Currently we model subvector inserts and extracts as shuffles,
potentially going as far as scalarizing. If the types are legal then
they can just be simple zip/unzip operations, or possible even no-ops.
Change the cost to a relatively small one to ensure that simple loops
featuring such operations between fixed and scalable vector types that
are effectively the same at a given sve width can be unrolled and
further optimized.
There are many tests that specify a target triple/CPU flags but no
DataLayout which can lead to IR being generated that has unusual
behaviour. This commit attempts to use the default DataLayout based
on the relevant flags if there is no explicit override on the command
line or in the IR file.
One thing that is not currently possible to differentiate from a missing
datalayout `target datalayout = ""` in the IR file since the current
APIs don't allow detecting this case. If it is considered useful to
support this case (instead of passing "-data-layout=" on the command
line), I can change IR parsers to track whether they have seen such a
directive and change the callback type.
Differential Revision: https://reviews.llvm.org/D141060
Loop unrolling tends to produce chains of
`%x1 = add %x0, 1; %x2 = add %x1, 1; ...` with one add per unrolled
iteration. This patch simplifies these adds to `%xN = add %x0, N`
directly during unrolling, rather than waiting for InstCombine to do so.
The motivation for this is that having a single add (rather than
an add chain) on the induction variable makes it a simple recurrence,
which we specially recognize in a number of places. This allows
InstCombine to directly perform folds with that knowledge, instead
of first folding the add chains, and then doing other folds in another
InstCombine iteration.
Due to the reduced number of InstCombine iterations, this also
results in a small compile-time improvement.
Differential Revision: https://reviews.llvm.org/D153540
Doesn't really matter for the larger purpose of the test, but
avoid the use of undef indices and instead use the loop induction
variable as index, which is what was likely intended here.
This is a follow-up to b71edfaa4ec3c998aadb35255ce2f60bba2940b0
since I forgot the lit.local.cfg files in that one.
Reformatting is done with `black`.
If you end up having problems merging this commit because you
have made changes to a python file, the best way to handle that
is to run git checkout --ours <yourfile> and then reformat it
with black.
If you run into any problems, post to discourse about it and
we will try to help.
RFC Thread below:
https://discourse.llvm.org/t/rfc-document-and-standardize-python-code-style
Reviewed By: barannikov88, kwk
Differential Revision: https://reviews.llvm.org/D150762
LICM will speculatively hoist code outside of loops. This requires removing information, like alias analysis (https://github.com/llvm/llvm-project/issues/53794), range information (https://bugs.llvm.org/show_bug.cgi?id=50550), among others. Prior to https://reviews.llvm.org/D99249 , LICM would only be run after LoopRotate. Running Loop Rotate prior to LICM prevents a instruction hoist from being speculative, if it was conditionally executed by the iteration (as is commonly emitted by clang and other frontends). Adding the additional LICM pass first, however, forces all of these instructions to be considered speculative, even if they are not speculative after LoopRotate. This destroys information, resulting in performance losses for discarding this additional information.
This PR modifies LICM to accept a ``speculative'' parameter which allows LICM to be set to perform information-loss speculative hoists or not. Phase ordering is then modified to not perform the information-losing speculative hoists until after loop rotate is performed, preserving this additional information.
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D119965
This is one of those wonderful "in theory X doesn't matter, but in practice is does" changes. In this particular case, we shift the IVs inserted by the runtime unroller to clamp iteration count of the loops* from decrementing to incrementing.
Why does this matter? A couple of reasons:
* SCEV doesn't have a native subtract node. Instead, all subtracts (A - B) are represented as A + -1 * B and drops any flags invalidated by such. As a result, SCEV is slightly less good at reasoning about edge cases involving decrementing addrecs than incrementing ones. (You can see this in the inferred flags in some of the test cases.)
* Other parts of the optimizer produce incrementing IVs, and they're common in idiomatic source language. We do have support for reversing IVs, but in general if we produce one of each, the pair will persist surprisingly far through the optimizer before being coalesced. (You can see this looking at nearby phis in the test cases.)
Note that if the hardware prefers decrementing (i.e. zero tested) loops, LSR should convert back immediately before codegen.
* Mostly irrelevant detail: The main loop of the prolog case is handled independently and will simple use the original IV with a changed start value. We could in theory use this scheme for all iteration clamping, but that's a larger and more invasive change.
The unrolling code was previously inserting new cloned blocks at the end of the function. The result of this with typical loop structures is that the new iterations are placed far from the initial iteration.
With unrolling, the general assumption is that the a) the loop is reasonable hot, and b) the first Count-1 copies of the loop are rarely (if ever) loop exiting. As such, placing Count-1 copies out of line is a fairly poor code placement choice. We'd much rather fall through into the hot (non-exiting) path. For code with branch profiles, later layout would fix this, but this may have a positive impact on non-PGO compiled code.
However, the real motivation for this change isn't performance. Its readability and human understanding. Having to jump around long distances in an IR file to trace an unrolled loop structure is error prone and tedious.
Addition of this pass has been botched.
There is no particular reason why it had to be sold as an inseparable part
of new-pm transition. It was added when old-pm was still the default,
and very *very* few users were actually tracking new-pm,
so it's effects weren't measured.
Which means, some of the turnoil of the new-pm transition
are actually likely regressions due to this pass.
Likewise, there has been a number of post-commit feedback
(post new-pm switch), namely
* https://reviews.llvm.org/D37467#2787157 (regresses HW-loops)
* https://reviews.llvm.org/D37467#2787259 (should not be in middle-end, should run after LSR, not before)
* https://reviews.llvm.org/D95789 (an attempt to fix bad loop backedge metadata)
and in the half year past, the pass authors (google) still haven't found time to respond to any of that.
Hereby it is proposed to backout the pass from the pipeline,
until someone who cares about it can address the issues reported,
and properly start the process of adding a new pass into the pipeline,
with proper performance evaluation.
Furthermore, neither google nor facebook reports any perf changes
from this change, so i'm dropping the pass completely.
It can always be re-reverted should/if anyone want to pick it up again.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D104099
Turns out simplifyLoopIVs sometimes returns a non-dead instruction in it's DeadInsts out param. I had done a bit of NFC cleanup which was only NFC if simplifyLoopIVs obeyed it's documentation. I'm simplfy dropping that part of the change.
Commit message from try 3:
Recommitting after fixing a bug found post commit. Amusingly, try 1 had been correct, and by reverting to incorporate last minute review feedback, I introduce the bug. Oops. :)
Original commit message:
The problem was that recursively deleting an instruction can delete instructions beyond the current iterator (via a dead phi), thus invalidating iteration. Test case added in LoopUnroll/dce.ll to cover this case.
LoopUnroll does a limited DCE pass after unrolling, but if you have a chain of dead instructions, it only deletes the last one. Improve the code to recursively delete all trivially dead instructions.
Differential Revision: https://reviews.llvm.org/D102511
Recommitting after fixing a bug found post commit. Amusingly, try 1 had been correct, and by reverting to incorporate last minute review feedback, I introduce the bug. Oops. :)
The problem was that recursively deleting an instruction can delete instructions beyond the current iterator (via a dead phi), thus invalidating iteration. Test case added in LoopUnroll/dce.ll to cover this case.
LoopUnroll does a limited DCE pass after unrolling, but if you have a chain of dead instructions, it only deletes the last one. Improve the code to recursively delete all trivially dead instructions.
Differential Revision: https://reviews.llvm.org/D102511
Recommitting after addressing a missed review comment, and updating an aarch64 test I'd missed.
LoopUnroll does a limited DCE pass after unrolling, but if you have a chain of dead instructions, it only deletes the last one. Improve the code to recursively delete all trivially dead instructions.
Differential Revision: https://reviews.llvm.org/D102511
At the moment, getMemoryOpCost returns 1 for all inputs if CostKind is
CodeSize or SizeAndLatency. This fools LoopUnroll into thinking memory
operations on large vectors have a cost of one, even if they will get
expanded to a large number of memory operations in the backend.
This patch updates getMemoryOpCost to return the cost for the type
legalization for both CodeSize and SizeAndLatency. This should more
accurately reflect the number of memory operations required.
I am not sure how latency should properly be included in SizeAndLatency
from the description, but returning the size cost should be clearly more
accurate.
This does not cause any binary changes when building
MultiSource/SPEC2000/SPEC2006 with -O3 -flto for AArch64, likely because
large vector memops are not really formed by code emitted from Clang.
But using the C/C++ matrix extension can easily result in code with very
large vector operations directly from Clang, e.g.
https://clang.godbolt.org/z/6xzxcTGvb
Reviewed By: samparker
Differential Revision: https://reviews.llvm.org/D100291
See discussion in D90554.
This is a partial un-revert of 32dd5870ee31. I'm adding
back the baseline tests first, so we don't have to
back-track as much in case there are still problems.
as it's causing crashes in the optimizer. A reduced testcase has been posted as a follow-up.
This reverts commit f7eac51b9b3f780c96ca41913293851c5acb465b.
Temporarily Revert "[CostModel] make default size cost for libcalls small (again)" as it depends upon the primary revert.
This reverts commit 8ec7ea3ddce7379e13e8dfb4a5260a6d2004aa1c.
Temporarily Revert "[CostModel] add tests for math library calls; NFC" as it depends upon the primary revert.
This reverts commit df09f825995b10da03f148133c119f52c94fd6e4.
Temporarily Revert "[LoopUnroll] add test for full unroll that is sensitive to cost-model; NFC" as it depends upon the primary revert.
This reverts commit 618d555e8d926a83161774df2035519c387269db.
This was changed recently with D90554 / f7eac51b9b3f
...because we had a regression testing blindspot for intrinsics
that are expected to be lowered to libcalls.
In general, we want the *size* cost for a scalar call to be cheap
even if the other costs are expensive - we expect it to just be
a branch with some optional stack manipulation.
It is likely that we will want to carve out some
exceptions/overrides to this rule as follow-up patches for
calls that have some general and/or target-specific difference
to the expected lowering.
This was noticed as a regression in unrolling, so we have a test
for that now along with a couple of direct cost model tests.
If the assumed scalarization costs for the oversized vector
calls are not realistic, that would be another follow-up
refinement of the cost models.
We use `< UP.Threshold` later on, so we should use LoopSize + 1, to
allow unrolling if the result won't exceed to loop size.
Fixes PR43305.
Reviewers: efriedma, dmgreen, paquette
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D67594
llvm-svn: 372084
Summary:
In the following cases, unrolling can be beneficial, even when
optimizing for code size:
1) very low trip counts
2) potential to constant fold most instructions after fully unrolling.
We can unroll in those cases, by setting the unrolling threshold to the
loop size. This might highlight some cost modeling issues and fixing
them will have a positive impact in general.
Reviewers: vsk, efriedma, dmgreen, paquette
Reviewed By: paquette
Differential Revision: https://reviews.llvm.org/D60265
llvm-svn: 358586
As it's causing some bot failures (and per request from kbarton).
This reverts commit r358543/ab70da07286e618016e78247e4a24fcb84077fda.
llvm-svn: 358546
Reappy r284044 after revert in r284051. Krzysztof fixed the error in r284049.
The original summary:
This patch tries to fully unroll loops having break statement like this
for (int i = 0; i < 8; i++) {
if (a[i] == value) {
found = true;
break;
}
}
GCC can fully unroll such loops, but currently LLVM cannot because LLVM only
supports loops having exact constant trip counts.
The upper bound of the trip count can be obtained from calling
ScalarEvolution::getMaxBackedgeTakenCount(). Part of the patch is the
refactoring work in SCEV to prevent duplicating code.
The feature of using the upper bound is enabled under the same circumstance
when runtime unrolling is enabled since both are used to unroll loops without
knowing the exact constant trip count.
llvm-svn: 284053
This patch tries to fully unroll loops having break statement like this
for (int i = 0; i < 8; i++) {
if (a[i] == value) {
found = true;
break;
}
}
GCC can fully unroll such loops, but currently LLVM cannot because LLVM only
supports loops having exact constant trip counts.
The upper bound of the trip count can be obtained from calling
ScalarEvolution::getMaxBackedgeTakenCount(). Part of the patch is the
refactoring work in SCEV to prevent duplicating code.
The feature of using the upper bound is enabled under the same circumstance
when runtime unrolling is enabled since both are used to unroll loops without
knowing the exact constant trip count.
Differential Revision: https://reviews.llvm.org/D24790
llvm-svn: 284044
As agreed in post-commit review of r265388, I'm switching the flag to
its original value until the 90% runtime performance regression on
SingleSource/Benchmarks/Stanford/Bubblesort is addressed.
llvm-svn: 277524
Summary:
It is incorrect to compare TripCount (which is BECount + 1)
with extraiters (or Count) to check if we should enter unrolled
loop or not, because TripCount can potentially overflow
(when BECount is max unsigned integer).
While comparing BECount with (Count - 1) is overflow safe and
therefore correct.
Reviewer: hfinkel
Differential Revision: http://reviews.llvm.org/D19256
From: Evgeny Stupachenko <evstupac@gmail.com>
llvm-svn: 267662
For inner one of nested loops, it is more likely to be a hot loop,
and the runtime check can be promoted out from patch 0001, so the
overhead is less, we can try a doubled threshold to unroll more loops.
llvm-svn: 231632
