This reverts commit bd7949bcd8.
Revert this patch since reviwers have different opinions regarding
the approach in post-commit review.
Will open RFC for further discussion.
Differential Revision: https://reviews.llvm.org/D132408
SimplifyCFG folds
bool foo() {
if (cond1) return false;
if (cond2) return false;
return true;
}
as
bool foo() {
if (cond1 | cond2) return false
return true;
}
'cond2' is called 'bonus insts' in branch folding since they introduce overhead
since the original CFG could do early exit but the folded CFG always executes
them. SimplifyCFG calculates the costs of 'bonus insts' of a folding a BB into
its predecessor BB which shares the destination. If it is below bonus-inst-threshold,
SimplifyCFG will fold that BB into its predecessor and cond2 will always be executed.
When SimplifyCFG calculates the cost of 'bonus insts', it only consider 'bonus' insts
in the current BB to be considered for folding. This causes issue for unrolled loops
which share destinations, e.g.
bool foo(int *a) {
for (int i = 0; i < 32; i++)
if (a[i] > 0) return false;
return true;
}
After unrolling, it becomes
bool foo(int *a) {
if(a[0]>0) return false
if(a[1]>0) return false;
//...
if(a[31]>0) return false;
return true;
}
SimplifyCFG will merge each BB with its predecessor BB,
and ends up with 32 'bonus insts' which are always executed, which
is much slower than the original CFG.
The root cause is that SimplifyCFG does not consider the
accumulated cost of 'bonus insts' which are folded from
different BB's.
This patch fixes that by introducing a ValueMap to track
costs of 'bonus insts' coming from different BB's into
the same BB, and cuts off if the accumulated cost
exceeds a threshold.
Reviewed by: Artem Belevich, Florian Hahn, Nikita Popov, Matt Arsenault
Differential Revision: https://reviews.llvm.org/D132408
We observe a hang within iterativelySimplifyCFG due to infinite
loop execution. Currently, there is no limit to this loop, so
in case of bug it just works forever. This patch adds an assert
that will break it after 1000 iterations if it didn't converge.
Added opt option -print-pipeline-passes to print a -passes compatible
string describing the built pass pipeline.
As an example:
$ opt -enable-new-pm=1 -adce -licm -simplifycfg -o /dev/null /dev/null -print-pipeline-passes
verify,function(adce),function(loop-mssa(licm)),function(simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts>),verify,BitcodeWriterPass
At the moment this is best-effort only and there are some known
limitations:
- Not all passes accepting parameters will print their parameters
(currently only implemented for simplifycfg).
- Some ClassName to pass-name mappings are not unique.
- Some ClassName to pass-name mappings are missing (e.g.
BitcodeWriterPass).
Differential Revision: https://reviews.llvm.org/D108298
Added opt option -print-pipeline-passes to print a -passes compatible
string describing the built pass pipeline.
As an example:
$ opt -enable-new-pm=1 -adce -licm -simplifycfg -o /dev/null /dev/null -print-pipeline-passes
verify,function(adce),function(loop-mssa(licm)),function(simplifycfg<bonus-inst-threshold=1;no-forward-switch-cond;no-switch-to-lookup;keep-loops;no-hoist-common-insts;no-sink-common-insts>),verify,BitcodeWriterPass
At the moment this is best-effort only and there are some known
limitations:
- Not all passes accepting parameters will print their parameters
(currently only implemented for simplifycfg).
- Some ClassName to pass-name mappings are not unique.
- Some ClassName to pass-name mappings are missing (e.g.
BitcodeWriterPass).
Similar to what we already do for `ret` terminators.
As noted by @rnk, clang seems to already generate a single `ret`/`resume`,
so this isn't likely to cause widespread changes.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D104849
Based ontop of D104598, which is a NFCI-ish refactoring.
Here, a restriction, that only empty blocks can be merged, is lifted.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D104597
This changes the approach taken to tail-merge the blocks
to always create a new block instead of trying to reuse some block,
and generalizes it to support dealing not with just the `ret` in the future.
This effectively lifts the CallBr restriction, although this isn't really intentional.
That is the only non-NFC change here, i'm not sure if it's reasonable/feasible to temporarily retain it.
Other restrictions of the transform remain.
Reviewed By: rnk
Differential Revision: https://reviews.llvm.org/D104598
In this case, it does the same thing as the original pattern does.
SimplifyCFG has a few lurking miscompilations about deleting blocks that
have their address taken, and consistently using DeleteDeadBlocks() instead
of a hand-rolled pattern will allow to weed those cases out easierly.
Currently all AA analyses marked as preserved are stateless, not taking
into account their dependent analyses. So there's no need to mark them
as preserved, they won't be invalidated unless their analyses are.
SCEVAAResults was the one exception to this, it was treated like a
typical analysis result. Make it like the others and don't invalidate
unless SCEV is invalidated.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D102032
If i change it to AssertingVH instead, a number of existing tests fail,
which means we don't consistently remove from the set when deleting blocks,
which means newly-created blocks may happen to appear in that set
if they happen to occupy the same memory chunk as did some block
that was in the set originally.
There are many places where we delete blocks,
and while we could probably consistently delete from LoopHeaders
when deleting a block in transforms located in SimplifyCFG.cpp itself,
transforms located elsewhere (Local.cpp/BasicBlockUtils.cpp) also may
delete blocks, and it doesn't seem good to teach them to deal with it.
Since we at most only ever delete from LoopHeaders,
let's just delegate to WeakVH to do that automatically.
But to be honest, personally, i'm not sure that the idea
behind LoopHeaders is sound.
This boils down to how we deal with early-increment iterator
over function's basic blocks: not only we need to early-increment,
after that we also need to skip all the blocks
that are scheduled for removal, as per DomTreeUpdater.
Thus supporting lazy DomTreeUpdater mode,
where the domtree updates (and thus block removals)
aren't applied immediately, but are delayed
until last possible moment.
Notably, this doesn't switch *every* case, remaining cases
don't actually pass sanity checks in non-permissve mode,
and therefore require further analysis.
Note that SimplifyCFG still defaults to not preserving DomTree by default,
so this is effectively a NFC change.
A first real transformation that didn't already knew how to do that,
but it's pretty tame - either change successor of all the predecessors
of a block and carefully delay deletion of the block until afterwards
the DomTree updates are appled, or add a successor to the block.
There wasn't a great test coverage for this, so i added extra, to be sure.
... so just ensure that we pass DomTreeUpdater it into it.
Fixes DomTree preservation for a large number of tests,
all of which are marked as such so that they do not regress.
... so just ensure that we pass DomTreeUpdater it into it.
Apparently, there were no dedicated tests just for that functionality,
so i'm adding one here.
Two observations:
1. Unavailability of DomTree makes it impossible to make
`FoldBranchToCommonDest()` transform in certain cases,
where the successor is dominated by predecessor,
because we then don't have PHI's, and can't recreate them,
well, without handrolling 'is dominated by' check,
which doesn't really look like a great solution to me.
2. Avoiding invalidating DomTree in SimplifyCFG will
decrease the number of `Dominator Tree Construction` by 5
(from 28 now, i.e. -18%) in `-O3` old-pm pipeline
(as per `llvm/test/Other/opt-O3-pipeline.ll`)
This might or might not be beneficial for compile time.
So the plan is to make SimplifyCFG preserve DomTree, and then
eventually make DomTree fully required and preserved by the pass.
Now, SimplifyCFG is ~7KLOC. I don't think it will be nice
to do all this uplifting in a single mega-commit,
nor would it be possible to review it in any meaningful way.
But, i believe, it should be possible to do this in smaller steps,
introducing the new behavior, in an optional way, off-by-default,
opt-in option, and gradually fixing transforms one-by-one
and adding the flag to appropriate test coverage.
Then, eventually, the default should be flipped,
and eventually^2 the flag removed.
And that is what is happening here - when the new off-by-default option
is specified, DomTree is required and is claimed to be preserved,
and SimplifyCFG-internal assertions verify that the DomTree is still OK.
SimplifyCFG's options should always be overridden by command line flags,
but they mistakenly weren't in the default constructor.
Reviewed By: ychen
Differential Revision: https://reviews.llvm.org/D87718
This was reverted in 503deec218
because it caused gigantic increase (3x) in branch mispredictions
in certain benchmarks on certain CPU's,
see https://reviews.llvm.org/D84108#2227365.
It has since been investigated and here are the results:
https://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20200907/827578.html
> It's an amazingly severe regression, but it's also all due to branch
> mispredicts (about 3x without this). The code layout looks ok so there's
> probably something else to deal with. I'm not sure there's anything we can
> reasonably do so we'll just have to take the hit for now and wait for
> another code reorganization to make the branch predictor a bit more happy :)
>
> Thanks for giving us some time to investigate and feel free to recommit
> whenever you'd like.
>
> -eric
So let's just reland this.
Original commit message:
I've been looking at missed vectorizations in one codebase.
One particular thing that stands out is that some of the loops
reach vectorizer in a rather mangled form, with weird PHI's,
and some of the loops aren't even in a rotated form.
After taking a more detailed look, that happened because
the loop's headers were too big by then. It is evident that
SimplifyCFG's common code hoisting transform is at fault there,
because the pattern it handles is precisely the unrotated
loop basic block structure.
Surprizingly, `SimplifyCFGOpt::HoistThenElseCodeToIf()` is enabled
by default, and is always run, unlike it's friend, common code sinking
transform, `SinkCommonCodeFromPredecessors()`, which is not enabled
by default and is only run once very late in the pipeline.
I'm proposing to harmonize this, and disable common code hoisting
until //late// in pipeline. Definition of //late// may vary,
here currently i've picked the same one as for code sinking,
but i suppose we could enable it as soon as right after
loop rotation happens.
Experimentation shows that this does indeed unsurprizingly help,
more loops got rotated, although other issues remain elsewhere.
Now, this undoubtedly seriously shakes phase ordering.
This will undoubtedly be a mixed bag in terms of both compile- and
run- time performance, codesize. Since we no longer aggressively
hoist+deduplicate common code, we don't pay the price of said hoisting
(which wasn't big). That may allow more loops to be rotated,
so we pay that price. That, in turn, that may enable all the transforms
that require canonical (rotated) loop form, including but not limited to
vectorization, so we pay that too. And in general, no deduplication means
more [duplicate] instructions going through the optimizations. But there's still
late hoisting, some of them will be caught late.
As per benchmarks i've run {F12360204}, this is mostly within the noise,
there are some small improvements, some small regressions.
One big regression i saw i fixed in rG8d487668d09fb0e4e54f36207f07c1480ffabbfd, but i'm sure
this will expose many more pre-existing missed optimizations, as usual :S
llvm-compile-time-tracker.com thoughts on this:
http://llvm-compile-time-tracker.com/compare.php?from=e40315d2b4ed1e38962a8f33ff151693ed4ada63&to=c8289c0ecbf235da9fb0e3bc052e3c0d6bff5cf9&stat=instructions
* this does regress compile-time by +0.5% geomean (unsurprizingly)
* size impact varies; for ThinLTO it's actually an improvement
The largest fallout appears to be in GVN's load partial redundancy
elimination, it spends *much* more time in
`MemoryDependenceResults::getNonLocalPointerDependency()`.
Non-local `MemoryDependenceResults` is widely-known to be, uh, costly.
There does not appear to be a proper solution to this issue,
other than silencing the compile-time performance regression
by tuning cut-off thresholds in `MemoryDependenceResults`,
at the cost of potentially regressing run-time performance.
D84609 attempts to move in that direction, but the path is unclear
and is going to take some time.
If we look at stats before/after diffs, some excerpts:
* RawSpeed (the target) {F12360200}
* -14 (-73.68%) loops not rotated due to the header size (yay)
* -272 (-0.67%) `"Number of live out of a loop variables"` - good for vectorizer
* -3937 (-64.19%) common instructions hoisted
* +561 (+0.06%) x86 asm instructions
* -2 basic blocks
* +2418 (+0.11%) IR instructions
* vanilla test-suite + RawSpeed + darktable {F12360201}
* -36396 (-65.29%) common instructions hoisted
* +1676 (+0.02%) x86 asm instructions
* +662 (+0.06%) basic blocks
* +4395 (+0.04%) IR instructions
It is likely to be sub-optimal for when optimizing for code size,
so one might want to change tune pipeline by enabling sinking/hoisting
when optimizing for size.
Reviewed By: mkazantsev
Differential Revision: https://reviews.llvm.org/D84108
This reverts commit 503deec218.
As disscussed in post-commit review starting with
https://reviews.llvm.org/D84108#2227365
while this appears to be mostly a win overall, especially code-size-wise,
this appears to shake //certain// code pattens in a way that is extremely
unfavorable for performance (+30% runtime regression)
on certain CPU's (i personally can't reproduce).
So until the behaviour is better understood, and a path forward is mapped,
let's back this out for now.
This reverts commit 1d51dc38d8.
I've been looking at missed vectorizations in one codebase.
One particular thing that stands out is that some of the loops
reach vectorizer in a rather mangled form, with weird PHI's,
and some of the loops aren't even in a rotated form.
After taking a more detailed look, that happened because
the loop's headers were too big by then. It is evident that
SimplifyCFG's common code hoisting transform is at fault there,
because the pattern it handles is precisely the unrotated
loop basic block structure.
Surprizingly, `SimplifyCFGOpt::HoistThenElseCodeToIf()` is enabled
by default, and is always run, unlike it's friend, common code sinking
transform, `SinkCommonCodeFromPredecessors()`, which is not enabled
by default and is only run once very late in the pipeline.
I'm proposing to harmonize this, and disable common code hoisting
until //late// in pipeline. Definition of //late// may vary,
here currently i've picked the same one as for code sinking,
but i suppose we could enable it as soon as right after
loop rotation happens.
Experimentation shows that this does indeed unsurprizingly help,
more loops got rotated, although other issues remain elsewhere.
Now, this undoubtedly seriously shakes phase ordering.
This will undoubtedly be a mixed bag in terms of both compile- and
run- time performance, codesize. Since we no longer aggressively
hoist+deduplicate common code, we don't pay the price of said hoisting
(which wasn't big). That may allow more loops to be rotated,
so we pay that price. That, in turn, that may enable all the transforms
that require canonical (rotated) loop form, including but not limited to
vectorization, so we pay that too. And in general, no deduplication means
more [duplicate] instructions going through the optimizations. But there's still
late hoisting, some of them will be caught late.
As per benchmarks i've run {F12360204}, this is mostly within the noise,
there are some small improvements, some small regressions.
One big regression i saw i fixed in rG8d487668d09fb0e4e54f36207f07c1480ffabbfd, but i'm sure
this will expose many more pre-existing missed optimizations, as usual :S
llvm-compile-time-tracker.com thoughts on this:
http://llvm-compile-time-tracker.com/compare.php?from=e40315d2b4ed1e38962a8f33ff151693ed4ada63&to=c8289c0ecbf235da9fb0e3bc052e3c0d6bff5cf9&stat=instructions
* this does regress compile-time by +0.5% geomean (unsurprizingly)
* size impact varies; for ThinLTO it's actually an improvement
The largest fallout appears to be in GVN's load partial redundancy
elimination, it spends *much* more time in
`MemoryDependenceResults::getNonLocalPointerDependency()`.
Non-local `MemoryDependenceResults` is widely-known to be, uh, costly.
There does not appear to be a proper solution to this issue,
other than silencing the compile-time performance regression
by tuning cut-off thresholds in `MemoryDependenceResults`,
at the cost of potentially regressing run-time performance.
D84609 attempts to move in that direction, but the path is unclear
and is going to take some time.
If we look at stats before/after diffs, some excerpts:
* RawSpeed (the target) {F12360200}
* -14 (-73.68%) loops not rotated due to the header size (yay)
* -272 (-0.67%) `"Number of live out of a loop variables"` - good for vectorizer
* -3937 (-64.19%) common instructions hoisted
* +561 (+0.06%) x86 asm instructions
* -2 basic blocks
* +2418 (+0.11%) IR instructions
* vanilla test-suite + RawSpeed + darktable {F12360201}
* -36396 (-65.29%) common instructions hoisted
* +1676 (+0.02%) x86 asm instructions
* +662 (+0.06%) basic blocks
* +4395 (+0.04%) IR instructions
It is likely to be sub-optimal for when optimizing for code size,
so one might want to change tune pipeline by enabling sinking/hoisting
when optimizing for size.
Reviewed By: mkazantsev
Differential Revision: https://reviews.llvm.org/D84108
Common code sinking is already guarded with a (with default-off!) flag,
so add a flag for hoisting, too.
D84108 will hopefully make hoisting off-by-default too.
This reverts commit 1067d3e176,
which reverted commit b2018198c3,
because it introduced a Dependency Cycle between Transforms/Scalar and
Transforms/Utils.
So let's just move SimplifyCFGOptions.h into Utils/, thus avoiding
the cycle.
This reverts commit b2018198c3.
This commit introduced a Dependency Cycle between Transforms/Scalar and
Transforms/Utils. Transforms/Scalar already depends on Transforms/Utils,
so if SimplifyCFGOptions.h is moved to Scalar, and Utils/Local.h still
depends on it, we have a cycle.
Taking so many parameters is simply unmaintainable.
We don't want to include the entire llvm/Transforms/Utils/Local.h into
llvm/Transforms/Scalar.h so i've split SimplifyCFGOptions into
it's own header.
SimplifyCFG should not merge empty return blocks and leave a CallBr behind
with a duplicated destination since the verifier will then trigger an
assert. This patch checks for this case and avoids the transformation.
CodeGenPrepare has a similar check which also has a FIXME comment about why
this is needed. It seems perhaps better if these two passes would eventually
instead update the CallBr instruction instead of just checking and avoiding.
This fixes https://bugs.llvm.org/show_bug.cgi?id=45062.
Review: Craig Topper
Differential Revision: https://reviews.llvm.org/D75620
This file lists every pass in LLVM, and is included by Pass.h, which is
very popular. Every time we add, remove, or rename a pass in LLVM, it
caused lots of recompilation.
I found this fact by looking at this table, which is sorted by the
number of times a file was changed over the last 100,000 git commits
multiplied by the number of object files that depend on it in the
current checkout:
recompiles touches affected_files header
342380 95 3604 llvm/include/llvm/ADT/STLExtras.h
314730 234 1345 llvm/include/llvm/InitializePasses.h
307036 118 2602 llvm/include/llvm/ADT/APInt.h
213049 59 3611 llvm/include/llvm/Support/MathExtras.h
170422 47 3626 llvm/include/llvm/Support/Compiler.h
162225 45 3605 llvm/include/llvm/ADT/Optional.h
158319 63 2513 llvm/include/llvm/ADT/Triple.h
140322 39 3598 llvm/include/llvm/ADT/StringRef.h
137647 59 2333 llvm/include/llvm/Support/Error.h
131619 73 1803 llvm/include/llvm/Support/FileSystem.h
Before this change, touching InitializePasses.h would cause 1345 files
to recompile. After this change, touching it only causes 550 compiles in
an incremental rebuild.
Reviewers: bkramer, asbirlea, bollu, jdoerfert
Differential Revision: https://reviews.llvm.org/D70211
to reflect the new license.
We understand that people may be surprised that we're moving the header
entirely to discuss the new license. We checked this carefully with the
Foundation's lawyer and we believe this is the correct approach.
Essentially, all code in the project is now made available by the LLVM
project under our new license, so you will see that the license headers
include that license only. Some of our contributors have contributed
code under our old license, and accordingly, we have retained a copy of
our old license notice in the top-level files in each project and
repository.
llvm-svn: 351636
Review feedback from r328165. Split out just the one function from the
file that's used by Analysis. (As chandlerc pointed out, the original
change only moved the header and not the implementation anyway - which
was fine for the one function that was used (since it's a
template/inlined in the header) but not in general)
llvm-svn: 333954
Yaxun (Sam) Liu