1  #version 450
 2  
 3  #include "generic_head.comp"
 4  #include "types.comp"
 5  
 6  #extension GL_EXT_control_flow_attributes : enable
 7  layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
 8  
 9  layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
10  layout (binding = 1) writeonly buffer D {D_TYPE data_d[];};
11  
12  layout (constant_id = 0) const uint BLOCK_SIZE = 32;
13  
14  shared FLOAT_TYPE tmp[BLOCK_SIZE];
15  
16  void main() {
17      const uint row = gl_WorkGroupID.x;
18      const uint col = gl_LocalInvocationID.x;
19  
20      tmp[col] = FLOAT_TYPE(0.0f);
21  
22      for (uint i = col; i < p.KX; i += BLOCK_SIZE) {
23          tmp[col] += FLOAT_TYPE(data_a[row*p.KX + i]);
24      }
25  
26      barrier();
27      [[unroll]] for (int s = int(BLOCK_SIZE) / 2; s > 0; s >>= 1) {
28          if (col < s) {
29              tmp[col] += tmp[col + s];
30          }
31          barrier();
32      }
33  
34      if (col == 0) {
35          data_d[row] = D_TYPE(tmp[0]);
36      }
37  }