#version 450 core
#extension GL_KHR_memory_scope_semantics : enable
#extension GL_KHR_cooperative_matrix : enable
#extension GL_EXT_shader_explicit_arithmetic_types : enable
#extension GL_EXT_buffer_reference : enable

layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;

const int X = 8;
layout(constant_id = 0) const int Y = 2;
const int Z = X*Y;

coopmat<float16_t, gl_ScopeSubgroup, Z, 8, gl_MatrixUseAccumulator> mC;
coopmat<float16_t, gl_ScopeSubgroup, Z, 8, gl_MatrixUseAccumulator> mC2[3];

layout(constant_id = 1) const float F = 3.0;

const coopmat<float, gl_ScopeSubgroup, Z, 8, gl_MatrixUseAccumulator> mD = coopmat<float, gl_ScopeSubgroup, Z, 8, gl_MatrixUseAccumulator>(0.0);
const coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> mD2 = coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator>(1);

struct S { int a; int b; int c; };

const S s = S(12, 23, 34);

layout(set = 0, binding = 0, buffer_reference) coherent buffer Block {
    float y[1024*1024];
    float x[];
} block;

layout(set = 0, binding = 0) coherent buffer Block16 {
    float16_t y[1024*1024];
    float16_t x[];

    Block b;
} block16;

coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> f16(coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> m) { return -m; }
coopmat<float, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> f32(coopmat<float, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> m) { return -m; }

layout(constant_id = 2) const int SC = 1;
coopmat<float16_t, gl_ScopeSubgroup, SC, SC, gl_MatrixUseAccumulator> scm[SC][SC];

// sized for coopmat<float16_t, gl_ScopeSubgroup, 16, 16, gl_MatrixUseAccumulator>
shared uvec4 shmatrix[16*16*2/16];

void main()
{
    coopmat<float, gl_ScopeSubgroup, 16, (2>1?8:4), gl_MatrixUseAccumulator> m = coopmat<float, gl_ScopeSubgroup, 16, (2>1?8:4), gl_MatrixUseAccumulator>(0.0);

    m = m + m;
    m = m - m;
    m = -m;
    m = 2.0*m;
    m = m*2.0;

    coopmat<float16_t, gl_ScopeSubgroup, 16, 8, gl_MatrixUseAccumulator> m2 = coopmat<float16_t, gl_ScopeSubgroup, 16, 8, gl_MatrixUseAccumulator>(m);

    float x = m[1];
    m[0] = x;

    coopMatLoad(m, block.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(m, block.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatLoad(m2, block16.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(m2, block16.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatLoad(m, block16.b.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(m, block16.b.x, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);

    coopmat<float16_t, gl_ScopeSubgroup, 16, 8, gl_MatrixUseA> A;
    coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseB> B;
    coopmat<float, gl_ScopeSubgroup, 16, 8, gl_MatrixUseAccumulator> C;
    coopmat<float, gl_ScopeSubgroup, 16, 8, gl_MatrixUseAccumulator> D;
    D = coopMatMulAdd(A, B, C);

    int l = D.length();

    coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> E;

    coopmat<float16_t, gl_ScopeSubgroup, Z, Z, gl_MatrixUseAccumulator> F = coopmat<float16_t, gl_ScopeSubgroup, Z, Z, gl_MatrixUseAccumulator>(0.0);

    coopmat<float, gl_ScopeSubgroup, 16, (2>1?8:4), gl_MatrixUseAccumulator> a[5];
    a[3][0] = 1.0;

    float md1 = mD[1];

    md1 += (m += m)[1234];

    mC2[1] = mC2[2];

    coopMatLoad(m, block.y, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(m, block.y, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatLoad(m2, block16.y, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(m2, block16.y, 16, 128, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);

    coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> p1;
    coopmat<float, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> p2;

    p1 = f16(p1);
    p2 = f32(p2);

    p1 = coopmat<float16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator>(0.0);
    p2 = coopmat<float, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator>(0.0);

    p1 /= p1;

    p1 *= float16_t(2.0);
    p2 *= 4.0;

    coopmat<float16_t, gl_ScopeSubgroup, 16, 8, gl_MatrixUseAccumulator> ms;
    coopMatLoad(ms, shmatrix, 1, 2, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
    coopMatStore(ms, shmatrix, 1, 2, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);

    coopmat<int8_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseA> ms8A;
    coopmat<int8_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseB> ms8B;
    coopmat<int8_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseAccumulator> ms8C;
    coopMatMulAdd(ms8A, ms8B, ms8C);
    coopMatMulAdd(ms8A, ms8B, ms8C, 0);
    coopMatMulAdd(ms8A, ms8B, ms8C, gl_MatrixOperandsSaturatingAccumulation);

    coopmat<int16_t, gl_ScopeSubgroup, 8, 8, gl_MatrixUseA> m16;
    coopMatStore(m16, shmatrix, 1, 2, gl_CooperativeMatrixLayoutRowBlockedInterleavedARM);
}