v8/test/cctest/compiler/test-code-generator.cc

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

594 lines
23 KiB
C++
Raw Normal View History

Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
// Copyright 2017 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include "src/assembler-inl.h"
#include "src/base/utils/random-number-generator.h"
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
#include "src/codegen.h"
#include "src/compilation-info.h"
#include "src/compiler/code-generator.h"
#include "src/compiler/instruction.h"
#include "src/compiler/linkage.h"
#include "src/isolate.h"
#include "src/objects-inl.h"
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
#include "test/cctest/cctest.h"
#include "test/cctest/compiler/function-tester.h"
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
namespace v8 {
namespace internal {
namespace compiler {
namespace {
int GetSlotSizeInBytes(MachineRepresentation rep) {
switch (rep) {
case MachineRepresentation::kTagged:
case MachineRepresentation::kFloat32:
return kPointerSize;
case MachineRepresentation::kFloat64:
return kDoubleSize;
case MachineRepresentation::kSimd128:
return kSimd128Size;
default:
break;
}
UNREACHABLE();
}
} // namespace
// Wrapper around the CodeGenerator with the ability to randomly generate moves
// and swaps which can then be executed. The `slots` map represents how many
// slots should be allocated per representation. Parallel moves will then be
// generated by randomly picking slots. Constants can be provided so that
// parallel moves may use them.
//
// At the moment, only the following representations are tested:
// - kTagged
// - kFloat32
// - kFloat64
// - kSimd128
// There is no need to test using Word32 or Word64 as they are the same as
// Tagged as far as the code generator is concerned.
class CodeGeneratorTester : public HandleAndZoneScope {
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
public:
CodeGeneratorTester(std::map<MachineRepresentation, int> slots =
std::map<MachineRepresentation, int>{},
std::initializer_list<Constant> constants = {})
: info_(ArrayVector("test"), main_isolate(), main_zone(), Code::STUB),
descriptor_(Linkage::GetStubCallDescriptor(
main_isolate(), main_zone(), VoidDescriptor(main_isolate()), 0,
CallDescriptor::kNoFlags, Operator::kNoProperties,
MachineType::AnyTagged(), 0)),
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
linkage_(descriptor_),
blocks_(main_zone()),
sequence_(main_isolate(), main_zone(), &blocks_),
rng_(CcTest::random_number_generator()),
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
frame_(descriptor_->CalculateFixedFrameSize()),
generator_(main_zone(), &frame_, &linkage_, &sequence_, &info_,
base::Optional<OsrHelper>(), kNoSourcePosition, nullptr) {
// Keep track of all supported representations depending on what kind of
// stack slots are supported.
for (const auto& slot : slots) {
supported_reps_.push_back(slot.first);
}
// Allocate new slots until we run out of them.
while (std::any_of(slots.cbegin(), slots.cend(),
[](const std::pair<MachineRepresentation, int>& entry) {
// True if there are slots left to allocate for this
// representation.
return entry.second > 0;
})) {
// Pick a random MachineRepresentation from supported_reps_.
MachineRepresentation rep = CreateRandomMachineRepresentation();
auto entry = slots.find(rep);
DCHECK(entry != slots.end());
// We may have picked a representation for which all slots have already
// been allocated.
if (entry->second > 0) {
// Keep a map of (MachineRepresentation . std::vector<int>) with
// allocated slots to pick from for each representation.
RegisterSlot(rep, frame_.AllocateSpillSlot(GetSlotSizeInBytes(rep)));
entry->second--;
}
}
for (auto constant : constants) {
int virtual_register = AllocateConstant(constant);
// Associate constants with their compatible representations.
// TODO(all): Test all types of constants.
switch (constant.type()) {
// Integer constants are always moved to a tagged location, whatever
// their sizes.
case Constant::kInt32:
case Constant::kInt64:
RegisterConstant(MachineRepresentation::kTagged, virtual_register);
break;
// FP constants may be moved to a tagged location using a heap number,
// or directly to a location of the same size.
case Constant::kFloat32:
RegisterConstant(MachineRepresentation::kTagged, virtual_register);
RegisterConstant(MachineRepresentation::kFloat32, virtual_register);
break;
case Constant::kFloat64:
RegisterConstant(MachineRepresentation::kTagged, virtual_register);
RegisterConstant(MachineRepresentation::kFloat64, virtual_register);
break;
default:
break;
}
}
// Force a frame to be created.
generator_.frame_access_state()->MarkHasFrame(true);
generator_.AssembleConstructFrame();
// TODO(all): Generate a stack check here so that we fail gracefully if the
// frame is too big.
}
int AllocateConstant(Constant constant) {
int virtual_register = sequence_.NextVirtualRegister();
sequence_.AddConstant(virtual_register, constant);
return virtual_register;
}
// Register a constant referenced by `virtual_register` as compatible with
// `rep`.
void RegisterConstant(MachineRepresentation rep, int virtual_register) {
auto entry = constants_.find(rep);
if (entry == constants_.end()) {
std::vector<int> vregs = {virtual_register};
constants_.emplace(rep, vregs);
} else {
entry->second.push_back(virtual_register);
}
}
void RegisterSlot(MachineRepresentation rep, int slot) {
auto entry = allocated_slots_.find(rep);
if (entry == allocated_slots_.end()) {
std::vector<int> slots = {slot};
allocated_slots_.emplace(rep, slots);
} else {
entry->second.push_back(slot);
}
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
}
enum PushTypeFlag {
kRegisterPush = CodeGenerator::kRegisterPush,
kStackSlotPush = CodeGenerator::kStackSlotPush,
kScalarPush = CodeGenerator::kScalarPush
};
enum OperandConstraint {
kNone,
// Restrict operands to non-constants. This is useful when generating a
// destination.
kCannotBeConstant
};
// Generate parallel moves at random. Note that they may not be compatible
// between each other as this doesn't matter to the code generator.
ParallelMove* GenerateRandomMoves(int size) {
ParallelMove* parallel_move = new (main_zone()) ParallelMove(main_zone());
for (int i = 0; i < size;) {
MachineRepresentation rep = CreateRandomMachineRepresentation();
MoveOperands mo(CreateRandomOperand(kNone, rep),
CreateRandomOperand(kCannotBeConstant, rep));
// It isn't valid to call `AssembleMove` and `AssembleSwap` with redundant
// moves.
if (mo.IsRedundant()) continue;
parallel_move->AddMove(mo.source(), mo.destination());
// Iterate only when a move was created.
i++;
}
return parallel_move;
}
ParallelMove* GenerateRandomSwaps(int size) {
ParallelMove* parallel_move = new (main_zone()) ParallelMove(main_zone());
for (int i = 0; i < size;) {
MachineRepresentation rep = CreateRandomMachineRepresentation();
InstructionOperand lhs = CreateRandomOperand(kCannotBeConstant, rep);
InstructionOperand rhs = CreateRandomOperand(kCannotBeConstant, rep);
MoveOperands mo(lhs, rhs);
// It isn't valid to call `AssembleMove` and `AssembleSwap` with redundant
// moves.
if (mo.IsRedundant()) continue;
// Canonicalize the swap: the register operand has to be the left hand
// side.
if (lhs.IsStackSlot() || lhs.IsFPStackSlot()) {
std::swap(lhs, rhs);
}
parallel_move->AddMove(lhs, rhs);
// Iterate only when a swap was created.
i++;
}
return parallel_move;
}
MachineRepresentation CreateRandomMachineRepresentation() {
int index = rng_->NextInt(static_cast<int>(supported_reps_.size()));
return supported_reps_[index];
}
InstructionOperand CreateRandomOperand(OperandConstraint constraint,
MachineRepresentation rep) {
// Only generate a Constant if the operand is a source and we have a
// constant with a compatible representation in stock.
bool generate_constant = (constraint != kCannotBeConstant) &&
(constants_.find(rep) != constants_.end());
switch (rng_->NextInt(generate_constant ? 3 : 2)) {
case 0:
return CreateRandomStackSlotOperand(rep);
case 1:
return CreateRandomRegisterOperand(rep);
case 2:
return CreateRandomConstant(rep);
}
UNREACHABLE();
}
InstructionOperand CreateRandomRegisterOperand(MachineRepresentation rep) {
int code;
const RegisterConfiguration* conf = RegisterConfiguration::Default();
switch (rep) {
case MachineRepresentation::kFloat32: {
int index = rng_->NextInt(conf->num_allocatable_float_registers());
code = conf->RegisterConfiguration::GetAllocatableFloatCode(index);
break;
}
case MachineRepresentation::kFloat64: {
int index = rng_->NextInt(conf->num_allocatable_double_registers());
code = conf->RegisterConfiguration::GetAllocatableDoubleCode(index);
break;
}
case MachineRepresentation::kSimd128: {
int index = rng_->NextInt(conf->num_allocatable_simd128_registers());
code = conf->RegisterConfiguration::GetAllocatableSimd128Code(index);
break;
}
case MachineRepresentation::kTagged: {
// Pick an allocatable register that is not the return register.
do {
int index = rng_->NextInt(conf->num_allocatable_general_registers());
code = conf->RegisterConfiguration::GetAllocatableGeneralCode(index);
} while (code == kReturnRegister0.code());
break;
}
default:
UNREACHABLE();
break;
}
return AllocatedOperand(LocationOperand::REGISTER, rep, code);
}
InstructionOperand CreateRandomStackSlotOperand(MachineRepresentation rep) {
int index = rng_->NextInt(static_cast<int>(allocated_slots_[rep].size()));
return AllocatedOperand(LocationOperand::STACK_SLOT, rep,
allocated_slots_[rep][index]);
}
InstructionOperand CreateRandomConstant(MachineRepresentation rep) {
int index = rng_->NextInt(static_cast<int>(constants_[rep].size()));
return ConstantOperand(constants_[rep][index]);
}
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
void CheckAssembleTailCallGaps(Instruction* instr,
int first_unused_stack_slot,
CodeGeneratorTester::PushTypeFlag push_type) {
generator_.AssembleTailCallBeforeGap(instr, first_unused_stack_slot);
#if defined(V8_TARGET_ARCH_ARM) || defined(V8_TARGET_ARCH_S390) || \
defined(V8_TARGET_ARCH_PPC)
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
// Only folding register pushes is supported on ARM.
bool supported = ((push_type & CodeGenerator::kRegisterPush) == push_type);
#elif defined(V8_TARGET_ARCH_X64) || defined(V8_TARGET_ARCH_IA32) || \
defined(V8_TARGET_ARCH_X87)
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
bool supported = ((push_type & CodeGenerator::kScalarPush) == push_type);
#else
bool supported = false;
#endif
if (supported) {
// Architectures supporting folding adjacent pushes should now have
// resolved all moves.
for (const auto& move :
*instr->parallel_moves()[Instruction::FIRST_GAP_POSITION]) {
CHECK(move->IsEliminated());
}
}
generator_.AssembleGaps(instr);
generator_.AssembleTailCallAfterGap(instr, first_unused_stack_slot);
}
void CheckAssembleMove(InstructionOperand* source,
InstructionOperand* destination) {
int start = generator_.tasm()->pc_offset();
generator_.AssembleMove(source, destination);
CHECK(generator_.tasm()->pc_offset() > start);
}
void CheckAssembleSwap(InstructionOperand* source,
InstructionOperand* destination) {
int start = generator_.tasm()->pc_offset();
generator_.AssembleSwap(source, destination);
CHECK(generator_.tasm()->pc_offset() > start);
}
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
Handle<Code> Finalize() {
InstructionOperand zero = ImmediateOperand(ImmediateOperand::INLINE, 0);
generator_.AssembleReturn(&zero);
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
generator_.FinishCode();
generator_.safepoints()->Emit(generator_.tasm(),
frame_.GetTotalFrameSlotCount());
return generator_.FinalizeCode();
}
void Disassemble() {
HandleScope scope(main_isolate());
Handle<Code> code = Finalize();
if (FLAG_print_code) {
code->Print();
}
}
void Run() {
HandleScope scope(main_isolate());
Handle<Code> code = Finalize();
if (FLAG_print_code) {
code->Print();
}
FunctionTester ft(code);
ft.Call();
}
v8::base::RandomNumberGenerator* rng() const { return rng_; }
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
private:
CompilationInfo info_;
CallDescriptor* descriptor_;
Linkage linkage_;
ZoneVector<InstructionBlock*> blocks_;
InstructionSequence sequence_;
std::vector<MachineRepresentation> supported_reps_;
std::map<MachineRepresentation, std::vector<int>> allocated_slots_;
std::map<MachineRepresentation, std::vector<int>> constants_;
v8::base::RandomNumberGenerator* rng_;
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
Frame frame_;
CodeGenerator generator_;
};
// The following fuzz tests will assemble a lot of moves, wrap them in
// executable native code and run them. At this time, we only check that
// something is actually generated, and that it runs on hardware or the
// simulator.
// TODO(all): It would be great to record the data on the stack after all moves
// are executed so that we could test the functionality in an architecture
// independent way. We would also have to make sure we generate moves compatible
// with each other as the gap-resolver tests do.
TEST(FuzzAssembleMove) {
// Test small and potentially large ranges separately. Note that the number of
// slots affects how much stack is allocated when running the generated code.
// This means we have to be careful not to exceed the stack limit, which is
// lower on Windows.
for (auto n : {64, 500}) {
std::map<MachineRepresentation, int> slots = {
{MachineRepresentation::kTagged, n},
{MachineRepresentation::kFloat32, n},
{MachineRepresentation::kFloat64, n}};
if (CpuFeatures::SupportsWasmSimd128()) {
// Generate fewer 128-bit slots.
slots.emplace(MachineRepresentation::kSimd128, n / 4);
}
CodeGeneratorTester c(
slots,
{Constant(0), Constant(1), Constant(2), Constant(3), Constant(4),
Constant(5), Constant(6), Constant(7),
Constant(static_cast<float>(0.1)), Constant(static_cast<float>(0.2)),
Constant(static_cast<float>(0.3)), Constant(static_cast<float>(0.4)),
Constant(static_cast<double>(0.5)), Constant(static_cast<double>(0.6)),
Constant(static_cast<double>(0.7)),
Constant(static_cast<double>(0.8))});
ParallelMove* moves = c.GenerateRandomMoves(1000);
for (const auto m : *moves) {
c.CheckAssembleMove(&m->source(), &m->destination());
}
c.Run();
}
}
TEST(FuzzAssembleSwap) {
// Test small and potentially large ranges separately. Note that the number of
// slots affects how much stack is allocated when running the generated code.
// This means we have to be careful not to exceed the stack limit, which is
// lower on Windows.
for (auto n : {64, 500}) {
std::map<MachineRepresentation, int> slots = {
{MachineRepresentation::kTagged, n},
{MachineRepresentation::kFloat32, n},
{MachineRepresentation::kFloat64, n}};
if (CpuFeatures::SupportsWasmSimd128()) {
// Generate fewer 128-bit slots.
slots.emplace(MachineRepresentation::kSimd128, n / 4);
}
CodeGeneratorTester c(slots);
ParallelMove* moves = c.GenerateRandomSwaps(1000);
for (const auto m : *moves) {
c.CheckAssembleSwap(&m->source(), &m->destination());
}
c.Run();
}
}
TEST(FuzzAssembleMoveAndSwap) {
// Test small and potentially large ranges separately. Note that the number of
// slots affects how much stack is allocated when running the generated code.
// This means we have to be careful not to exceed the stack limit, which is
// lower on Windows.
for (auto n : {64, 500}) {
std::map<MachineRepresentation, int> slots = {
{MachineRepresentation::kTagged, n},
{MachineRepresentation::kFloat32, n},
{MachineRepresentation::kFloat64, n}};
if (CpuFeatures::SupportsWasmSimd128()) {
// Generate fewer 128-bit slots.
slots.emplace(MachineRepresentation::kSimd128, n / 4);
}
CodeGeneratorTester c(
slots,
{Constant(0), Constant(1), Constant(2), Constant(3), Constant(4),
Constant(5), Constant(6), Constant(7),
Constant(static_cast<float>(0.1)), Constant(static_cast<float>(0.2)),
Constant(static_cast<float>(0.3)), Constant(static_cast<float>(0.4)),
Constant(static_cast<double>(0.5)), Constant(static_cast<double>(0.6)),
Constant(static_cast<double>(0.7)),
Constant(static_cast<double>(0.8))});
for (int i = 0; i < 1000; i++) {
// Randomly alternate between swaps and moves.
if (c.rng()->NextInt(2) == 0) {
MoveOperands* move = c.GenerateRandomMoves(1)->at(0);
c.CheckAssembleMove(&move->source(), &move->destination());
} else {
MoveOperands* move = c.GenerateRandomSwaps(1)->at(0);
c.CheckAssembleSwap(&move->source(), &move->destination());
}
}
c.Run();
}
}
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
TEST(AssembleTailCallGap) {
const RegisterConfiguration* conf = RegisterConfiguration::Default();
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
// This test assumes at least 4 registers are allocatable.
CHECK_LE(4, conf->num_allocatable_general_registers());
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
auto r0 = AllocatedOperand(LocationOperand::REGISTER,
MachineRepresentation::kTagged,
conf->GetAllocatableGeneralCode(0));
auto r1 = AllocatedOperand(LocationOperand::REGISTER,
MachineRepresentation::kTagged,
conf->GetAllocatableGeneralCode(1));
auto r2 = AllocatedOperand(LocationOperand::REGISTER,
MachineRepresentation::kTagged,
conf->GetAllocatableGeneralCode(2));
auto r3 = AllocatedOperand(LocationOperand::REGISTER,
MachineRepresentation::kTagged,
conf->GetAllocatableGeneralCode(3));
auto slot_minus_4 = AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, -4);
auto slot_minus_3 = AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, -3);
auto slot_minus_2 = AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, -2);
auto slot_minus_1 = AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, -1);
// Avoid slot 0 for architectures which use it store the return address.
int first_slot = V8_TARGET_ARCH_STORES_RETURN_ADDRESS_ON_STACK ? 1 : 0;
auto slot_0 = AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, first_slot);
auto slot_1 =
AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, first_slot + 1);
auto slot_2 =
AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, first_slot + 2);
auto slot_3 =
AllocatedOperand(LocationOperand::STACK_SLOT,
MachineRepresentation::kTagged, first_slot + 3);
// These tests all generate series of moves that the code generator should
// detect as adjacent pushes. Depending on the architecture, we make sure
// these moves get eliminated.
// Also, disassembling with `--print-code` is useful when debugging.
{
// Generate a series of register pushes only.
CodeGeneratorTester c;
Instruction* instr = Instruction::New(c.main_zone(), kArchNop);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r3, slot_0);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r2, slot_1);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r1, slot_2);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r0, slot_3);
c.CheckAssembleTailCallGaps(instr, first_slot + 4,
CodeGeneratorTester::kRegisterPush);
c.Disassemble();
}
{
// Generate a series of stack pushes only.
CodeGeneratorTester c;
Instruction* instr = Instruction::New(c.main_zone(), kArchNop);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_4, slot_0);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_3, slot_1);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_2, slot_2);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_1, slot_3);
c.CheckAssembleTailCallGaps(instr, first_slot + 4,
CodeGeneratorTester::kStackSlotPush);
c.Disassemble();
}
{
// Generate a mix of stack and register pushes.
CodeGeneratorTester c;
Instruction* instr = Instruction::New(c.main_zone(), kArchNop);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_2, slot_0);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r1, slot_1);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(slot_minus_1, slot_2);
instr
->GetOrCreateParallelMove(Instruction::FIRST_GAP_POSITION,
c.main_zone())
Reland "[arm] Restrict grouping pushes before a TailCall to registers only" This is a reland of a72b2f88a8cb44f12002f1168c639e07db9d5600 Original change's description: > [arm] Restrict grouping pushes before a TailCall to registers only > > We optimize parallel moves performed before a TailCall by grouping adjacent > pushes. This way, we may use a single instruction to push multiple registers at > once. However, we also have support for pushing immediates and stack slots for > which the benefit is questionnable therefore this patch removes support for > them. > > Concerning immediate pushes, it looks like a mistake since we do not have > support for this case in `AssembleMove` so this patch removes it. Furthermore, > if we add a test for this case, we see that a `push ip` instruction is > generated, effectively pushing whatever was in `ip` at the time instead of > pushing a constant. > > Concerning stack slot pushes, we generate a more or less equivalent sequence of > instructions. > > Finally, grouping floating point pushes is not used anywhere so this patch > removes support for this also. > > Bug: v8:6553 > Change-Id: I9b820d33361fc442dd813f66e1f96cda41009110 > Reviewed-on: https://chromium-review.googlesource.com/567191 > Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> > Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> > Cr-Commit-Position: refs/heads/master@{#46718} Bug: v8:6553 Change-Id: Ib9a55dae7cc5db6185d163c56088ff23426d04bb Reviewed-on: https://chromium-review.googlesource.com/576087 Reviewed-by: Benedikt Meurer <bmeurer@chromium.org> Commit-Queue: Pierre Langlois <pierre.langlois@arm.com> Cr-Commit-Position: refs/heads/master@{#46754}
2017-07-13 09:55:16 +00:00
->AddMove(r0, slot_3);
c.CheckAssembleTailCallGaps(instr, first_slot + 4,
CodeGeneratorTester::kScalarPush);
c.Disassemble();
}
}
} // namespace compiler
} // namespace internal
} // namespace v8