perf.txt

https://github.com/pytorch/pytorch/pull/3526 Reuse intermediate results over multiple backwards grad_inputs
https://github.com/pytorch/pytorch/pull/3509 Optimizer: optimize transposes in variety of circumstances
https://github.com/pytorch/pytorch/pull/3409 Add Tensor Core ops to RNNs for Volta
https://github.com/pytorch/pytorch/pull/3370 Follow up #3211 (sparse broadcast_coalesced, reduce_add_coalesced)
https://github.com/pytorch/pytorch/pull/3336 Prevent numerical issues with poisson_nll_loss when log_input=False
https://github.com/pytorch/pytorch/pull/2764 [Done]parallelize elementwise operation with openmp
https://github.com/pytorch/pytorch/pull/6110 Fix bilinear performance regression
https://github.com/pytorch/pytorch/pull/6078 Exp, log, sin, cos vectorized
https://github.com/pytorch/pytorch/pull/6062 Enable MKLDNN convolution forward and backward
https://github.com/pytorch/pytorch/pull/6026 Speed up sum over a dimension
https://github.com/pytorch/pytorch/pull/5913 Optimize unique sorting by using std::vector+sort instead of std::set
https://github.com/pytorch/pytorch/pull/5782 fused GLU backward
https://github.com/pytorch/pytorch/pull/5747 Save self.numel() for backward computation instead of self
https://github.com/pytorch/pytorch/pull/5722 Add optimization to norm for common norms
https://github.com/pytorch/pytorch/pull/5710 improve occupancy for cuda rngs
https://github.com/pytorch/pytorch/pull/5680  implement TripletMarginLoss as a native function
https://github.com/pytorch/pytorch/pull/5646 implement CosineEmbeddingLoss as a native function and add reduce arg
https://github.com/pytorch/pytorch/pull/5640 Revert "implement CosineEmbeddingLoss as a native function and add reduce arg"
https://github.com/pytorch/pytorch/pull/5447 implement CosineEmbeddingLoss as a native function and add reduce arg
https://github.com/pytorch/pytorch/pull/5433 speed up CPU EmbeddingBag (indexSelectAdd op)
https://github.com/pytorch/pytorch/pull/5346 Implement MarginRankingLoss as native function and add reduce=True arg to it
https://github.com/pytorch/pytorch/pull/5279 Speed-up nn.Linear for the 3d input case
https://github.com/pytorch/pytorch/pull/5080 Implement hinge_embedding_loss as a native function.
https://github.com/pytorch/pytorch/pull/5064 DDP: 10% of NCCL backend perf improvements with mixed-prec support
https://github.com/pytorch/pytorch/pull/5054 Use fast integer division algorithm to avoid division ops inside kernels.
https://github.com/pytorch/pytorch/pull/5010 add AVX2 implementation for sigmoid function
https://github.com/pytorch/pytorch/pull/4924 add reduce=True argument to MultiLabelMarginLoss
https://github.com/pytorch/pytorch/pull/4870 Slightly improve DistributedDataParallel (single-GPU binding) multi-process distributed training performance
https://github.com/pytorch/pytorch/pull/4824 parallelize vol2col and col2vol of Conv3D with CPU backend
https://github.com/pytorch/pytorch/pull/4803 More efficient squeeze() backward in edge case
https://github.com/pytorch/pytorch/pull/4705 adds reduce argument to BCEWithLogitsLoss interface
https://github.com/pytorch/pytorch/pull/4312 Vectorize normal_
https://github.com/pytorch/pytorch/pull/4231 Add reduce arg to BCELoss
https://github.com/pytorch/pytorch/pull/4183 Allowing usage of GPU Direct within PyTorch for the Broadcast operation
https://github.com/pytorch/pytorch/pull/4174 Rearrange dimensions for pointwise operations for better performance.
https://github.com/pytorch/pytorch/pull/4094 Implement pin_memory() as a NativeFunction