peterbell10 comments

Results 134 comments of


                                            peterbell10

RFC-0025 Improving incremental builds

CUDA builds don't require any additional work for the main changes here. `.cu` files can use the per-operator headers as they would any other header and `TensorBase` was used first...

logaddexp2: Use log1p and exp2

> you can change `-abs(a-b)` to be `min - max` I think the former is better actually, since `min`/`max` has 4 cycles of latency while `abs` and `neg` are bitwise...

logaddexp2: Use log1p and exp2

@pytorchbot merge

Vectorize torch.exp2 on CPU and add complex support

@pytorchbot merge -f "CI failure is present on master"

Replace log(1 + x) with log1p(x)

@lezcano I did see the distribution uses but not sure what our reproducibility guarantees are, so didn't want to touch it. For the logit functions, I tried `log(x) - log1p(-x)`...

Replace exp(x) - 1 with expm1(x)

@pytorchbot merge

[caffe2] Remove IValue include from operator.h

@dagitses PTAL

[caffe2] Export operators to c10 without including ATen/Tensor.h

There is still a caffe2 job in PyTorch trunk CI so I think it still has value.

Add test tracking operators without decompositions

@pytorchbot merge

Add test tracking operators without decompositions

@pytorchbot merge -f "Dynamo failure is pre-existing on master"