Jake Hemstad

[REVIEW] NULL support for hash-based group by

8

This PR adds support for NULL values to the hash-based groupby implementation. The `build_aggregation_table` kernel was updated such that if a row is invalid, then it is not inserted into...

[FEA] Invalidate `get_size()` after user calls `get_device_mutable_view()`

1

**Is your feature request related to a problem? Please describe.** The value of `static_map::get_size()` is only guaranteed to be correct if a user goes through the bulk `static_map::insert` API. I...

### Is your feature request related to a problem? Please describe. As a user of cuco, there are times where I want to treat existing memory as if it were...

[FEA] Refactor of open address data structures

8

**Is your feature request related to a problem? Please describe.** There is a significant amount of redundancy among the `static_map/static_multimap/static_reduction_map` classes. This is a large maintenance overhead and means optimizations...

Generating vector load instructions is important for performance. However, the compiler can be very fussy about _actually_ generating them. For example, https://godbolt.org/z/sfdaj695P shows prototyping code where all versions of nvcc...

We currently only have benchmarks for the bulk APIs: https://github.com/NVIDIA/cuCollections/tree/dev/benchmarks We should add benchmarks for the device-side APIs as well.

[FEA] Add option to select between Struct of Arrays vs Array of Structs

1

**Is your feature request related to a problem? Please describe.** All of the current hash map implementations use an Array of Structs storage implementation. There are instances where it could...

Use NV_IF_TARGET instead of CUDA_ARCH

1

This reminds me that in order to be friendly to nvc++, we should probably be switching to use `NV_IF_TARGET`. That'll be a different issue though. _Originally posted by @jrhemstad in...

**Is your feature request related to a problem? Please describe.** After @PointKernel's excellent work in https://github.com/NVIDIA/cuCollections/pull/141 to ensure each cuco header is standalone, we should add some tests to ensure...

[FEA] Hash map design that supports reduce by key

2

**Is your feature request related to a problem? Please describe.** I would like to be able to implement a hash-based reduce-by-key operation. A `thrust::reduce_by_key` operation requires first doing a `sort_by_key`....