Gather crash

[attila-dusnoki-htec]

The following models are crashing on the gather operator:

• https://huggingface.co/FFusion/FFusionXL-BASE/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/aegon-h/sdxl/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/Sdxl-Turbo-Olive-Onnx/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/juggernaut-xl-v5-Olive-Onnx/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/stable-diffusion-xl-base-1.0-Olive-Onnx/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/musiclang/musiclang-4k/blob/main/onnx/decoder_model_merged_quantized.onnx The crash comes from this line: https://github.com/ROCm/AMDMIGraphX/blob/b00489b383b931dcbd395ab316d5a6423e5b97ed/src/include/migraphx/op/gather.hpp#L133

Related MIGraphX issue: https://github.com/ROCm/AMDMIGraphX/issues/2838

[gyulaz-htec]

These models requires the --input-dim and --fill1 to be set:

• https://huggingface.co/FFusion/FFusionXL-BASE/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/aegon-h/sdxl/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/Sdxl-Turbo-Olive-Onnx/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/juggernaut-xl-v5-Olive-Onnx/blob/main/text_encoder_2/model.onnx
• https://huggingface.co/softwareweaver/stable-diffusion-xl-base-1.0-Olive-Onnx/blob/main/text_encoder_2/model.onnx with the following: --input-dim @input_ids 1 77 --fill1 input_ids