#version 450 core

#extension GL_EXT_bfloat16 : require
#extension GL_KHR_cooperative_matrix : enable
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_EXT_shader_explicit_arithmetic_types : enable
#extension GL_EXT_scalar_block_layout : enable

layout(local_size_x = 1, local_size_y = 1, local_size_z = 1) in;

const bfloat16_t bc10 = bfloat16_t(10);
const bf16vec2 b2c20 = bf16vec2(20);
layout(constant_id = 1) const bfloat16_t bsc10 = bfloat16_t(10);

struct S { bfloat16_t b; };
const S sc = S(bc10);

const bfloat16_t bca[2] = {bc10, bc10};

shared bfloat16_t bfs[10];

layout(scalar) buffer B {
    bf16vec3 b3;
    bf16vec2 b2;
    bfloat16_t b1;
} buf;

bfloat16_t funcbf16(bfloat16_t x)
{
    return x;
}

bfloat16_t funcf32(float32_t x)
{
    return bfloat16_t(x);
}

bfloat16_t funcf64(float64_t x)
{
    return bfloat16_t(x);
}

void main()
{
    float f = 2.0;
    bfloat16_t b = bfloat16_t(1.0);
    bfloat16_t(f);
    bf16vec2 b2 = bf16vec2(f);

    uint8_t u8 = uint8_t(5);
    uint16_t u16 = uint16_t(5);
    uint32_t u32 = 5;
    uint64_t u64 = 5;
    int8_t i8 = int8_t(6);
    int16_t i16 = int16_t(6);
    int32_t i32 = 6;
    int64_t i64 = 6;
    bfloat16_t bf16 = bfloat16_t(7);
    float16_t f16 = float16_t(7);
    float32_t f32 = 7;
    float64_t f64 = 7;
    b = bfloat16_t(u8);
    b = bfloat16_t(u16);
    b = bfloat16_t(u32);
    b = bfloat16_t(u64);
    b = bfloat16_t(i8);
    b = bfloat16_t(i16);
    b = bfloat16_t(i32);
    b = bfloat16_t(i64);
    b = bfloat16_t(bf16);
    b = bfloat16_t(f16);
    b = bfloat16_t(f32);
    b = bfloat16_t(f64);
    u8 = uint8_t(b);
    u16 = uint16_t(b);
    u32 = uint32_t(b);
    u64 = uint64_t(b);
    i8 = int8_t(b);
    i16 = int16_t(b);
    i32 = int32_t(b);
    i64 = int64_t(b);
    bf16 = bfloat16_t(b);
    f16 = float16_t(b);
    f32 = float32_t(b);
    f64 = float64_t(b);

    f32 = b;
    f64 = b;

    b = buf.b1;
    b2 = buf.b2;
    bf16vec3 b3 = buf.b3;

    dot(b2, b2);

    coopmat<bfloat16_t, gl_ScopeSubgroup, 16, 16, gl_MatrixUseA> cmA = coopmat<bfloat16_t, gl_ScopeSubgroup, 16, 16, gl_MatrixUseA>(3.0);

    coopmat<float, gl_ScopeSubgroup, 16, 16, gl_MatrixUseA> cmAf = coopmat<float, gl_ScopeSubgroup, 16, 16, gl_MatrixUseA>(cmA);

    funcbf16(b);
    funcf32(b);
    funcf64(b);

    int16_t i16_1 = bfloat16BitsToIntEXT(b);
    i16vec2 i16_2 = bfloat16BitsToIntEXT(b2);
    i16vec3 i16_3 = bfloat16BitsToIntEXT(bf16vec3(b2, b));
    i16vec4 i16_4 = bfloat16BitsToIntEXT(bf16vec4(b2, b2));

    uint16_t u16_1 = bfloat16BitsToUintEXT(b);
    u16vec2  u16_2 = bfloat16BitsToUintEXT(b2);
    u16vec3  u16_3 = bfloat16BitsToUintEXT(bf16vec3(b2, b));
    u16vec4  u16_4 = bfloat16BitsToUintEXT(bf16vec4(b2, b2));

    bfloat16_t b16_1 = intBitsToBFloat16EXT(i16_1);
    bf16vec2   b16_2 = intBitsToBFloat16EXT(i16_2);
    bf16vec3   b16_3 = intBitsToBFloat16EXT(i16_3);
    bf16vec4   b16_4 = intBitsToBFloat16EXT(i16_4);

    b16_1 = uintBitsToBFloat16EXT(u16_1);
    b16_2 = uintBitsToBFloat16EXT(u16_2);
    b16_3 = uintBitsToBFloat16EXT(u16_3);
    b16_4 = uintBitsToBFloat16EXT(u16_4);

}

bfloat16_t func2(bf16vec4 v[2], int i)
{
    bfloat16_t b;
    b = (bf16vec2(v[0]))[i];
    b = (bf16vec2(v[1])).y;

    v[i].z = v[i][i];

    S s;
    s.b = b;

    b = (i != 0) ? b : bfloat16_t(2);

    if (i == 2) {
        b = v[0].x;
    } else {
        b = v[1].y;
    }

    return b;
}