Jeff Bolz cea22b3075 vulkan: For coopmat2 FA, use fp16 accumulators for the final result (llama/19376) ... The cpu and cuda backends use fp16 for the VKQ accumulator type, this change does the same for vulkan. This helps particularly with large head sizes which are very register-limited. I tried this for the coopmat1 path and it slowed down a bit. I didn't try for scalar. I applied the softmax bias that the cuda backend uses to avoid overflow, although I was not able to reproduce the original bug without it.		2026-02-08 09:29:10 +02:00
..
cmake	cmake: fix ggml-shaders-gen compiler paths containing spaces (llama/12747)	2025-04-24 20:39:16 +03:00
vulkan-shaders	vulkan: For coopmat2 FA, use fp16 accumulators for the final result (llama/19376)	2026-02-08 09:29:10 +02:00
CMakeLists.txt	vulkan: Improve build time for MSVC (llama/16545)	2025-10-15 09:29:17 +03:00
ggml-vulkan.cpp	vulkan: make FA mask/softcap enables spec constants (llama/19309)	2026-02-08 09:29:10 +02:00