happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

vulkan: support noncontig i32 copy (#17328)

This commit is contained in:
Jeff Bolz 2025-11-18 00:41:24 -06:00 committed by GitHub
parent 0de8878c96
commit da95bf2a85
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
ggml/src/ggml-vulkan/ggml-vulkan.cpp
View File

@ -13644,10 +13644,11 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
} }
// We can handle copying from a type to the same type if it's // We can handle copying from a type to the same type if it's
// contiguous (memcpy). We use f16 or f32 shaders to do the copy, // either not quantized or is quantized and contiguous.
// We use f16 or f32 shaders to do the copy,
// so the type/block size must be a multiple of 4. // so the type/block size must be a multiple of 4.
if (src0_type == src1_type && if (src0_type == src1_type &&
ggml_is_contiguous(op->src[0]) && ggml_is_contiguous(op) && (!ggml_is_quantized(src0_type) || (ggml_is_contiguous(op->src[0]) && ggml_is_contiguous(op))) &&
(ggml_type_size(src0_type) % 2) == 0) { (ggml_type_size(src0_type) % 2) == 0) {
return true; return true;
} }