mlx_native/ops/
mod.rs

1//! GPU kernel host-side dispatch functions.
2//!
3//! Each submodule implements dispatch for a specific kernel family.
4
5pub mod argmax;
6pub mod argsort;
7pub mod moe_mm_id_map0;
8pub mod top_k;
9pub mod copy;
10pub mod cumsum;
11pub mod dense_gemm;
12pub mod dense_gemv_bf16;
13pub mod dense_mm_bf16;
14pub mod dense_mm_f16;
15pub mod dense_mm_f32_f32;
16pub mod dequant_to_f16;
17pub mod elementwise;
18pub mod embedding;
19pub mod gather;
20pub mod gather_bench;
21pub mod hadamard;
22pub mod hadamard_quantize_kv;
23pub mod encode_helpers;
24pub mod fused_dual_proj_q4_0;
25pub mod fused_gate_up_silu_iq4_nl;
26#[allow(non_snake_case)]
27pub mod fused_gate_up_silu_q4_K;
28#[allow(non_snake_case)]
29pub mod fused_gate_up_silu_q5_K;
30#[allow(non_snake_case)]
31pub mod fused_gate_up_silu_q6_K;
32pub mod fused_gate_up_silu_q8_0;
33pub mod fused_head_norm_rope;
34pub mod fused_norm_add;
35pub mod fused_residual_norm;
36pub mod gelu;
37pub mod kv_cache_copy;
38pub mod l2_norm;
39pub mod log_elementwise;
40pub mod row_sum;
41pub mod moe_dispatch;
42pub mod moe_gate;
43pub mod moe_softmax_topk;
44pub mod mul_mv_ext;
45pub mod moe_weighted_reduce;
46pub mod qkv_split;
47pub mod repeat_tiled;
48pub mod quantized_matmul;
49pub mod quantized_matmul_ggml;
50pub mod quantized_matmul_id;
51pub mod quantized_matmul_id_ggml;
52pub mod rms_norm;
53pub mod rope;
54pub mod rope_multi;
55pub mod rope_train;
56pub mod vision_2d_rope;
57pub mod scale_mask_softmax;
58pub mod sigmoid_mul;
59pub mod silu_mul;
60pub mod compute_g_beta;
61pub mod ssm_norm_gate;
62pub mod flash_attn_prefill;
63pub mod flash_attn_prefill_blk;
64pub mod flash_attn_train;
65pub mod flash_attn_prefill_d512;
66pub mod flash_attn_prefill_mask;
67pub mod flash_attn_vec;
68pub mod flash_attn_vec_tq;
69// ADR-037 Phase E1.1 (2026-05-22) — EAGLE-3 + dynamic tree port.
70pub mod tree_attention;
71pub mod flash_attn_vec_tq_hb;
72pub mod flash_attn_vec_hybrid;
73pub mod flash_attn_vec_peer_port_f16;
74pub mod flash_attn_vec_reduce_tq_hb_undo;
75pub mod fwht_standalone;
76pub mod chunk_gated_delta_rule;
77pub mod chunk_gated_delta_rule_bank_split;
78pub mod chunk_gated_delta_rule_tri_solve_invert;
79pub mod gated_delta_net;
80pub mod gated_delta_net_decode;
81pub mod gated_delta_net_chunk;
82pub mod gated_delta_net_chunk_o;
83pub mod gated_delta_net_kkt;
84pub mod gated_delta_net_recompute_wu;
85pub mod tq_dequantize_kv;
86pub mod sdpa;
87pub mod sdpa_decode;
88pub mod sdpa_sliding;
89pub mod softcap;
90pub mod softmax;
91pub mod softmax_backward;
92pub mod softmax_sample;
93pub mod ssm_conv;
94pub mod conv1d_depthwise_causal;
95pub mod exp_elementwise;
96pub mod outer_product;
97pub mod take_along_axis;
98pub mod divide_elementwise;
99pub mod sqrt_elementwise;
100pub mod transpose;
101pub mod tri_solve;
102pub mod qdq_legacy;
103pub mod rms_norm_backward;
104pub mod slice_concat_2d;
105pub mod silu_backward;
106pub mod embedding_autograd;
107pub mod adam_update;
108pub mod qdq_affine;
109pub mod qmm_affine;
110// ADR-021 Qwen3-VL ViT prelude GPU port (5 new kernels):
111pub mod im2col_2d_3ch;
112pub mod add_bias_row_2d;
113pub mod bilinear_resize_2d;
114pub mod block_merge_2x2;
115pub mod feature_concat;
mlx_native/ops/mod.rs

mlx_native/ops/
mod.rs
