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.
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.
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.
This PR updates the torch-to-tosa conversion with following changes:
- Support torch.none as min/max input argument for tosa.clamp op
- Support negative value as start index for tosa.slice op
- Add tosa.logical_or lowering support
e2e test:
python -m e2e_testing.main --config=tosa
LIT tests:
cmake --build build --target tools/torch-mlir/all
---------
Co-authored-by: Ze Zhang <ze.zhang@getcruise.com>
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.
For easier tracking of issues, sort the TOSA passing list. It is still
significantly smaller then the XFAIL list would be.
Resolves#2620, at least until the xfail list gets smaller than the
passing list.
Signed-off-by: Eric Kunze <eric.kunze@arm.com>
The linalg Op `linalg.conv_2d_ngchw_fgchw` had a bug where
1. Weights were accessed as G,F,C,H,W instead of as F,G,C,H,W
2. Output was accessed as N,F,G,H,W instead of as N,G,F,H,W
Now this has been fixed in
https://github.com/llvm/llvm-project/pull/73855 which broke the
torch-mlir lowering to that Op.
This patch switches lowering in torch-mlir to the newly introduced
`linalg.conv_2d_ngchw_gfchw` op which accesses weights in an order that
is compatible with PyTorch's memory layout.
Fix https://github.com/llvm/torch-mlir/issues/2622
llvm-project: bbd2b08b95fe76bea138c1b03c1cd42ed3ee04df
stablehlo: ab709fe48de88c67717abfbd7ef17425eb95ddaf
These commits were chosen in order to account for an MLIR API break from
3dbac2c007
which required a patch to stablehlo. We integrate a bit beyond that
commit to deal with some revert/reapply cycles in the intervening range
which were discovered in another downstream.
Further, it requires adaptation to the stablehlo API breaks introduced
from https://github.com/openxla/stablehlo/pull/1872 which are along for
the ride.
Since some stablehlo builders were changed to directly take int64_t
array refs, also traced that up some call stacks to eliminate some
signed/unsigned mismatches that result.
Also adds a few TOSA tests to the passing set that seem to work now.
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.
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.
Rob Suderman
Xida Ren (Cedar)
John Wu
Ze Zhang
Sungsoon Cho
lonely eagle
Ilija Kalinić
zjgarvey
Vivek Khandelwal
JianzheXiao
Eric Kunze
Felix Schneider
Stella Laurenzo
Frederik Harwath
James Newling
Yuanqiang Liu
Jiawei Wu