vulkan: small mul_mat_vec optimizations #10665
0cc4m
merged
44 commits into
ggml-org:master
from
vulkan
dot and delta optimization
b7ad2345
server : fix default draft model parameters (#10586)
be2d0048
github : minify link [no ci]
ed8649f8
github : minify link [no ci] (revert)
ca7c2135
metal : small-batch mat-mul kernels (#10581)
d6753d70
readme : add option, update default value, fix formatting (#10271)
d37b7e09
llama : add missing LLAMA_API for llama_chat_builtin_templates (#10636)
f697baf8
metal : add `GGML_OP_CONV_TRANSPOSE_1D` kernels (ggml/1026)
e92a46be
feat: add `GGML_UNARY_OP_ARGMAX` Metal kernel (ggml/1019)
2b155906
CUDA: remove unnecessary warp reduce in FA (ggml/1032)
0df0452a
sync : ggml
69c7f204
scripts : remove amx sync
f8fe71ab
server : (web ui) Various improvements, now use vite as bundler (#10599)
70f0346f
vulkan: optimize and reenable split_k (#10637)
fa9abd6c
clip : add sycl support (#10574)
0fa9dc4c
Add docs for creating a static build (#10268) (#10630)
0a81a82f
Avoid using __fp16 on ARM with old nvcc (#10616)
9075271c
fix typo of README.md (#10605)
4153d57c
SYCL : Move to compile time oneMKL interface backend selection for NV…
e147054b
remove a multiply
062f256e
merge
fe811349
Merge https://github.com/ggerganov/llama.cpp into vulkan
c403d895
remove a multiply
5fbaf121
additional small optimizations
2f56bac7
Merge https://github.com/ggerganov/llama.cpp into vulkan
591894a0
Merge branch 'ggerganov:master' into vulkan
0b1b7c85
0cc4m
commented
on 2024-12-06
Merge branch 'vulkan' of https://github.com/netrunnereve/llama.cpp in…
4eefebc8
Merge branch 'ggerganov:master' into vulkan
32b994e8
Merge branch 'vulkan' of https://github.com/netrunnereve/llama.cpp in…
4b65c6b9
remove ifdefs
f5a15fc6
cleanup
bd17bc45
double the number of rows per workgroup
4a185ad3
Update ggml-vulkan.cpp
984d4707
Vulkan: Add VK_EXT_subgroup_size_control support to ensure full subgr…
595c1a7d
only increase the number of rows for amd and subgroup size 64
6de28665
merge
bfecabeb
fix missing NUM_ROWS for mul_mat_vec_iq4_nl_f16_f32, untested
1c163674
0cc4m
commented
on 2024-12-08
Merge branch '0cc4m/vulkan-subgroup-size-control' of https://github.c…
8972f1d3
Merge https://github.com/ggerganov/llama.cpp into vulkan
c7bc42ce
use subgroup min and max to check for gcn (requires https://github.co…
9af9e801
manual merge ggml-vulkan.cpp
d9c6bf16
fix conflict
8b13f2d0
jeffbolznv
approved these changes
on 2024-12-12
set min and max subgroup size in any case
1aa26d78
Also double the number of rows for Intel GPUs
20b47d4d
Login to write a write a comment.
Login via GitHub