Op Support: aten._scaled_dot_product_flash_attention.default

[shivadbhavsar]

High level op: https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html

• [ ] Add acc op for torch.nn.functional.scaled_dot_product_attention
• [ ] Implement acc converter for acc op
• [ ] Implement aten converter for aten._scaled_dot_product_flash_attention.default that maps to the acc converter
• [ ] Test cases for both acc converter (under fx) and aten converter (under dynamo)

[shivadbhavsar]

Used in TIMM models:

• cait_m36_384
• crossvit_9_240
• deit_base_distilled_patch16_224
• mobilevit_s
• pit_b_224

Note: requires rocm/pytorch:rocm6.2_ubuntu22.04_py3.10_pytorch_release_2.3.0 for aten export to include this op