This revision adds custom rewrites for patterns that arise during linalg structured
ops vectorization. These patterns allow the composition of linalg promotion,
vectorization and removal of redundant copies.
The patterns are voluntarily limited and restrictive atm.
More robust behavior will be implemented once more powerful side effect modeling and analyses are available on view/subview.
On the transfer_read side, the following pattern is rewritten:
```
%alloc = ...
[optional] %view = std.view %alloc ...
%subView = subview %allocOrView ...
[optional] linalg.fill(%allocOrView, %cst) ...
...
linalg.copy(%in, %subView) ...
vector.transfer_read %allocOrView[...], %cst ...
```
into
```
[unchanged] %alloc = ...
[unchanged] [optional] %view = std.view %alloc ...
[unchanged] [unchanged] %subView = subview %allocOrView ...
...
vector.transfer_read %in[...], %cst ...
```
On the transfer_write side, the following pattern is rewriten:
```
%alloc = ...
[optional] %view = std.view %alloc ...
%subView = subview %allocOrView...
...
vector.transfer_write %..., %allocOrView[...]
linalg.copy(%subView, %out)
```
Differential Revision: https://reviews.llvm.org/D80728
This utility factors out the machinery required to add iterArgs and yield values to an scf.ForOp.
Differential Revision: https://reviews.llvm.org/D80656
Buffer placement can now operates on functions that return buffers. These
buffers escape from the deallocation phase of buffer placement.
Differential Revision: https://reviews.llvm.org/D80696
https://reviews.llvm.org/D79246 introduces alignment propagation for vector transfer operations. Unfortunately, the alignment calculation is incorrect and can result in crashes.
This revision fixes the calculation by using the natural alignment of the memref elemental type, instead of the resulting vector type.
If more alignment is desired, it can be done in 2 ways:
1. use a proper vector.type_cast to transform a memref<axbxcxdxf32> into a memref<axbxvector<cxdxf32>> giving a natural alignment of vector<cxdxf32>
2. add an alignment attribute to vector transfer operations and propagate it.
With this change the alignment in the relevant tests goes down from 128 to 4.
Lastly, a few minor cleanups are performed and the custom `isMinorIdentityMap` is deprecated.
Differential Revision: https://reviews.llvm.org/D80734
operands of Generic ops.
Unit-extent dimensions are typically used for achieving broadcasting
behavior. The pattern added (along with canonicalization patterns
added previously) removes the use of unit-extent dimensions, and
instead uses a more canonical representation of the computation. This
new pattern is not added as a canonicalization for now since it
entails adding additional reshape operations. A pass is added to
exercise these patterns, along with an API entry to populate a
patterns list with these patterns.
Differential Revision: https://reviews.llvm.org/D79766
This allows constructing operand adaptor from existing op (useful for commonalizing verification as I want to do in a follow up).
I also add ability to use member initializers for the generated adaptor constructors for convenience.
Differential Revision: https://reviews.llvm.org/D80667
The operand of `from_extent_tensor` is now of the same index type as the result
type of the inverse operation `to_extent_tensor`.
Differential Revision: https://reviews.llvm.org/D80283
Make ConvertKernelFuncToCubin pass to be generic:
- Rename to ConvertKernelFuncToBlob.
- Allow specifying triple, target chip, target features.
- Initializing LLVM backend is supplied by a callback function.
- Lowering process from MLIR module to LLVM module is via another callback.
- Change mlir-cuda-runner to adopt the revised pass.
- Add new tests for lowering to ROCm HSA code object (HSACO).
- Tests for CUDA and ROCm are kept in separate directories.
Differential Revision: https://reviews.llvm.org/D80142
The operation `num_elements` determines the number of elements for a given
shape.
That is the product of its dimensions.
Differential Revision: https://reviews.llvm.org/D80281
Add the two conversion operations `index_to_size` and `size_to_index` to the
shape dialect.
This facilitates the conversion of index types between the shape and the
standard dialect.
Differential Revision: https://reviews.llvm.org/D80280
This changes will catch error where C++ op are used without being
registered, either through creation with the OpBuilder or when trying to
cast to the C++ op.
Differential Revision: https://reviews.llvm.org/D80651
Summary:
Index is the proper type for storing shapes when constant folding, so
this fixes the previous code (which was using i64).
Differential Revision: https://reviews.llvm.org/D80600
I just spent a bunch of time debugging a mysterious bug that ended being due to my SmallVector getting passed to the Args&... overload instead of the MutableArrayRef overload, with disastrous results.
I appreciate the intent of this API, but for a function that does a bunch of unsafe casts, adding in potential overload confusion is just too much C++ footgun. If we end up needing this functionality, having something like a separate `packArgs(Args&...) -> SmallVector` overload would be preferable.
Turns out this API is unused and untested (even out of tree as far as I can tell, modulo the optional passing of no args to the other invoke as I fixed in this patch), so it's an easy fix -- just delete it and touch up the other overload.
Differential Revision: https://reviews.llvm.org/D80607
This allocation of a workgroup memory is lowered to a
spv.globalVariable. Only static size allocation with element type
being int or float is handled. The lowering does account for the
element type that are not supported in the lowered spv.module based on
the extensions/capabilities and adjusts the number of elements to get
the same byte length.
Differential Revision: https://reviews.llvm.org/D80411
Summary:
This includes a basic implementation for the OpenMP parallel
operation without a custom pretty-printer and parser.
The if, num_threads, private, shared, first_private, last_private,
proc_bind and default clauses are included in this implementation.
Currently the reduction clause is omitted as it is more complex and
requires analysis to see if we can share implementation with the loop
dialect. The allocate clause is also omitted.
A discussion about the design of this operation can be found here:
https://llvm.discourse.group/t/openmp-parallel-operation-design-issues/686
The current OpenMP Specification can be found here:
https://www.openmp.org/wp-content/uploads/OpenMP-API-Specification-5.0.pdf
Co-authored-by: Kiran Chandramohan <kiran.chandramohan@arm.com>
Reviewers: jdoerfert
Subscribers: mgorny, yaxunl, kristof.beyls, guansong, mehdi_amini, rriddle, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, aartbik, liufengdb, Joonsoo, grosul1, frgossen, Kayjukh, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D79410
Take advantage of equality constrains to generate the type inference interface.
This is used for equality and trivially built types. The type inference method
is only generated when no type inference trait is specified already.
This reorders verification that changes some test error messages.
Differential Revision: https://reviews.llvm.org/D80484
Now that OpBuilder is available in `build` functions, it becomes possible to
populate the "then" and "else" regions directly when building the "if"
operation. This is desirable in more structured forms of builders, especially
in when conditionals are mixed with loops. Provide new `build` APIs taking
callbacks for body constructors, similarly to scf::ForOp, and replace more
clunky edsc::BlockBuilder uses with these. The original APIs remain available
and go through the new implementation.
Differential Revision: https://reviews.llvm.org/D80527
alloc/dealloc/copies.
Add options to LinalgPromotion to use callbacks for implementating the
allocation, deallocation of buffers used for the promoted subviews,
and to copy data into and from the original subviews to the allocated
buffers.
Also some misc. cleanup of the code.
Differential Revision: https://reviews.llvm.org/D80365
Modifying the loop nest builder for generating scf.parallel loops to
not generate scf.parallel loops for non-parallel iterator types in
Linalg operations. The existing implementation incorrectly generated
scf.parallel for all tiled loops. It is rectified by refactoring logic
used while lowering to loops that accounted for this.
Differential Revision: https://reviews.llvm.org/D80188
Summary:
This op extracts an extent from a shape.
This also is the first op which constant folds to shape.const_size,
which revealed that shape.const_size needs a folder (ConstantLike ops
seem to always need folders for the constant folding infra to work).
Differential Revision: https://reviews.llvm.org/D80394
This revision expands the types of vector contractions that can be lowered to vector.outerproduct.
All 8 permutation cases are support.
The idiomatic manipulation of AffineMap written declaratively makes this straightforward.
In the process a bug with the vector.contract verifier was uncovered.
The vector shape verification part of the contract op is rewritten to use AffineMap composition.
One bug in the vector `ops.mlir` test is fixed and a new case not yet captured is added
to the vector`invalid.mlir` test.
Differential Revision: https://reviews.llvm.org/D80393
Summary:
Add DynamicMemRefType which can reference one of the statically ranked StridedMemRefType or a UnrankedMemRefType so that runner utils only need to be implemented once.
There is definitely room for more clean up and unification, but I will keep that for follow-ups.
Reviewers: nicolasvasilache
Reviewed By: nicolasvasilache
Subscribers: mehdi_amini, rriddle, jpienaar, shauheen, antiagainst, nicolasvasilache, arpith-jacob, mgester, lucyrfox, liufengdb, stephenneuendorffer, Joonsoo, grosul1, frgossen, Kayjukh, jurahul, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D80513
This revision adds the additional lowering and exposes the patterns at a finer granularity for better programmatic reuse. The unit test makes use of the finer grained pattern for simpler checks.
As the ContractionOpLowering is exposed programmatically, cleanup opportunities appear and static class methods are turned into free functions with static visibility.
Differential Revision: https://reviews.llvm.org/D80375
This still allows `if (value)` while requiring an explicit cast when not
in a boolean context. This means things like `std::set<Value>` will no
longer compile.
Differential Revision: https://reviews.llvm.org/D80497
* Enables using with more variadic sized operands;
* Generate convenience accessors for attributes;
- The accessor are named the same as their name in ODS and returns attribute
type (not convenience type) and no derived attributes.
This is first step to changing adapter to support verifying argument
constraints before the op is even created. This does not change the name of
adaptor nor does it require it except for ops with variadic operands to keep this change smaller.
Considered creating separate adapter but decided against that given operands also require attributes in general (and definitely for verification of operands and attributes).
Differential Revision: https://reviews.llvm.org/D80420
Enable inset/extract/construct composite ops as well as access chain for
cooperative matrix. ConstantComposite requires more change and will be done in
a separate patch. Also fix the getNumElements function for coopMatrix per
feedback from Jeff Bolz. The number of element is implementation dependent so
it cannot be known at compile time.
Differential Revision: https://reviews.llvm.org/D80321
Adds support for cooperative matrix support for arithmetic and cast
instructions. It also adds cooperative matrix store, muladd and matrixlength
instructions which are part of the extension.
Differential Revision: https://reviews.llvm.org/D80181
Due to similar APIs between CUDA and ROCm (HIP),
ConvertGpuLaunchFuncToCudaCalls pass could be used on both platforms with some
refactoring.
In this commit:
- Migrate ConvertLaunchFuncToCudaCalls from GPUToCUDA to GPUCommon, and rename.
- Rename runtime wrapper APIs be platform-neutral.
- Let GPU binary annotation attribute be specifiable as a PassOption.
- Naming changes within the implementation and tests.
Subsequent patches would introduce ROCm-specific tests and runtime wrapper
APIs.
Differential Revision: https://reviews.llvm.org/D80167
This reverts commit cdb6f05e2d.
The build is broken with:
You have called ADD_LIBRARY for library obj.MLIRGPUtoCUDATransforms without any source files. This typically indicates a problem with your CMakeLists.txt file
Due to similar APIs between CUDA and ROCm (HIP),
ConvertGpuLaunchFuncToCudaCalls pass could be used on both platforms with some
refactoring.
In this commit:
- Migrate ConvertLaunchFuncToCudaCalls from GPUToCUDA to GPUCommon, and rename.
- Rename runtime wrapper APIs be platform-neutral.
- Let GPU binary annotation attribute be specifiable as a PassOption.
- Naming changes within the implementation and tests.
Subsequent patches would introduce ROCm-specific tests and runtime wrapper
APIs.
Differential Revision: https://reviews.llvm.org/D80167
The subview semantics changes recently to allow for more natural
representation of constant offsets and strides. The legalization of
subview op for lowering to SPIR-V needs to account for this.
Also change the linearization to use the strides from the affine map
of a memref.
Differential Revision: https://reviews.llvm.org/D80270
Summary:
Previously, the only support partial lowering from vector transfers to SCF was
going through loops. This requires a dedicated allocation and extra memory
roundtrips because LLVM aggregates cannot be indexed dynamically (for more
details see the [deep-dive](https://mlir.llvm.org/docs/Dialects/Vector/#deeperdive)).
This revision allows specifying full unrolling which removes this additional roundtrip.
This should be used carefully though because full unrolling will spill, negating the
benefits of removing the interim alloc in the first place.
Proper heuristics are left for a later time.
Differential Revision: https://reviews.llvm.org/D80100
The SingleBlockImplicitTerminator op trait provides a function
`ensureRegionTerminator` that injects an appropriate terminator into the block
if necessary, which is used during operation constructing and parsing.
Currently, this function directly modifies the IR using low-level APIs on
Operation and Block. If this function is called from a conversion pattern,
these manipulations are not reflected in the ConversionPatternRewriter and thus
cannot be undone or, worse, lead to tricky memory errors and malformed IR.
Change `ensureRegionTerminator` to take an instance of `OpBuilder` instead of
`Builder`, and use it to construct the block and the terminator when required.
Maintain overloads taking an instance of `Builder` and creating a simple
`OpBuilder` to use in parsers, which don't have an `OpBuilder` and cannot
interact with the dialect conversion mechanism. This change was one of the
reasons to make `<OpTy>::build` accept an `OpBuilder`.
Differential Revision: https://reviews.llvm.org/D80138
Summary:
This revision refactors the Linalg tiling pass to be written as pattern applications and retires the use of the folder in Linalg tiling.
In the early days, tiling was written as a pass that would create (partially) folded and canonicalized operations on the fly for better composability.
As this evolves towards composition of patterns, the pass-specific folder is counter-productive and is retired.
The tiling options struct evolves to take a tile size creation function which allows materializing tile sizes on the fly (in particular constant tile sizes). This plays better with folding and DCE.
With the folder going away in Tiling, the check on whether subviews are the same in linalg fusion needs to be more robust. This revision also implements such a check.
In the current form, there are still some canonicalizations missing due to AffineMin/Max ops fed by scf::ForOp. These will be improved at a later time.
Differential Revision: https://reviews.llvm.org/D80267
Summary:
Additionally, this adds traits and builder methods to AssumingYieldOp
and names the input witness to the AssumingOp.
Differential Revision: https://reviews.llvm.org/D80187
Nicolas Vasilache