Gather nodes are vectorized as simply vector of the scalars instead of
relying on the actual node. It leads to the fact that in some cases
we may miss incorrect transformation (non-matching set of scalars is
just ended as a gather node instead of possible vector/gather node).
Better to rely on the actual nodes, it allows to improve stability and
better detect missed cases.
Differential Revision: https://reviews.llvm.org/D135174
Replace custom code to check if only the first lane is used by generic
helper `onlyFirstLaneUsed`. This enables VPlan-based sinking in a few
additional cases and was suggested in D133760.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D136368
Epilogue loop vectorization is a feature in the vectorize intended to avoid running fully scalar code when the vector length of the main loop turns out to be either longer than the trip count of the actual loop, or with a huge remainder.
In practice, this feature appears to not have been well tuned. I honestly don't think it should be on by default at all, but it definitely shouldn't be on for RISCV. Note that other targets have also disabled it, but they've done so via disabling interleaving - which is, well, completely unrelated - and we don't want to do that for RISCV.
In the near term, many examples I'm seeing have terrible codegen for epilogue vectorization. We are greatly increasing code size for little value at reasonable VLEN values for small types. In the long term, the cases that epilogue vectorization are intended to handle are likely better handled via tail folding on RISCV.
As an aside, I also don't really trust the correctness of epilogue vectorization. The code structure is such that otherwise straight forward changes sometimes break only epilogue vectorization. The reuse of an existing vplan without careful validation opens significant room for nasty bugs. Given how rarely the code is exercised, that is not a good combination.
As such, this patch introduces a TTI hook, and completely disables epilogue vectorization on RISCV.
Differential Revision: https://reviews.llvm.org/D136695
Improve O(N^2) to O(N) in some cases, reduce number of allocations by
reserving memory.
Also, improve analysis of loads reduction values to avoid analysis
of not vectorizable cases.
The code in buildScalarSteps already properly handles creating the
scalar induction values with VF = 1. Use it directly instead of using
extra code to handle that case.
Suggested by @Ayal in D133760.
@Ayal suggested a better named helper than using `!getDef()` to check if
a value is invariant across all parts.
The property we are using here is that the VPValue is defined outside
any vector loop region. There's a TODO left to handle recipes defined in
pre-header blocks.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D133666
Generalized the cost model estimation. Improved cost model estimation
for repeated scalars (no need to count their cost anymore), improved
cost model for extractelement instructions.
cpu2017
511.povray_r 0.57
520.omnetpp_r -0.98
521.wrf_r -0.01
525.x264_r 3.59 <+
526.blender_r -0.12
531.deepsjeng_r -0.07
538.imagick_r -1.42
Geometric mean: 0.21
Differential Revision: https://reviews.llvm.org/D115757
Generalized the cost model estimation. Improved cost model estimation
for repeated scalars (no need to count their cost anymore), improved
cost model for extractelement instructions.
cpu2017
511.povray_r 0.57
520.omnetpp_r -0.98
521.wrf_r -0.01
525.x264_r 3.59 <+
526.blender_r -0.12
531.deepsjeng_r -0.07
538.imagick_r -1.42
Geometric mean: 0.21
Differential Revision: https://reviews.llvm.org/D115757
We can't assume that operand 0 is the negated operand because
the matcher handles "fsub -0.0, X" (and also +0.0 with FMF).
By capturing the extract within the match, we avoid the bug
and make the transform more robust (can't assume that this
pass will only see canonical IR).
Instead of checking whether a map entry exists to decide if we should
initialize it or add to it, we can rely on the map entry being constructed
and initialized to 0 before the addition happens.
For the std::max case, I've made a reference to the map entry to
avoid looking it up twice.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D135977
When generating masked gathers nodes, SLP vectorizer accounts the cost
of the GEPs for loads as part of the scalar-vector transformation cost
estimation. But it does not do it for vectorized loads/stores, while it
may completely remove some of the GEPs completely. Because of this in
some cases masked gather operation can be much more profitable rather
than regular vectorization (masked-gather cost + vector GEP - scalar
loads + GEPs comparing to vectorized loads - scalar loads).
Added the analysis of the removed scalarGEPs for vectorized load/store nodes for better cost estimation.
Differential Revision: https://reviews.llvm.org/D135282
Need to set the insertpoint for extractelement to point to the first
instruction in the node to avoid possible crash during external uses
combine process. Without it we may endup with the incorrect
transformation.
Differential Revision: https://reviews.llvm.org/D135591
Freeze instruction in some cases makes codegen worse, so need to be very
careful when emitting it. Instead improve analysis in isUndefVector
function to generate mask of unused elements and use it in the analysis.
Differential Revision: https://reviews.llvm.org/D135382
optimizeInductions may leave dead recipes which can prevent sinking.
Sinking on the other hand should not introduce new dead recipes, so
clean up dead recipes before sinking.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D133762
insertelt DestVec, (fneg (extractelt SrcVec, Index)), Index --> shuffle DestVec, (fneg SrcVec), Mask
This is a specialized form of what could be a more general fold for a binop.
It's also possible that fneg is overlooked by SLP in this kind of
insert/extract pattern since it's a unary op.
This shows up in the motivating example from #issue 58139, but it won't solve
it (that probably requires some x86-specific backend changes). There are also
some small enhancements (see TODO comments) that can be done as follow-up
patches.
Differential Revision: https://reviews.llvm.org/D135278
Added analysis for invariant extractelement instructions and improved
detection of the CSE blocks for generated extractelement instructions.
Differential Revision: https://reviews.llvm.org/D135279
In the canonical form of the shuffle the poison/undef operand is the
second operand, the patch tries to emit canonical form for partial
vectorization of the buildvector sequence.
Also, this patch starts emitting freeze instruction for shuffles with undef indices if the second shuffle operan is undef, not poison. It is an initial step to D93818, where undef mask element are treated as returning poison value.
Differential Revision: https://reviews.llvm.org/D134377
Use LoopAccessInfoManager directly instead of various GetLAA lambdas.
Depends on D134608.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D134609
At the moment, LoopAccessAnalysis is a loop analysis for the new pass
manager. The issue with that is that LAI caches SCEV expressions and
modifications in a loop may impact SCEV expressions in other loops, but
we do not have a convenient way to invalidate LAI for other loops
withing a loop pipeline.
To avoid this issue, turn it into a function analysis which returns a
manager object that keeps track of the individual LAI objects per loop.
Fixes#50940.
Fixes#51669.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D134606
Revert rGef89409a59f3b79ae143b33b7d8e6ee6285aa42f "Fix 'unused-lambda-capture' gcc warning. NFCI."
Revert rG926ccfef032d206dcbcdf74ca1e3a9ebf4d1be45 "[SLP] ScalarizationOverheadBuilder - demand all elements for scalarization if the extraction index is unknown / out of bounds"
Revert ScalarizationOverheadBuilder sequence from D134605 - when accumulating extraction costs by Type (instead of specific Value), we are not distinguishing enough when they are coming from the same source or not, and we always just count the cost once. This needs addressing before we can use getScalarizationOverhead properly.
Factor out the logic to create induction resume values for a specific
induction. This will be used in D92132 to support widened IVs during
epilogue vectorization.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D134211
Interestingly, MathExtras.h doesn't use <cmath> declaration, so move it out of
that header and include it when needed.
No functional change intended, but there's no longer a transitive include
fromMathExtras.h to cmath.
Fixes#57572
Generally LICM pass is responsible for sinking out code that calculates
invariant address inside loop as it only needed to be calculated once.
But in rare case it does not happen we will not be vectorizing the
loop.
Differential Revision: https://reviews.llvm.org/D133687
This is purely NFC restructure in advance of a change which actually exposes zero strides. This is mostly because I find this interface confusing each time I look at it.
Follow up to D133580; adjust the cost model to prefer uniform store lowering for scalable stores which are unpredicated.
The impact here isn't in the uniform store lowering quality itself. InstCombine happily converts the scatter form into the single store form. The main impact is in letting the rest of the cost model make choices based on the knowledge that the vector will be scalarized on use.
Differential Revision: https://reviews.llvm.org/D134460
Instead of accumulating all extraction costs separately and then adjusting for repeated subvector extractions, this patch collects all the extractions and then converts to calls to getScalarizationOverhead to improve the accuracy of the costs.
I'm not entirely satisfied with the getExtractWithExtendCost handling yet - this still just adds all the getExtractWithExtendCost costs together - it really needs to be replaced with a "getScalarizationOverheadWithExtend", but that will require further refactoring first.
This replaces my initial attempt in D124769.
Differential Revision: https://reviews.llvm.org/D134605
The dependent code has been changed quite a lot since 151c144 which
b73d2c8 effectively reverts. Now we run into a case where lowering
didn't expect/support the behavior pre 151c144 any longer.
Update the code dealing with scalable pointer inductions to also check
for uniformity in combination with isScalarAfterVectorization. This
should ensure scalable pointer inductions are handled properly during
epilogue vectorization.
Fixes#57912.
When store vectorization is infeasible, it's helpful to have a debug logging indication of why. A case I've hit a couple times now is accidentally using -march instead of -mtriple and getting the default TTI results. This causes max-vf to become 1, and thus hits the added logging line.
We allow the target to report different costs depending on properties of the operands; given this, we have to make sure we pass the right set of operands and account for the fact that different scalar instructions can have operands with different properties.
As a motivating example, consider a set of multiplies which each multiply by a constant (but not all the same constant). Most of the constants are power of two (but not all).
If the target doesn't have support for non-uniform constant immediates, this will likely require constant materialization and a non-uniform multiply. However, depending on the balance of target costs for constant scalar multiplies vs a single vector multiply, this might or might not be a profitable vectorization.
This ends up basically being a rewrite of the existing code. Normally, I'd scope the change more narrowly, but I kept noticing things which seemed highly suspicious, and none of the existing code appears to have any test coverage at all. I think this is a case where simply throwing out the existing code and starting from scratch is reasonable.
This is a follow on to Alexey's D126885, but also handles the arithmetic instruction case since the existing code appears to have the same problem.
Differential Revision: https://reviews.llvm.org/D132566
The mul by constant costmodels handle power-of-2 constants, but not negated-power-of-2, despite the backends handling both.
This patch adds the OperandValueProperties::OP_NegatedPowerOf2 enum and wires it for use for basic mul cost analysis and SLP handling.
Fixes#50778
Differential Revision: https://reviews.llvm.org/D111968
This extends the previously added uniform store case to handle stores of loop varying values to a loop invariant address. Note that the placement of this code only allows unpredicated stores; this is important for correctness. (That is "IsPredicated" is always false at this point in the function.)
This patch does not include scalable types. The diff felt "large enough" as it were; I'll handle that in a separate patch. (It requires some changes to cost modeling.)
Differential Revision: https://reviews.llvm.org/D133580
The compiler does not reorder the gather nodes with reused scalars, just
does it for opernads of the user nodes. This currently does not affect
the compiler but breaks internal logic of the SLP graph. In future, it
is supposed to actually use all nodes instead of just list of operands
and this will affect the vectorization result.
Also, did some early check to avoid complex logic in cost estimation
analysis, should improve compiler time a bit.
Epilogue vectorization uses isScalarAfterVectorization to check if
widened versions for inductions need to be generated and bails out in
those cases.
At the moment, there are scenarios where isScalarAfterVectorization
returns true but VPWidenPointerInduction::onlyScalarsGenerated would
return false, causing widening.
This can lead to widened phis with incorrect start values being created
in the epilogue vector body.
This patch addresses the issue by storing the cost-model decision in
VPWidenPointerInductionRecipe and restoring the behavior before 151c144.
This effectively reverts 151c144, but the long-term fix is to properly
support widened inductions during epilogue vectorization
Fixes#57712.
My understanding is that NoImplicitFloat, despite it's name, is
supposed to disable all vectors not just float vectors.
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D134084
Added the mask and the analysis of the buildvector sequence in the
isUndefVector function, improves codegen and cost estimation.
Metric: SLP.NumVectorInstructions
Program SLP.NumVectorInstructions
results results0 diff
test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test 27362.00 27360.00 -0.0%
Metric: size..text
Program size..text
results results0 diff
test-suite :: External/SPEC/CFP2017rate/508.namd_r/508.namd_r.test 805299.00 806035.00 0.1%
526.blender_r - some extra code is vectorized.
508.namd_r - some extra code is optimized out.
Differential Revision: https://reviews.llvm.org/D133891
Adding the pre-header to CSEBlocks ensures instructions are CSE'd even
after hoisting.
This was original discovered by @atrick a while ago.
Reviewed By: ABataev
Differential Revision: https://reviews.llvm.org/D133649
If the reused scalars are clustered, i.e. each part of the reused mask
contains all elements of the original scalars exactly once, we can
reorder those clusters to improve the whole ordering of of the clustered
vectors.
Differential Revision: https://reviews.llvm.org/D133524
Use VPExpandSCEVRecipe to expand the step of pointer inductions. This
cleanup addresses a corresponding FIXME.
It should be NFC, as steps for pointer induction must be constants,
which makes expansion trivial.
This extends the safe-divisor widening scheme recently added for scalable vectors to handle fixed vectors as well.
Differential Revision: https://reviews.llvm.org/D132591
VPReplicateRecipe::isUniform actually means uniform-per-parts, hence a
scalar instruction is generated per-part.
This is a potential alternative D132892. For now the current patch only
catches cases where the address is trivially invariant (defined outside
VPlan), while D132892 catches any address that is considered invariant
by SCEV AFAICT.
It should be possible to hoist fully invariant recipes feeding loads out
of the vector loop region as well, but in practice LICM should do that
already.
This version of the patch artificially limits this to loads to make it
easier to compare, but this restriction should be easily liftable.
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D133019
Need either follow the original order of the operands for bool logical
ops, or emit freeze instruction to avoid poison propagation.
Differential Revision: https://reviews.llvm.org/D126877
This patch moves the cost-based decision whether to use an intrinsic or
library call to the point where the recipe is created. This untangles
code-gen from the cost model and also avoids doing some extra work as
the information is already computed at construction.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D132585
This is NOT nfc. Specifically, the following behavior changes:
* Pointers are now allowed. Both uniform, and constants.
* FP uniform non-constants can now be recognized.
* FP undefs are no longer considered constant. This matches int behavior which we had tests for. FP behavior was untested. Its not clear to me int behavior is reasonable, but it's what tests seem to expect, so go with minimum impact for now.
The pointer operands for the ScatterVectorize node may contain
non-instruction values and they are not checked for "already being
vectorized". Need to check that such pointers are already vectorized and
gather them instead of trying to build vectorize node to avoid compiler
crash.
Differential Revision: https://reviews.llvm.org/D132949
Removed EnableFP parameter in getOperandInfo function since it is not
needed, the operands kinds also controlled by the operation code, which
allows to remove extra check for the type of the operands. Also, added
analysis for uniform constant float values.
This change currently does not trigger any changes in the code since TTI
does not do analysis for constant floats, so it can be considered NFC.
Tested with llvm-test-suite + SPEC2017, no changes.
Differential Revision: https://reviews.llvm.org/D132886
I keep finding myself needing to rule this out as a possible source of scalarization, so add debug output like we have for other instructions we decide to scalarize.
This patch changes order of searching for reductions vs other vectorization possibilities.
The idea is if we do not match a reduction it won't be harmful for further attempts to
find vectorizable operations on a vector build sequences. But doing it in the opposite
order we have good chance to ruin opportunity to match a reduction later.
We also don't want to try vectorizing binary operations too early as 2-way vectorization
may effectively prohibit wider ones leading to producing less effective code.
Differential Revision: https://reviews.llvm.org/D132590
When estimating the cost of the in-tree vectorized scalars in
buildvector sequences, need to take into account the vectorized
insertelement instruction. The top of the buildvector seuences is the
topmost vectorized insertelement instruction, because it will have
> than 1 use after the vectorization.
For the affected test case improves througput from 21 to 16 (per
llvm-mca).
Differential Revision: https://reviews.llvm.org/D132740
This addresses a suggestion to simplify the check from D131989. This
also makes it easier to ensure that VPHeaderPHIRecipe::classof checks
for all header phi ids.
Add verification that VPHeaderPHIRecipes are only in header VPBBs. Also
adds missing checks for VPPointerInductionRecipe to
VPHeaderPHIRecipe::classof.
Split off from D119661.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D131989
I'd extracted isUniform, and Florian moved isUniformAfterVectorization out of VPlan at basically the same time. Let's go ahead and merge them.
For the VPTransformState::get path, a VPValue without a def (which corresponds to an external IR value outside of VPLan) is explicitly handled above the uniform check. On the scalarizeInstruction path, I'm less sure why the change isn't visible, but test cases which would seem likely to hit it were already being handled as uniform through some other mechanism. It would be correct to consider values defined outside of vplan uniform here.
Alexey Bataev