Elias Stehle

icon indicating a website link http://eliasstehle.com

icon company NVIDIA icon location Zurich icon location Research scientist and CUDA C++ developer. Helping to leverage GPUs as the compute platform for data processing.

Results 23 issues of Elias Stehle

Add support for large `num_items` to `DevicePartition::ThreeWayPartition`

Simply instantiating the kernel template with different offset types unfortunately does degrade performance by as much as 1.66-fold. This needs further investigation on alternative approaches, like streaming, bit-packed tile state,...

Adds support for large number of items to `DeviceScan::*ByKey` family of algorithms

2 comment

## Description Closes https://github.com/NVIDIA/cccl/issues/2458 `ScanByKey` used to have the tile state comprising (1) the accumulated value and (2) the `OffsetT`. The `OffsetT` part is used by `ReduceByKey` to figure the...

Add support for large `num_items` to `DeviceScan::*ByKey`

Simply instantiating the `scan_by_key` kernel template with a 64-bit offset type turns out that it incurs some noticeable performance downside of -~35% (or 1.5-fold slowdown) . Hence, we want to...