This is an enhancement motivated by https://llvm.org/PR16739
(see D92858 for another).
We can look through a GEP to find a base pointer that may be
safe to use for a vector load. If so, then we shuffle (shift)
the necessary vector element over to index 0.
Alive2 proof based on 1 of the regression tests:
https://alive2.llvm.org/ce/z/yPJLkh
The vector translation is independent of endian (verify by
changing to leading 'E' in the datalayout string).
Differential Revision: https://reviews.llvm.org/D93229
Here's another minimal step suggested by D93229 / D93397 .
(I'm trying to be extra careful in these changes because
load transforms are easy to get wrong.)
We can optimistically choose the greater alignment of a
load and its pointer operand. As the test diffs show, this
can improve what would have been unaligned vector loads
into aligned loads.
When we enhance with gep offsets, we will need to adjust
the alignment calculation to include that offset.
Differential Revision: https://reviews.llvm.org/D93406
As discussed in D93229, we only need a minimal alignment constraint
when querying whether a hypothetical vector load is safe. We still
pass/use the potentially stronger alignment attribute when checking
costs and creating the new load.
There's already a test that changes with the minimum code change,
so splitting this off as a preliminary commit independent of any
gep/offset enhancements.
Differential Revision: https://reviews.llvm.org/D93397
This patch changes the type of cost variables (for instance: Cost, ExtractCost,
SpillCost) to use InstructionCost.
This patch also changes the type of cost variables to InstructionCost in other
functions that use the result of getTreeCost()
This patch is part of a series of patches to use InstructionCost instead of
unsigned/int for the cost model functions.
See this thread for context:
http://lists.llvm.org/pipermail/llvm-dev/2020-November/146408.html
Depends on D91174
Differential Revision: https://reviews.llvm.org/D93049
Given we haven't yet enabled multiple exiting blocks, this is currently non functional, but it's an obvious extension which cleans up a later patch.
I don't think this is worth review (as it's pretty obvious), if anyone disagrees, feel feel to revert or comment and I will.
This should be purely non-functional. When touching this code for another reason, I found the handling of the PredicateOrDontVectorize piece here very confusing. Let's make it an explicit state (instead of an implicit combination of two variables), and use early return for options/hint processing.
This patch turns updates VPWidenSelectRecipe to manage the value
it defines using VPDef.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90560
This patch turns updates VPWidenGEPRecipe to manage the value it defines
using VPDef. The VPValue is used during VPlan construction and
codegeneration instead of the plain IR reference where possible.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90561
This patch turns updates VPWidenREcipe to manage the value it defines
using VPDef.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90559
As noted in D93229, the transform from scalar load to vector load
potentially leaks poison from the extra vector elements that are
being loaded.
We could use freeze here (and x86 codegen at least appears to be
the same either way), but we already have a shuffle in this logic
to optionally change the vector size, so let's allow that
instruction to serve both purposes.
Differential Revision: https://reviews.llvm.org/D93238
D82227 has added a proper check to limit PHI vectorization to the
maximum vector register size. That unfortunately resulted in at
least a couple of regressions on SystemZ and x86.
This change reverts PHI handling from D82227 and replaces it with
a more general check in SLPVectorizerPass::tryToVectorizeList().
Moved to tryToVectorizeList() it allows to restart vectorization
if initial chunk fails.
However, this function is more general and handles not only PHI
but everything which SLP handles. If vectorization factor would
be limited to maximum vector register size it would limit much
more vectorization than before leading to further regressions.
Therefore a new TTI callback getMaximumVF() is added with the
default 0 to preserve current behavior and limit nothing. Then
targets can decide what is better for them.
The callback gets ElementSize just like a similar getMinimumVF()
function and the main opcode of the chain. The latter is to avoid
regressions at least on the AMDGPU. We can have loads and stores
up to 128 bit wide, and <2 x 16> bit vector math on some
subtargets, where the rest shall not be vectorized. I.e. we need
to differentiate based on the element size and operation itself.
Differential Revision: https://reviews.llvm.org/D92059
This patch updates VPWidenMemoryInstructionRecipe to use VPDef
to manage the value it produces instead of inheriting from VPValue.
Reviewed By: gilr
Differential Revision: https://reviews.llvm.org/D90563
Vector element size could be different for different store chains.
This patch prevents wrong computation of maximum number of elements
for that case.
Differential Revision: https://reviews.llvm.org/D93192
When it comes to the scalar cost of any predicated block, the loop
vectorizer by default regards this predication as a sign that it is
looking at an if-conversion and divides the scalar cost of the block by
2, assuming it would only be executed half the time. This however makes
no sense if the predication has been introduced to tail predicate the
loop.
Original patch by Anna Welker
Differential Revision: https://reviews.llvm.org/D86452
This is the first in a series of patches that attempts to migrate
existing cost instructions to return a new InstructionCost class
in place of a simple integer. This new class is intended to be
as light-weight and simple as possible, with a full range of
arithmetic and comparison operators that largely mirror the same
sets of operations on basic types, such as integers. The main
advantage to using an InstructionCost is that it can encode a
particular cost state in addition to a value. The initial
implementation only has two states - Normal and Invalid - but these
could be expanded over time if necessary. An invalid state can
be used to represent an unknown cost or an instruction that is
prohibitively expensive.
This patch adds the new class and changes the getInstructionCost
interface to return the new class. Other cost functions, such as
getUserCost, etc., will be migrated in future patches as I believe
this to be less disruptive. One benefit of this new class is that
it provides a way to unify many of the magic costs in the codebase
where the cost is set to a deliberately high number to prevent
optimisations taking place, e.g. vectorization. It also provides
a route to represent the extremely high, and unknown, cost of
scalarization of scalable vectors, which is not currently supported.
Differential Revision: https://reviews.llvm.org/D91174
This is an enhancement to load vectorization that is motivated by
a pattern in https://llvm.org/PR16739.
Unfortunately, it's still not enough to make a difference there.
We will have to handle multi-use cases in some better way to avoid
creating multiple overlapping loads.
Differential Revision: https://reviews.llvm.org/D92858
For stores chain vectorization we choose the size of vector
elements to ensure we fit to minimum and maximum vector register
size for the number of elements given. This patch corrects vector
element size choosing the width of value truncated just before
storing instead of the width of value stored.
Fixes PR46983
Differential Revision: https://reviews.llvm.org/D92824
* Steps are scaled by `vscale`, a runtime value.
* Changes to circumvent the cost-model for now (temporary)
so that the cost-model can be implemented separately.
This can vectorize the following loop [1]:
void loop(int N, double *a, double *b) {
#pragma clang loop vectorize_width(4, scalable)
for (int i = 0; i < N; i++) {
a[i] = b[i] + 1.0;
}
}
[1] This source-level example is based on the pragma proposed
separately in D89031. This patch only implements the LLVM part.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D91077
This patch removes a number of asserts that VF is not scalable, even though
the code where this assert lives does nothing that prevents VF being scalable.
Reviewed By: dmgreen
Differential Revision: https://reviews.llvm.org/D91060
It is possible to merge reuse and reorder shuffles and reduce the total
cost of the ivectorization tree/number of final instructions.
Differential Revision: https://reviews.llvm.org/D92668
The initial step of the uniform-after-vectorization (lane-0 demanded only) analysis was very awkwardly written. It would revisit use list of each pointer operand of a widened load/store. As a result, it was in the worst case O(N^2) where N was the number of instructions in a loop, and had restricted operand Value types to reduce the size of use lists.
This patch replaces the original algorithm with one which is at most O(2N) in the number of instructions in the loop. (The key observation is that each use of a potentially interesting pointer is visited at most twice, once on first scan, once in the use list of *it's* operand. Only instructions within the loop have their uses scanned.)
In the process, we remove a restriction which required the operand of the uniform mem op to itself be an instruction. This allows detection of uniform mem ops involving global addresses.
Differential Revision: https://reviews.llvm.org/D92056
This is yet another attempt at providing support for epilogue
vectorization following discussions raised in RFC http://llvm.1065342.n5.nabble.com/llvm-dev-Proposal-RFC-Epilog-loop-vectorization-tt106322.html#none
and reviews D30247 and D88819.
Similar to D88819, this patch achieve epilogue vectorization by
executing a single vplan twice: once on the main loop and a second
time on the epilogue loop (using a different VF). However it's able
to handle more loops, and generates more optimal control flow for
cases where the trip count is too small to execute any code in vector
form.
Reviewed By: SjoerdMeijer
Differential Revision: https://reviews.llvm.org/D89566
This might be a small improvement in readability, but the
real motivation is to make it easier to adapt the code to
deal with intrinsics like 'maxnum' and/or integer min/max.
There is potentially help in doing that with D92086, but
we might also just add specialized wrappers here to deal
with the expected patterns.
In this patch I have added support for a new loop hint called
vectorize.scalable.enable that says whether we should enable scalable
vectorization or not. If a user wants to instruct the compiler to
vectorize a loop with scalable vectors they can now do this as
follows:
br i1 %exitcond, label %for.end, label %for.body, !llvm.loop !2
...
!2 = !{!2, !3, !4}
!3 = !{!"llvm.loop.vectorize.width", i32 8}
!4 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
Setting the hint to false simply reverts the behaviour back to the
default, using fixed width vectors.
Differential Revision: https://reviews.llvm.org/D88962
This is yet another attempt at providing support for epilogue
vectorization following discussions raised in RFC http://llvm.1065342.n5.nabble.com/llvm-dev-Proposal-RFC-Epilog-loop-vectorization-tt106322.html#none
and reviews D30247 and D88819.
Similar to D88819, this patch achieve epilogue vectorization by
executing a single vplan twice: once on the main loop and a second
time on the epilogue loop (using a different VF). However it's able
to handle more loops, and generates more optimal control flow for
cases where the trip count is too small to execute any code in vector
form.
Reviewed By: SjoerdMeijer
Differential Revision: https://reviews.llvm.org/D89566
In the following loop the dependence distance is 2 and can only be
vectorized if the vector length is no larger than this.
void foo(int *a, int *b, int N) {
#pragma clang loop vectorize(enable) vectorize_width(4)
for (int i=0; i<N; ++i) {
a[i + 2] = a[i] + b[i];
}
}
However, when specifying a VF of 4 via a loop hint this loop is
vectorized. According to [1][2], loop hints are ignored if the
optimization is not safe to apply.
This patch introduces a check to bail of vectorization if the user
specified VF is greater than the maximum feasible VF, unless explicitly
forced with '-force-vector-width=X'.
[1] https://llvm.org/docs/LangRef.html#llvm-loop-vectorize-and-llvm-loop-interleave
[2] https://clang.llvm.org/docs/LanguageExtensions.html#extensions-for-loop-hint-optimizations
Reviewed By: sdesmalen, fhahn, Meinersbur
Differential Revision: https://reviews.llvm.org/D90687
This patch replaces the attribute `unsigned VF` in the class
IntrinsicCostAttributes by `ElementCount VF`.
This is a non-functional change to help upcoming patches to compute the cost
model for scalable vector inside this class.
Differential Revision: https://reviews.llvm.org/D91532
Instruction ExtractValue wasn't handled in
LoopVectorizationCostModel::getInstructionCost(). As a result, it was modeled
as a mul which is not really accurate. Since it is free (most of the times),
this now gets a cost of 0 using getInstructionCost.
This is a follow-up of D92208, that required changing this regression test.
In a follow up I will look at InsertValue which also isn't handled yet.
Differential Revision: https://reviews.llvm.org/D92317
VPPredInstPHIRecipe is one of the recipes that was missed during the
initial conversion. This patch adjusts the recipe to also manage its
operand using VPUser.
Interleave groups also depend on the values they store. Manage the
stored values as VPUser operands. This is currently a NFC, but is
required to allow VPlan transforms and to manage generated vector values
exclusively in VPTransformState.
Update VPReplicateRecipe to inherit from VPValue. This still does not
update scalarizeInstruction to set the result for the VPValue of
VPReplicateRecipe, because this first requires tracking scalar values in
VPTransformState.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D91500
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
This is a follow-up to 00a6601136 to make
isa<VPReductionRecipe> work and unifies the VPValue ID names, by making
sure they all consistently start with VPV*.
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 patch introduces a new VPDef class, which can be used to
manage VPValues defined by recipes/VPInstructions.
The idea here is to mirror VPUser for values defined by a recipe. A
VPDef can produce either zero (e.g. a store recipe), one (most recipes)
or multiple (VPInterleaveRecipe) result VPValues.
To traverse the def-use chain from a VPDef to its users, one has to
traverse the users of all values defined by a VPDef.
VPValues now contain a pointer to their corresponding VPDef, if one
exists. To traverse the def-use chain upwards from a VPValue, we first
need to check if the VPValue is defined by a VPDef. If it does not have
a VPDef, this means we have a VPValue that is not directly defined
iniside the plan and we are done.
If we have a VPDef, it is defined inside the region by a recipe, which
is a VPUser, and the upwards def-use chain traversal continues by
traversing all its operands.
Note that we need to add an additional field to to VPVAlue to link them
to their defs. The space increase is going to be offset by being able to
remove the SubclassID field in future patches.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D90558
For the scattered operands of load instructions it makes sense
to use gathering load intrinsic, which can lower to native instruction
for X86/AVX512 and ARM/SVE. This also enables building
vectorization tree with entries containing scattered operands.
The next step is to add scattered store.
Fixes PR47629 and PR47623
Differential Revision: https://reviews.llvm.org/D90445
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
No longer rely on an external tool to build the llvm component layout.
Instead, leverage the existing `add_llvm_componentlibrary` cmake function and
introduce `add_llvm_component_group` to accurately describe component behavior.
These function store extra properties in the created targets. These properties
are processed once all components are defined to resolve library dependencies
and produce the header expected by llvm-config.
Differential Revision: https://reviews.llvm.org/D90848
This reverts commits:
* [LoopVectorizer] NFCI: Calculate register usage based on TLI.getTypeLegalizationCost.
b873aba394.
* [LoopVectorizer] Silence warning in GetRegUsage.
9ff701100a.
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 408c4408fa.
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.
This reverts the revert commit a1b53db324.
This patch includes a fix for a reported issue, caused by
matchSelectPattern returning UMIN for selects of pointers in
some cases by looking to some connected casts.
For now, ensure integer instrinsics are only returned for selects of
ints or int vectors.
This reverts commit 1922570489.
This appears to cause a crash in the following example
a, b, c;
l() {
int e = a, f = l, g, h, i, j;
float *d = c, *k = b;
for (;;)
for (; g < f; g++) {
k[h] = d[i];
k[h - 1] = d[j];
h += e << 1;
i += e;
}
}
clang -cc1 -triple i386-unknown-linux-gnu -emit-obj -target-cpu pentium-m -O1 -vectorize-loops -vectorize-slp reduced.c
llvm::Type *llvm::Type::getWithNewBitWidth(unsigned int) const: Assertion `isIntOrIntVectorTy() && "Original type expected to be a vector of integers or a scalar integer."' failed.
As per the comment in VPRecipeBase, clients should not rely on
getVPRecipeID, as it may change in the future. It should only be used in
classof implementations. Use isa instead in getFirstNonPhi.
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
Some architectures do not have general vector select instructions (e.g.
AArch64). But some cmp/select patterns can be vectorized using other
instructions/intrinsics.
One example is using min/max instructions for certain patterns.
This patch updates the cost calculations for selects in the SLP
vectorizer to consider using min/max intrinsics.
This patch does not change SLP vectorizer's codegen itself to actually
generate those intrinsics, but relies on the backends to lower the
vector cmps & selects. This keeps things simple on the SLP side and
works well in practice for AArch64.
This exposes additional SLP vectorization opportunities in some
benchmarks on AArch64 (-O3 -flto).
Metric: SLP.NumVectorInstructions
Program base slp diff
test-suite...ications/JM/ldecod/ldecod.test 502.00 697.00 38.8%
test-suite...ications/JM/lencod/lencod.test 1023.00 1414.00 38.2%
test-suite...-typeset/consumer-typeset.test 56.00 65.00 16.1%
test-suite...6/464.h264ref/464.h264ref.test 804.00 822.00 2.2%
test-suite...006/453.povray/453.povray.test 3335.00 3357.00 0.7%
test-suite...CFP2000/177.mesa/177.mesa.test 2110.00 2121.00 0.5%
test-suite...:: External/Povray/povray.test 2378.00 2382.00 0.2%
Reviewed By: RKSimon, samparker
Differential Revision: https://reviews.llvm.org/D89969
These logically belong together since it's a base commit plus
followup fixes to less common build configurations.
The patches are:
Revert "CfgInterface: rename interface() to getInterface()"
This reverts commit a74fc48158.
Revert "Wrap CfgTraitsFor in namespace llvm to please GCC 5"
This reverts commit f2a06875b6.
Revert "Try to make GCC5 happy about the CfgTraits thing"
This reverts commit 03a5f7ce12.
Revert "Introduce CfgTraits abstraction"
This reverts commit c0cdd22c72.
The warning would fire when calling isDereferenceableAndAlignedInLoop
with a scalable load. Calling isDereferenceableAndAlignedInLoop with a
scalable load would result in the use of the now deprecated implicit
cast of TypeSize to uint64_t through the overloaded operator.
This patch fixes this issue by:
- no longer considering vector loads as candidates in
canVectorizeWithIfConvert. This doesn't make sense in the context of
identifying scalar loads to vectorize.
- making use of getFixedSize inside isDereferenceableAndAlignedInLoop --
this removes the dependency on the deprecated interface, and will
trigger an assertion error if the function is ever called with a
scalable type.
Reviewed By: sdesmalen
Differential Revision: https://reviews.llvm.org/D89798
The CfgTraits abstraction simplfies writing algorithms that are
generic over the type of CFG, and enables writing such algorithms
as regular non-template code that operates on opaque references
to CFG blocks and values.
Implementations of CfgTraits provide operations on the concrete
CFG types, e.g. `IrCfgTraits::BlockRef` is `BasicBlock *`.
CfgInterface is an abstract base class which provides operations
on opaque types CfgBlockRef and CfgValueRef. Those opaque types
encapsulate a `void *`, but the meaning depends on the concrete
CFG type. For example, MachineCfgTraits -- for use with MachineIR
in SSA form -- encodes a Register inside CfgValueRef. Converting
between concrete references and opaque/generic ones is done by
CfgTraits::{fromGeneric,toGeneric}. Convenience methods
CfgTraits::{un}wrap{Iterator,Range} are available as well.
Writing algorithms in terms of CfgInterface adds some overhead
(virtual method calls, plus in same cases it removes the
opportunity to inline iterators), but can be much more convenient
since generic algorithms can be written as non-templates.
This patch adds implementations of CfgTraits for all CFGs on
which dominator trees are calculated, so that the dominator
tree can be ported to this machinery. Only IrCfgTraits (LLVM IR)
and MachineCfgTraits (Machine IR in SSA form) are complete, the
other implementations are limited to the absolute minimum
required to make the upcoming dominator tree changes work.
v5:
- fix MachineCfgTraits::blockdef_iterator and allow it to iterate over
the instructions in a bundle
- use MachineBasicBlock::printName
v6:
- implement predecessors/successors for all CfgTraits implementations
- fix error in unwrapRange
- rename toGeneric/fromGeneric into wrapRef/unwrapRef to have naming
that is consistent with {wrap,unwrap}{Iterator,Range}
- use getVRegDef instead of getUniqueVRegDef
v7:
- std::forward fix in wrapping_iterator
- fix typos
v8:
- cleanup operators on CfgOpaqueType
- address other review comments
Change-Id: Ia75f4f268fded33fca11218a7d578c9aec1f3f4d
Differential Revision: https://reviews.llvm.org/D83088
We can not bitcast pointers across different address spaces, and VectorCombine
should be careful when it attempts to find the original source of the loaded
data.
Differential Revision: https://reviews.llvm.org/D89577
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 710aceb645
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
Update the code responsible for deleting VPBBs and recipes to properly
update users and release operands.
This is another preparation for D84680 & following patches towards
enabling modeling def-use chains in VPlan.
This adds a helper to convert a VPRecipeBase pointer to a VPUser, for
recipes that inherit from VPUser. Once VPRecipeBase directly inherits
from VPUser this helper can be removed.
When updating operands of a VPUser, we also have to adjust the list of
users for the new and old VPValues. This is required once we start
transitioning recipes to become VPValues.
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.
These were only really used for 2 things. One was to check if the operand matches the phi if it exists. The other was for the createOp method to build the reduction.
For the first case we still have the operation we just need to know how to index its operands. So I've modified getLHS/getRHS to just use the opcode/kind to know how to find the right operands on an instruction that is now passed in.
For the other case we had to create an OperationData object to set the LHS/RHS values and copy the opcode/kind from another object. We would then just call createOp on that temporary object. Instead I've made LHS/RHS arguments to createOp and removed all these temporary objects.
Differential Revision: https://reviews.llvm.org/D88193
All of the callers already have an Instruction *. Many of them
from a dyn_cast.
Also update the OperationData constructor to use a Instruction&
to remove a dyn_cast and make it clear that the pointer is non-null.
Differential Revision: https://reviews.llvm.org/D88132
This refactors VPuser to not inherit from VPValue to facilitate
introducing operations that introduce multiple VPValues (e.g.
VPInterleaveRecipe).
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D84679
This provides a convenient way to print VPValues and recipes in a
debugger. In particular it saves the user from instantiating
VPSlotTracker to print recipes or values.
The implementation of gather() should be reduced too,
but this change by itself makes things a little clearer:
we don't try to gather to a different type or
number-of-values than whatever is passed in as the value
list itself.
If some leaves have the same instructions to be vectorized, we may
incorrectly evaluate the best order for the root node (it is built for the
vector of instructions without repeated instructions and, thus, has less
elements than the root node). In this case we just can not try to reorder
the tree + we may calculate the wrong number of nodes that requre the
same reordering.
For example, if the root node is \<a+b, a+c, a+d, f+e\>, then the leaves
are \<a, a, a, f\> and \<b, c, d, e\>. When we try to vectorize the first
leaf, it will be shrink to \<a, b\>. If instructions in this leaf should
be reordered, the best order will be \<1, 0\>. We need to extend this
order for the root node. For the root node this order should look like
\<3, 0, 1, 2\>. This patch allows extension of the orders of the nodes
with the reused instructions.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D45263
If some leaves have the same instructions to be vectorized, we may
incorrectly evaluate the best order for the root node (it is built for the
vector of instructions without repeated instructions and, thus, has less
elements than the root node). In this case we just can not try to reorder
the tree + we may calculate the wrong number of nodes that requre the
same reordering.
For example, if the root node is \<a+b, a+c, a+d, f+e\>, then the leaves
are \<a, a, a, f\> and \<b, c, d, e\>. When we try to vectorize the first
leaf, it will be shrink to \<a, b\>. If instructions in this leaf should
be reordered, the best order will be \<1, 0\>. We need to extend this
order for the root node. For the root node this order should look like
\<3, 0, 1, 2\>. This patch allows extension of the orders of the nodes
with the reused instructions.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D45263
As discussed in:
https://llvm.org/PR47558
...there are several potential fixes/follow-ups visible
in the test case, but this is the quickest and safest
fix of the perf regression.
This is one (small) part of improving PR41312:
https://llvm.org/PR41312
As shown there and in the smaller tests here, if we have some member of the
reduction values that does not match the others, we want to push it to the
end (bring the matching members forward and together).
In the regression tests, we have 5 candidates for the 4 slots of the reduction.
If the one "wrong" compare is grouped with the others, it prevents forming the
ideal v4i1 compare reduction.
Differential Revision: https://reviews.llvm.org/D87772
~~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
For scalable type, the aggregated size is unknown at compile-time.
Skip instructions with scalable type to ensure the list of instructions
for vectorizeSimpleInstructions does not contains any scalable-vector instructions.
Reviewed By: RKSimon
Differential Revision: https://reviews.llvm.org/D87550
Similar to the tsan suppression in
`Utils/VNCoercion.cpp:getLoadLoadClobberFullWidthSize` (rL175034; load widening used by GVN),
the D81766 optimization should be suppressed under tsan due to potential
spurious data race reports:
struct A {
int i;
const short s; // the load cannot be vectorized because
int modify; // it overlaps with bytes being concurrently modified
long pad1, pad2;
};
// __tsan_read16 does not know that some bytes are undef and accessing is safe
Similarly, under asan, users can mark memory regions with
`__asan_poison_memory_region`. A widened load can lead to a spurious
use-after-poison error. hwasan/memtag should be similarly suppressed.
`mustSuppressSpeculation` suppresses asan/hwasan/tsan but not memtag, so
we need to exclude memtag in `vectorizeLoadInsert`.
Note, memtag suppression can be relaxed if the load is aligned to the
its granule (usually 16), but that is out of scope of this patch.
Reviewed By: spatel, vitalybuka
Differential Revision: https://reviews.llvm.org/D87538
Forward declare AAResults instead of the (old) AliasAnalysis type.
Remove includes from SLPVectorizer.cpp that are already included in SLPVectorizer.h.
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
The test example based on PR47450 shows that we can
match non-byte-sized shifts, but those won't ever be
bswap opportunities. This isn't a full fix (we'd still
match if the shifts were by 8-bits for example), but
this should be enough until there's evidence that we
need to do more (this is a borderline case for
vectorization in the first place).
Sanjay Patel