//
// Copyright (C) 2009-2021 Intel Corporation
//
// SPDX-License-Identifier: MIT
//
//

module morton_builder;

kernel_module morton_kernels ("morton/pre_sort.cl")
{
    kernel opencl_build_kernel_init                                     < kernelFunction="init" >;
    kernel opencl_build_morton_kernel_create_morton_codes_indirect      < kernelFunction="create_morton_codes_indirect" >;
    kernel opencl_build_morton_kernel_init_bottom_up_indirect           < kernelFunction="init_bottom_up_indirect" >;
}

kernel_module morton_kernels ("morton/post_sort.cl")
{
    links lsc_intrinsics;

    kernel opencl_build_morton_kernel_build_bottom_up_indirect          < kernelFunction="build_bottom_up_indirect" >;
}

kernel_module morton_kernels ("morton/phase0.cl")
{
    links lsc_intrinsics;

    kernel opencl_build_morton_kernel_parallel_build_phase0             < kernelFunction="parallel_build_phase0" >;
    kernel opencl_build_morton_kernel_parallel_build_phase0_local_sync  < kernelFunction="parallel_build_phase0_local_sync" >;
}

kernel_module morton_kernels ("morton/phase1.cl")
{
    links lsc_intrinsics;

    kernel opencl_build_morton_kernel_parallel_build_phase1_Indirect    < kernelFunction="parallel_build_phase1_Indirect_SG" >;
    kernel opencl_build_morton_kernel_parallel_build_phase1_root        < kernelFunction="parallel_build_phase1_Indirect_global_root" >;
}

kernel_module morton_kernels ("morton/phase2.cl")
{
    links lsc_intrinsics;

    kernel opencl_build_morton_kernel_parallel_build_phase2_refit       < kernelFunction="parallel_build_phase2_refit" >;
    kernel opencl_build_morton_kernel_parallel_build_phase2_refit_local < kernelFunction="parallel_build_phase2_refit_local" >;
}

import struct MKBuilderState "structs.grl";

/*
metakernel begin(
    MKBuilderState state,
    qword morton_code_buffer,
    dword primLeafType,
    dword numHwThreads)
{
    dispatch opencl_build_kernel_init(1, 1, 1) args(
        state.build_globals
        );

    control(wait_idle);


    dispatch opencl_build_morton_kernel_create_morton_codes(numHwThreads, 1, 1) args(
        state.build_globals,
        state.bvh_buffer,
        state.build_primref_buffer,
        morton_code_buffer);

    control(wait_idle);

}

metakernel build_bottom_up(
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer,
    dword numHwThreads)
{
    dispatch opencl_build_morton_kernel_init_bottom_up(numHwThreads, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up);

    control(wait_idle);

    dispatch opencl_build_morton_kernel_build_bottom_up(numHwThreads, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        morton_code_buffer);

    control(wait_idle);

}


metakernel parallel_build(
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer,
    dword numHwThreads)
{
    dispatch opencl_build_morton_kernel_parallel_build_phase0(1, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        state.bvh_buffer);

    control(wait_idle);

    dispatch opencl_build_morton_kernel_parallel_build_phase1(numHwThreads, 1, 1) args(
        state.build_globals,
        morton_code_buffer,
        state.build_primref_buffer,
        buildrecords_bottom_up,
        state.bvh_buffer);

   control(wait_idle);

}

*/

metakernel NewMorton_pre_sort(
    qword num_primrefs_counter,
    MKBuilderState state,
    qword morton_code_buffer,
    qword morton_code_buffer_tmp,
    qword buildrecords_bottom_up,
    dword use_new_morton_sort)
{


    {
        REG1 = 15;
        REG2 = 4;
        REG0 = load_dword( num_primrefs_counter );

        REG0 = REG0 + REG1;     // JDB TODO:  TGL will need to do this computation in the EU and store it in globals
        REG1 = ~REG1;
        REG0 = REG0 & REG1;
        REG0 = REG0 >> REG2;
    }

    dispatch opencl_build_kernel_init(1, 1, 1) args( state.build_globals );

    DISPATCHDIM_X = REG0.lo;
    DISPATCHDIM_Y = 1;
    DISPATCHDIM_Z = 1;

    /*
    // new bottom-up kernel does not need this
    dispatch_indirect opencl_build_morton_kernel_init_bottom_up_indirect args(
        state.build_globals,
        buildrecords_bottom_up);
        */
    dispatch_indirect opencl_build_morton_kernel_create_morton_codes_indirect args(
        state.build_globals,
        state.bvh_buffer,
        state.build_primref_buffer,
        morton_code_buffer,
        morton_code_buffer_tmp,
        use_new_morton_sort);


}



metakernel NewMorton_post_sort(
    qword num_primrefs_counter,
    qword num_buildrecords_counter,
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer )
{

    {
        REG1 = 15;
        REG2 = 4;
        REG0 = load_dword( num_primrefs_counter );

        REG0 = REG0 + REG1;     // JDB TODO:  TGL will need to do this computation in the EU and store it in globals
        REG1 = ~REG1;
        REG0 = REG0 & REG1;
        REG0 = REG0 >> REG2;
    }

    DISPATCHDIM_X = REG0.lo;
    DISPATCHDIM_Y = 1;
    DISPATCHDIM_Z = 1;

    dispatch_indirect opencl_build_morton_kernel_build_bottom_up_indirect args(
        state.build_globals,
        buildrecords_bottom_up,
        morton_code_buffer);


    /*
   dispatch opencl_build_morton_kernel_build_bottom_up(16, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        morton_code_buffer);
        */

    control(wait_idle);

    dispatch opencl_build_morton_kernel_parallel_build_phase0(1, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        state.bvh_buffer);

    control(wait_idle);

    DISPATCHDIM_X = load_dword( num_buildrecords_counter );

    dispatch_indirect opencl_build_morton_kernel_parallel_build_phase1_Indirect args(
        state.build_globals,
        morton_code_buffer,
        state.build_primref_buffer,
        buildrecords_bottom_up,
        state.bvh_buffer);

   control(wait_idle);

}

metakernel NewMorton_bottom_up(
    qword num_primrefs_counter,
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer )
{

    {
        REG1 = 15;
        REG2 = 4;
        REG0 = load_dword( num_primrefs_counter );

        REG0 = REG0 + REG1;     // JDB TODO:  TGL will need to do this computation in the EU and store it in globals
        REG1 = ~REG1;
        REG0 = REG0 & REG1;
        REG0 = REG0 >> REG2;
    }

    DISPATCHDIM_X = REG0.lo;
    DISPATCHDIM_Y = 1;
    DISPATCHDIM_Z = 1;

    dispatch_indirect opencl_build_morton_kernel_build_bottom_up_indirect args(
        state.build_globals,
        buildrecords_bottom_up,
        morton_code_buffer);
}


metakernel NewMorton_phase0(
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_p0_refit_startpoints)
{

    dispatch opencl_build_morton_kernel_parallel_build_phase0(1, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        state.bvh_buffer,
        morton_p0_refit_startpoints);
}

metakernel NewMorton_phase0_local_sync(
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword p0_boxless_nodes)
{

    dispatch opencl_build_morton_kernel_parallel_build_phase0_local_sync(1, 1, 1) args(
        state.build_globals,
        buildrecords_bottom_up,
        state.bvh_buffer,
        p0_boxless_nodes);
}


metakernel NewMorton_phase1(
    qword num_buildrecords_counter,
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer)
{

    DISPATCHDIM_X = load_dword( num_buildrecords_counter );

    dispatch_indirect opencl_build_morton_kernel_parallel_build_phase1_Indirect args(
        state.build_globals,
        morton_code_buffer,
        state.build_primref_buffer,
        buildrecords_bottom_up,
        state.bvh_buffer);
}

metakernel NewMorton_phase1_root(
    qword num_buildrecords_counter,
    MKBuilderState state,
    qword buildrecords_bottom_up,
    qword morton_code_buffer)
{
    dispatch opencl_build_morton_kernel_parallel_build_phase1_root(1, 1, 1) args(
        state.build_globals,
        morton_code_buffer,
        state.build_primref_buffer,
        buildrecords_bottom_up,
        state.bvh_buffer);
}

metakernel NewMorton_phase2(
    qword num_leaves_counter,
    MKBuilderState state,
    qword bottom_node_ids )
{

    DISPATCHDIM_X = load_dword( num_leaves_counter );

    dispatch_indirect opencl_build_morton_kernel_parallel_build_phase2_refit args(
        state.bvh_buffer,
        bottom_node_ids);
}

metakernel NewMorton_phase2_local(
    MKBuilderState state,
    qword p0_boxless_nodes)
{

    dispatch opencl_build_morton_kernel_parallel_build_phase2_refit_local(1, 1, 1) args(
        state.build_globals,
        state.bvh_buffer,
        p0_boxless_nodes);
}