nikitaved

`mul`: sparse-dense + sparse-sparse with 0-dims support take 2.

4

This one is a copy of https://github.com/pytorch/pytorch/pull/81556 https://github.com/pytorch/pytorch/pull/82717 These got reverted due to issues with torchvision. CC @kit1980 , could you please take over from here?

Faster mul(sparse, sparse) with broadcasting in dense dims.

1

This is a combo PR of https://github.com/pytorch/pytorch/pull/84929 and https://github.com/pytorch/pytorch/pull/83428. Preliminary benchmarks (square matrices of shape (n, n)). Script ```python import torch import math from IPython import get_ipython from itertools import...

csr <-> csc, csc <-> csc, bsr <-> bsc, bsc <-> bsc, bsr <-> bsr conversions

2

As per title. Required to enable a wider selection of sparse formats for `nn.functional.linear`.

Improve `bsr @ strided` performance in `baddmm` for `bfloat16/half` with Triton kernels.

68

As per title. Additionally we also introduce support for: - Rectangular block sizes which are powers of 2 and at least 16 (triton's `dot` limitation). - Batch support with broadcasting...

COO intersection kernel: respect value intersection order

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #92976 * __->__ #92242 * #91714 * #91704 * #92248

COO intersection primitives: performance improvement

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #92976 * #92242 * #91714 * #91704 * #92248

retire sparse_mask_helper

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #92976 * #92242 * __->__ #91714 * #91704 * #92248 cc @pearu @cpuhrsch @amjames @bhosmer

_sparse_coo_tensor_with_dims_and_tensors backward: simplify and optimize

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #92976 * #92242 * #91714 * __->__ #91704 * #92248 cc @pearu @cpuhrsch @amjames @bhosmer @ezyang @albanD @zou3519 @gqchen @soulitzer @Lezcano @Varal7

sparse_mask: extend lhs to sparse COO tensors

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #92976 * #92242 * #91714 * #91704 * __->__ #92248 cc @alexsamardzic @pearu @cpuhrsch @amjames @bhosmer

Enable `bsr_dense_mm` Triton kernel in `nn.functional.linear`.

1

As per title. cc @alexsamardzic @pearu @cpuhrsch @amjames @bhosmer