GitLab

Enable simplify qdq to work with FP8 types, add warning for fp8 when loading mxr

Updated onnx.proto files to be same as onnx==1.14.1 that is used by MIGraphX.   Added more switch cases in onnx parser to handle FP8 dtypes.

Ref Unique Operator only

Add min and max reduction modes
Implement duplicate index handling
Add corresponding reference, onnx, and verify tests.

Simplify what we parse the ONNX operator ConstantOfShape into a literal for a static shape

Introduce dilations attribute to pooling operators reference implementation.

This PR introduces the support of Windows resource files to Embed.cmake. It is ON by default on Windows, and when it is OFF *.cpp files will be used. The same applies to Linux - ON -> *.o (LD) or OFF -> *.cpp .

This PR fixes building resources on Linux with ld and objcopy commands.

Converts dynamic reshapes with constant output dimensions to a static reshape instruction
Intended to be used after split_single_dyn_dim pass and after find_static_dimensions_of makes shape calculations doable at compile-time.

Handles all 4 Fp8 dtypes listed here : https://onnx.ai/onnx/technical/float8.html
Follows saturation/clipping logic from table there as well : https://onnx.ai/onnx/technical/float8.html#cast
Only adding fp8e4m3fnuz in MIGraphX IR for now.

Reworked the simplify_qdq pass to support:

Per-axis quantization (ie. allow 1D scales and zero points)
Allow broadcast and transpose ops between dq and quant_op

Since ONNX opset version 14, layout attribute has been introduced to LSTM operator which allows two predefined layouts for input and output shapes.  Add corresponding reference, onnx, and verify tests.