]> git.djapps.eu Git - pkg/ggml/sources/ggml/commit
overview / pkg / ggml / sources / ggml / commit
summary | shortlog | log | commit | commitdiff | tree
(parent: 9bcb7e0)
vulkan: fuse adds (llama/15252)
authorJeff Bolz <redacted>
Sat, 16 Aug 2025 16:48:22 +0000 (11:48 -0500)
committerGeorgi Gerganov <redacted>
Mon, 18 Aug 2025 16:15:25 +0000 (19:15 +0300)
commit8c021e3dd32d8c648cf66dc84be9bf546f572fe8
tree0ce897c1a61dd3252982e6cf3f4939e7e3fbeb2dtree
parent9bcb7e0f5b3201e3b75947043e69526fe5001e02commit | diff
vulkan: fuse adds (llama/15252)

* vulkan: fuse adds

Fuse adds that have the same shape, which are common in MoE models.
It will currently fuse up to 6 adds, because we assume no more than
8 descriptors per dispatch. But this could be changed.

* check runtimeDescriptorArray feature

* disable multi_add for Intel due to likely driver bug
src/ggml-vulkan/ggml-vulkan.cpp diff | blob | history
src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp diff | blob | history
src/ggml-vulkan/vulkan-shaders/multi_add.comp [new file with mode: 0644] blob
src/ggml-vulkan/vulkan-shaders/utils.comp [new file with mode: 0644] blob
src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp diff | blob | history
tests/test-backend-ops.cpp diff | blob | history
Packaging of ggml-org/ggml