In cases where the same target function is used for multiple
CID ranges, we only perform the inlining once. However, if the
CID range used for the initial inlining is a single CID, we will
set the type of the redefinition to that CID.
Detect this case and clear the concrete type associated with
the redefinition if it was given one.
Change-Id: I9f9bdd7c21e0dc1ac537f8facece0010630bd9aa
Cq-Include-Trybots: luci.dart.try:vm-kernel-win-debug-x64-try,vm-kernel-win-release-ia32-try,vm-kernel-win-product-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/105221
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
In cases where the same target function is used for multiple
CID ranges, we only perform the inlining once. However, if the
CID range used for the initial inlining is a single CID, we will
set the type of the redefinition to that CID.
Detect this case and clear the concrete type associated with
the redefinition if it was given one.
Change-Id: I3c27c46c71e10e71624ce3b969cb741eb9b2915f
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/104781
Commit-Queue: Teagan Strickland <sstrickl@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Based on the unified typed data layout, we can now inline accesses to
typed data interface classes if there are no 3rd party implementations
of those interfaces.
Example: If a receiver is of type Uint8List and we call `[]` or `[]=` we
will inline the byte access.
Instead of changing the existing inliner / call specializer we add this
as an extra pass: If the inliner / call specializer infer that the
receiver type is e.g. internal typed data then it will perform the
inlining itself using more optimized LoadIndexed instruction.
=> Only if those existing optimization passes have not been able to inline
the access will we, later on in the compilation pipeline, run a
specialized pass which will inline the accesses using LoadUntagged +
LoadIndexed (which is slightly less efficient than using only LoadIndexed
for internal typed data).
As a first step this is only done for AOT.
For ease of writing tests matching certain IR graphs this CL also adds a
IR pattern matcher.
Issue https://github.com/dart-lang/sdk/issues/35154
Cq-Include-Trybots: luci.dart.try:vm-canary-linux-debug-try, vm-dartkb-linux-debug-x64-try, vm-dartkb-linux-release-x64-try, vm-kernel-asan-linux-release-x64-try, vm-kernel-checked-linux-release-x64-try, vm-kernel-linux-debug-ia32-try, vm-kernel-linux-debug-simdbc64-try, vm-kernel-linux-debug-x64-try, vm-kernel-linux-product-x64-try, vm-kernel-linux-release-ia32-try, vm-kernel-linux-release-simarm-try, vm-kernel-linux-release-simarm64-try, vm-kernel-linux-release-simdbc64-try, vm-kernel-linux-release-x64-try, vm-kernel-optcounter-threshold-linux-release-ia32-try, vm-kernel-optcounter-threshold-linux-release-x64-try, vm-kernel-precomp-android-release-arm-try, vm-kernel-precomp-bare-linux-release-simarm-try, vm-kernel-precomp-bare-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-x64-try, vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-product-x64-try, vm-kernel-precomp-linux-release-simarm-try, vm-kernel-precomp-linux-release-simarm64-try, vm-kernel-precomp-linux-release-x64-try, vm-kernel-precomp-obfuscate-linux-release-x64-try, vm-kernel-precomp-win-release-simarm64-try, vm-kernel-precomp-win-release-x64-try, vm-kernel-reload-linux-debug-x64-try, vm-kernel-reload-linux-release-x64-try, vm-kernel-reload-rollback-linux-debug-x64-try, vm-kernel-reload-rollback-linux-release-x64-try, vm-kernel-win-debug-ia32-try, vm-kernel-win-debug-x64-try, vm-kernel-win-product-x64-try, vm-kernel-win-release-ia32-try, vm-kernel-win-release-x64-try
Change-Id: I5f2e01a55f46b473f64478b05679f65b9fd7c4c8
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/98662
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
The two existing places where IR is built are changed to use this new
helper. It supports running normal JIT/AOT passes as well as a
user-specified set of passes.
In order to allow vm/cc tests to make assertions about AOT pipeline this CL
enables the DART_PRECOMPILER define in run_vm_tests binary (similar to
gen_snapshot, run_vm_tests has now JIT and AOT support)
Change-Id: Ib51a024a81e0291e89d20860b8b9a2762611426c
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/98482
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
In general we would like to keep previous results of type propagation passes, however this
leads to situations where CompileType objects owned by definitions are cached as reaching
types at uses that are no longer directly connected to the definition itself.
This leads to situations when type propagator updates the type of the definition in place
via Definition::UpdateType, but does not enqueue an instruction which uses this type which
leads to incorrect fix-point computation.
As a fix we make the fact that some CompileType objects are owned by definitions explicit
by assigning [CompileType::owner_] field. Whenever we encounter situation that reaching type's
owner does not match current definition flowing into use we clone compile type to disconnect
it from original owner and prevent hidden mutation.
This CL also fixes two minor issues:
- incorrect computation of non-nullable type from a nullable CompileType with a cid;
- assings CompileType to LoadIndexedUnsafe instructions emitted by prologue builder;
Fixes https://github.com/dart-lang/sdk/issues/36156
Bug: 36156
Change-Id: I6536cfe1fda033cba3524336bfbdb871b88a8b3f
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-android-release-arm-try, vm-kernel-precomp-bare-linux-release-simarm-try, vm-kernel-precomp-bare-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-x64-try, vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-product-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/98358
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This helper class can be used for building IL in SSA form. It was
previously used to build intrinsic IL, but it can also be used to
build IL for tests.
This move required introducing a constant that describes offset
of the last parameter from SP value on entry (kLastParamFrom
This CL also changes the meaning of ParameterInstr(index, SPREG) to be
[SP + index + kLastParamSlotFromEntrySp], where previously it was
something like [SP + (index - 1) + kLastParamSlotFromEntrySp].
Change-Id: Ib75ecb3a94f9cde74d6f91028d3cbea8f5e4bc42
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/98336
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Parameter instructions inside CatchBlockEntry's initial definitions
are similar to Phi instructions: Parameter value at entry to the
catch would be taken from corresponding slot in the environment
attached to the instruction which threw the exception which is
being caught.
Thus similarly to Phi-s Parameter instructions can be redundant (if they
always evaluate to the same value - no matter where control arrives
from) or dead (if their value never reaches real instruction).
We already had an analysis to discover subset of redundant Parameter-s
which always evaluate to a constant value.
This CL introduces analysis to eliminate dead Parameter-s.
This should significantly reduce size of metadata generated for try/catch
and reduce spilling within functions with try/catch.
On Flutter Gallery: reduces Isolate part of snapshot by 15k (0.7%), Instructions
part is reduced by 2.7K (0.06%).
Change-Id: I345813e326152ca2ed2d4d5353bb64730d7545d4
Cq-Include-Trybots: luci.dart.try:vm-kernel-precomp-android-release-arm-try, vm-kernel-precomp-bare-linux-release-simarm-try, vm-kernel-precomp-bare-linux-release-simarm64-try, vm-kernel-precomp-bare-linux-release-x64-try, vm-kernel-precomp-linux-debug-x64-try, vm-kernel-precomp-linux-product-x64-try
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/97110
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
This change replaces kernel AST declarations of fields and functions
with bytecode declarations.
Size of dilp files is reduced by 11-12%.
Startup latency:
Time to the first full frame: 1.945s -> 1.687s
FinalizeClass: 554ms -> 277ms
FinishClassLoading: 296ms -> 156ms
There are following regressions in bytecode mode, which will be fixed
in future:
* dart:mirrors are not supported yet (implementation of mirrors relies
on reading kernel AST in certain cases).
As the result, lib_2/mirrors/* tests fail.
* native extensions are not supported yet (annotations on libraries
and classes in AST are cleaned up as they could reference members
which are now removed from AST).
As the result, standalone_2/entrypoints_verification_test test fails.
* language_2/spread_collections/const_error_test/* tests fail
due to https://github.com/dart-lang/sdk/issues/36286.
Change-Id: I5130f401fd7b84038b136136e7ccc1a6e51b6cea
Reviewed-on: https://dart-review.googlesource.com/c/sdk/+/97561
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Ryan Macnak <rmacnak@google.com>
This is the next step towards preventing compiler from directly peeking
into runtime and instead interact with runtime through a well defined
surface.
This CL decouples the hand-written intrinsifier code from the runtime:
* the intrinsifier is split up into a GraphIntrinsifier and AsmIntrinsifier
* the recognized methods list is moved to a separate .h file
* all intrinsifier code is moved into dart::compiler namespace
* the AsmIntrinsifier is only interacting with RT through runtime_api.h
Issue https://github.com/dart-lang/sdk/issues/31709
Change-Id: I0a73ad620e051dd49c9db7da3241212b3b74ccdd
Reviewed-on: https://dart-review.googlesource.com/c/92740
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
This is the next step towards preventing compiler from directly peeking
into runtime and instead interact with runtime through a well defined
surface.
This CL decouples the hand-written stub codes from the runtime. The
target architecture dependent stubs are moved to
dart::compiler::StubCodeCompiler which use dart::compiler::target:*
for accessing any runtime related code.
The generation of type testing stubs is moved to separate files for the
time being.
Issue https://github.com/dart-lang/sdk/issues/31709
Change-Id: Icd0995b18a7bac496b1e12231cf437943f5c94f1
Reviewed-on: https://dart-review.googlesource.com/c/92720
Reviewed-by: Martin Kustermann <kustermann@google.com>
Commit-Queue: Martin Kustermann <kustermann@google.com>
Auto-Submit: Martin Kustermann <kustermann@google.com>
This is the next step towards preventing compiler from directly peeking
into runtime and instead interact with runtime through a well defined
surface.
This CL decouples the hand-written stub codes from the runtime. The
target architecture dependent stubs are moved to
dart::compiler::StubCodeCompiler which use dart::compiler::target:*
for accessing any runtime related code.
The generation of type testing stubs is moved to separate files for the
time being.
Issue https://github.com/dart-lang/sdk/issues/31709
Change-Id: I1b4f1cca0acb704b30b80eca7f634734772389b5
Reviewed-on: https://dart-review.googlesource.com/c/92138
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Aart Bik <ajcbik@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Rationale:
This is a start with a graph checker that verifies
the consistency of the flow graph after each compiler
pass with the goal of detecting errors as early as
possible. The checks add overhead, so are only run
in debug mode. The objective is that all written
and unwritten assumptions on the flow graph that
are relatively easy to check will be converted
into actual code.
Change-Id: Iad9927d5b0fd87a43dc6a7369174d52761d89c9e
Reviewed-on: https://dart-review.googlesource.com/c/91143
Commit-Queue: Aart Bik <ajcbik@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This is the next step towards preventing compiler from directly peeking
into runtime and instead interact with runtime through a well defined
surface. The goal of the refactoring to locate all places where compiler
accesses some runtime information and partion those accesses into two
categories:
- creating objects in the host runtime (e.g. allocating strings, numbers, etc)
during compilation;
- accessing properties of the target runtime (e.g. offsets of fields) to
embed those into the generated code;
This change introduces dart::compiler and dart::compiler::target namespaces.
All code in the compiler will gradually be moved into dart::compiler namespace.
One of the motivations for this change is to be able to prevent access to
globally defined host constants like kWordSize by shadowing them in the
dart::compiler namespace.
The nested namespace dart::compiler::target hosts all information about
target runtime that compiler could access, e.g. compiler::target::kWordSize
defines word size of the target which will eventually be made different
from the host kWordSize (defined by dart::kWordSize).
The API for compiler to runtime interaction is placed into compiler_api.h.
Note that we still permit runtime to access compiler internals directly -
this is not going to be decoupled as part of this work.
Issue https://github.com/dart-lang/sdk/issues/31709
Change-Id: If4396d295879391becfa6c38d4802bbff81f5b20
Reviewed-on: https://dart-review.googlesource.com/c/90242
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
There might be a race between background compiler and mutator where
mutator changes guarded state of the field after Slot was created from
it.
A situation is possible where we have a clone of a field with its
guarded state set to unknown, however Slot::Get for this field returns
a Slot created from the previous clone of the same field with a known
guarded state. In this case we must add *old* clone from which the Slot
was created to guarded fields and not the new clone, because new clone
has no guarded state to begin with and thus
ParsedFunction::AddToGuardedFields(...) would simply ignore it.
Such slots with inconsistent guarded state that are not in the current
list of guarded fields arise due to unsuccessful inlining attempts.
If we built and discard the graph, then guarded fields associated with
that graph are also discarded. However the slot itself stays behind in
the compilation global cache.
TEST=vm/cc/SlotFromGuardedField
Bug: b/121271056
Change-Id: Ib6ee5ec4922c033c3a71bdc46da74ead47b1edd0
Reviewed-on: https://dart-review.googlesource.com/c/88575
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Start ordering instructions before writing them out in AOT snapshots.
This is done by:
* discovery of relevant code objects
* ordering of the code objects (currently simply discovery order)
* relocate any pc-relative calls by patching up pc-relative calls with
the actual distance
* producing image writer commands which tell it in which order and at
which .text offsets to write out instructions
Note: Our compiler currently does not emit any pc-relative calls. This
will be added later on.
Issue https://github.com/dart-lang/sdk/issues/33274
Change-Id: I0153c32a7427ea395e80489efadd5513fe7ca371
Reviewed-on: https://dart-review.googlesource.com/c/85745
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Make StoreIntanceField and LoadField instructions use a single uniform
abstraction: Slot (used to be called NativeFieldDesc), which represents
either a real Dart field (i.e. a field that has a corresponding Field object)
or a native VM field that does not have a corresponding Field object.
This refactoring eliminates raw stores/loads that were just using offsets
before - now we always know what kind of slots we are accessing and
this yields better aliasing information.
Change-Id: I2f48332d58258219565bd961764e8cc9dd4d75ce
Reviewed-on: https://dart-review.googlesource.com/c/74582
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Rationale:
The more one knows about loops, the better!
This introduces a rigorous framework for detecting
induction variables (more precise "sequence" variables).
Even though this is already pretty general, in the
future we can expand this on a "need to" base, recognizing
more operators (shifts, negate, etc.) and more classes
of induction (perhaps geometric, polynomial, etc.).
Change-Id: I82a14515e8ae946d520ee470cd31046f6a58af7d
Reviewed-on: https://dart-review.googlesource.com/c/81743
Commit-Queue: Aart Bik <ajcbik@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Rationale:
Started a more structured implementation of detecting
loops and the loop hierarchy in the Dart flow graph.
This new framework can replace some of the more ad-hoc
approach taken now, as already started in this CL, and
also forms a foundation for more optimizations later.
https://github.com/dart-lang/sdk/issues/34473
Change-Id: I75130cb6863324910e3d068e4f2628c25fdf880c
Reviewed-on: https://dart-review.googlesource.com/c/79147
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Aart Bik <ajcbik@google.com>
Also move deopt id computation logic into a separate class and add a comment
explaining while deopt ids are incremented by 2.
Change-Id: Ife489be7d10c7198a8e7adf9e97e0c516d78ea55
Reviewed-on: https://dart-review.googlesource.com/72685
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This CL continues refactoring aimed at isolating StreamingFlowGraphBuilder.
StreamingFlowGraphBuilder dependency is removed from classes
* StreamingConstantEvaluator (it still depends on FlowGraphBuilder)
* ConstantHelper
* SimpleExpressionConverter
and several methods in object.cc.
StreamingConstantEvaluator and ConstantHelper are moved from
kernel_binary_flowgraph{.h, .cc} to a new source file
constant_evaluator{.h, .cc}. StreamingConstantEvaluator is renamed
to ConstantEvaluator.
KernelFingerprintHelper and KernelSourceFingerprintHelper are moved to
a new source file kernel_fingerprints{.h, .cc}.
Instances of kernel::FlowGraphBuilder no longer contain back reference
to a StreamingFlowGraphBuilder. In order to drop this circular dependency
TranslateFinallyFinalizers() is moved from FlowGraphBuilder to
StreamingFlowGraphBuilder.
Change-Id: Id550d22b3567dea9512328a900935bd6145a8107
Reviewed-on: https://dart-review.googlesource.com/64463
Commit-Queue: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Zach Anderson <zra@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
After previous refactoring, StreamingScopeBuilder no longer depends on
StreamingFlowGraphBuilder.
This CL moves StreamingScopeBuilder to a separate file
vm/compiler/frontend/scope_builder{.h, .cc}.
Also:
* StreamingScopeBuilder is renamed to ScopeBuilder.
* IntMap is moved to hash_map.h.
* Minor cleanup of unused forward declarations and friend classes.
Change-Id: I5e780a53f37935c0a1d563d4fbaec3a92ae13d6a
Reviewed-on: https://dart-review.googlesource.com/63692
Reviewed-by: Zach Anderson <zra@google.com>
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
This change continues refactoring to split kernel_binary_flowgraph and
kernel_to_il and move out parts which have no dependencies on flow graph
construction.
Bytecode reading is moved to a separate file bytecode_reader{.h, .cc}.
Metadata helpers, TypeTranslator, ActiveClass and other Active* are moved
to kernel_translation_helper.
Removed dependency of ICData tags on IL Instruction tags, so ICData objects
can be created by bytecode reader without depending on IL.
Unused AlternativeScriptScope is removed.
Change-Id: Ic9f98af82697b48013fb89802076b56833acf55d
Reviewed-on: https://dart-review.googlesource.com/63262
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Régis Crelier <regis@google.com>
Commit-Queue: Alexander Markov <alexmarkov@google.com>
We add two things:
* --print_instruction_stats makes compiler dump per IL instruction size
breakdown (how many bytes of code were produced from specific instruction
kinds). This was largely implemented by kustermann@ in
https://codereview.chromium.org/2584613002/ and this CL does only few changes
to the original implementation, namely more uniform handling of slow-path code
and puts statistics object into RawInstructions (which has free space due to
alignment) instead of RawCode.
* --print_instructions_sizes_to=symbols.json makes compiler dump per Instruction
object size breakdown into a JSON file. This JSON file can later be processed
with pkg/vm/tool/run_binary_size_analysis.dart script to produce interactive
binary size diagram similar to runtime/third_party/binary_size tool.
Change-Id: Ied4965b9a0a91b3025eefbe981ecd47cdcf782d6
Reviewed-on: https://dart-review.googlesource.com/50501
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Passes bodies are moved into compiler_pass{.cc,.h}.
Invoking a pass is just INVOKE_PASS(Name) now, instead of putting a bunch of
if-s and calls in compiler.cc or precompiler.cc.
We also consolidate ability to print IL and enable-disable passes under a
single flag --compiler-passes, e.g.
--compiler-passes=-Inlining,-CSE disable inlining and CSE passes
--compiler-passes=Inlining+ print IL after Inlining pass and all
subsequent passes
Change-Id: I90ff54b04a54f20099f5bf38dd45b16b8e3c4781
Reviewed-on: https://dart-review.googlesource.com/43968
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
R=vegorov@google.com
This removes the IA32 disassembler and uses the X64 disassembler for
both variants. Instructions that were in the IA32 assembler, but not
supported by the X64 disassembler have been added. It also adds some
regression tests for a lot of the disassembler output.
Bug:
Change-Id: I243abbb04c3a77810ce96ca74f7f42a5a1aea0cf
Reviewed-on: https://dart-review.googlesource.com/22982
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Erik Corry <erikcorry@google.com>
Rewrite most SIMD instructions on ARM using this new way.
Our current way for defining instruction backends -- a pair of two virtual
methods called MakeLocationSummary and EmitNativeCode, leads to unnecessary
duplicated and verbose code. Code generation happens in three steps:
1. For each instruction in the graph MakeLocationSummary is called to
constructing a location summary object encoding register allocation constraints;
2. When all register constraints are collected a register allocation is performed
and results are filled back into the location summaries;
3. For each instruction in the graph EmitNativeCode is called. It unpacks
location summary attached to the instruction into actual machine registers and
emits native code.
There is usually a lot of duplication between declaring register constraints in
MLS and unpacking them in ENC which this CL is trying to remove.
The new way is centered on the concept of an *emitter function* which encodes
in its signature register constraints for a particular instruction.
We use a combination of templates and macroses to enable writing
DEFINE_BACKEND(BinaryFloat32x4Op,
(QRegister result, QRegister left, QRegister right)) {
// ...
}
Instead of
LocationSummary* BinaryFloat32x4OpInstr::MakeLocationSummary(Zone* zone,
bool opt) const {
const intptr_t kNumInputs = 2;
const intptr_t kNumTemps = 0;
LocationSummary* summary = new (zone)
LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
summary->set_in(0, Location::RequiresFpuRegister());
summary->set_in(1, Location::RequiresFpuRegister());
summary->set_out(0, Location::RequiresFpuRegister());
return summary;
}
void BinaryFloat32x4OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
const QRegister left = locs()->in(0).fpu_reg();
const QRegister right = locs()->in(1).fpu_reg();
const QRegister result = locs()->out(0).fpu_reg();
// ...
}
This change also introduces a new, more handy way to work with S/D components of QRegisters, QRegister_ wrapper type.
Bug: https://github.com/dart-lang/sdk/issues/30949
Change-Id: I7bb0fc9672c89acc3d3d9b5e9859ae5a5471f420
Reviewed-on: https://dart-review.googlesource.com/11820
Commit-Queue: Martin Kustermann <kustermann@google.com>
Reviewed-by: Martin Kustermann <kustermann@google.com>
This reverts commit 04aa2b0186.
Reason for revert: Compilation is failing on some Windows builders. It may be the version of MSVC running on those machines that makes the difference. But I think we need to revert until we figure it out. The error is "runtime\vm\compiler\backend\locations_helpers_test.cc(111): error C2466: cannot allocate an array of constant size 0"
Original change's description:
> [VM] Introduce new way to define instruction backends.
>
> Rewrite most SIMD instructions on ARM using this new way.
>
> Our current way for defining instruction backends -- a pair of two virtual
> methods called MakeLocationSummary and EmitNativeCode, leads to unnecessary
> duplicated and verbose code. Code generation happens in three steps:
>
> 1. For each instruction in the graph MakeLocationSummary is called to
> constructing a location summary object encoding register allocation constraints;
> 2. When all register constraints are collected a register allocation is performed
> and results are filled back into the location summaries;
> 3. For each instruction in the graph EmitNativeCode is called. It unpacks
> location summary attached to the instruction into actual machine registers and
> emits native code.
>
> There is usually a lot of duplication between declaring register constraints in
> MLS and unpacking them in ENC which this CL is trying to remove.
>
> The new way is centered on the concept of an *emitter function* which encodes
> in its signature register constraints for a particular instruction.
>
> We use a combination of templates and macroses to enable writing
>
> DEFINE_BACKEND(BinaryFloat32x4Op,
> (QRegister result, QRegister left, QRegister right)) {
> // ...
> }
>
> Instead of
>
> LocationSummary* BinaryFloat32x4OpInstr::MakeLocationSummary(Zone* zone,
> bool opt) const {
> const intptr_t kNumInputs = 2;
> const intptr_t kNumTemps = 0;
> LocationSummary* summary = new (zone)
> LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
> summary->set_in(0, Location::RequiresFpuRegister());
> summary->set_in(1, Location::RequiresFpuRegister());
> summary->set_out(0, Location::RequiresFpuRegister());
> return summary;
> }
>
> void BinaryFloat32x4OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
> const QRegister left = locs()->in(0).fpu_reg();
> const QRegister right = locs()->in(1).fpu_reg();
> const QRegister result = locs()->out(0).fpu_reg();
> // ...
> }
>
> This change also introduces a new, more handy way to work with S/D components of QRegisters, QRegister_ wrapper type.
>
> Bug: https://github.com/dart-lang/sdk/issues/30949
> Change-Id: I7f2beb106d1458facf4a3d75cae123e1fc25d8b5
> Reviewed-on: https://dart-review.googlesource.com/11507
> Reviewed-by: Zach Anderson <zra@google.com>
> Reviewed-by: Alexander Markov <alexmarkov@google.com>
> Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
TBR=vegorov@google.com,alexmarkov@google.com,zra@google.com
Change-Id: Icfaabe58351a61d4eb50c4f9ac3bbd9677339fe7
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: https://github.com/dart-lang/sdk/issues/30949
Reviewed-on: https://dart-review.googlesource.com/11780
Reviewed-by: William Hesse <whesse@google.com>
Rewrite most SIMD instructions on ARM using this new way.
Our current way for defining instruction backends -- a pair of two virtual
methods called MakeLocationSummary and EmitNativeCode, leads to unnecessary
duplicated and verbose code. Code generation happens in three steps:
1. For each instruction in the graph MakeLocationSummary is called to
constructing a location summary object encoding register allocation constraints;
2. When all register constraints are collected a register allocation is performed
and results are filled back into the location summaries;
3. For each instruction in the graph EmitNativeCode is called. It unpacks
location summary attached to the instruction into actual machine registers and
emits native code.
There is usually a lot of duplication between declaring register constraints in
MLS and unpacking them in ENC which this CL is trying to remove.
The new way is centered on the concept of an *emitter function* which encodes
in its signature register constraints for a particular instruction.
We use a combination of templates and macroses to enable writing
DEFINE_BACKEND(BinaryFloat32x4Op,
(QRegister result, QRegister left, QRegister right)) {
// ...
}
Instead of
LocationSummary* BinaryFloat32x4OpInstr::MakeLocationSummary(Zone* zone,
bool opt) const {
const intptr_t kNumInputs = 2;
const intptr_t kNumTemps = 0;
LocationSummary* summary = new (zone)
LocationSummary(zone, kNumInputs, kNumTemps, LocationSummary::kNoCall);
summary->set_in(0, Location::RequiresFpuRegister());
summary->set_in(1, Location::RequiresFpuRegister());
summary->set_out(0, Location::RequiresFpuRegister());
return summary;
}
void BinaryFloat32x4OpInstr::EmitNativeCode(FlowGraphCompiler* compiler) {
const QRegister left = locs()->in(0).fpu_reg();
const QRegister right = locs()->in(1).fpu_reg();
const QRegister result = locs()->out(0).fpu_reg();
// ...
}
This change also introduces a new, more handy way to work with S/D components of QRegisters, QRegister_ wrapper type.
Bug: https://github.com/dart-lang/sdk/issues/30949
Change-Id: I7f2beb106d1458facf4a3d75cae123e1fc25d8b5
Reviewed-on: https://dart-review.googlesource.com/11507
Reviewed-by: Zach Anderson <zra@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Originally these passes were a single confusingly named flow_graph_optimizer
pass, they later were completely split by duplicating flow_graph_optimizer
twice as jit_optimizer.cc and aot_optimizer.cc and tweaking them in few
places.
Duplication has been done in anticipation that these passes will diverge
considerably, however simple diffing reveals that in 2 years they have
not actually diverged much at all.
Diffing also reveals numerous bugs and inconsistencies between jit_optimizer
and aot_optimizer.
It does not make sense to keep these files duplicated so this change merges
them back together - extracting most of their common behaviour into
CallSpecializer base class.
The separation is not entirely clean - for simplicity we introduce
the knowledge about precompiled_mode into CallSpecializer base class.
You can look for FLAG_precompiled_mode and FLAG_use_field_guards in the
call_specializer.cc to see where base class has to be aware about both AOT and
JIT mode.
This change also renames XyzOptimizer to XyzCallSpecializer because the main
optimization these classes were doing was specialization of calls based on
type feedback or inferred types.
Bug: https://github.com/dart-lang/sdk/issues/30575
Change-Id: I0b062c4b7549d08b1bee9303d92a3fb549f54e21
Reviewed-on: https://dart-review.googlesource.com/3640
Reviewed-by: Ryan Macnak <rmacnak@google.com>
Reviewed-by: Alexander Markov <alexmarkov@google.com>
Reviewed-by: Zach Anderson <zra@google.com>
New folder structure (nested under vm/):
- compiler/
- jit/ - JIT specific code
- aot/ - AOT specific code
- backend/ - all middle-end and back-end code (IL, flow graph)
- assembler/ - assemblers and disassemblers
- frontend/ - front ends (AST -> IL, Kernel -> IL)
compiler/README.md would be the documentation root for the compiler
pipeline
Bug: https://github.com/dart-lang/sdk/issues/30575
Change-Id: I2dfd9688793bff737f7632ddc77fca766875ce36
Reviewed-on: https://dart-review.googlesource.com/2940
Reviewed-by: Vyacheslav Egorov <vegorov@google.com>
Commit-Queue: Vyacheslav Egorov <vegorov@google.com>
Teagan Strickland