We do not support average pool when `countIncludePad is set to false.
However if the input is unpadded then the setting of the boolean is
unneeded. Extended use by checking if padding is zero before rejecting
the lowering.
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.
Per the RFC and numerous conversations on Discord, this rebuilds the
torch-mlir CI and discontinues the infra and coupling to the binary
releases
(https://discourse.llvm.org/t/rfc-discontinuing-pytorch-1-binary-releases/76371).
I iterated on this to get latency back to about what it was with the old
(much larger and non-ephemeral) runners: About 4m - 4.5m for an
incremental change.
Behind the scenes changes:
* Uses a new runner pool operated by AMD. It is currently set to manual
scaling and has two runners (32-core, 64GiB RAM) while we get some
traction. We can either fiddle with some auto-scaling or use a schedule
to give it an increase during certain high traffic hours.
* Builds are now completely isolated and cannot have run-to-run
interference like we were getting before (i.e. lock file/permissions
stuff).
* The GHA runner is installed directly into a manylinux 2.28 container
with upgraded dev tools. This eliminates the need to do sub-invocations
of docker on Linux in order to run on the same OS that is used to build
wheels.
* While not using it now, this setup was cloned from another project
that posts the built artifacts to the job and fans out testing. Might be
useful here later.
* Uses a special git cache that lets us have ephemeral runners and still
check out the repo and deps (incl. llvm) in ~13s.
* Running in an Azure VM Scale Set.
In-repo changes:
* Disables (but does not yet delete):
* Old buildAndTest.yml jobs
* releaseSnapshotPackage.yml
* Adds a new `ci.yml` pipeline and scripts the steps in `build_tools/ci`
(by decomposing the existing `build_linux_packages.sh` for in-tree
builds and modularizing it a bit better).
* Test framework changes:
* Adds a `TORCH_MLIR_TEST_CONCURRENCY` env var that can be used to bound
the multiprocess concurrency. Ended up not using this in the final
version but is useful to have as a knob.
* Changes the default concurrency to `nproc * 0.8 + 1` vs `nproc * 1.1`.
We're running on systems with significantly less virtual memory and I
did a bit of fiddling to find a good tradeoff.
* Changed multiprocess mode to spawn instead of fork. Otherwise, I was
getting instability (as discussed on discord).
* Added MLIR configuration to disable multithreaded contexts globally
for the project. Constantly spawning `nproc * nproc` threads (more than
that actually) was OOM'ing.
* Added a test timeout of 5 minutes. If a multiprocess worker crashes,
the framework can get wedged indefinitely (and then will just be reaped
after multiple hours). We should fix this, but this at least keeps the
CI pool from wedging with stuck jobs.
Functional changes needing followup:
* No matter what I did, I couldn't get the LTC tests to work, and I'm
not 100% sure they were being run in the old setup as the scripts were a
bit twisty. I disabled them and left a comment.
* Dropped out-of-tree build variants. These were not providing much
signal and increase CI needs by 50%.
* Dropped MacOS and Windows builds. Now that we are "just a library" and
not building releases, there is less pressure to test these commit by
commit. Further, since we bump torch-mlir to known good commits on these
platforms, it has been a long time since either of these jobs have
provided much signal (and they take ~an hour+ to run). We can add them
back later post-submit if ever needed.
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.
The logic here is very similar to the conversion for AdaptiveAvgPool1d
#2661 with a few modifications:
1. buffVal = -inf instead of 0
2. the main linalg generic op accumulates a max, instead of a sum, to
the first output tensor
3. avg pooling requires dividing the sum pool by the kernel width, which
we stored as an auxilliary tensor (kSizeTensor). Here, the auxiliary
tensor will be recording the indices. Strangely enough, the only
signature available for this function is to return indices, and it
appears that they must be computed whether the user desires them or not.
See
[pytorch/torch/nn/functional.py](https://github.com/pytorch/pytorch/blob/main/torch/nn/functional.py#L1174).
Before writing other adaptive pooling conversions, the logic of this
decomposition should be rolled into a helper function that will work for
both max and avg pooling ops. Even the auxiliary tensor should likely be
automated. This code was written in a slightly more tedious way than
strictly necessary (often using loops to fill SmallVectors up to rank-2,
which is only two in this case), in order to more easily facilitate the
transition to a helper function.
convolution with [time,batch,channel] ordering, as opposed to the
default [batch, channel, time]. Currently implementing by transposing
the input and output, but may need to get its own implementation in the
future because this is supposed to be an op that gives a speedup. This
is used by fairseq
(https://github.com/facebookresearch/fairseq/issues/172).
(in case you were wondering like me, this is different from transposed
convolution. Transposed convolution has fractional strides).
---------
Co-authored-by: Xida Ren <xida.ren.dev@gmail.com>
Co-authored-by: Frederik Harwath <frederik.harwath@amd.com>
Handle both `torch.dequantize` and `torch.quantize_per_tensor` including
the op based quantization parameter tracking. This includes adding
`qint32` to torch types as it was missing during the initial type
inclusion.
For testing we only have `torch.int8` and `torch.float` types on
function boundaries as the `qint8` types require passing the scale
and zero point quantization information which is not supported yet.
Adaptive pooling ops can only be decomposed into their non-adaptive
counterparts in trivial cases.
For example, the current decomposition for AtenAdaptiveAvgPool1dOp in
DecomposeComplexOps.cpp supports outSize = inSize (i.e., do literally
nothing), and outSize = 1 (i.e., do a batched average).
The reason adaptive pooling ops are difficult to lower to linalg is that
they are not constantly strided. They are computed by taking an input
tensor of shape (N, C, Hin), and an output size Hout, and computing the
output tensor at position (n,c, h) in the following way:
1. compute st(h) = (h*Hin)//Hout
2. compute en(h) = 1 + ((h+1)*Hin -1)//Hout
3. apply a computation (max or avg) to the slice: INPUT[n, c,
st(h):en(h)]
The provided sample implementation (for ConvertAtenAdaptiveAvgPool1dOp)
uses tensor.extract to access the input tensor inside the payload of a
linalg generic op. This is likely an unattractive use of linalg generic
ops, which is why I am asking for some more targeted feedback on the
validity of this approach before attempting to support the many other
adaptive pooling ops.
Specifically:
- Is the performance of this implementation bad enough to warrant
targeting different dialects entirely? e.g. TMtensor/linalg ext/ etc.
- If the provided implementation is of acceptable performance to the
community, then is it permissable to remove the Adaptive pooling
decompositions from DecomposeComplexOps.cpp? Based on the current
structure of the -torch-decompose-complex-ops pass, it does not seem
possible to only decompose the adaptive ops in special cases (it seems
to get stuck in an infinite loop on a match failure). I would be happy
to instead incorporate the case logic into the conversion directly, and
remove the decompositions once they are rendered completely obsolete.
As long as this approach is acceptable, I can clean up the
implementation with some helper functions, and quickly add support for
each of the remaining Adaptive pooling ops.
Adds a lowering to Linalg for reflection_pad1d. Based on ideas/code from draft PR
https://github.com/llvm/torch-mlir/pull/2693.
---------
Co-authored-by: Kumar Deepak <kumar@xilinx.com>
This replaces the lowering of aten.cat with tensor.concat, allowing more
efficient handling of concatenations in downstream flows. The refbackend
populates concat decomposition patterns that can be used to recover the
previous lowering.
`AtenStackOp` defines this folder for list operand containing single
element:
```
OpFoldResult AtenStackOp::fold(FoldAdaptor adaptor) {
auto list = getOperand(0).getDefiningOp<PrimListConstructOp>();
if (!list || !list->hasOneUse() || list.getElements().size() != 1)
return nullptr;
return list.getElements()[0];
}
```
However, unlike `AtenCatOp`, `AtenStackOp` cannot be folded away for
single element list operand because the result from a stack operation
contains an additional dimension (of size 1, like expand_shape).
This PR removes the `AtenStackOp::fold` method, and adds an e2e test for
single element list input case, which fails on current `main` as
follows:
```
Unexpected outcome summary: (linalg)
****** Failed tests - 1 tests
FAIL - "TensorsStackSingleElementListModule_basic"
@ trace item #0 - call to "forward"
@ output of call to "forward"
ERROR: shape (torch.Size([10, 32])) is not equal to golden shape (torch.Size([10, 1, 32]))
```
Thanks Chris Lalau Keraly for the bug report.
This commit adds the OnnxToTorch support for BitwiseXor, BitwiseOr, Div, Equal, Cast,
Ceil, Floor, Cos, and Clip op.
This commit also adds the TorchToLinalg support for aten.clamp.Tensor and aten.clamp_min.Tensor op.
Signed-Off By: vivekkhandelwal1424@gmail.com
Despite aten.mm requiring the input and output types match, we still opt
to maintain signedness semantics in case later passes try to do any sort
of integer type narrowing.
Adds a lowering for the torch.aten.argmin operator to linalg via decomposition into torch.aten.min.dim.
---------
Co-authored-by: Franz Haniel <franz.haniel@amd.com>
The function `getTypeForScalarType` currently takes an argument to
specify the signedness of integer types. This is leakage of backend
specific requirements into the torch dialect world. Because
`getTypeForScalarType` is a utility function for the torch dialect, it
should only produce types that match the sign conventions used by
PyTorch (regular integers are signed and unsigned integers are
unsigned).
This commit removes the signedness argument from
`getTypeForScalarType`, and moves the backend specific handling of
integer types to the backend code.
This commit adds the OnnxToTorch support for Atan, Bitshift, BitwiseAnd,
and BitwiseNot op.
This commit also adds the TorchToLinalg support for AtenBitwiseLeftShiftTensorOp.
Signed-Off By: vivekkhandelwal@nod-labs.com
The aten.reshape ops in the decomposition are replaced with prims.collapse
and prims.split_dim ops, which means that the cases where the lowering of
reshape from torch to linalg which are not supported, are avoided.
Essentially, by using the collapse and split_dim ops instead of the
reshape ops, we are not "losing" the information that the reshapes do not
arbitrarily mix dimensions. Which makes lowering easy.
3 additional tests added:
- fully dynamic,
- dynamic only the spatial dimensions,
- dynamic only in the non-spatial dimensions.
Adds support for lowering to prims split_op.
Similar design to collapse op lowering in
https://github.com/llvm/torch-mlir/pull/2572, with some
small differences, because the split_dim op (in pytorch) is
view-changing whereas the collapse is not. The difference
means that
1) it must be registered in the function Torch::isViewLikeOp
2) it must be be added to the "expected fail" set for the torch dynamo backend.
… AtenBernoulli_FloatOp
It fixing case like: `%2110 = torch.aten.arange.start_out %int1,
%int1517, %int1, %2109 : !torch.int, !torch.int, !torch.int,
!torch.tensor -> !torch.tensor`.
`aten.arange.start_out` doesn't have value semantics also, means`%2110`
is an alias for %2109.
So I decompose it to `aten.arange.start` + `torch.contents.overwrite`.
The complex decomposition logic is target to handle cases like view and
dtype cast which I add in e2e tests.
Steps taken:
1) add generator code to torch_ods_gen.py, run update_torch_ods.sh
2) add (custom) shape and type inference generator code to
abstract_interp_lib_gen.py, run update_abstract_interp_lib.sh
3) Implement lowering to tensor.collapse_dims. Requires the `start` and
`end` values to be constant, else lowering fails
4) Update xfail_sets.py (append to LTC_XFAIL_SET) after running
/tools/e2e_test.sh --filter Collapse --verbose -c XX for all support
backends (XX).
Motivation:
- Supporting the collapse operation will be useful for lowering of
pixel_shuffle (see Issue #2559)
For static tests (that is when the shape is know) for example:
```
@annotate_args([None, ([3, 18, 2, 2], torch.float32, True)])
```
The e2e passes. But only if the replacement op's return type is set as
undefined (optional shape and type must be explicitly made unset),
otherwise there's a error about the function return type.
For dynamic cases, for example if the above is replaced with
```
@annotate_args([None, ([-1, -1, -1, -1], torch.float32, True)])
```
There is a failure to lower to linalg from torch ("view op explicitly
labelled as illegal"). This seems to be because the support for lowering
from torch to linalg with dynamic shapes is limited.
Rob Suderman
Ilija Kalinić
Yuanqiang Liu
Stella Laurenzo
zjgarvey
Xida Ren (Cedar)
Franz Haniel
John Wu
Sungsoon Cho
lonely eagle
lisaliu1
Frederik Harwath
Aart Bik
kumardeepakamd
Quinn Dawkins
JianzheXiao
Sambhav Jain
Vivek Khandelwal
Frederik Harwath
Ramiro Leal-Cavazos
Mi Jiazhi
James Newling
Jiawei Wu