llama.cpp
Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing
#10597
Merged
Merged
Vulkan: VK_KHR_cooperative_matrix support to speed up prompt processing #10597
0cc4m
merged
12 commits into
master
from
0cc4m/vulkan-coopmat
Vulkan: Implement VK_KHR_cooperative_matrix support in the matrix mat…
a2b5822c
Improve performance with better q4_k and q5_k dequant and store unrol…
21c18ee2
Add Vulkan MUL_MAT and MUL_MAT_ID accumulator precision selection
412d5616
Rework mulmat shader selection and compilation logic, avoid compiling…
f3936a4b
Vulkan: Implement accumulator switch for specific mul mat mat shaders
0cb02c01
Vulkan: Unroll more loops for more mul mat mat performance
56c67df5
Vulkan: Add VK_AMD_shader_core_properties2 support to read Compute Un…
13a068e2
Disable coopmat support on AMD proprietary driver
3a58a015
Remove redundant checks
f54afb4f
Add environment variable GGML_VK_DISABLE_COOPMAT to disable VK_KHR_co…
c66d0080
jeffbolznv
approved these changes
on 2024-12-05
Fix rebase typo
e1d17a20
Fix coopmat2 MUL_MAT_ID pipeline selection
eeaf0b94
Login to write a write a comment.
Login via GitHub