vulkan: use GL_NV_cooperative_matrix_decode_vector for faster matmul (llama/23541)
J
Jeff Bolz committed
1b590bbb9ae834d31d1116e804249296bc83762c
Parent: c5cde8c
Committed by Georgi Gerganov <ggerganov@gmail.com>
on 5/29/2026, 6:47:30 AM