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.
* implemented ceil_mode== true support for lowering aten.max_pool2d to tosa
* add e2e test for lowering aten.max_pool2d to tosa with ceil_mode=true
---------
Co-authored-by: Lisa Liu <lingl@xilinx.com>
* 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.
This commit adds a check that `defaultDtype` exists in the RefineTypes
handling of `AtenSumOp` before accessing the method `isInteger`, which
crashes the program is `defaultDtype` is null.
The handling of `defaultDtype` is the same as the one used for the
`AtenSumDimIntListOp`.
* implemented lowering torch.aten.constant_pad_nd to tosa
* add constant_pad_nd e2e tests to TOSA_PASS_SET
* add PadModule_basic & PadWithNoneValModule_basic to TOSA_PASS_SET
---------
Co-authored-by: Lisa Liu <lingl@xilinx.com>
Fixes a broken bazel build from https://github.com/llvm/torch-mlir/pull/1887 causing the following build error:
```
ERROR: /root/.cache/bazel/_bazel_root/b89349c08f7224396763d14fe35cba11/external/torch-mlir/BUILD.bazel:819:10: Compiling tools/torch-mlir-opt/torch-mlir-opt.cpp failed: (Exit 1): clang failed: error executing command /usr/lib/llvm-16/bin/clang -U_FORTIFY_SOURCE -fstack-protector -Wall -Wthread-safety -Wself-assign -Wunused-but-set-parameter -Wno-free-nonheap-object -fcolor-diagnostics -fno-omit-frame-pointer ... (remaining 366 arguments skipped)
Use --sandbox_debug to see verbose messages from the sandbox and retain the sandbox build root for debugging
external/torch-mlir/tools/torch-mlir-opt/torch-mlir-opt.cpp:16:10: fatal error: 'stablehlo/dialect/Register.h' file not found
#include "stablehlo/dialect/Register.h"
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
Target @torch-mlir//:torch-mlir-opt failed to build
```
GHA workflow with the fix:
https://github.com/sjain-stanford/torch-mlir/actions/runs/4421486154
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
* Adding stablehlo dialects support for torch-mlir-opt tool.
* Update torch-mlir-opt.cpp
Fixed the build error according to build configuration for macOS.
The current implementation of `getScalarValue` does not check that the
input to a `ValueTensorLiteralOp` is an i64 before extracting the
value, and it does not check that the result type of the
`PrimNumToTensorScalarOp` is also an i64. This leads to crashes or
invalid IR generated when the `input` is something other than an i64
tensor or `!torch.int`.
This commit addresses those issues. In addition, the function
`getScalarValue` is renamed to `getScalarIntValue` to make it clear
that it *only* extracts scalar integers.
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.
Roll PyTorch Action
Michael Feliz
Ramiro Leal-Cavazos
Zhekun Zhang
lisaliu1
Sean Silva
Yuanqiang Liu
Matthias Gehre
Sambhav Jain
Jiahao Li
gpetters94
Ziheng Jiang
Kan Chen
Eric Kunze
Maksim Levental
Priya Savithiri
Vivek Khandelwal