The goal of this patch is to maximize CPU utilization on multi-socket or high core count systems, so that parallel computations such as LLD/ThinLTO can use all hardware threads in the system. Before this patch, on Windows, a maximum of 64 hardware threads could be used at most, in some cases dispatched only on one CPU socket.
== Background ==
Windows doesn't have a flat cpu_set_t like Linux. Instead, it projects hardware CPUs (or NUMA nodes) to applications through a concept of "processor groups". A "processor" is the smallest unit of execution on a CPU, that is, an hyper-thread if SMT is active; a core otherwise. There's a limit of 32-bit processors on older 32-bit versions of Windows, which later was raised to 64-processors with 64-bit versions of Windows. This limit comes from the affinity mask, which historically is represented by the sizeof(void*). Consequently, the concept of "processor groups" was introduced for dealing with systems with more than 64 hyper-threads.
By default, the Windows OS assigns only one "processor group" to each starting application, in a round-robin manner. If the application wants to use more processors, it needs to programmatically enable it, by assigning threads to other "processor groups". This also means that affinity cannot cross "processor group" boundaries; one can only specify a "preferred" group on start-up, but the application is free to allocate more groups if it wants to.
This creates a peculiar situation, where newer CPUs like the AMD EPYC 7702P (64-cores, 128-hyperthreads) are projected by the OS as two (2) "processor groups". This means that by default, an application can only use half of the cores. This situation could only get worse in the years to come, as dies with more cores will appear on the market.
== The problem ==
The heavyweight_hardware_concurrency() API was introduced so that only *one hardware thread per core* was used. Once that API returns, that original intention is lost, only the number of threads is retained. Consider a situation, on Windows, where the system has 2 CPU sockets, 18 cores each, each core having 2 hyper-threads, for a total of 72 hyper-threads. Both heavyweight_hardware_concurrency() and hardware_concurrency() currently return 36, because on Windows they are simply wrappers over std:🧵:hardware_concurrency() -- which can only return processors from the current "processor group".
== The changes in this patch ==
To solve this situation, we capture (and retain) the initial intention until the point of usage, through a new ThreadPoolStrategy class. The number of threads to use is deferred as late as possible, until the moment where the std::threads are created (ThreadPool in the case of ThinLTO).
When using hardware_concurrency(), setting ThreadCount to 0 now means to use all the possible hardware CPU (SMT) threads. Providing a ThreadCount above to the maximum number of threads will have no effect, the maximum will be used instead.
The heavyweight_hardware_concurrency() is similar to hardware_concurrency(), except that only one thread per hardware *core* will be used.
When LLVM_ENABLE_THREADS is OFF, the threading APIs will always return 1, to ensure any caller loops will be exercised at least once.
Differential Revision: https://reviews.llvm.org/D71775
Summary:
This revision adds support for printing pass options as part of the normal help description. This also moves registered passes and pipelines into different sections of the help.
Example:
```
Compiler passes to run
--pass-pipeline - ...
Passes:
--affine-data-copy-generate - ...
--convert-gpu-to-spirv - ...
--workgroup-size=<long> - ...
--test-options-pass - ...
--list=<int> - ...
--string=<string> - ...
--string-list=<string> - ...
Pass Pipelines:
--test-options-pass-pipeline - ...
--list=<int> - ...
--string=<string> - ...
--string-list=<string> - ...
```
Differential Revision: https://reviews.llvm.org/D74246
This is how it should've been and brings it more in line with
std::string_view. There should be no functional change here.
This is mostly mechanical from a custom clang-tidy check, with a lot of
manual fixups. It uncovers a lot of minor inefficiencies.
This doesn't actually modify StringRef yet, I'll do that in a follow-up.
This change refactors pass options to be more similar to how statistics are modeled. More specifically, the options are specified directly on the pass instead of in a separate options class. (Note that the behavior and specification for pass pipelines remains the same.) This brings about several benefits:
* The specification of options is much simpler
* The round-trip format of a pass can be generated automatically
* This gives a somewhat deeper integration with "configuring" a pass, which we could potentially expose to users in the future.
PiperOrigin-RevId: 286953824
This is an initial step to refactoring the representation of OpResult as proposed in: https://groups.google.com/a/tensorflow.org/g/mlir/c/XXzzKhqqF_0/m/v6bKb08WCgAJ
This change will make it much simpler to incrementally transition all of the existing code to use value-typed semantics.
PiperOrigin-RevId: 286844725
This adds an additional filtering mode for printing after a pass that checks to see if the pass actually changed the IR before printing it. This "change" detection is implemented using a SHA1 hash of the current operation and its children.
PiperOrigin-RevId: 284291089
This allows for more interesting behavior from users, e.g. enabling the ability to dump the IR to a separate file for each pass invocation.
PiperOrigin-RevId: 284059447
Statistics are a way to keep track of what the compiler is doing and how effective various optimizations are. It is useful to see what optimizations are contributing to making a particular program run faster. Pass-instance specific statistics take this even further as you can see the effect of placing a particular pass at specific places within the pass pipeline, e.g. they could help answer questions like "what happens if I run CSE again here".
Statistics can be added to a pass by simply adding members of type 'Pass::Statistics'. This class takes as a constructor arguments: the parent pass pointer, a name, and a description. Statistics can be dumped by the pass manager in a similar manner to how pass timing information is dumped, i.e. via PassManager::enableStatistics programmatically; or -pass-statistics and -pass-statistics-display via the command line pass manager options.
Below is an example:
struct MyPass : public OperationPass<MyPass> {
Statistic testStat{this, "testStat", "A test statistic"};
void runOnOperation() {
...
++testStat;
...
}
};
$ mlir-opt -pass-pipeline='func(my-pass,my-pass)' foo.mlir -pass-statistics
Pipeline Display:
===-------------------------------------------------------------------------===
... Pass statistics report ...
===-------------------------------------------------------------------------===
'func' Pipeline
MyPass
(S) 15 testStat - A test statistic
MyPass
(S) 6 testStat - A test statistic
List Display:
===-------------------------------------------------------------------------===
... Pass statistics report ...
===-------------------------------------------------------------------------===
MyPass
(S) 21 testStat - A test statistic
PiperOrigin-RevId: 284022014
This causes the AsmPrinter to use a local value numbering when printing the IR, allowing for the printer to be used safely in a local context, e.g. to ensure thread-safety when printing the IR. This means that the IR printing instrumentation can also be used during multi-threading when module-scope is disabled. Operation::dump and DiagnosticArgument(Operation*) are also updated to always print local scope, as this is the most common use case when debugging.
PiperOrigin-RevId: 279988203
On failure, the IR is likely to be in an invalid state, meaning the custom printer for some operations may now crash. Using the generic op form prevents this from happening.
PiperOrigin-RevId: 274104146
This cl adds support for generating a .mlir file containing a reproducer for crashes and failures that happen during pass execution. The reproducer contains a comment detailing the configuration of the pass manager(e.g. the textual description of the pass pipeline that the pass manager was executing), along with the original input module.
Example Output:
// configuration: -pass-pipeline='func(cse, canonicalize), inline'
// note: verifyPasses=false
module {
...
}
PiperOrigin-RevId: 274088134
Allow printing out pipelines in a format that is as close as possible to the
textual pass pipeline format. Individual passes can override the print function
in order to format any options that may have been used to construct that pass.
PiperOrigin-RevId: 273813627
This allows individual passes to define options structs and for these options to be parsed per instance of the pass while building the pass pipeline from the command line provided textual specification.
The user can specify these per-instance pipeline options like so:
```
struct MyPassOptions : public PassOptions<MyPassOptions> {
Option<int> exampleOption{*this, "flag-name", llvm:🆑:desc("...")};
List<int> exampleListOption{*this, "list-flag-name", llvm:🆑:desc("...")};
};
static PassRegistration<MyPass, MyPassOptions> pass("my-pass", "description");
```
PiperOrigin-RevId: 273650140
For the cases where there are multiple levels of nested pass managers, the parent thread ID is not enough to distinguish the parent of a given pass pipeline. Passing in the parent pass gives an exact anchor point.
PiperOrigin-RevId: 272105461
OperationPass' are defined exactly the same way as they are now:
class DerivedPass : public OperationPass<DerivedPass>;
OpPass' are now defined as OperationPass, but with an additional template parameter for the operation type:
class DerivedPass : public OperationPass<DerivedPass, FuncOp>;
PiperOrigin-RevId: 269122410
This allows for users other than those on the command line to apply a textual description of a pipeline to a given pass manager.
PiperOrigin-RevId: 269017028
This allows for explicitly specifying the pipeline to add to the pass manager. This includes the nesting structure, as well as the passes/pipelines to run. A textual pipeline string is defined as a series of names, each of which may in itself recursively contain a nested pipeline description. A name is either the name of a registered pass, or pass pipeline, (e.g. "cse") or the name of an operation type (e.g. "func").
For example, the following pipeline:
$ mlir-opt foo.mlir -cse -canonicalize -lower-to-llvm
Could now be specified as:
$ mlir-opt foo.mlir -pass-pipeline='func(cse, canonicalize), lower-to-llvm'
This will allow for running pipelines on nested operations, like say spirv modules. This does not remove any of the current functionality, and in fact can be used in unison. The new option is available via 'pass-pipeline'.
PiperOrigin-RevId: 268954279
Some compilers will try to auto-generate the destructor, instead of using the user provided destructor, when creating a default move constructor.
PiperOrigin-RevId: 268067367
This allows for parallelizing across pipelines of multiple operation types. AdaptorPasses can now hold pass managers for multiple operation types and will dispatch based upon the operation being operated on.
PiperOrigin-RevId: 268017344
This is done via a new set of instrumentation hooks runBeforePipeline/runAfterPipeline, that signal the lifetime of a pass pipeline on a specific operation type. These hooks also provide the parent thread of the pipeline, allowing for accurate merging of timers running on different threads.
PiperOrigin-RevId: 267909193
This change generalizes the structure of the pass manager to allow arbitrary nesting pass managers for other operations, at any level. The only user visible change to existing code is the fact that a PassManager must now provide an MLIRContext on construction. A new class `OpPassManager` has been added that represents a pass manager on a specific operation type. `PassManager` will remain the top-level entry point into the pipeline, with OpPassManagers being nested underneath. OpPassManagers will still be implicitly nested if the operation type on the pass differs from the pass manager. To explicitly build a pipeline, the 'nest' methods on OpPassManager may be used:
// Pass manager for the top-level module.
PassManager pm(ctx);
// Nest a pipeline operating on FuncOp.
OpPassManager &fpm = pm.nest<FuncOp>();
fpm.addPass(...);
// Nest a pipeline under the FuncOp pipeline that operates on spirv::ModuleOp
OpPassManager &spvModulePM = pm.nest<spirv::ModuleOp>();
// Nest a pipeline on FuncOps inside of the spirv::ModuleOp.
OpPassManager &spvFuncPM = spvModulePM.nest<FuncOp>();
To help accomplish this a new general OperationPass is added that operates on opaque Operations. This pass can be inserted in a pass manager of any type to operate on any operation opaquely. An example of this opaque OperationPass is a VerifierPass, that simply runs the verifier opaquely on the current operation.
/// Pass to verify an operation and signal failure if necessary.
class VerifierPass : public OperationPass<VerifierPass> {
void runOnOperation() override {
Operation *op = getOperation();
if (failed(verify(op)))
signalPassFailure();
markAllAnalysesPreserved();
}
};
PiperOrigin-RevId: 266840344
This pass class generalizes the current functionality between FunctionPass and ModulePass, and allows for operating on any operation type. The pass manager currently only supports OpPasses operating on FuncOp and ModuleOp, but this restriction will be relaxed in follow-up changes. A utility class OpPassBase<OpT> allows for generically referring to operation specific passes: e.g. FunctionPassBase == OpPassBase<FuncOp>.
PiperOrigin-RevId: 266442239
The pass manager is moving towards being able to run on operations at arbitrary nesting. An operation may have both parent and child operations, and the AnalysisManager must be able to handle this generalization. The AnalysisManager class now contains generic 'getCachedParentAnalysis' and 'getChildAnalysis/getCachedChildAnalysis' functions to query analyses on parent/child operations. This removes the hard coded nesting relationship between Module/Function.
PiperOrigin-RevId: 266003636
Switch to C++14 standard method as llvm::make_unique has been removed (
https://reviews.llvm.org/D66259). Also mark some targets as c++14 to ease next
integrates.
PiperOrigin-RevId: 263953918
Since raw pointers are always passed around for IR construct without
implying any ownership transfer, it can be error prone to have implicit
ownership transferred the same way.
For example this code can seem harmless:
Pass *pass = ....
pm.addPass(pass);
pm.addPass(pass);
pm.run(module);
PiperOrigin-RevId: 263053082
The verifier passes are NO-OP and are only useful to print after in the case of failure. This removes a lot of unnecessary clutter when printing after/before all passes.
PiperOrigin-RevId: 257836310
There is already a more general 'getParentOfType' method, and 'getModule' is likely to be misused as functions get placed within different regions than ModuleOp.
PiperOrigin-RevId: 257442243
Modules can now contain more than just Functions, this just updates the iteration API to reflect that. The 'begin'/'end' methods have also been updated to iterate over opaque Operations.
PiperOrigin-RevId: 257099084
As Functions/Modules becomes operations, these methods will conflict with the 'verify' hook already on derived operation types.
PiperOrigin-RevId: 256246112
As with Functions, Module will soon become an operation, which are value-typed. This eases the transition from Module to ModuleOp. A new class, OwningModuleRef is provided to allow for owning a reference to a Module, and will auto-delete the held module on destruction.
PiperOrigin-RevId: 256196193
Move the data members out of Function and into a new impl storage class 'FunctionStorage'. This allows for Function to become value typed, which will greatly simplify the transition of Function to FuncOp(given that FuncOp is also value typed).
PiperOrigin-RevId: 255983022
The Diagnostic class contains all of the information necessary to report a diagnostic to the DiagnosticEngine. It should generally not be constructed directly, and instead used transitively via InFlightDiagnostic. A diagnostic is currently comprised of several different elements:
* A severity level.
* A source Location.
* A list of DiagnosticArguments that help compose and comprise the output message.
* A DiagnosticArgument represents any value that may be part of the diagnostic, e.g. string, integer, Type, Attribute, etc.
* Arguments can be added to the diagnostic via the stream(<<) operator.
* (In a future cl) A list of attached notes.
* These are in the form of other diagnostics that provide supplemental information to the main diagnostic, but do not have context on their own.
The InFlightDiagnostic class represents an RAII wrapper around a Diagnostic that is set to be reported with the diagnostic engine. This allows for the user to modify a diagnostic that is inflight. The internally wrapped diagnostic can be reported directly or automatically upon destruction.
These classes allow for more natural composition of diagnostics by removing the restriction that the message of a diagnostic is comprised of a single Twine. They should also allow for nice incremental improvements to the diagnostics experience in the future, e.g. formatv style diagnostics.
Simple Example:
emitError(loc, "integer bitwidth is limited to " + Twine(IntegerType::kMaxWidth) + " bits");
emitError(loc) << "integer bitwidth is limited to " << IntegerType::kMaxWidth << " bits";
--
PiperOrigin-RevId: 246526439
When multi-threading is enabled in the pass manager the meaning of the display
slightly changes. First, a new timing column is added, `User Time`, that
displays the total time spent across all threads. Secondly, the `Wall Time`
column displays the longest individual time spent amongst all of the threads.
This means that the `Wall Time` column will continue to give an indicator on the
perceived time, or clock time, whereas the `User Time` will display the total
cpu time.
Example:
$ mlir-opt foo.mlir -experimental-mt-pm -cse -canonicalize -convert-to-llvmir -pass-timing
===-------------------------------------------------------------------------===
... Pass execution timing report ...
===-------------------------------------------------------------------------===
Total Execution Time: 0.0078 seconds
---User Time--- ---Wall Time--- --- Name ---
0.0175 ( 88.3%) 0.0055 ( 70.4%) Function Pipeline
0.0018 ( 9.3%) 0.0006 ( 8.1%) CSE
0.0013 ( 6.3%) 0.0004 ( 5.8%) (A) DominanceInfo
0.0017 ( 8.7%) 0.0006 ( 7.1%) FunctionVerifier
0.0128 ( 64.6%) 0.0039 ( 50.5%) Canonicalizer
0.0011 ( 5.7%) 0.0004 ( 4.7%) FunctionVerifier
0.0004 ( 2.1%) 0.0004 ( 5.2%) ModuleVerifier
0.0010 ( 5.3%) 0.0010 ( 13.4%) LLVMLowering
0.0009 ( 4.3%) 0.0009 ( 11.0%) ModuleVerifier
0.0198 (100.0%) 0.0078 (100.0%) Total
PiperOrigin-RevId: 240636269
a pointer. This makes it consistent with all the other methods in
FunctionPass, as well as with ModulePass::getModule(). NFC.
PiperOrigin-RevId: 240257910
* print-ir-before=(comma-separated-pass-list)
- Print the IR before each of the passes provided within the pass list.
* print-ir-before-all
- Print the IR before every pass in the pipeline.
* print-ir-after=(comma-separated-pass-list)
- Print the IR after each of the passes provided within the pass list.
* print-ir-after-all
- Print the IR after every pass in the pipeline.
* print-ir-module-scope
- Always print the Module IR, even for non module passes.
PiperOrigin-RevId: 238523649
Below shows the output for an example mlir-opt command line.
mlir-opt foo.mlir -verify-each=false -cse -canonicalize -cse -cse -pass-timing
list view (-pass-timing-display=list):
* In this mode the results are displayed in a list sorted by total time; with each pass/analysis instance aggregated into one unique result. This mode is similar to the output of 'time-passes' in llvm-opt.
===-------------------------------------------------------------------------===
... Pass execution timing report ...
===-------------------------------------------------------------------------===
Total Execution Time: 0.0097 seconds (0.0096 wall clock)
---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name ---
0.0051 ( 58.3%) 0.0001 ( 12.2%) 0.0052 ( 53.8%) 0.0052 ( 53.8%) Canonicalizer
0.0025 ( 29.1%) 0.0005 ( 58.2%) 0.0031 ( 31.9%) 0.0031 ( 32.0%) CSE
0.0011 ( 12.6%) 0.0003 ( 29.7%) 0.0014 ( 14.3%) 0.0014 ( 14.2%) DominanceInfo
0.0087 (100.0%) 0.0009 (100.0%) 0.0097 (100.0%) 0.0096 (100.0%) Total
pipeline view (-pass-timing-display=pipeline):
* In this mode the results are displayed in a nested pipeline view that mirrors the internal pass pipeline that is being executed in the pass manager. This view is useful for understanding specifically which parts of the pipeline are taking the most time, and can also be used to identify when analyses are being invalidated and recomputed.
===-------------------------------------------------------------------------===
... Pass execution timing report ...
===-------------------------------------------------------------------------===
Total Execution Time: 0.0082 seconds (0.0081 wall clock)
---User Time--- --System Time-- --User+System-- ---Wall Time--- --- Name ---
0.0042 (100.0%) 0.0039 (100.0%) 0.0082 (100.0%) 0.0081 (100.0%) Function Pipeline
0.0005 ( 11.6%) 0.0008 ( 21.1%) 0.0013 ( 16.1%) 0.0013 ( 16.2%) CSE
0.0002 ( 5.0%) 0.0004 ( 9.3%) 0.0006 ( 7.0%) 0.0006 ( 7.0%) (A) DominanceInfo
0.0026 ( 61.8%) 0.0018 ( 45.6%) 0.0044 ( 54.0%) 0.0044 ( 54.1%) Canonicalizer
0.0005 ( 11.7%) 0.0005 ( 13.0%) 0.0010 ( 12.3%) 0.0010 ( 12.4%) CSE
0.0003 ( 6.1%) 0.0003 ( 8.3%) 0.0006 ( 7.2%) 0.0006 ( 7.1%) (A) DominanceInfo
0.0002 ( 3.8%) 0.0001 ( 2.8%) 0.0003 ( 3.3%) 0.0003 ( 3.3%) CSE
0.0042 (100.0%) 0.0039 (100.0%) 0.0082 (100.0%) 0.0081 (100.0%) Total
PiperOrigin-RevId: 237825367
* before/after pass execution
* after a pass fails
* before/after an analysis is computed
After getting this infrastructure in place, we can start providing common developer utilities like pass timing, IR printing after pass execution, etc.
PiperOrigin-RevId: 237709692
- change this for consistency - everything else similar takes/returns a
Function pointer - the FuncBuilder ctor,
Block/Value/Instruction::getFunction(), etc.
- saves a whole bunch of &s everywhere
PiperOrigin-RevId: 236928761
An analysis can be any class, but it must provide the following:
* A constructor for a given IR unit.
struct MyAnalysis {
// Compute this analysis with the provided module.
MyAnalysis(Module *module);
};
Analyses can be accessed from a Pass by calling either the 'getAnalysisResult<AnalysisT>' or 'getCachedAnalysisResult<AnalysisT>' methods. A FunctionPass may query for a cached analysis on the parent module with 'getCachedModuleAnalysisResult'. Similary, a ModulePass may query an analysis, it doesn't need to be cached, on a child function with 'getFunctionAnalysisResult'.
By default, when running a pass all cached analyses are set to be invalidated. If no transformation was performed, a pass can use the method 'markAllAnalysesPreserved' to preserve all analysis results. As noted above, preserving specific analyses is not yet supported.
PiperOrigin-RevId: 236505642
The definitions of derived passes have now changed and passes must adhere to the following:
* Inherit from a CRTP base class FunctionPass/ModulePass.
- This class provides several necessary utilities for the transformation:
. Access to the IR unit being transformed (getFunction/getModule)
. Various utilities for pass identification and registration.
* Provide a 'PassResult runOn(Function|Module)()' method to transform the IR.
- This replaces the runOn* functions from before.
This patch also introduces the notion of the PassManager. This allows for simplified construction of pass pipelines and acts as the sole interface for executing passes. This is important as FunctionPass will no longer have a 'runOnModule' method.
PiperOrigin-RevId: 235952008
Benjamin Kramer