#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_NV_cooperative_matrix2 : enable
#extension GL_EXT_buffer_reference : enable

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

buffer BufType {
   float16_t x[];
} Buf;

layout(buffer_reference, std430, buffer_reference_align = 2) buffer fp16Buf {
    float16_t f;
};


float16_t decode(const in fp16Buf b, const in uint32_t blockCoords[2], const in uint32_t coordInBlock[2])
{
    return b.f;
}

struct S {
   f16vec2 x;
};


layout(std430, binding  = 0) buffer SBuf {
    S s[];
} sbuf;

layout(constant_id = 0) const uint32_t Clamp = gl_CooperativeMatrixClampModeConstantNV;

void main()
{
    coopmat<float16_t, gl_ScopeWorkgroup, 64, 32, gl_MatrixUseA> A;

    tensorLayoutNV<2> t = createTensorLayoutNV(2);
    tensorLayoutNV<3, 1> t2 = createTensorLayoutNV(3, 1);

    t = setTensorLayoutBlockSizeNV(t, 4, 8);
    t = setTensorLayoutDimensionNV(t, 256, 512);
    t = sliceTensorLayoutNV(t, 128, 32, 256, 32);

    tensorViewNV<5> v = createTensorViewNV(5);
    v = setTensorViewDimensionsNV(v, 10, 11, 12, 13, 14);
    v = setTensorViewStrideNV(v, 10, 11, 12, 13, 15);
    v = setTensorViewClipNV(v, 0, 16, 0, 16);

    tensorViewNV<5, true> v2 = createTensorViewNV(5, true);
    tensorViewNV<2, true, 1, 0> v3 = createTensorViewNV(2, true, 1, 0);

    coopMatLoadTensorNV(A, Buf.x, 0, t);
    coopMatStoreTensorNV(A, Buf.x, 0, t);

    coopMatLoadTensorNV(A, Buf.x, 0, t, v);
    coopMatStoreTensorNV(A, Buf.x, 0, t, v);

    coopMatLoadTensorNV(A, Buf.x, 0, t, decode);
    coopMatLoadTensorNV(A, Buf.x, 0, t, v, decode);

    coopMatLoadTensorNV(A, sbuf.s, 1, t, v, decode);

    tensorLayoutNV<2, Clamp> tc = createTensorLayoutNV(2, Clamp);

    coopmat<float16_t, gl_ScopeWorkgroup, 64, 32, gl_MatrixUseA> Arr[2];
    coopMatLoadTensorNV(Arr[1], Buf.x, 0, t);
}