the format required
// by the corresponding fields in the logical instruction.
// If it can not be encoded, the function returns false, and the values pointed
// to by n, imm_s and imm_r are undefined.
bool Assembler::IsImmLogical(uint64_t value,
unsigned width,
unsigned* n,
unsigned* imm_s,
unsigned* imm_r) {
DCHECK((n != nullptr) && (imm_s != nullptr) && (imm_r != nullptr));
DCHECK((width == kWRegSizeInBits) || (width == kXRegSizeInBits));
bool negate = false;
// Logical immediates are encoded using parameters n, imm_s and imm_r using
// the following table:
//
// N imms immr size S R
// 1 ssssss rrrrrr 64 UInt(ssssss) UInt(rrrrrr)
// 0 0sssss xrrrrr 32 UInt(sssss) UInt(rrrrr)
// 0 10ssss xxrrrr 16 UInt(ssss) UInt(rrrr)
// 0 110sss xxxrrr 8 UInt(sss) UInt(rrr)
// 0 1110ss xxxxrr 4 UInt(ss) UInt(rr)
// 0 11110s xxxxxr 2 UInt(s) UInt(r)
// (s bits must not be all set)
//
// A pattern is constructed of size bits, where the least significant S+1 bits
// are set. The pattern is rotated right by R, and repeated across a 32 or
// 64-bit value, depending on destination register width.
//
// Put another way: the basic format of a logical immediate is a single
// contiguous stretch of 1 bits, repeated across the whole word at intervals
// given by a power of 2. To identify them quickly, we first locate the
// lowest stretch of 1 bits, then the next 1 bit above that; that combination
// is different for every logical immediate, so it gives us all the
// information we need to identify the only logical immediate that our input
// could be, and then we simply check if that's the value we actually have.
//
// (The rotation parameter does give the possibility of the stretch of 1 bits
// going 'round the end' of the word. To deal with that, we observe that in
// any situation where that happens the bitwise NOT of the value is also a
// valid logical immediate. So we simply invert the input whenever its low bit
// is set, and then we know that the rotated case can't arise.)
if (value & 1) {
// If the low bit is 1, negate the value, and set a flag to remember that we
// did (so that we can adjust the return values appropriately).
negate = true;
value = ~value;
}
if (width == kWRegSizeInBits) {
// To handle 32-bit logical immediates, the very easiest thing is to repeat
// the input value twice to make a 64-bit word. The correct encoding of that
// as a logical immediate will also be the correct encoding of the 32-bit
// value.
// The most-significant 32 bits may not be zero (ie. negate is true) so
// shift the value left before duplicating it.
value <<= kWRegSizeInBits;
value |= value >> kWRegSizeInBits;
}
// The basic analysis idea: imagine our input word looks like this.
//
// 0011111000111110001111100011111000111110001111100011111000111110
// c b a
// |<--d-->|
//
// We find the lowest set bit (as an actual power-of-2 value, not its index)
// and call it a. Then we add a to our original number, which wipes out the
// bottommost stretch of set bits and replaces it with a 1 carried into the
// next zero bit. Then we look for the new lowest set bit, which is in
// position b, and subtract it, so now our number is just like the original
// but with the lowest stretch of set bits completely gone. Now we find the
// lowest set bit again, which is position c in the diagram above. Then we'll
// measure the distance d between bit positions a and c (using CLZ), and that
// tells us that the only valid logical immediate that could possibly be equal
// to this number is the one in which a stretch of bits running from a to just
// below b is replicated every d bits.
uint64_t a = LargestPowerOf2Divisor(value);
uint64_t value_plus_a = value + a;
uint64_t b = LargestPowerOf2Divisor(value_plus_a);
uint64_t value_plus_a_minus_b = value_plus_a - b;
uint64_t c = LargestPowerOf2Divisor(value_plus_a_minus_b);
int d, clz_a, out_n;
uint64_t mask;
if (c != 0) {
// The general case, in which there is more than one stretch of set bits.
// Compute the repeat distance d, and set up a bitmask covering the basic
// unit of repetition (i.e. a word with the bottom d bits set). Also, in all
// of these cases the N bit of the output will be zero.
clz_a = CountLeadingZeros(a, kXRegSizeInBits);
int clz_c = CountLeadingZeros(c, kXRegSizeInBits);
d = clz_a - clz_c;
mask = ((uint64_t{1} << d) - 1);
out_n = 0;
} else {
// Handle degenerate cases.
//
// If any of those 'find lowest set bit' operations didn't find a set bit at
// all, then the word will have been zero thereafter, so in particular the
// last lowest_set_bit operation will have returned zero. So we can test for
// all the special case conditions in one go by seeing if c is zero.
if (a == 0) {
// The input was zero (or all 1 bits, which will come to here too after we
// inverted it at the start of the function), for which we just return
// false.
return false;
} else {
// Otherwise, if c was zero but a was not, then there's just one stretch
// of set bits in our word, meaning that we have the trivial case of
// d == 64 and only one 'repetition'. Set up all the same variables as in
// the general case above, and set the N bit in the output.
clz_a = CountLeadingZeros(a, kXRegSizeInBits);
d = 64;
mask = ~uint64_t{0};
out_n = 1;
}
}
// If the repeat period d is not a power of two, it can't be encoded.
if (!base::bits::IsPowerOfTwo(d)) {
return false;
}
if (((b - a) & ~mask) != 0) {
// If the bit stretch (b - a) does not fit within the mask derived from the
// repeat period, then fail.
return false;
}
// The only possible option is b - a repeated every d bits. Now we're going to
// actually construct the valid logical immediate derived from that
// specification, and see if it equals our original input.
//
// To repeat a value every d bits, we multiply it by a number of the form
// (1 + 2^d + 2^(2d) + ...), i.e. 0x0001000100010001 or similar. These can
// be derived using a table lookup on CLZ(d).
static const uint64_t multipliers[] = {
0x0000000000000001UL,
0x0000000100000001UL,
0x0001000100010001UL,
0x0101010101010101UL,
0x1111111111111111UL,
0x5555555555555555UL,
};
int multiplier_idx = CountLeadingZeros(d, kXRegSizeInBits) - 57;
// Ensure that the index to the multipliers array is within bounds.
DCHECK((multiplier_idx >= 0) &&
(static_cast<size_t>(multiplier_idx) < arraysize(multipliers)));
uint64_t multiplier = multipliers[multiplier_idx];
uint64_t candidate = (b - a) * multiplier;
if (value != candidate) {
// The candidate pattern doesn't match our input value, so fail.
return false;
}
// We have a match! This is a valid logical immediate, so now we have to
// construct the bits and pieces of the instruction encoding that generates
// it.
// Count the set bits in our basic stretch. The special case of clz(0) == -1
// makes the answer come out right for stretches that reach the very top of
// the word (e.g. numbers like 0xFFFFC00000000000).
int clz_b = (b == 0) ? -1 : CountLeadingZeros(b, kXRegSizeInBits);
int s = clz_a - clz_b;
// Decide how many bits to rotate right by, to put the low bit of that basic
// stretch in position a.
int r;
if (negate) {
// If we inverted the input right at the start of this function, here's
// where we compensate: the number of set bits becomes the number of clear
// bits, and the rotation count is based on position b rather than position
// a (since b is the location of the 'lowest' 1 bit after inversion).
s = d - s;
r = (clz_b + 1) & (d - 1);
} else {
r = (clz_a + 1) & (d - 1);
}
// Now we're done, except for having to encode the S output in such a way that
// it gives both the number of set bits and the length of the repeated
// segment. The s field is encoded like this:
//
// imms size S
// ssssss 64 UInt(ssssss)
// 0sssss 32 UInt(sssss)
// 10ssss 16 UInt(ssss)
// 110sss 8 UInt(sss)
// 1110ss 4 UInt(ss)
// 11110s 2 UInt(s)
//
// So we 'or' (-d << 1) with our computed s to form imms.
*n = out_n;
*imm_s = ((-d << 1) | (s - 1)) & 0x3F;
*imm_r = r;
return true;
}
bool Assembler::IsImmConditionalCompare(int64_t immediate) {
return is_uint5(immediate);
}
bool Assembler::IsImmFP32(float imm) {
// Valid values will have the form:
// aBbb.bbbc.defg.h000.0000.0000.0000.0000
uint32_t bits = bit_cast<uint32_t>(imm);
// bits[19..0] are cleared.
if ((bits & 0x7FFFF) != 0) {
return false;
}
// bits[29..25] are all set or all cleared.
uint32_t b_pattern = (bits >> 16) & 0x3E00;
if (b_pattern != 0 && b_pattern != 0x3E00) {
return false;
}
// bit[30] and bit[29] are opposite.
if (((bits ^ (bits << 1)) & 0x40000000) == 0) {
return false;
}
return true;
}
bool Assembler::IsImmFP64(double imm) {
// Valid values will have the form:
// aBbb.bbbb.bbcd.efgh.0000.0000.0000.0000
// 0000.0000.0000.0000.0000.0000.0000.0000
uint64_t bits = bit_cast<uint64_t>(imm);
// bits[47..0] are cleared.
if ((bits & 0xFFFFFFFFFFFFL) != 0) {
return false;
}
// bits[61..54] are all set or all cleared.
uint32_t b_pattern = (bits >> 48) & 0x3FC0;
if (b_pattern != 0 && b_pattern != 0x3FC0) {
return false;
}
// bit[62] and bit[61] are opposite.
if (((bits ^ (bits << 1)) & 0x4000000000000000L) == 0) {
return false;
}
return true;
}
void Assembler::GrowBuffer() {
if (!own_buffer_) FATAL("external code buffer is too small");
// Compute new buffer size.
CodeDesc desc; // the new buffer
if (buffer_size_ < 1 * MB) {
desc.buffer_size = 2 * buffer_size_;
} else {
desc.buffer_size = buffer_size_ + 1 * MB;
}
// Some internal data structures overflow for very large buffers,
// they must ensure that kMaximalBufferSize is not too large.
if (desc.buffer_size > kMaximalBufferSize) {
V8::FatalProcessOutOfMemory("Assembler::GrowBuffer");
}
byte* buffer = reinterpret_cast<byte*>(buffer_);
// Set up new buffer.
desc.buffer = NewArray<byte>(desc.buffer_size);
desc.origin = this;
desc.instr_size = pc_offset();
desc.reloc_size =
static_cast<int>((buffer + buffer_size_) - reloc_info_writer.pos());
// Copy the data.
intptr_t pc_delta = desc.buffer - buffer;
intptr_t rc_delta = (desc.buffer + desc.buffer_size) -
(buffer + buffer_size_);
memmove(desc.buffer, buffer, desc.instr_size);
memmove(reloc_info_writer.pos() + rc_delta,
reloc_info_writer.pos(), desc.reloc_size);
// Switch buffers.
DeleteArray(buffer_);
buffer_ = desc.buffer;
buffer_size_ = desc.buffer_size;
pc_ = reinterpret_cast<byte*>(pc_) + pc_delta;
reloc_info_writer.Reposition(reloc_info_writer.pos() + rc_delta,
reloc_info_writer.last_pc() + pc_delta);
// None of our relocation types are pc relative pointing outside the code
// buffer nor pc absolute pointing inside the code buffer, so there is no need
// to relocate any emitted relocation entries.
// Relocate internal references.
for (auto pos : internal_reference_positions_) {
intptr_t* p = reinterpret_cast<intptr_t*>(buffer_ + pos);
*p += pc_delta;
}
// Pending relocation entries are also relative, no need to relocate.
}
void Assembler::RecordRelocInfo(RelocInfo::Mode rmode, intptr_t data) {
// We do not try to reuse pool constants.
RelocInfo rinfo(reinterpret_cast<byte*>(pc_), rmode, data, nullptr);
bool write_reloc_info = true;
if ((rmode == RelocInfo::COMMENT) ||
(rmode == RelocInfo::INTERNAL_REFERENCE) ||
(rmode == RelocInfo::CONST_POOL) || (rmode == RelocInfo::VENEER_POOL) ||
(rmode == RelocInfo::DEOPT_SCRIPT_OFFSET) ||
(rmode == RelocInfo::DEOPT_INLINING_ID) ||
(rmode == RelocInfo::DEOPT_REASON) || (rmode == RelocInfo::DEOPT_ID)) {
// Adjust code for new modes.
DCHECK(RelocInfo::IsComment(rmode) || RelocInfo::IsDeoptReason(rmode) ||
RelocInfo::IsDeoptId(rmode) || RelocInfo::IsDeoptPosition(rmode) ||
RelocInfo::IsInternalReference(rmode) ||
RelocInfo::IsConstPool(rmode) || RelocInfo::IsVeneerPool(rmode));
// These modes do not need an entry in the constant pool.
} else {
write_reloc_info = constpool_.RecordEntry(data, rmode);
// Make sure the constant pool is not emitted in place of the next
// instruction for which we just recorded relocation info.
BlockConstPoolFor(1);
}
if (!RelocInfo::IsNone(rmode) && write_reloc_info) {
// Don't record external references unless the heap will be serialized.
if (rmode == RelocInfo::EXTERNAL_REFERENCE &&
!serializer_enabled() && !emit_debug_code()) {
return;
}
DCHECK_GE(buffer_space(), kMaxRelocSize); // too late to grow buffer here
reloc_info_writer.Write(&rinfo);
}
}
void Assembler::BlockConstPoolFor(int instructions) {
int pc_limit = pc_offset() + instructions * kInstructionSize;
if (no_const_pool_before_ < pc_limit) {
no_const_pool_before_ = pc_limit;
// Make sure the pool won't be blocked for too long.
DCHECK(pc_limit < constpool_.MaxPcOffset());
}
if (next_constant_pool_check_ < no_const_pool_before_) {
next_constant_pool_check_ = no_const_pool_before_;
}
}
void Assembler::CheckConstPool(bool force_emit, bool require_jump) {
// Some short sequence of instruction mustn't be broken up by constant pool
// emission, such sequences are protected by calls to BlockConstPoolFor and
// BlockConstPoolScope.
if (is_const_pool_blocked()) {
// Something is wrong if emission is forced and blocked at the same time.
DCHECK(!force_emit);
return;
}
// There is nothing to do if there are no pending constant pool entries.
if (constpool_.IsEmpty()) {
// Calculate the offset of the next check.
SetNextConstPoolCheckIn(kCheckConstPoolInterval);
return;
}
// We emit a constant pool when:
// * requested to do so by parameter force_emit (e.g. after each function).
// * the distance to the first instruction accessing the constant pool is
// kApproxMaxDistToConstPool or more.
// * the number of entries in the pool is kApproxMaxPoolEntryCount or more.
int dist = constpool_.DistanceToFirstUse();
int count = constpool_.EntryCount();
if (!force_emit &&
(dist < kApproxMaxDistToConstPool) &&
(count < kApproxMaxPoolEntryCount)) {
return;
}
// Emit veneers for branches that would go out of range during emission of the
// constant pool.
int worst_case_size = constpool_.WorstCaseSize();
CheckVeneerPool(false, require_jump,
kVeneerDistanceMargin + worst_case_size);
// Check that the code buffer is large enough before emitting the constant
// pool (this includes the gap to the relocation information).
int needed_space = worst_case_size + kGap + 1 * kInstructionSize;
while (buffer_space() <= needed_space) {
GrowBuffer();
}
Label size_check;
bind(&size_check);
constpool_.Emit(require_jump);
DCHECK(SizeOfCodeGeneratedSince(&size_check) <=
static_cast<unsigned>(worst_case_size));
// Since a constant pool was just emitted, move the check offset forward by
// the standard interval.
SetNextConstPoolCheckIn(kCheckConstPoolInterval);
}
bool Assembler::ShouldEmitVeneer(int max_reachable_pc, int margin) {
// Account for the branch around the veneers and the guard.
int protection_offset = 2 * kInstructionSize;
return pc_offset() > max_reachable_pc - margin - protection_offset -
static_cast<int>(unresolved_branches_.size() * kMaxVeneerCodeSize);
}
void Assembler::RecordVeneerPool(int location_offset, int size) {
RelocInfo rinfo(buffer_ + location_offset, RelocInfo::VENEER_POOL,
static_cast<intptr_t>(size), nullptr);
reloc_info_writer.Write(&rinfo);
}
void Assembler::EmitVeneers(bool force_emit, bool need_protection, int margin) {
BlockPoolsScope scope(this);
RecordComment("[ Veneers");
// The exact size of the veneer pool must be recorded (see the comment at the
// declaration site of RecordConstPool()), but computing the number of
// veneers that will be generated is not obvious. So instead we remember the
// current position and will record the size after the pool has been
// generated.
Label size_check;
bind(&size_check);
int veneer_pool_relocinfo_loc = pc_offset();
Label end;
if (need_protection) {
b(&end);
}
EmitVeneersGuard();
Label veneer_size_check;
std::multimap<int, FarBranchInfo>::iterator it, it_to_delete;
it = unresolved_branches_.begin();
while (it != unresolved_branches_.end()) {
if (force_emit || ShouldEmitVeneer(it->first, margin)) {
Instruction* branch = InstructionAt(it->second.pc_offset_);
Label* label = it->second.label_;
#ifdef DEBUG
bind(&veneer_size_check);
#endif
// Patch the branch to point to the current position, and emit a branch
// to the label.
Instruction* veneer = reinterpret_cast<Instruction*>(pc_);
RemoveBranchFromLabelLinkChain(branch, label, veneer);
branch->SetImmPCOffsetTarget(isolate_data(), veneer);
b(label);
#ifdef DEBUG
DCHECK(SizeOfCodeGeneratedSince(&veneer_size_check) <=
static_cast<uint64_t>(kMaxVeneerCodeSize));
veneer_size_check.Unuse();
#endif
it_to_delete = it++;
unresolved_branches_.erase(it_to_delete);
} else {
++it;
}
}
// Record the veneer pool size.
int pool_size = static_cast<int>(SizeOfCodeGeneratedSince(&size_check));
RecordVeneerPool(veneer_pool_relocinfo_loc, pool_size);
if (unresolved_branches_.empty()) {
next_veneer_pool_check_ = kMaxInt;
} else {
next_veneer_pool_check_ =
unresolved_branches_first_limit() - kVeneerDistanceCheckMargin;
}
bind(&end);
RecordComment("]");
}
void Assembler::CheckVeneerPool(bool force_emit, bool require_jump,
int margin) {
// There is nothing to do if there are no pending veneer pool entries.
if (unresolved_branches_.empty()) {
DCHECK_EQ(next_veneer_pool_check_, kMaxInt);
return;
}
DCHECK(pc_offset() < unresolved_branches_first_limit());
// Some short sequence of instruction mustn't be broken up by veneer pool
// emission, such sequences are protected by calls to BlockVeneerPoolFor and
// BlockVeneerPoolScope.
if (is_veneer_pool_blocked()) {
DCHECK(!force_emit);
return;
}
if (!require_jump) {
// Prefer emitting veneers protected by an existing instruction.
margin *= kVeneerNoProtectionFactor;
}
if (force_emit || ShouldEmitVeneers(margin)) {
EmitVeneers(force_emit, require_jump, margin);
} else {
next_veneer_pool_check_ =
unresolved_branches_first_limit() - kVeneerDistanceCheckMargin;
}
}
int Assembler::buffer_space() const {
return static_cast<int>(reloc_info_writer.pos() -
reinterpret_cast<byte*>(pc_));
}
void Assembler::RecordConstPool(int size) {
// We only need this for debugger support, to correctly compute offsets in the
// code.
RecordRelocInfo(RelocInfo::CONST_POOL, static_cast<intptr_t>(size));
}
void PatchingAssembler::PatchAdrFar(int64_t target_offset) {
// The code at the current instruction should be:
// adr rd, 0
// nop (adr_far)
// nop (adr_far)
// movz scratch, 0
// Verify the expected code.
Instruction* expected_adr = InstructionAt(0);
CHECK(expected_adr->IsAdr() && (expected_adr->ImmPCRel() == 0));
int rd_code = expected_adr->Rd();
for (int i = 0; i < kAdrFarPatchableNNops; ++i) {
CHECK(InstructionAt((i + 1) * kInstructionSize)->IsNop(ADR_FAR_NOP));
}
Instruction* expected_movz =
InstructionAt((kAdrFarPatchableNInstrs - 1) * kInstructionSize);
CHECK(expected_movz->IsMovz() &&
(expected_movz->ImmMoveWide() == 0) &&
(expected_movz->ShiftMoveWide() == 0));
int scratch_code = expected_movz->Rd();
// Patch to load the correct address.
Register rd = Register::XRegFromCode(rd_code);
Register scratch = Register::XRegFromCode(scratch_code);
// Addresses are only 48 bits.
adr(rd, target_offset & 0xFFFF);
movz(scratch, (target_offset >> 16) & 0xFFFF, 16);
movk(scratch, (target_offset >> 32) & 0xFFFF, 32);
DCHECK_EQ(target_offset >> 48, 0);
add(rd, rd, scratch);
}
} // namespace internal
} // namespace v8
#endif // V8_TARGET_ARCH_ARM64