]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
vulkan: optimize mul_mat_id loading row ids into shared memory (llama/15427)
authorJeff Bolz <redacted>
Sat, 23 Aug 2025 06:31:54 +0000 (01:31 -0500)
committerGeorgi Gerganov <redacted>
Fri, 5 Sep 2025 09:54:01 +0000 (12:54 +0300)
commitb238604c3949144794cebc9fe980388a5197e01e
tree735a0fee0cef7c5ea780c232775b6d3f5e6d80e1
parent899845299d5224e15ba4f91e6ee0c3ca98042d7b
vulkan: optimize mul_mat_id loading row ids into shared memory (llama/15427)

- Spread the work across the whole workgroup. Using more threads seems to
far outweigh the synchronization overhead.
- Specialize the code for when the division is by a power of two.
src/ggml-vulkan/ggml-vulkan.cpp
src/ggml-vulkan/vulkan-shaders/mul_mm.comp
src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp