mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-03-08 13:00:17 +00:00
38 lines
876 B
Plaintext
38 lines
876 B
Plaintext
#version 450
|
|
|
|
#include "types.comp"
|
|
|
|
#extension GL_EXT_shader_16bit_storage : require
|
|
#extension GL_EXT_control_flow_attributes : enable
|
|
|
|
layout (push_constant) uniform parameter
|
|
{
|
|
uint ne;
|
|
uint ne0;
|
|
uint ne1;
|
|
uint nb1;
|
|
uint nb01;
|
|
uint nadd;
|
|
} p;
|
|
|
|
layout(local_size_x = 512, local_size_y = 1, local_size_z = 1) in;
|
|
|
|
layout (binding = 0) readonly buffer X {A_TYPE data_a[];};
|
|
layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
|
|
|
|
void main() {
|
|
const uint i = gl_GlobalInvocationID.z * 262144 + gl_GlobalInvocationID.y * 512 + gl_GlobalInvocationID.x;
|
|
|
|
if (i >= p.ne) {
|
|
return;
|
|
}
|
|
|
|
uint i1 = i / p.ne0;
|
|
uint i0 = i % p.ne0;
|
|
|
|
float sum = data_a[i1*p.nb01 + i0] + data_a[i1*p.nb01 + i0 + p.ne0];
|
|
for (int j = 2; j < p.nadd; ++j) sum += data_a[i1*p.nb01 + i0 + j*p.ne0];
|
|
|
|
data_d[i1*p.nb1 + i0] = D_TYPE(sum);
|
|
}
|