Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

sycl: add fused top-k MoE (#25217)

Browse Source 
* sycl: add fused top-k MoE

* sycl: address review: GGML_SYCL_ENABLE_FUSION env, move fusion dispatch to topk-moe

* sycl: print GGML_SYCL_ENABLE_FUSION at startup like other env vars

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Frosty40
2026-07-13 09:56:41 +03:00
committed by GitHub
co-authored by Claude Fable 5
parent e474bba7af
commit efb3036c18
6 changed files with 645 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-sycl/topk-moe.hpp
+12
View File
@@ -0,0 +1,12 @@
#ifndef GGML_SYCL_TOPK_MOE_HPP
#define GGML_SYCL_TOPK_MOE_HPP
#include "common.hpp"
// Detect a fusable op subgraph starting at cgraph node `i` and, if found, dispatch the fused
// kernel. Returns the number of *following* nodes consumed (0 = no fusion applies at i).
int ggml_sycl_fuse(ggml_backend_sycl_context & ctx, ggml_cgraph * cgraph, int i);
int ggml_sycl_fuse_topk_moe(ggml_backend_sycl_context & ctx, ggml_cgraph * cgraph, int i);
#endif // GGML_SYCL_TOPK_MOE_HPP