Commit Graph

4273 Commits

Author SHA1 Message Date
thomasraoux ae4cea38f1 [mlir] Add support for tensor.extract to comprehensive bufferization
Differential Revision: https://reviews.llvm.org/D105870
2021-07-13 09:54:46 -07:00
Aart Bik 7039dfc6dd [mlir][memref] adjust integration tests to new lowering passes
these tests run under the emulator and thus were overlooked

Reviewed By: ftynse

Differential Revision: https://reviews.llvm.org/D105855
2021-07-13 09:14:41 -07:00
Nicolas Vasilache af55335924 [mlir][Linalg] Better support for bufferizing non-tensor results.
Clean up corner cases related to elemental tensor / buffer type return values that would previously fail.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D105857
2021-07-13 10:27:40 +00:00
Nicolas Vasilache e312fc49ae [mlir][Linalg] Add layout specification support to bufferization.
Previously, linalg bufferization always had to be conservative at function boundaries and assume the most dynamic strided memref layout.
This revision introduce the mechanism to specify a  linalg.buffer_layout function argument attribute that carries an affine map used to set a less pessimistic layout.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D105859
2021-07-13 10:22:18 +00:00
Aart Bik 45b3cfe843 [mlir][sparse] add support for AND and OR operations
Integral AND and OR follow the simple conjunction and disjuction rules
for lattice building. This revision also completes some of the Merge
refactoring by moving the remainder parts that are merger specific from
sparsification into utils files.

Reviewed By: gussmith23

Differential Revision: https://reviews.llvm.org/D105851
2021-07-12 17:47:18 -07:00
Hanhan Wang 50529affcd [mlir][Linalg] Add 3D pooling named ops to Linalg.
Reviewed By: gysit, hanchung

Differential Revision: https://reviews.llvm.org/D105329
2021-07-12 17:26:02 -07:00
Rob Suderman f2832c2295 [mlir][tosa] Added shape propagation for TOSA pool operations.
Pool operations perform the same shape propagation. Included the shape
propagation and tests for these avg_pool2d and max_pool2d.

Differential Revision: https://reviews.llvm.org/D105665
2021-07-12 15:40:49 -07:00
Aart Bik 622eb169f6 [mlir][sparse] add restrictive versions of division support
Right now, we only accept x/c with nonzero c, since this
conceptually can be treated as a x*(1/c) conjunction for both
FP and INT as far as lattice computations go. The codegen
keeps the division though to preserve precise semantics.

See discussion:
https://llvm.discourse.group/t/sparse-tensors-in-mlir/3389/28

Reviewed By: gussmith23

Differential Revision: https://reviews.llvm.org/D105731
2021-07-12 14:59:48 -07:00
Rob Suderman 5a4e776010 [mlir][tosa] Added more shape inference for tosa ops
Added shape inference for:
- scatter
- gather
- transpose
- slice
- pad
- concat
- reduction operations

Also updated reshape for more aggressive shape inference.

Differential Revision: https://reviews.llvm.org/D105383
2021-07-12 10:04:49 -07:00
Nicolas Vasilache 6b1668397f [mlir][Linalg] Improve comprehensive bufferization for scf.yield.
Previously, comprehensive bufferization of scf.yield did not have enough information
to detect whether an enclosing scf::for bbargs would bufferize to a buffer equivalent
to that of the matching scf::yield operand.
As a consequence a separate sanity check step would be required to determine whether
bufferization occured properly.
This late check would miss the case of calling a function in an loop.

Instead, we now pass and update aliasInfo during bufferization and it is possible to
imrpove bufferization of scf::yield and drop that post-pass check.

Add an example use case that was failing previously.
This slightly modifies the error conditions, which are also updated as part of this
revision.

Differential Revision: https://reviews.llvm.org/D105803
2021-07-12 10:36:25 +00:00
Alex Zinenko 26e59cc19f [mlir] factor math-to-llvm out of standard-to-llvm
After the Math has been split out of the Standard dialect, the
conversion to the LLVM dialect remained as a huge monolithic pass.
This is undesirable for the same complexity management reasons as having
a huge Standard dialect itself, and is even more confusing given the
existence of a separate dialect. Extract the conversion of the Math
dialect operations to LLVM into a separate library and a separate
conversion pass.

Reviewed By: silvas

Differential Revision: https://reviews.llvm.org/D105702
2021-07-12 11:09:42 +02:00
Jacques Pienaar 51cbe4e587 [mlir] Fix broadcasting check with 1 values
The trait was inconsistent with the other broadcasting logic here. And
also fix printing here to use ? rather than -1 in the error.

Differential Revision: https://reviews.llvm.org/D105748
2021-07-11 20:41:33 -07:00
Itai Zukerman ebbe149a6f [mlir] Gated calls to getAsm{Result,BlockArgument}Names on whether printing ops in generic form.
Depends On D105300

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D105301
2021-07-10 17:19:37 +00:00
Alex Zinenko c282d55a38 [mlir] add support for reductions in OpenMP WsLoopOp
Use a modeling similar to SCF ParallelOp to support arbitrary parallel
reductions. The two main differences are: (1) reductions are named and declared
beforehand similarly to functions using a special op that provides the neutral
element, the reduction code and optionally the atomic reduction code; (2)
reductions go through memory instead because this is closer to the OpenMP
semantics.

See https://llvm.discourse.group/t/rfc-openmp-reduction-support/3367.

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D105358
2021-07-09 17:54:20 +02:00
Alex Zinenko d4df3825bd [mlir] don't drop undef initializers in translation to LLVM IR
LLVM IR allows globals with external linkage to have initializers, including
undef. The translation was incorrectly using undef as a indicator that the
initializer should be ignored in translation, leading to the impossibility to
create an external global with an explicit undef initializer. Fix this and use
nullptr as a marker instead.

Reviewed By: wsmoses

Differential Revision: https://reviews.llvm.org/D105631
2021-07-09 17:52:43 +02:00
Alex Zinenko 75e5f0aac9 [mlir] factor memref-to-llvm lowering out of std-to-llvm
After the MemRef has been split out of the Standard dialect, the
conversion to the LLVM dialect remained as a huge monolithic pass.
This is undesirable for the same complexity management reasons as having
a huge Standard dialect itself, and is even more confusing given the
existence of a separate dialect. Extract the conversion of the MemRef
dialect operations to LLVM into a separate library and a separate
conversion pass.

Reviewed By: herhut, silvas

Differential Revision: https://reviews.llvm.org/D105625
2021-07-09 14:49:52 +02:00
Yi Zhang 7c35aae35b Mark TensorDialect legal and PadTensor op illegal
`GeneralizePadTensorOpPattern` might generate `tensor.dim` op so the
TensorDialect should be marked legal. This pattern should also
transform all `linalg.pad_tensor` ops so mark those as illegal. Those
changes are missed from a previous change in
https://reviews.llvm.org/D105293

Reviewed By: silvas

Differential Revision: https://reviews.llvm.org/D105642
2021-07-08 15:02:22 -07:00
Nicolas Vasilache 4747e1b83b [mlir][Linalg] Fix tensor.extract_slice(linalg.init_tensor) canonicalization for rank-reducing extract
Differential Revision: https://reviews.llvm.org/D105636
2021-07-08 18:13:51 +00:00
Nicolas Vasilache 31f80393bc Revert "[mlir][MemRef] Fix DimOp folding of OffsetSizeAndStrideInterface."
This reverts commit 6c0fd4db79.

This simple implementation is unfortunately not extensible and needs to be reverted.
The extensible way should be to extend https://reviews.llvm.org/D104321.
2021-07-08 10:09:00 +00:00
Tobias Gysi abfa950d86 [mlir][linalg][python] Add exp and log to the OpDSL.
Introduce the exp and log function in OpDSL. Add the soft plus operator to test the emitted IR in Python and C++.

Reviewed By: nicolasvasilache

Differential Revision: https://reviews.llvm.org/D105420
2021-07-08 08:48:23 +00:00
Nicolas Vasilache 6c0fd4db79 [mlir][MemRef] Fix DimOp folding of OffsetSizeAndStrideInterface.
This addresses the issue reported in

https://llvm.discourse.group/t/rank-reducing-memref-subview-offsetsizeandstrideopinterface-interface-issues/3805

Differential Revision: https://reviews.llvm.org/D105558
2021-07-08 08:30:24 +00:00
Tobias Gysi 511af1b1ad [mlir][linalg] Tighter StructuredOp Verification.
Verify the number of results matches exactly the number of output tensors. Simplify the FillOp verification since part of it got redundant.

Differential Revision: https://reviews.llvm.org/D105427
2021-07-08 06:53:36 +00:00
William S. Moses 9a11c70c18 [SCF] Handle lowering of Execute region to Standard CFG
Lower SCF.executeregionop to llvm by essentially inlining the region and replacing the return

Differential Revision: https://reviews.llvm.org/D105567
2021-07-07 15:27:21 -04:00
William S. Moses eaf22ba011 [MLIR] Provide lowering of std switch to llvm switch
This patch allows lowering of std switch to llvm switch

Differential Revision: https://reviews.llvm.org/D105580
2021-07-07 15:25:55 -04:00
Sean Silva da289a174f [mlir] Allow conversion to LLVM for ElementsAttr's with size 0
The code seems to correctly handle the case that this assertion was
guarding.

Differential Revision: https://reviews.llvm.org/D105379
2021-07-07 11:15:20 -07:00
thomasraoux 291025389c [mlir][vector] Refactor Vector Unrolling and remove Tuple ops
Simplify vector unrolling pattern to be more aligned with rest of the
patterns and be closer to vector distribution.
The new implementation uses ExtractStridedSlice/InsertStridedSlice
instead of the Tuple ops. After this change the ops based on Tuple don't
have any more used so they can be removed.

This allows removing signifcant amount of dead code and will allow
extending the unrolling code going forward.

Differential Revision: https://reviews.llvm.org/D105381
2021-07-07 11:11:26 -07:00
Nicolas Vasilache d0b282e10b [mlir][Linalg] Rewrite PadTensorOp to enable its comprehensive bufferization.
Add the rewrite of PadTensorOp to InitTensor + InsertSlice before the
bufferization analysis starts.

This is exercised via a more advanced integration test.

Since the new behavior triggers folding, 2 tests need to be updated.
One of those seems to exhibit a folding issue with `switch` and is modified.

Differential Revision: https://reviews.llvm.org/D105549
2021-07-07 12:39:22 +00:00
Yi Zhang 35df2f6fbd Refactor GenericPadTensorOpVectorizationPattern
Refactor the original code to rewrite a PadTensorOp into a
sequence of InitTensorOp, FillOp and InsertSliceOp without
vectorization by default. `GenericPadTensorOpVectorizationPattern`
provides a customized OptimizeCopyFn to vectorize the
copying step.

Reviewed By: silvas, nicolasvasilache, springerm

Differential Revision: https://reviews.llvm.org/D105293
2021-07-07 11:44:32 +00:00
Nicolas Vasilache 9a0af63d05 [mlir][Linalg] Proper handling of ForOp and TiledLoopOp
The `bufferizesToMemoryRead` condition was too optimistics in the case
of operands that map to a block argument.
This is the case for ForOp and TiledLoopOp.
For such ops, forward the call to all uses of the matching BBArg.

Differential Revision: https://reviews.llvm.org/D105540
2021-07-07 11:34:05 +00:00
Adrian Kuegel 6e80e3bd1b Add Log1pOp to complex dialect.
Also add a lowering pattern from Complex to Standard/Math dialect.

Differential Revision: https://reviews.llvm.org/D105538
2021-07-07 11:33:54 +02:00
Nicolas Vasilache 0c4e538d8f [mlir][Linalg] Add an InitTensor -> DimOp canonicalization pattern.
Differential Revision: https://reviews.llvm.org/D105537
2021-07-07 08:44:54 +00:00
Srishti Srivastava 0c1a7730f5 [MLIR] Simplify affine.if having yield values and trivial conditions
When an affine.if operation is returning/yielding results and has a
trivially true or false condition, then its 'then' or 'else' block,
respectively, is promoted to the affine.if's parent block and then, the
affine.if operation is replaced by the correct results/yield values.
Relevant test cases are also added.

Signed-off-by: Srishti Srivastava <srishti.srivastava@polymagelabs.com>

Differential Revision: https://reviews.llvm.org/D105418
2021-07-07 13:02:10 +05:30
Uday Bondhugula 4acf3807e3 [MLIR] Split out GPU ops library from Transforms
Split out GPU ops library from GPU transforms. This allows libraries to
depend on GPU Ops without needing/building its transforms.

Differential Revision: https://reviews.llvm.org/D105472
2021-07-07 11:26:49 +05:30
Uday Bondhugula 715137d0c8 [MLIR] Fix memref get constant bound size and shape method
Fix FlatAffineConstraints::getConstantBoundOnDimSize to ensure that
returned bounds on dim size are always non-negative regardless of the
constraints on that dimension. Add an assertion at the user.

Differential Revision: https://reviews.llvm.org/D105171
2021-07-05 23:00:41 +05:30
Simon Camphausen 4ff440b0ef [mlir] Change custom syntax of emitc.include op to resemble C
This changes the custom syntax of the emitc.include operation for standard includes.

Reviewed By: marbre

Differential Revision: https://reviews.llvm.org/D105281
2021-07-05 16:40:05 +02:00
Adrian Kuegel bf17ee1950 Add MulOp lowering from Complex dialect to Standard/Math dialect.
The lowering handles special cases with NaN or infinity like C++.

Differential Revision: https://reviews.llvm.org/D105270
2021-07-05 12:51:51 +02:00
Markus Böck a96911c49b [mlir] Escape strings of opaque attributes
Opaque attributes that currently contain string literals can't currently be properly roundtripped as they are not printed as escaped strings. This leads to incorrect tokens being generated and the parser to almost certainly fail. This patch simply uses llvm::printEscapedString from LLVM. It escapes all non printable characters and quotes to \xx hex literals, and backslashes to two backslashes. This syntax is supported by MLIRs Lexer as well. The same function is also currently in use for the same purpose in printSymbolReference, printAttribute for StringAttr and many more in AsmPrinter.cpp.

Differential Revision: https://reviews.llvm.org/D105405
2021-07-05 12:13:36 +02:00
Adrian Kuegel 380fa71fb0 [mlir] Add LogOp lowering from Complex dialect to Standard/Math dialect.
Differential Revision: https://reviews.llvm.org/D105342
2021-07-05 09:33:45 +02:00
Chia-hung Duan db9df434fa [mlir-tblgen] Avoid ODS verifier duplication
Different constraints may share the same predicate, in this case, we
will generate duplicate ODS verification function.

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D104369
2021-07-05 10:09:41 +08:00
Aart Bik b8a021dbe3 [mlir][sparse] support for negation and subtractions
This revision extends the sparse compiler support from fp/int addition and multiplication to fp/int negation and subtraction, thereby increasing the scope of sparse kernels that can be compiled.

Reviewed By: gussmith23

Differential Revision: https://reviews.llvm.org/D105306
2021-07-02 15:55:05 -07:00
Adam Paszke 35d4593e6b Add C API files for the LLVM dialect
For now only expose a builder for the LLVM pointer type.

Reviewed By: jpienaar, ftynse

Differential Revision: https://reviews.llvm.org/D105346
2021-07-02 11:55:44 -07:00
MaheshRavishankar cdf7b661c2 [mlir][Linalg] Fix incorrect logic in deciding when to fuse reshapes by linearization.
Fusion by linearization should not happen when
- The reshape is expanding and it is a consumer
- The reshape is collapsing and is a producer.

The bug introduced in this logic by some recent refactoring resulted
in a crash.
To enforce this (negetive) use case, add a test that reproduces the
error and verifies the fix.

Differential Revision: https://reviews.llvm.org/D104970
2021-07-02 11:16:21 -07:00
Tobias Gysi f239026f89 [mlir][linalg][python] Add min operation in OpDSL.
Add the min operation to OpDSL and introduce a min pooling operation to test the implementation. The patch is a sibling of the max operation patch https://reviews.llvm.org/D105203 and the min operation is again lowered to a compare and select pair.

Differential Revision: https://reviews.llvm.org/D105345
2021-07-02 16:27:30 +00:00
Nicolas Vasilache ad0050c607 [mlir][Linalg] Add comprehensive bufferization support for TiledLoopOp (14/n)
Differential Revision: https://reviews.llvm.org/D105335
2021-07-02 14:21:08 +00:00
Adrian Kuegel 791ddb79f1 Add LogOp to Complex dialect.
Differential Revision: https://reviews.llvm.org/D105337
2021-07-02 13:15:47 +02:00
Tobias Gysi 6944f7da25 [mlir][linalg][python] Introduce python integration test folder.
Introduce an integration test folder in the test/python subfolder and move the opsrun.py test into the newly created folder. The test verifies named operations end-to-end using both the yaml and the python path.

Differential Revision: https://reviews.llvm.org/D105276
2021-07-02 07:20:34 +00:00
Tobias Gysi 3b95400f78 [mlir][linalg][python] Add max operation in OpDSL
Add the max operation to the OpDSL and introduce a max pooling operation to test the implementation. As MLIR has no builtin max operation, the max function is lowered to a compare and select pair.

Differential Revision: https://reviews.llvm.org/D105203
2021-07-02 07:12:37 +00:00
Matthias Springer e895a670f8 [mlir] Move BufferizeDimOp to Tensor/Transforms/Bufferize.cpp
Differential Revision: https://reviews.llvm.org/D105256
2021-07-02 10:05:59 +09:00
Rob Suderman 8dea784b3e [mlir][tosa] Add tosa shape inference with InferReturnTypeComponent
Added InferReturnTypeComponents for NAry operations, reshape, and reverse.
With the additional tosa-infer-shapes pass, we can infer/propagate shapes
across a set of TOSA operations. Current version does not modify the
FuncOp type by inserting an unrealized conversion cast prior to any new
non-matchin returns.

Differential Revision: https://reviews.llvm.org/D105312
2021-07-01 16:04:26 -07:00
William S. Moses e86fe368db [MLIR] Allow Affine scalar replacement to handle inner operations
Affine scalar replacement (and other affine passes, though not fixed here) don't properly handle operations with nested regions. This patch fixes the pass and two affine utilities to function properly given a non-affine internal region

This patch prevents the pass from throwing an internal compiler error when running on the added test case.

Differential Revision: https://reviews.llvm.org/D105058
2021-07-01 15:12:59 -04:00