[Metal] [Performance] Add radix select implementation for efficient partition operations#3069

Closed

NripeshN wants to merge 21 commits intoml-explore:mainfrom

NripeshN:cuda-radix-select

Commits on Jan 27, 2026

Add radix select implementation for efficient partition operations

NripeshN
authored and
awni
committed
Add radix select implementation for Metal partition operations

NripeshN
authored and
awni
committed
Fix Metal radix select compilation errors

NripeshN
authored and
awni
committed
Add multi-pass radix select kernel stubs and benchmark script

NripeshN
authored and
awni
committed
Implement multi-pass radix select for Metal partition operations

NripeshN
authored and
awni
committed
reset cuda

NripeshN
authored and
awni
committed
Enhance radix select implementation for Metal with optimized streaming kernel

NripeshN
authored and
awni
committed
clean comments

NripeshN
authored and
awni
committed
lint

NripeshN
authored and
awni
committed
Add non-contiguous streaming kernel for radix select operations

NripeshN
authored and
awni
committed
cuda implementation

NripeshN
authored and
awni
committed
nits
awni
committed
fallback to merge sort when rows small and n.elements big
NripeshN
committed
Enhance radix select algorithm with dynamic thresholding for sort fallback
NripeshN
committed
cuda revert
NripeshN
committed
Revert so called "optimisations" that caused overhead
NripeshN
committed
remove benchmark
awni
committed
cuda attempt cub::DeviceTopK
NripeshN
committed

Commits on Jan 31, 2026

Commits on Feb 7, 2026

Fix for compilation on metal
angeloskath
committed