I've upstreamed the necessary quantized linalg Op with the
"channel-first" ordering used by torch
(https://github.com/llvm/llvm-project/pull/107740) for 2d convolution.
This patch changes the lowering for the quantized 2d case of
`aten.convolution` accordingly, which saves three transpositions per
convolution (input, weights, result) and therefore removes the
requirement to try to optimize these away in downstream passes.
Adds onnx ConvTranspose support for autopadding
(https://github.com/nod-ai/SHARK-ModelDev/issues/839).
- Adds support for attribute auto_pad="SAME_UPPER" or "SAME_LOWER" which
will automatically calculate padding of input based on output shape.
- Adds support, during auto-padding, for output_shape=[H,W] which
overrides the default output shape of input_shape[i]*stride[i] (for
spatial dimensions only).
- Adds lit test for auto-padding.
- Tests are added by https://github.com/nod-ai/SHARK-TestSuite/pull/370
NOTE: ConvTranspose still doesn't support asymmetric padding, therefore
multiple original onnx tests still won't pass.
- Add Torch to TOSA legalization for the following ops:
+ aten.empty.memory_format
+ aten.scatter.src
+ aten.slice_scatter
+ aten.diag_embed
- Update xfail_sets.py with new e2e results
- Update basic.mlir with new LIT tests
Change-Id: I817ecf207bcfcf97ca54f30c10c76c4f0f4145ae
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Torch-to-linalg pass fails for `EmbeddingBag` when the training only
specific properties of the operator are set to `true.` For instance,
this operator's `sparse` input/property is training-specific, and if the
value of this property is `true,` the existing lowering bails out.
However, we don't need to check for training-specific parameters and
bailout from the legalization since we don't care about these properties
during the eval/inference mode.
---------
Co-authored-by: Hanumanth Hanumantharayappa <hhanuman@ah-hhanuman-l.dhcp.mathworks.com>
# Description
Implementation of the op for `torch.aten.unfold`: [TorchToLinalg Op
Support #347](https://github.com/nod-ai/SHARK-ModelDev/issues/849)
Documentation of op can be found here: [PyTorch
Docs](https://pytorch.org/docs/stable/generated/torch.Tensor.unfold.html)
For this op, we apply a sliding window of some `size` along a single
`dimension`, with `step` in between iterations.
`Declaration: aten::unfold(Tensor(a) self, int dimension, int size, int
step) -> Tensor(a)`
The resulting `unfolded` tensor modifies the shape of `dimension` to be
equal to the number of blocks that the sliding windows extracts/inserts,
with an additional dimension of `size` appended (the number of cols of
the output tensor directly translates from the size of the sliding
window).
So if we had a tensor of rank 3 (A x B x C), with dimension = 1, size =
2 and step = 2:
(A x B x C) |=> (A x (B - size) // step + 1 x C x size)
After extracting the window from the input tensor, we insert the (1 x
size) slice into the output tensor. We can make this simpler by mapping
the output indices from the input indices, like they do in the official
implementation:
[PyTorch
Code](https://github.com/pytorch/pytorch/blob/main/torch/_inductor/lowering.py#L1694)
- Support Bidirectional LSTM (utilising the forward LSTM layer with
flipped Inputs and Outputs)
- Support layout 1
- Support default cases for attr `clip` and `input_forget`
- Support returning partial outputs (1-3)
- fixes for alt_e2e_tests lstm tests (1,2,3)
This commit adds the support for the 1-d depthwise convolution as a
special case of 1-d group convolution.
Signed-Off By: Vivek Khandelwal <vivekkhandelwal1424@gmail.com>
This commit adds the support for negative step values in
aten.slice.Tensor op. Although, PyTorch does not allow negative step
value for slice op but the Onnx.Slice op supports negative step value
which eventually lowers to torch.aten.slice.Tensor op. Hence, the
support is added for handling those kind of values during the
Torch->Linalg lowering of aten.slice.Tensor op.
Signed-Off By: Vivek Khandelwal <vivekkhandelwal1424@gmail.com>
- Add Torch to TOSA lowering for aten.fill.Scalar/Tensor, aten.flip, and
aten.round
- Fix torchScalarToTosaTensor function to correctly convert Torch scalar
input to TOSA tensor
- Update xfail_sets.py with new e2e results
- Update basic.mlir with LIT tests for new ops
Change-Id: If1e42c2e582710dd8ad0465eed29806fbcdbde41
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
- Add Torch to TOSA legalization for aten.index_select
- Fix createOneDimTfIndices function in TosaLegalizeCommon.cpp to
correctly convert Torch indices to TF-style indices, which is used in
convertGatherNdOp
- Update e2e tests in xfail_sets.py
- Update basic.mlir with new LIT test for aten.index_select
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Change-Id: I52519246183949353a3cf22f0a685fe3df8ec8ff
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Addresses ~200 onnx model compile failures in
<https://github.com/nod-ai/SHARK-TestSuite> related to
<https://github.com/iree-org/iree/issues/18631>.
This change simplifies the result of the generated broadcast op
substantially, but reduces the case coverage slightly.
The case which will become unsupported:
- trying to actually broadcast a dynamic dim that is secretly 1.
When does this case appear in practical scenarios?
- for a model where onnx shape inference cannot figure out that a dim
should be 1.
Why do I think we should not support this case for now?
1. For all models with dynamic dim expand ops, the previous path
uniformly generates uglier linalg IR (making it harder for IREE to fuse
properly with other ops).
2. For models failing shape inference castastrophically enough to fail
to see a dim is statically 1, we can try to apply constant folding in
the onnx model before importing.
Leaving this as a draft PR, since it may be more appropriate to fix the
compilation failure in IREE rather than torch-mlir.
### Example of broadcast required in previous path:
```mlir
%300 = linalg.generic {indexing_maps = [#map11], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} outs(%299 : tensor<?x12x?x?xi1>) {
^bb0(%out: i1):
%306 = linalg.index 0 : index
%307 = linalg.index 3 : index
%308 = arith.index_cast %285 : i64 to index
%309 = arith.cmpi eq, %308, %c1 : index
%310 = arith.select %309, %c0, %306 : index
%311 = arith.index_cast %286 : i64 to index
%312 = arith.cmpi eq, %311, %c1 : index
%313 = arith.select %312, %c0, %307 : index
%extracted_79 = tensor.extract %reshape_78[%310, %c0, %c0, %313] : tensor<?x1x1x?xi1>
linalg.yield %extracted_79 : i1
} -> tensor<?x12x?x?xi1>
```
### Example of broadcast with simplified shape list:
```mlir
%409 = linalg.generic {indexing_maps = [#map15, #map11], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%reshape_135 : tensor<?x1x1x?xi1>) outs(%408 : tensor<?x12x?x?xi1>) {
^bb0(%in: i1, %out: i1):
linalg.yield %in : i1
} -> tensor<?x12x?x?xi1>
```
- Add lowering from Torch to TOSA for aten.diagonal
- Clean up some code
- Update xfail_sets.py with the new e2e results
- Update basic_mlir with the new op mlir test
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Change-Id: I99bed685455752d09ed96edd837c4dfbee152701
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
- When the signal tensor is real, onnx allows its shape to be
`[batch][length]` as well as `[batch][length][1]`.
- Onnx also allows to specify `frame_length` together with `window` (not
empty), given that it matches the window size.
- Adding checks on signal and result shapes.
Current version does not work for a mixture of dynamic and static shaped
batch dimensions. Rework to grab the correct dynamic shapes.
---------
Co-authored-by: dan <danimal197@gmail.com>
Previously, if the value was absent, this conversion was creating a
dense resource of value 0 with shape equal to the result shape, then
later re-extracting a splat value. This only works if the shape is
statically known, and even when the shape is known, this is completely
unnecessary since the value's shape should be `[1]` and not the result
shape.
This patch simply sets the `splatvalue` to a `torch.constant.float 0.0`
when the onnx op's `value` attr is absent, and adds `nullptr` checks to
the subsequent conditionals to avoid them in the case where an `attr` is
not given.
Addresses <https://github.com/nod-ai/SHARK-Turbine/issues/831>.
- Add Torch to TOSA legalization for the following reduction ops:
+ aten.min.dim
+ aten.min
+ aten.max
+ aten.prod
+ aten.prod.dim_int
+ aten.all.dim
- Add dtype casting support for reduce sum and prod ops
- Extend aten.max.dim legalization to a template to support aten.min.dim
legalization
- Update end-to-end tests sets in xfail_sets.py
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Change-Id: I854dd6c0c55e570c1fb7242f20c85cf64d6e7fe0
Signed-off-by: Justin Ngo <justin.ngo@arm.com>
Follow up cleanup for [this
PR](https://github.com/llvm/torch-mlir/pull/3689), which introduced a
decomposition for `aten.fmod.Tensor`. This means that the lowering for
this operator in linalg is no longer needed.
Thanks to @vivekkhandelwal1 for pointing this out.
---------
Co-authored-by: Srinath Avadhanula <srinath.avadhanula@getcruise.com>
Enabled mask and is_causal parameters for torch.aten.scaled_dot_product
attention + relevant comments + tests.
The tests added highlight the new capabilities introduced in this PR,
including:
Attention with F16 mask
Attention with Boolean mask
Causal attention with same Q K V shapes
Causal attention without Q K V shapes
Made sure that one cannot input both mask and is_causal.
The lowering pattern for `aten.T` uses transposition implemented via
`linalg.generic`. For downstream passes it is advantageous to use named
ops wherever possible, so this patch changes the lowering to use
`linalg.transpose` instead.
This PR add `floordiv` to the `PY_BUILTIN_TO_TORCH_OP`. For
`aten.mul.int` and `aten.floordiv.int` ops, we add new Canonicalization
Patterns as follow:
```
%1 = torch.aten.mul.int %input, %const-5
%2 = torch.aten.mul.int %1, %const-6
```
Will be replaced by
`torch.aten.mul.int %input, %const-30`
And
```
%1 = torch.aten.mul.int %input, %const-5
%2 = torch.aten.floordiv.int %1, %const-5
```
Will directly return `%input`
This PR also relaxes the `float` type constraint in TorchToTosa for the
`AtenRsubScalarOp` conversion.
To test:
`cmake --build build --target check-torch-mlir-all`
Supports the result with dynamic shape and scalar indices like
```
func.func @test_gather_scalar(%arg0: !torch.vtensor<[3,4,5],f32>, %arg1: !torch.vtensor<[], si64>) -> !torch.vtensor<[?,?],f32> attributes {torch.onnx_meta.opset_version = 13 : si64} {
%0 = torch.operator "onnx.Gather"(%arg0, %arg1) {torch.onnx.axis = 0 : si64} : (!torch.vtensor<[3,4,5],f32>, !torch.vtensor<[], si64>) -> !torch.vtensor<[?,?],f32>
return %0 : !torch.vtensor<[?,?],f32>
}
```
`Torch::AtenSqueezeOp` is referring to the result shape, so it will
failed on lowering if the result shape is dynamic.
The current implementation uses a `linalg.generic` to broadcast the bias
tensor for the lowering of convolutions. This is suboptimal for later
pattern matching. This patch changes it to use the respective named op,
`linalg.broadcast`, instead.
The `axis` attribute is optionally available. Added support by computing
the pad based on the axis values.
---------
Signed-off-by: Rob Suderman <rob.suderman@gmail.com>
- This PR adds new (and equivalent) more tensorized impl of
MelWeightMatrix which lowers all the way to linalg.
- [Ref Pytorch
Impl](https://gist.github.com/PhaneeshB/4e6dfcded3007b1b686fbe28f07a67cd)
- Thanks to @rsuderman for pointing out the difficulties [earlier
impl](#3503) posed during lowering to linalg and also for providing a
better numpy impl 🙏
Felix Schneider
David Tanner
Justin Ngo
Hanumanth04
yyp0
Vivek Khandelwal
Stephen Baione
Phaneesh Barwaria
jinchen
Chi_Liu
Rob Suderman
zjgarvey
Kyle Wang
Samu Tamminen
Yuanqiang Liu
Xida Ren (Cedar)
giacs-epic
Srinath Avadhanula
rohan-tan-bhowmik
Ze Zhang
lingzhiz1998
Ian Wood