git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit

]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit

overview / pkg / ggml / sources / llama.cpp / commit

author	Hongqiang Wang <redacted>
	Wed, 15 Jul 2026 16:08:40 +0000 (09:08 -0700)
committer	GitHub <redacted>
	Wed, 15 Jul 2026 16:08:40 +0000 (09:08 -0700)
commit	c3d47e696b1187a27e896aa828d48ff9a33fc679
tree	3ff37fbd6333e9a86a9a1e433c18a76dc7a6b693	tree
parent	f6f12e43fa869ef0e008b99ed97dc4006bbb8907	commit \| diff

opencl: fix two issues on flash attention for Adreno a7x (#25697)

* opencl: route `sub_group_shuffle_xor` to qcom ext when KHR ext is unavailable

KHR `sub_group_shuffle_xor` is not defined by compiler when
`cl_qcom_subgroup_shuffle` is present, causing certain FA
kernels fail to build. Define the KHR shuffle_xor using
the qcom extension.

* opencl: skip FA kernels with mixed and quant types for A7x to avoid compiler crash

Packaging of ggml-org/llama.cpp

RSS Atom

ggml/src/ggml-opencl/ggml-opencl.cpp		diff \| blob \| history
ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl		diff \| blob \| history
ggml/src/ggml-opencl/kernels/flash_attn_f32_q4_0.cl		diff \| blob \| history
ggml/src/ggml-opencl/kernels/flash_attn_f32_q8_0.cl		diff \| blob \| history
ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_l4.cl		diff \| blob \| history