Pearu Peterson

Take a sparse histogram input of a given dimensionality and output a single column dense grid, i.e. `input (x, y) => count(*)` to a dense array/raster of `count(*)`, which would...

See also https://github.com/Quansight/omnisci/issues/38

As an entry point, see  https://github.com/omnisci/omniscidb-internal/blob/master/ThriftHandler/DBHandler.cpp#L6174

Static or runtime (or both) bounds checking for reads/writes to make functions safe. Static bounds checking could be potentially written as an LLVM pass that does simple min/max access range...

Add blocksparse_int_addmm. Eliminate unnecessary contiguous calls which leads to performance increase.

1

As in the title. This PR is created on top of https://github.com/pytorch/ao/pull/821 and requires https://github.com/pytorch/pytorch/pull/136104 . The diff of `jcaip/int8-bsr` and `pearu/int8-bsr` branches is ``` diff --git a/torchao/dtypes/affine_quantized_tensor.py b/torchao/dtypes/affine_quantized_tensor.py index...

printf that can be called from Numba jit-decorated function

5

## Feature request Having the following implementation ``` from numba import extending, cgutils, types @extending.intrinsic def printf(typingctx, format_type, *args): """printf that can be called from Numba jit-decorated functions. """ if...

How to prevent landing PRs on sparse tensors that should be rejected?

4

Recently, https://github.com/pytorch/pytorch/pull/169807 was submitted that added out-of-bounds checks for inputs of constructing a sparse COO tensor. Sounds reasonable, right? No, it is not right because the corresponding checks already exist...