pytorch

a07a0190 - enable deterministic path for index_put with accumulate=False on CPU and CUDA (#57839)

enable deterministic path for index_put with accumulate=False on CPU and CUDA (#57839) Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/57839 we reuse the `index_put_accum_kernel`, rename it to `index_put_deterministic_kernel` and add a bool `accumulate` in `index_backward_kernel` Test Plan: buck test mode/opt //caffe2/test:torch -- test_index_put_non_accumulate_deterministic ✓ Pass: caffe2/test:torch - test_index_put_non_accumulate_deterministic_cpu (test_torch.TestTorchDeviceTypeCPU) (5.120) Summary Pass: 1 Skip: 1 ↻ caffe2/test:torch - test_index_put_non_accumulate_deterministic_meta (test_torch.TestTorchDeviceTypeMETA) ListingSuccess: 1 buck test mode/opt //caffe2/test:torch_cuda -- test_index_put_non_accumulate_deterministic ✓ ListingSuccess: caffe2/test:torch_cuda - main (6.397) ✓ Pass: caffe2/test:torch_cuda - test_index_put_non_accumulate_deterministic_cuda (test_torch.TestTorchDeviceTypeCUDA) (26.030) ✓ Pass: caffe2/test:torch_cuda - main (26.030) Summary Pass: 2 ListingSuccess: 1 Reviewed By: ngimel Differential Revision: D28290699 fbshipit-source-id: df8bbe7af2e72017566161b05b85737fda4ceb3f