Tobias Ribizel

Add transform_output_iterator

6

oneDPL has a variety of useful fancy iterators and views (transform, counting, iota, zip, permutation ...), but is missing an important type of iterator that can be used to fuse...

reduce_by_segment compilation failure for non-primitive types

3

Take the following minimal examples of `reduce_by_segment` calls ```cpp #include #include #include #include void foo() { int* a = nullptr; double* b = nullptr; oneapi::dpl::reduce_by_segment(oneapi::dpl::execution::par_unseq, a, a, b, a, b);...

Add BGZIP `data_chunk_reader`

3

## Description This adds a BGZIP `data_chunk_reader` usable with `multibyte_split`. The BGZIP format is a modified GZIP format that consists of multiple blocks of at most 65536 bytes compressed data...

Add BGZIP multibyte_split benchmark

2

## Description This refactors #11652 to extract the BGZIP IO and adds another `source_type` to the `multibyte_split` benchmark, creating a compressed file using `zlib`. For reviews, only changes starting from...

**Is your feature request related to a problem? Please describe.** To parse files from a regular language like CSV, we can only safely parse data inside a byte range if...

Unified tests

3

This is a big one 😀 Most of our tests are 90% - 100% identical between different executors. In terms of maintenance, I don't think there is value in keeping...

Deprecating `gko::range`

2

I think we should deprecate our range implementation and the surrounding code in favor of the `accessor` interface. The reasons are 1. they are nothing but a thin wrapper over...

Gracefully handle overflows

4

There are many places where integer computations might overflow in Ginkgo. I think the most important one is the `prefix_sum` kernel. So I would like to suggest that we add...

ATM, we only attach loggers to the device executor in our benchmarks. That means that host operations like they are used in RCM and symbolic Cholesky are missing entirely from...

Move implementation from headers to source files

5

This PR moves almost all function implementations from headers to source files. This should make the headers easier to read and maybe also faster to compile. To finish this up,...