Skip to content

Conversation

@mgehre-amd
Copy link
Collaborator

DO NOT merge, just to look at our diff

@mgehre-amd mgehre-amd changed the title Feature/onnx_to_tosa feature/onnx-to-tosa Jun 26, 2023
chaitanyakamarapu and others added 29 commits August 21, 2025 09:41
…ixing_pad_value

[backport ] convtranspose fixing pad value
… remove code duplication

Signed-off-by: Rickert, Jonas <[email protected]>
…for.qdq

Enable Pow to Mul canonicalization for quantized exponents
Re-remove attempt to create rewriter from not existing declarative rewriter
…with_single_conv

Improving convtranspose decomposition for stride 2x2
gerion-amd and others added 30 commits December 9, 2025 01:39
Quant types as operands and results
…lude-scale-layernorm-recompose

[AIESW-18974] Do not include the scale in LayerNorm recomposing but fuse it later
Add QuantDialect as dependent for QuantTypes pass
Signed-off-by: Jonas Rickert <[email protected]>
Do not try to decompose onnx.SoftmaxCrossEntropyLoss if its unranked
#503)

* feat: add new SameOperandsAndResultShape to some onnx ops

* test: update lit test
…ted-data

FrontendDialectHelper: fail gracefully when parsing truncated data files
- Add XFE custom ops dialect and operations
- Implement XFE.cpp with shape inference helpers
- Add YAML-based schema definition for XFE ops
- Update gen_onnx_mlir.py to support custom ops generation
- Add utilities: gen_onnx_yaml.py, test_custom_ops.py, verify_yaml_schema.py
- Add comprehensive documentation: QUICK_START.md, CUSTOM_OPS_README.md, CHANGES_SUMMARY.md
- Update CMakeLists.txt and ONNX.td to integrate XFE ops

Signed-off-by: Sameek Banerjee <[email protected]>

Update Documentation and fix build issues.

Separate Shape Inference Implementation

Add test case for shape inference

Add pass to convert to NHWC version of conv

Fix Linter

Fix black format error

Address the LIT failures. Remove Un-necessary Documentation

Regenerate for quant types + refactors

Rename to Channel Last

Add Logic for shape inference for ND cases

Rename Pass to convert and add all ops

Clang fix

Add extra logic to genereate fold and verify methods

Black reformat

Clang Fix

Add XFE Prefix + other changes

Remove Channel Last as suffix

Minor changes
A length value of -1 is a valid value and makes the MemoryBuffer find
out the file length by itself.

The offset can still be to big so read it specifically, too.
FronendDialectHelper: fix data length checking
[AIESW-14742] merge pad op into avgpool pad attribute.
* feat: back to back maxpool optimization

* test: add lit tests

* fix: stop ksize from getting too large

* feat: extend to quantized cases

* fix: check if upper maxpool covers input fully

* fix: formatting

* fix: check for static shapes

* fix: formatting

* fix: check for null defining op
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.