Struct Glm5NextDsaLayerKernels 

pub struct Glm5NextDsaLayerKernels {
    pub gemm: KernelHandle,
    pub gemm_f32: KernelHandle,
    pub gemv: KernelHandle,
    pub gemv_f32: KernelHandle,
    pub gemv_batchm: KernelHandle,
    pub rms_norm: KernelHandle,
    pub latent_write: KernelHandle,
}

2026-09-25: The projection, norm and latent-write kernels of a DSA block. The selection kernels are in Glm5NextDsaKernels, the decode kernel in attend.

Fields

gemm: KernelHandle

2026-09-25: dense_gemm_bf16, C = A @ B^T, BF16 out.

gemm_f32: KernelHandle

2026-09-25: The same with FP32 out, for the selector’s q_idx and head weights.

gemv: KernelHandle

2026-09-25: M = 1 kernels for gemm and gemm_f32. gemv_f32 is KernelHandle(0) on a target without dense_gemv_bf16_fp32out, and dense_mm_bf16 then uses the tile GEMM.

gemv_f32: KernelHandle

gemv_batchm: KernelHandle

2026-09-25: dense_gemv_bf16_batchm: 2..=DENSE_GEMV_BATCHM_MAX_M rows in one pass over the weight; KernelHandle(0) when the target lacks it.

rms_norm: KernelHandle

2026-09-25: rms_norm_vanilla; see the module header.

latent_write: KernelHandle

2026-09-25: glm5next_mla_latent_write_fp8: RMSNorm, FP8 quantisation and the paged slot write of the KV latent.

Implementations

impl Glm5NextDsaLayerKernels

pub fn resolve(gpu: &dyn GpuBackend) -> Result<Self>

Trait Implementations

impl Clone for Glm5NextDsaLayerKernels

fn clone(&self) -> Glm5NextDsaLayerKernels

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for Glm5NextDsaLayerKernels