So that the CumSum Op in OPT can get the constant that it requires to be lowered to TMTensor
---------
Co-authored-by: Rob Suderman <rob.suderman@gmail.com>
Co-authored-by: Xida Ren <xida.ren.dev@gmail.com>
Linalg has quantized specific operations. We can lower to these
operations when there is a known zeropoint and scale operations. This
allows the `convolution` to occur with lower bitwidth's, improving the
overall performance.
We were seeing some assertion failures after some checks around folders
were tightened up in LLVM:
https://github.com/llvm/llvm-project/pull/75887 . This PR essentially
moves the logic that used to be applied at the LLVM level into the
folder, which seems to be the suggested fix.
I'm not sure if the IR that caused issues for us _should_ be valid?
```
%1 = torch.aten.detach %arg0 : !torch.tensor<[1],f32> -> !torch.tensor
```
A better fix might be to create a verifier ensuring the result of
`aten.detach` has the same type as its operand.
---------
Co-authored-by: aaron-stgeorge <aaron.stgeorge@getcruise.com>
This adds an encoding field to the torch type, using the interfaces for
printing, parsing, and verification. Note that although this change
prepares adding sparsity to the torch type (as illustrated by the round
trip and invalid tests), nothing in this change depends on the actual
contents of the encoding field!
This includes custom op matching for decomposed operations and fusing
dequantization into dense operations. As a validation we compare
to the dequant+mm torch implementation.
This lifts the core of the jit_ir_importer and ltc out of the pt1
project, making them peers to it. As a side-effect of this layering, now
the "MLIR bits" (dialects, etc) are not commingled with the various
parts of the pt1 project, allowing pt1 and ltc to overlay cleanly onto a
more fundamental "just MLIR" Python core. Prior to this, the Python
namespace was polluted to the point that this could not happen.
That "just MLIR" Python core will be introduced in a followup, which
will create the space to upstream the FX and ONNX pure Python importers.
This primary non-NFC change to the API is:
* `torch_mlir.dialects.torch.importer.jit_ir` ->
`torch_mlir.jit_ir_importer`.
The rest is source code layering so that we can make the pt1 project
optional without losing the other features.
Progress on #2546.
- adds support for an optional verifier to the generated torch op
tablegen (GeneratedTorchOps.td)
- uses the above to add a verifier for the torch permute op.
Motivation: I hit an unclear error from linalg while developing a
decomposition pass for pixel_shuffle. The error would have been clearer
if the problem had been detected earlier in the invalid aten.permute op.
Testing: new tests added. To run added tests, from the base directory
run
```
./build/bin/llvm-lit test/Dialect/Torch/invalid.mlir
```
This is a first step towards the structure we discussed here:
https://gist.github.com/stellaraccident/931b068aaf7fa56f34069426740ebf20
There are two primary goals:
1. Separate the core project (C++ dialects and conversions) from the
hard PyTorch dependencies. We move all such things into projects/pt1 as
a starting point since they are presently entangled with PT1-era APIs.
Additional work can be done to disentangle components from that
(specifically LTC is identified as likely ultimately living in a
`projects/ltc`).
2. Create space for native PyTorch2 Dynamo-based infra to be upstreamed
without needing to co-exist with the original TorchScript path.
Very little changes in this path with respect to build layering or
options. These can be updated in a followup without commingling
directory structure changes.
This also takes steps toward a couple of other layering enhancements:
* Removes the llvm-external-projects/torch-mlir-dialects sub-project,
collapsing it into the main tree.
* Audits and fixes up the core C++ build to account for issues found
while moving things. This is just an opportunistic pass through but
roughly ~halves the number of build actions for the project from the
high 4000's to the low 2000's.
It deviates from the discussed plan by having a `projects/` tree instead
of `compat/`. As I was thinking about it, this will better accommodate
the follow-on code movement.
Once things are roughly in place and the CI passing, followups will
focus on more in-situ fixes and cleanups.
NonValueSemantic Ops like Add_, div_, etc. expect result DType to be the
same as the first input. However, current implementation would result in
wrong result type for case like:
```python
a = torch.randn(3, 3).half() # float16
b = torch.randn(3, 3) # float32
a += b # i.e. torch.ops.aten.add_(a, b)
```
torch expects `a` to be float16, but dtype refinement would infer
float32 type, since it's replaced by `aten.add`.
Strict symbolic shapes allow us to assume numpy-style dynamic broadcasts
never occur. This allows us to strengthen the folder for broadcasts to
cases where the rank is the same and all shapes match (including dynamic
sentinel values).
Corresponding commits:
* mlir-hlo: 16886a108eff5197f816ca0f1950cc5ff1b078d9
* stablehlo: 77a59815a82b34f7b08ed2d42a711d9920682d0e
* llvm-project: 4acc3ffbb0af5631bc7916aeff3570f448899647
* Adapt to ByteCodeOpInterface changes.
* Adapt to RegionBranchPoint changes: https://reviews.llvm.org/D159116
* Adapt inferReturnTypes to get the value from properties.
* Adapt invalid.mlir to properties syntax
* [TOSA] Align with custom assembly format change.
* [TOSA] handle change of axis to int32 type
* [TOSA] Restore improper convert to i32
Landing with Windows broken (it cannot be fixed because of the way the mlir-hlo dep is inserted). Will followup with an untangling.
---------
Co-authored-by: TatWai Chong <tatwai.chong@arm.com>
Co-authored-by: Eric Kunze <eric.kunze@arm.com>
* view_as_real test case, allow dtype in testutils.randn
* abstract python upstream func implemented
* fixed upstream dtype func, implemented view_as_real backend op
* formatted AtenViewAsRealOp, removed change in e2etest/framework
* removed test suit from reshape_like.py, because it's moved to basic.py
* implemented C-API wrapper for mlirComplexF128 type
* fixed torch.complex dtype width in MLIR and Torch MLIR, deleted float16 dtype dict
* Changed IR input of aten fft_fft unit test
* code refactored
* code refactored and fixed ci test
* refactored: removed white spaces, and rolled back to having both input/output affine expr
* refactored: deleted output affine expr to reduce redundancy
* xfail ltc backend
* removed ComplexImag and ComplexReal from torchdynamo xfail set
* copied and pasted from main branch as there's no change to be made in this file
* refactored abstract_interp_lib_gen.py
* refactored: torchtypes.td, formatted, removed commented out code
* Support brevitas custom op (#2320)
* f16 change for brevitas
* Adapt the change of brevitas quant custom op name
* Add unit tests
* Make brevitas conversions isolated
* Address the comments
---------
Co-authored-by: dan <danimal197@gmail.com>
The implementation at this place was a remnent of the times the pipeline was
run only once.
Rely instead on the backend verification, after optimizations have had an
opportunity to resolve some uncertainties. (e.g. `!torch.optional`).
* RecomposeComplexOps: Remove dead slice op
* lib/Dialect/Torch/IR/TorchOps.cpp: Fold slice ops even when they are on non-value tensors
* lib/Conversion/TorchToTosa/TorchToTosa.cpp: Fix slice start/end out of range/none
* lib/Dialect/Torch/IR/TorchOps.cpp: AtenSliceTensorOp::fold: Fold slices that go from 0:int_max
* More tests for aten.split.Tensor
In PyTorch, the `NumberType` is equal to `Union[int, float,
complex]`. However, the abstract interpretation library was treating
the `NumberType` as `Union[int, float]`, resulting in type mismatches
when reifying certain dtype functions. This commit fixes the type
inconsistency by having the abstract interpretation functions take as
an input a `Union[int, float, complex]` for the ops that take
`!torch.number` inputs.
When `use_tracing=True` is used to import a model into Torch-MLIR,
several casts get inserted in the IR to bridge the untyped inputs and
outputs with the typed body of the computation. These casts create
extra aliases of tensors that cause the current analysis in
`maximize-value-semantics` to fail.
In particular, the `maximize-value-semantics` analysis assumes that the
only valid alias right after an overwrite is the overwritten
alias. So, if there is a use of a casted version of the overwritten
alias after the overwrite, the analysis fails.
This commit improves the analysis by identifying all cast-like aliases
of the overwritten alias and allowing such aliases to be used after an
overwrite.
Because this issue only arises when using tracing, it cannot be
currently tested e2e, so only lit test is added.
This commit adds dtype functions for all the torch ops that did not
previously have one and removes the pass `RefineTypes`, since the
abstract interpretation library now takes care of all the dtype
propagation.
All dtype functions added are tested except for
- `aten.embedding`
- `aten._embedding_bag`
- `aten.embedding_bag`
These functions need a change to the testing framework to allow
specifying the actual data inside the tensor used for testing. I will
fix this in a follow up patch.
Co-authored-by: Jiahao Li <liplus17@163.com>
To keep things simple in shape functions, `Scalar` inputs are
considered `float`s. This means that when inserting the shape
functions into the IR, we must cast any `!torch.number`s into `float`s
so that the operand type matches the expected type in the shape
function. This commit adds the cast from `Scalar` to `float`.
The original design for the dtype functions outlined in
https://github.com/llvm/torch-mlir/issues/1462 was unable to properly
handle ops that take optional tensors as an input when the optional
tensor has a value of None. By the time the op gets imported into
torch-mlir, if an optional value is None, all information about the
original type is lost from the op type signature, preventing
torch-mlir from knowing if a value of None was from an optional tensor
or not, which was crucial in the original design since each tensor
argument must be turned into two separate arguments for the dtype
function.
This commit changes the interface to dtype functions such that each
tensor turns into a tuple of two ints, the first representing the rank
of the tensor and the second the dtype of the tensor. Since now there
is a one-to-one correspondence between the operands of an op and the
operands of its dtype function, there is no ambiguity about which
operand of the op corresponds with which operand of the dtype
function.
To test the implementation, this commit defines dtype function for
convolution op, which takes one optional tensor as an argument.
* LowerToBackendContract: Explicitly error out on unimplemented operator
But only reject torch.operator when results are invalid.
Otherwise it might be a custom op that the backend supports.
Currently, the op `torch.tensor_static_info_cast` will not get
canonicalized away if the result type has any shape or dtype
information. This is because `isValidSubtype` only returns true when
the tensor types being compared are exactly the same or the supertype
has no shape and dtype information. Being unable to canonicalize away
the `torch.tensor_static_info_cast` gets in the way of further
optimizations, such as shape propagation.
This commit improves `isValidSubtype` by adding logic that compares
the shapes and dtypes of the two tensor types to determine of one type
is indeed a valid subtype of the other.
Fixes https://github.com/llvm/torch-mlir/issues/1926
The data-flow analysis does not always propagate information to the
entire graph. This results in some lattice elements being
uninitialized. Currently the lattice elements are not checked to see
if they are uninitialized before rewriting the graph, potentially
resulting in invalid IR (see
https://github.com/llvm/torch-mlir/issues/1896).
This commit adds handling for uninitialized lattice elements.
This commit replaces the `tanh` dtype function, which was being used
to test the implementation of dtype functions in
a710237437, with a dtype function for
`expm1`. The dtype function for `expm1` is identical to the `tanh`
one, so the same level of testing is maintained.
Currently, there are ops getting dtype information from the
`RefineTypes` pass and ops getting dtype information from the
`TorchDtypeRefinementPipeline`. Since each pass can only propagete
dtype information for the ops it knows how to handle, some models with
many ops handled in both passes require the two dtype propagation
passes to execute many times, reaching the iteration limit set in the
`LowerToBackendContractPass`. To temporarily avoid this issue while
the migration to `TorchDtypeRefinementPipeline` is finished, this
commit switches `tanh` to `expm1`, since the latter is used a lot less
in large models.
In order to verify if a given IR satisfies the backend contract, the
verifier needs to know if decompositions took place, and if so, which
ops were decomposed and which were not.
This commit adds two arguments to `verifyBackendContractPass` to
specify if decompositions took place and which ops to consider backend
legal, similar to the arguments of `LowerToBackendContractPass`.
* [custom op] Generalize shape library logic to work with dtypes
This commit generalizes the shape library logic, so that dtype rules
for ops can also be expressed using the same mechanism. In other
words, each op can now have a shape function and a dtype function
specified in Python that is imported during lowering to calculate the
shapes and dtypes throught a program. For more information about how
to specify a dtype function, see the updated
`docs/adding_a_shape_and_dtype_function.md`.
For those not familiar with how the shape library works, the file
`docs/calculations_lib.md` provides an overview.
The current implementation of `DecomposeComplexOps` fails if an op
expected to be decomposed does not get decomposed in the first
iteration of the `createTorchSimplificationPipeline` in
`LowerToBackendContractPass`. However, some graphs require multiple
iterations of `createTorchSimplificationPipeline` to fully propagate
all statically knowable information, such as dtypes and shapes, to the
entire graph, sometimes resulting in the need to run
`DecomposeComplexOps` more than once.
This commit changes `DecomposeComplexOps` to use a greedy algorithm
for pattern application and moves the legalization check of ops to the
`LowerToBackendContractPass` to allow for the `DecomposeComplexOps` to
run more than once.
- Support for non-prefixed accessors has been removed. See:
https://reviews.llvm.org/D136727
- Rename `operands` to `methodOperands` in `prim.CallMethod` since the
name `operands` overlaps with a builtin method name. See:
https://reviews.llvm.org/D136727
- Add passes in refbackend to lower memref.subview. See:
https://reviews.llvm.org/D136377
- Replace `CopyToValueTensorOps` first in `RewriteViewLikeSubgraph` in
maximize-value-semantics.
The current implementation of the `RewriteViewLikeSubgraph` pass in
maximize-value-semantics creates temporarily invalid IR. In
particular, given a forward slice starting from a
`CopyToNonValueTensorOp` and ending in `CopyToValueTensorOp`s, the
pass first replaces all uses of the `CopyToNonValueTensorOp` with
its operand, which results in all the `CopyToValueTensorOp` users
having their operand have type `!torch.vtensor`, which is invalid.
The correct way to do things is to first replace all the
`CopyToValueTensorOp`s with their operand, and then replace all uses
of the `CopyToNonValueTensorOp` with its operand.
This only started failing now because the generated accessor
`getOperand` for the `CopyToValueTensorOp` now returns a
`TypedValue<NonValueTensorType>`, which has an assert checking that
the value returned is of the expected type.
Summary of changes:
- Replace call to `MemoryEffectOpInterface::hasNoEffect`
with `isMemoryEffectFree`.
- Make fix for the dynamic dims, since
`kDynamicSize` value changed to
`std::numeric_limits<int64_t>::min()` from `-1` in llvm
- `makeShapeLLVMCompatible` and `makeShapeTorchCompatible`
utilities convert shapes in order to remain consistent
with the Torch and MLIR semantics.
- Update tags
llvm: 147fe9de29dc13c14835127b35280c4d95c8e8ba
mhlo: 1944b5fa6062ec4c065d726c9c5d64f1487ee8c5
Signed-Off By: Vivek Khandelwal<vivek@nod-labs.com>
This commit fixes the aten.mean and aten.mean.dim op decomposition
for supporting large-sized inputs.
This commit also fixes the formatting for the file stats.py
Signed-Off By: Vivek Khandelwal<vivek@nod-labs.com>
This commit removes almost all of the valsem ops, since the value
semantics version of the ops now exist in PyTorch. The only op missing
is `aten.bernoulli_.float`. In addition, this commit also simplifies
the implementation of `aten.fill.Scalar` by moving it to the pattern
that converts elementwise ops.
This commit makes the following changes needed to update bump LLVM:
- Replace `linalg.init_tensor` with `tensor.empty` (see:
https://reviews.llvm.org/D135129)
- Replace `NoSideEffect` with `Pure` (see
https://reviews.llvm.org/D135505)
- Replace `body` region accessor for `ReduceOp` and `ReduceWindowOp`
with `getBody`
- Fix incorrect use of `tosa::ReduceSumOp` in `AtenNativeLayerNormOp`
conversion pattern. The result type of `tosa::ReduceSumOp` must have
the same rank as the input type. (see:
https://www.mlplatform.org/tosa/tosa_spec.html#_reduce_sum)
Co-authored-by: Ashay Rane <ashay@users.noreply.github.com>
Co-authored-by: Ashay Rane <ashay@users.noreply.github.com>
This commit adds lowering of `aten.div.int` and `aten.bitwise_or.Tensor`
ops. Both these ops are required in order to support bloom_560m model.
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
We were already hitting many cases where backends different in terms of
the legal ops that they wanted. This caused unnecessary coupling between
the backends. Examples:
- https://github.com/llvm/torch-mlir/pull/1161
- https://github.com/llvm/torch-mlir/pull/862
This PR centralizes all compilation to go through `torch_mlir.compile`
so that we can keep the logic centralized there. We should move these
lists closer to each backend. Especially cases like
https://github.com/llvm/torch-mlir/pull/862 where blocking a
decomposition is necessary to avoid a crash emphasize that the set of
decompositions is tightly coupled to the backend, and should be
"controlled by the backend" and not something arbitrarily tweakable.
Also:
- Fix a small bug in the way we passed through the backendLegalOps
option.
- Add better error messages in `torch_mlir.compile` for import errors.
One of the simplifications made by the pass `RefinePublicReturn`
currently only happens if the tensor in question only has one
user. However, the current method of checking this does not correctly
handle the case of a user having multiple uses of the same
tensor. This commit makes sure only unique users are considered.
This is a first step towards formalizing the set of ops in our backend
contract. The goal is to eventually formalize `torch` dialect ops into 3
categories:
1. Legal in backend contract
2. Illegal in backend contract
3. Conditionally legal in backend contract
The "conditionally legal" set are the ops that we can optionally
decompose for backends.
This patch adds relevant pass options for this throughout the compiler,
in preparation for a new set of traits which will formalize this
classification.
This introduces a new pass LowerToBackendContract (better name very
welcome) which performs the bulk of the simplifications that we do,
such as
- shape refinement
- dtype refinement
- maximizing value semantics
- inlining global slots
- decomposing complex ops
The key difference from before is that it iterates the set of
transformations, which can help to break a number of "catch-22" issues
where one simplification depends on another, the latest example being
here:
https://github.com/llvm/torch-mlir/issues/1131
This also exposed that RefineTypes was sometimes crashing/asserting for
certain inputs. This commit hardens it a bit.
Rather than a per-global-slot initializer region, we now have one for
the whole module. For example, it might look like this:
```
torch.global_slot "private" @tensor : !torch.tensor
torch.global_slot "private" @list : !torch.list<tensor>
torch.global_slot.module_initializer {
%0 = torch.tensor.literal(dense<0.0> : tensor<f32>) : !torch.tensor
%1 = torch.prim.ListConstruct %0 : (!torch.tensor) -> !torch.list<tensor>
torch.initialize.global_slots [
@tensor(%0 : !torch.tensor)
@list(%1 : !torch.list<tensor>)
]
}
```
This new structure allows GlobalizeObjectGraph to create the initializer in a
much simpler way, avoiding the need to reason about whether different slots
alias each other. Reasoning about whether slots alias each other now is the
responsibility of InlineGlobalSlots, which has to do a much more complicated
analysis, implemented using MLIR's dataflow analysis framework.
Recommended review order:
- Check out the new IR constructs in the .mlir files of various passes
- Op definitions (*.td)
- Changes to GlobalizeObjectGraph pass.
- InlineGlobalSlots pass (~total rewrite)
- Misc changes:
- Moving torchMlirAdjustStaticInformation for sharing with C++ code.
- EraseModuleInitializer pass
To make this a bit nicer, it would be good to have a `torch.module` op
with an initializer region attached. That would be more invasive though.
This change has highlighted certain aspects of our project layering
which are worth calling out. None of our backends can handle global
slots, so we enforce that there are no global slots before backend
lowering. At an earlier stage in the project, we had aspirations of
transparently handling mutable global state and such, but for reasons
described below, that is no longer a goal. So really global slots should
be seen as a progressive lowering step as part of inlining all the
IValue's in the original program (GlobalizeObjectGraph is also one such
step).
Over time, with insights from work like IREE-JAX, it has become clear
that there isn't a reliable programming model we can compile for users
where we just transparently handle mutable global state (and some other
things, like lists and dictionaries). There is a need for an "outer
program" that orchestrates more restricted subroutines of the kind we
can handle in our compile flow here. The benefit of that is that it
decouples considerations like shapes, dtypes, etc. from the program
constructs used in the outer program. As long as the outer program can
efficiently invoke (pipelining/async/etc.) high-performance
data-parallel numerical subroutines of the kind we compile in our flow
here, then there is a complete programming model. This is also
consistent with the direction of upstream PyTorch which is becoming more
tracing-based (which inherently loses a lot of program structure, which
then has to be applied back with an "outer program" orchestrating the
traced subroutines).
- Includes a canonicalizer for `aten.add.t`needed for successfully lowering the shape function
- Only offers support for statically sized index tensors when there is more than one
- Dynamic shape support remains for single indexing tensors
This commit adds verifiers to the ops `ToBuiltinTensorOp` and
`FromBuiltinTensorOp` that make sure that the input and output have
the same shape and data type.
This commit adds the decomposition for `aten.var.dim` op.
This commit also make changes in the decomposition for `aten.var` op.
Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>
This patch adds a new pass `torch-verify-conversion-to-value-semantics`,
which looks for non-value semantics tensors to catch such tensors early
during compilation.
This pass requires `torch-refine-public-return` pass to ensure that
return operations are updated to use value tensors, followed by the
canonicalize pass to remove any dead ops that may use or produce
non-value tensors.
Prior to this patch, the canonicalizers for `AtenSizeOp` and
`AtenSizeIntOp` succeeded only if the tensor operand's type information
included the size of the requested dimension(s). We can extend the set
of optimizable cases by propagating types across operations whose result
type matches the input tensor type.
Specifically, this patch enables the canonicalizers for `AtenSizeOp` and
`AtenSizeIntOp` to see past `tensor_static_info_cast`,
`copy.to_vtensor`, and `copy.to_tensor` ops until it reaches the first
op whose result type contains size information for the requested
dimensions, with a maximum bound of 6 parent lookups to avoid indefinite
compilation times. All other encountered ops cause the canonicalizer to
give up.
Xida Ren (Cedar)
Rob Suderman
Aaron St George
Aart Bik
Han-Chung Wang
Zhekun(Josh) Zhang
Stella Laurenzo
James Newling
Yuanqiang Liu
Quinn Dawkins
Bruce Kim
JianzheXiao
jinchen62
Jiawei Wu
Alexandre Rames
Matthias Gehre
Ramiro Leal-Cavazos
Jiawei Wu
Vivek Khandelwal
Maksim Levental
Ziheng Jiang
Gleb Kazantaev
ataheridezfouli-groq
Ahmed S. Taei
Tanyo Kwok
Daniel Ellis
Xiafei Qiu
Gaurav Shukla
gpetters94
Sean Silva
PhaneeshB
Kevin Kiningham
Ashay Rane