char-fork/llama.cpp
git clone https://git.t4t.associates/char-fork/llama.cpp
6c97bffd6
char/aliceai-t5-moe
1kernel void kernel_diag_f32 ( 2global const char * src0 , 3ulong offset0 , 4global char * dst , 5ulong offsetd , 6ulong nb01 , 7ulong nb02 , 8ulong nb03 , 9int ne0 , 10ulong nb0 , 11ulong nb2 , 12ulong nb3 13) { 14 src0 = src0 + offset0; 15 dst = dst + offsetd; 16 17 int i3 = get_group_id(2); 18 int i2 = get_group_id(1); 19 int i1 = get_group_id(0); 20 21 global const float * src0_ptr = (global const float *)(src0 + i2*nb02 + i3*nb03); 22 global float * dst_ptr = (global float *)(dst + i1*nb01 + i2*nb2 + i3*nb3); 23 24 for (int i0 = get_local_id(0); i0 < ne0; i0 += get_local_size(0)) { 25 dst_ptr[i0] = i0 == i1 ? src0_ptr[i0] : 0.0f; 26 } 27}