2015-03-07 17:02:36 +08:00
|
|
|
//===- Parsing, selection, and construction of pass pipelines -------------===//
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
/// \file
|
|
|
|
///
|
2015-03-07 17:02:36 +08:00
|
|
|
/// This file provides the implementation of the PassBuilder based on our
|
|
|
|
/// static pass registry as well as related functionality. It also provides
|
|
|
|
/// helpers to aid in analyzing, debugging, and testing passes and pass
|
|
|
|
/// pipelines.
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
///
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
#include "llvm/Passes/PassBuilder.h"
|
2016-02-29 06:16:03 +08:00
|
|
|
#include "llvm/ADT/StringSwitch.h"
|
2016-02-14 07:32:00 +08:00
|
|
|
#include "llvm/Analysis/AliasAnalysis.h"
|
2016-02-20 11:46:03 +08:00
|
|
|
#include "llvm/Analysis/AliasAnalysisEvaluator.h"
|
2016-12-19 16:22:17 +08:00
|
|
|
#include "llvm/Analysis/AssumptionCache.h"
|
2016-02-14 07:46:24 +08:00
|
|
|
#include "llvm/Analysis/BasicAliasAnalysis.h"
|
2016-05-06 05:13:27 +08:00
|
|
|
#include "llvm/Analysis/BlockFrequencyInfo.h"
|
|
|
|
#include "llvm/Analysis/BlockFrequencyInfoImpl.h"
|
2016-05-05 10:59:57 +08:00
|
|
|
#include "llvm/Analysis/BranchProbabilityInfo.h"
|
2016-09-17 00:56:30 +08:00
|
|
|
#include "llvm/Analysis/CFGPrinter.h"
|
2016-07-06 08:26:41 +08:00
|
|
|
#include "llvm/Analysis/CFLAndersAliasAnalysis.h"
|
|
|
|
#include "llvm/Analysis/CFLSteensAliasAnalysis.h"
|
2014-04-21 19:12:00 +08:00
|
|
|
#include "llvm/Analysis/CGSCCPassManager.h"
|
2016-03-10 19:24:11 +08:00
|
|
|
#include "llvm/Analysis/CallGraph.h"
|
2016-04-19 07:55:01 +08:00
|
|
|
#include "llvm/Analysis/DemandedBits.h"
|
2016-05-13 06:19:39 +08:00
|
|
|
#include "llvm/Analysis/DependenceAnalysis.h"
|
2016-02-26 01:54:15 +08:00
|
|
|
#include "llvm/Analysis/DominanceFrontier.h"
|
2016-03-11 17:15:11 +08:00
|
|
|
#include "llvm/Analysis/GlobalsModRef.h"
|
2016-07-17 06:51:33 +08:00
|
|
|
#include "llvm/Analysis/IVUsers.h"
|
2014-02-06 12:37:03 +08:00
|
|
|
#include "llvm/Analysis/LazyCallGraph.h"
|
2016-06-14 06:01:25 +08:00
|
|
|
#include "llvm/Analysis/LazyValueInfo.h"
|
2016-07-03 05:18:40 +08:00
|
|
|
#include "llvm/Analysis/LoopAccessAnalysis.h"
|
2015-01-20 18:58:50 +08:00
|
|
|
#include "llvm/Analysis/LoopInfo.h"
|
2016-03-10 08:55:30 +08:00
|
|
|
#include "llvm/Analysis/MemoryDependenceAnalysis.h"
|
2017-04-12 04:06:36 +08:00
|
|
|
#include "llvm/Analysis/MemorySSA.h"
|
2016-08-12 21:53:02 +08:00
|
|
|
#include "llvm/Analysis/ModuleSummaryAnalysis.h"
|
2016-07-19 00:29:21 +08:00
|
|
|
#include "llvm/Analysis/OptimizationDiagnosticInfo.h"
|
2016-02-26 01:54:07 +08:00
|
|
|
#include "llvm/Analysis/PostDominators.h"
|
2016-06-04 06:54:26 +08:00
|
|
|
#include "llvm/Analysis/ProfileSummaryInfo.h"
|
2016-02-26 01:54:25 +08:00
|
|
|
#include "llvm/Analysis/RegionInfo.h"
|
[PM] Port ScalarEvolution to the new pass manager.
This change makes ScalarEvolution a stand-alone object and just produces
one from a pass as needed. Making this work well requires making the
object movable, using references instead of overwritten pointers in
a number of places, and other refactorings.
I've also wired it up to the new pass manager and added a RUN line to
a test to exercise it under the new pass manager. This includes basic
printing support much like with other analyses.
But there is a big and somewhat scary change here. Prior to this patch
ScalarEvolution was never *actually* invalidated!!! Re-running the pass
just re-wired up the various other analyses and didn't remove any of the
existing entries in the SCEV caches or clear out anything at all. This
might seem OK as everything in SCEV that can uses ValueHandles to track
updates to the values that serve as SCEV keys. However, this still means
that as we ran SCEV over each function in the module, we kept
accumulating more and more SCEVs into the cache. At the end, we would
have a SCEV cache with every value that we ever needed a SCEV for in the
entire module!!! Yowzers. The releaseMemory routine would dump all of
this, but that isn't realy called during normal runs of the pipeline as
far as I can see.
To make matters worse, there *is* actually a key that we don't update
with value handles -- there is a map keyed off of Loop*s. Because
LoopInfo *does* release its memory from run to run, it is entirely
possible to run SCEV over one function, then over another function, and
then lookup a Loop* from the second function but find an entry inserted
for the first function! Ouch.
To make matters still worse, there are plenty of updates that *don't*
trip a value handle. It seems incredibly unlikely that today GVN or
another pass that invalidates SCEV can update values in *just* such
a way that a subsequent run of SCEV will incorrectly find lookups in
a cache, but it is theoretically possible and would be a nightmare to
debug.
With this refactoring, I've fixed all this by actually destroying and
recreating the ScalarEvolution object from run to run. Technically, this
could increase the amount of malloc traffic we see, but then again it is
also technically correct. ;] I don't actually think we're suffering from
tons of malloc traffic from SCEV because if we were, the fact that we
never clear the memory would seem more likely to have come up as an
actual problem before now. So, I've made the simple fix here. If in fact
there are serious issues with too much allocation and deallocation,
I can work on a clever fix that preserves the allocations (while
clearing the data) between each run, but I'd prefer to do that kind of
optimization with a test case / benchmark that shows why we need such
cleverness (and that can test that we actually make it faster). It's
possible that this will make some things faster by making the SCEV
caches have higher locality (due to being significantly smaller) so
until there is a clear benchmark, I think the simple change is best.
Differential Revision: http://reviews.llvm.org/D12063
llvm-svn: 245193
2015-08-17 10:08:17 +08:00
|
|
|
#include "llvm/Analysis/ScalarEvolution.h"
|
2016-02-20 12:01:45 +08:00
|
|
|
#include "llvm/Analysis/ScalarEvolutionAliasAnalysis.h"
|
2016-02-20 12:03:06 +08:00
|
|
|
#include "llvm/Analysis/ScopedNoAliasAA.h"
|
2015-01-15 19:39:46 +08:00
|
|
|
#include "llvm/Analysis/TargetLibraryInfo.h"
|
2015-02-01 18:11:22 +08:00
|
|
|
#include "llvm/Analysis/TargetTransformInfo.h"
|
2016-02-20 12:04:52 +08:00
|
|
|
#include "llvm/Analysis/TypeBasedAliasAnalysis.h"
|
2016-06-25 04:13:42 +08:00
|
|
|
#include "llvm/CodeGen/PreISelIntrinsicLowering.h"
|
2016-07-08 11:32:49 +08:00
|
|
|
#include "llvm/CodeGen/UnreachableBlockElim.h"
|
2015-01-14 18:19:28 +08:00
|
|
|
#include "llvm/IR/Dominators.h"
|
2014-01-12 20:15:39 +08:00
|
|
|
#include "llvm/IR/IRPrintingPasses.h"
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
#include "llvm/IR/PassManager.h"
|
2014-01-20 19:34:08 +08:00
|
|
|
#include "llvm/IR/Verifier.h"
|
2014-01-12 20:15:39 +08:00
|
|
|
#include "llvm/Support/Debug.h"
|
2016-02-29 06:16:03 +08:00
|
|
|
#include "llvm/Support/Regex.h"
|
2015-02-01 18:11:22 +08:00
|
|
|
#include "llvm/Target/TargetMachine.h"
|
2016-06-05 13:12:23 +08:00
|
|
|
#include "llvm/Transforms/GCOVProfiler.h"
|
[PM] Port the always inliner to the new pass manager in a much more
minimal and boring form than the old pass manager's version.
This pass does the very minimal amount of work necessary to inline
functions declared as always-inline. It doesn't support a wide array of
things that the legacy pass manager did support, but is alse ... about
20 lines of code. So it has that going for it. Notably things this
doesn't support:
- Array alloca merging
- To support the above, bottom-up inlining with careful history
tracking and call graph updates
- DCE of the functions that become dead after this inlining.
- Inlining through call instructions with the always_inline attribute.
Instead, it focuses on inlining functions with that attribute.
The first I've omitted because I'm hoping to just turn it off for the
primary pass manager. If that doesn't pan out, I can add it here but it
will be reasonably expensive to do so.
The second should really be handled by running global-dce after the
inliner. I don't want to re-implement the non-trivial logic necessary to
do comdat-correct DCE of functions. This means the -O0 pipeline will
have to be at least 'always-inline,global-dce', but that seems
reasonable to me. If others are seriously worried about this I'd like to
hear about it and understand why. Again, this is all solveable by
factoring that logic into a utility and calling it here, but I'd like to
wait to do that until there is a clear reason why the existing
pass-based factoring won't work.
The final point is a serious one. I can fairly easily add support for
this, but it seems both costly and a confusing construct for the use
case of the always inliner running at -O0. This attribute can of course
still impact the normal inliner easily (although I find that
a questionable re-use of the same attribute). I've started a discussion
to sort out what semantics we want here and based on that can figure out
if it makes sense ta have this complexity at O0 or not.
One other advantage of this design is that it should be quite a bit
faster due to checking for whether the function is a viable candidate
for inlining exactly once per function instead of doing it for each call
site.
Anyways, hopefully a reasonable starting point for this pass.
Differential Revision: https://reviews.llvm.org/D23299
llvm-svn: 278896
2016-08-17 10:56:20 +08:00
|
|
|
#include "llvm/Transforms/IPO/AlwaysInliner.h"
|
2017-02-10 07:46:27 +08:00
|
|
|
#include "llvm/Transforms/IPO/ArgumentPromotion.h"
|
2016-05-05 08:51:09 +08:00
|
|
|
#include "llvm/Transforms/IPO/ConstantMerge.h"
|
2016-07-09 11:25:35 +08:00
|
|
|
#include "llvm/Transforms/IPO/CrossDSOCFI.h"
|
2016-06-12 17:16:39 +08:00
|
|
|
#include "llvm/Transforms/IPO/DeadArgumentElimination.h"
|
2016-05-05 10:37:32 +08:00
|
|
|
#include "llvm/Transforms/IPO/ElimAvailExtern.h"
|
2015-12-27 16:13:45 +08:00
|
|
|
#include "llvm/Transforms/IPO/ForceFunctionAttrs.h"
|
2016-02-18 19:03:11 +08:00
|
|
|
#include "llvm/Transforms/IPO/FunctionAttrs.h"
|
2016-07-19 05:22:24 +08:00
|
|
|
#include "llvm/Transforms/IPO/FunctionImport.h"
|
2016-05-04 03:39:15 +08:00
|
|
|
#include "llvm/Transforms/IPO/GlobalDCE.h"
|
2016-04-26 08:28:01 +08:00
|
|
|
#include "llvm/Transforms/IPO/GlobalOpt.h"
|
2016-11-21 08:28:23 +08:00
|
|
|
#include "llvm/Transforms/IPO/GlobalSplit.h"
|
2015-12-27 16:41:34 +08:00
|
|
|
#include "llvm/Transforms/IPO/InferFunctionAttrs.h"
|
[PM] Provide an initial, minimal port of the inliner to the new pass manager.
This doesn't implement *every* feature of the existing inliner, but
tries to implement the most important ones for building a functional
optimization pipeline and beginning to sort out bugs, regressions, and
other problems.
Notable, but intentional omissions:
- No alloca merging support. Why? Because it isn't clear we want to do
this at all. Active discussion and investigation is going on to remove
it, so for simplicity I omitted it.
- No support for trying to iterate on "internally" devirtualized calls.
Why? Because it adds what I suspect is inappropriate coupling for
little or no benefit. We will have an outer iteration system that
tracks devirtualization including that from function passes and
iterates already. We should improve that rather than approximate it
here.
- Optimization remarks. Why? Purely to make the patch smaller, no other
reason at all.
The last one I'll probably work on almost immediately. But I wanted to
skip it in the initial patch to try to focus the change as much as
possible as there is already a lot of code moving around and both of
these *could* be skipped without really disrupting the core logic.
A summary of the different things happening here:
1) Adding the usual new PM class and rigging.
2) Fixing minor underlying assumptions in the inline cost analysis or
inline logic that don't generally hold in the new PM world.
3) Adding the core pass logic which is in essence a loop over the calls
in the nodes in the call graph. This is a bit duplicated from the old
inliner, but only a handful of lines could realistically be shared.
(I tried at first, and it really didn't help anything.) All told,
this is only about 100 lines of code, and most of that is the
mechanics of wiring up analyses from the new PM world.
4) Updating the LazyCallGraph (in the new PM) based on the *newly
inlined* calls and references. This is very minimal because we cannot
form cycles.
5) When inlining removes the last use of a function, eagerly nuking the
body of the function so that any "one use remaining" inline cost
heuristics are immediately refined, and queuing these functions to be
completely deleted once inlining is complete and the call graph
updated to reflect that they have become dead.
6) After all the inlining for a particular function, updating the
LazyCallGraph and the CGSCC pass manager to reflect the
function-local simplifications that are done immediately and
internally by the inline utilties. These are the exact same
fundamental set of CG updates done by arbitrary function passes.
7) Adding a bunch of test cases to specifically target CGSCC and other
subtle aspects in the new PM world.
Many thanks to the careful review from Easwaran and Sanjoy and others!
Differential Revision: https://reviews.llvm.org/D24226
llvm-svn: 290161
2016-12-20 11:15:32 +08:00
|
|
|
#include "llvm/Transforms/IPO/Inliner.h"
|
2016-04-27 04:15:52 +08:00
|
|
|
#include "llvm/Transforms/IPO/Internalize.h"
|
2016-07-12 02:10:06 +08:00
|
|
|
#include "llvm/Transforms/IPO/LowerTypeTests.h"
|
2016-06-28 00:50:18 +08:00
|
|
|
#include "llvm/Transforms/IPO/PartialInlining.h"
|
2016-05-06 05:05:36 +08:00
|
|
|
#include "llvm/Transforms/IPO/SCCP.h"
|
2015-10-31 07:28:12 +08:00
|
|
|
#include "llvm/Transforms/IPO/StripDeadPrototypes.h"
|
2016-06-15 05:44:19 +08:00
|
|
|
#include "llvm/Transforms/IPO/WholeProgramDevirt.h"
|
2015-12-27 16:13:45 +08:00
|
|
|
#include "llvm/Transforms/InstCombine/InstCombine.h"
|
2016-04-19 01:47:38 +08:00
|
|
|
#include "llvm/Transforms/InstrProfiling.h"
|
2016-05-06 13:49:19 +08:00
|
|
|
#include "llvm/Transforms/PGOInstrumentation.h"
|
2016-05-28 07:20:16 +08:00
|
|
|
#include "llvm/Transforms/SampleProfile.h"
|
2015-10-31 07:13:18 +08:00
|
|
|
#include "llvm/Transforms/Scalar/ADCE.h"
|
2016-06-15 14:18:01 +08:00
|
|
|
#include "llvm/Transforms/Scalar/AlignmentFromAssumptions.h"
|
2016-05-25 09:57:04 +08:00
|
|
|
#include "llvm/Transforms/Scalar/BDCE.h"
|
2016-07-19 00:29:17 +08:00
|
|
|
#include "llvm/Transforms/Scalar/ConstantHoisting.h"
|
2016-07-07 07:26:29 +08:00
|
|
|
#include "llvm/Transforms/Scalar/CorrelatedValuePropagation.h"
|
2016-04-23 03:40:41 +08:00
|
|
|
#include "llvm/Transforms/Scalar/DCE.h"
|
2016-05-18 05:38:13 +08:00
|
|
|
#include "llvm/Transforms/Scalar/DeadStoreElimination.h"
|
2015-02-01 18:51:23 +08:00
|
|
|
#include "llvm/Transforms/Scalar/EarlyCSE.h"
|
2016-06-25 07:32:02 +08:00
|
|
|
#include "llvm/Transforms/Scalar/Float2Int.h"
|
2016-06-04 06:54:26 +08:00
|
|
|
#include "llvm/Transforms/Scalar/GVN.h"
|
2016-06-14 08:49:23 +08:00
|
|
|
#include "llvm/Transforms/Scalar/GuardWidening.h"
|
2017-01-11 17:43:56 +08:00
|
|
|
#include "llvm/Transforms/Scalar/IVUsersPrinter.h"
|
2016-06-06 02:01:19 +08:00
|
|
|
#include "llvm/Transforms/Scalar/IndVarSimplify.h"
|
2016-06-14 08:51:09 +08:00
|
|
|
#include "llvm/Transforms/Scalar/JumpThreading.h"
|
2016-07-13 06:42:24 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LICM.h"
|
2017-01-11 17:43:56 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopAccessAnalysisPrinter.h"
|
2016-08-13 12:11:27 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopDataPrefetch.h"
|
2016-07-15 02:28:29 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopDeletion.h"
|
2016-07-19 00:29:27 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopDistribute.h"
|
2016-07-13 02:45:51 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopIdiomRecognize.h"
|
2016-07-16 00:42:11 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopInstSimplify.h"
|
2017-01-27 09:32:26 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopLoadElimination.h"
|
2017-01-11 17:43:56 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopPassManager.h"
|
2017-01-26 00:00:44 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopPredication.h"
|
2016-05-04 06:02:31 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopRotation.h"
|
2016-05-04 05:47:32 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopSimplifyCFG.h"
|
2017-01-20 16:42:19 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopSink.h"
|
2016-07-19 05:41:50 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopStrengthReduce.h"
|
2016-07-20 07:54:23 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LoopUnrollPass.h"
|
2016-05-14 06:52:35 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LowerAtomic.h"
|
2015-01-24 19:13:02 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LowerExpectIntrinsic.h"
|
2016-07-29 06:08:41 +08:00
|
|
|
#include "llvm/Transforms/Scalar/LowerGuardIntrinsic.h"
|
2016-06-14 10:44:55 +08:00
|
|
|
#include "llvm/Transforms/Scalar/MemCpyOptimizer.h"
|
2016-06-18 03:10:09 +08:00
|
|
|
#include "llvm/Transforms/Scalar/MergedLoadStoreMotion.h"
|
2016-07-22 06:28:52 +08:00
|
|
|
#include "llvm/Transforms/Scalar/NaryReassociate.h"
|
2016-12-23 00:35:02 +08:00
|
|
|
#include "llvm/Transforms/Scalar/NewGVN.h"
|
2016-05-26 07:38:53 +08:00
|
|
|
#include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h"
|
2016-04-27 07:39:29 +08:00
|
|
|
#include "llvm/Transforms/Scalar/Reassociate.h"
|
2016-05-18 23:18:25 +08:00
|
|
|
#include "llvm/Transforms/Scalar/SCCP.h"
|
2015-09-12 17:09:14 +08:00
|
|
|
#include "llvm/Transforms/Scalar/SROA.h"
|
[PM/LoopUnswitch] Introduce a new, simpler loop unswitch pass.
Currently, this pass only focuses on *trivial* loop unswitching. At that
reduced problem it remains significantly better than the current loop
unswitch:
- Old pass is worse than cubic complexity. New pass is (I think) linear.
- New pass is much simpler in its design by focusing on full unswitching. (See
below for details on this).
- New pass doesn't carry state for thresholds between pass iterations.
- New pass doesn't carry state for correctness (both miscompile and
infloop) between pass iterations.
- New pass produces substantially better code after unswitching.
- New pass can handle more trivial unswitch cases.
- New pass doesn't recompute the dominator tree for the entire function
and instead incrementally updates it.
I've ported all of the trivial unswitching test cases from the old pass
to the new one to make sure that major functionality isn't lost in the
process. For several of the test cases I've worked to improve the
precision and rigor of the CHECKs, but for many I've just updated them
to handle the new IR produced.
My initial motivation was the fact that the old pass carried state in
very unreliable ways between pass iterations, and these mechansims were
incompatible with the new pass manager. However, I discovered many more
improvements to make along the way.
This pass makes two very significant assumptions that enable most of these
improvements:
1) Focus on *full* unswitching -- that is, completely removing whatever
control flow construct is being unswitched from the loop. In the case
of trivial unswitching, this means removing the trivial (exiting)
edge. In non-trivial unswitching, this means removing the branch or
switch itself. This is in opposition to *partial* unswitching where
some part of the unswitched control flow remains in the loop. Partial
unswitching only really applies to switches and to folded branches.
These are very similar to full unrolling and partial unrolling. The
full form is an effective canonicalization, the partial form needs
a complex cost model, cannot be iterated, isn't canonicalizing, and
should be a separate pass that runs very late (much like unrolling).
2) Leverage LLVM's Loop machinery to the fullest. The original unswitch
dates from a time when a great deal of LLVM's loop infrastructure was
missing, ineffective, and/or unreliable. As a consequence, a lot of
complexity was added which we no longer need.
With these two overarching principles, I think we can build a fast and
effective unswitcher that fits in well in the new PM and in the
canonicalization pipeline. Some of the remaining functionality around
partial unswitching may not be relevant today (not many test cases or
benchmarks I can find) but if they are I'd like to add support for them
as a separate layer that runs very late in the pipeline.
Purely to make reviewing and introducing this code more manageable, I've
split this into first a trivial-unswitch-only pass and in the next patch
I'll add support for full non-trivial unswitching against a *fixed*
threshold, exactly like full unrolling. I even plan to re-use the
unrolling thresholds, as these are incredibly similar cost tradeoffs:
we're cloning a loop body in order to end up with simplified control
flow. We should only do that when the total growth is reasonably small.
One of the biggest changes with this pass compared to the previous one
is that previously, each individual trivial exiting edge from a switch
was unswitched separately as a branch. Now, we unswitch the entire
switch at once, with cases going to the various destinations. This lets
us unswitch multiple exiting edges in a single operation and also avoids
numerous extremely bad behaviors, where we would introduce 1000s of
branches to test for thousands of possible values, all of which would
take the exact same exit path bypassing the loop. Now we will use
a switch with 1000s of cases that can be efficiently lowered into
a jumptable. This avoids relying on somehow forming a switch out of the
branches or getting horrible code if that fails for any reason.
Another significant change is that this pass actively updates the CFG
based on unswitching. For trivial unswitching, this is actually very
easy because of the definition of loop simplified form. Doing this makes
the code coming out of loop unswitch dramatically more friendly. We
still should run loop-simplifycfg (at the least) after this to clean up,
but it will have to do a lot less work.
Finally, this pass makes much fewer attempts to simplify instructions
based on the unswitch. Something like loop-instsimplify, instcombine, or
GVN can be used to do increasingly powerful simplifications based on the
now dominating predicate. The old simplifications are things that
something like loop-instsimplify should get today or a very, very basic
loop-instcombine could get. Keeping that logic separate is a big
simplifying technique.
Most of the code in this pass that isn't in the old one has to do with
achieving specific goals:
- Updating the dominator tree as we go
- Unswitching all cases in a switch in a single step.
I think it is still shorter than just the trivial unswitching code in
the old pass despite having this functionality.
Differential Revision: https://reviews.llvm.org/D32409
llvm-svn: 301576
2017-04-28 02:45:20 +08:00
|
|
|
#include "llvm/Transforms/Scalar/SimpleLoopUnswitch.h"
|
2015-12-27 16:13:45 +08:00
|
|
|
#include "llvm/Transforms/Scalar/SimplifyCFG.h"
|
2016-04-23 03:54:10 +08:00
|
|
|
#include "llvm/Transforms/Scalar/Sink.h"
|
2016-08-02 05:48:33 +08:00
|
|
|
#include "llvm/Transforms/Scalar/SpeculativeExecution.h"
|
2016-07-07 07:48:41 +08:00
|
|
|
#include "llvm/Transforms/Scalar/TailRecursionElimination.h"
|
2016-06-16 05:51:30 +08:00
|
|
|
#include "llvm/Transforms/Utils/AddDiscriminators.h"
|
2016-07-23 02:04:25 +08:00
|
|
|
#include "llvm/Transforms/Utils/BreakCriticalEdges.h"
|
2016-06-10 03:44:46 +08:00
|
|
|
#include "llvm/Transforms/Utils/LCSSA.h"
|
Conditionally eliminate library calls where the result value is not used
Summary:
This pass shrink-wraps a condition to some library calls where the call
result is not used. For example:
sqrt(val);
is transformed to
if (val < 0)
sqrt(val);
Even if the result of library call is not being used, the compiler cannot
safely delete the call because the function can set errno on error
conditions.
Note in many functions, the error condition solely depends on the incoming
parameter. In this optimization, we can generate the condition can lead to
the errno to shrink-wrap the call. Since the chances of hitting the error
condition is low, the runtime call is effectively eliminated.
These partially dead calls are usually results of C++ abstraction penalty
exposed by inlining. This optimization hits 108 times in 19 C/C++ programs
in SPEC2006.
Reviewers: hfinkel, mehdi_amini, davidxl
Subscribers: modocache, mgorny, mehdi_amini, xur, llvm-commits, beanz
Differential Revision: https://reviews.llvm.org/D24414
llvm-svn: 284542
2016-10-19 05:36:27 +08:00
|
|
|
#include "llvm/Transforms/Utils/LibCallsShrinkWrap.h"
|
2016-07-09 11:03:01 +08:00
|
|
|
#include "llvm/Transforms/Utils/LoopSimplify.h"
|
2016-08-13 01:28:27 +08:00
|
|
|
#include "llvm/Transforms/Utils/LowerInvoke.h"
|
2016-06-14 11:22:22 +08:00
|
|
|
#include "llvm/Transforms/Utils/Mem2Reg.h"
|
2016-09-17 00:56:30 +08:00
|
|
|
#include "llvm/Transforms/Utils/NameAnonGlobals.h"
|
Add PredicateInfo utility and printing pass
Summary:
This patch adds a utility to build extended SSA (see "ABCD: eliminating
array bounds checks on demand"), and an intrinsic to support it. This
is then used to get functionality equivalent to propagateEquality in
GVN, in NewGVN (without having to replace instructions as we go). It
would work similarly in SCCP or other passes. This has been talked
about a few times, so i built a real implementation and tried to
productionize it.
Copies are inserted for operands used in assumes and conditional
branches that are based on comparisons (see below for more)
Every use affected by the predicate is renamed to the appropriate
intrinsic result.
E.g.
%cmp = icmp eq i32 %x, 50
br i1 %cmp, label %true, label %false
true:
ret i32 %x
false:
ret i32 1
will become
%cmp = icmp eq i32, %x, 50
br i1 %cmp, label %true, label %false
true:
; Has predicate info
; branch predicate info { TrueEdge: 1 Comparison: %cmp = icmp eq i32 %x, 50 }
%x.0 = call @llvm.ssa_copy.i32(i32 %x)
ret i32 %x.0
false:
ret i23 1
(you can use -print-predicateinfo to get an annotated-with-predicateinfo dump)
This enables us to easily determine what operations are affected by a
given predicate, and how operations affected by a chain of
predicates.
Reviewers: davide, sanjoy
Subscribers: mgorny, llvm-commits, Prazek
Differential Revision: https://reviews.llvm.org/D29519
Update for review comments
Fix a bug Nuno noticed where we are giving information about and/or on edges where the info is not useful and easy to use wrong
Update for review comments
llvm-svn: 294351
2017-02-08 05:10:46 +08:00
|
|
|
#include "llvm/Transforms/Utils/PredicateInfo.h"
|
2016-07-08 05:14:36 +08:00
|
|
|
#include "llvm/Transforms/Utils/SimplifyInstructions.h"
|
2016-07-26 04:52:00 +08:00
|
|
|
#include "llvm/Transforms/Utils/SymbolRewriter.h"
|
2016-07-10 06:56:50 +08:00
|
|
|
#include "llvm/Transforms/Vectorize/LoopVectorize.h"
|
2016-07-09 11:11:29 +08:00
|
|
|
#include "llvm/Transforms/Vectorize/SLPVectorizer.h"
|
2016-06-14 11:22:22 +08:00
|
|
|
|
2016-02-26 20:17:54 +08:00
|
|
|
#include <type_traits>
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
|
|
|
|
using namespace llvm;
|
|
|
|
|
2017-02-12 13:38:04 +08:00
|
|
|
static cl::opt<unsigned> MaxDevirtIterations("pm-max-devirt-iterations",
|
|
|
|
cl::ReallyHidden, cl::init(4));
|
2017-05-23 00:41:57 +08:00
|
|
|
static cl::opt<bool>
|
|
|
|
RunPartialInlining("enable-npm-partial-inlining", cl::init(false),
|
|
|
|
cl::Hidden, cl::ZeroOrMore,
|
|
|
|
cl::desc("Run Partial inlinining pass"));
|
2017-02-12 13:38:04 +08:00
|
|
|
|
2017-05-23 07:41:40 +08:00
|
|
|
static cl::opt<bool>
|
2017-05-23 07:47:11 +08:00
|
|
|
RunNewGVN("enable-npm-newgvn", cl::init(false),
|
2017-05-23 07:41:40 +08:00
|
|
|
cl::Hidden, cl::ZeroOrMore,
|
|
|
|
cl::desc("Run NewGVN instead of GVN"));
|
|
|
|
|
2017-06-10 23:20:03 +08:00
|
|
|
static cl::opt<bool> EnableEarlyCSEMemSSA(
|
|
|
|
"enable-npm-earlycse-memssa", cl::init(false), cl::Hidden,
|
|
|
|
cl::desc("Enable the EarlyCSE w/ MemorySSA pass for the new PM (default = off)"));
|
|
|
|
|
2017-04-27 08:28:03 +08:00
|
|
|
static cl::opt<bool> EnableGVNHoist(
|
|
|
|
"enable-npm-gvn-hoist", cl::init(false), cl::Hidden,
|
|
|
|
cl::desc("Enable the GVN hoisting pass for the new PM (default = off)"));
|
|
|
|
|
2017-06-04 07:18:29 +08:00
|
|
|
static cl::opt<bool> EnableGVNSink(
|
|
|
|
"enable-npm-gvn-sink", cl::init(false), cl::Hidden,
|
|
|
|
cl::desc("Enable the GVN hoisting pass for the new PM (default = off)"));
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
static Regex DefaultAliasRegex(
|
|
|
|
"^(default|thinlto-pre-link|thinlto|lto-pre-link|lto)<(O[0123sz])>$");
|
2016-02-29 06:16:03 +08:00
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
static bool isOptimizingForSize(PassBuilder::OptimizationLevel Level) {
|
|
|
|
switch (Level) {
|
|
|
|
case PassBuilder::O0:
|
|
|
|
case PassBuilder::O1:
|
|
|
|
case PassBuilder::O2:
|
|
|
|
case PassBuilder::O3:
|
|
|
|
return false;
|
|
|
|
|
|
|
|
case PassBuilder::Os:
|
|
|
|
case PassBuilder::Oz:
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
llvm_unreachable("Invalid optimization level!");
|
|
|
|
}
|
|
|
|
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
namespace {
|
|
|
|
|
2014-01-12 17:34:22 +08:00
|
|
|
/// \brief No-op module pass which does nothing.
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
struct NoOpModulePass {
|
2016-08-09 08:28:38 +08:00
|
|
|
PreservedAnalyses run(Module &M, ModuleAnalysisManager &) {
|
2016-06-17 08:11:01 +08:00
|
|
|
return PreservedAnalyses::all();
|
|
|
|
}
|
2014-01-11 19:52:05 +08:00
|
|
|
static StringRef name() { return "NoOpModulePass"; }
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
};
|
|
|
|
|
2015-01-06 10:50:06 +08:00
|
|
|
/// \brief No-op module analysis.
|
2016-03-11 18:33:22 +08:00
|
|
|
class NoOpModuleAnalysis : public AnalysisInfoMixin<NoOpModuleAnalysis> {
|
|
|
|
friend AnalysisInfoMixin<NoOpModuleAnalysis>;
|
2016-11-24 01:53:26 +08:00
|
|
|
static AnalysisKey Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
|
|
|
public:
|
2015-01-06 10:50:06 +08:00
|
|
|
struct Result {};
|
2016-08-09 08:28:38 +08:00
|
|
|
Result run(Module &, ModuleAnalysisManager &) { return Result(); }
|
2015-01-06 10:50:06 +08:00
|
|
|
static StringRef name() { return "NoOpModuleAnalysis"; }
|
|
|
|
};
|
|
|
|
|
2014-04-21 19:12:00 +08:00
|
|
|
/// \brief No-op CGSCC pass which does nothing.
|
|
|
|
struct NoOpCGSCCPass {
|
[PM] Introduce basic update capabilities to the new PM's CGSCC pass
manager, including both plumbing and logic to handle function pass
updates.
There are three fundamentally tied changes here:
1) Plumbing *some* mechanism for updating the CGSCC pass manager as the
CG changes while passes are running.
2) Changing the CGSCC pass manager infrastructure to have support for
the underlying graph to mutate mid-pass run.
3) Actually updating the CG after function passes run.
I can separate them if necessary, but I think its really useful to have
them together as the needs of #3 drove #2, and that in turn drove #1.
The plumbing technique is to extend the "run" method signature with
extra arguments. We provide the call graph that intrinsically is
available as it is the basis of the pass manager's IR units, and an
output parameter that records the results of updating the call graph
during an SCC passes's run. Note that "...UpdateResult" isn't a *great*
name here... suggestions very welcome.
I tried a pretty frustrating number of different data structures and such
for the innards of the update result. Every other one failed for one
reason or another. Sometimes I just couldn't keep the layers of
complexity right in my head. The thing that really worked was to just
directly provide access to the underlying structures used to walk the
call graph so that their updates could be informed by the *particular*
nature of the change to the graph.
The technique for how to make the pass management infrastructure cope
with mutating graphs was also something that took a really, really large
number of iterations to get to a place where I was happy. Here are some
of the considerations that drove the design:
- We operate at three levels within the infrastructure: RefSCC, SCC, and
Node. In each case, we are working bottom up and so we want to
continue to iterate on the "lowest" node as the graph changes. Look at
how we iterate over nodes in an SCC running function passes as those
function passes mutate the CG. We continue to iterate on the "lowest"
SCC, which is the one that continues to contain the function just
processed.
- The call graph structure re-uses SCCs (and RefSCCs) during mutation
events for the *highest* entry in the resulting new subgraph, not the
lowest. This means that it is necessary to continually update the
current SCC or RefSCC as it shifts. This is really surprising and
subtle, and took a long time for me to work out. I actually tried
changing the call graph to provide the opposite behavior, and it
breaks *EVERYTHING*. The graph update algorithms are really deeply
tied to this particualr pattern.
- When SCCs or RefSCCs are split apart and refined and we continually
re-pin our processing to the bottom one in the subgraph, we need to
enqueue the newly formed SCCs and RefSCCs for subsequent processing.
Queuing them presents a few challenges:
1) SCCs and RefSCCs use wildly different iteration strategies at
a high level. We end up needing to converge them on worklist
approaches that can be extended in order to be able to handle the
mutations.
2) The order of the enqueuing need to remain bottom-up post-order so
that we don't get surprising order of visitation for things like
the inliner.
3) We need the worklists to have set semantics so we don't duplicate
things endlessly. We don't need a *persistent* set though because
we always keep processing the bottom node!!!! This is super, super
surprising to me and took a long time to convince myself this is
correct, but I'm pretty sure it is... Once we sink down to the
bottom node, we can't re-split out the same node in any way, and
the postorder of the current queue is fixed and unchanging.
4) We need to make sure that the "current" SCC or RefSCC actually gets
enqueued here such that we re-visit it because we continue
processing a *new*, *bottom* SCC/RefSCC.
- We also need the ability to *skip* SCCs and RefSCCs that get merged
into a larger component. We even need the ability to skip *nodes* from
an SCC that are no longer part of that SCC.
This led to the design you see in the patch which uses SetVector-based
worklists. The RefSCC worklist is always empty until an update occurs
and is just used to handle those RefSCCs created by updates as the
others don't even exist yet and are formed on-demand during the
bottom-up walk. The SCC worklist is pre-populated from the RefSCC, and
we push new SCCs onto it and blacklist existing SCCs on it to get the
desired processing.
We then *directly* update these when updating the call graph as I was
never able to find a satisfactory abstraction around the update
strategy.
Finally, we need to compute the updates for function passes. This is
mostly used as an initial customer of all the update mechanisms to drive
their design to at least cover some real set of use cases. There are
a bunch of interesting things that came out of doing this:
- It is really nice to do this a function at a time because that
function is likely hot in the cache. This means we want even the
function pass adaptor to support online updates to the call graph!
- To update the call graph after arbitrary function pass mutations is
quite hard. We have to build a fairly comprehensive set of
data structures and then process them. Fortunately, some of this code
is related to the code for building the cal graph in the first place.
Unfortunately, very little of it makes any sense to share because the
nature of what we're doing is so very different. I've factored out the
one part that made sense at least.
- We need to transfer these updates into the various structures for the
CGSCC pass manager. Once those were more sanely worked out, this
became relatively easier. But some of those needs necessitated changes
to the LazyCallGraph interface to make it significantly easier to
extract the changed SCCs from an update operation.
- We also need to update the CGSCC analysis manager as the shape of the
graph changes. When an SCC is merged away we need to clear analyses
associated with it from the analysis manager which we didn't have
support for in the analysis manager infrsatructure. New SCCs are easy!
But then we have the case that the original SCC has its shape changed
but remains in the call graph. There we need to *invalidate* the
analyses associated with it.
- We also need to invalidate analyses after we *finish* processing an
SCC. But the analyses we need to invalidate here are *only those for
the newly updated SCC*!!! Because we only continue processing the
bottom SCC, if we split SCCs apart the original one gets invalidated
once when its shape changes and is not processed farther so its
analyses will be correct. It is the bottom SCC which continues being
processed and needs to have the "normal" invalidation done based on
the preserved analyses set.
All of this is mostly background and context for the changes here.
Many thanks to all the reviewers who helped here. Especially Sanjoy who
caught several interesting bugs in the graph algorithms, David, Sean,
and others who all helped with feedback.
Differential Revision: http://reviews.llvm.org/D21464
llvm-svn: 279618
2016-08-24 17:37:14 +08:00
|
|
|
PreservedAnalyses run(LazyCallGraph::SCC &C, CGSCCAnalysisManager &,
|
|
|
|
LazyCallGraph &, CGSCCUpdateResult &UR) {
|
2014-04-21 19:12:00 +08:00
|
|
|
return PreservedAnalyses::all();
|
|
|
|
}
|
|
|
|
static StringRef name() { return "NoOpCGSCCPass"; }
|
|
|
|
};
|
|
|
|
|
2015-01-06 10:50:06 +08:00
|
|
|
/// \brief No-op CGSCC analysis.
|
2016-03-11 18:33:22 +08:00
|
|
|
class NoOpCGSCCAnalysis : public AnalysisInfoMixin<NoOpCGSCCAnalysis> {
|
|
|
|
friend AnalysisInfoMixin<NoOpCGSCCAnalysis>;
|
2016-11-24 01:53:26 +08:00
|
|
|
static AnalysisKey Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
|
|
|
public:
|
2015-01-06 10:50:06 +08:00
|
|
|
struct Result {};
|
[PM] Introduce basic update capabilities to the new PM's CGSCC pass
manager, including both plumbing and logic to handle function pass
updates.
There are three fundamentally tied changes here:
1) Plumbing *some* mechanism for updating the CGSCC pass manager as the
CG changes while passes are running.
2) Changing the CGSCC pass manager infrastructure to have support for
the underlying graph to mutate mid-pass run.
3) Actually updating the CG after function passes run.
I can separate them if necessary, but I think its really useful to have
them together as the needs of #3 drove #2, and that in turn drove #1.
The plumbing technique is to extend the "run" method signature with
extra arguments. We provide the call graph that intrinsically is
available as it is the basis of the pass manager's IR units, and an
output parameter that records the results of updating the call graph
during an SCC passes's run. Note that "...UpdateResult" isn't a *great*
name here... suggestions very welcome.
I tried a pretty frustrating number of different data structures and such
for the innards of the update result. Every other one failed for one
reason or another. Sometimes I just couldn't keep the layers of
complexity right in my head. The thing that really worked was to just
directly provide access to the underlying structures used to walk the
call graph so that their updates could be informed by the *particular*
nature of the change to the graph.
The technique for how to make the pass management infrastructure cope
with mutating graphs was also something that took a really, really large
number of iterations to get to a place where I was happy. Here are some
of the considerations that drove the design:
- We operate at three levels within the infrastructure: RefSCC, SCC, and
Node. In each case, we are working bottom up and so we want to
continue to iterate on the "lowest" node as the graph changes. Look at
how we iterate over nodes in an SCC running function passes as those
function passes mutate the CG. We continue to iterate on the "lowest"
SCC, which is the one that continues to contain the function just
processed.
- The call graph structure re-uses SCCs (and RefSCCs) during mutation
events for the *highest* entry in the resulting new subgraph, not the
lowest. This means that it is necessary to continually update the
current SCC or RefSCC as it shifts. This is really surprising and
subtle, and took a long time for me to work out. I actually tried
changing the call graph to provide the opposite behavior, and it
breaks *EVERYTHING*. The graph update algorithms are really deeply
tied to this particualr pattern.
- When SCCs or RefSCCs are split apart and refined and we continually
re-pin our processing to the bottom one in the subgraph, we need to
enqueue the newly formed SCCs and RefSCCs for subsequent processing.
Queuing them presents a few challenges:
1) SCCs and RefSCCs use wildly different iteration strategies at
a high level. We end up needing to converge them on worklist
approaches that can be extended in order to be able to handle the
mutations.
2) The order of the enqueuing need to remain bottom-up post-order so
that we don't get surprising order of visitation for things like
the inliner.
3) We need the worklists to have set semantics so we don't duplicate
things endlessly. We don't need a *persistent* set though because
we always keep processing the bottom node!!!! This is super, super
surprising to me and took a long time to convince myself this is
correct, but I'm pretty sure it is... Once we sink down to the
bottom node, we can't re-split out the same node in any way, and
the postorder of the current queue is fixed and unchanging.
4) We need to make sure that the "current" SCC or RefSCC actually gets
enqueued here such that we re-visit it because we continue
processing a *new*, *bottom* SCC/RefSCC.
- We also need the ability to *skip* SCCs and RefSCCs that get merged
into a larger component. We even need the ability to skip *nodes* from
an SCC that are no longer part of that SCC.
This led to the design you see in the patch which uses SetVector-based
worklists. The RefSCC worklist is always empty until an update occurs
and is just used to handle those RefSCCs created by updates as the
others don't even exist yet and are formed on-demand during the
bottom-up walk. The SCC worklist is pre-populated from the RefSCC, and
we push new SCCs onto it and blacklist existing SCCs on it to get the
desired processing.
We then *directly* update these when updating the call graph as I was
never able to find a satisfactory abstraction around the update
strategy.
Finally, we need to compute the updates for function passes. This is
mostly used as an initial customer of all the update mechanisms to drive
their design to at least cover some real set of use cases. There are
a bunch of interesting things that came out of doing this:
- It is really nice to do this a function at a time because that
function is likely hot in the cache. This means we want even the
function pass adaptor to support online updates to the call graph!
- To update the call graph after arbitrary function pass mutations is
quite hard. We have to build a fairly comprehensive set of
data structures and then process them. Fortunately, some of this code
is related to the code for building the cal graph in the first place.
Unfortunately, very little of it makes any sense to share because the
nature of what we're doing is so very different. I've factored out the
one part that made sense at least.
- We need to transfer these updates into the various structures for the
CGSCC pass manager. Once those were more sanely worked out, this
became relatively easier. But some of those needs necessitated changes
to the LazyCallGraph interface to make it significantly easier to
extract the changed SCCs from an update operation.
- We also need to update the CGSCC analysis manager as the shape of the
graph changes. When an SCC is merged away we need to clear analyses
associated with it from the analysis manager which we didn't have
support for in the analysis manager infrsatructure. New SCCs are easy!
But then we have the case that the original SCC has its shape changed
but remains in the call graph. There we need to *invalidate* the
analyses associated with it.
- We also need to invalidate analyses after we *finish* processing an
SCC. But the analyses we need to invalidate here are *only those for
the newly updated SCC*!!! Because we only continue processing the
bottom SCC, if we split SCCs apart the original one gets invalidated
once when its shape changes and is not processed farther so its
analyses will be correct. It is the bottom SCC which continues being
processed and needs to have the "normal" invalidation done based on
the preserved analyses set.
All of this is mostly background and context for the changes here.
Many thanks to all the reviewers who helped here. Especially Sanjoy who
caught several interesting bugs in the graph algorithms, David, Sean,
and others who all helped with feedback.
Differential Revision: http://reviews.llvm.org/D21464
llvm-svn: 279618
2016-08-24 17:37:14 +08:00
|
|
|
Result run(LazyCallGraph::SCC &, CGSCCAnalysisManager &, LazyCallGraph &G) {
|
2016-06-17 08:11:01 +08:00
|
|
|
return Result();
|
|
|
|
}
|
2015-01-06 10:50:06 +08:00
|
|
|
static StringRef name() { return "NoOpCGSCCAnalysis"; }
|
|
|
|
};
|
|
|
|
|
2014-01-12 17:34:22 +08:00
|
|
|
/// \brief No-op function pass which does nothing.
|
|
|
|
struct NoOpFunctionPass {
|
2016-08-09 08:28:15 +08:00
|
|
|
PreservedAnalyses run(Function &F, FunctionAnalysisManager &) {
|
2016-06-17 08:11:01 +08:00
|
|
|
return PreservedAnalyses::all();
|
|
|
|
}
|
2014-01-12 17:34:22 +08:00
|
|
|
static StringRef name() { return "NoOpFunctionPass"; }
|
|
|
|
};
|
|
|
|
|
2015-01-06 10:50:06 +08:00
|
|
|
/// \brief No-op function analysis.
|
2016-03-11 18:33:22 +08:00
|
|
|
class NoOpFunctionAnalysis : public AnalysisInfoMixin<NoOpFunctionAnalysis> {
|
|
|
|
friend AnalysisInfoMixin<NoOpFunctionAnalysis>;
|
2016-11-24 01:53:26 +08:00
|
|
|
static AnalysisKey Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
|
|
|
public:
|
2015-01-06 10:50:06 +08:00
|
|
|
struct Result {};
|
2016-08-09 08:28:15 +08:00
|
|
|
Result run(Function &, FunctionAnalysisManager &) { return Result(); }
|
2015-01-06 10:50:06 +08:00
|
|
|
static StringRef name() { return "NoOpFunctionAnalysis"; }
|
|
|
|
};
|
|
|
|
|
2016-02-25 15:23:08 +08:00
|
|
|
/// \brief No-op loop pass which does nothing.
|
|
|
|
struct NoOpLoopPass {
|
2017-01-11 14:23:21 +08:00
|
|
|
PreservedAnalyses run(Loop &L, LoopAnalysisManager &,
|
|
|
|
LoopStandardAnalysisResults &, LPMUpdater &) {
|
2016-06-17 08:11:01 +08:00
|
|
|
return PreservedAnalyses::all();
|
|
|
|
}
|
2016-02-25 15:23:08 +08:00
|
|
|
static StringRef name() { return "NoOpLoopPass"; }
|
|
|
|
};
|
|
|
|
|
|
|
|
/// \brief No-op loop analysis.
|
2016-03-11 18:33:22 +08:00
|
|
|
class NoOpLoopAnalysis : public AnalysisInfoMixin<NoOpLoopAnalysis> {
|
|
|
|
friend AnalysisInfoMixin<NoOpLoopAnalysis>;
|
2016-11-24 01:53:26 +08:00
|
|
|
static AnalysisKey Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
|
|
|
public:
|
2016-02-25 15:23:08 +08:00
|
|
|
struct Result {};
|
2017-01-11 14:23:21 +08:00
|
|
|
Result run(Loop &, LoopAnalysisManager &, LoopStandardAnalysisResults &) {
|
|
|
|
return Result();
|
|
|
|
}
|
2016-02-25 15:23:08 +08:00
|
|
|
static StringRef name() { return "NoOpLoopAnalysis"; }
|
|
|
|
};
|
|
|
|
|
2016-11-24 01:53:26 +08:00
|
|
|
AnalysisKey NoOpModuleAnalysis::Key;
|
|
|
|
AnalysisKey NoOpCGSCCAnalysis::Key;
|
|
|
|
AnalysisKey NoOpFunctionAnalysis::Key;
|
|
|
|
AnalysisKey NoOpLoopAnalysis::Key;
|
2016-03-11 18:22:49 +08:00
|
|
|
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
} // End anonymous namespace.
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
void PassBuilder::registerModuleAnalyses(ModuleAnalysisManager &MAM) {
|
2016-06-17 15:15:29 +08:00
|
|
|
#define MODULE_ANALYSIS(NAME, CREATE_PASS) \
|
2016-02-18 17:45:17 +08:00
|
|
|
MAM.registerPass([&] { return CREATE_PASS; });
|
2015-01-06 10:21:37 +08:00
|
|
|
#include "PassRegistry.def"
|
|
|
|
}
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
void PassBuilder::registerCGSCCAnalyses(CGSCCAnalysisManager &CGAM) {
|
2016-06-17 15:15:29 +08:00
|
|
|
#define CGSCC_ANALYSIS(NAME, CREATE_PASS) \
|
2016-02-18 17:45:17 +08:00
|
|
|
CGAM.registerPass([&] { return CREATE_PASS; });
|
2015-01-06 10:21:37 +08:00
|
|
|
#include "PassRegistry.def"
|
|
|
|
}
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
void PassBuilder::registerFunctionAnalyses(FunctionAnalysisManager &FAM) {
|
2016-06-17 15:15:29 +08:00
|
|
|
#define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \
|
2016-02-18 17:45:17 +08:00
|
|
|
FAM.registerPass([&] { return CREATE_PASS; });
|
2015-01-06 10:21:37 +08:00
|
|
|
#include "PassRegistry.def"
|
|
|
|
}
|
|
|
|
|
2016-02-25 15:23:08 +08:00
|
|
|
void PassBuilder::registerLoopAnalyses(LoopAnalysisManager &LAM) {
|
2016-06-17 15:15:29 +08:00
|
|
|
#define LOOP_ANALYSIS(NAME, CREATE_PASS) \
|
2016-02-25 15:23:08 +08:00
|
|
|
LAM.registerPass([&] { return CREATE_PASS; });
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
}
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
FunctionPassManager
|
|
|
|
PassBuilder::buildFunctionSimplificationPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
assert(Level != O0 && "Must request optimizations!");
|
|
|
|
FunctionPassManager FPM(DebugLogging);
|
|
|
|
|
|
|
|
// Form SSA out of local memory accesses after breaking apart aggregates into
|
|
|
|
// scalars.
|
|
|
|
FPM.addPass(SROA());
|
|
|
|
|
|
|
|
// Catch trivial redundancies
|
2017-06-10 23:20:03 +08:00
|
|
|
FPM.addPass(EarlyCSEPass(EnableEarlyCSEMemSSA));
|
2016-12-22 14:59:15 +08:00
|
|
|
|
2017-06-02 07:08:14 +08:00
|
|
|
// Hoisting of scalars and load expressions.
|
|
|
|
if (EnableGVNHoist)
|
|
|
|
FPM.addPass(GVNHoistPass());
|
|
|
|
|
2017-06-04 07:18:29 +08:00
|
|
|
// Global value numbering based sinking.
|
|
|
|
if (EnableGVNSink) {
|
|
|
|
FPM.addPass(GVNSinkPass());
|
|
|
|
FPM.addPass(SimplifyCFGPass());
|
|
|
|
}
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
// Speculative execution if the target has divergent branches; otherwise nop.
|
|
|
|
FPM.addPass(SpeculativeExecutionPass());
|
|
|
|
|
|
|
|
// Optimize based on known information about branches, and cleanup afterward.
|
|
|
|
FPM.addPass(JumpThreadingPass());
|
|
|
|
FPM.addPass(CorrelatedValuePropagationPass());
|
|
|
|
FPM.addPass(SimplifyCFGPass());
|
|
|
|
FPM.addPass(InstCombinePass());
|
|
|
|
|
|
|
|
if (!isOptimizingForSize(Level))
|
|
|
|
FPM.addPass(LibCallsShrinkWrapPass());
|
|
|
|
|
|
|
|
FPM.addPass(TailCallElimPass());
|
|
|
|
FPM.addPass(SimplifyCFGPass());
|
|
|
|
|
|
|
|
// Form canonically associated expression trees, and simplify the trees using
|
|
|
|
// basic mathematical properties. For example, this will form (nearly)
|
|
|
|
// minimal multiplication trees.
|
|
|
|
FPM.addPass(ReassociatePass());
|
|
|
|
|
|
|
|
// Add the primary loop simplification pipeline.
|
|
|
|
// FIXME: Currently this is split into two loop pass pipelines because we run
|
|
|
|
// some function passes in between them. These can and should be replaced by
|
|
|
|
// loop pass equivalenst but those aren't ready yet. Specifically,
|
|
|
|
// `SimplifyCFGPass` and `InstCombinePass` are used. We have
|
|
|
|
// `LoopSimplifyCFGPass` which isn't yet powerful enough, and the closest to
|
|
|
|
// the other we have is `LoopInstSimplify`.
|
|
|
|
LoopPassManager LPM1(DebugLogging), LPM2(DebugLogging);
|
|
|
|
|
|
|
|
// Rotate Loop - disable header duplication at -Oz
|
|
|
|
LPM1.addPass(LoopRotatePass(Level != Oz));
|
|
|
|
LPM1.addPass(LICMPass());
|
2017-05-26 09:24:11 +08:00
|
|
|
LPM1.addPass(SimpleLoopUnswitchPass());
|
2016-12-22 14:59:15 +08:00
|
|
|
LPM2.addPass(IndVarSimplifyPass());
|
2017-01-27 07:21:17 +08:00
|
|
|
LPM2.addPass(LoopIdiomRecognizePass());
|
2016-12-22 14:59:15 +08:00
|
|
|
LPM2.addPass(LoopDeletionPass());
|
2017-06-01 19:39:39 +08:00
|
|
|
// FIXME: The old pass manager has a hack to disable loop unrolling during
|
|
|
|
// ThinLTO when using sample PGO. Need to either fix it or port some
|
|
|
|
// workaround.
|
Increases full-unroll threshold.
Summary:
The default threshold for fully unroll is too conservative. This patch doubles the full-unroll threshold
This change will affect the following speccpu2006 benchmarks (performance numbers were collected from Intel Sandybridge):
Performance:
403 0.11%
433 0.51%
445 0.48%
447 3.50%
453 1.49%
464 0.75%
Code size:
403 0.56%
433 0.96%
445 2.16%
447 2.96%
453 0.94%
464 8.02%
The compiler time overhead is similar with code size.
Reviewers: davidxl, mkuper, mzolotukhin, hfinkel, chandlerc
Reviewed By: hfinkel, chandlerc
Subscribers: mehdi_amini, zzheng, efriedma, haicheng, hfinkel, llvm-commits
Differential Revision: https://reviews.llvm.org/D28368
llvm-svn: 295538
2017-02-18 11:46:51 +08:00
|
|
|
LPM2.addPass(LoopUnrollPass::createFull(Level));
|
2017-01-27 07:21:17 +08:00
|
|
|
|
2017-02-10 07:54:57 +08:00
|
|
|
// We provide the opt remark emitter pass for LICM to use. We only need to do
|
|
|
|
// this once as it is immutable.
|
|
|
|
FPM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>());
|
2016-12-22 14:59:15 +08:00
|
|
|
FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM1)));
|
|
|
|
FPM.addPass(SimplifyCFGPass());
|
|
|
|
FPM.addPass(InstCombinePass());
|
|
|
|
FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM2)));
|
|
|
|
|
|
|
|
// Eliminate redundancies.
|
|
|
|
if (Level != O1) {
|
|
|
|
// These passes add substantial compile time so skip them at O1.
|
|
|
|
FPM.addPass(MergedLoadStoreMotionPass());
|
2017-05-23 07:41:40 +08:00
|
|
|
if (RunNewGVN)
|
|
|
|
FPM.addPass(NewGVNPass());
|
|
|
|
else
|
|
|
|
FPM.addPass(GVN());
|
2016-12-22 14:59:15 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Specially optimize memory movement as it doesn't look like dataflow in SSA.
|
|
|
|
FPM.addPass(MemCpyOptPass());
|
|
|
|
|
|
|
|
// Sparse conditional constant propagation.
|
|
|
|
// FIXME: It isn't clear why we do this *after* loop passes rather than
|
|
|
|
// before...
|
|
|
|
FPM.addPass(SCCPPass());
|
|
|
|
|
|
|
|
// Delete dead bit computations (instcombine runs after to fold away the dead
|
|
|
|
// computations, and then ADCE will run later to exploit any new DCE
|
|
|
|
// opportunities that creates).
|
|
|
|
FPM.addPass(BDCEPass());
|
|
|
|
|
|
|
|
// Run instcombine after redundancy and dead bit elimination to exploit
|
|
|
|
// opportunities opened up by them.
|
|
|
|
FPM.addPass(InstCombinePass());
|
|
|
|
|
|
|
|
// Re-consider control flow based optimizations after redundancy elimination,
|
|
|
|
// redo DCE, etc.
|
|
|
|
FPM.addPass(JumpThreadingPass());
|
|
|
|
FPM.addPass(CorrelatedValuePropagationPass());
|
|
|
|
FPM.addPass(DSEPass());
|
|
|
|
FPM.addPass(createFunctionToLoopPassAdaptor(LICMPass()));
|
|
|
|
|
|
|
|
// Finally, do an expensive DCE pass to catch all the dead code exposed by
|
|
|
|
// the simplifications and basic cleanup after all the simplifications.
|
|
|
|
FPM.addPass(ADCEPass());
|
|
|
|
FPM.addPass(SimplifyCFGPass());
|
|
|
|
FPM.addPass(InstCombinePass());
|
|
|
|
|
|
|
|
return FPM;
|
|
|
|
}
|
|
|
|
|
2017-02-13 23:26:22 +08:00
|
|
|
static void addPGOInstrPasses(ModulePassManager &MPM, bool DebugLogging,
|
|
|
|
PassBuilder::OptimizationLevel Level,
|
|
|
|
bool RunProfileGen, std::string ProfileGenFile,
|
|
|
|
std::string ProfileUseFile) {
|
|
|
|
// Generally running simplification passes and the inliner with an high
|
|
|
|
// threshold results in smaller executables, but there may be cases where
|
|
|
|
// the size grows, so let's be conservative here and skip this simplification
|
|
|
|
// at -Os/Oz.
|
|
|
|
if (!isOptimizingForSize(Level)) {
|
|
|
|
InlineParams IP;
|
|
|
|
|
|
|
|
// In the old pass manager, this is a cl::opt. Should still this be one?
|
|
|
|
IP.DefaultThreshold = 75;
|
|
|
|
|
|
|
|
// FIXME: The hint threshold has the same value used by the regular inliner.
|
|
|
|
// This should probably be lowered after performance testing.
|
|
|
|
// FIXME: this comment is cargo culted from the old pass manager, revisit).
|
|
|
|
IP.HintThreshold = 325;
|
|
|
|
|
|
|
|
CGSCCPassManager CGPipeline(DebugLogging);
|
|
|
|
|
|
|
|
CGPipeline.addPass(InlinerPass(IP));
|
|
|
|
|
|
|
|
FunctionPassManager FPM;
|
|
|
|
FPM.addPass(SROA());
|
|
|
|
FPM.addPass(EarlyCSEPass()); // Catch trivial redundancies.
|
|
|
|
FPM.addPass(SimplifyCFGPass()); // Merge & remove basic blocks.
|
|
|
|
FPM.addPass(InstCombinePass()); // Combine silly sequences.
|
|
|
|
|
|
|
|
// FIXME: Here the old pass manager inserts peephole extensions.
|
|
|
|
// Add them when they're supported.
|
|
|
|
CGPipeline.addPass(createCGSCCToFunctionPassAdaptor(std::move(FPM)));
|
|
|
|
|
|
|
|
MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPipeline)));
|
|
|
|
}
|
|
|
|
|
2017-05-25 15:15:09 +08:00
|
|
|
// Delete anything that is now dead to make sure that we don't instrument
|
|
|
|
// dead code. Instrumentation can end up keeping dead code around and
|
|
|
|
// dramatically increase code size.
|
|
|
|
MPM.addPass(GlobalDCEPass());
|
|
|
|
|
2017-02-13 23:26:22 +08:00
|
|
|
if (RunProfileGen) {
|
|
|
|
MPM.addPass(PGOInstrumentationGen());
|
|
|
|
|
2017-06-25 08:26:43 +08:00
|
|
|
FunctionPassManager FPM;
|
|
|
|
FPM.addPass(createFunctionToLoopPassAdaptor(LoopRotatePass()));
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
|
|
|
|
|
2017-02-13 23:26:22 +08:00
|
|
|
// Add the profile lowering pass.
|
|
|
|
InstrProfOptions Options;
|
|
|
|
if (!ProfileGenFile.empty())
|
|
|
|
Options.InstrProfileOutput = ProfileGenFile;
|
2017-06-25 08:26:43 +08:00
|
|
|
Options.DoCounterPromotion = true;
|
2017-02-13 23:26:22 +08:00
|
|
|
MPM.addPass(InstrProfiling(Options));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!ProfileUseFile.empty())
|
|
|
|
MPM.addPass(PGOInstrumentationUse(ProfileUseFile));
|
|
|
|
}
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
ModulePassManager
|
2017-06-01 19:39:39 +08:00
|
|
|
PassBuilder::buildModuleSimplificationPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
2016-12-22 14:59:15 +08:00
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
|
|
|
// Do basic inference of function attributes from known properties of system
|
|
|
|
// libraries and other oracles.
|
|
|
|
MPM.addPass(InferFunctionAttrsPass());
|
|
|
|
|
|
|
|
// Create an early function pass manager to cleanup the output of the
|
|
|
|
// frontend.
|
2016-02-29 06:16:03 +08:00
|
|
|
FunctionPassManager EarlyFPM(DebugLogging);
|
|
|
|
EarlyFPM.addPass(SimplifyCFGPass());
|
|
|
|
EarlyFPM.addPass(SROA());
|
|
|
|
EarlyFPM.addPass(EarlyCSEPass());
|
|
|
|
EarlyFPM.addPass(LowerExpectIntrinsicPass());
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(EarlyFPM)));
|
2016-12-22 14:59:15 +08:00
|
|
|
|
|
|
|
// Interprocedural constant propagation now that basic cleanup has occured
|
|
|
|
// and prior to optimizing globals.
|
|
|
|
// FIXME: This position in the pipeline hasn't been carefully considered in
|
|
|
|
// years, it should be re-analyzed.
|
|
|
|
MPM.addPass(IPSCCPPass());
|
|
|
|
|
|
|
|
// Optimize globals to try and fold them into constants.
|
|
|
|
MPM.addPass(GlobalOptPass());
|
|
|
|
|
|
|
|
// Promote any localized globals to SSA registers.
|
|
|
|
// FIXME: Should this instead by a run of SROA?
|
|
|
|
// FIXME: We should probably run instcombine and simplify-cfg afterward to
|
|
|
|
// delete control flows that are dead once globals have been folded to
|
|
|
|
// constants.
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass()));
|
|
|
|
|
|
|
|
// Remove any dead arguments exposed by cleanups and constand folding
|
|
|
|
// globals.
|
|
|
|
MPM.addPass(DeadArgumentEliminationPass());
|
|
|
|
|
|
|
|
// Create a small function pass pipeline to cleanup after all the global
|
|
|
|
// optimizations.
|
|
|
|
FunctionPassManager GlobalCleanupPM(DebugLogging);
|
|
|
|
GlobalCleanupPM.addPass(InstCombinePass());
|
|
|
|
GlobalCleanupPM.addPass(SimplifyCFGPass());
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(GlobalCleanupPM)));
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
// Add all the requested passes for PGO, if requested.
|
2017-02-13 23:26:22 +08:00
|
|
|
if (PGOOpt) {
|
2017-02-24 06:15:18 +08:00
|
|
|
assert(PGOOpt->RunProfileGen || PGOOpt->SamplePGO ||
|
|
|
|
!PGOOpt->ProfileUseFile.empty());
|
2017-02-13 23:26:22 +08:00
|
|
|
addPGOInstrPasses(MPM, DebugLogging, Level, PGOOpt->RunProfileGen,
|
|
|
|
PGOOpt->ProfileGenFile, PGOOpt->ProfileUseFile);
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
// Indirect call promotion that promotes intra-module targes only.
|
|
|
|
MPM.addPass(PGOIndirectCallPromotion(false, PGOOpt && PGOOpt->SamplePGO));
|
|
|
|
}
|
2017-02-13 23:26:22 +08:00
|
|
|
|
2017-02-12 13:34:04 +08:00
|
|
|
// Require the GlobalsAA analysis for the module so we can query it within
|
|
|
|
// the CGSCC pipeline.
|
|
|
|
MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>());
|
2016-12-22 14:59:15 +08:00
|
|
|
|
2017-05-05 00:58:45 +08:00
|
|
|
// Require the ProfileSummaryAnalysis for the module so we can query it within
|
|
|
|
// the inliner pass.
|
|
|
|
MPM.addPass(RequireAnalysisPass<ProfileSummaryAnalysis, Module>());
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
// Now begin the main postorder CGSCC pipeline.
|
|
|
|
// FIXME: The current CGSCC pipeline has its origins in the legacy pass
|
|
|
|
// manager and trying to emulate its precise behavior. Much of this doesn't
|
|
|
|
// make a lot of sense and we should revisit the core CGSCC structure.
|
|
|
|
CGSCCPassManager MainCGPipeline(DebugLogging);
|
|
|
|
|
|
|
|
// Note: historically, the PruneEH pass was run first to deduce nounwind and
|
|
|
|
// generally clean up exception handling overhead. It isn't clear this is
|
|
|
|
// valuable as the inliner doesn't currently care whether it is inlining an
|
|
|
|
// invoke or a call.
|
|
|
|
|
|
|
|
// Run the inliner first. The theory is that we are walking bottom-up and so
|
|
|
|
// the callees have already been fully optimized, and we want to inline them
|
|
|
|
// into the callers so that our optimizations can reflect that.
|
|
|
|
// FIXME; Customize the threshold based on optimization level.
|
|
|
|
MainCGPipeline.addPass(InlinerPass());
|
|
|
|
|
|
|
|
// Now deduce any function attributes based in the current code.
|
|
|
|
MainCGPipeline.addPass(PostOrderFunctionAttrsPass());
|
|
|
|
|
2017-02-10 07:54:57 +08:00
|
|
|
// When at O3 add argument promotion to the pass pipeline.
|
|
|
|
// FIXME: It isn't at all clear why this should be limited to O3.
|
|
|
|
if (Level == O3)
|
|
|
|
MainCGPipeline.addPass(ArgumentPromotionPass());
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
// Lastly, add the core function simplification pipeline nested inside the
|
|
|
|
// CGSCC walk.
|
|
|
|
MainCGPipeline.addPass(createCGSCCToFunctionPassAdaptor(
|
|
|
|
buildFunctionSimplificationPipeline(Level, DebugLogging)));
|
|
|
|
|
2017-02-12 13:38:04 +08:00
|
|
|
// We wrap the CGSCC pipeline in a devirtualization repeater. This will try
|
|
|
|
// to detect when we devirtualize indirect calls and iterate the SCC passes
|
|
|
|
// in that case to try and catch knock-on inlining or function attrs
|
|
|
|
// opportunities. Then we add it to the module pipeline by walking the SCCs
|
|
|
|
// in postorder (or bottom-up).
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(
|
2017-02-12 13:38:04 +08:00
|
|
|
createModuleToPostOrderCGSCCPassAdaptor(createDevirtSCCRepeatedPass(
|
|
|
|
std::move(MainCGPipeline), MaxDevirtIterations, DebugLogging)));
|
2016-12-22 14:59:15 +08:00
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
return MPM;
|
|
|
|
}
|
|
|
|
|
|
|
|
ModulePassManager
|
|
|
|
PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
|
|
|
// Optimize globals now that the module is fully simplified.
|
|
|
|
MPM.addPass(GlobalOptPass());
|
2016-12-22 14:59:15 +08:00
|
|
|
|
2017-05-23 00:41:57 +08:00
|
|
|
// Run partial inlining pass to partially inline functions that have
|
|
|
|
// large bodies.
|
|
|
|
if (RunPartialInlining)
|
|
|
|
MPM.addPass(PartialInlinerPass());
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
// Remove avail extern fns and globals definitions since we aren't compiling
|
|
|
|
// an object file for later LTO. For LTO we want to preserve these so they
|
|
|
|
// are eligible for inlining at link-time. Note if they are unreferenced they
|
|
|
|
// will be removed by GlobalDCE later, so this only impacts referenced
|
|
|
|
// available externally globals. Eventually they will be suppressed during
|
|
|
|
// codegen, but eliminating here enables more opportunity for GlobalDCE as it
|
|
|
|
// may make globals referenced by available external functions dead and saves
|
|
|
|
// running remaining passes on the eliminated functions.
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(EliminateAvailableExternallyPass());
|
|
|
|
|
|
|
|
// Do RPO function attribute inference across the module to forward-propagate
|
|
|
|
// attributes where applicable.
|
|
|
|
// FIXME: Is this really an optimization rather than a canonicalization?
|
|
|
|
MPM.addPass(ReversePostOrderFunctionAttrsPass());
|
|
|
|
|
2017-02-12 13:34:04 +08:00
|
|
|
// Re-require GloblasAA here prior to function passes. This is particularly
|
2016-12-22 14:59:15 +08:00
|
|
|
// useful as the above will have inlined, DCE'ed, and function-attr
|
|
|
|
// propagated everything. We should at this point have a reasonably minimal
|
|
|
|
// and richly annotated call graph. By computing aliasing and mod/ref
|
|
|
|
// information for all local globals here, the late loop passes and notably
|
|
|
|
// the vectorizer will be able to use them to help recognize vectorizable
|
|
|
|
// memory operations.
|
2017-02-12 13:34:04 +08:00
|
|
|
MPM.addPass(RequireAnalysisPass<GlobalsAA, Module>());
|
2016-12-22 14:59:15 +08:00
|
|
|
|
|
|
|
FunctionPassManager OptimizePM(DebugLogging);
|
|
|
|
OptimizePM.addPass(Float2IntPass());
|
|
|
|
// FIXME: We need to run some loop optimizations to re-rotate loops after
|
|
|
|
// simplify-cfg and others undo their rotation.
|
|
|
|
|
|
|
|
// Optimize the loop execution. These passes operate on entire loop nests
|
|
|
|
// rather than on each loop in an inside-out manner, and so they are actually
|
|
|
|
// function passes.
|
2017-01-27 08:50:21 +08:00
|
|
|
|
|
|
|
// First rotate loops that may have been un-rotated by prior passes.
|
|
|
|
OptimizePM.addPass(createFunctionToLoopPassAdaptor(LoopRotatePass()));
|
|
|
|
|
|
|
|
// Distribute loops to allow partial vectorization. I.e. isolate dependences
|
|
|
|
// into separate loop that would otherwise inhibit vectorization. This is
|
|
|
|
// currently only performed for loops marked with the metadata
|
|
|
|
// llvm.loop.distribute=true or when -enable-loop-distribute is specified.
|
2016-12-22 14:59:15 +08:00
|
|
|
OptimizePM.addPass(LoopDistributePass());
|
2017-01-27 08:50:21 +08:00
|
|
|
|
|
|
|
// Now run the core loop vectorizer.
|
2016-12-22 14:59:15 +08:00
|
|
|
OptimizePM.addPass(LoopVectorizePass());
|
2017-01-27 08:50:21 +08:00
|
|
|
|
2017-01-27 09:32:26 +08:00
|
|
|
// Eliminate loads by forwarding stores from the previous iteration to loads
|
|
|
|
// of the current iteration.
|
|
|
|
OptimizePM.addPass(LoopLoadEliminationPass());
|
2017-01-27 08:50:21 +08:00
|
|
|
|
|
|
|
// Cleanup after the loop optimization passes.
|
2016-12-22 14:59:15 +08:00
|
|
|
OptimizePM.addPass(InstCombinePass());
|
|
|
|
|
2017-01-27 08:50:21 +08:00
|
|
|
|
|
|
|
// Now that we've formed fast to execute loop structures, we do further
|
|
|
|
// optimizations. These are run afterward as they might block doing complex
|
|
|
|
// analyses and transforms such as what are needed for loop vectorization.
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
// Optimize parallel scalar instruction chains into SIMD instructions.
|
|
|
|
OptimizePM.addPass(SLPVectorizerPass());
|
|
|
|
|
2017-01-27 08:50:21 +08:00
|
|
|
// Cleanup after all of the vectorizers.
|
2016-12-22 14:59:15 +08:00
|
|
|
OptimizePM.addPass(SimplifyCFGPass());
|
|
|
|
OptimizePM.addPass(InstCombinePass());
|
|
|
|
|
|
|
|
// Unroll small loops to hide loop backedge latency and saturate any parallel
|
2017-01-27 08:50:21 +08:00
|
|
|
// execution resources of an out-of-order processor. We also then need to
|
|
|
|
// clean up redundancies and loop invariant code.
|
|
|
|
// FIXME: It would be really good to use a loop-integrated instruction
|
|
|
|
// combiner for cleanup here so that the unrolling and LICM can be pipelined
|
|
|
|
// across the loop nests.
|
Increases full-unroll threshold.
Summary:
The default threshold for fully unroll is too conservative. This patch doubles the full-unroll threshold
This change will affect the following speccpu2006 benchmarks (performance numbers were collected from Intel Sandybridge):
Performance:
403 0.11%
433 0.51%
445 0.48%
447 3.50%
453 1.49%
464 0.75%
Code size:
403 0.56%
433 0.96%
445 2.16%
447 2.96%
453 0.94%
464 8.02%
The compiler time overhead is similar with code size.
Reviewers: davidxl, mkuper, mzolotukhin, hfinkel, chandlerc
Reviewed By: hfinkel, chandlerc
Subscribers: mehdi_amini, zzheng, efriedma, haicheng, hfinkel, llvm-commits
Differential Revision: https://reviews.llvm.org/D28368
llvm-svn: 295538
2017-02-18 11:46:51 +08:00
|
|
|
OptimizePM.addPass(createFunctionToLoopPassAdaptor(LoopUnrollPass::create(Level)));
|
2017-01-27 08:50:21 +08:00
|
|
|
OptimizePM.addPass(InstCombinePass());
|
2017-02-10 07:54:57 +08:00
|
|
|
OptimizePM.addPass(RequireAnalysisPass<OptimizationRemarkEmitterAnalysis, Function>());
|
2017-01-27 08:50:21 +08:00
|
|
|
OptimizePM.addPass(createFunctionToLoopPassAdaptor(LICMPass()));
|
2016-12-22 14:59:15 +08:00
|
|
|
|
|
|
|
// Now that we've vectorized and unrolled loops, we may have more refined
|
|
|
|
// alignment information, try to re-derive it here.
|
|
|
|
OptimizePM.addPass(AlignmentFromAssumptionsPass());
|
|
|
|
|
2017-01-27 08:50:21 +08:00
|
|
|
// LoopSink pass sinks instructions hoisted by LICM, which serves as a
|
|
|
|
// canonicalization pass that enables other optimizations. As a result,
|
|
|
|
// LoopSink pass needs to be a very late IR pass to avoid undoing LICM
|
|
|
|
// result too early.
|
|
|
|
OptimizePM.addPass(LoopSinkPass());
|
|
|
|
|
|
|
|
// And finally clean up LCSSA form before generating code.
|
|
|
|
OptimizePM.addPass(InstSimplifierPass());
|
|
|
|
|
2017-04-26 20:02:41 +08:00
|
|
|
// LoopSink (and other loop passes since the last simplifyCFG) might have
|
|
|
|
// resulted in single-entry-single-exit or empty blocks. Clean up the CFG.
|
|
|
|
OptimizePM.addPass(SimplifyCFGPass());
|
|
|
|
|
2017-01-27 08:50:21 +08:00
|
|
|
// Add the core optimizing pipeline.
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(OptimizePM)));
|
|
|
|
|
|
|
|
// Now we need to do some global optimization transforms.
|
|
|
|
// FIXME: It would seem like these should come first in the optimization
|
|
|
|
// pipeline and maybe be the bottom of the canonicalization pipeline? Weird
|
|
|
|
// ordering here.
|
|
|
|
MPM.addPass(GlobalDCEPass());
|
|
|
|
MPM.addPass(ConstantMergePass());
|
|
|
|
|
|
|
|
return MPM;
|
2016-02-29 06:16:03 +08:00
|
|
|
}
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
ModulePassManager
|
|
|
|
PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
assert(Level != O0 && "Must request optimizations for the default pipeline!");
|
|
|
|
|
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
|
|
|
// Force any function attributes we want the rest of the pipeline to observe.
|
|
|
|
MPM.addPass(ForceFunctionAttrsPass());
|
|
|
|
|
|
|
|
// Add the core simplification pipeline.
|
|
|
|
MPM.addPass(buildModuleSimplificationPipeline(Level, DebugLogging));
|
|
|
|
|
|
|
|
// Now add the optimization pipeline.
|
|
|
|
MPM.addPass(buildModuleOptimizationPipeline(Level, DebugLogging));
|
|
|
|
|
|
|
|
return MPM;
|
|
|
|
}
|
|
|
|
|
|
|
|
ModulePassManager
|
|
|
|
PassBuilder::buildThinLTOPreLinkDefaultPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
assert(Level != O0 && "Must request optimizations for the default pipeline!");
|
|
|
|
|
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
|
|
|
// Force any function attributes we want the rest of the pipeline to observe.
|
|
|
|
MPM.addPass(ForceFunctionAttrsPass());
|
|
|
|
|
|
|
|
// If we are planning to perform ThinLTO later, we don't bloat the code with
|
|
|
|
// unrolling/vectorization/... now. Just simplify the module as much as we
|
|
|
|
// can.
|
|
|
|
MPM.addPass(buildModuleSimplificationPipeline(Level, DebugLogging));
|
|
|
|
|
|
|
|
// Run partial inlining pass to partially inline functions that have
|
|
|
|
// large bodies.
|
|
|
|
// FIXME: It isn't clear whether this is really the right place to run this
|
|
|
|
// in ThinLTO. Because there is another canonicalization and simplification
|
|
|
|
// phase that will run after the thin link, running this here ends up with
|
|
|
|
// less information than will be available later and it may grow functions in
|
|
|
|
// ways that aren't beneficial.
|
|
|
|
if (RunPartialInlining)
|
|
|
|
MPM.addPass(PartialInlinerPass());
|
|
|
|
|
|
|
|
// Reduce the size of the IR as much as possible.
|
|
|
|
MPM.addPass(GlobalOptPass());
|
|
|
|
|
|
|
|
// Rename anon globals to be able to export them in the summary.
|
|
|
|
MPM.addPass(NameAnonGlobalPass());
|
|
|
|
|
|
|
|
return MPM;
|
|
|
|
}
|
|
|
|
|
|
|
|
ModulePassManager
|
|
|
|
PassBuilder::buildThinLTODefaultPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
// FIXME: The summary index is not hooked in the new pass manager yet.
|
|
|
|
// When it's going to be hooked, enable WholeProgramDevirt and LowerTypeTest
|
|
|
|
// here.
|
|
|
|
|
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
|
|
|
// Force any function attributes we want the rest of the pipeline to observe.
|
|
|
|
MPM.addPass(ForceFunctionAttrsPass());
|
|
|
|
|
|
|
|
// During the ThinLTO backend phase we perform early indirect call promotion
|
|
|
|
// here, before globalopt. Otherwise imported available_externally functions
|
|
|
|
// look unreferenced and are removed.
|
|
|
|
MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */,
|
|
|
|
PGOOpt && PGOOpt->SamplePGO &&
|
|
|
|
!PGOOpt->ProfileUseFile.empty()));
|
|
|
|
|
|
|
|
// Add the core simplification pipeline.
|
|
|
|
MPM.addPass(buildModuleSimplificationPipeline(Level, DebugLogging));
|
|
|
|
|
|
|
|
// Now add the optimization pipeline.
|
|
|
|
MPM.addPass(buildModuleOptimizationPipeline(Level, DebugLogging));
|
|
|
|
|
|
|
|
return MPM;
|
|
|
|
}
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
ModulePassManager
|
|
|
|
PassBuilder::buildLTOPreLinkDefaultPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
assert(Level != O0 && "Must request optimizations for the default pipeline!");
|
2016-02-29 06:16:03 +08:00
|
|
|
// FIXME: We should use a customized pre-link pipeline!
|
2016-12-22 14:59:15 +08:00
|
|
|
return buildPerModuleDefaultPipeline(Level, DebugLogging);
|
2016-02-29 06:16:03 +08:00
|
|
|
}
|
|
|
|
|
2016-12-22 14:59:15 +08:00
|
|
|
ModulePassManager PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level,
|
|
|
|
bool DebugLogging) {
|
|
|
|
assert(Level != O0 && "Must request optimizations for the default pipeline!");
|
|
|
|
ModulePassManager MPM(DebugLogging);
|
|
|
|
|
2017-01-24 08:57:39 +08:00
|
|
|
// Remove unused virtual tables to improve the quality of code generated by
|
|
|
|
// whole-program devirtualization and bitset lowering.
|
|
|
|
MPM.addPass(GlobalDCEPass());
|
|
|
|
|
|
|
|
// Force any function attributes we want the rest of the pipeline to observe.
|
|
|
|
MPM.addPass(ForceFunctionAttrsPass());
|
|
|
|
|
|
|
|
// Do basic inference of function attributes from known properties of system
|
|
|
|
// libraries and other oracles.
|
|
|
|
MPM.addPass(InferFunctionAttrsPass());
|
|
|
|
|
|
|
|
if (Level > 1) {
|
|
|
|
// Indirect call promotion. This should promote all the targets that are
|
|
|
|
// left by the earlier promotion pass that promotes intra-module targets.
|
|
|
|
// This two-step promotion is to save the compile time. For LTO, it should
|
|
|
|
// produce the same result as if we only do promotion here.
|
2017-02-24 06:15:18 +08:00
|
|
|
MPM.addPass(PGOIndirectCallPromotion(true /* InLTO */,
|
|
|
|
PGOOpt && PGOOpt->SamplePGO));
|
2017-01-24 08:57:39 +08:00
|
|
|
|
|
|
|
// Propagate constants at call sites into the functions they call. This
|
|
|
|
// opens opportunities for globalopt (and inlining) by substituting function
|
|
|
|
// pointers passed as arguments to direct uses of functions.
|
|
|
|
MPM.addPass(IPSCCPPass());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now deduce any function attributes based in the current code.
|
|
|
|
MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
|
|
|
|
PostOrderFunctionAttrsPass()));
|
|
|
|
|
|
|
|
// Do RPO function attribute inference across the module to forward-propagate
|
|
|
|
// attributes where applicable.
|
|
|
|
// FIXME: Is this really an optimization rather than a canonicalization?
|
|
|
|
MPM.addPass(ReversePostOrderFunctionAttrsPass());
|
|
|
|
|
|
|
|
// Use inragne annotations on GEP indices to split globals where beneficial.
|
|
|
|
MPM.addPass(GlobalSplitPass());
|
2016-02-29 06:16:03 +08:00
|
|
|
|
2017-01-24 08:57:39 +08:00
|
|
|
// Run whole program optimization of virtual call when the list of callees
|
|
|
|
// is fixed.
|
|
|
|
MPM.addPass(WholeProgramDevirtPass());
|
|
|
|
|
|
|
|
// Stop here at -O1.
|
|
|
|
if (Level == 1)
|
|
|
|
return MPM;
|
|
|
|
|
|
|
|
// Optimize globals to try and fold them into constants.
|
|
|
|
MPM.addPass(GlobalOptPass());
|
|
|
|
|
|
|
|
// Promote any localized globals to SSA registers.
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(PromotePass()));
|
|
|
|
|
|
|
|
// Linking modules together can lead to duplicate global constant, only
|
|
|
|
// keep one copy of each constant.
|
|
|
|
MPM.addPass(ConstantMergePass());
|
|
|
|
|
|
|
|
// Remove unused arguments from functions.
|
|
|
|
MPM.addPass(DeadArgumentEliminationPass());
|
|
|
|
|
|
|
|
// Reduce the code after globalopt and ipsccp. Both can open up significant
|
|
|
|
// simplification opportunities, and both can propagate functions through
|
|
|
|
// function pointers. When this happens, we often have to resolve varargs
|
|
|
|
// calls, etc, so let instcombine do this.
|
|
|
|
// FIXME: add peephole extensions here as the legacy PM does.
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(InstCombinePass()));
|
|
|
|
|
|
|
|
// Note: historically, the PruneEH pass was run first to deduce nounwind and
|
|
|
|
// generally clean up exception handling overhead. It isn't clear this is
|
|
|
|
// valuable as the inliner doesn't currently care whether it is inlining an
|
|
|
|
// invoke or a call.
|
|
|
|
// Run the inliner now.
|
|
|
|
MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(InlinerPass()));
|
|
|
|
|
|
|
|
// Optimize globals again after we ran the inliner.
|
|
|
|
MPM.addPass(GlobalOptPass());
|
|
|
|
|
|
|
|
// Garbage collect dead functions.
|
|
|
|
// FIXME: Add ArgumentPromotion pass after once it's ported.
|
|
|
|
MPM.addPass(GlobalDCEPass());
|
|
|
|
|
|
|
|
FunctionPassManager FPM(DebugLogging);
|
|
|
|
|
|
|
|
// The IPO Passes may leave cruft around. Clean up after them.
|
|
|
|
// FIXME: add peephole extensions here as the legacy PM does.
|
|
|
|
FPM.addPass(InstCombinePass());
|
|
|
|
FPM.addPass(JumpThreadingPass());
|
|
|
|
|
|
|
|
// Break up allocas
|
|
|
|
FPM.addPass(SROA());
|
|
|
|
|
|
|
|
// Run a few AA driver optimizations here and now to cleanup the code.
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
|
|
|
|
|
|
|
|
MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(
|
|
|
|
PostOrderFunctionAttrsPass()));
|
|
|
|
// FIXME: here we run IP alias analysis in the legacy PM.
|
|
|
|
|
|
|
|
FunctionPassManager MainFPM;
|
|
|
|
|
|
|
|
// FIXME: once we fix LoopPass Manager, add LICM here.
|
|
|
|
// FIXME: once we provide support for enabling MLSM, add it here.
|
|
|
|
// FIXME: once we provide support for enabling NewGVN, add it here.
|
2017-05-23 07:41:40 +08:00
|
|
|
if (RunNewGVN)
|
|
|
|
MainFPM.addPass(NewGVNPass());
|
|
|
|
else
|
|
|
|
MainFPM.addPass(GVN());
|
2017-01-24 08:57:39 +08:00
|
|
|
|
|
|
|
// Remove dead memcpy()'s.
|
|
|
|
MainFPM.addPass(MemCpyOptPass());
|
|
|
|
|
|
|
|
// Nuke dead stores.
|
|
|
|
MainFPM.addPass(DSEPass());
|
|
|
|
|
|
|
|
// FIXME: at this point, we run a bunch of loop passes:
|
|
|
|
// indVarSimplify, loopDeletion, loopInterchange, loopUnrool,
|
|
|
|
// loopVectorize. Enable them once the remaining issue with LPM
|
|
|
|
// are sorted out.
|
|
|
|
|
|
|
|
MainFPM.addPass(InstCombinePass());
|
|
|
|
MainFPM.addPass(SimplifyCFGPass());
|
|
|
|
MainFPM.addPass(SCCPPass());
|
|
|
|
MainFPM.addPass(InstCombinePass());
|
|
|
|
MainFPM.addPass(BDCEPass());
|
|
|
|
|
|
|
|
// FIXME: We may want to run SLPVectorizer here.
|
|
|
|
// After vectorization, assume intrinsics may tell us more
|
|
|
|
// about pointer alignments.
|
|
|
|
#if 0
|
|
|
|
MainFPM.add(AlignmentFromAssumptionsPass());
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// FIXME: add peephole extensions to the PM here.
|
|
|
|
MainFPM.addPass(InstCombinePass());
|
|
|
|
MainFPM.addPass(JumpThreadingPass());
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(MainFPM)));
|
|
|
|
|
|
|
|
// Create a function that performs CFI checks for cross-DSO calls with
|
|
|
|
// targets in the current module.
|
|
|
|
MPM.addPass(CrossDSOCFIPass());
|
|
|
|
|
|
|
|
// Lower type metadata and the type.test intrinsic. This pass supports
|
|
|
|
// clang's control flow integrity mechanisms (-fsanitize=cfi*) and needs
|
|
|
|
// to be run at link time if CFI is enabled. This pass does nothing if
|
|
|
|
// CFI is disabled.
|
|
|
|
// Enable once we add support for the summary in the new PM.
|
|
|
|
#if 0
|
2017-02-10 05:45:01 +08:00
|
|
|
MPM.addPass(LowerTypeTestsPass(Summary ? PassSummaryAction::Export :
|
|
|
|
PassSummaryAction::None,
|
2017-01-24 08:57:39 +08:00
|
|
|
Summary));
|
|
|
|
#endif
|
|
|
|
|
|
|
|
// Add late LTO optimization passes.
|
|
|
|
// Delete basic blocks, which optimization passes may have killed.
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(SimplifyCFGPass()));
|
|
|
|
|
|
|
|
// Drop bodies of available eternally objects to improve GlobalDCE.
|
|
|
|
MPM.addPass(EliminateAvailableExternallyPass());
|
|
|
|
|
|
|
|
// Now that we have optimized the program, discard unreachable functions.
|
|
|
|
MPM.addPass(GlobalDCEPass());
|
2016-12-22 14:59:15 +08:00
|
|
|
|
2017-01-24 08:57:39 +08:00
|
|
|
// FIXME: Enable MergeFuncs, conditionally, after ported, maybe.
|
2016-12-22 14:59:15 +08:00
|
|
|
return MPM;
|
2016-02-29 06:16:03 +08:00
|
|
|
}
|
|
|
|
|
2016-12-24 04:38:19 +08:00
|
|
|
AAManager PassBuilder::buildDefaultAAPipeline() {
|
|
|
|
AAManager AA;
|
|
|
|
|
|
|
|
// The order in which these are registered determines their priority when
|
|
|
|
// being queried.
|
|
|
|
|
|
|
|
// First we register the basic alias analysis that provides the majority of
|
|
|
|
// per-function local AA logic. This is a stateless, on-demand local set of
|
|
|
|
// AA techniques.
|
|
|
|
AA.registerFunctionAnalysis<BasicAA>();
|
|
|
|
|
|
|
|
// Next we query fast, specialized alias analyses that wrap IR-embedded
|
|
|
|
// information about aliasing.
|
|
|
|
AA.registerFunctionAnalysis<ScopedNoAliasAA>();
|
|
|
|
AA.registerFunctionAnalysis<TypeBasedAA>();
|
|
|
|
|
|
|
|
// Add support for querying global aliasing information when available.
|
2016-12-24 13:11:17 +08:00
|
|
|
// Because the `AAManager` is a function analysis and `GlobalsAA` is a module
|
|
|
|
// analysis, all that the `AAManager` can do is query for any *cached*
|
2017-02-12 13:34:04 +08:00
|
|
|
// results from `GlobalsAA` through a readonly proxy.
|
2016-12-24 04:38:19 +08:00
|
|
|
AA.registerModuleAnalysis<GlobalsAA>();
|
|
|
|
|
|
|
|
return AA;
|
|
|
|
}
|
|
|
|
|
2016-08-03 15:44:48 +08:00
|
|
|
static Optional<int> parseRepeatPassName(StringRef Name) {
|
|
|
|
if (!Name.consume_front("repeat<") || !Name.consume_back(">"))
|
|
|
|
return None;
|
|
|
|
int Count;
|
|
|
|
if (Name.getAsInteger(0, Count) || Count <= 0)
|
|
|
|
return None;
|
|
|
|
return Count;
|
|
|
|
}
|
|
|
|
|
2016-12-28 19:07:33 +08:00
|
|
|
static Optional<int> parseDevirtPassName(StringRef Name) {
|
|
|
|
if (!Name.consume_front("devirt<") || !Name.consume_back(">"))
|
|
|
|
return None;
|
|
|
|
int Count;
|
|
|
|
if (Name.getAsInteger(0, Count) || Count <= 0)
|
|
|
|
return None;
|
|
|
|
return Count;
|
|
|
|
}
|
|
|
|
|
2017-06-01 19:39:39 +08:00
|
|
|
/// Tests whether a pass name starts with a valid prefix for a default pipeline
|
|
|
|
/// alias.
|
|
|
|
static bool startsWithDefaultPipelineAliasPrefix(StringRef Name) {
|
|
|
|
return Name.startswith("default") || Name.startswith("thinlto") ||
|
|
|
|
Name.startswith("lto");
|
|
|
|
}
|
|
|
|
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
static bool isModulePassName(StringRef Name) {
|
2016-02-29 06:16:03 +08:00
|
|
|
// Manually handle aliases for pre-configured pipeline fragments.
|
2017-06-01 19:39:39 +08:00
|
|
|
if (startsWithDefaultPipelineAliasPrefix(Name))
|
2016-02-29 06:16:03 +08:00
|
|
|
return DefaultAliasRegex.match(Name);
|
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
// Explicitly handle pass manager names.
|
|
|
|
if (Name == "module")
|
|
|
|
return true;
|
|
|
|
if (Name == "cgscc")
|
|
|
|
return true;
|
|
|
|
if (Name == "function")
|
|
|
|
return true;
|
|
|
|
|
2016-08-03 15:44:48 +08:00
|
|
|
// Explicitly handle custom-parsed pass names.
|
|
|
|
if (parseRepeatPassName(Name))
|
|
|
|
return true;
|
|
|
|
|
2016-06-17 15:15:29 +08:00
|
|
|
#define MODULE_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) \
|
|
|
|
return true;
|
2015-01-06 10:10:51 +08:00
|
|
|
#define MODULE_ANALYSIS(NAME, CREATE_PASS) \
|
2015-01-06 12:49:44 +08:00
|
|
|
if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true;
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2014-04-21 19:12:00 +08:00
|
|
|
static bool isCGSCCPassName(StringRef Name) {
|
2016-08-03 11:21:41 +08:00
|
|
|
// Explicitly handle pass manager names.
|
|
|
|
if (Name == "cgscc")
|
|
|
|
return true;
|
|
|
|
if (Name == "function")
|
|
|
|
return true;
|
|
|
|
|
2016-08-03 15:44:48 +08:00
|
|
|
// Explicitly handle custom-parsed pass names.
|
|
|
|
if (parseRepeatPassName(Name))
|
|
|
|
return true;
|
2016-12-28 19:07:33 +08:00
|
|
|
if (parseDevirtPassName(Name))
|
|
|
|
return true;
|
2016-08-03 15:44:48 +08:00
|
|
|
|
2016-06-17 15:15:29 +08:00
|
|
|
#define CGSCC_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) \
|
|
|
|
return true;
|
2015-01-06 10:10:51 +08:00
|
|
|
#define CGSCC_ANALYSIS(NAME, CREATE_PASS) \
|
2015-01-06 12:49:44 +08:00
|
|
|
if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true;
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
2014-04-21 19:12:00 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2014-01-12 17:34:22 +08:00
|
|
|
static bool isFunctionPassName(StringRef Name) {
|
2016-08-03 11:21:41 +08:00
|
|
|
// Explicitly handle pass manager names.
|
|
|
|
if (Name == "function")
|
|
|
|
return true;
|
|
|
|
if (Name == "loop")
|
|
|
|
return true;
|
|
|
|
|
2016-08-03 15:44:48 +08:00
|
|
|
// Explicitly handle custom-parsed pass names.
|
|
|
|
if (parseRepeatPassName(Name))
|
|
|
|
return true;
|
|
|
|
|
2016-06-17 15:15:29 +08:00
|
|
|
#define FUNCTION_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) \
|
|
|
|
return true;
|
2015-01-06 10:10:51 +08:00
|
|
|
#define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \
|
2015-01-06 12:49:44 +08:00
|
|
|
if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true;
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
2014-01-12 17:34:22 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-02-25 15:23:08 +08:00
|
|
|
static bool isLoopPassName(StringRef Name) {
|
2016-08-03 11:21:41 +08:00
|
|
|
// Explicitly handle pass manager names.
|
|
|
|
if (Name == "loop")
|
|
|
|
return true;
|
|
|
|
|
2016-08-03 15:44:48 +08:00
|
|
|
// Explicitly handle custom-parsed pass names.
|
|
|
|
if (parseRepeatPassName(Name))
|
|
|
|
return true;
|
|
|
|
|
2016-06-17 15:15:29 +08:00
|
|
|
#define LOOP_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) \
|
|
|
|
return true;
|
2016-02-25 15:23:08 +08:00
|
|
|
#define LOOP_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == "require<" NAME ">" || Name == "invalidate<" NAME ">") \
|
|
|
|
return true;
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
Optional<std::vector<PassBuilder::PipelineElement>>
|
|
|
|
PassBuilder::parsePipelineText(StringRef Text) {
|
|
|
|
std::vector<PipelineElement> ResultPipeline;
|
|
|
|
|
|
|
|
SmallVector<std::vector<PipelineElement> *, 4> PipelineStack = {
|
|
|
|
&ResultPipeline};
|
|
|
|
for (;;) {
|
|
|
|
std::vector<PipelineElement> &Pipeline = *PipelineStack.back();
|
|
|
|
size_t Pos = Text.find_first_of(",()");
|
|
|
|
Pipeline.push_back({Text.substr(0, Pos), {}});
|
|
|
|
|
|
|
|
// If we have a single terminating name, we're done.
|
|
|
|
if (Pos == Text.npos)
|
|
|
|
break;
|
|
|
|
|
|
|
|
char Sep = Text[Pos];
|
|
|
|
Text = Text.substr(Pos + 1);
|
|
|
|
if (Sep == ',')
|
|
|
|
// Just a name ending in a comma, continue.
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (Sep == '(') {
|
|
|
|
// Push the inner pipeline onto the stack to continue processing.
|
|
|
|
PipelineStack.push_back(&Pipeline.back().InnerPipeline);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(Sep == ')' && "Bogus separator!");
|
|
|
|
// When handling the close parenthesis, we greedily consume them to avoid
|
|
|
|
// empty strings in the pipeline.
|
|
|
|
do {
|
|
|
|
// If we try to pop the outer pipeline we have unbalanced parentheses.
|
|
|
|
if (PipelineStack.size() == 1)
|
|
|
|
return None;
|
|
|
|
|
|
|
|
PipelineStack.pop_back();
|
|
|
|
} while (Text.consume_front(")"));
|
|
|
|
|
|
|
|
// Check if we've finished parsing.
|
|
|
|
if (Text.empty())
|
|
|
|
break;
|
|
|
|
|
|
|
|
// Otherwise, the end of an inner pipeline always has to be followed by
|
|
|
|
// a comma, and then we can continue.
|
|
|
|
if (!Text.consume_front(","))
|
|
|
|
return None;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (PipelineStack.size() > 1)
|
|
|
|
// Unbalanced paretheses.
|
|
|
|
return None;
|
|
|
|
|
|
|
|
assert(PipelineStack.back() == &ResultPipeline &&
|
|
|
|
"Wrong pipeline at the bottom of the stack!");
|
|
|
|
return {std::move(ResultPipeline)};
|
|
|
|
}
|
|
|
|
|
|
|
|
bool PassBuilder::parseModulePass(ModulePassManager &MPM,
|
|
|
|
const PipelineElement &E, bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
|
|
|
auto &Name = E.Name;
|
|
|
|
auto &InnerPipeline = E.InnerPipeline;
|
|
|
|
|
|
|
|
// First handle complex passes like the pass managers which carry pipelines.
|
|
|
|
if (!InnerPipeline.empty()) {
|
|
|
|
if (Name == "module") {
|
|
|
|
ModulePassManager NestedMPM(DebugLogging);
|
|
|
|
if (!parseModulePassPipeline(NestedMPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
MPM.addPass(std::move(NestedMPM));
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name == "cgscc") {
|
|
|
|
CGSCCPassManager CGPM(DebugLogging);
|
|
|
|
if (!parseCGSCCPassPipeline(CGPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
MPM.addPass(createModuleToPostOrderCGSCCPassAdaptor(std::move(CGPM),
|
|
|
|
DebugLogging));
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name == "function") {
|
|
|
|
FunctionPassManager FPM(DebugLogging);
|
|
|
|
if (!parseFunctionPassPipeline(FPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
MPM.addPass(createModuleToFunctionPassAdaptor(std::move(FPM)));
|
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 15:44:48 +08:00
|
|
|
if (auto Count = parseRepeatPassName(Name)) {
|
|
|
|
ModulePassManager NestedMPM(DebugLogging);
|
|
|
|
if (!parseModulePassPipeline(NestedMPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
2016-08-04 11:52:53 +08:00
|
|
|
MPM.addPass(createRepeatedPass(*Count, std::move(NestedMPM)));
|
2016-08-03 15:44:48 +08:00
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
// Normal passes can't have pipelines.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-02-29 06:16:03 +08:00
|
|
|
// Manually handle aliases for pre-configured pipeline fragments.
|
2017-06-01 19:39:39 +08:00
|
|
|
if (startsWithDefaultPipelineAliasPrefix(Name)) {
|
2016-02-29 06:16:03 +08:00
|
|
|
SmallVector<StringRef, 3> Matches;
|
|
|
|
if (!DefaultAliasRegex.match(Name, &Matches))
|
|
|
|
return false;
|
|
|
|
assert(Matches.size() == 3 && "Must capture two matched strings!");
|
|
|
|
|
2016-07-26 02:34:51 +08:00
|
|
|
OptimizationLevel L = StringSwitch<OptimizationLevel>(Matches[2])
|
|
|
|
.Case("O0", O0)
|
|
|
|
.Case("O1", O1)
|
|
|
|
.Case("O2", O2)
|
|
|
|
.Case("O3", O3)
|
|
|
|
.Case("Os", Os)
|
|
|
|
.Case("Oz", Oz);
|
2016-12-22 14:59:15 +08:00
|
|
|
if (L == O0)
|
|
|
|
// At O0 we do nothing at all!
|
|
|
|
return true;
|
2016-02-29 06:16:03 +08:00
|
|
|
|
|
|
|
if (Matches[1] == "default") {
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(buildPerModuleDefaultPipeline(L, DebugLogging));
|
2017-06-01 19:39:39 +08:00
|
|
|
} else if (Matches[1] == "thinlto-pre-link") {
|
|
|
|
MPM.addPass(buildThinLTOPreLinkDefaultPipeline(L, DebugLogging));
|
|
|
|
} else if (Matches[1] == "thinlto") {
|
|
|
|
MPM.addPass(buildThinLTODefaultPipeline(L, DebugLogging));
|
2016-02-29 06:16:03 +08:00
|
|
|
} else if (Matches[1] == "lto-pre-link") {
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(buildLTOPreLinkDefaultPipeline(L, DebugLogging));
|
2016-02-29 06:16:03 +08:00
|
|
|
} else {
|
|
|
|
assert(Matches[1] == "lto" && "Not one of the matched options!");
|
2016-12-22 14:59:15 +08:00
|
|
|
MPM.addPass(buildLTODefaultPipeline(L, DebugLogging));
|
2016-02-29 06:16:03 +08:00
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
// Finally expand the basic registered passes from the .inc file.
|
2014-04-21 16:08:50 +08:00
|
|
|
#define MODULE_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
|
|
|
MPM.addPass(CREATE_PASS); \
|
|
|
|
return true; \
|
2014-02-06 12:37:03 +08:00
|
|
|
}
|
2015-01-06 10:10:51 +08:00
|
|
|
#define MODULE_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == "require<" NAME ">") { \
|
2016-08-20 02:36:06 +08:00
|
|
|
MPM.addPass( \
|
|
|
|
RequireAnalysisPass< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type, Module>()); \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true; \
|
2015-01-06 12:49:44 +08:00
|
|
|
} \
|
|
|
|
if (Name == "invalidate<" NAME ">") { \
|
2016-02-26 20:30:18 +08:00
|
|
|
MPM.addPass(InvalidateAnalysisPass< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>()); \
|
2015-01-06 12:49:44 +08:00
|
|
|
return true; \
|
2015-01-06 10:10:51 +08:00
|
|
|
}
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
bool PassBuilder::parseCGSCCPass(CGSCCPassManager &CGPM,
|
|
|
|
const PipelineElement &E, bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
|
|
|
auto &Name = E.Name;
|
|
|
|
auto &InnerPipeline = E.InnerPipeline;
|
|
|
|
|
|
|
|
// First handle complex passes like the pass managers which carry pipelines.
|
|
|
|
if (!InnerPipeline.empty()) {
|
|
|
|
if (Name == "cgscc") {
|
|
|
|
CGSCCPassManager NestedCGPM(DebugLogging);
|
|
|
|
if (!parseCGSCCPassPipeline(NestedCGPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
// Add the nested pass manager with the appropriate adaptor.
|
|
|
|
CGPM.addPass(std::move(NestedCGPM));
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name == "function") {
|
|
|
|
FunctionPassManager FPM(DebugLogging);
|
|
|
|
if (!parseFunctionPassPipeline(FPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
// Add the nested pass manager with the appropriate adaptor.
|
|
|
|
CGPM.addPass(
|
|
|
|
createCGSCCToFunctionPassAdaptor(std::move(FPM), DebugLogging));
|
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 15:44:48 +08:00
|
|
|
if (auto Count = parseRepeatPassName(Name)) {
|
|
|
|
CGSCCPassManager NestedCGPM(DebugLogging);
|
|
|
|
if (!parseCGSCCPassPipeline(NestedCGPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
2016-08-04 11:52:53 +08:00
|
|
|
CGPM.addPass(createRepeatedPass(*Count, std::move(NestedCGPM)));
|
2016-08-03 15:44:48 +08:00
|
|
|
return true;
|
|
|
|
}
|
2016-12-28 19:07:33 +08:00
|
|
|
if (auto MaxRepetitions = parseDevirtPassName(Name)) {
|
|
|
|
CGSCCPassManager NestedCGPM(DebugLogging);
|
|
|
|
if (!parseCGSCCPassPipeline(NestedCGPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
CGPM.addPass(createDevirtSCCRepeatedPass(std::move(NestedCGPM),
|
|
|
|
*MaxRepetitions, DebugLogging));
|
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
// Normal passes can't have pipelines.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now expand the basic registered passes from the .inc file.
|
2014-04-21 19:12:00 +08:00
|
|
|
#define CGSCC_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
|
|
|
CGPM.addPass(CREATE_PASS); \
|
|
|
|
return true; \
|
|
|
|
}
|
2015-01-06 10:10:51 +08:00
|
|
|
#define CGSCC_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == "require<" NAME ">") { \
|
2016-02-26 20:30:18 +08:00
|
|
|
CGPM.addPass(RequireAnalysisPass< \
|
2016-08-20 02:36:06 +08:00
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type, \
|
[PM] Introduce basic update capabilities to the new PM's CGSCC pass
manager, including both plumbing and logic to handle function pass
updates.
There are three fundamentally tied changes here:
1) Plumbing *some* mechanism for updating the CGSCC pass manager as the
CG changes while passes are running.
2) Changing the CGSCC pass manager infrastructure to have support for
the underlying graph to mutate mid-pass run.
3) Actually updating the CG after function passes run.
I can separate them if necessary, but I think its really useful to have
them together as the needs of #3 drove #2, and that in turn drove #1.
The plumbing technique is to extend the "run" method signature with
extra arguments. We provide the call graph that intrinsically is
available as it is the basis of the pass manager's IR units, and an
output parameter that records the results of updating the call graph
during an SCC passes's run. Note that "...UpdateResult" isn't a *great*
name here... suggestions very welcome.
I tried a pretty frustrating number of different data structures and such
for the innards of the update result. Every other one failed for one
reason or another. Sometimes I just couldn't keep the layers of
complexity right in my head. The thing that really worked was to just
directly provide access to the underlying structures used to walk the
call graph so that their updates could be informed by the *particular*
nature of the change to the graph.
The technique for how to make the pass management infrastructure cope
with mutating graphs was also something that took a really, really large
number of iterations to get to a place where I was happy. Here are some
of the considerations that drove the design:
- We operate at three levels within the infrastructure: RefSCC, SCC, and
Node. In each case, we are working bottom up and so we want to
continue to iterate on the "lowest" node as the graph changes. Look at
how we iterate over nodes in an SCC running function passes as those
function passes mutate the CG. We continue to iterate on the "lowest"
SCC, which is the one that continues to contain the function just
processed.
- The call graph structure re-uses SCCs (and RefSCCs) during mutation
events for the *highest* entry in the resulting new subgraph, not the
lowest. This means that it is necessary to continually update the
current SCC or RefSCC as it shifts. This is really surprising and
subtle, and took a long time for me to work out. I actually tried
changing the call graph to provide the opposite behavior, and it
breaks *EVERYTHING*. The graph update algorithms are really deeply
tied to this particualr pattern.
- When SCCs or RefSCCs are split apart and refined and we continually
re-pin our processing to the bottom one in the subgraph, we need to
enqueue the newly formed SCCs and RefSCCs for subsequent processing.
Queuing them presents a few challenges:
1) SCCs and RefSCCs use wildly different iteration strategies at
a high level. We end up needing to converge them on worklist
approaches that can be extended in order to be able to handle the
mutations.
2) The order of the enqueuing need to remain bottom-up post-order so
that we don't get surprising order of visitation for things like
the inliner.
3) We need the worklists to have set semantics so we don't duplicate
things endlessly. We don't need a *persistent* set though because
we always keep processing the bottom node!!!! This is super, super
surprising to me and took a long time to convince myself this is
correct, but I'm pretty sure it is... Once we sink down to the
bottom node, we can't re-split out the same node in any way, and
the postorder of the current queue is fixed and unchanging.
4) We need to make sure that the "current" SCC or RefSCC actually gets
enqueued here such that we re-visit it because we continue
processing a *new*, *bottom* SCC/RefSCC.
- We also need the ability to *skip* SCCs and RefSCCs that get merged
into a larger component. We even need the ability to skip *nodes* from
an SCC that are no longer part of that SCC.
This led to the design you see in the patch which uses SetVector-based
worklists. The RefSCC worklist is always empty until an update occurs
and is just used to handle those RefSCCs created by updates as the
others don't even exist yet and are formed on-demand during the
bottom-up walk. The SCC worklist is pre-populated from the RefSCC, and
we push new SCCs onto it and blacklist existing SCCs on it to get the
desired processing.
We then *directly* update these when updating the call graph as I was
never able to find a satisfactory abstraction around the update
strategy.
Finally, we need to compute the updates for function passes. This is
mostly used as an initial customer of all the update mechanisms to drive
their design to at least cover some real set of use cases. There are
a bunch of interesting things that came out of doing this:
- It is really nice to do this a function at a time because that
function is likely hot in the cache. This means we want even the
function pass adaptor to support online updates to the call graph!
- To update the call graph after arbitrary function pass mutations is
quite hard. We have to build a fairly comprehensive set of
data structures and then process them. Fortunately, some of this code
is related to the code for building the cal graph in the first place.
Unfortunately, very little of it makes any sense to share because the
nature of what we're doing is so very different. I've factored out the
one part that made sense at least.
- We need to transfer these updates into the various structures for the
CGSCC pass manager. Once those were more sanely worked out, this
became relatively easier. But some of those needs necessitated changes
to the LazyCallGraph interface to make it significantly easier to
extract the changed SCCs from an update operation.
- We also need to update the CGSCC analysis manager as the shape of the
graph changes. When an SCC is merged away we need to clear analyses
associated with it from the analysis manager which we didn't have
support for in the analysis manager infrsatructure. New SCCs are easy!
But then we have the case that the original SCC has its shape changed
but remains in the call graph. There we need to *invalidate* the
analyses associated with it.
- We also need to invalidate analyses after we *finish* processing an
SCC. But the analyses we need to invalidate here are *only those for
the newly updated SCC*!!! Because we only continue processing the
bottom SCC, if we split SCCs apart the original one gets invalidated
once when its shape changes and is not processed farther so its
analyses will be correct. It is the bottom SCC which continues being
processed and needs to have the "normal" invalidation done based on
the preserved analyses set.
All of this is mostly background and context for the changes here.
Many thanks to all the reviewers who helped here. Especially Sanjoy who
caught several interesting bugs in the graph algorithms, David, Sean,
and others who all helped with feedback.
Differential Revision: http://reviews.llvm.org/D21464
llvm-svn: 279618
2016-08-24 17:37:14 +08:00
|
|
|
LazyCallGraph::SCC, CGSCCAnalysisManager, LazyCallGraph &, \
|
|
|
|
CGSCCUpdateResult &>()); \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true; \
|
2015-01-06 12:49:44 +08:00
|
|
|
} \
|
|
|
|
if (Name == "invalidate<" NAME ">") { \
|
2016-02-26 20:30:18 +08:00
|
|
|
CGPM.addPass(InvalidateAnalysisPass< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>()); \
|
2015-01-06 12:49:44 +08:00
|
|
|
return true; \
|
2015-01-06 10:10:51 +08:00
|
|
|
}
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
2014-04-21 19:12:00 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
bool PassBuilder::parseFunctionPass(FunctionPassManager &FPM,
|
|
|
|
const PipelineElement &E,
|
|
|
|
bool VerifyEachPass, bool DebugLogging) {
|
|
|
|
auto &Name = E.Name;
|
|
|
|
auto &InnerPipeline = E.InnerPipeline;
|
|
|
|
|
|
|
|
// First handle complex passes like the pass managers which carry pipelines.
|
|
|
|
if (!InnerPipeline.empty()) {
|
|
|
|
if (Name == "function") {
|
|
|
|
FunctionPassManager NestedFPM(DebugLogging);
|
|
|
|
if (!parseFunctionPassPipeline(NestedFPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
// Add the nested pass manager with the appropriate adaptor.
|
|
|
|
FPM.addPass(std::move(NestedFPM));
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (Name == "loop") {
|
|
|
|
LoopPassManager LPM(DebugLogging);
|
|
|
|
if (!parseLoopPassPipeline(LPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
// Add the nested pass manager with the appropriate adaptor.
|
|
|
|
FPM.addPass(createFunctionToLoopPassAdaptor(std::move(LPM)));
|
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 15:44:48 +08:00
|
|
|
if (auto Count = parseRepeatPassName(Name)) {
|
|
|
|
FunctionPassManager NestedFPM(DebugLogging);
|
|
|
|
if (!parseFunctionPassPipeline(NestedFPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
2016-08-04 11:52:53 +08:00
|
|
|
FPM.addPass(createRepeatedPass(*Count, std::move(NestedFPM)));
|
2016-08-03 15:44:48 +08:00
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
// Normal passes can't have pipelines.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now expand the basic registered passes from the .inc file.
|
2014-04-21 16:08:50 +08:00
|
|
|
#define FUNCTION_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
|
|
|
FPM.addPass(CREATE_PASS); \
|
|
|
|
return true; \
|
2014-01-12 20:15:39 +08:00
|
|
|
}
|
2015-01-06 10:10:51 +08:00
|
|
|
#define FUNCTION_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == "require<" NAME ">") { \
|
2016-08-20 02:36:06 +08:00
|
|
|
FPM.addPass( \
|
|
|
|
RequireAnalysisPass< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type, Function>()); \
|
2015-01-06 10:10:51 +08:00
|
|
|
return true; \
|
2015-01-06 12:49:44 +08:00
|
|
|
} \
|
|
|
|
if (Name == "invalidate<" NAME ">") { \
|
2016-02-26 20:30:18 +08:00
|
|
|
FPM.addPass(InvalidateAnalysisPass< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>()); \
|
2015-01-06 12:49:44 +08:00
|
|
|
return true; \
|
2015-01-06 10:10:51 +08:00
|
|
|
}
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
2014-01-12 17:34:22 +08:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-08-03 17:14:03 +08:00
|
|
|
bool PassBuilder::parseLoopPass(LoopPassManager &LPM, const PipelineElement &E,
|
2016-08-03 11:21:41 +08:00
|
|
|
bool VerifyEachPass, bool DebugLogging) {
|
2016-08-03 15:44:48 +08:00
|
|
|
StringRef Name = E.Name;
|
2016-08-03 11:21:41 +08:00
|
|
|
auto &InnerPipeline = E.InnerPipeline;
|
|
|
|
|
|
|
|
// First handle complex passes like the pass managers which carry pipelines.
|
|
|
|
if (!InnerPipeline.empty()) {
|
|
|
|
if (Name == "loop") {
|
|
|
|
LoopPassManager NestedLPM(DebugLogging);
|
|
|
|
if (!parseLoopPassPipeline(NestedLPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
|
|
|
// Add the nested pass manager with the appropriate adaptor.
|
2016-08-03 17:14:03 +08:00
|
|
|
LPM.addPass(std::move(NestedLPM));
|
2016-08-03 11:21:41 +08:00
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 15:44:48 +08:00
|
|
|
if (auto Count = parseRepeatPassName(Name)) {
|
|
|
|
LoopPassManager NestedLPM(DebugLogging);
|
|
|
|
if (!parseLoopPassPipeline(NestedLPM, InnerPipeline, VerifyEachPass,
|
|
|
|
DebugLogging))
|
|
|
|
return false;
|
2016-08-04 11:52:53 +08:00
|
|
|
LPM.addPass(createRepeatedPass(*Count, std::move(NestedLPM)));
|
2016-08-03 15:44:48 +08:00
|
|
|
return true;
|
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
// Normal passes can't have pipelines.
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now expand the basic registered passes from the .inc file.
|
2016-02-25 15:23:08 +08:00
|
|
|
#define LOOP_PASS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
2016-08-03 17:14:03 +08:00
|
|
|
LPM.addPass(CREATE_PASS); \
|
2016-02-25 15:23:08 +08:00
|
|
|
return true; \
|
|
|
|
}
|
|
|
|
#define LOOP_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == "require<" NAME ">") { \
|
2016-08-03 17:14:03 +08:00
|
|
|
LPM.addPass(RequireAnalysisPass< \
|
2017-01-11 14:23:21 +08:00
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type, Loop, \
|
|
|
|
LoopAnalysisManager, LoopStandardAnalysisResults &, \
|
|
|
|
LPMUpdater &>()); \
|
2016-02-25 15:23:08 +08:00
|
|
|
return true; \
|
|
|
|
} \
|
|
|
|
if (Name == "invalidate<" NAME ">") { \
|
2016-08-03 17:14:03 +08:00
|
|
|
LPM.addPass(InvalidateAnalysisPass< \
|
2016-02-26 20:30:18 +08:00
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>()); \
|
2016-02-25 15:23:08 +08:00
|
|
|
return true; \
|
|
|
|
}
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-02-18 17:45:17 +08:00
|
|
|
bool PassBuilder::parseAAPassName(AAManager &AA, StringRef Name) {
|
2016-03-11 17:15:11 +08:00
|
|
|
#define MODULE_ALIAS_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
|
|
|
AA.registerModuleAnalysis< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>(); \
|
|
|
|
return true; \
|
|
|
|
}
|
2016-02-18 17:45:17 +08:00
|
|
|
#define FUNCTION_ALIAS_ANALYSIS(NAME, CREATE_PASS) \
|
|
|
|
if (Name == NAME) { \
|
2016-02-26 20:17:54 +08:00
|
|
|
AA.registerFunctionAnalysis< \
|
|
|
|
std::remove_reference<decltype(CREATE_PASS)>::type>(); \
|
2016-02-18 17:45:17 +08:00
|
|
|
return true; \
|
|
|
|
}
|
|
|
|
#include "PassRegistry.def"
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2016-02-25 15:23:08 +08:00
|
|
|
bool PassBuilder::parseLoopPassPipeline(LoopPassManager &LPM,
|
2016-08-03 11:21:41 +08:00
|
|
|
ArrayRef<PipelineElement> Pipeline,
|
2016-02-25 15:23:08 +08:00
|
|
|
bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
2016-08-03 11:21:41 +08:00
|
|
|
for (const auto &Element : Pipeline) {
|
|
|
|
if (!parseLoopPass(LPM, Element, VerifyEachPass, DebugLogging))
|
|
|
|
return false;
|
|
|
|
// FIXME: No verifier support for Loop passes!
|
2016-02-25 15:23:08 +08:00
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
return true;
|
2016-02-25 15:23:08 +08:00
|
|
|
}
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
bool PassBuilder::parseFunctionPassPipeline(FunctionPassManager &FPM,
|
2016-08-03 11:21:41 +08:00
|
|
|
ArrayRef<PipelineElement> Pipeline,
|
2015-03-07 17:02:36 +08:00
|
|
|
bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
2016-08-03 11:21:41 +08:00
|
|
|
for (const auto &Element : Pipeline) {
|
|
|
|
if (!parseFunctionPass(FPM, Element, VerifyEachPass, DebugLogging))
|
|
|
|
return false;
|
|
|
|
if (VerifyEachPass)
|
|
|
|
FPM.addPass(VerifierPass());
|
2014-01-12 17:34:22 +08:00
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
return true;
|
2014-01-12 17:34:22 +08:00
|
|
|
}
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
bool PassBuilder::parseCGSCCPassPipeline(CGSCCPassManager &CGPM,
|
2016-08-03 11:21:41 +08:00
|
|
|
ArrayRef<PipelineElement> Pipeline,
|
2015-03-07 17:02:36 +08:00
|
|
|
bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
2016-08-03 11:21:41 +08:00
|
|
|
for (const auto &Element : Pipeline) {
|
|
|
|
if (!parseCGSCCPass(CGPM, Element, VerifyEachPass, DebugLogging))
|
|
|
|
return false;
|
|
|
|
// FIXME: No verifier support for CGSCC passes!
|
2014-04-21 19:12:00 +08:00
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
return true;
|
2014-04-21 19:12:00 +08:00
|
|
|
}
|
|
|
|
|
2016-05-16 18:13:37 +08:00
|
|
|
void PassBuilder::crossRegisterProxies(LoopAnalysisManager &LAM,
|
|
|
|
FunctionAnalysisManager &FAM,
|
|
|
|
CGSCCAnalysisManager &CGAM,
|
|
|
|
ModuleAnalysisManager &MAM) {
|
|
|
|
MAM.registerPass([&] { return FunctionAnalysisManagerModuleProxy(FAM); });
|
|
|
|
MAM.registerPass([&] { return CGSCCAnalysisManagerModuleProxy(CGAM); });
|
|
|
|
CGAM.registerPass([&] { return ModuleAnalysisManagerCGSCCProxy(MAM); });
|
|
|
|
FAM.registerPass([&] { return CGSCCAnalysisManagerFunctionProxy(CGAM); });
|
|
|
|
FAM.registerPass([&] { return ModuleAnalysisManagerFunctionProxy(MAM); });
|
|
|
|
FAM.registerPass([&] { return LoopAnalysisManagerFunctionProxy(LAM); });
|
|
|
|
LAM.registerPass([&] { return FunctionAnalysisManagerLoopProxy(FAM); });
|
|
|
|
}
|
|
|
|
|
2015-03-07 17:02:36 +08:00
|
|
|
bool PassBuilder::parseModulePassPipeline(ModulePassManager &MPM,
|
2016-08-03 11:21:41 +08:00
|
|
|
ArrayRef<PipelineElement> Pipeline,
|
2015-03-07 17:02:36 +08:00
|
|
|
bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
2016-08-03 11:21:41 +08:00
|
|
|
for (const auto &Element : Pipeline) {
|
|
|
|
if (!parseModulePass(MPM, Element, VerifyEachPass, DebugLogging))
|
|
|
|
return false;
|
|
|
|
if (VerifyEachPass)
|
|
|
|
MPM.addPass(VerifierPass());
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
}
|
2016-08-03 11:21:41 +08:00
|
|
|
return true;
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
// Primary pass pipeline description parsing routine.
|
|
|
|
// FIXME: Should this routine accept a TargetMachine or require the caller to
|
|
|
|
// pre-populate the analysis managers with target-specific stuff?
|
2015-03-07 17:02:36 +08:00
|
|
|
bool PassBuilder::parsePassPipeline(ModulePassManager &MPM,
|
|
|
|
StringRef PipelineText, bool VerifyEachPass,
|
|
|
|
bool DebugLogging) {
|
2016-08-03 11:21:41 +08:00
|
|
|
auto Pipeline = parsePipelineText(PipelineText);
|
|
|
|
if (!Pipeline || Pipeline->empty())
|
|
|
|
return false;
|
|
|
|
|
|
|
|
// If the first name isn't at the module layer, wrap the pipeline up
|
|
|
|
// automatically.
|
|
|
|
StringRef FirstName = Pipeline->front().Name;
|
|
|
|
|
|
|
|
if (!isModulePassName(FirstName)) {
|
|
|
|
if (isCGSCCPassName(FirstName))
|
|
|
|
Pipeline = {{"cgscc", std::move(*Pipeline)}};
|
|
|
|
else if (isFunctionPassName(FirstName))
|
|
|
|
Pipeline = {{"function", std::move(*Pipeline)}};
|
|
|
|
else if (isLoopPassName(FirstName))
|
|
|
|
Pipeline = {{"function", {{"loop", std::move(*Pipeline)}}}};
|
|
|
|
else
|
|
|
|
// Unknown pass name!
|
2014-04-21 19:12:00 +08:00
|
|
|
return false;
|
2014-01-12 17:34:22 +08:00
|
|
|
}
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
|
2016-08-03 11:21:41 +08:00
|
|
|
return parseModulePassPipeline(MPM, *Pipeline, VerifyEachPass, DebugLogging);
|
[PM] Add (very skeletal) support to opt for running the new pass
manager. I cannot emphasize enough that this is a WIP. =] I expect it
to change a great deal as things stabilize, but I think its really
important to get *some* functionality here so that the infrastructure
can be tested more traditionally from the commandline.
The current design is looking something like this:
./bin/opt -passes='module(pass_a,pass_b,function(pass_c,pass_d))'
So rather than custom-parsed flags, there is a single flag with a string
argument that is parsed into the pass pipeline structure. This makes it
really easy to have nice structural properties that are very explicit.
There is one obvious and important shortcut. You can start off the
pipeline with a pass, and the minimal context of pass managers will be
built around the entire specified pipeline. This makes the common case
for tests super easy:
./bin/opt -passes=instcombine,sroa,gvn
But this won't introduce any of the complexity of the fully inferred old
system -- we only ever do this for the *entire* argument, and we only
look at the first pass. If the other passes don't fit in the pass
manager selected it is a hard error.
The other interesting aspect here is that I'm not relying on any
registration facilities. Such facilities may be unavoidable for
supporting plugins, but I have alternative ideas for plugins that I'd
like to try first. My plan is essentially to build everything without
registration until we hit an absolute requirement.
Instead of registration of pass names, there will be a library dedicated
to parsing pass names and the pass pipeline strings described above.
Currently, this is directly embedded into opt for simplicity as it is
very early, but I plan to eventually pull this into a library that opt,
bugpoint, and even Clang can depend on. It should end up as a good home
for things like the existing PassManagerBuilder as well.
There are a bunch of FIXMEs in the code for the parts of this that are
just stubbed out to make the patch more incremental. A quick list of
what's coming up directly after this:
- Support for function passes and building the structured nesting.
- Support for printing the pass structure, and FileCheck tests of all of
this code.
- The .def-file based pass name parsing.
- IR priting passes and the corresponding tests.
Some obvious things that I'm not going to do right now, but am
definitely planning on as the pass manager work gets a bit further:
- Pull the parsing into library, including the builders.
- Thread the rest of the target stuff into the new pass manager.
- Wire support for the new pass manager up to llc.
- Plugin support.
Some things that I'd like to have, but are significantly lower on my
priority list. I'll get to these eventually, but they may also be places
where others want to contribute:
- Adding nice error reporting for broken pass pipeline descriptions.
- Typo-correction for pass names.
llvm-svn: 198998
2014-01-11 16:16:35 +08:00
|
|
|
}
|
2016-02-18 17:45:17 +08:00
|
|
|
|
|
|
|
bool PassBuilder::parseAAPipeline(AAManager &AA, StringRef PipelineText) {
|
2016-12-24 04:38:19 +08:00
|
|
|
// If the pipeline just consists of the word 'default' just replace the AA
|
|
|
|
// manager with our default one.
|
|
|
|
if (PipelineText == "default") {
|
|
|
|
AA = buildDefaultAAPipeline();
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2016-02-18 17:45:17 +08:00
|
|
|
while (!PipelineText.empty()) {
|
|
|
|
StringRef Name;
|
|
|
|
std::tie(Name, PipelineText) = PipelineText.split(',');
|
|
|
|
if (!parseAAPassName(AA, Name))
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|