llama.cpp/vulkan-shaders/dequant_funcs.comp

#if !defined(DATA_A_F32) && !defined(DATA_A_F16)
#extension GL_EXT_shader_explicit_arithmetic_types_int8 : require
#endif

#if defined(DATA_A_F32)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    return vec2(data_a[a_offset + ib], data_a[a_offset + ib + 1]);
}
#endif

#if defined(DATA_A_F16)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    return vec2(data_a[a_offset + ib], data_a[a_offset + ib + 1]);
}
#endif

#if defined(DATA_A_Q4_0)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    const float d = float(data_a[a_offset + ib].d);
    const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
    return (vec2(vui & 0xF, vui >> 4) - 8.0f) * d;
}
#endif

#if defined(DATA_A_Q4_1)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    const float d = float(data_a[a_offset + ib].d);
    const float m = float(data_a[a_offset + ib].m);
    const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
    return vec2(vui & 0xF, vui >> 4) * d + m;
}
#endif

#if defined(DATA_A_Q5_0)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    const float d = float(data_a[a_offset + ib].d);
    const uint uint_qh = uint(data_a[a_offset + ib].qh[1]) << 16 | data_a[a_offset + ib].qh[0];
    const ivec2 qh = ivec2(((uint_qh >> iqs) << 4) & 0x10, (uint_qh >> (iqs + 12)) & 0x10);
    const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
    return (vec2((vui & 0xF) | qh.x, (vui >> 4) | qh.y) - 16.0f) * d;
}
#endif

#if defined(DATA_A_Q5_1)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    const float d = float(data_a[a_offset + ib].d);
    const float m = float(data_a[a_offset + ib].m);
    const uint uint_qh = data_a[a_offset + ib].qh;
    const ivec2 qh = ivec2(((uint_qh >> iqs) << 4) & 0x10, (uint_qh >> (iqs + 12)) & 0x10);
    const uint vui = uint(data_a[a_offset + ib].qs[iqs]);
    return vec2((vui & 0xF) | qh.x, (vui >> 4) | qh.y) * d + m;
}
#endif

#if defined(DATA_A_Q8_0)
vec2 dequantize(uint ib, uint iqs, uint a_offset) {
    const float d = float(data_a[a_offset + ib].d);
    return vec2(int(data_a[a_offset + ib].qs[iqs]), int(data_a[a_offset + ib].qs[iqs + 1])) * d;
}
#endif