308 : [args_ptr]
"r" (&ka), [offsetof_B_stride]
"I" (offsetof(KernelArgs, B_stride)), [offsetof_Bpanel]
"I" (offsetof(KernelArgs, Bpanel)), [offsetof_K]
"I" (offsetof(KernelArgs,
K)), [offsetof_N]
"I" (offsetof(KernelArgs,
N)), [offsetof_cur_B_ptr]
"I" (offsetof(KernelArgs, cur_B_ptr))