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

vulkan: add DeepSeek-V4 hyper-connection fused ops (DSV4_HC_COMB/PRE/POST) (#26578)

Browse Source 
* vulkan: add DeepSeek-V4 hyper-connection fused ops (DSV4_HC_COMB/PRE/POST)

CUDA has these ops from the DeepSeek-V4 merge and Metal gained them in
PR 26459. Vulkan was the last major backend running the unfused primitive
chain. On DeepSeek-V4-Flash the unfused Sinkhorn comb chain alone takes
about 32% of decode op time on gfx1151 (Strix Halo), spread over roughly
16k dispatches per token.

dsv4_hc_comb runs the full 20-iteration Sinkhorn in registers. A token's
4x4 comb matrix lives in 16 consecutive subgroup lanes, with idst in bits
0-1 and isrc in bits 2-3 to match the CPU reference layout, so
subgroupShuffleXor by 1|2 reduces rows and by 4|8 reduces columns. One
dispatch replaces about 137 strictly ordered node executions per site.
The shuffle masks never cross a 16-lane boundary, so a subgroup of size
64 packs 4 independent tokens.

dsv4_hc_pre and dsv4_hc_post handle the elementwise stream collapse and
fan-out, with per-token coefficients staged in shared memory.

GGML_VK_DISABLE_DSV4_HC disables all three ops. The _COMB, _PRE and
_POST variants gate each op independently so a single kernel can be
bisected against the unfused graph.

Adds eval cases at the production n_iter=20 across batch sizes that
cross subgroup and workgroup boundaries.

* vulkan: dsv4 hc review fixes

Drop the per-op env-var disables and device flags, the stride divisibility
check (ggml guarantees it) and the workgroup-count fallback in supports_op.
Trim the comb shader comments to the lane layout.

---------

Co-authored-by: Kevin Hopper <no-reply@maestro.press>
This commit is contained in:
Kevin Hopper
2026-09-07 15:24:03 +02:00
committed by GitHub
co-authored by Kevin Hopper
parent 0c963452ea
commit 7a333e7240
6 changed files with 462 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp
+3
View File
@@ -1042,6 +1042,9 @@ void process_shaders() {
string_to_spv("fwht_f32", "fwht.comp", {});
string_to_spv("fwht_shmem_f32", "fwht.comp", {{"FWHT_SHMEM", "1"}});
string_to_spv("count_equal_i32", "count_equal.comp", merge_maps(base_dict, {{"A_TYPE", "int"}, {"B_TYPE", "int"}, {"D_TYPE", "int"}}));
string_to_spv("dsv4_hc_comb_f32", "dsv4_hc_comb.comp", {});
string_to_spv("dsv4_hc_pre_f32", "dsv4_hc_pre.comp", {});
string_to_spv("dsv4_hc_post_f32", "dsv4_hc_post.comp", {});
string_to_spv("cumsum_f32", "cumsum.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("cumsum_multipass1_f32", "cumsum_multipass1.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));
string_to_spv("cumsum_multipass2_f32", "cumsum_multipass2.comp", merge_maps(base_dict, {{"A_TYPE", "float"}, {"D_TYPE", "float"}}));