hanzo-ml 0.10.2

Minimalist ML framework.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

#version 450
// conv_transpose1d. input [b,c_in,l_in], weight [c_in,c_out,k], out [b,c_out,l_out].
// out[b,oc,ox] = sum over ic,k where (ox+padding-k*dilation) == ix*stride, ix in [0,l_in).
layout(local_size_x = 64) in;
layout(set = 0, binding = 0) readonly  buffer In  { float inp[]; };
layout(set = 0, binding = 1) readonly  buffer W   { float w[]; };
layout(set = 0, binding = 2) writeonly buffer Out { float o[]; };
layout(push_constant) uniform Pc {
    uint b_size; uint c_in; uint c_out; uint l_in; uint l_out;
    uint k_size; uint padding; uint stride; uint dilation;
};
void main() {
    uint gid = gl_GlobalInvocationID.x;
    uint total = b_size * c_out * l_out;
    if (gid >= total) { return; }
    uint ox = gid % l_out;
    uint oc = (gid / l_out) % c_out;
    uint bb = gid / (l_out * c_out);
    float acc = 0.0;
    for (uint k = 0u; k < k_size; k++) {
        int num = int(ox + padding) - int(k * dilation);
        if (num < 0) { continue; }
        if (uint(num) % stride != 0u) { continue; }
        uint ix = uint(num) / stride;
        if (ix >= l_in) { continue; }
        for (uint ic = 0u; ic < c_in; ic++) {
            acc += inp[(bb * c_in + ic) * l_in + ix] * w[(ic * c_out + oc) * k_size + k];
        }
    }
    o[(bb * c_out + oc) * l_out + ox] = acc;
}