#version 450

layout(local_size_x = 256) in;
layout(binding = 0) readonly buffer X {
    float x[];
};
layout(binding = 1) readonly buffer S {
    float scale[];
};
layout(binding = 2) readonly buffer H {
    float shift[];
};
layout(binding = 3) writeonly buffer D {
    float dst[];
};
layout(push_constant) uniform Parameters {
    uint ne[4], scale_ne[4], shift_ne[4];
    uint x_stride[4], scale_stride[4], shift_stride[4];
    uint offsets[4];
    uint count, has_shift;
}
p;

void main() {
    uint group = gl_WorkGroupID.y * gl_NumWorkGroups.x + gl_WorkGroupID.x;
    if (group >= (p.count + 255) / 256) {
        return;
    }
    uint i = group * 256 + gl_LocalInvocationID.x;
    if (i >= p.count) {
        return;
    }
    uint index = i;
    uint xi = p.offsets[0], si = p.offsets[1], hi = p.offsets[2];
    for (uint d = 0; d < 4; ++d) {
        uint coord = index % p.ne[d];
        index /= p.ne[d];
        xi += coord * p.x_stride[d];
        si += (coord % p.scale_ne[d]) * p.scale_stride[d];
        hi += (coord % p.shift_ne[d]) * p.shift_stride[d];
    }
    precise float product = x[xi] * scale[si];
    precise float value   = x[xi] + product;
    if (p.has_shift != 0) {
        value += shift[hi];
    }
    dst[p.offsets[3] + i] = value;
}