Dynamic broadcast/reshape/iota take i32/i64 shape inputs, but users of
reification expect index shapes. Insert an appropriate cast if necessary.
PiperOrigin-RevId: 380613128
The rank specialization case for shapes which are either of the same shape or a
scalar doesn't need to compute the final result shape.
PiperOrigin-RevId: 380129316
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50073
-- Lowering of `GatherOp` is added from lmhlo to Affine. The lowering
has been added as a part of `lhlo-legalize-to-affine` pass.
Signed-off-by: Abhishek Varma <abhishek.varma@polymagelabs.com>
Copybara import of the project:
--
5b3dcd4ab31a69f305cd079b869ee35ba6dc8bf5 by Abhishek Varma <abhishek.varma@polymagelabs.com>:
[MLIR] Add GatherOp lowering from lmhlo to Affine.
-- Lowering of `GatherOp` is added from lmhlo to Affine. The lowering
has been added as a part of `lhlo-legalize-to-affine` pass.
Signed-off-by: Abhishek Varma <abhishek.varma@polymagelabs.com>
PiperOrigin-RevId: 380052157
The pass was replaced by the new generalized rank specialization and the two
passes `mhlo-rank-specialization-cluster` and `mhlo-rank-specialization-to-scf`.
PiperOrigin-RevId: 379935562
Merge `mhlo.select` into rank specialization clusters. Infer shape equalities
correctly from `mhlo.select` (and also from `mhlo.clamp`). This allows to lower
the relu kernel completely flattened.
PiperOrigin-RevId: 379925793
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50271
support hlo-to-lhlo conversion for GatherOp and DynamicGatherOp
Copybara import of the project:
--
117a1b1bcaac7ecc5224b02863eede5c1b9618fe by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] Bufferize GatherOp and DynamicGatherOp
PiperOrigin-RevId: 379801972
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50020
This pass implements the logic to group kLoop/kInput fusion patterns on
buffer level. The reason for this is that we can avoid a lot of
headaches to handle `shape-only` consumers specially (e.g. memref.dim,
shape.shapeOf) since shapes are already resolved in buffer world. It may
be better to move this pass to tensor level after more shape
inference/constraint infras are ready on mhlo level.
Copybara import of the project:
--
e31f8344b59aa9860097197585215ea1689b8ff4 by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] support fusion on buffer
This pass implements the logic to group kLoop/kInput fusion patterns on
buffer level. The reason for this is that we can avoid a lot of
headaches to handle `shape-only` consumers specially (e.g. memref.dim,
shape.shapeOf) since shapes are already resolved in buffer world. It may
be better to move this pass to tensor level after more shape
inference/constraint infras are ready on mhlo level.
--
35f2eb2791241b0ab5db1ddcaf1b4006278ddccf by Wenyi Zhao <reyizero@gmail.com>:
fix
--
923c8d61f7fe00a2a0df22d5be396508f0667964 by Wenyi Zhao <reyizero@gmail.com>:
fix sanity check failure
PiperOrigin-RevId: 379743424
The operations allow for a limited form of broadcasting which allows some
operands to be scalars. As such they are neither strictly `Elementwise`, nor
`Broadcasting`. They do fulfill the requirements for `BroadcastingElementwise`
though.
PiperOrigin-RevId: 379719961
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/49919
We are porting our MLIR-based dynamic shape compiler to tf community (From OP def, Patttern, to Optimization pass, etc).
This is the 5th PR about tf2mhlo pattern conversion, which including ConvertUnpackOpDynamic, ConvertSignOpDynamic, ConvertSigmoidGradOpDynamic.
The rest pattern conversions we will add:
- ConvertSqueezeOpxxx
- ConvertStridedSliceOpxxx
- ConvertPrintOp
Copybara import of the project:
--
21b3c3eb05b12956bcdb8b98cc54d9371dbf034d by azazhu <azazhu@gmail.com>:
[MLIR][DISC] pattern conversion from tf2mhlo: ConvertUnpackOpDynamic, ConvertSignOpDynamic, ConvertSigmoidGradOpDynamic
--
634630a4e2e426357290650bd579b35efecab5b3 by azazhu <azazhu@gmail.com>:
[MLIR][DISC] refine ConvertUnpackOpDynamic, ConvertSignOpDynamic, ConvertSigmoidGradOpDynamic
--
39a2bedd6dafb369ae960c5197b7a352bfdfbc80 by azazhu <azazhu@gmail.com>:
add RealDynamicSliceOp's canonicalize and fix CI
--
a1c38dd0963d602ed4812da0d77a096a95920ddb by azazhu <azazhu@gmail.com>:
fix CI for ConvertUnpackOpDynamic
--
5a8b4eb389ed6dc554104356c37f2f1550802b8c by azazhu <azazhu@gmail.com>:
fix typo in ConvertSigmoidGradOpDynamic
PiperOrigin-RevId: 379521079
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50236
support hlo-to-lhlo conversion for TransposeOp and ConcatenateOp
Copybara import of the project:
--
62860e717f2a14fbd3ddfb634aa6ff132d245a72 by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] Bufferize TransposeOp and ConcatenateOp
--
ce2ff57c1edee1172cd2f36346cc0b34ec1c7467 by Wenyi Zhao <reyizero@gmail.com>:
fix
PiperOrigin-RevId: 379330954
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50191
DISC is a e2e flow, including both compiler side and runtime side. For
runtime side, we have different targeting environments (e.g. tensorflow,
pytorch, or sometimes even a standalone binary). In order to simplify
the design of the compiler side, we design a Runtime Abstraction Layer
(RAL) to sperate the compiler side and runtime side. Thus the compiler
side only need to target RAL itself and it is the responsibility of RAL
to handle the differences between different targeting environments.
One of the most important functions of RAL is to manage stateful
resources. To this end, it provides a context object, and hides all
stateful operations behind this context, thus the compiler side itself
doesn't need to care about the resource initialization. For example, a
kernel must be loaded before it can be launched on GPU. However, the
loading operation should only be taken once during the whole lifetime of
the context in order to achieve the best performance. Based on the
initialization-free interfaces provided by RAL, compiler side can focus
on its core optimization logic and lets the RAL to manage the resource
status.
The context mentioned above is passed as a parameter to the entry
function and all RAL APIs should always use the context as their first
argument. This CR also provides a pass to help to ensure this property.
The pass rewrites the entry function to make sure their first argument
is the context. For entry function, the pass also rewrites its inputs
and outputs. To be concrete, all the original inputs and outputs of the
entry function are received from and sent to RAL through a sequence of
RAL API calls correspondingly. The motivation behind this is to hide the
implementation details of I/Os. This design may also potentially enable
partial execution of the compiled module when some of the inputs are
ready.
Copybara import of the project:
--
c4f20a89aed71181e75bcc5265723b88bde23240 by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] Add RAL (Runtime abstraction layer) Dialect
DISC is a e2e flow, including both compiler side and runtime side. For
runtime side, we have different targeting environments (e.g. tensorflow,
pytorch, or sometimes even a standalone binary). In order to simplify
the design of the compiler side, we design a Runtime Abstraction Layer
(RAL) to sperate the compiler side and runtime side. Thus the compiler
side only need to target RAL itself and it is the responsibility of RAL
to handle the differences between different targeting environments.
One of the most important functions of RAL is to manage stateful
resources. To this end, it provides a context object, and hides all
stateful operations behind this context, thus the compiler side itself
doesn't need to care about the resource initialization. For example, a
kernel must be loaded before it can be launched on GPU. However, the
loading operation should only be taken once during the whole lifetime of
the context in order to achieve the best performance. Based on the
initialization-free interfaces provided by RAL, compiler side can focus
on its core optimization logic and lets the RAL to manage the resource
status.
The context mentioned above is passed as a parameter to the entry
function and all RAL APIs should always use the context as their first
argument. This CR also provides a pass to help to ensure this property.
The pass rewrites the entry function to make sure their first argument
is the context. For entry function, the pass also rewrites its inputs
and outputs. To be concrete, all the original inputs and outputs of the
entry function are received from and sent to RAL through a sequence of
RAL API calls correspondingly. The motivation behind this is to hide the
implementation details of I/Os. This design may also potentially enable
partial execution of the compiled module when some of the inputs are
ready.
--
1991d4f80ab6087943956e1c0fec4940a22ab08d by Wenyi Zhao <reyizero@gmail.com>:
fix
PiperOrigin-RevId: 379317586
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50211
support hlo-to-lhlo conversion for RealDynamicSliceOp and ReduceOp
Copybara import of the project:
--
c417b336670a1fc256f7026dfe8080e46d13d79a by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] Bufferize RealDynamicSliceOp and ReduceOp
PiperOrigin-RevId: 378972113
This just adds support for it in the op, but keeps the production/uses as is (e.g., single tensor or tuple) matching what XLA export requires. In follow up here, would be to add pass for export to retuple and then the canonical form could be changed. Tuple'ing given control flow via regions & multi-result operations does not add representational power and all the get_tuple_element ops obscure the computation.
The old form allowed single tensor or tuple. The new variadic number of tensor or tuples as tuples may be nested, so the input could have (Tensor<..>, Tuple<Tensor<...>, Tuple<...>, ...>, Tensor<...>) and HLO_Tensor doesn't allow Tuples.
PiperOrigin-RevId: 378934388
Find shape equivalence classes among the operands and use them for better rank
specialization. If all operands are known to be of the same shape, we can
flatten them to rank one. If there are two shape equivalence classes, we can
generalize the scalar rank specialization cases.
PiperOrigin-RevId: 378844575
Imported from GitHub PR https://github.com/tensorflow/tensorflow/pull/50100
support hlo-to-lhlo conversion for DynamicIotaOp and DynamicPadOp
Copybara import of the project:
--
c3aae94954e35d3f8ad265f619ef9765665a5115 by Wenyi Zhao <reyizero@gmail.com>:
[MLIR][DISC] Bufferize DynamicIotaOp and DynamicPadOp
--
adc6996d70b804d61310d56a33fac975d70c8636 by Wenyi Zhao <reyizero@gmail.com>:
minor
PiperOrigin-RevId: 378733284
We should upcast F16 to F32 to prevent precision loss.
E.g. cosh(-9) would evaluate to 4042 previously instead of 4052.
This allows to enable the MLIR generated kernel for F16 type.
Also move template instantiation for Sinh to inside the #ifdef block.
This was missed in a previous commit.
PiperOrigin-RevId: 378635042
When merging rank specialization clusters, avoid duplicating operands. A fewer
number of operands usually allows better rank specialization.
PiperOrigin-RevId: 378445946
colin.liang
A. Unique TensorFlower
Benjamin Kramer
Adrian Kuegel
Abhishek Varma
Mehdi Amini
Wenyi Zhao
Hanhan Wang
Feiwen
Chris Jones
Geoffrey Martin-Noble
Rahul Joshi
Jacques Pienaar