MaxSafeRegisterWidth is a misnomer since it actually returns the maximum
safe vector width. Register suggests it relates directly to a physical
register where it could be a vector spanning one or more physical
registers.
Reviewed By: sdesmalen
Differential Revision: https://reviews.llvm.org/D91727
Similar to other patches, this makes VPWidenRecipe a VPValue. Because of
the way it interacts with the reduction code it also slightly alters the
way that VPValues are registered, removing the up front NeedDef and
using getOrAddVPValue to create them on-demand if needed instead.
Differential Revision: https://reviews.llvm.org/D88447
This converts the VPReductionRecipe into a VPValue, like other
VPRecipe's in preparation for traversing def-use chains. It also makes
it a VPUser, now storing the used VPValues as operands.
It doesn't yet change how the VPReductionRecipes are created. It will
need to call replaceAllUsesWith from the original recipe they replace,
but that is not done yet as VPWidenRecipe need to be created first.
Differential Revision: https://reviews.llvm.org/D88382
Some older code - and code copied from older code - still directly tested against the singelton result of SE::getCouldNotCompute. Using the isa<SCEVCouldNotCompute> form is both shorter, and more readable.
Fix PR47390.
The primary induction should be considered alive when folding tail by masking,
because it will be used by said masking; even when it may otherwise appear
useless: feeding only its own 'bump', which is correctly considered dead, and
as the 'bump' of another induction variable, which may wrongfully want to
consider its bump = the primary induction, dead.
Differential Revision: https://reviews.llvm.org/D92017
A uniform load is one which loads from a uniform address across all lanes. As currently implemented, we cost model such loads as if we did a single scalar load + a broadcast, but the actual lowering replicates the load once per lane.
This change tweaks the lowering to use the REPLICATE strategy by marking such loads (and the computation leading to their memory operand) as uniform after vectorization. This is a useful change in itself, but it's real purpose is to pave the way for a following change which will generalize our uniformity logic.
In review discussion, there was an issue raised with coupling cost modeling with the lowering strategy for uniform inputs. The discussion on that item remains unsettled and is pending larger architectural discussion. We decided to move forward with this patch as is, and revise as warranted once the bigger picture design questions are settled.
Differential Revision: https://reviews.llvm.org/D91398
This change introduces a new IR intrinsic named `llvm.pseudoprobe` for pseudo-probe block instrumentation. Please refer to https://reviews.llvm.org/D86193 for the whole story.
A pseudo probe is used to collect the execution count of the block where the probe is instrumented. This requires a pseudo probe to be persisting. The LLVM PGO instrumentation also instruments in similar places by placing a counter in the form of atomic read/write operations or runtime helper calls. While these operations are very persisting or optimization-resilient, in theory we can borrow the atomic read/write implementation from PGO counters and cut it off at the end of compilation with all the atomics converted into binary data. This was our initial design and we’ve seen promising sample correlation quality with it. However, the atomics approach has a couple issues:
1. IR Optimizations are blocked unexpectedly. Those atomic instructions are not going to be physically present in the binary code, but since they are on the IR till very end of compilation, they can still prevent certain IR optimizations and result in lower code quality.
2. The counter atomics may not be fully cleaned up from the code stream eventually.
3. Extra work is needed for re-targeting.
We choose to implement pseudo probes based on a special LLVM intrinsic, which is expected to have most of the semantics that comes with an atomic operation but does not block desired optimizations as much as possible. More specifically the semantics associated with the new intrinsic enforces a pseudo probe to be virtually executed exactly the same number of times before and after an IR optimization. The intrinsic also comes with certain flags that are carefully chosen so that the places they are probing are not going to be messed up by the optimizer while most of the IR optimizations still work. The core flags given to the special intrinsic is `IntrInaccessibleMemOnly`, which means the intrinsic accesses memory and does have a side effect so that it is not removable, but is does not access memory locations that are accessible by any original instructions. This way the intrinsic does not alias with any original instruction and thus it does not block optimizations as much as an atomic operation does. We also assign a function GUID and a block index to an intrinsic so that they are uniquely identified and not merged in order to achieve good correlation quality.
Let's now look at an example. Given the following LLVM IR:
```
define internal void @foo2(i32 %x, void (i32)* %f) !dbg !4 {
bb0:
%cmp = icmp eq i32 %x, 0
br i1 %cmp, label %bb1, label %bb2
bb1:
br label %bb3
bb2:
br label %bb3
bb3:
ret void
}
```
The instrumented IR will look like below. Note that each `llvm.pseudoprobe` intrinsic call represents a pseudo probe at a block, of which the first parameter is the GUID of the probe’s owner function and the second parameter is the probe’s ID.
```
define internal void @foo2(i32 %x, void (i32)* %f) !dbg !4 {
bb0:
%cmp = icmp eq i32 %x, 0
call void @llvm.pseudoprobe(i64 837061429793323041, i64 1)
br i1 %cmp, label %bb1, label %bb2
bb1:
call void @llvm.pseudoprobe(i64 837061429793323041, i64 2)
br label %bb3
bb2:
call void @llvm.pseudoprobe(i64 837061429793323041, i64 3)
br label %bb3
bb3:
call void @llvm.pseudoprobe(i64 837061429793323041, i64 4)
ret void
}
```
Reviewed By: wmi
Differential Revision: https://reviews.llvm.org/D86490
rGf571fe6df585127d8b045f8e8f5b4e59da9bbb73 led to a warning of an unused
variable for MaxSafeDepDist (written but not used). It seems this
variable and assignment can be safely removed.
The assertion that vector widths are <= 256 elements was hard wired in the LV code. Eg, VE allows for vectors up to 512 elements. Test again the TTI vector register bit width instead - this is an NFC for non-asserting builds.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D91518
This relands https://reviews.llvm.org/D91059 and reverts commit
30fded75b48bcbc034120154a57a00c7f3d07e06.
GetRegUsage now returns 0 when Ty is not a valid vector element type.
This patch turns VPWidenGEPRecipe into a VPValue and uses it
during VPlan construction and codegeneration instead of the plain IR
reference where possible.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D84683
This reverts the revert commit c8d73d939fa4fda9c87b3979225d02d63062bd68.
It includes a fix for cases where we missed inserting VPValues
for some selects, which should fix PR48142.
This reverts commits:
* [LoopVectorizer] NFCI: Calculate register usage based on TLI.getTypeLegalizationCost.
b873aba3943c067a5efd5303cbdf5aeb0732cf88.
* [LoopVectorizer] Silence warning in GetRegUsage.
9ff701100a868b7b680aac5c54e9db21a55531fd.
This patch silences the warning:
error: lambda capture 'DL' is not used [-Werror,-Wunused-lambda-capture]
auto GetRegUsage = [&DL, &TTI=TTI](Type *Ty, ElementCount VF) {
~^~~
1 error generated.
Introduced in:
https://reviews.llvm.org/rGb873aba3943c067a5efd5303cbdf5aeb0732cf88
This is more accurate than dividing the bitwidth based on the element count by the
maximum register size, as it can just reuse whatever has been calculated for
legalization of these types.
This change is also necessary when calculating register usage for scalable vectors, where
the legalization of these types cannot be done based on the widest register size, because
that does not take the 'vscale' component into account.
Reviewed By: SjoerdMeijer
Differential Revision: https://reviews.llvm.org/D91059
This patch turns VPWidenSelectRecipe into a VPValue and uses it
during VPlan construction and codegeneration instead of the plain IR
reference where possible.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D84682
Interfaces changed to take `ElementCount` as parameters:
* LoopVectorizationPlanner::buildVPlans
* LoopVectorizationPlanner::buildVPlansWithVPRecipes
* LoopVectorizationCostModel::selectVectorizationFactor
This patch is NFC for fixed-width vectors.
Reviewed By: dmgreen, ctetreau
Differential Revision: https://reviews.llvm.org/D90879
This patch turns VPWidenCall into a VPValue and uses it
during VPlan construction and codegeneration instead of the plain IR
reference where possible.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D84681
This patch changes the type of Start, End in VFRange to be an ElementCount
instead of `unsigned`. This is done as preparation to make VPlans for
scalable vectors, but is otherwise NFC.
Reviewed By: dmgreen, fhahn, vkmr
Differential Revision: https://reviews.llvm.org/D90715
This reverts the revert commit 408c4408facc3a79ee4ff7e9983cc972f797e176.
This version of the patch includes a fix for a crash caused by
treating ICmp/FCmp constant expressions as instructions.
Original message:
On some targets, like AArch64, vector selects can be efficiently lowered
if the vector condition is a compare with a supported predicate.
This patch adds a new argument to getCmpSelInstrCost, to indicate the
predicate of the feeding select condition. Note that it is not
sufficient to use the context instruction when querying the cost of a
vector select starting from a scalar one, because the condition of the
vector select could be composed of compares with different predicates.
This change greatly improves modeling the costs of certain
compare/select patterns on AArch64.
I am also planning on putting up patches to make use of the new argument in
SLPVectorizer & LV.
On some targets, like AArch64, vector selects can be efficiently lowered
if the vector condition is a compare with a supported predicate.
This patch adds a new argument to getCmpSelInstrCost, to indicate the
predicate of the feeding select condition. Note that it is not
sufficient to use the context instruction when querying the cost of a
vector select starting from a scalar one, because the condition of the
vector select could be composed of compares with different predicates.
This change greatly improves modeling the costs of certain
compare/select patterns on AArch64.
I am also planning on putting up patches to make use of the new argument in
SLPVectorizer & LV.
Reviewed By: dmgreen, RKSimon
Differential Revision: https://reviews.llvm.org/D90070
This is an initial cleanup of the way LoopVersioning interacts with LAA.
Currently LoopVersioning has 2 ways of initializing things:
1. Passing LAI and passing UseLAIChecks = true
2. Passing UseLAIChecks = false, followed by calling setSCEVChecks and
setAliasChecks.
Both ways of initializing lead to the same result and the duplication
seems more complicated than necessary.
This patch removes the UseLAIChecks flag from the constructor and the
setSCEVChecks & setAliasChecks helpers and move initialization
exclusively to the constructor.
This simplifies things, by providing a single way to initialize
LoopVersioning and reducing duplication.
Reviewed By: Meinersbur, lebedev.ri
Differential Revision: https://reviews.llvm.org/D84406
This reverts the revert commit 710aceb645e7dba4de7053eef2c616311b9163d4
and includes a fix for a memsan failure.
Original message:
This patch turns VPMemoryInstructionRecipe into a VPValue and uses it
during VPlan construction and codegeneration instead of the plain IR
reference where possible.
LV fails with assertion checking that UF > 0. We already set UF to 1 if it is 0 except the case when IC > MaxInterleaveCount. The fix is to set UF to 1 for that case as well.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D87679
This patch turns VPMemoryInstructionRecipe into a VPValue and uses it
during VPlan construction and codegeneration instead of the plain IR
reference where possible.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D84680
I have introduced a new template PolySize class, where the template
parameter determines the type of quantity, i.e. for an element
count this is just an unsigned value. The ElementCount class is
now just a simple derivation of PolySize<unsigned>, whereas TypeSize
is more complicated because it still needs to contain the uint64_t
cast operator, since there are still many places in the code that
rely upon this implicit cast. As such the class also still needs
some of it's own operators.
I've tried to minimise the amount of code in the base PolySize
class, which led to a couple of changes:
1. In some places we were relying on '==' operator comparisons
between ElementCounts and the scalar value 1. I didn't put this
operator in the new PolySize class, and thought it was actually
clearer to use the isScalar() function instead.
2. I removed the isByteSized function and replaced it with calls
to isKnownMultipleOf(8).
I've also renamed NextPowerOf2 to be coefficientNextPowerOf2 so
that it's more consistent with coefficientDivideBy.
Differential Revision: https://reviews.llvm.org/D88409
This expands upon the inloop reductions added in e9761688e41cb9e976,
allowing them to be inserted into tail folded loops. Reductions are
generates with the form:
x = select(mask, vecop, zero)
v = vecreduce.add(x)
c = add chain, v
Where zero here is chosen as the identity value for add reductions. The
backend is then expected to fold the select and the vecreduce into a
single predicated instruction.
Most of the code is fairly straight forward, except for the creation of
blockmasks which need to ensure they are created in dominance order. The
order they are added is altered to be after any phis, keeping the
requirements for the underlying IR.
Differential Revision: https://reviews.llvm.org/D84451
We currently collect the ICmp and Add from an induction variable,
marking them as dead so that vplan values are not created for them. This
extends that to include any single use trunk from the ICmp, which allows
the Add to more readily be removed too.
This can help with costing vplan nodes, as the ICmp and Add are more
reliably removed and are not double-counted.
Differential Revision: https://reviews.llvm.org/D88873
Now that VPUser is not inheriting from VPValue, we can take the next
step and turn the recipes that already manage their operands via VPUser
into VPUsers directly. This is another small step towards traversing
def-use chains in VPlan.
This is NFC with respect to the generated code, but makes the interface
more powerful.
~~D65060 uncovered that trying to use BFI in loop passes can lead to non-deterministic behavior when blocks are re-used while retaining old BFI data.~~
~~To make sure BFI is preserved through loop passes a Value Handle (VH) callback is registered on blocks themselves. When a block is freed it now also wipes out the accompanying BFI entry such that stale BFI data can no longer persist resolving the determinism issue. ~~
~~An optimistic approach would be to incrementally update BFI information throughout the loop passes rather than only invalidating them on removed blocks. The issues with that are:~~
~~1. It is not clear how BFI information should be incrementally updated: If a block is duplicated does its BFI information come with? How about if it's split/modified/moved around? ~~
~~2. Assuming we can address these problems the implementation here will be a massive undertaking. ~~
~~There's a known need of BFI in LICM analysis which requires correct but not incrementally updated BFI data. A follow-up change can register BFI in all loop passes so this preserved but potentially lossy data is available to any loop pass that wants it.~~
See: D75341 for an identical implementation of preserving BFI via VH callbacks. The previous statements do still apply but this change no longer has to be in this diff because it's already upstream 😄 .
This diff also moves BFI to be a part of LoopStandardAnalysisResults since the previous method using getCachedResults now (correctly!) statically asserts (D72893) that this data isn't static through the loop passes.
Testing
Ninja check
Reviewed By: asbirlea, nikic
Differential Revision: https://reviews.llvm.org/D86156
This allows the backend to tell the vectorizer to produce inloop
reductions through a TTI hook.
For the moment on ARM under MVE this means allowing integer add
reductions of the correct size. In the future this can include integer
min/max too, under -Os.
Differential Revision: https://reviews.llvm.org/D75512
Interleave for small loops that have reductions inside,
which breaks dependencies and expose.
This gives very significant performance improvements for some benchmarks.
Because small loops could be in very hot functions in real applications.
Differential Revision: https://reviews.llvm.org/D81416
addRuntimeChecks uses SCEVExpander, which relies on the DT/LoopInfo to
be up-to-date. Changing the CFG afterwards may invalidate some inserted
instructions, especially LCSSA phis.
Reorder the code to first update the CFG and then create the runtime
checks. This should not have any impact on the generated code, as we
adjust the CFG and generate runtime checks together.
Fixes PR47343.
Move bail out when optimizing for size before runtime check generation.
In that case, we do not use the result of the expansion, the expanded
instruction will be dead and cleaned up later.
By doing the check before expanding the runtime-checks, we can save a
bit of unnecessary work.
This patch changes ElementCount so that the Min and Scalable
members are now private and can only be accessed via the get
functions getKnownMinValue() and isScalable(). In addition I've
added some other member functions for more commonly used operations.
Hopefully this makes the class more useful and will reduce the
need for calling getKnownMinValue().
Differential Revision: https://reviews.llvm.org/D86065
This implements 2 different vectorisation fallback strategies if tail-folding
fails: 1) don't vectorise at all, or 2) vectorise using a scalar epilogue. This
can be controlled with option -prefer-predicate-over-epilogue, that has been
changed to take a numeric value corresponding to the tail-folding preference
and preferred fallback.
Patch by: Pierre van Houtryve, Sjoerd Meijer.
Differential Revision: https://reviews.llvm.org/D79783
