Module gather 

GPU-accelerated gather / index_select along dim=0.

Gathers rows from a 2D source tensor using an index array: output[i, :] = src[indices[i], :]

Used for MoE scale factor gathering by expert index.

Statics

GATHER_SHADER_SOURCE

MSL source for the gather kernel (embedded at compile time).

Functions

dispatch_gather_f32

Dispatch a gather / index_select operation on the GPU.

register

Register gather shader source with the given kernel registry.