KBaba7/llama.cpp
0
1#version 4502 3#extension GL_EXT_shader_16bit_storage : require4 5layout (push_constant) uniform parameter6{7 uint nb1;8 uint dim;9 uint max_period;10} p;11 12#include "types.comp"13 14#extension GL_EXT_control_flow_attributes : enable15#define BLOCK_SIZE 25616 17layout(local_size_x = BLOCK_SIZE, local_size_y = 1, local_size_z = 1) in;18 19layout (binding = 0) readonly buffer X {A_TYPE data_a[];};20layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};21 22void main() {23 const uint i = gl_WorkGroupID.y;24 const uint j = gl_GlobalInvocationID.x;25 const uint d_offset = i * p.nb1;26 27 if (p.dim % 2 != 0 && j == ((p.dim + 1) / 2)) {28 data_d[d_offset + p.dim] = 0.f;29 }30 31 const uint half_dim = p.dim / 2;32 if (j >= half_dim) {33 return;34 }35 36 const float timestep = float(data_a[i]);37 const float freq = float(exp(-log(p.max_period) * j / half_dim));38 const float arg = timestep * freq;39 data_d[d_offset + j] = D_TYPE(cos(arg));40 data_d[d_offset + j + half_dim] = D_TYPE(sin(arg));41}42 