Break SampleProfileLoader into to a base and a derived class.
Base class (SampleProfileLoaderBaseImpl) includes the common
code for IR and MachineIR (CodeGen) sample loader.
It will be templatelized in the later patch.
Inline and Probe related code will remain in the derived class of
SampleProfileLoader and stays in SampleProfile.cpp.
We need to refactor some functions:
(1) getInstWeight() to enable the code sharing -- put the core into
getInstWeightImpl().
(2) emitAnnotation() and propagateWeights() to carve out the code
specific to SampleProfileLoader.
(3) make getInstWeight() and findFunctionSamples() virtual and override
in SampleProfileLoader as they need to access the fields in the derived
class.
Differential Revision: https://reviews.llvm.org/D95832
The IR/MIR pseudo probe intrinsics don't get materialized into real machine instructions and therefore they don't incur runtime cost directly. However, they come with indirect cost by blocking certain optimizations. Some of the blocking are intentional (such as blocking code merge) for better counts quality while the others are accidental. This change unblocks perf-critical optimizations that do not affect counts quality. They include:
1. IR InstCombine, sinking load operation to shorten lifetimes.
2. MIR LiveRangeShrink, similar to #1
3. MIR TwoAddressInstructionPass, i.e, opeq transform
4. MIR function argument copy elision
5. IR stack protection. (though not perf-critical but nice to have).
Reviewed By: wmi
Differential Revision: https://reviews.llvm.org/D95982
This was reported as PR49104. The reproducer uses varargs but the issue
is the same, we know an argument is dead but can't change the signature
for some reason. The PR49104 situation was: We are in an CG-SCC
traversal and we remove all the uses of an argument and proof it thereby
dead. However, if we do not remove the argument, via signature rewrite,
we need to ensure that the `undef` we introduce at the call site doesn't
clash with a `noundef` attribute.
Type tests used only by assumes were original for devirtualization, but
are meant to be kept through the first invocation of LTT so that they
can be used for additional optimization. In the regular LTO case where
the IR is analyzed we may find a resolution for the type test and end up
rewriting the associated vtable global, which can have implications on
section splitting. Simply ignore these type tests.
Fixes PR48245.
Differential Revision: https://reviews.llvm.org/D96083
This patch detaches SampleProfileLoader from class
SampleCoverageTracker. We plan to move SampleProfileLoader
to a template class. This would remain SampleCoverageTracker
as a class.
Also make callsiteIsHot() as a file static function.
Differential Revision: https://reviews.llvm.org/D95823
Sample re-annotation is required in LTO time to achieve a reasonable post-inline profile quality. However, we have seen that such LTO-time re-annotation degrades profile quality. This is mainly caused by preLTO code duplication that is done by passes such as loop unrolling, jump threading, indirect call promotion etc, where samples corresponding to a source location are aggregated multiple times due to the duplicates. In this change we are introducing a concept of distribution factor for pseudo probes so that samples can be distributed for duplicated probes scaled by a factor. We hope that optimizations duplicating code well-maintain the branch frequency information (BFI) based on which probe distribution factors are calculated. Distribution factors are updated at the end of preLTO pipeline to reflect an estimated portion of the real execution count.
This change also introduces a pseudo probe verifier that can be run after each IR passes to detect duplicated pseudo probes.
A saturated distribution factor stands for 1.0. A pesudo probe will carry a factor with the value ranged from 0.0 to 1.0. A 64-bit integral distribution factor field that represents [0.0, 1.0] is associated to each block probe. Unfortunately this cannot be done for callsite probes due to the size limitation of a 32-bit Dwarf discriminator. A 7-bit distribution factor is used instead.
Changes are also needed to the sample profile inliner to deal with prorated callsite counts. Call sites duplicated by PreLTO passes, when later on inlined in LTO time, should have the callees’s probe prorated based on the Prelink-computed distribution factors. The distribution factors should also be taken into account when computing hotness for inline candidates. Also, Indirect call promotion results in multiple callisites. The original samples should be distributed across them. This is fixed by adjusting the callisites' distribution factors.
Reviewed By: wmi
Differential Revision: https://reviews.llvm.org/D93264
Refactoring SampleProfileLoader::inlineHotFunctions to use helpers from CSSPGO inlining and reduce similar code in the inlining loop, plus minor cleanup for AFDO path.
This is resubmit of D95024, with build break and overtighten assertion fixed.
Test Plan:
Refactoring SampleProfileLoader::inlineHotFunctions to use helpers from CSSPGO inlining and reduce similar code in the inlining loop, plus minor cleanup for AFDO path.
Test Plan:
Differential Revision: https://reviews.llvm.org/D95024
This change implemented call site prioritized BFS profile guided inlining for sample profile loader. The new inlining strategy maximize the benefit of context-sensitive profile as mentioned in the follow up discussion of CSSPGO RFC. The change will not affect today's AutoFDO as it's opt-in. CSSPGO now defaults to the new FDO inliner, but can fall back to today's replay inliner using a switch (`-sample-profile-prioritized-inline=0`).
Motivation
With baseline AutoFDO, the inliner in sample profile loader only replays previous inlining, and the use of profile is only for pruning previous inlining that turned out to be cold. Due to the nature of replay, the FDO inliner is simple with hotness being the only decision factor. It has the following limitations that we're improving now for CSSPGO.
- It doesn't take inline candidate size into account. Since it's doing replay, the size growth is bounded by previous CGSCC inlining. With context-sensitive profile, FDO inliner is no longer limited by previous inlining, so we need to take size into account to avoid significant size bloat.
- The way it looks at hotness is not accurate. It uses total samples in an inlinee as proxy for hotness, while what really matters for an inline decision is the call site count. This is an unfortunate fall back because call site count and callee entry count are not reliable due to dwarf based correlation, especially for inlinees. Now paired with pseudo-probe, we have accurate call site count and callee's entry count, so we can use that to gauge hotness more accurately.
- It treats all call sites from a block as hot as long as there's one call site considered hot. This is normally true, but since total samples is used as hotness proxy, this transitiveness within block magnifies the inacurate hotness heuristic. With pseduo-probe and the change above, this is no longer an issue for CSSPGO.
New FDO Inliner
Putting all the requirement for CSSPGO together, we need a top-down call site prioritized BFS inliner. Here're reasons why each component is needed.
- Top-down: We need a top-down inliner to better leverage context-sensitive profile, so inlining is driven by accurate context profile, and post-inline is also accurate. This is already implemented in https://reviews.llvm.org/D70655.
- Size Cap: For top-down inliner, taking function size into account for inline decision alone isn't sufficient to control size growth. We also need to explicitly cap size growth because with top-down inlining, we can grow inliner size significantly with large number of smaller inlinees even if each individually passes the cost/size check.
- Prioritize call sites: With size cap, inlining order also becomes important, because if we stop inlining due to size budget limit, we'd want to use budget towards the most beneficial call sites.
- BFS inline: Same as call site prioritization, if we stop inlining due to size budget limit, we want a balanced inline tree, rather than going deep on one call path.
Note that the new inliner avoids repeatedly evaluating same set of call site, so it should help with compile time too. For this reason, we could transition today's FDO inliner to use a queue with equal priority to avoid wasted reevaluation of same call site (TODO).
Speculative indirect call promotion and inlining is also supported now with CSSPGO just like baseline AutoFDO.
Tunings and knobs
I created tuning knobs for size growth/cap control, and for hot threshold separate from CGSCC inliner. The default values are selected based on initial tuning with CSSPGO.
Results
Evaluated with an internal LLVM fork couple months ago, plus another change to adjust hot-threshold cutoff for context profile (will send up after this one), the new inliner show ~1% geomean perf win on spec2006 with CSSPGO, while reducing code size too. The measurement was done using train-train setup, MonoLTO w/ new pass manager and pseudo-probe. Note that this is just a starting point - we hope that the new inliner will open up more opportunity with CSSPGO, but it will certainly take more time and effort to make it fully calibrated and ready for bigger workloads (we're working on it).
Differential Revision: https://reviews.llvm.org/D94001
Fixing up a couple places where `getCallSiteIdentifier` is needed to support pseudo-probe-based callsites.
Also fixing an issue in the extbinary profile reader where the metadata section is not fully scanned based on the number of profiles loaded only for the current module.
Reviewed By: wmi, wenlei
Differential Revision: https://reviews.llvm.org/D95791
This change brings up support of context-sensitive profiles in the format of extended binary. Existing sample profile reader/writer/merger code is being tweaked to reflect the fact of bracketed input contexts, like (`[...]`). The paired brackets are also needed in extbinary profiles because we don't yet have an otherwise good way to tell calling contexts apart from regular function names since the context delimiter `@` can somehow serve as a part of the C++ mangled names.
Reviewed By: wmi, wenlei
Differential Revision: https://reviews.llvm.org/D95547
Identify dynamically exported symbols (--export-dynamic[-symbol=],
--dynamic-list=, or definitions needed to preempt shared objects) and
prevent their LTO visibility from being upgraded.
This helps avoid use of whole program devirtualization when there may
be overrides in dynamic libraries.
Differential Revision: https://reviews.llvm.org/D91583
Imported functions and variable get the visibility from the module supplying the
definition. However, non-imported definitions do not get the visibility from
(ELF) the most constraining visibility among all modules (Mach-O) the visibility
of the prevailing definition.
This patch
* adds visibility bits to GlobalValueSummary::GVFlags
* computes the result visibility and propagates it to all definitions
Protected/hidden can imply dso_local which can enable some optimizations (this
is stronger than GVFlags::DSOLocal because the implied dso_local can be
leveraged for ELF -shared while default visibility dso_local has to be cleared
for ELF -shared).
Note: we don't have summaries for declarations, so for ELF if a declaration has
the most constraining visibility, the result visibility may not be that one.
Differential Revision: https://reviews.llvm.org/D92900
As it looks like NewPM generally is using SimpleLoopUnswitch
instead of LoopUnswitch, this patch also use SimpleLoopUnswitch
in the ExtraVectorizerPasses sequence (compared with LegacyPM
which use the LoopUnswitch pass).
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D95457
Loop peeling removes conditions from loop bodies that become invariant
after a small number of iterations. When triggered, this leads to fewer
compares and possibly PHIs in loop bodies, enabling further
optimizations. The current cost-model of loop peeling should be quite
conservative/safe, i.e. only peel if a condition in the loop becomes
known after peeling.
For example, see PR47671, where loop peeling enables vectorization by
removing a PHI the vectorizer does not understand. Granted, the
loop-vectorizer could also be taught about constant PHIs, but loop
peeling is likely to enable other optimizations as well.
This has an impact on quite a few benchmarks from
MultiSource/SPEC2000/SPEC2006 on X86 with -O3 -flto, for example
Same hash: 186 (filtered out)
Remaining: 51
Metric: loop-vectorize.LoopsVectorized
Program base patch diff
test-suite...ve-susan/automotive-susan.test 8.00 9.00 12.5%
test-suite...nal/skidmarks10/skidmarks.test 35.00 31.00 -11.4%
test-suite...lications/sqlite3/sqlite3.test 41.00 43.00 4.9%
test-suite...s/ASC_Sequoia/AMGmk/AMGmk.test 25.00 26.00 4.0%
test-suite...006/450.soplex/450.soplex.test 88.00 89.00 1.1%
test-suite...TimberWolfMC/timberwolfmc.test 120.00 119.00 -0.8%
test-suite.../CINT2006/403.gcc/403.gcc.test 215.00 216.00 0.5%
test-suite...006/447.dealII/447.dealII.test 957.00 958.00 0.1%
test-suite...ternal/HMMER/hmmcalibrate.test 75.00 75.00 0.0%
Same hash: 186 (filtered out)
Remaining: 51
Metric: loop-vectorize.LoopsAnalyzed
Program base patch diff
test-suite...ks/Prolangs-C/agrep/agrep.test 440.00 434.00 -1.4%
test-suite...nal/skidmarks10/skidmarks.test 312.00 308.00 -1.3%
test-suite...marks/7zip/7zip-benchmark.test 6399.00 6323.00 -1.2%
test-suite...lications/minisat/minisat.test 134.00 135.00 0.7%
test-suite...rks/FreeBench/pifft/pifft.test 295.00 297.00 0.7%
test-suite...TimberWolfMC/timberwolfmc.test 1879.00 1869.00 -0.5%
test-suite...pplications/treecc/treecc.test 689.00 691.00 0.3%
test-suite...T2000/300.twolf/300.twolf.test 1593.00 1597.00 0.3%
test-suite.../Benchmarks/Bullet/bullet.test 1394.00 1392.00 -0.1%
test-suite...ications/JM/ldecod/ldecod.test 1431.00 1429.00 -0.1%
test-suite...6/464.h264ref/464.h264ref.test 2229.00 2230.00 0.0%
test-suite...lications/sqlite3/sqlite3.test 2590.00 2589.00 -0.0%
test-suite...ications/JM/lencod/lencod.test 2732.00 2733.00 0.0%
test-suite...006/453.povray/453.povray.test 3395.00 3394.00 -0.0%
Note the -11% regression in number of loops vectorized for skidmarks. I
suspect this corresponds to the fact that those loops are gone now (see
the reduction in number of loops analyzed by LV).
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D88471
This change leverages the work done in D83743 to replay in the SampleProfile inliner to also be used in the CGSCC inliner. NOTE: currently restricted to non-ML advisors only.
The added switch `-cgscc-inline-replay=<remarks file>` will replay the inlining decisions in that file where the remarks file is generated via `-Rpass=inline`. The aim here is to make it easier to analyze changes that would modify inlining heuristics to be separated from this behavior. Doing so allows easier examination of assembly and runtime behavior compared to the baseline rather than trying to dig through the large churn caused by inlining.
In LTO compilation, since inlining is done twice you can separately specify replay by passing the flag to the FE (`-cgscc-inline-replay=`) and to the linker (`-Wl,cgscc-inline-replay=`) with the remarks generated from their respective places.
Testing on mysqld by comparing the inline decisions between base (generates remarks.txt) and diff (replay using identical input/tools with remarks.txt) and examining the inlining sites with `diff` shows 14,000 mismatches out of 247,341 for a ~94% replay accuracy. I believe this gap can be narrowed further though for the general case we may never achieve full accuracy. For my personal use, this is close enough to be representative: I set the baseline as the one generated by the replay on identical input/toolset and compare that to my modified input/toolset using the same replay.
Testing:
ninja check-llvm
newly added test correctly replays CGSCC inlining decisions
Reviewed By: mtrofin, wenlei
Differential Revision: https://reviews.llvm.org/D94334
This reverts commit 53176c168061d6f26dcf3ce4fa59288b7d67255e, which
introduceed a layering violation. LLVM's IR library can't include
headers from Analysis.
or claimRV calls in the IR
Background:
This patch makes changes to the front-end and middle-end that are
needed to fix a longstanding problem where llvm breaks ARC's autorelease
optimization (see the link below) by separating calls from the marker
instructions or retainRV/claimRV calls. The backend changes are in
https://reviews.llvm.org/D92569.
https://clang.llvm.org/docs/AutomaticReferenceCounting.html#arc-runtime-objc-autoreleasereturnvalue
What this patch does to fix the problem:
- The front-end annotates calls with attribute "clang.arc.rv"="retain"
or "clang.arc.rv"="claim", which indicates the call is implicitly
followed by a marker instruction and a retainRV/claimRV call that
consumes the call result. This is currently done only when the target
is arm64 and the optimization level is higher than -O0.
- ARC optimizer temporarily emits retainRV/claimRV calls after the
annotated calls in the IR and removes the inserted calls after
processing the function.
- ARC contract pass emits retainRV/claimRV calls after the annotated
calls. It doesn't remove the attribute on the call since the backend
needs it to emit the marker instruction. The retainRV/claimRV calls
are emitted late in the pipeline to prevent optimization passes from
transforming the IR in a way that makes it harder for the ARC
middle-end passes to figure out the def-use relationship between the
call and the retainRV/claimRV calls (which is the cause of PR31925).
- The function inliner removes the autoreleaseRV call in the callee that
returns the result if nothing in the callee prevents it from being
paired up with the calls annotated with "clang.arc.rv"="retain/claim"
in the caller. If the call is annotated with "claim", a release call
is inserted since autoreleaseRV+claimRV is equivalent to a release. If
it cannot find an autoreleaseRV call, it tries to transfer the
attributes to a function call in the callee. This is important since
ARC optimizer can remove the autoreleaseRV call returning the callee
result, which makes it impossible to pair it up with the retainRV or
claimRV call in the caller. If that fails, it simply emits a retain
call in the IR if the call is annotated with "retain" and does nothing
if it's annotated with "claim".
- This patch teaches dead argument elimination pass not to change the
return type of a function if any of the calls to the function are
annotated with attribute "clang.arc.rv". This is necessary since the
pass can incorrectly determine nothing in the IR uses the function
return, which can happen since the front-end no longer explicitly
emits retainRV/claimRV calls in the IR, and change its return type to
'void'.
Future work:
- Use the attribute on x86-64.
- Fix the auto upgrader to convert call+retainRV/claimRV pairs into
calls annotated with the attributes.
rdar://71443534
Differential Revision: https://reviews.llvm.org/D92808
turning off SampleFDO silently.
Currently sample loader pass turns off SampleFDO optimization silently when
it sees error in reading the profile. This behavior will defeat the tests
which could have caught those bad/incompatible profile problems. This patch
change the behavior to report error.
Differential Revision: https://reviews.llvm.org/D95269
If a function doesn't contain loops and does not call non-willreturn
functions, then it is willreturn. Loops are detected by checking
for backedges in the function. We don't attempt to handle finite
loops at this point.
Differential Revision: https://reviews.llvm.org/D94633
This reverts commit d97f776be5f8cd3cd446fe73827cd355f6bab4e1.
The original problem was due to build failures in shared lib builds. D95079
moved ImportedFunctionsInliningStatistics under Analysis, unblocking
this.
This is related to D94982. We want to call these APIs from the Analysis
component, so we can't leave them under Transforms.
Differential Revision: https://reviews.llvm.org/D95079
When using 2 InlinePass instances in the same CGSCC - one for other
mandatory inlinings, the other for the heuristic-driven ones - the order
in which the ImportedFunctionStats would be output-ed would depend on
the destruction order of the inline passes, which is not deterministic.
This patch moves the ImportedFunctionStats responsibility to the
InlineAdvisor to address this problem.
Differential Revision: https://reviews.llvm.org/D94982
Currently LLVM is relying on ValueTracking's `isKnownNonZero` to attach `nonnull`, which can return true when the value is poison.
To make the semantics of `nonnull` consistent with the behavior of `isKnownNonZero`, this makes the semantics of `nonnull` to accept poison, and return poison if the input pointer isn't null.
This makes many transformations like below legal:
```
%p = gep inbounds %x, 1 ; % p is non-null pointer or poison
call void @f(%p) ; instcombine converts this to call void @f(nonnull %p)
```
Instead, this semantics makes propagation of `nonnull` to caller illegal.
The reason is that, passing poison to `nonnull` does not immediately raise UB anymore, so such program is still well defined, if the callee does not use the argument.
Having `noundef` attribute there re-allows this.
```
define void @f(i8* %p) { ; functionattr cannot mark %p nonnull here anymore
call void @g(i8* nonnull %p) ; .. because @g never raises UB if it never uses %p.
ret void
}
```
Another attribute that needs to be updated is `align`. This patch updates the semantics of align to accept poison as well.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D90529
separate sections.
For ThinLTO, all the function profiles without context has been annotated to
outline functions if possible in prelink phase. In postlink phase, profile
annotation in postlink phase is only meaningful for function profile with
context. If the profile is large, it is better to split the profile into two
parts, one with context and one without, so the profile reading in postlink
phase only has to read the part with context. To have the profile splitting,
we extend the ExtBinary format to support different section arrangement. It
will be flexible to add other section layout in the future without the need
to create new class inheriting from ExtBinary class.
Differential Revision: https://reviews.llvm.org/D94435
D84108 exposed a bad interaction between inlining and loop-rotation
during regular LTO, which is causing notable regressions in at least
CINT2006/473.astar.
The problem boils down to: we now rotate a loop just before the vectorizer
which requires duplicating a function call in the preheader when compiling
the individual files ('prepare for LTO'). But this then prevents further
inlining of the function during LTO.
This patch tries to resolve this issue by making LoopRotate more
conservative with respect to rotating loops that have inline-able calls
during the 'prepare for LTO' stage.
I think this change intuitively improves the current situation in
general. Loop-rotate tries hard to avoid creating headers that are 'too
big'. At the moment, it assumes all inlining already happened and the
cost of duplicating a call is equal to just doing the call. But with LTO,
inlining also happens during full LTO and it is possible that a previously
duplicated call is actually a huge function which gets inlined
during LTO.
From the perspective of LV, not much should change overall. Most loops
calling user-provided functions won't get vectorized to start with
(unless we can infer that the function does not touch memory, has no
other side effects). If we do not inline the 'inline-able' call during
the LTO stage, we merely delayed loop-rotation & vectorization. If we
inline during LTO, chances should be very high that the inlined code is
itself vectorizable or the user call was not vectorizable to start with.
There could of course be scenarios where we inline a sufficiently large
function with code not profitable to vectorize, which would have be
vectorized earlier (by scalarzing the call). But even in that case,
there probably is no big performance impact, because it should be mostly
down to the cost-model to reject vectorization in that case. And then
the version with scalarized calls should also not be beneficial. In a way,
LV should have strictly more information after inlining and make more
accurate decisions (barring cost-model issues).
There is of course plenty of room for things to go wrong unexpectedly,
so we need to keep a close look at actual performance and address any
follow-up issues.
I took a look at the impact on statistics for
MultiSource/SPEC2000/SPEC2006. There are a few benchmarks with fewer
loops rotated, but no change to the number of loops vectorized.
Reviewed By: sanwou01
Differential Revision: https://reviews.llvm.org/D94232
Expanding from D94808 - we ensure the same InlineAdvisor is used by both
InlinerPass instances. The notion of mandatory inlining is moved into
the core InlineAdvisor: advisors anyway have to handle that case, so
this change also factors out that a bit better.
Differential Revision: https://reviews.llvm.org/D94825
to Pass.h.
In some compiler passes like SampleProfileLoaderPass, we want to know which
LTO/ThinLTO phase the pass is in. Currently the phase is represented in enum
class PassBuilder::ThinLTOPhase, so it is only available in PassBuilder and
it also cannot represent phase in full LTO. The patch extends it to include
full LTO phases and move it from PassBuilder.h to Pass.h, then it is much
easier for PassBuilder to communiate with each pass about current LTO phase.
Differential Revision: https://reviews.llvm.org/D94613
