blob: a615d6749679d068c4ac29ace9c5075ae1d531e2 [file] [log] [blame]
// Copyright (c) 1994-2006 Sun Microsystems Inc.
// 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.
//
// - Redistribution 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 Sun Microsystems or the names of 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.
// The original source code covered by the above license above has been
// modified significantly by Google Inc.
// Copyright 2012 the V8 project authors. All rights reserved.
#include "src/arm/assembler-arm.h"
#if V8_TARGET_ARCH_ARM
#include "src/arm/assembler-arm-inl.h"
#include "src/assembler-inl.h"
#include "src/base/bits.h"
#include "src/base/cpu.h"
#include "src/code-stubs.h"
#include "src/macro-assembler.h"
#include "src/objects-inl.h"
namespace v8 {
namespace internal {
static const unsigned kArmv6 = 0u;
static const unsigned kArmv7 = kArmv6 | (1u << ARMv7);
static const unsigned kArmv7WithSudiv = kArmv7 | (1u << ARMv7_SUDIV);
static const unsigned kArmv8 = kArmv7WithSudiv | (1u << ARMv8);
static unsigned CpuFeaturesFromCommandLine() {
unsigned result;
if (strcmp(FLAG_arm_arch, "armv8") == 0) {
result = kArmv8;
} else if (strcmp(FLAG_arm_arch, "armv7+sudiv") == 0) {
result = kArmv7WithSudiv;
} else if (strcmp(FLAG_arm_arch, "armv7") == 0) {
result = kArmv7;
} else if (strcmp(FLAG_arm_arch, "armv6") == 0) {
result = kArmv6;
} else {
fprintf(stderr, "Error: unrecognised value for --arm-arch ('%s').\n",
FLAG_arm_arch);
fprintf(stderr,
"Supported values are: armv8\n"
" armv7+sudiv\n"
" armv7\n"
" armv6\n");
FATAL("arm-arch");
}
// If any of the old (deprecated) flags are specified, print a warning, but
// otherwise try to respect them for now.
// TODO(jbramley): When all the old bots have been updated, remove this.
if (FLAG_enable_armv7.has_value || FLAG_enable_vfp3.has_value ||
FLAG_enable_32dregs.has_value || FLAG_enable_neon.has_value ||
FLAG_enable_sudiv.has_value || FLAG_enable_armv8.has_value) {
// As an approximation of the old behaviour, set the default values from the
// arm_arch setting, then apply the flags over the top.
bool enable_armv7 = (result & (1u << ARMv7)) != 0;
bool enable_vfp3 = (result & (1u << ARMv7)) != 0;
bool enable_32dregs = (result & (1u << ARMv7)) != 0;
bool enable_neon = (result & (1u << ARMv7)) != 0;
bool enable_sudiv = (result & (1u << ARMv7_SUDIV)) != 0;
bool enable_armv8 = (result & (1u << ARMv8)) != 0;
if (FLAG_enable_armv7.has_value) {
fprintf(stderr,
"Warning: --enable_armv7 is deprecated. "
"Use --arm_arch instead.\n");
enable_armv7 = FLAG_enable_armv7.value;
}
if (FLAG_enable_vfp3.has_value) {
fprintf(stderr,
"Warning: --enable_vfp3 is deprecated. "
"Use --arm_arch instead.\n");
enable_vfp3 = FLAG_enable_vfp3.value;
}
if (FLAG_enable_32dregs.has_value) {
fprintf(stderr,
"Warning: --enable_32dregs is deprecated. "
"Use --arm_arch instead.\n");
enable_32dregs = FLAG_enable_32dregs.value;
}
if (FLAG_enable_neon.has_value) {
fprintf(stderr,
"Warning: --enable_neon is deprecated. "
"Use --arm_arch instead.\n");
enable_neon = FLAG_enable_neon.value;
}
if (FLAG_enable_sudiv.has_value) {
fprintf(stderr,
"Warning: --enable_sudiv is deprecated. "
"Use --arm_arch instead.\n");
enable_sudiv = FLAG_enable_sudiv.value;
}
if (FLAG_enable_armv8.has_value) {
fprintf(stderr,
"Warning: --enable_armv8 is deprecated. "
"Use --arm_arch instead.\n");
enable_armv8 = FLAG_enable_armv8.value;
}
// Emulate the old implications.
if (enable_armv8) {
enable_vfp3 = true;
enable_neon = true;
enable_32dregs = true;
enable_sudiv = true;
}
// Select the best available configuration.
if (enable_armv7 && enable_vfp3 && enable_32dregs && enable_neon) {
if (enable_sudiv) {
if (enable_armv8) {
result = kArmv8;
} else {
result = kArmv7WithSudiv;
}
} else {
result = kArmv7;
}
} else {
result = kArmv6;
}
}
return result;
}
// Get the CPU features enabled by the build.
// For cross compilation the preprocessor symbols such as
// CAN_USE_ARMV7_INSTRUCTIONS and CAN_USE_VFP3_INSTRUCTIONS can be used to
// enable ARMv7 and VFPv3 instructions when building the snapshot. However,
// these flags should be consistent with a supported ARM configuration:
// "armv6": ARMv6 + VFPv2
// "armv7": ARMv7 + VFPv3-D32 + NEON
// "armv7+sudiv": ARMv7 + VFPv4-D32 + NEON + SUDIV
// "armv8": ARMv8 (+ all of the above)
static constexpr unsigned CpuFeaturesFromCompiler() {
// TODO(jbramley): Once the build flags are simplified, these tests should
// also be simplified.
// Check *architectural* implications.
#if defined(CAN_USE_ARMV8_INSTRUCTIONS) && !defined(CAN_USE_ARMV7_INSTRUCTIONS)
#error "CAN_USE_ARMV8_INSTRUCTIONS should imply CAN_USE_ARMV7_INSTRUCTIONS"
#endif
#if defined(CAN_USE_ARMV8_INSTRUCTIONS) && !defined(CAN_USE_SUDIV)
#error "CAN_USE_ARMV8_INSTRUCTIONS should imply CAN_USE_SUDIV"
#endif
#if defined(CAN_USE_ARMV7_INSTRUCTIONS) != defined(CAN_USE_VFP3_INSTRUCTIONS)
// V8 requires VFP, and all ARMv7 devices with VFP have VFPv3. Similarly,
// VFPv3 isn't available before ARMv7.
#error "CAN_USE_ARMV7_INSTRUCTIONS should match CAN_USE_VFP3_INSTRUCTIONS"
#endif
#if defined(CAN_USE_NEON) && !defined(CAN_USE_ARMV7_INSTRUCTIONS)
#error "CAN_USE_NEON should imply CAN_USE_ARMV7_INSTRUCTIONS"
#endif
// Find compiler-implied features.
#if defined(CAN_USE_ARMV8_INSTRUCTIONS) && \
defined(CAN_USE_ARMV7_INSTRUCTIONS) && defined(CAN_USE_SUDIV) && \
defined(CAN_USE_NEON) && defined(CAN_USE_VFP3_INSTRUCTIONS)
return kArmv8;
#elif defined(CAN_USE_ARMV7_INSTRUCTIONS) && defined(CAN_USE_SUDIV) && \
defined(CAN_USE_NEON) && defined(CAN_USE_VFP3_INSTRUCTIONS)
return kArmv7WithSudiv;
#elif defined(CAN_USE_ARMV7_INSTRUCTIONS) && defined(CAN_USE_NEON) && \
defined(CAN_USE_VFP3_INSTRUCTIONS)
return kArmv7;
#else
return kArmv6;
#endif
}
void CpuFeatures::ProbeImpl(bool cross_compile) {
dcache_line_size_ = 64;
unsigned command_line = CpuFeaturesFromCommandLine();
// Only use statically determined features for cross compile (snapshot).
if (cross_compile) {
supported_ |= command_line & CpuFeaturesFromCompiler();
return;
}
#ifndef __arm__
// For the simulator build, use whatever the flags specify.
supported_ |= command_line;
#else // __arm__
// Probe for additional features at runtime.
base::CPU cpu;
// Runtime detection is slightly fuzzy, and some inferences are necessary.
unsigned runtime = kArmv6;
// NEON and VFPv3 imply at least ARMv7-A.
if (cpu.has_neon() && cpu.has_vfp3_d32()) {
DCHECK(cpu.has_vfp3());
runtime |= kArmv7;
if (cpu.has_idiva()) {
runtime |= kArmv7WithSudiv;
if (cpu.architecture() >= 8) {
runtime |= kArmv8;
}
}
}
// Use the best of the features found by CPU detection and those inferred from
// the build system. In both cases, restrict available features using the
// command-line. Note that the command-line flags are very permissive (kArmv8)
// by default.
supported_ |= command_line & CpuFeaturesFromCompiler();
supported_ |= command_line & runtime;
// Additional tuning options.
// ARM Cortex-A9 and Cortex-A5 have 32 byte cachelines.
if (cpu.implementer() == base::CPU::ARM &&
(cpu.part() == base::CPU::ARM_CORTEX_A5 ||
cpu.part() == base::CPU::ARM_CORTEX_A9)) {
dcache_line_size_ = 32;
}
#endif
DCHECK_IMPLIES(IsSupported(ARMv7_SUDIV), IsSupported(ARMv7));
DCHECK_IMPLIES(IsSupported(ARMv8), IsSupported(ARMv7_SUDIV));
}
void CpuFeatures::PrintTarget() {
const char* arm_arch = nullptr;
const char* arm_target_type = "";
const char* arm_no_probe = "";
const char* arm_fpu = "";
const char* arm_thumb = "";
const char* arm_float_abi = nullptr;
#if !defined __arm__
arm_target_type = " simulator";
#endif
#if defined ARM_TEST_NO_FEATURE_PROBE
arm_no_probe = " noprobe";
#endif
#if defined CAN_USE_ARMV8_INSTRUCTIONS
arm_arch = "arm v8";
#elif defined CAN_USE_ARMV7_INSTRUCTIONS
arm_arch = "arm v7";
#else
arm_arch = "arm v6";
#endif
#if defined CAN_USE_NEON
arm_fpu = " neon";
#elif defined CAN_USE_VFP3_INSTRUCTIONS
# if defined CAN_USE_VFP32DREGS
arm_fpu = " vfp3";
# else
arm_fpu = " vfp3-d16";
# endif
#else
arm_fpu = " vfp2";
#endif
#ifdef __arm__
arm_float_abi = base::OS::ArmUsingHardFloat() ? "hard" : "softfp";
#elif USE_EABI_HARDFLOAT
arm_float_abi = "hard";
#else
arm_float_abi = "softfp";
#endif
#if defined __arm__ && (defined __thumb__) || (defined __thumb2__)
arm_thumb = " thumb";
#endif
printf("target%s%s %s%s%s %s\n",
arm_target_type, arm_no_probe, arm_arch, arm_fpu, arm_thumb,
arm_float_abi);
}
void CpuFeatures::PrintFeatures() {
printf("ARMv8=%d ARMv7=%d VFPv3=%d VFP32DREGS=%d NEON=%d SUDIV=%d",
CpuFeatures::IsSupported(ARMv8), CpuFeatures::IsSupported(ARMv7),
CpuFeatures::IsSupported(VFPv3), CpuFeatures::IsSupported(VFP32DREGS),
CpuFeatures::IsSupported(NEON), CpuFeatures::IsSupported(SUDIV));
#ifdef __arm__
bool eabi_hardfloat = base::OS::ArmUsingHardFloat();
#elif USE_EABI_HARDFLOAT
bool eabi_hardfloat = true;
#else
bool eabi_hardfloat = false;
#endif
printf(" USE_EABI_HARDFLOAT=%d\n", eabi_hardfloat);
}
// -----------------------------------------------------------------------------
// Implementation of RelocInfo
// static
const int RelocInfo::kApplyMask = 0;
bool RelocInfo::IsCodedSpecially() {
// The deserializer needs to know whether a pointer is specially coded.  Being
// specially coded on ARM means that it is a movw/movt instruction. We don't
// generate those for relocatable pointers.
return false;
}
bool RelocInfo::IsInConstantPool() {
return Assembler::is_constant_pool_load(pc_);
}
Address RelocInfo::embedded_address() const {
return Assembler::target_address_at(pc_, constant_pool_);
}
uint32_t RelocInfo::embedded_size() const {
return reinterpret_cast<uint32_t>(
Assembler::target_address_at(pc_, constant_pool_));
}
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) {
Assembler::set_target_address_at(isolate, pc_, constant_pool_,
reinterpret_cast<Address>(size), flush_mode);
}
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();
}
// -----------------------------------------------------------------------------
// Implementation of Operand and MemOperand
// See assembler-arm-inl.h for inlined constructors
Operand::Operand(Handle<HeapObject> handle) {
rm_ = no_reg;
value_.immediate = reinterpret_cast<intptr_t>(handle.address());
rmode_ = RelocInfo::EMBEDDED_OBJECT;
}
Operand::Operand(Register rm, ShiftOp shift_op, int shift_imm) {
DCHECK(is_uint5(shift_imm));
rm_ = rm;
rs_ = no_reg;
shift_op_ = shift_op;
shift_imm_ = shift_imm & 31;
if ((shift_op == ROR) && (shift_imm == 0)) {
// ROR #0 is functionally equivalent to LSL #0 and this allow us to encode
// RRX as ROR #0 (See below).
shift_op = LSL;
} else if (shift_op == RRX) {
// encoded as ROR with shift_imm == 0
DCHECK_EQ(shift_imm, 0);
shift_op_ = ROR;
shift_imm_ = 0;
}
}
Operand::Operand(Register rm, ShiftOp shift_op, Register rs) {
DCHECK(shift_op != RRX);
rm_ = rm;
rs_ = no_reg;
shift_op_ = shift_op;
rs_ = rs;
}
Operand Operand::EmbeddedNumber(double value) {
int32_t smi;
if (DoubleToSmiInteger(value, &smi)) return Operand(Smi::FromInt(smi));
Operand result(0, RelocInfo::EMBEDDED_OBJECT);
result.is_heap_object_request_ = true;
result.value_.heap_object_request = HeapObjectRequest(value);
return result;
}
Operand Operand::EmbeddedCode(CodeStub* stub) {
Operand result(0, RelocInfo::CODE_TARGET);
result.is_heap_object_request_ = true;
result.value_.heap_object_request = HeapObjectRequest(stub);
return result;
}
MemOperand::MemOperand(Register rn, int32_t offset, AddrMode am)
: rn_(rn), rm_(no_reg), offset_(offset), am_(am) {
// Accesses below the stack pointer are not safe, and are prohibited by the
// ABI. We can check obvious violations here.
if (rn == sp) {
if (am == Offset) DCHECK_LE(0, offset);
if (am == NegOffset) DCHECK_GE(0, offset);
}
}
MemOperand::MemOperand(Register rn, Register rm, AddrMode am)
: rn_(rn), rm_(rm), shift_op_(LSL), shift_imm_(0), am_(am) {}
MemOperand::MemOperand(Register rn, Register rm, ShiftOp shift_op,
int shift_imm, AddrMode am)
: rn_(rn),
rm_(rm),
shift_op_(shift_op),
shift_imm_(shift_imm & 31),
am_(am) {
DCHECK(is_uint5(shift_imm));
}
NeonMemOperand::NeonMemOperand(Register rn, AddrMode am, int align)
: rn_(rn), rm_(am == Offset ? pc : sp) {
DCHECK((am == Offset) || (am == PostIndex));
SetAlignment(align);
}
NeonMemOperand::NeonMemOperand(Register rn, Register rm, int align)
: rn_(rn), rm_(rm) {
SetAlignment(align);
}
void NeonMemOperand::SetAlignment(int align) {
switch (align) {
case 0:
align_ = 0;
break;
case 64:
align_ = 1;
break;
case 128:
align_ = 2;
break;
case 256:
align_ = 3;
break;
default:
UNREACHABLE();
break;
}
}
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(constant_pool_entry_address(pc, 0 /* unused */)) =
object.address();
}
}
// -----------------------------------------------------------------------------
// Specific instructions, constants, and masks.
// str(r, MemOperand(sp, 4, NegPreIndex), al) instruction (aka push(r))
// register r is not encoded.
const Instr kPushRegPattern = al | B26 | 4 | NegPreIndex | sp.code() * B16;
// ldr(r, MemOperand(sp, 4, PostIndex), al) instruction (aka pop(r))
// register r is not encoded.
const Instr kPopRegPattern = al | B26 | L | 4 | PostIndex | sp.code() * B16;
// ldr rd, [pc, #offset]
const Instr kLdrPCImmedMask = 15 * B24 | 7 * B20 | 15 * B16;
const Instr kLdrPCImmedPattern = 5 * B24 | L | pc.code() * B16;
// vldr dd, [pc, #offset]
const Instr kVldrDPCMask = 15 * B24 | 3 * B20 | 15 * B16 | 15 * B8;
const Instr kVldrDPCPattern = 13 * B24 | L | pc.code() * B16 | 11 * B8;
// blxcc rm
const Instr kBlxRegMask =
15 * B24 | 15 * B20 | 15 * B16 | 15 * B12 | 15 * B8 | 15 * B4;
const Instr kBlxRegPattern =
B24 | B21 | 15 * B16 | 15 * B12 | 15 * B8 | BLX;
const Instr kBlxIp = al | kBlxRegPattern | ip.code();
const Instr kMovMvnMask = 0x6D * B21 | 0xF * B16;
const Instr kMovMvnPattern = 0xD * B21;
const Instr kMovMvnFlip = B22;
const Instr kMovLeaveCCMask = 0xDFF * B16;
const Instr kMovLeaveCCPattern = 0x1A0 * B16;
const Instr kMovwPattern = 0x30 * B20;
const Instr kMovtPattern = 0x34 * B20;
const Instr kMovwLeaveCCFlip = 0x5 * B21;
const Instr kMovImmedMask = 0x7F * B21;
const Instr kMovImmedPattern = 0x1D * B21;
const Instr kOrrImmedMask = 0x7F * B21;
const Instr kOrrImmedPattern = 0x1C * B21;
const Instr kCmpCmnMask = 0xDD * B20 | 0xF * B12;
const Instr kCmpCmnPattern = 0x15 * B20;
const Instr kCmpCmnFlip = B21;
const Instr kAddSubFlip = 0x6 * B21;
const Instr kAndBicFlip = 0xE * B21;
// A mask for the Rd register for push, pop, ldr, str instructions.
const Instr kLdrRegFpOffsetPattern = al | B26 | L | Offset | fp.code() * B16;
const Instr kStrRegFpOffsetPattern = al | B26 | Offset | fp.code() * B16;
const Instr kLdrRegFpNegOffsetPattern =
al | B26 | L | NegOffset | fp.code() * B16;
const Instr kStrRegFpNegOffsetPattern = al | B26 | NegOffset | fp.code() * B16;
const Instr kLdrStrInstrTypeMask = 0xFFFF0000;
Assembler::Assembler(IsolateData isolate_data, void* buffer, int buffer_size)
: AssemblerBase(isolate_data, buffer, buffer_size),
pending_32_bit_constants_(),
pending_64_bit_constants_(),
scratch_register_list_(ip.bit()) {
pending_32_bit_constants_.reserve(kMinNumPendingConstants);
pending_64_bit_constants_.reserve(kMinNumPendingConstants);
reloc_info_writer.Reposition(buffer_ + buffer_size_, pc_);
next_buffer_check_ = 0;
code_target_sharing_blocked_nesting_ = 0;
const_pool_blocked_nesting_ = 0;
no_const_pool_before_ = 0;
first_const_pool_32_use_ = -1;
first_const_pool_64_use_ = -1;
last_bound_pos_ = 0;
if (CpuFeatures::IsSupported(VFP32DREGS)) {
// Register objects tend to be abstracted and survive between scopes, so
// it's awkward to use CpuFeatures::VFP32DREGS with CpuFeatureScope. To make
// its use consistent with other features, we always enable it if we can.
EnableCpuFeature(VFP32DREGS);
}
}
Assembler::~Assembler() {
DCHECK_EQ(const_pool_blocked_nesting_, 0);
DCHECK_EQ(code_target_sharing_blocked_nesting_, 0);
}
void Assembler::GetCode(Isolate* isolate, CodeDesc* desc) {
// Emit constant pool if necessary.
int constant_pool_offset = 0;
CheckConstPool(true, false);
DCHECK(pending_32_bit_constants_.empty());
DCHECK(pending_64_bit_constants_.empty());
AllocateAndInstallRequestedHeapObjects(isolate);
// Set up code descriptor.
desc->buffer = buffer_;
desc->buffer_size = buffer_size_;
desc->instr_size = pc_offset();
desc->reloc_size = (buffer_ + buffer_size_) - reloc_info_writer.pos();
desc->constant_pool_size =
(constant_pool_offset ? desc->instr_size - constant_pool_offset : 0);
desc->origin = this;
desc->unwinding_info_size = 0;
desc->unwinding_info = nullptr;
}
void Assembler::Align(int m) {
DCHECK(m >= 4 && base::bits::IsPowerOfTwo(m));
DCHECK_EQ(pc_offset() & (kInstrSize - 1), 0);
while ((pc_offset() & (m - 1)) != 0) {
nop();
}
}
void Assembler::CodeTargetAlign() {
// Preferred alignment of jump targets on some ARM chips.
Align(8);
}
Condition Assembler::GetCondition(Instr instr) {
return Instruction::ConditionField(instr);
}
bool Assembler::IsBranch(Instr instr) {
return (instr & (B27 | B25)) == (B27 | B25);
}
int Assembler::GetBranchOffset(Instr instr) {
DCHECK(IsBranch(instr));
// Take the jump offset in the lower 24 bits, sign extend it and multiply it
// with 4 to get the offset in bytes.
return ((instr & kImm24Mask) << 8) >> 6;
}
bool Assembler::IsLdrRegisterImmediate(Instr instr) {
return (instr & (B27 | B26 | B25 | B22 | B20)) == (B26 | B20);
}
bool Assembler::IsVldrDRegisterImmediate(Instr instr) {
return (instr & (15 * B24 | 3 * B20 | 15 * B8)) == (13 * B24 | B20 | 11 * B8);
}
int Assembler::GetLdrRegisterImmediateOffset(Instr instr) {
DCHECK(IsLdrRegisterImmediate(instr));
bool positive = (instr & B23) == B23;
int offset = instr & kOff12Mask; // Zero extended offset.
return positive ? offset : -offset;
}
int Assembler::GetVldrDRegisterImmediateOffset(Instr instr) {
DCHECK(IsVldrDRegisterImmediate(instr));
bool positive = (instr & B23) == B23;
int offset = instr & kOff8Mask; // Zero extended offset.
offset <<= 2;
return positive ? offset : -offset;
}
Instr Assembler::SetLdrRegisterImmediateOffset(Instr instr, int offset) {
DCHECK(IsLdrRegisterImmediate(instr));
bool positive = offset >= 0;
if (!positive) offset = -offset;
DCHECK(is_uint12(offset));
// Set bit indicating whether the offset should be added.
instr = (instr & ~B23) | (positive ? B23 : 0);
// Set the actual offset.
return (instr & ~kOff12Mask) | offset;
}
Instr Assembler::SetVldrDRegisterImmediateOffset(Instr instr, int offset) {
DCHECK(IsVldrDRegisterImmediate(instr));
DCHECK((offset & ~3) == offset); // Must be 64-bit aligned.
bool positive = offset >= 0;
if (!positive) offset = -offset;
DCHECK(is_uint10(offset));
// Set bit indicating whether the offset should be added.
instr = (instr & ~B23) | (positive ? B23 : 0);
// Set the actual offset. Its bottom 2 bits are zero.
return (instr & ~kOff8Mask) | (offset >> 2);
}
bool Assembler::IsStrRegisterImmediate(Instr instr) {
return (instr & (B27 | B26 | B25 | B22 | B20)) == B26;
}
Instr Assembler::SetStrRegisterImmediateOffset(Instr instr, int offset) {
DCHECK(IsStrRegisterImmediate(instr));
bool positive = offset >= 0;
if (!positive) offset = -offset;
DCHECK(is_uint12(offset));
// Set bit indicating whether the offset should be added.
instr = (instr & ~B23) | (positive ? B23 : 0);
// Set the actual offset.
return (instr & ~kOff12Mask) | offset;
}
bool Assembler::IsAddRegisterImmediate(Instr instr) {
return (instr & (B27 | B26 | B25 | B24 | B23 | B22 | B21)) == (B25 | B23);
}
Instr Assembler::SetAddRegisterImmediateOffset(Instr instr, int offset) {
DCHECK(IsAddRegisterImmediate(instr));
DCHECK_GE(offset, 0);
DCHECK(is_uint12(offset));
// Set the offset.
return (instr & ~kOff12Mask) | offset;
}
Register Assembler::GetRd(Instr instr) {
return Register::from_code(Instruction::RdValue(instr));
}
Register Assembler::GetRn(Instr instr) {
return Register::from_code(Instruction::RnValue(instr));
}
Register Assembler::GetRm(Instr instr) {
return Register::from_code(Instruction::RmValue(instr));
}
bool Assembler::IsPush(Instr instr) {
return ((instr & ~kRdMask) == kPushRegPattern);
}
bool Assembler::IsPop(Instr instr) {
return ((instr & ~kRdMask) == kPopRegPattern);
}
bool Assembler::IsStrRegFpOffset(Instr instr) {
return ((instr & kLdrStrInstrTypeMask) == kStrRegFpOffsetPattern);
}
bool Assembler::IsLdrRegFpOffset(Instr instr) {
return ((instr & kLdrStrInstrTypeMask) == kLdrRegFpOffsetPattern);
}
bool Assembler::IsStrRegFpNegOffset(Instr instr) {
return ((instr & kLdrStrInstrTypeMask) == kStrRegFpNegOffsetPattern);
}
bool Assembler::IsLdrRegFpNegOffset(Instr instr) {
return ((instr & kLdrStrInstrTypeMask) == kLdrRegFpNegOffsetPattern);
}
bool Assembler::IsLdrPcImmediateOffset(Instr instr) {
// Check the instruction is indeed a
// ldr<cond> <Rd>, [pc +/- offset_12].
return (instr & kLdrPCImmedMask) == kLdrPCImmedPattern;
}
bool Assembler::IsVldrDPcImmediateOffset(Instr instr) {
// Check the instruction is indeed a
// vldr<cond> <Dd>, [pc +/- offset_10].
return (instr & kVldrDPCMask) == kVldrDPCPattern;
}
bool Assembler::IsBlxReg(Instr instr) {
// Check the instruction is indeed a
// blxcc <Rm>
return (instr & kBlxRegMask) == kBlxRegPattern;
}
bool Assembler::IsBlxIp(Instr instr) {
// Check the instruction is indeed a
// blx ip
return instr == kBlxIp;
}
bool Assembler::IsTstImmediate(Instr instr) {
return (instr & (B27 | B26 | I | kOpCodeMask | S | kRdMask)) ==
(I | TST | S);
}
bool Assembler::IsCmpRegister(Instr instr) {
return (instr & (B27 | B26 | I | kOpCodeMask | S | kRdMask | B4)) ==
(CMP | S);
}
bool Assembler::IsCmpImmediate(Instr instr) {
return (instr & (B27 | B26 | I | kOpCodeMask | S | kRdMask)) ==
(I | CMP | S);
}
Register Assembler::GetCmpImmediateRegister(Instr instr) {
DCHECK(IsCmpImmediate(instr));
return GetRn(instr);
}
int Assembler::GetCmpImmediateRawImmediate(Instr instr) {
DCHECK(IsCmpImmediate(instr));
return instr & kOff12Mask;
}
// Labels refer to positions in the (to be) generated code.
// There are bound, linked, and unused labels.
//
// Bound labels refer to known positions in the already
// generated code. pos() is the position the label refers to.
//
// Linked labels refer to unknown positions in the code
// to be generated; pos() is the position of the last
// instruction using the label.
//
// The linked labels form a link chain by making the branch offset
// in the instruction steam to point to the previous branch
// instruction using the same label.
//
// The link chain is terminated by a branch offset pointing to the
// same position.
int Assembler::target_at(int pos) {
Instr instr = instr_at(pos);
if (is_uint24(instr)) {
// Emitted link to a label, not part of a branch.
return instr;
}
DCHECK_EQ(5 * B25, instr & 7 * B25); // b, bl, or blx imm24
int imm26 = ((instr & kImm24Mask) << 8) >> 6;
if ((Instruction::ConditionField(instr) == kSpecialCondition) &&
((instr & B24) != 0)) {
// blx uses bit 24 to encode bit 2 of imm26
imm26 += 2;
}
return pos + kPcLoadDelta + imm26;
}
void Assembler::target_at_put(int pos, int target_pos) {
Instr instr = instr_at(pos);
if (is_uint24(instr)) {
DCHECK(target_pos == pos || target_pos >= 0);
// Emitted link to a label, not part of a branch.
// Load the position of the label relative to the generated code object
// pointer in a register.
// The existing code must be a single 24-bit label chain link, followed by
// nops encoding the destination register. See mov_label_offset.
// Extract the destination register from the first nop instructions.
Register dst =
Register::from_code(Instruction::RmValue(instr_at(pos + kInstrSize)));
// In addition to the 24-bit label chain link, we expect to find one nop for
// ARMv7 and above, or two nops for ARMv6. See mov_label_offset.
DCHECK(IsNop(instr_at(pos + kInstrSize), dst.code()));
if (!CpuFeatures::IsSupported(ARMv7)) {
DCHECK(IsNop(instr_at(pos + 2 * kInstrSize), dst.code()));
}
// Here are the instructions we need to emit:
// For ARMv7: target24 => target16_1:target16_0
// movw dst, #target16_0
// movt dst, #target16_1
// For ARMv6: target24 => target8_2:target8_1:target8_0
// mov dst, #target8_0
// orr dst, dst, #target8_1 << 8
// orr dst, dst, #target8_2 << 16
uint32_t target24 = target_pos + (Code::kHeaderSize - kHeapObjectTag);
DCHECK(is_uint24(target24));
if (is_uint8(target24)) {
// If the target fits in a byte then only patch with a mov
// instruction.
PatchingAssembler patcher(isolate_data(),
reinterpret_cast<byte*>(buffer_ + pos), 1);
patcher.mov(dst, Operand(target24));
} else {
uint16_t target16_0 = target24 & kImm16Mask;
uint16_t target16_1 = target24 >> 16;
if (CpuFeatures::IsSupported(ARMv7)) {
// Patch with movw/movt.
if (target16_1 == 0) {
PatchingAssembler patcher(isolate_data(),
reinterpret_cast<byte*>(buffer_ + pos), 1);
CpuFeatureScope scope(&patcher, ARMv7);
patcher.movw(dst, target16_0);
} else {
PatchingAssembler patcher(isolate_data(),
reinterpret_cast<byte*>(buffer_ + pos), 2);
CpuFeatureScope scope(&patcher, ARMv7);
patcher.movw(dst, target16_0);
patcher.movt(dst, target16_1);
}
} else {
// Patch with a sequence of mov/orr/orr instructions.
uint8_t target8_0 = target16_0 & kImm8Mask;
uint8_t target8_1 = target16_0 >> 8;
uint8_t target8_2 = target16_1 & kImm8Mask;
if (target8_2 == 0) {
PatchingAssembler patcher(isolate_data(),
reinterpret_cast<byte*>(buffer_ + pos), 2);
patcher.mov(dst, Operand(target8_0));
patcher.orr(dst, dst, Operand(target8_1 << 8));
} else {
PatchingAssembler patcher(isolate_data(),
reinterpret_cast<byte*>(buffer_ + pos), 3);
patcher.mov(dst, Operand(target8_0));
patcher.orr(dst, dst, Operand(target8_1 << 8));
patcher.orr(dst, dst, Operand(target8_2 << 16));
}
}
}
return;
}
int imm26 = target_pos - (pos + kPcLoadDelta);
DCHECK_EQ(5 * B25, instr & 7 * B25); // b, bl, or blx imm24
if (Instruction::ConditionField(instr) == kSpecialCondition) {
// blx uses bit 24 to encode bit 2 of imm26
DCHECK_EQ(0, imm26 & 1);
instr = (instr & ~(B24 | kImm24Mask)) | ((imm26 & 2) >> 1) * B24;
} else {
DCHECK_EQ(0, imm26 & 3);
instr &= ~kImm24Mask;
}
int imm24 = imm26 >> 2;
DCHECK(is_int24(imm24));
instr_at_put(pos, instr | (imm24 & kImm24Mask));
}
void Assembler::print(const Label* L) {
if (L->is_unused()) {
PrintF("unused label\n");
} else if (L->is_bound()) {
PrintF("bound label to %d\n", L->pos());
} else if (L->is_linked()) {
Label l;
l.link_to(L->pos());
PrintF("unbound label");
while (l.is_linked()) {
PrintF("@ %d ", l.pos());
Instr instr = instr_at(l.pos());
if ((instr & ~kImm24Mask) == 0) {
PrintF("value\n");
} else {
DCHECK_EQ(instr & 7 * B25, 5 * B25); // b, bl, or blx
Condition cond = Instruction::ConditionField(instr);
const char* b;
const char* c;
if (cond == kSpecialCondition) {
b = "blx";
c = "";
} else {
if ((instr & B24) != 0)
b = "bl";
else
b = "b";
switch (cond) {
case eq: c = "eq"; break;
case ne: c = "ne"; break;
case hs: c = "hs"; break;
case lo: c = "lo"; break;
case mi: c = "mi"; break;
case pl: c = "pl"; break;
case vs: c = "vs"; break;
case vc: c = "vc"; break;
case hi: c = "hi"; break;
case ls: c = "ls"; break;
case ge: c = "ge"; break;
case lt: c = "lt"; break;
case gt: c = "gt"; break;
case le: c = "le"; break;
case al: c = ""; break;
default:
c = "";
UNREACHABLE();
}
}
PrintF("%s%s\n", b, c);
}
next(&l);
}
} else {
PrintF("label in inconsistent state (pos = %d)\n", L->pos_);
}
}
void Assembler::bind_to(Label* L, int pos) {
DCHECK(0 <= pos && pos <= pc_offset()); // must have a valid binding position
while (L->is_linked()) {
int fixup_pos = L->pos();
next(L); // call next before overwriting link with target at fixup_pos
target_at_put(fixup_pos, pos);
}
L->bind_to(pos);
// Keep track of the last bound label so we don't eliminate any instructions
// before a bound label.
if (pos > last_bound_pos_)
last_bound_pos_ = pos;
}
void Assembler::bind(Label* L) {
DCHECK(!L->is_bound()); // label can only be bound once
bind_to(L, pc_offset());
}
void Assembler::next(Label* L) {
DCHECK(L->is_linked());
int link = target_at(L->pos());
if (link == L->pos()) {
// Branch target points to the same instruction. This is the end of the link
// chain.
L->Unuse();
} else {
DCHECK_GE(link, 0);
L->link_to(link);
}
}
namespace {
// Low-level code emission routines depending on the addressing mode.
// If this returns true then you have to use the rotate_imm and immed_8
// that it returns, because it may have already changed the instruction
// to match them!
bool FitsShifter(uint32_t imm32, uint32_t* rotate_imm, uint32_t* immed_8,
Instr* instr) {
// imm32 must be unsigned.
for (int rot = 0; rot < 16; rot++) {
uint32_t imm8 = base::bits::RotateLeft32(imm32, 2 * rot);
if ((imm8 <= 0xFF)) {
*rotate_imm = rot;
*immed_8 = imm8;
return true;
}
}
// If the opcode is one with a complementary version and the complementary
// immediate fits, change the opcode.
if (instr != nullptr) {
if ((*instr & kMovMvnMask) == kMovMvnPattern) {
if (FitsShifter(~imm32, rotate_imm, immed_8, nullptr)) {
*instr ^= kMovMvnFlip;
return true;
} else if ((*instr & kMovLeaveCCMask) == kMovLeaveCCPattern) {
if (CpuFeatures::IsSupported(ARMv7)) {
if (imm32 < 0x10000) {
*instr ^= kMovwLeaveCCFlip;
*instr |= Assembler::EncodeMovwImmediate(imm32);
*rotate_imm = *immed_8 = 0; // Not used for movw.
return true;
}
}
}
} else if ((*instr & kCmpCmnMask) == kCmpCmnPattern) {
if (FitsShifter(-static_cast<int>(imm32), rotate_imm, immed_8, nullptr)) {
*instr ^= kCmpCmnFlip;
return true;
}
} else {
Instr alu_insn = (*instr & kALUMask);
if (alu_insn == ADD ||
alu_insn == SUB) {
if (FitsShifter(-static_cast<int>(imm32), rotate_imm, immed_8,
nullptr)) {
*instr ^= kAddSubFlip;
return true;
}
} else if (alu_insn == AND ||
alu_insn == BIC) {
if (FitsShifter(~imm32, rotate_imm, immed_8, nullptr)) {
*instr ^= kAndBicFlip;
return true;
}
}
}
}
return false;
}
// We have to use the temporary register for things that can be relocated even
// if they can be encoded in the ARM's 12 bits of immediate-offset instruction
// space. There is no guarantee that the relocated location can be similarly
// encoded.
bool MustOutputRelocInfo(RelocInfo::Mode rmode, const Assembler* assembler) {
if (rmode == RelocInfo::EXTERNAL_REFERENCE) {
if (assembler != nullptr && assembler->predictable_code_size()) return true;
return assembler->serializer_enabled();
} else if (RelocInfo::IsNone(rmode)) {
return false;
}
return true;
}
bool UseMovImmediateLoad(const Operand& x, const Assembler* assembler) {
DCHECK_NOT_NULL(assembler);
if (x.MustOutputRelocInfo(assembler)) {
// Prefer constant pool if data is likely to be patched.
return false;
} else {
// Otherwise, use immediate load if movw / movt is available.
return CpuFeatures::IsSupported(ARMv7);
}
}
} // namespace
bool Operand::MustOutputRelocInfo(const Assembler* assembler) const {
return v8::internal::MustOutputRelocInfo(rmode_, assembler);
}
int Operand::InstructionsRequired(const Assembler* assembler,
Instr instr) const {
DCHECK_NOT_NULL(assembler);
if (rm_.is_valid()) return 1;
uint32_t dummy1, dummy2;
if (MustOutputRelocInfo(assembler) ||
!FitsShifter(immediate(), &dummy1, &dummy2, &instr)) {
// The immediate operand cannot be encoded as a shifter operand, or use of
// constant pool is required. First account for the instructions required
// for the constant pool or immediate load
int instructions;
if (UseMovImmediateLoad(*this, assembler)) {
DCHECK(CpuFeatures::IsSupported(ARMv7));
// A movw / movt immediate load.
instructions = 2;
} else {
// A small constant pool load.
instructions = 1;
}
if ((instr & ~kCondMask) != 13 * B21) { // mov, S not set
// For a mov or mvn instruction which doesn't set the condition
// code, the constant pool or immediate load is enough, otherwise we need
// to account for the actual instruction being requested.
instructions += 1;
}
return instructions;
} else {
// No use of constant pool and the immediate operand can be encoded as a
// shifter operand.
return 1;
}
}
void Assembler::Move32BitImmediate(Register rd, const Operand& x,
Condition cond) {
if (UseMovImmediateLoad(x, this)) {
// UseMovImmediateLoad should return false when we need to output
// relocation info, since we prefer the constant pool for values that
// can be patched.
DCHECK(!x.MustOutputRelocInfo(this));
UseScratchRegisterScope temps(this);
// Re-use the destination register as a scratch if possible.
Register target = rd != pc ? rd : temps.Acquire();
if (CpuFeatures::IsSupported(ARMv7)) {
uint32_t imm32 = static_cast<uint32_t>(x.immediate());
CpuFeatureScope scope(this, ARMv7);
movw(target, imm32 & 0xFFFF, cond);
movt(target, imm32 >> 16, cond);
}
if (target.code() != rd.code()) {
mov(rd, target, LeaveCC, cond);
}
} else {
int32_t immediate;
if (x.IsHeapObjectRequest()) {
RequestHeapObject(x.heap_object_request());
immediate = 0;
} else {
immediate = x.immediate();
}
ConstantPoolAddEntry(pc_offset(), x.rmode_, immediate);
ldr_pcrel(rd, 0, cond);
}
}
void Assembler::AddrMode1(Instr instr, Register rd, Register rn,
const Operand& x) {
CheckBuffer();
uint32_t opcode = instr & kOpCodeMask;
bool set_flags = (instr & S) != 0;
DCHECK((opcode == ADC) || (opcode == ADD) || (opcode == AND) ||
(opcode == BIC) || (opcode == EOR) || (opcode == ORR) ||
(opcode == RSB) || (opcode == RSC) || (opcode == SBC) ||
(opcode == SUB) || (opcode == CMN) || (opcode == CMP) ||
(opcode == TEQ) || (opcode == TST) || (opcode == MOV) ||
(opcode == MVN));
// For comparison instructions, rd is not defined.
DCHECK(rd.is_valid() || (opcode == CMN) || (opcode == CMP) ||
(opcode == TEQ) || (opcode == TST));
// For move instructions, rn is not defined.
DCHECK(rn.is_valid() || (opcode == MOV) || (opcode == MVN));
DCHECK(rd.is_valid() || rn.is_valid());
DCHECK_EQ(instr & ~(kCondMask | kOpCodeMask | S), 0);
if (!AddrMode1TryEncodeOperand(&instr, x)) {
DCHECK(x.IsImmediate());
// Upon failure to encode, the opcode should not have changed.
DCHECK(opcode == (instr & kOpCodeMask));
Condition cond = Instruction::ConditionField(instr);
if ((opcode == MOV) && !set_flags) {
// Generate a sequence of mov instructions or a load from the constant
// pool only for a MOV instruction which does not set the flags.
DCHECK(!rn.is_valid());
Move32BitImmediate(rd, x, cond);
} else if ((opcode == ADD) && !set_flags && (rd == rn) &&
(scratch_register_list_ == 0)) {
// Split the operation into a sequence of additions if we cannot use a
// scratch register. In this case, we cannot re-use rn and the assembler
// does not have any scratch registers to spare.
uint32_t imm = x.immediate();
do {
// The immediate encoding format is composed of 8 bits of data and 4
// bits encoding a rotation. Each of the 16 possible rotations accounts
// for a rotation by an even number.
// 4 bits -> 16 rotations possible
// -> 16 rotations of 2 bits each fits in a 32-bit value.
// This means that finding the even number of trailing zeroes of the
// immediate allows us to more efficiently split it:
int trailing_zeroes = base::bits::CountTrailingZeros(imm) & ~1u;
uint32_t mask = (0xFF << trailing_zeroes);
add(rd, rd, Operand(imm & mask), LeaveCC, cond);
imm = imm & ~mask;
} while (!ImmediateFitsAddrMode1Instruction(imm));
add(rd, rd, Operand(imm), LeaveCC, cond);
} else {
// The immediate operand cannot be encoded as a shifter operand, so load
// it first to a scratch register and change the original instruction to
// use it.
UseScratchRegisterScope temps(this);
// Re-use the destination register if possible.
Register scratch =
(rd.is_valid() && rd != rn && rd != pc) ? rd : temps.Acquire();
mov(scratch, x, LeaveCC, cond);
AddrMode1(instr, rd, rn, Operand(scratch));
}
return;
}
if (!rd.is_valid()) {
// Emit a comparison instruction.
emit(instr | rn.code() * B16);
} else if (!rn.is_valid()) {
// Emit a move instruction. If the operand is a register-shifted register,
// then prevent the destination from being PC as this is unpredictable.
DCHECK(!x.IsRegisterShiftedRegister() || rd != pc);
emit(instr | rd.code() * B12);
} else {
emit(instr | rn.code() * B16 | rd.code() * B12);
}
if (rn == pc || x.rm_ == pc) {
// Block constant pool emission for one instruction after reading pc.
BlockConstPoolFor(1);
}
}
bool Assembler::AddrMode1TryEncodeOperand(Instr* instr, const Operand& x) {
if (x.IsImmediate()) {
// Immediate.
uint32_t rotate_imm;
uint32_t immed_8;
if (x.MustOutputRelocInfo(this) ||
!FitsShifter(x.immediate(), &rotate_imm, &immed_8, instr)) {
// Let the caller handle generating multiple instructions.
return false;
}
*instr |= I | rotate_imm * B8 | immed_8;
} else if (x.IsImmediateShiftedRegister()) {
*instr |= x.shift_imm_ * B7 | x.shift_op_ | x.rm_.code();
} else {
DCHECK(x.IsRegisterShiftedRegister());
// It is unpredictable to use the PC in this case.
DCHECK(x.rm_ != pc && x.rs_ != pc);
*instr |= x.rs_.code() * B8 | x.shift_op_ | B4 | x.rm_.code();
}
return true;
}
void Assembler::AddrMode2(Instr instr, Register rd, const MemOperand& x) {
DCHECK((instr & ~(kCondMask | B | L)) == B26);
// This method does not handle pc-relative addresses. ldr_pcrel() should be
// used instead.
DCHECK(x.rn_ != pc);
int am = x.am_;
if (!x.rm_.is_valid()) {
// Immediate offset.
int offset_12 = x.offset_;
if (offset_12 < 0) {
offset_12 = -offset_12;
am ^= U;
}
if (!is_uint12(offset_12)) {
// Immediate offset cannot be encoded, load it first to a scratch
// register.
UseScratchRegisterScope temps(this);
// Allow re-using rd for load instructions if possible.
bool is_load = (instr & L) == L;
Register scratch =
(is_load && rd != x.rn_ && rd != pc) ? rd : temps.Acquire();
mov(scratch, Operand(x.offset_), LeaveCC,
Instruction::ConditionField(instr));
AddrMode2(instr, rd, MemOperand(x.rn_, scratch, x.am_));
return;
}
DCHECK_GE(offset_12, 0); // no masking needed
instr |= offset_12;
} else {
// Register offset (shift_imm_ and shift_op_ are 0) or scaled
// register offset the constructors make sure than both shift_imm_
// and shift_op_ are initialized.
DCHECK(x.rm_ != pc);
instr |= B25 | x.shift_imm_*B7 | x.shift_op_ | x.rm_.code();
}
DCHECK((am & (P | W)) == P || x.rn_ != pc); // no pc base with writeback
emit(instr | am | x.rn_.code()*B16 | rd.code()*B12);
}
void Assembler::AddrMode3(Instr instr, Register rd, const MemOperand& x) {
DCHECK((instr & ~(kCondMask | L | S6 | H)) == (B4 | B7));
DCHECK(x.rn_.is_valid());
// This method does not handle pc-relative addresses. ldr_pcrel() should be
// used instead.
DCHECK(x.rn_ != pc);
int am = x.am_;
bool is_load = (instr & L) == L;
if (!x.rm_.is_valid()) {
// Immediate offset.
int offset_8 = x.offset_;
if (offset_8 < 0) {
offset_8 = -offset_8;
am ^= U;
}
if (!is_uint8(offset_8)) {
// Immediate offset cannot be encoded, load it first to a scratch
// register.
UseScratchRegisterScope temps(this);
// Allow re-using rd for load instructions if possible.
Register scratch =
(is_load && rd != x.rn_ && rd != pc) ? rd : temps.Acquire();
mov(scratch, Operand(x.offset_), LeaveCC,
Instruction::ConditionField(instr));
AddrMode3(instr, rd, MemOperand(x.rn_, scratch, x.am_));
return;
}
DCHECK_GE(offset_8, 0); // no masking needed
instr |= B | (offset_8 >> 4) * B8 | (offset_8 & 0xF);
} else if (x.shift_imm_ != 0) {
// Scaled register offsets are not supported, compute the offset separately
// to a scratch register.
UseScratchRegisterScope temps(this);
// Allow re-using rd for load instructions if possible.
Register scratch =
(is_load && rd != x.rn_ && rd != pc) ? rd : temps.Acquire();
mov(scratch, Operand(x.rm_, x.shift_op_, x.shift_imm_), LeaveCC,
Instruction::ConditionField(instr));
AddrMode3(instr, rd, MemOperand(x.rn_, scratch, x.am_));
return;
} else {
// Register offset.
DCHECK((am & (P | W)) == P || x.rm_ != pc); // no pc index with writeback
instr |= x.rm_.code();
}
DCHECK((am & (P | W)) == P || x.rn_ != pc); // no pc base with writeback
emit(instr | am | x.rn_.code()*B16 | rd.code()*B12);
}
void Assembler::AddrMode4(Instr instr, Register rn, RegList rl) {
DCHECK((instr & ~(kCondMask | P | U | W | L)) == B27);
DCHECK_NE(rl, 0);
DCHECK(rn != pc);
emit(instr | rn.code()*B16 | rl);
}
void Assembler::AddrMode5(Instr instr, CRegister crd, const MemOperand& x) {
// Unindexed addressing is not encoded by this function.
DCHECK_EQ((B27 | B26),
(instr & ~(kCondMask | kCoprocessorMask | P | U | N | W | L)));
DCHECK(x.rn_.is_valid() && !x.rm_.is_valid());
int am = x.am_;
int offset_8 = x.offset_;
DCHECK_EQ(offset_8 & 3, 0); // offset must be an aligned word offset
offset_8 >>= 2;
if (offset_8 < 0) {
offset_8 = -offset_8;
am ^= U;
}
DCHECK(is_uint8(offset_8)); // unsigned word offset must fit in a byte
DCHECK((am & (P | W)) == P || x.rn_ != pc); // no pc base with writeback
// Post-indexed addressing requires W == 1; different than in AddrMode2/3.
if ((am & P) == 0)
am |= W;
DCHECK_GE(offset_8, 0); // no masking needed
emit(instr | am | x.rn_.code()*B16 | crd.code()*B12 | offset_8);
}
int Assembler::branch_offset(Label* L) {
int target_pos;
if (L->is_bound()) {
target_pos = L->pos();
} else {
if (L->is_linked()) {
// Point to previous instruction that uses the link.
target_pos = L->pos();
} else {
// First entry of the link chain points to itself.
target_pos = pc_offset();
}
L->link_to(pc_offset());
}
// Block the emission of the constant pool, since the branch instruction must
// be emitted at the pc offset recorded by the label.
if (!is_const_pool_blocked()) BlockConstPoolFor(1);
return target_pos - (pc_offset() + kPcLoadDelta);
}
// Branch instructions.
void Assembler::b(int branch_offset, Condition cond) {
DCHECK_EQ(branch_offset & 3, 0);
int imm24 = branch_offset >> 2;
CHECK(is_int24(imm24));
emit(cond | B27 | B25 | (imm24 & kImm24Mask));
if (cond == al) {
// Dead code is a good location to emit the constant pool.
CheckConstPool(false, false);
}
}
void Assembler::bl(int branch_offset, Condition cond) {
DCHECK_EQ(branch_offset & 3, 0);
int imm24 = branch_offset >> 2;
CHECK(is_int24(imm24));
emit(cond | B27 | B25 | B24 | (imm24 & kImm24Mask));
}
void Assembler::blx(int branch_offset) {
DCHECK_EQ(branch_offset & 1, 0);
int h = ((branch_offset & 2) >> 1)*B24;
int imm24 = branch_offset >> 2;
CHECK(is_int24(imm24));
emit(kSpecialCondition | B27 | B25 | h | (imm24 & kImm24Mask));
}
void Assembler::blx(Register target, Condition cond) {
DCHECK(target != pc);
emit(cond | B24 | B21 | 15*B16 | 15*B12 | 15*B8 | BLX | target.code());
}
void Assembler::bx(Register target, Condition cond) {
DCHECK(target != pc); // use of pc is actually allowed, but discouraged
emit(cond | B24 | B21 | 15*B16 | 15*B12 | 15*B8 | BX | target.code());
}
void Assembler::b(Label* L, Condition cond) {
CheckBuffer();
b(branch_offset(L), cond);
}
void Assembler::bl(Label* L, Condition cond) {
CheckBuffer();
bl(branch_offset(L), cond);
}
void Assembler::blx(Label* L) {
CheckBuffer();
blx(branch_offset(L));
}
// Data-processing instructions.
void Assembler::and_(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | AND | s, dst, src1, src2);
}
void Assembler::eor(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | EOR | s, dst, src1, src2);
}
void Assembler::sub(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | SUB | s, dst, src1, src2);
}
void Assembler::sub(Register dst, Register src1, Register src2, SBit s,
Condition cond) {
sub(dst, src1, Operand(src2), s, cond);
}
void Assembler::rsb(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | RSB | s, dst, src1, src2);
}
void Assembler::add(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | ADD | s, dst, src1, src2);
}
void Assembler::add(Register dst, Register src1, Register src2, SBit s,
Condition cond) {
add(dst, src1, Operand(src2), s, cond);
}
void Assembler::adc(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | ADC | s, dst, src1, src2);
}
void Assembler::sbc(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | SBC | s, dst, src1, src2);
}
void Assembler::rsc(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | RSC | s, dst, src1, src2);
}
void Assembler::tst(Register src1, const Operand& src2, Condition cond) {
AddrMode1(cond | TST | S, no_reg, src1, src2);
}
void Assembler::tst(Register src1, Register src2, Condition cond) {
tst(src1, Operand(src2), cond);
}
void Assembler::teq(Register src1, const Operand& src2, Condition cond) {
AddrMode1(cond | TEQ | S, no_reg, src1, src2);
}
void Assembler::cmp(Register src1, const Operand& src2, Condition cond) {
AddrMode1(cond | CMP | S, no_reg, src1, src2);
}
void Assembler::cmp(Register src1, Register src2, Condition cond) {
cmp(src1, Operand(src2), cond);
}
void Assembler::cmp_raw_immediate(
Register src, int raw_immediate, Condition cond) {
DCHECK(is_uint12(raw_immediate));
emit(cond | I | CMP | S | src.code() << 16 | raw_immediate);
}
void Assembler::cmn(Register src1, const Operand& src2, Condition cond) {
AddrMode1(cond | CMN | S, no_reg, src1, src2);
}
void Assembler::orr(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | ORR | s, dst, src1, src2);
}
void Assembler::orr(Register dst, Register src1, Register src2, SBit s,
Condition cond) {
orr(dst, src1, Operand(src2), s, cond);
}
void Assembler::mov(Register dst, const Operand& src, SBit s, Condition cond) {
// Don't allow nop instructions in the form mov rn, rn to be generated using
// the mov instruction. They must be generated using nop(int/NopMarkerTypes).
DCHECK(!(src.IsRegister() && src.rm() == dst && s == LeaveCC && cond == al));
AddrMode1(cond | MOV | s, dst, no_reg, src);
}
void Assembler::mov(Register dst, Register src, SBit s, Condition cond) {
mov(dst, Operand(src), s, cond);
}
void Assembler::mov_label_offset(Register dst, Label* label) {
if (label->is_bound()) {
mov(dst, Operand(label->pos() + (Code::kHeaderSize - kHeapObjectTag)));
} else {
// Emit the link to the label in the code stream followed by extra nop
// instructions.
// If the label is not linked, then start a new link chain by linking it to
// itself, emitting pc_offset().
int link = label->is_linked() ? label->pos() : pc_offset();
label->link_to(pc_offset());
// When the label is bound, these instructions will be patched with a
// sequence of movw/movt or mov/orr/orr instructions. They will load the
// destination register with the position of the label from the beginning
// of the code.
//
// The link will be extracted from the first instruction and the destination
// register from the second.
// For ARMv7:
// link
// mov dst, dst
// For ARMv6:
// link
// mov dst, dst
// mov dst, dst
//
// When the label gets bound: target_at extracts the link and target_at_put
// patches the instructions.
CHECK(is_uint24(link));
BlockConstPoolScope block_const_pool(this);
emit(link);
nop(dst.code());
if (!CpuFeatures::IsSupported(ARMv7)) {
nop(dst.code());
}
}
}
void Assembler::movw(Register reg, uint32_t immediate, Condition cond) {
DCHECK(IsEnabled(ARMv7));
emit(cond | 0x30*B20 | reg.code()*B12 | EncodeMovwImmediate(immediate));
}
void Assembler::movt(Register reg, uint32_t immediate, Condition cond) {
DCHECK(IsEnabled(ARMv7));
emit(cond | 0x34*B20 | reg.code()*B12 | EncodeMovwImmediate(immediate));
}
void Assembler::bic(Register dst, Register src1, const Operand& src2,
SBit s, Condition cond) {
AddrMode1(cond | BIC | s, dst, src1, src2);
}
void Assembler::mvn(Register dst, const Operand& src, SBit s, Condition cond) {
AddrMode1(cond | MVN | s, dst, no_reg, src);
}
void Assembler::asr(Register dst, Register src1, const Operand& src2, SBit s,
Condition cond) {
if (src2.IsRegister()) {
mov(dst, Operand(src1, ASR, src2.rm()), s, cond);
} else {
mov(dst, Operand(src1, ASR, src2.immediate()), s, cond);
}
}
void Assembler::lsl(Register dst, Register src1, const Operand& src2, SBit s,
Condition cond) {
if (src2.IsRegister()) {
mov(dst, Operand(src1, LSL, src2.rm()), s, cond);
} else {
mov(dst, Operand(src1, LSL, src2.immediate()), s, cond);
}
}
void Assembler::lsr(Register dst, Register src1, const Operand& src2, SBit s,
Condition cond) {
if (src2.IsRegister()) {
mov(dst, Operand(src1, LSR, src2.rm()), s, cond);
} else {
mov(dst, Operand(src1, LSR, src2.immediate()), s, cond);
}
}
// Multiply instructions.
void Assembler::mla(Register dst, Register src1, Register src2, Register srcA,
SBit s, Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc && srcA != pc);
emit(cond | A | s | dst.code()*B16 | srcA.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
void Assembler::mls(Register dst, Register src1, Register src2, Register srcA,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc && srcA != pc);
DCHECK(IsEnabled(ARMv7));
emit(cond | B22 | B21 | dst.code()*B16 | srcA.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
void Assembler::sdiv(Register dst, Register src1, Register src2,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc);
DCHECK(IsEnabled(SUDIV));
emit(cond | B26 | B25 | B24 | B20 | dst.code() * B16 | 0xF * B12 |
src2.code() * B8 | B4 | src1.code());
}
void Assembler::udiv(Register dst, Register src1, Register src2,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc);
DCHECK(IsEnabled(SUDIV));
emit(cond | B26 | B25 | B24 | B21 | B20 | dst.code() * B16 | 0xF * B12 |
src2.code() * B8 | B4 | src1.code());
}
void Assembler::mul(Register dst, Register src1, Register src2, SBit s,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc);
// dst goes in bits 16-19 for this instruction!
emit(cond | s | dst.code() * B16 | src2.code() * B8 | B7 | B4 | src1.code());
}
void Assembler::smmla(Register dst, Register src1, Register src2, Register srcA,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc && srcA != pc);
emit(cond | B26 | B25 | B24 | B22 | B20 | dst.code() * B16 |
srcA.code() * B12 | src2.code() * B8 | B4 | src1.code());
}
void Assembler::smmul(Register dst, Register src1, Register src2,
Condition cond) {
DCHECK(dst != pc && src1 != pc && src2 != pc);
emit(cond | B26 | B25 | B24 | B22 | B20 | dst.code() * B16 | 0xF * B12 |
src2.code() * B8 | B4 | src1.code());
}
void Assembler::smlal(Register dstL,
Register dstH,
Register src1,
Register src2,
SBit s,
Condition cond) {
DCHECK(dstL != pc && dstH != pc && src1 != pc && src2 != pc);
DCHECK(dstL != dstH);
emit(cond | B23 | B22 | A | s | dstH.code()*B16 | dstL.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
void Assembler::smull(Register dstL,
Register dstH,
Register src1,
Register src2,
SBit s,
Condition cond) {
DCHECK(dstL != pc && dstH != pc && src1 != pc && src2 != pc);
DCHECK(dstL != dstH);
emit(cond | B23 | B22 | s | dstH.code()*B16 | dstL.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
void Assembler::umlal(Register dstL,
Register dstH,
Register src1,
Register src2,
SBit s,
Condition cond) {
DCHECK(dstL != pc && dstH != pc && src1 != pc && src2 != pc);
DCHECK(dstL != dstH);
emit(cond | B23 | A | s | dstH.code()*B16 | dstL.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
void Assembler::umull(Register dstL,
Register dstH,
Register src1,
Register src2,
SBit s,
Condition cond) {
DCHECK(dstL != pc && dstH != pc && src1 != pc && src2 != pc);
DCHECK(dstL != dstH);
emit(cond | B23 | s | dstH.code()*B16 | dstL.code()*B12 |
src2.code()*B8 | B7 | B4 | src1.code());
}
// Miscellaneous arithmetic instructions.
void Assembler::clz(Register dst, Register src, Condition cond) {
DCHECK(dst != pc && src != pc);
emit(cond | B24 | B22 | B21 | 15*B16 | dst.code()*B12 |
15*B8 | CLZ | src.code());
}
// Saturating instructions.
// Unsigned saturate.
void Assembler::usat(Register dst,
int satpos,
const Operand& src,
Condition cond) {
DCHECK(dst != pc && src.rm_ != pc);
DCHECK((satpos >= 0) && (satpos <= 31));
DCHECK(src.IsImmediateShiftedRegister());
DCHECK((src.shift_op_ == ASR) || (src.shift_op_ == LSL));
int sh = 0;
if (src.shift_op_ == ASR) {
sh = 1;
}
emit(cond | 0x6 * B24 | 0xE * B20 | satpos * B16 | dst.code() * B12 |
src.shift_imm_ * B7 | sh * B6 | 0x1 * B4 | src.rm_.code());
}
// Bitfield manipulation instructions.
// Unsigned bit field extract.
// Extracts #width adjacent bits from position #lsb in a register, and
// writes them to the low bits of a destination register.
// ubfx dst, src, #lsb, #width
void Assembler::ubfx(Register dst,
Register src,
int lsb,
int width,
Condition cond) {
DCHECK(IsEnabled(ARMv7));
DCHECK(dst != pc && src != pc);
DCHECK((lsb >= 0) && (lsb <= 31));
DCHECK((width >= 1) && (width <= (32 - lsb)));
emit(cond | 0xF * B23 | B22 | B21 | (width - 1) * B16 | dst.code() * B12 |
lsb * B7 | B6 | B4 | src.code());
}
// Signed bit field extract.
// Extracts #width adjacent bits from position #lsb in a register, and
// writes them to the low bits of a destination register. The extracted
// value is sign extended to fill the destination register.
// sbfx dst, src, #lsb, #width
void Assembler::sbfx(Register dst,
Register src,
int lsb,
int width,
Condition cond) {
DCHECK(IsEnabled(ARMv7));
DCHECK(dst != pc && src != pc);
DCHECK((lsb >= 0) && (lsb <= 31));
DCHECK((width >= 1) && (width <= (32 - lsb)));
emit(cond | 0xF * B23 | B21 | (width - 1) * B16 | dst.code() * B12 |
lsb * B7 | B6 | B4 | src.code());
}
// Bit field clear.
// Sets #width adjacent bits at position #lsb in the destination register
// to zero, preserving the value of the other bits.
// bfc dst, #lsb, #width
void Assembler::bfc(Register dst, int lsb, int width, Condition cond) {
DCHECK(IsEnabled(ARMv7));
DCHECK(dst != pc);
DCHECK((lsb >= 0) && (lsb <= 31));
DCHECK((width >= 1) && (width <= (32 - lsb)));
int msb = lsb + width - 1;
emit(cond | 0x1F * B22 | msb * B16 | dst.code() * B12 | lsb * B7 | B4 | 0xF);
}
// Bit field insert.
// Inserts #width adjacent bits from the low bits of the source register
// into position #lsb of the destination register.
// bfi dst, src, #lsb, #width
void Assembler::bfi(Register dst,
Register src,
int lsb,
int width,
Condition cond) {
DCHECK(IsEnabled(ARMv7));
DCHECK(dst != pc && src != pc);
DCHECK((lsb >= 0) && (lsb <= 31));
DCHECK((width >= 1) && (width <= (32 - lsb)));
int msb = lsb + width - 1;
emit(cond | 0x1F * B22 | msb * B16 | dst.code() * B12 | lsb * B7 | B4 |
src.code());
}
void Assembler::pkhbt(Register dst,
Register src1,
const Operand& src2,
Condition cond ) {
// Instruction details available in ARM DDI 0406C.b, A8.8.125.
// cond(31-28) | 01101000(27-20) | Rn(19-16) |
// Rd(15-12) | imm5(11-7) | 0(6) | 01(5-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2.IsImmediateShiftedRegister());
DCHECK(src2.rm() != pc);
DCHECK((src2.shift_imm_ >= 0) && (src2.shift_imm_ <= 31));
DCHECK(src2.shift_op() == LSL);
emit(cond | 0x68*B20 | src1.code()*B16 | dst.code()*B12 |
src2.shift_imm_*B7 | B4 | src2.rm().code());
}
void Assembler::pkhtb(Register dst,
Register src1,
const Operand& src2,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.125.
// cond(31-28) | 01101000(27-20) | Rn(19-16) |
// Rd(15-12) | imm5(11-7) | 1(6) | 01(5-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2.IsImmediateShiftedRegister());
DCHECK(src2.rm() != pc);
DCHECK((src2.shift_imm_ >= 1) && (src2.shift_imm_ <= 32));
DCHECK(src2.shift_op() == ASR);
int asr = (src2.shift_imm_ == 32) ? 0 : src2.shift_imm_;
emit(cond | 0x68*B20 | src1.code()*B16 | dst.code()*B12 |
asr*B7 | B6 | B4 | src2.rm().code());
}
void Assembler::sxtb(Register dst, Register src, int rotate, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.233.
// cond(31-28) | 01101010(27-20) | 1111(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6A * B20 | 0xF * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src.code());
}
void Assembler::sxtab(Register dst, Register src1, Register src2, int rotate,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.233.
// cond(31-28) | 01101010(27-20) | Rn(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6A * B20 | src1.code() * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src2.code());
}
void Assembler::sxth(Register dst, Register src, int rotate, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.235.
// cond(31-28) | 01101011(27-20) | 1111(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6B * B20 | 0xF * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src.code());
}
void Assembler::sxtah(Register dst, Register src1, Register src2, int rotate,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.235.
// cond(31-28) | 01101011(27-20) | Rn(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6B * B20 | src1.code() * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src2.code());
}
void Assembler::uxtb(Register dst, Register src, int rotate, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.274.
// cond(31-28) | 01101110(27-20) | 1111(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6E * B20 | 0xF * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src.code());
}
void Assembler::uxtab(Register dst, Register src1, Register src2, int rotate,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.271.
// cond(31-28) | 01101110(27-20) | Rn(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6E * B20 | src1.code() * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src2.code());
}
void Assembler::uxtb16(Register dst, Register src, int rotate, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.275.
// cond(31-28) | 01101100(27-20) | 1111(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6C * B20 | 0xF * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src.code());
}
void Assembler::uxth(Register dst, Register src, int rotate, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.276.
// cond(31-28) | 01101111(27-20) | 1111(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6F * B20 | 0xF * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src.code());
}
void Assembler::uxtah(Register dst, Register src1, Register src2, int rotate,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.273.
// cond(31-28) | 01101111(27-20) | Rn(19-16) |
// Rd(15-12) | rotate(11-10) | 00(9-8)| 0111(7-4) | Rm(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(rotate == 0 || rotate == 8 || rotate == 16 || rotate == 24);
emit(cond | 0x6F * B20 | src1.code() * B16 | dst.code() * B12 |
((rotate >> 1) & 0xC) * B8 | 7 * B4 | src2.code());
}
void Assembler::rbit(Register dst, Register src, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.144.
// cond(31-28) | 011011111111(27-16) | Rd(15-12) | 11110011(11-4) | Rm(3-0)
DCHECK(IsEnabled(ARMv7));
DCHECK(dst != pc);
DCHECK(src != pc);
emit(cond | 0x6FF * B16 | dst.code() * B12 | 0xF3 * B4 | src.code());
}
// Status register access instructions.
void Assembler::mrs(Register dst, SRegister s, Condition cond) {
DCHECK(dst != pc);
emit(cond | B24 | s | 15*B16 | dst.code()*B12);
}
void Assembler::msr(SRegisterFieldMask fields, const Operand& src,
Condition cond) {
DCHECK_NE(fields & 0x000F0000, 0); // At least one field must be set.
DCHECK(((fields & 0xFFF0FFFF) == CPSR) || ((fields & 0xFFF0FFFF) == SPSR));
Instr instr;
if (src.IsImmediate()) {
// Immediate.
uint32_t rotate_imm;
uint32_t immed_8;
if (src.MustOutputRelocInfo(this) ||
!FitsShifter(src.immediate(), &rotate_imm, &immed_8, nullptr)) {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
// Immediate operand cannot be encoded, load it first to a scratch
// register.
Move32BitImmediate(scratch, src);
msr(fields, Operand(scratch), cond);
return;
}
instr = I | rotate_imm*B8 | immed_8;
} else {
DCHECK(src.IsRegister()); // Only rm is allowed.
instr = src.rm_.code();
}
emit(cond | instr | B24 | B21 | fields | 15*B12);
}
// Load/Store instructions.
void Assembler::ldr(Register dst, const MemOperand& src, Condition cond) {
AddrMode2(cond | B26 | L, dst, src);
}
void Assembler::str(Register src, const MemOperand& dst, Condition cond) {
AddrMode2(cond | B26, src, dst);
}
void Assembler::ldrb(Register dst, const MemOperand& src, Condition cond) {
AddrMode2(cond | B26 | B | L, dst, src);
}
void Assembler::strb(Register src, const MemOperand& dst, Condition cond) {
AddrMode2(cond | B26 | B, src, dst);
}
void Assembler::ldrh(Register dst, const MemOperand& src, Condition cond) {
AddrMode3(cond | L | B7 | H | B4, dst, src);
}
void Assembler::strh(Register src, const MemOperand& dst, Condition cond) {
AddrMode3(cond | B7 | H | B4, src, dst);
}
void Assembler::ldrsb(Register dst, const MemOperand& src, Condition cond) {
AddrMode3(cond | L | B7 | S6 | B4, dst, src);
}
void Assembler::ldrsh(Register dst, const MemOperand& src, Condition cond) {
AddrMode3(cond | L | B7 | S6 | H | B4, dst, src);
}
void Assembler::ldrd(Register dst1, Register dst2,
const MemOperand& src, Condition cond) {
DCHECK(src.rm() == no_reg);
DCHECK(dst1 != lr); // r14.
DCHECK_EQ(0, dst1.code() % 2);
DCHECK_EQ(dst1.code() + 1, dst2.code());
AddrMode3(cond | B7 | B6 | B4, dst1, src);
}
void Assembler::strd(Register src1, Register src2,
const MemOperand& dst, Condition cond) {
DCHECK(dst.rm() == no_reg);
DCHECK(src1 != lr); // r14.
DCHECK_EQ(0, src1.code() % 2);
DCHECK_EQ(src1.code() + 1, src2.code());
AddrMode3(cond | B7 | B6 | B5 | B4, src1, dst);
}
void Assembler::ldr_pcrel(Register dst, int imm12, Condition cond) {
AddrMode am = Offset;
if (imm12 < 0) {
imm12 = -imm12;
am = NegOffset;
}
DCHECK(is_uint12(imm12));
emit(cond | B26 | am | L | pc.code() * B16 | dst.code() * B12 | imm12);
}
// Load/Store exclusive instructions.
void Assembler::ldrex(Register dst, Register src, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.75.
// cond(31-28) | 00011001(27-20) | Rn(19-16) | Rt(15-12) | 111110011111(11-0)
DCHECK(dst != pc);
DCHECK(src != pc);
emit(cond | B24 | B23 | B20 | src.code() * B16 | dst.code() * B12 | 0xF9F);
}
void Assembler::strex(Register src1, Register src2, Register dst,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.212.
// cond(31-28) | 00011000(27-20) | Rn(19-16) | Rd(15-12) | 11111001(11-4) |
// Rt(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(src1 != dst);
DCHECK(src1 != src2);
emit(cond | B24 | B23 | dst.code() * B16 | src1.code() * B12 | 0xF9 * B4 |
src2.code());
}
void Assembler::ldrexb(Register dst, Register src, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.76.
// cond(31-28) | 00011101(27-20) | Rn(19-16) | Rt(15-12) | 111110011111(11-0)
DCHECK(dst != pc);
DCHECK(src != pc);
emit(cond | B24 | B23 | B22 | B20 | src.code() * B16 | dst.code() * B12 |
0xF9F);
}
void Assembler::strexb(Register src1, Register src2, Register dst,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.213.
// cond(31-28) | 00011100(27-20) | Rn(19-16) | Rd(15-12) | 11111001(11-4) |
// Rt(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(src1 != dst);
DCHECK(src1 != src2);
emit(cond | B24 | B23 | B22 | dst.code() * B16 | src1.code() * B12 |
0xF9 * B4 | src2.code());
}
void Assembler::ldrexh(Register dst, Register src, Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.78.
// cond(31-28) | 00011111(27-20) | Rn(19-16) | Rt(15-12) | 111110011111(11-0)
DCHECK(dst != pc);
DCHECK(src != pc);
emit(cond | B24 | B23 | B22 | B21 | B20 | src.code() * B16 |
dst.code() * B12 | 0xF9F);
}
void Assembler::strexh(Register src1, Register src2, Register dst,
Condition cond) {
// Instruction details available in ARM DDI 0406C.b, A8.8.215.
// cond(31-28) | 00011110(27-20) | Rn(19-16) | Rd(15-12) | 11111001(11-4) |
// Rt(3-0)
DCHECK(dst != pc);
DCHECK(src1 != pc);
DCHECK(src2 != pc);
DCHECK(src1 != dst);
DCHECK(src1 != src2);
emit(cond | B24 | B23 | B22 | B21 | dst.code() * B16 | src1.code() * B12 |
0xF9 * B4 | src2.code());
}
// Preload instructions.
void Assembler::pld(const MemOperand& address) {
// Instruction details available in ARM DDI 0406C.b, A8.8.128.
// 1111(31-28) | 0111(27-24) | U(23) | R(22) | 01(21-20) | Rn(19-16) |
// 1111(15-12) | imm5(11-07) | type(6-5) | 0(4)| Rm(3-0) |
DCHECK(address.rm() == no_reg);
DCHECK(address.am() == Offset);
int U = B23;
int offset = address.offset();
if (offset < 0) {
offset = -offset;
U = 0;
}
DCHECK_LT(offset, 4096);
emit(kSpecialCondition | B26 | B24 | U | B22 | B20 |
address.rn().code() * B16 | 0xF * B12 | offset);
}
// Load/Store multiple instructions.
void Assembler::ldm(BlockAddrMode am,
Register base,
RegList dst,
Condition cond) {
// ABI stack constraint: ldmxx base, {..sp..} base != sp is not restartable.
DCHECK(base == sp || (dst & sp.bit()) == 0);
AddrMode4(cond | B27 | am | L, base, dst);
// Emit the constant pool after a function return implemented by ldm ..{..pc}.
if (cond == al && (dst & pc.bit()) != 0) {
// There is a slight chance that the ldm instruction was actually a call,
// in which case it would be wrong to return into the constant pool; we
// recognize this case by checking if the emission of the pool was blocked
// at the pc of the ldm instruction by a mov lr, pc instruction; if this is
// the case, we emit a jump over the pool.
CheckConstPool(true, no_const_pool_before_ == pc_offset() - kInstrSize);
}
}
void Assembler::stm(BlockAddrMode am,
Register base,
RegList src,
Condition cond) {
AddrMode4(cond | B27 | am, base, src);
}
// Exception-generating instructions and debugging support.
// Stops with a non-negative code less than kNumOfWatchedStops support
// enabling/disabling and a counter feature. See simulator-arm.h .
void Assembler::stop(const char* msg, Condition cond, int32_t code) {
#ifndef __arm__
DCHECK_GE(code, kDefaultStopCode);
{
BlockConstPoolScope block_const_pool(this);
if (code >= 0) {
svc(kStopCode + code, cond);
} else {
svc(kStopCode + kMaxStopCode, cond);
}
}
#else // def __arm__
if (cond != al) {
Label skip;
b(&skip, NegateCondition(cond));
bkpt(0);
bind(&skip);
} else {
bkpt(0);
}
#endif // def __arm__
}
void Assembler::bkpt(uint32_t imm16) {
DCHECK(is_uint16(imm16));
emit(al | B24 | B21 | (imm16 >> 4) * B8 | BKPT | (imm16 & 0xF));
}
void Assembler::svc(uint32_t imm24, Condition cond) {
DCHECK(is_uint24(imm24));
emit(cond | 15*B24 | imm24);
}
void Assembler::dmb(BarrierOption option) {
if (CpuFeatures::IsSupported(ARMv7)) {
// Details available in ARM DDI 0406C.b, A8-378.
emit(kSpecialCondition | 0x57FF * B12 | 5 * B4 | option);
} else {
// Details available in ARM DDI 0406C.b, B3-1750.
// CP15DMB: CRn=c7, opc1=0, CRm=c10, opc2=5, Rt is ignored.
mcr(p15, 0, r0, cr7, cr10, 5);
}
}
void Assembler::dsb(BarrierOption option) {
if (CpuFeatures::IsSupported(ARMv7)) {
// Details available in ARM DDI 0406C.b, A8-380.
emit(kSpecialCondition | 0x57FF * B12 | 4 * B4 | option);
} else {
// Details available in ARM DDI 0406C.b, B3-1750.
// CP15DSB: CRn=c7, opc1=0, CRm=c10, opc2=4, Rt is ignored.
mcr(p15, 0, r0, cr7, cr10, 4);
}
}
void Assembler::isb(BarrierOption option) {
if (CpuFeatures::IsSupported(ARMv7)) {
// Details available in ARM DDI 0406C.b, A8-389.
emit(kSpecialCondition | 0x57FF * B12 | 6 * B4 | option);
} else {
// Details available in ARM DDI 0406C.b, B3-1750.
// CP15ISB: CRn=c7, opc1=0, CRm=c5, opc2=4, Rt is ignored.
mcr(p15, 0, r0, cr7, cr5, 4);
}
}
// Coprocessor instructions.
void Assembler::cdp(Coprocessor coproc,
int opcode_1,
CRegister crd,
CRegister crn,
CRegister crm,
int opcode_2,
Condition cond) {
DCHECK(is_uint4(opcode_1) && is_uint3(opcode_2));
emit(cond | B27 | B26 | B25 | (opcode_1 & 15)*B20 | crn.code()*B16 |
crd.code()*B12 | coproc*B8 | (opcode_2 & 7)*B5 | crm.code());
}
void Assembler::cdp2(Coprocessor coproc, int opcode_1, CRegister crd,
CRegister crn, CRegister crm, int opcode_2) {
cdp(coproc, opcode_1, crd, crn, crm, opcode_2, kSpecialCondition);
}
void Assembler::mcr(Coprocessor coproc,
int opcode_1,
Register rd,
CRegister crn,
CRegister crm,
int opcode_2,
Condition cond) {
DCHECK(is_uint3(opcode_1) && is_uint3(opcode_2));
emit(cond | B27 | B26 | B25 | (opcode_1 & 7)*B21 | crn.code()*B16 |
rd.code()*B12 | coproc*B8 | (opcode_2 & 7)*B5 | B4 | crm.code());
}
void Assembler::mcr2(Coprocessor coproc, int opcode_1, Register rd,
CRegister crn, CRegister crm, int opcode_2) {
mcr(coproc, opcode_1, rd, crn, crm, opcode_2, kSpecialCondition);
}
void Assembler::mrc(Coprocessor coproc,
int opcode_1,
Register rd,
CRegister crn,
CRegister crm,
int opcode_2,
Condition cond) {
DCHECK(is_uint3(opcode_1) && is_uint3(opcode_2));
emit(cond | B27 | B26 | B25 | (opcode_1 & 7)*B21 | L | crn.code()*B16 |
rd.code()*B12 | coproc*B8 | (opcode_2 & 7)*B5 | B4 | crm.code());
}
void Assembler::mrc2(Coprocessor coproc, int opcode_1, Register rd,
CRegister crn, CRegister crm, int opcode_2) {
mrc(coproc, opcode_1, rd, crn, crm, opcode_2, kSpecialCondition);
}
void Assembler::ldc(Coprocessor coproc,
CRegister crd,
const MemOperand& src,
LFlag l,
Condition cond) {
AddrMode5(cond | B27 | B26 | l | L | coproc * B8, crd, src);
}
void Assembler::ldc(Coprocessor coproc,
CRegister crd,
Register rn,
int option,
LFlag l,
Condition cond) {
// Unindexed addressing.
DCHECK(is_uint8(option));
emit(cond | B27 | B26 | U | l | L | rn.code()*B16 | crd.code()*B12 |
coproc*B8 | (option & 255));
}
void Assembler::ldc2(Coprocessor coproc, CRegister crd, const MemOperand& src,
LFlag l) {
ldc(coproc, crd, src, l, kSpecialCondition);
}
void Assembler::ldc2(Coprocessor coproc, CRegister crd, Register rn, int option,
LFlag l) {
ldc(coproc, crd, rn, option, l, kSpecialCondition);
}
// Support for VFP.
void Assembler::vldr(const DwVfpRegister dst,
const Register base,
int offset,
const Condition cond) {
// Ddst = MEM(Rbase + offset).
// Instruction details available in ARM DDI 0406C.b, A8-924.
// cond(31-28) | 1101(27-24)| U(23) | D(22) | 01(21-20) | Rbase(19-16) |
// Vd(15-12) | 1011(11-8) | offset
DCHECK(VfpRegisterIsAvailable(dst));
int u = 1;
if (offset < 0) {
CHECK_NE(offset, kMinInt);
offset = -offset;
u = 0;
}
int vd, d;
dst.split_code(&vd, &d);
DCHECK_GE(offset, 0);
if ((offset % 4) == 0 && (offset / 4) < 256) {
emit(cond | 0xD*B24 | u*B23 | d*B22 | B20 | base.code()*B16 | vd*B12 |
0xB*B8 | ((offset / 4) & 255));
} else {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
// Larger offsets must be handled by computing the correct address in a
// scratch register.
DCHECK(base != scratch);
if (u == 1) {
add(scratch, base, Operand(offset));
} else {
sub(scratch, base, Operand(offset));
}
emit(cond | 0xD * B24 | d * B22 | B20 | scratch.code() * B16 | vd * B12 |
0xB * B8);
}
}
void Assembler::vldr(const DwVfpRegister dst,
const MemOperand& operand,
const Condition cond) {
DCHECK(VfpRegisterIsAvailable(dst));
DCHECK(operand.am_ == Offset);
if (operand.rm().is_valid()) {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
add(scratch, operand.rn(),
Operand(operand.rm(), operand.shift_op_, operand.shift_imm_));
vldr(dst, scratch, 0, cond);
} else {
vldr(dst, operand.rn(), operand.offset(), cond);
}
}
void Assembler::vldr(const SwVfpRegister dst,
const Register base,
int offset,
const Condition cond) {
// Sdst = MEM(Rbase + offset).
// Instruction details available in ARM DDI 0406A, A8-628.
// cond(31-28) | 1101(27-24)| U001(23-20) | Rbase(19-16) |
// Vdst(15-12) | 1010(11-8) | offset
int u = 1;
if (offset < 0) {
offset = -offset;
u = 0;
}
int sd, d;
dst.split_code(&sd, &d);
DCHECK_GE(offset, 0);
if ((offset % 4) == 0 && (offset / 4) < 256) {
emit(cond | u*B23 | d*B22 | 0xD1*B20 | base.code()*B16 | sd*B12 |
0xA*B8 | ((offset / 4) & 255));
} else {
// Larger offsets must be handled by computing the correct address in a
// scratch register.
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
DCHECK(base != scratch);
if (u == 1) {
add(scratch, base, Operand(offset));
} else {
sub(scratch, base, Operand(offset));
}
emit(cond | d * B22 | 0xD1 * B20 | scratch.code() * B16 | sd * B12 |
0xA * B8);
}
}
void Assembler::vldr(const SwVfpRegister dst,
const MemOperand& operand,
const Condition cond) {
DCHECK(operand.am_ == Offset);
if (operand.rm().is_valid()) {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
add(scratch, operand.rn(),
Operand(operand.rm(), operand.shift_op_, operand.shift_imm_));
vldr(dst, scratch, 0, cond);
} else {
vldr(dst, operand.rn(), operand.offset(), cond);
}
}
void Assembler::vstr(const DwVfpRegister src,
const Register base,
int offset,
const Condition cond) {
// MEM(Rbase + offset) = Dsrc.
// Instruction details available in ARM DDI 0406C.b, A8-1082.
// cond(31-28) | 1101(27-24)| U(23) | D(22) | 00(21-20) | Rbase(19-16) |
// Vd(15-12) | 1011(11-8) | (offset/4)
DCHECK(VfpRegisterIsAvailable(src));
int u = 1;
if (offset < 0) {
CHECK_NE(offset, kMinInt);
offset = -offset;
u = 0;
}
DCHECK_GE(offset, 0);
int vd, d;
src.split_code(&vd, &d);
if ((offset % 4) == 0 && (offset / 4) < 256) {
emit(cond | 0xD*B24 | u*B23 | d*B22 | base.code()*B16 | vd*B12 | 0xB*B8 |
((offset / 4) & 255));
} else {
// Larger offsets must be handled by computing the correct address in the a
// scratch register.
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
DCHECK(base != scratch);
if (u == 1) {
add(scratch, base, Operand(offset));
} else {
sub(scratch, base, Operand(offset));
}
emit(cond | 0xD * B24 | d * B22 | scratch.code() * B16 | vd * B12 |
0xB * B8);
}
}
void Assembler::vstr(const DwVfpRegister src,
const MemOperand& operand,
const Condition cond) {
DCHECK(VfpRegisterIsAvailable(src));
DCHECK(operand.am_ == Offset);
if (operand.rm().is_valid()) {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
add(scratch, operand.rn(),
Operand(operand.rm(), operand.shift_op_, operand.shift_imm_));
vstr(src, scratch, 0, cond);
} else {
vstr(src, operand.rn(), operand.offset(), cond);
}
}
void Assembler::vstr(const SwVfpRegister src,
const Register base,
int offset,
const Condition cond) {
// MEM(Rbase + offset) = SSrc.
// Instruction details available in ARM DDI 0406A, A8-786.
// cond(31-28) | 1101(27-24)| U000(23-20) | Rbase(19-16) |
// Vdst(15-12) | 1010(11-8) | (offset/4)
int u = 1;
if (offset < 0) {
CHECK_NE(offset, kMinInt);
offset = -offset;
u = 0;
}
int sd, d;
src.split_code(&sd, &d);
DCHECK_GE(offset, 0);
if ((offset % 4) == 0 && (offset / 4) < 256) {
emit(cond | u*B23 | d*B22 | 0xD0*B20 | base.code()*B16 | sd*B12 |
0xA*B8 | ((offset / 4) & 255));
} else {
// Larger offsets must be handled by computing the correct address in a
// scratch register.
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
DCHECK(base != scratch);
if (u == 1) {
add(scratch, base, Operand(offset));
} else {
sub(scratch, base, Operand(offset));
}
emit(cond | d * B22 | 0xD0 * B20 | scratch.code() * B16 | sd * B12 |
0xA * B8);
}
}
void Assembler::vstr(const SwVfpRegister src,
const MemOperand& operand,
const Condition cond) {
DCHECK(operand.am_ == Offset);
if (operand.rm().is_valid()) {
UseScratchRegisterScope temps(this);
Register scratch = temps.Acquire();
add(scratch, operand.rn(),
Operand(operand.rm(), operand.shift_op_, operand.shift_imm_));
vstr(src, scratch, 0, cond);