opendilab/DI-treetensor

PyTorch OP List(P0)

PaParaZz1 opened this issue · 3 comments

reference: https://pytorch.org/docs/1.8.0/torch.html

common

  • numel
  • cpu
  • cuda
  • to

Creation Ops

  • torch.zeros_like
  • torch.randn_like
  • torch.randint_like
  • torch.ones_like
  • torch.full_like
  • torch.empty_like
  • torch.zeros
  • torch.randn
  • torch.randint
  • torch.ones
  • torch.full
  • torch.empty

Indexing, Slicing, Joining, Mutating Ops

  • cat
  • chunk
  • gather
  • index_select
  • masked_select
  • reshape
  • scatter
  • split
  • squeeze
  • stack
  • tile
  • unbind
  • unsqueeze
  • where

Math Ops

Pointwise Ops
  • add
  • sub
  • mul
  • div
  • pow
  • neg
  • abs
  • sign
  • floor
  • ceil
  • round
  • sigmoid
  • clamp
  • exp
  • exp2
  • sqrt
  • log
  • log10
  • log2
Reduction Ops
  • argmax
  • argmin
  • all
  • any
  • max
  • min
  • dist
  • logsumexp
  • mean
  • median
  • norm
  • prod
  • std
  • sum
  • unique
Comparison Ops
  • argsort
  • eq
  • ge
  • gt
  • isfinite
  • isinf
  • isnan
  • le
  • lt
  • ne
  • sort
  • topk
Other Ops
  • cdist
  • clone
  • flip

BLAS and LAPACK Ops

  • addbmm
  • addmm
  • bmm
  • dot
  • matmul
  • mm

Add the todo checkboxes into the issue.

In-place OP List:

In-place random sampling
  • normal_
  • log_normal_
  • random_
  • uniform_
Calculation
  • abs_
  • add_
  • addbmm_
  • addmm_
  • ceil_
  • clamp_
  • cos_
  • div_
  • exp_
  • floor_
  • log_
  • log2_
  • mul_
  • neg_
  • pow_
  • round_
  • sigmoid_
  • sign_
  • sin_
  • sqrt_
  • square_
  • squeeze_
  • sub_
  • tan_
  • tanh_
  • unsqueeze_
Other
  • requires_grad_

All the ops has been supported by dynamic features.