Files
sdk/runtime/vm/instructions_arm.h
T
Ryan Macnak 1e24fe7d69 [vm, compiler] Specialize unoptimized monomorphic and megamorphic calls.
dart-bytecode, arm64:            +4.742% geomean
dart-bytecode-jit-unopt, arm64: +12.73% geomean
dart2js-compile, x64:            +3.635% geomean

In the polymorphic and unlinked cases, call to a stub the does a linear scan against an ICData.

In the monomorphic case, call to a prologue of the expected target function that checks the expected receiver class. There is additional indirection in the JIT version compared to the AOT version to also tick a usage counter so the inliner can make good decisions.

In the megamorphic case, call to a stub that does a hash table lookup against a MegamorphicCache.

Megamorphic call sites face a loss of precision in usage counts. The call site count is not recorded and the usage counter of the target function is used as an approximation.

Monomorphic and megamorphic calls sites are reset to the polymorphic/unlinked state on hot reload.

Monomorphic and megamorphic calls sites do not check the stepping state, so they are reset to the polymorphic/unlinked state when stepping begins and disabled.

Back-edges now increment the usage counter in addition to checking it. This ensures function with loops containing monomorphic calls will eventually cross the optimization threshold.

Fixed backwards use of kMonomorphicEntryOffset and kPolymorphicEntryOffset.

Fixed C stack overflow when bouncing between the KBC interpreter and a simulator.

Bug: https://github.com/dart-lang/sdk/issues/26780
Bug: https://github.com/dart-lang/sdk/issues/36409
Bug: https://github.com/dart-lang/sdk/issues/36731
Change-Id: I78a49cccd962703a459288e71ce246ed845df474
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/102820
Commit-Queue: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
2019-06-12 21:56:53 +00:00

271 lines
8.1 KiB
C++

// Copyright (c) 2013, the Dart project authors. Please see the AUTHORS file
// for details. All rights reserved. Use of this source code is governed by a
// BSD-style license that can be found in the LICENSE file.
// Classes that describe assembly patterns as used by inline caches.
#ifndef RUNTIME_VM_INSTRUCTIONS_ARM_H_
#define RUNTIME_VM_INSTRUCTIONS_ARM_H_
#ifndef RUNTIME_VM_INSTRUCTIONS_H_
#error Do not include instructions_arm.h directly; use instructions.h instead.
#endif
#include "vm/allocation.h"
#include "vm/compiler/assembler/assembler.h"
#include "vm/constants.h"
#include "vm/native_function.h"
namespace dart {
class ICData;
class Code;
class Object;
class ObjectPool;
class RawCode;
class RawICData;
class RawObject;
class InstructionPattern : public AllStatic {
public:
// Decodes a load sequence ending at 'end' (the last instruction of the
// load sequence is the instruction before the one at end). Returns the
// address of the first instruction in the sequence. Returns the register
// being loaded and the loaded object in the output parameters 'reg' and
// 'obj' respectively.
static uword DecodeLoadObject(uword end,
const ObjectPool& object_pool,
Register* reg,
Object* obj);
// Decodes a load sequence ending at 'end' (the last instruction of the
// load sequence is the instruction before the one at end). Returns the
// address of the first instruction in the sequence. Returns the register
// being loaded and the loaded immediate value in the output parameters
// 'reg' and 'value' respectively.
static uword DecodeLoadWordImmediate(uword end,
Register* reg,
intptr_t* value);
// Encodes a load immediate sequence ending at 'end' (the last instruction of
// the load sequence is the instruction before the one at end).
//
// Supports only a subset of [DecodeLoadWordImmediate], namely:
// movw r, #lower16
// movt r, #upper16
static void EncodeLoadWordImmediate(uword end, Register reg, intptr_t value);
// Decodes a load sequence ending at 'end' (the last instruction of the
// load sequence is the instruction before the one at end). Returns the
// address of the first instruction in the sequence. Returns the register
// being loaded and the index in the pool being read from in the output
// parameters 'reg' and 'index' respectively.
// IMPORANT: When generating code loading values from pool on ARM use
// LoadWordFromPool macro instruction instead of emitting direct load.
// The macro instruction takes care of pool offsets that can't be
// encoded as immediates.
static uword DecodeLoadWordFromPool(uword end,
Register* reg,
intptr_t* index);
};
class CallPattern : public ValueObject {
public:
CallPattern(uword pc, const Code& code);
RawCode* TargetCode() const;
void SetTargetCode(const Code& code) const;
private:
const ObjectPool& object_pool_;
intptr_t target_code_pool_index_;
DISALLOW_COPY_AND_ASSIGN(CallPattern);
};
class ICCallPattern : public ValueObject {
public:
ICCallPattern(uword pc, const Code& code);
RawObject* Data() const;
void SetData(const Object& data) const;
RawCode* TargetCode() const;
void SetTargetCode(const Code& code) const;
private:
const ObjectPool& object_pool_;
intptr_t target_pool_index_;
intptr_t data_pool_index_;
DISALLOW_COPY_AND_ASSIGN(ICCallPattern);
};
class NativeCallPattern : public ValueObject {
public:
NativeCallPattern(uword pc, const Code& code);
RawCode* target() const;
void set_target(const Code& target) const;
NativeFunction native_function() const;
void set_native_function(NativeFunction target) const;
private:
const ObjectPool& object_pool_;
uword end_;
intptr_t native_function_pool_index_;
intptr_t target_code_pool_index_;
DISALLOW_COPY_AND_ASSIGN(NativeCallPattern);
};
// Instance call that can switch between a direct monomorphic call, an IC call,
// and a megamorphic call.
// load guarded cid load ICData load MegamorphicCache
// load monomorphic target <-> load ICLookup stub -> load MMLookup stub
// call target.entry call stub.entry call stub.entry
class SwitchableCallPatternBase : public ValueObject {
public:
explicit SwitchableCallPatternBase(const Code& code);
RawObject* data() const;
void SetData(const Object& data) const;
protected:
const ObjectPool& object_pool_;
intptr_t data_pool_index_;
intptr_t target_pool_index_;
private:
DISALLOW_COPY_AND_ASSIGN(SwitchableCallPatternBase);
};
// See [SwitchableCallBase] for a switchable calls in general.
//
// The target slot is always a [Code] object: Either the code of the
// monomorphic function or a stub code.
class SwitchableCallPattern : public SwitchableCallPatternBase {
public:
SwitchableCallPattern(uword pc, const Code& code);
RawCode* target() const;
void SetTarget(const Code& target) const;
private:
DISALLOW_COPY_AND_ASSIGN(SwitchableCallPattern);
};
// See [SwitchableCallBase] for a switchable calls in general.
//
// The target slot is always a direct entrypoint address: Either the entry point
// of the monomorphic function or a stub entry point.
class BareSwitchableCallPattern : public SwitchableCallPatternBase {
public:
BareSwitchableCallPattern(uword pc, const Code& code);
RawCode* target() const;
void SetTarget(const Code& target) const;
private:
DISALLOW_COPY_AND_ASSIGN(BareSwitchableCallPattern);
};
class ReturnPattern : public ValueObject {
public:
explicit ReturnPattern(uword pc);
// bx_lr = 1.
static const int kLengthInBytes = 1 * Instr::kInstrSize;
int pattern_length_in_bytes() const { return kLengthInBytes; }
bool IsValid() const;
private:
const uword pc_;
};
class PcRelativeCallPattern : public ValueObject {
public:
// 24 bit signed integer which will get multiplied by 4.
static const intptr_t kLowerCallingRange = -(1 << 25);
static const intptr_t kUpperCallingRange = (1 << 25) - 1;
explicit PcRelativeCallPattern(uword pc) : pc_(pc) {}
static const int kLengthInBytes = 1 * Instr::kInstrSize;
int32_t distance() {
#if !defined(DART_PRECOMPILED_RUNTIME)
return Assembler::DecodeBranchOffset(*reinterpret_cast<int32_t*>(pc_));
#else
UNREACHABLE();
return 0;
#endif
}
void set_distance(int32_t distance) {
#if !defined(DART_PRECOMPILED_RUNTIME)
int32_t* word = reinterpret_cast<int32_t*>(pc_);
*word = Assembler::EncodeBranchOffset(distance, *word);
#else
UNREACHABLE();
#endif
}
bool IsValid() const;
private:
uword pc_;
};
// Instruction pattern for a tail call to a signed 32-bit PC-relative offset
//
// The AOT compiler can emit PC-relative calls. If the destination of such a
// call is not in range for the "bl.<cond> <offset>" instruction, the AOT
// compiler will emit a trampoline which is in range. That trampoline will
// then tail-call to the final destination (also via PC-relative offset, but it
// supports a full signed 32-bit offset).
//
// The pattern of the trampoline looks like:
//
// movw TMP, #lower16
// movt TMP, #upper16
// add PC, PC, TMP lsl #0
//
class PcRelativeTrampolineJumpPattern : public ValueObject {
public:
explicit PcRelativeTrampolineJumpPattern(uword pattern_start)
: pattern_start_(pattern_start) {
USE(pattern_start_);
}
static const int kLengthInBytes = 3 * Instr::kInstrSize;
void Initialize();
int32_t distance();
void set_distance(int32_t distance);
bool IsValid() const;
private:
// This offset must be applied to account for the fact that
// a) the actual "branch" is only in the 3rd instruction
// b) when reading the PC it reports current instruction + 8
static const intptr_t kDistanceOffset = -4 * Instr::kInstrSize;
// add PC, PC, TMP lsl #0
static const uint32_t kAddPcEncoding =
(ADD << kOpcodeShift) | (AL << kConditionShift) | (PC << kRnShift) |
(PC << kRdShift) | (TMP << kRmShift);
uword pattern_start_;
};
} // namespace dart
#endif // RUNTIME_VM_INSTRUCTIONS_ARM_H_