1  #version 450
 2  
 3  #include "types.comp"
 4  #include "generic_binary_head.comp"
 5  
 6  void main() {
 7      const uint i00 = gl_GlobalInvocationID.x;
 8      const uint i10 = gl_GlobalInvocationID.y;
 9      const uint i11 = (gl_GlobalInvocationID.z)/p.ne12;
10      const uint i12 = (gl_GlobalInvocationID.z)%p.ne12;
11  
12      if (i00 >= p.ne00) {
13          return;
14      }
15  
16      const uint i01 = data_b[i10*p.nb10 + i11*p.nb11 + i12*p.nb12];
17  
18      const uint a_offset = i01*p.nb01 + i11*p.nb02 + i12*p.nb03;
19      const uint d_offset = i10*p.nb21 + i11*p.nb22 + i12*p.nb23;
20  
21  #ifndef OPTIMIZATION_ERROR_WORKAROUND
22      data_d[d_offset + i00] = D_TYPE(data_a[a_offset + i00]);
23  #else
24      data_d[d_offset + i00] = data_a[a_offset + i00];
25  #endif
26  }