Commit Graph

302 Commits

Author SHA1 Message Date
A. Unique TensorFlower 4620410f18 Integrate LLVM at llvm/llvm-project@b25546a4b4
Updates LLVM usage to match
[b25546a4b406](https://github.com/llvm/llvm-project/commit/b25546a4b406)

PiperOrigin-RevId: 377077163
2021-06-02 09:32:59 -07:00
A. Unique TensorFlower 75a1c450ea [MLIR][KernelGen] Fix Windows build failure
Fix usage of default constructor. Instead, always use the parameterized
constructor and make the maximum supported rank explicit.

PiperOrigin-RevId: 377037155
2021-06-02 05:34:44 -07:00
wyzhao 968d4b8709 PR #49598: [MLIR][DISC] legalize tensor_load inserted during hlo-to-lhlo conversion
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/49598

This PR implements logic for lowering memref.tensor_load ops that are
inserted during `mhlo-legalize-to-lmhlo`
Copybara import of the project:

--
80eb377af4e02182e1aecc943a41ca5d7d1c2100 by Wenyi Zhao <reyizero@gmail.com>:

[MLIR][DISC] legalize tensor_load inserted during hlo-to-lhlo conversion

This PR implements logic for lowering memref.tensor_load ops that are
inserted during `mhlo-legalize-to-lmhlo`.

--
ac452fe3dcd591211cd5c59be9189fe2f7153b41 by Wenyi Zhao <reyizero@gmail.com>:

minor fix

--
6b36017f8632a06adbc3e05a62975fa641d0260f by Wenyi Zhao <reyizero@gmail.com>:

minor refine

--
846005cc76d0033112e47825c2e9a97790b6925f by Wenyi Zhao <reyizero@gmail.com>:

minor fix

--
f6a4becaa287d5ca323b2d152a4d0ae053730fd9 by Wenyi Zhao <reyizero@gmail.com>:

fix

--
5555749f60f7fce8f57962860ef65efccf0362ba by Wenyi Zhao <reyizero@gmail.com>:

fix

--
8873b9b6d9315c1199ca9f7c133ecf377ecd2fa6 by Wenyi Zhao <reyizero@gmail.com>:

fix

PiperOrigin-RevId: 376942547
2021-06-01 16:27:56 -07:00
A. Unique TensorFlower d1828625ab [MLIR][KernelGen] Make maximum supported rank in rank specialization configurable
The maximum supported target rank of 5 is sufficient for all operations but
`select`. Make the maximum target rank configurable in the rank specialization.
This reduces the number of generated kernels for operations that don't require
it.

PiperOrigin-RevId: 376822496
2021-06-01 06:54:31 -07:00
A. Unique TensorFlower c7c245eaf1 [MLIR][KernelGen] Add MLIR-generated Xlogy kernel
Add the first MLIR-generated kernel that relies on an in-TF lowering. Fusion for
this kernel relies on the generalized rank specialization for operation groups.

PiperOrigin-RevId: 376805435
2021-06-01 04:48:18 -07:00
A. Unique TensorFlower f16e5a3a67 [MLIR][HLO] Use canonicalization patterns in broadcast propagation pass
Replace local canonicalization patterns with those from upstream.

PiperOrigin-RevId: 376794178
2021-06-01 03:14:31 -07:00
A. Unique TensorFlower 31536431e0 [MLIR][HLO] Eliminate duplicate broadcastable constraints
PiperOrigin-RevId: 376718433
2021-05-31 13:50:23 -07:00
A. Unique TensorFlower 0f341012c6 [MLIR][HLO] Eliminate duplicate broadcastable constraints
PiperOrigin-RevId: 376715240
2021-05-31 13:08:02 -07:00
A. Unique TensorFlower 511a1db4f3 [MLIR][HLO] Use canonicalization patterns in broadcast propagation pass
Replace local canonicalization patterns with those from upstream.

PiperOrigin-RevId: 376708719
2021-05-31 12:01:26 -07:00
A. Unique TensorFlower 5f5db13715 [MLIR][HLO] Use canonicalization patterns in broadcast propagation pass
Replace local canonicalization patterns with those from upstream.

PiperOrigin-RevId: 376707588
2021-05-31 11:44:50 -07:00
A. Unique TensorFlower cc1b22e618 [HLO][Linalg] Support scalar broadcasts in point-wise converter
This is needed for operations that support this limited form of broadcasting,
namely `mhlo.select`.

PiperOrigin-RevId: 376655844
2021-05-31 03:50:23 -07:00
Hanhan Wang 402b74ed7f Fix type bug in mhlo.dynamic-update-slice lowering.
The operand type can be f32. We should not use operand type to do clamp
operations.

PiperOrigin-RevId: 376286524
2021-05-27 17:53:49 -07:00
Adrian Kuegel a4fa6afa07 [mlir][hlo] Avoid dyn_cast_or_null when called with getDefiningOp result (NFC)
PiperOrigin-RevId: 376110457
2021-05-27 00:20:42 -07:00
Hanhan Wang 28c411606f Add support for lowering mhlo.dynamic-update-slice ops to Linalg and std ops.
PiperOrigin-RevId: 376042810
2021-05-26 15:31:05 -07:00
Robert Suderman 26a0053d7d Remove linalg.indexed_generic from mhlo lowerings to linalg
IndexedGeneric is going away. Transition to using linalg.Index instead.

PiperOrigin-RevId: 376002501
2021-05-26 12:24:23 -07:00
A. Unique TensorFlower 4ebcebf31c [MLIR][HLO] Exploit scalar properties in rank specialization lowering
Take advantage of the fact that scalars are already ranked and that they are
neutral elements to broadcasting. Do not reshape scalars, do not consider them
for broadcasting, and materialize ranked operations on scalars accordingly.

PiperOrigin-RevId: 375968371
2021-05-26 09:59:13 -07:00
A. Unique TensorFlower cb46298a07 [MLIR][HLO] Support all smaller ranks in rank specialization cases
Rank specialization cases can be applied to all argument tensors of smaller
ranks than the expected maximum rank. This is crucial if all operands are
effectively scalars and the maximum reduced rank is 0.

PiperOrigin-RevId: 375712020
2021-05-25 08:38:53 -07:00
wyzhao b93e54d8a4 PR #49454: [MLIR][DISC] Upgrade to use the new `reifyReturnTypeShapes` interface.
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/49454

The new interface is more safe to be used during dialect conversion
(e.g. converting from tensor world to buffer world).
Copybara import of the project:

--
a6968072d59bec3c3bbaef0121d297e807c37c91 by Wenyi Zhao <reyizero@gmail.com>:

[MLIR][DISC] Upgrade to use the new `reifyReturnTypeShapes` interface.

The new interface is more safe to be used during dialect conversion
(e.g. converting from tensor world to buffer world).

--
55e7c6b7f2f99b99e226645a57e2433fae3e90ed by Wenyi Zhao <reyizero@gmail.com>:

minor fix

PiperOrigin-RevId: 375500273
2021-05-24 10:11:55 -07:00
Stella Laurenzo 28c4112f35 Downgrade some emitErrors in patterns to notifyMatchFailure.
PiperOrigin-RevId: 375160543
2021-05-21 14:09:23 -07:00
Hanhan Wang 1ba4c714c9 Add support for lowering mhlo.scatter ops to Linalg.
This only works for updating tensors, not add/min/max computations. It requires
the index depth to be 1 because of the limitation in Linalg. We can not compare
multiple indices without packing indices.

PiperOrigin-RevId: 375137721
2021-05-21 12:17:14 -07:00
A. Unique TensorFlower 97e6103933 [MLIR][HLO] Reshape to scalars in rank specialization
Scalars were incorrectly casted to scalar tensors when they have to be reshaped.

PiperOrigin-RevId: 375049088
2021-05-21 03:12:16 -07:00
A. Unique TensorFlower 3daf65578a [MLIR][HLO] Add scalar cases for binary rank specialization
For rank specialization clusters that have only two operands, we can materialize
two extra cases in which either of them is a scalar. This avoids redundant index
computations in these cases.

PiperOrigin-RevId: 375037390
2021-05-21 01:35:44 -07:00
Hanhan Wang cd8f585cf7 [MHLO:Linalg] Add support for lowering torch_index_select of unsigned tensors
Also fixes typos in tests.

PiperOrigin-RevId: 374979460
2021-05-20 17:03:05 -07:00
A. Unique TensorFlower 57aeb5ab16 Integrate LLVM at llvm/llvm-project@0316f3e649
Updates LLVM usage to match
[0316f3e64972](https://github.com/llvm/llvm-project/commit/0316f3e64972)

PiperOrigin-RevId: 374855085
2021-05-20 06:09:40 -07:00
A. Unique TensorFlower c62fd89663 [MLIR][HLO] Add equal shapes case to rank specialization
Also restructure lowering implementation to facilitate the addition or removal
of special cases.

PiperOrigin-RevId: 374626365
2021-05-19 05:38:42 -07:00
Stella Laurenzo 71394fb301 Properly handle if DynamicBroadcastInDimOp shape is not of index type.
* The op defines this to be index, any integer, or pred (i1).
* Many TensorFlow legalizations produce integers for the shape.

PiperOrigin-RevId: 374566113
2021-05-18 21:12:11 -07:00
Stella Laurenzo 0fe07e3814 Separate CHLO transforms for expanding compositions and lowering broadcasts.
* The former is typically invariant regardless of backend.
* The latter may need to be done differently depending on capabilities of the lowering target.

PiperOrigin-RevId: 374492924
2021-05-18 13:33:59 -07:00
A. Unique TensorFlower 6af3d2df91 [MLIR][HLO] Add rank specialization with multiple non-scalar operands
Add lowering pattern for rank specialization clusters with more than one
non-scalar operand. The lowering resembles that of the `TransformUnrankedHlo`
pass and switches cases for maximal ranks from 1 through 8.

PiperOrigin-RevId: 374377002
2021-05-18 03:02:45 -07:00
A. Unique TensorFlower 474e419729 [MLIR][HLO] Generalize rank specialization with single operand
The pattern can be generalized to also rank specialize operations with a single
non-scalar operand. Also extract helper functions that can be reused in
following specializations.

PiperOrigin-RevId: 374198381
2021-05-17 08:12:55 -07:00
A. Unique TensorFlower c514c73390 [MLIR][HLO] Extend rank specialization clustering pass
Also cluster operations that operate on same shape operands. These implicitly
satisfy the broadcasting semantics requirement. Also, add test cases for some
cases that appear in the current MLIR-generated kernels.

PiperOrigin-RevId: 374191950
2021-05-17 07:31:36 -07:00
A. Unique TensorFlower ccd70d5717 [MLIR][HLO] Add `rank-specialization-to-scf` pass
Currently the lowering is only implemented for the unary case. The n-ary case
will follow.

PiperOrigin-RevId: 374162772
2021-05-17 03:56:23 -07:00
Ben Vanik b06baae910 Fixing nondeterminism in pattern application.
The ReduceRegion* patterns are matching on the same ops as the PointwiseToLinalg*
patterns and on certain toolchains (MSVC) the order can be wrong. If the pointwise
runs first then it converts the op *within* the reduction before the reduction one
runs, leading to nested linalg op weirdness.

PiperOrigin-RevId: 373848269
2021-05-14 12:57:39 -07:00
Hanhan Wang d764806c1e [MHLO:Linalg] Add support for lowering reshape of unsigned tensors
PiperOrigin-RevId: 373461627
2021-05-12 15:14:29 -07:00
A. Unique TensorFlower 313d24bc8f [MLIR][HLO] Add `rank-specialization-cluster` pass
Add a pass to cluster unranked C/HLO operations in one
`chlo.rank_specialization_cluster` op. The C/HLO operations are moved to the
body of the operation. Later passes can use this to rank-specialize all these
operations together.

PiperOrigin-RevId: 373336725
2021-05-12 03:46:01 -07:00
A. Unique TensorFlower d8c40b691c [MLIR][HLO] Add `shape.broadcast` canonicalization to unblock broadcast moving
PiperOrigin-RevId: 372120309
2021-05-05 07:16:49 -07:00
Geoffrey Martin-Noble ac68145565 [MHLO:Linalg] Add support for lowering concat of unsigned tensors
Nothing about concat here really. Just need to plumb through the type
conversion.

PiperOrigin-RevId: 372012957
2021-05-04 15:57:54 -07:00
Geoffrey Martin-Noble 5a60793b31 [MHLO:Linalg] Add support for lowering dynamic-slice on unsigned ints
PiperOrigin-RevId: 371979004
2021-05-04 13:08:36 -07:00
Adrian Kuegel bae9936b98 Replace !any_of with none_of
No functional change.

PiperOrigin-RevId: 371868221
2021-05-04 01:55:36 -07:00
Adrian Kuegel 384b87fad0 Lower ReluGrad via chlo::BroadcastSelect.
This allows to get rid of the constraint that it needs to have a static shape.

PiperOrigin-RevId: 371862452
2021-05-04 01:03:02 -07:00
Benjamin Kramer f4414fcd66 [MHLO:Linalg] Add support for lowering unsigned ops
This strips away the signedness with a type converter, using unrealized
conversion casts. The rest is mostly mechanically pushing the original op down
the pipeline so lowerings can see the original types.

Signed types stay signless for now. This can be changed in the HLO bridge later.

I did a pass over all ops and added unsigned lowerings where they were missing.
There may be more.

Currently the lowering will die at a later stage because it doesn't understand
the unrealized casts.

PiperOrigin-RevId: 371077494
2021-04-29 02:27:35 -07:00
A. Unique TensorFlower 0569b7f7a4 [MLIR][MHLO] Generalize extent tensor cast elimination in bcast moving
PiperOrigin-RevId: 370112887
2021-04-23 10:52:50 -07:00
A. Unique TensorFlower 21e9365718 [MLIR][MHLO] Generalize extent tensor cast elimination in bcast moving
PiperOrigin-RevId: 370085141
2021-04-23 08:31:11 -07:00
A. Unique TensorFlower da5d252143 [MLIR] Merge extent tensor casts into `shape_of` ops in broadcast moving
PiperOrigin-RevId: 370058002
2021-04-23 04:44:01 -07:00
A. Unique TensorFlower 890a79641e Integrate LLVM at llvm/llvm-project@37e1458128
Updates LLVM usage to match
[37e145812855](https://github.com/llvm/llvm-project/commit/37e145812855)

PiperOrigin-RevId: 370020161
2021-04-22 22:57:08 -07:00
Hanhan Wang 49df46893c Add support for lowering variadic mhlo.reduce op.
Also add more lowering for body ops. Some MinOp and MaxOp can be legalized to
SelectOp + CompareOp.

PiperOrigin-RevId: 369891551
2021-04-22 09:50:49 -07:00
Benjamin Kramer 4d435a817e [mhlo:linalg] Add support for lowering mhlo.concatenate to Linalg ops.
This uses a indexed linalg.generic, which is rather awkward standalone but
allows fusing into the output of the concatenate and avoid to ever materialize
it in memory. I think this is the only way to get that with the current linalg
stack, fusion across a concatenate would require more infrastructure.

PiperOrigin-RevId: 369677652
2021-04-21 10:01:08 -07:00
A. Unique TensorFlower 9374a1c0c5 [MLIR] Fix merge of assuming ops
Assuming ops can only be merged if their witnesses will dominate the merged
assuming op. This is not the case if the second op's witness is a result of the
first.

PiperOrigin-RevId: 369192868
2021-04-19 04:21:08 -07:00
Adrian Kuegel db9f298505 Generate Equal and NotEqual kernels for complex types.
PiperOrigin-RevId: 368586877
2021-04-15 00:35:52 -07:00
Rahul Joshi c75cbf4ac7 [MLIR][NFC] Rename ReduceOp operands() => inputs().
- Rename to avoid confusion as operands generally includes all operands of an operation

PiperOrigin-RevId: 368479524
2021-04-14 12:08:23 -07:00
Prashant Kumar 236e7db5c0 PR #47315: [MLIR] Add concatenateOp lowering from lmhlo to Affine.
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/47315

Lowering of `concatenateOp` is added from lmhlo to Affine. The lowering
has been added as a part of `lhlo-legalize-to-affine` pass.

Signed-off-by: Prashant Kumar <prashantk@polymagelabs.com>
Copybara import of the project:

--
15314e4579f7a6901cf3475eff25962a34772eaf by Prashant Kumar <prashantk@polymagelabs.com>:

[MLIR] Add concatenateOp lowering from lmhlo to Affine.

Lowering of `concatenateOp` is added from lmhlo to Affine. The lowering
has been added as a part of `lhlo-legalize-to-affine` pass.

Signed-off-by: Prashant Kumar <prashantk@polymagelabs.com>
PiperOrigin-RevId: 368465992
2021-04-14 11:06:38 -07:00