- 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 🙏
This commit adds the shape info for the tensors created during the
decomposition of GroupNorm op.
Signed-Off By: Vivek Khandelwal <vivekkhandelwal1424@gmail.com>
Set PyTorch and TorchVision version to nightly release 2024-08-18.
This commit also updates the `scaled_dot_product_attention` op.
A new attribute `enable_gqa` has been added. As of now, only the
default value for the same is supported.
Signed-Off By: Vivek Khandelwal <vivekkhandelwal1424@gmail.com>
Discovered in https://github.com/llvm/torch-mlir/issues/3104
Most likely when building with stablehlo, while waiting for it missing
dependency was generated to location shared with another dependency.
This commit extends the OnnxToTorch lowering for BatchNormalization op
for supporting the case when training=True.
Signed-Off By: Vivek Khandelwal <vivekkhandelwal1424@gmail.com>
The `layout` attribute was not considered for the `onnx.RNN` operation.
Added support for the attribute to transpose the inputs / outputs of the
RNN when valid.
The einsum lowering was missing the behavior for duplicate indices in
the equation. This amounts to a diagonalization along duplicate pairs of
indices in the equation.
Closes#3575
The PyTorch remainder operator is meant to compute the Python modulus
operator entrywise:
https://pytorch.org/docs/stable/generated/torch.remainder.html#torch.remainder
In python the modulus operator is meant to always return a result with
the same sign as the divisor:
https://docs.python.org/3/reference/expressions.html#binary-arithmetic-operations
In other words, torch.aten.remainder should return a Python-style
modulus instead of a C-style modulus. However the remainder operator was
simply translated into arith.ModSI or arith.ModF, which both effectively
compute the C-style modulus. Now the lowering has been modified so that
the modulus operator works properly with negative numbers, both in the
dividend, and the divisor.
This patch adds basic support for lowering graphs with per-channel
quantization. Per-channel quantized ops have to be excluded from
`FuseQuantizedOps` for now but can be used in QDQ quantized form.
Using this patch, we're able to import and execute (on the linalg
backend) graphs with per-channel quantization applied using the "new"
PyTorch 2.0 Export Quantization.
The saga of aligning onnx and torch padding conventions continues.
```python
onnx_pads = [low_x, low_y, low_z, high_x, high_y, high_z]
torch_pads = [low_z, high_z, low_y, high_y, low_x, high_x]
```
So not only is the lexicographical ordering hierarchy swapped (low/high
x spatial-dim -> spatial-dim x low/high) but the ordering in the the
spatial-dim specification is also reversed.
This patch properly reverses the pad ordering (and actually uses the
`shuffledPadding` to pad).
`onnx.Shape` can select only a subset of indices using attributes. Add
support for these attributes.
---------
Co-authored-by: zjgarvey <47986913+zjgarvey@users.noreply.github.com>
Following up from the discussion in
<https://github.com/llvm/torch-mlir/pull/3550>, I've edited the lowering
to prevent OOB extracts in a more direct fashion (i.e., just clamping
directly).
I don't think this affects the lit tests at all, but I've tested the
changes in our external test suite at
<https://github.com/nod-ai/SHARK-TestSuite/tree/main/>. I found the
issue when I was unexpectedly getting `nan`'s along the output image
border for a resize test there.
Change linalg.matmul_unsigned to linalg.matmul with unsigned type_fn
Signed-off-by: Max Dawkins <max.dawkins@gmail.com>
Co-authored-by: Max Dawkins <max.dawkins@gmail.com>
There were two issues related to `ignore_index` being set
(1) the onnx-to-linalg pass as not reading the value correctly (2) the
mean pass was not considering the `ignore_index` value
For (2) when taking the mean we need to know how many of the values were
considered in the sum and therefore we cannot divide by the total number
of elements. Adding a summation across the total number should correct
this issue.
The static uneven divisible AdaptiveAvgPool2d means that although the
input size is not an integer multiple of ouput size, but the kernel and
stride size can also be fixed (not dynamic). The derivation logic of
kernel and stride size is consistent with
torch/_decomp/decomposations.py:adaptive_avg_pool2d as described in the
following:
1. Stride Size
Firstly , derive the start index in each reduce operation according to
the output size (`n`), `start_index = ([0, 1, ..., n - 1] * input_size)
// output_size`. For each index `k`, if `k * (input_size % output_size)
< output_size`, then the current and previous stride keeps the same as
`input_size // output_size`. So suppose `(n-1) * (input_size %
output_size) < output_size`, the stride in the whole AdaptiveAvgPool2d
process keeps static, as `input_size // output_size`.
2. Kernel Size
torch/_decomp/decomposations.py:adaptive_avg_pool2d calculates a static
kernel size when the input/output sizes satisfy either of the two
conditions, `input_size % output_size == 0` or `output_size %
(input_size % output_size) == 0`. Here if `input_size % output_size ==
0`, then the kernel size equals `input_size // output_size`, otherwise
`input_size // output_size + 1.`
Torch has all scalars represented as i64 and f64 types which results in
extraneous trunc-extf commands. We can rework this by elliding
widen-narrow cases away.
Current StableHlo lowering strategy works well when `src` tensor's rank
is no bigger than `dst` tensor's. The new patch make it succeed in other
cases. The following is an example.
```
%190 = torch.prim.ListConstruct %arg4 : (!torch.vtensor<[1,1024],si64>) -> !torch.list<vtensor>
%191 = torch.aten.index_put.hacked_twin %189, %190, %186, %true : !torch.vtensor<[1024,768],f32>, !torch.list<vtensor>, !torch.vtensor<[1,1024,768],f32>, !torch.bool -> !torch.vtensor<[1024,768],f32>
```
- Adds support for lowering depthwise + quantized convolution ops to
linalg::DepthwiseConv2DNhwcHwcQOp
- Changed the variable name for groupSize (which is really C/G) to the
more appropriate numGroups (G).
- Discovered in e2e testing that linalg does not accept (Cin = groups &&
Cout = K*groups for K>1) as a "depthwise" conv, so this also updates the
case-checking to reflect this issue.
Pytorch and ONNX apparently round to nearest, ties go to nearest even,
but we were using `math::round` for the torch-to-linalg conversion of
`quantize_per_tensor`, which rounds away from zero on ties.
This PR adds a conversion in the TorchOnnxToTorch pass for the ONNX
Multinomial operation. It also adds a TorchToLinalg lowering for the
`aten.Multinomial` op and does a light refactor of some repeated code
that generates random floating point numbers in
`TorchToLinalg/Random.cpp`.
This patch adds a few misc pad op related changes:
1. Addresses issue <https://github.com/llvm/torch-mlir/issues/3457>
2. Addresses issue <https://github.com/llvm/torch-mlir/issues/3442>
3. Fixes the padding order for asymmetrically padded onnx.Conv ops
4. Enables passing quantization through those onnx.Conv op pre-paddings
5. Modifies the torch-to-linalg lowering of AtenReplicationPad2d op to
enable support for input rank != 4
Unfortunately, even with all of these changes, the e2e tests for the
ReplicationPad2d still fail the onnx config, since the torch export
procedure for rearranging the pad order is complicated enough that the
padding ints end up not being able to fold back to constants.
The LpNormalization lowering was previously just computing the norm,
which is incorrect. This computes the norm then divides the input tensor
by it's norm.
I've tested this against some simple onnx models locally. I'll look into
adding a test case for this in an external test suite.
Register `aten.fake_quantize_per_channel_affine` and
`aten.fake_quantize_per_tensor_affine.tensor_qparams` ops
---------
Co-authored-by: Ze Zhang <ze.zhang@getcruise.com>
Fix the pad tensor rearrangement such that we change the representation
from [x1_begin, x2_begin, ..., x1_end, x2_end,...] to [xn_begin, xn_end,
...., x2_begin, x2_end, x1_begin, x1_end] where x1, x2 .. xn are the
dimensions of the pads tensor argument.
---------
Co-authored-by: zjgarvey <zjgarvey@gmail.com>
Co-authored-by: zjgarvey <47986913+zjgarvey@users.noreply.github.com>
* use lhs tensor's element type as compute type when rhs is scalar.
* previously `a != 1.0`(a is a fp32 tensor) will lowering to `%6 =
stablehlo.compare EQ, %4, %5, FLOAT : (tensor<2x5xf64>, tensor<2x5xf64>)
-> tensor<2x5xi1>`
* now it will lowering to `%6 = stablehlo.compare EQ, %4, %5, FLOAT :
(tensor<2x5xf32>, tensor<2x5xf32>) -> tensor<2x5xi1>`
Addresses an issue with onnx.Gather lowering to linalg:
<https://github.com/nod-ai/SHARK-Turbine/issues/242>
The builder for tensor.expand_shape, without an explicitly provided
output shape, fails to infer an output shape in the case of multiple
dynamic reassociation dims. I tried adding the output shape explicitly
for tensor.expand_shape, but ran into compilation issues later on (see
<https://github.com/iree-org/iree/issues/17760>).
This PR adds support by lowering this op to tensor.reshape when multiple
dynamic reassociation dims are provided.
Due to the custom operation parser, the print and parser were expecting
two different forms.
One having the dictionary before the value and the other after.
Following the format of the other constants ops, the constant.int will
follow the `value attr-dict` format. Updated the parser accordingly.
This bump triggered an upstream assert. Includes a WAR for #3506.
Also includes several things I needed to do to repro:
* When TORCH_MLIR_TEST_CONCURRENCY=1, test runs will be printed.
* Added TORCH_MLIR_TEST_VERBOSE=1 handling to enable verbose mode
(useful on CI).
---------
Co-authored-by: Stella Laurenzo <stellaraccident@gmail.com>
Phaneesh Barwaria
Vivek Khandelwal
lingzhiz1998
Ian Wood
zjgarvey
Rob Suderman
yyp0
Yevhenii Havrylko
pkapris-syrmia
Branko Trifkovic
Yuanqiang Liu
aldesilv
Felix Schneider
Marius Brehler
Chi_Liu
Rob Suderman
Gaurav Shukla
Jiawei Wu
Suraj Sudhir
Ivan Butygin
Vinayak Dev
pdhirajkumarprasad
Ze Zhang
bosko-syrmia
jinchen
Arham Khan
rohan-tan-bhowmik
Xinyu Yang
Xida Ren (Cedar)
Matthias Gehre
Sagar Kulkarni
Scott Todd
Max191
Christopher McGirr
Aart Bik