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

module leaf_builder;

kernel_module leaf_kernels ("bvh_build_leaf.cl")
{
    links lsc_intrinsics;

    kernel opencl_kernel_primref_to_quads                   < kernelFunction="primref_to_quads" >;
    kernel opencl_kernel_primref_to_procedurals             < kernelFunction="primref_to_procedurals" >;
    kernel opencl_kernel_create_HW_instance_nodes           < kernelFunction="create_HW_instance_nodes" >;
    kernel opencl_kernel_create_HW_instance_nodes_pointers  < kernelFunction="create_HW_instance_nodes_pointers" >;
}

import struct MKBuilderState "structs.grl";
import struct MKSizeEstimate "structs.grl";

const Instances_GROUPSIZE = 16;

metakernel buildLeafDXR_instances(
            MKBuilderState state,
            qword build_primref_index_buffers,
            qword srcInstanceDescrArray,
            dword stride,
            dword offset,
            dword numPrims)
{
    define num_groups (numPrims+Instances_GROUPSIZE-1)/Instances_GROUPSIZE;
    dispatch opencl_kernel_create_HW_instance_nodes(num_groups,1,1) args(
        state.build_globals,
        build_primref_index_buffers,
        state.build_primref_buffer,
        state.bvh_buffer,
        srcInstanceDescrArray,
        stride,
        offset);
}

metakernel buildLeafDXR_instances_indirect(
            MKBuilderState state,
            qword build_primref_index_buffers,
            qword srcInstanceDescrArray,
            qword indirectBuildRangeInfo,
            dword stride,
            dword offset)
{
    define num_groups  REG0;
    define groupsize_1 REG1; // groupsize - 1
    define C_4         REG2;

    // init with primitiveCount
    num_groups = load_dword(indirectBuildRangeInfo);
    groupsize_1 = 15; // Instances_GROUPSIZE - 1
    C_4 = 4;          // log_2(Instances_GROUPSIZE)

    num_groups = num_groups + groupsize_1;
    num_groups = num_groups >> C_4; // num_groups / Instances_GROUPSIZE;

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

    dispatch_indirect opencl_kernel_create_HW_instance_nodes args(
        state.build_globals,
        build_primref_index_buffers,
        state.build_primref_buffer,
        state.bvh_buffer,
        srcInstanceDescrArray,
        stride,
        offset);
}

metakernel buildLeafDXR_instances_pointers(
            MKBuilderState state,
            qword build_primref_index_buffers,
            qword srcInstanceDescrArrayPtr,
            dword stride,
            dword offset,
            dword numPrims)
{
    define num_groups (numPrims+Instances_GROUPSIZE-1)/Instances_GROUPSIZE;
    dispatch opencl_kernel_create_HW_instance_nodes_pointers(num_groups,1,1) args(
        state.build_globals,
        build_primref_index_buffers,
        state.build_primref_buffer,
        state.bvh_buffer,
        srcInstanceDescrArrayPtr,
        stride,
        offset);
}

metakernel buildLeafDXR_instances_pointers_indirect(
            MKBuilderState state,
            qword build_primref_index_buffers,
            qword srcInstanceDescrArrayPtr,
            qword indirectBuildRangeInfo,
            dword stride,
            dword offset)
{
    define num_groups  REG0;
    define groupsize_1 REG1; // groupsize - 1
    define C_4         REG2;

    // init with primitiveCount
    num_groups = load_dword(indirectBuildRangeInfo);
    groupsize_1 = 15; // Instances_GROUPSIZE - 1
    C_4 = 4;          // log_2(Instances_GROUPSIZE)

    num_groups = num_groups + groupsize_1;
    num_groups = num_groups >> C_4; // num_groups / Instances_GROUPSIZE;

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

    dispatch_indirect opencl_kernel_create_HW_instance_nodes_pointers args(
        state.build_globals,
        build_primref_index_buffers,
        state.build_primref_buffer,
        state.bvh_buffer,
        srcInstanceDescrArrayPtr,
        stride,
        offset);
}

metakernel buildLeafDXR_procedurals(
            MKBuilderState state,
            qword build_primref_index_buffers,
            dword stride,
            dword offset,
            qword p_numPrimitives)
{
    define C_1                  REG0;
    define REG_PRIMS_PER_WG     REG1;
    define REG_PRIMS_PER_WG_SHR REG2;

    C_1 = 1;
    REG_PRIMS_PER_WG = 16;
    REG_PRIMS_PER_WG_SHR = 4;// We cannot use div, so we use shift right instead (shift by 4 = div by 16 elements)

    define reg_numPrimitives  REG3;
    define reg_num_wgs        REG4;

    reg_numPrimitives = load_dword(p_numPrimitives);
    reg_num_wgs = reg_numPrimitives + REG_PRIMS_PER_WG;
    reg_num_wgs = reg_num_wgs - C_1;
    reg_num_wgs = reg_num_wgs >> REG_PRIMS_PER_WG_SHR;

    DISPATCHDIM_X = reg_num_wgs;
    DISPATCHDIM_Y = 1;
    DISPATCHDIM_Z = 1;

    dispatch_indirect opencl_kernel_primref_to_procedurals args(
        state.build_globals,
        state.build_primref_buffer,
        build_primref_index_buffers,
        state.bvh_buffer,
        state.geomDesc_buffer,
        stride,
        offset);
}

metakernel buildLeafDXR_quads(
            MKBuilderState state,
            qword build_primref_index_buffers,
            dword stride,
            dword offset,
            qword p_numPrimitives,
            dword allow_update)
{
    define C_1                  REG0;
    define REG_PRIMS_PER_WG     REG1;
    define SHIFT                REG2;

    C_1 = 1;
    REG_PRIMS_PER_WG = 32;
    SHIFT = 4;// We cannot use div, so we use shift right instead (shift by 4 = div by 16 elements)

    define reg_numPrimitives  REG3;
    define reg_num_wgs        REG4;

    reg_numPrimitives = load_dword(p_numPrimitives);
    reg_num_wgs = reg_numPrimitives + REG_PRIMS_PER_WG;
    reg_num_wgs = reg_num_wgs - C_1;
    reg_num_wgs = reg_num_wgs >> SHIFT;
    reg_num_wgs = reg_num_wgs >> C_1;

    DISPATCHDIM_X = reg_num_wgs;
    DISPATCHDIM_Y = 1;
    DISPATCHDIM_Z = 1;

    dispatch_indirect opencl_kernel_primref_to_quads args(
        state.build_globals,
        state.build_primref_buffer,
        build_primref_index_buffers,
        state.bvh_buffer,
        state.geomDesc_buffer,
        stride,
        offset,
        allow_update);
}