// Copyright (C) 2022 RozK
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program.  If not, see <http://www.gnu.org/licenses/>.

#include "../render.hpp"
#include "render_opengles.hpp"
#include "../display/display_glx.hpp"
#include "../cmp_memcpy.hpp"
#include <cstdlib>
#include <cstdio>
#include <cstring>

typedef void (*rk_DrawElementsInstancedBaseInstanceFunc)(rk_uint, rk_uint, rk_uint, const void *, rk_uint, rk_uint);
typedef void (*rk_MultiDrawElementsIndirectFunc)(rk_uint, rk_uint, const void *, rk_uint, rk_uint);

static rk_DrawElementsInstancedBaseInstanceFunc rk_DrawElementsInstancedBaseInstance = nullptr;
static rk_MultiDrawElementsIndirectFunc rk_MultiDrawElementsIndirect = nullptr;

struct rk_bucket {
    unsigned size;
    unsigned count;
    rk_ushort * indices;
};

static unsigned rk_nbuckets = 0;
static rk_bucket * rk_buckets = nullptr;

static void rk_gl_printf(char const * message) {
    printf("[GL] %s\n", message);
}

static void rk_printf(char const * message) {
    printf("[RK] %s\n", message);
}

static void rk_debug_message_callback(
    GLenum source,
    GLenum type,
    GLuint id,
    GLenum severity,
    GLsizei length,
    GLchar const * message,
    void const * userParam) {
    printf("[GL] (id=%d) %s\n", id, message);
}

void rk_render_initialize(
    rk_bool debug) {
    GLubyte const * const vendor = glGetString(GL_VENDOR);
    GLubyte const * const renderer = glGetString(GL_RENDERER);
    printf("[GL] vendor: %s, renderer: %s\n", vendor, renderer);
    GLubyte const * const version = glGetString(GL_VERSION);
    GLubyte const * const language = glGetString(GL_SHADING_LANGUAGE_VERSION);
    printf("[GL] version: %s, language: %s\n", version, language);

    if (debug) {
        GLint max_texture_layers = 0;
        glGetIntegerv(GL_MAX_ARRAY_TEXTURE_LAYERS, &max_texture_layers);
        printf("[GL] Max texture layers: %d\n", max_texture_layers);
        GLint max_vertex_attribs = 0;
        glGetIntegerv(GL_MAX_VERTEX_ATTRIBS, &max_vertex_attribs);
        printf("[GL] Max vertex attribs: %d\n", max_vertex_attribs);
        GLint max_vertex_bindings = 0;
        glGetIntegerv(GL_MAX_VERTEX_ATTRIB_BINDINGS, &max_vertex_bindings);
        printf("[GL] Max vertex bindings: %d\n", max_vertex_bindings);

        glDebugMessageCallback(rk_debug_message_callback, nullptr);
        glEnable(GL_DEBUG_OUTPUT);
    }

    char const * const gl_exts = reinterpret_cast<char const *>(glGetString(GL_EXTENSIONS));
    // printf("[GL] %s\n", gl_exts);
    rk_DrawElementsInstancedBaseInstance = reinterpret_cast<rk_DrawElementsInstancedBaseInstanceFunc>(
        rk_resolve_extension(gl_exts, "GL_EXT_base_instance", "DrawElementsInstancedBaseInstance"));
    if (rk_DrawElementsInstancedBaseInstance) {
        rk_gl_printf("Using extension GL_EXT_base_instance::DrawElementsInstancedBaseInstance.");
        rk_MultiDrawElementsIndirect = reinterpret_cast<rk_MultiDrawElementsIndirectFunc>(
            rk_resolve_extension(gl_exts, "GL_EXT_multi_draw_indirect", "MultiDrawElementsIndirectEXT"));
        if (rk_MultiDrawElementsIndirect) {
            rk_gl_printf("Using extension GL_EXT_multi_draw_indirect::MultiDrawElementsIndirectEXT.");
        }
    }

    glDisable(GL_BLEND);
    glDisable(GL_DITHER);
    glEnable(GL_DEPTH_TEST);
    glDisable(GL_SCISSOR_TEST);
    glDisable(GL_STENCIL_TEST);
    glEnable(GL_CULL_FACE);
    glFrontFace(GL_CCW);
    glCullFace(GL_BACK);
    glHint(GL_GENERATE_MIPMAP_HINT, GL_NICEST);
}

void rk_render_terminate() {
}

static char rk_infolog[1024];

static void rk_print_shader_infolog(GLuint shader) {
    GLsizei length;
    glGetShaderInfoLog(shader, sizeof(rk_infolog) - 1, &length, rk_infolog);
    if (length > 0) {
        rk_gl_printf(rk_infolog);
    }
}

static void rk_print_program_infolog(GLuint program) {
    GLsizei length;
    glGetProgramInfoLog(program, sizeof(rk_infolog) - 1, &length, rk_infolog);
    if (length > 0) {
        rk_gl_printf(rk_infolog);
    }
}

rk_shader_t rk_create_shader(
    rk_uint const vert_nlines,
    rk_char const * const * vert_lines,
    rk_uint const frag_nlines,
    rk_char const * const * const frag_lines) {
    if (!vert_nlines || !vert_lines || !frag_nlines || !frag_lines) {
        rk_printf("rk_load_shader(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    rk_printf("Compiling vertex shader...");
    while (glGetError() != GL_NO_ERROR);
    GLuint const vert = glCreateShader(GL_VERTEX_SHADER);
    glShaderSource(vert, vert_nlines, vert_lines, nullptr);
    glCompileShader(vert);
    GLint vert_success = 0;
    glGetShaderiv(vert, GL_COMPILE_STATUS, &vert_success);
    if (!vert_success || glGetError() != GL_NO_ERROR) {
        rk_print_shader_infolog(vert);
        glDeleteShader(vert);
        return RK_INVALID_HANDLE;
    }
    rk_printf("Compiling fragment shader...");
    while (glGetError() != GL_NO_ERROR);
    GLuint const frag = glCreateShader(GL_FRAGMENT_SHADER);
    glShaderSource(frag, frag_nlines, frag_lines, nullptr);
    glCompileShader(frag);
    GLint frag_success = 0;
    glGetShaderiv(frag, GL_COMPILE_STATUS, &frag_success);
    if (!frag_success || glGetError() != GL_NO_ERROR) {
        rk_print_shader_infolog(frag);
        glDeleteShader(vert);
        glDeleteShader(frag);
        return RK_INVALID_HANDLE;
    }
    rk_printf("Linking program...");
    while (glGetError() != GL_NO_ERROR);
    GLuint prog = glCreateProgram();
    glAttachShader(prog, vert);
    glAttachShader(prog, frag);
    glLinkProgram(prog);
    GLint prog_success = 0;
    glGetProgramiv(prog, GL_LINK_STATUS, &prog_success);
    if (!prog_success || glGetError() != GL_NO_ERROR) {
        rk_print_program_infolog(prog);
        glDeleteShader(vert);
        glDeleteShader(frag);
        glDeleteProgram(prog);
        return RK_INVALID_HANDLE;
    }
    rk_printf("Done.");
    glReleaseShaderCompiler();
    rk_shader * const shader = new rk_shader;
    shader->vertex = vert;
    shader->fragment = frag;
    shader->program = prog;
    return reinterpret_cast<rk_shader_t>(shader);
}

rk_input_t rk_resolve_input(
    rk_shader_t _shader,
    rk_char const * name) {
    rk_shader const * const shader = reinterpret_cast<rk_shader const *>(_shader);
    if (!shader || !name) {
        rk_printf("rk_resolve_input(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    GLint const uniform = glGetUniformLocation(shader->program, name);
    if (uniform < 0) {
        printf("[RK] rk_resolve_input(): uniform %s not found.\n", name);
    }
    return reinterpret_cast<rk_input_t>(uniform + 1);
}

rk_param_t rk_resolve_param(
    rk_shader_t _shader,
    rk_char const * name) {
    rk_shader const * const shader = reinterpret_cast<rk_shader const *>(_shader);
    if (!shader || !name) {
        rk_printf("rk_resolve_param(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    GLint const location = glGetAttribLocation(shader->program, name);
    if (location < 0) {
        printf("[RK] rk_resolve_param(): attrib %s not found.\n", name);
    }
    return reinterpret_cast<rk_param_t>(location + 1);
}

rk_texture_t rk_create_texture(
    rk_texture_format format,
    rk_uint width,
    rk_uint height,
    rk_uint nlevels,
    rk_texture_flags flags,
    rk_ubyte const * pixels) {
    if (!width || !height || !pixels) {
        rk_printf("rk_create_texture(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    GLint internal_format;
    GLenum source_format;
    GLenum source_type;
    switch (format) {
        case RK_TEXTURE_FORMAT_SRGB8_A8:
            internal_format = GL_SRGB8_ALPHA8;
            source_format = GL_RGBA;
            source_type = GL_UNSIGNED_BYTE;
            break;
        case RK_TEXTURE_FORMAT_RGBA8:
            internal_format = GL_RGBA8;
            source_format = GL_RGBA;
            source_type = GL_UNSIGNED_BYTE;
            break;
        case RK_TEXTURE_FORMAT_RGB10_A2:
            internal_format = GL_RGB10_A2;
            source_format = GL_RGBA;
            source_type = GL_UNSIGNED_INT_2_10_10_10_REV;
            break;
        case RK_TEXTURE_FORMAT_FLOAT_32:
            internal_format = GL_R32F;
            source_format = GL_RED;
            source_type =  GL_FLOAT;
            break;
        default:
            rk_printf("rk_create_texture(): invalid texture format.");
            return RK_INVALID_HANDLE;
            break;
    }
    rk_texture * const texture = new rk_texture;
    glGenTextures(1, &texture->texture);
    GLenum target;
    if (nlevels) {
        if (flags & RK_TEXTURE_FLAG_3D) {
            target = GL_TEXTURE_3D;
        } else {
            target = GL_TEXTURE_2D_ARRAY;
        }
        glBindTexture(target, texture->texture);
        //TODO: glTexStorage3D
        glTexImage3D(target, 0, internal_format, width, height, nlevels, 0, source_format, source_type, pixels);
    } else {
        target = GL_TEXTURE_2D;
        glBindTexture(target, texture->texture);
        //TODO: glTexStorage2D
        glTexImage2D(target, 0, internal_format, width, height, 0, source_format, source_type, pixels);
    }
    if (flags & RK_TEXTURE_FLAG_MIPMAPS) {
        if (flags & RK_TEXTURE_FLAG_MIN_LINEAR) {
            glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_LINEAR);
        } else {
            glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_NEAREST_MIPMAP_LINEAR);
        }
    } else {
        if (flags & RK_TEXTURE_FLAG_MIN_LINEAR) {
            glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
        } else {
            glTexParameteri(target, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
        }
    }
    if (flags & RK_TEXTURE_FLAG_MAG_LINEAR) {
        glTexParameteri(target, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
    } else {
        glTexParameteri(target, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
    }
    glTexParameteri(target, GL_TEXTURE_WRAP_S, GL_REPEAT);
    glTexParameteri(target, GL_TEXTURE_WRAP_T, GL_REPEAT);
    glTexParameteri(target, GL_TEXTURE_WRAP_R, GL_REPEAT);
    if (flags & RK_TEXTURE_FLAG_MIPMAPS) {
        glGenerateMipmap(target);
    }
    texture->nlevels = nlevels;
    glBindTexture(target, 0);
    return reinterpret_cast<rk_texture_t>(texture);
}

rk_triangles_t rk_create_triangles(
    rk_uint nvertices,
    rk_vec3 const * vertices) {
    if (!nvertices || !vertices) {
        rk_printf("rk_create_triangles(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    rk_triangles * const triangles = new rk_triangles;
    triangles->size = nvertices;
    glGenVertexArrays(1, &triangles->array);
    glBindVertexArray(triangles->array);
    glGenBuffers(1, &triangles->vertices);
    glBindBuffer(GL_ARRAY_BUFFER, triangles->vertices);
    glBufferData(GL_ARRAY_BUFFER, nvertices * sizeof(rk_vec3), vertices, GL_STATIC_DRAW);
    glEnableVertexAttribArray(0);
    glVertexAttribPointer(0, 3, GL_FLOAT, GL_FALSE, 0, 0);
    glBindBuffer(GL_ARRAY_BUFFER, 0);
    glBindVertexArray(0);
    return reinterpret_cast<rk_triangles_t>(triangles);
}

rk_vertices_t rk_create_vertices(
    rk_vertex_format const * format,
    rk_uint nvertices,
    rk_ubyte const * _vertices,
    rk_uint nindices,
    rk_ushort const * indices,
    rk_uint nmeshes,
    rk_mesh const * meshes) {
    if (!format || !nvertices || !_vertices || !nindices || !indices) {
        rk_printf("rk_create_vertices(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    unsigned format_size = 0;
    unsigned vertex_size = 0;
    for (rk_vertex_format const * f = format; *f; ++f, ++format_size) {
        switch (*f & RK_VERTEX_FORMAT_MASK) {
            case RK_VERTEX_FORMAT_VEC3_FLOAT:
                vertex_size += sizeof(rk_vec3_float);
                break;
            case RK_VERTEX_FORMAT_VEC3_INT10:
                vertex_size += sizeof(rk_vec3_int10);
                break;
            case RK_VERTEX_FORMAT_VEC3_UINT10:
                vertex_size += sizeof(rk_vec3_uint10);
                break;
            default:
                rk_printf("rk_create_vertices(): invalid vertex format.");
                return RK_INVALID_HANDLE;
                break;
        }
    }
    if (!format_size) {
        rk_printf("rk_create_vertices(): empty vertex format.");
        return RK_INVALID_HANDLE;
    }
    rk_vertices * const vertices = new rk_vertices;
    vertices->nvertices = nvertices;
    vertices->nindices = nindices;
    vertices->nmeshes = nmeshes;
    vertices->format = new rk_vertex_format[format_size + 1];
    memcpy(vertices->format, format, (format_size + 1) * sizeof(rk_vertex_format));
    vertices->vertices = new rk_ubyte[nvertices * vertex_size];
    memcpy(vertices->vertices, _vertices, nvertices * vertex_size);
    vertices->indices = new rk_ushort[nindices];
    memcpy(vertices->indices, indices, nindices * sizeof(rk_ushort));
    vertices->meshes = new rk_mesh[nmeshes];
    memcpy(vertices->meshes, meshes, nmeshes * sizeof(rk_mesh));
    vertices->vertices_buffer = 0;
    vertices->indices_buffer = 0;
    return reinterpret_cast<rk_vertices_t>(vertices);
}

static void rk_buckets_alloc(
    rk_batch const & batch) {
    unsigned const count = batch.vertices->nmeshes;
    unsigned const size = batch.max_size;
    bool reallocated = false;
    if (!rk_nbuckets) {
        rk_nbuckets = count;
        rk_buckets = reinterpret_cast<rk_bucket *>(malloc(count * sizeof(rk_bucket)));
        for (unsigned index = 0; index < count; ++index) {
            rk_bucket & bucket = rk_buckets[index];
            bucket.size = size;
            bucket.indices = reinterpret_cast<rk_ushort *>(malloc(size * sizeof(rk_ushort)));
        }
        reallocated = true;
    }
    else if (count <= rk_nbuckets) {
        for (unsigned index = 0; index < count; ++index) {
            rk_bucket & bucket = rk_buckets[index];
            if (bucket.size < size) {
                bucket.size = size;
                bucket.indices = reinterpret_cast<rk_ushort *>(realloc(bucket.indices, size * sizeof(rk_ushort)));
                reallocated = true;
            }
        }
    }
    else {
        rk_buckets = reinterpret_cast<rk_bucket *>(realloc(rk_buckets, count * sizeof(rk_bucket)));
        for (unsigned index = 0; index < rk_nbuckets; ++index) {
            rk_bucket & bucket = rk_buckets[index];
            if (bucket.size < size) {
                bucket.size = size;
                bucket.indices = reinterpret_cast<rk_ushort *>(realloc(bucket.indices, size * sizeof(rk_ushort)));
            }
        }
        for (unsigned index = rk_nbuckets; index < count; ++index) {
            rk_bucket & bucket = rk_buckets[index];
            bucket.size = size;
            bucket.indices = reinterpret_cast<rk_ushort *>(malloc(size * sizeof(rk_ushort)));
        }
        rk_nbuckets = count;
        reallocated = true;
    }
    if (reallocated) {
        unsigned total_size = rk_nbuckets * sizeof(rk_bucket);
        for (unsigned index = 0; index < rk_nbuckets; ++index) {
            rk_bucket const & bucket = rk_buckets[index];
            total_size += bucket.size * sizeof(rk_ushort);
        }
        printf("[RK] rk_buckets_alloc() -> %d KiB\n", total_size / 1024);
    }
}

static void rk_pack_vec3_float(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_vec3_float * __restrict dst = reinterpret_cast<rk_vec3_float *>(_dst);
    rk_vec3_float const * const __restrict src = reinterpret_cast<rk_vec3_float const *>(_src);
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        *dst = src[*index];
    }
}

static void rk_pack_vec3_short(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_vec3_short * __restrict dst = reinterpret_cast<rk_vec3_short *>(_dst);
    rk_vec3_float const * const __restrict src = reinterpret_cast<rk_vec3_float const *>(_src);
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_vec3_float const & input = src[*index];
        dst->x = static_cast<rk_short>(input.x);
        dst->y = static_cast<rk_short>(input.y);
        dst->z = static_cast<rk_short>(input.z);
        dst->pad = 0;
    }
}

static void rk_pack_vec3_short_norm(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_vec3_short * __restrict dst = reinterpret_cast<rk_vec3_short *>(_dst);
    rk_vec3_float const * const __restrict src = reinterpret_cast<rk_vec3_float const *>(_src);
    #define _convert(s) (static_cast<rk_short>((s) * ((s) < 0.f ? 32768.f : 32767.f)))
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_vec3_float const & input = src[*index];
        dst->x = _convert(input.x);
        dst->y = _convert(input.y);
        dst->z = _convert(input.z);
        dst->pad = 0;
    }
    #undef _convert
}

static void rk_pack_vec3_int10(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_vec3_int10 * __restrict dst = reinterpret_cast<rk_vec3_int10 *>(_dst);
    rk_vec3_float const * const __restrict src = reinterpret_cast<rk_vec3_float const *>(_src);
    #define _convert(s) (static_cast<rk_int>((s)) & 1023)
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_vec3_float const & input = src[*index];
        *dst = _convert(input.x) | (_convert(input.y) << 10) | (_convert(input.z) << 20);
    }
    #undef _convert
}

static void rk_pack_vec3_int10_norm(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_vec3_int10 * __restrict dst = reinterpret_cast<rk_vec3_int10 *>(_dst);
    rk_vec3_float const * const __restrict src = reinterpret_cast<rk_vec3_float const *>(_src);
    #define _convert(s) (static_cast<rk_int>((s) * ((s) < 0.f ? 512.f : 511.f)) & 1023)
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_vec3_float const & input = src[*index];
        *dst = _convert(input.x) | (_convert(input.y) << 10) | (_convert(input.z) << 20);
    }
    #undef _convert
}

static void rk_pack_mat3_float(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_mat3_float * __restrict dst = reinterpret_cast<rk_mat3_float *>(_dst);
    rk_mat3_float const * const __restrict src = reinterpret_cast<rk_mat3_float const *>(_src);
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        *dst = src[*index];
    }
    #undef _convert
}

static void rk_pack_mat3_int10(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_mat3_int10 * __restrict dst = reinterpret_cast<rk_mat3_int10 *>(_dst);
    rk_mat3_float const * const __restrict src = reinterpret_cast<rk_mat3_float const *>(_src);
    #define _convert(s) (static_cast<rk_int>((s)) & 1023)
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_mat3_float const & input = src[*index];
        dst->x = _convert(input.x.x) | (_convert(input.x.y) << 10) | (_convert(input.x.z) << 20);
        dst->y = _convert(input.y.x) | (_convert(input.y.y) << 10) | (_convert(input.y.z) << 20);
        dst->z = _convert(input.z.x) | (_convert(input.z.y) << 10) | (_convert(input.z.z) << 20);
    }
    #undef _convert
}

static void rk_pack_mat3_int10_norm(
    unsigned const count,
    rk_ushort const * const __restrict indices,
    rk_ubyte * __restrict _dst,
    rk_ubyte const * const __restrict _src) {
    rk_ushort const * const last_index = indices + count;
    rk_mat3_int10 * __restrict dst = reinterpret_cast<rk_mat3_int10 *>(_dst);
    rk_mat3_float const * const __restrict src = reinterpret_cast<rk_mat3_float const *>(_src);
    #define _convert(s) (static_cast<rk_int>((s) * ((s) < 0.f ? 512.f : 511.f)) & 1023)
    for (rk_ushort const * __restrict index = indices; index < last_index; ++index, ++dst) {
        rk_mat3_float const & input = src[*index];
        dst->x = _convert(input.x.x) | (_convert(input.x.y) << 10) | (_convert(input.x.z) << 20);
        dst->y = _convert(input.y.x) | (_convert(input.y.y) << 10) | (_convert(input.y.z) << 20);
        dst->z = _convert(input.z.x) | (_convert(input.z.y) << 10) | (_convert(input.z.z) << 20);
    }
    #undef _convert
}

rk_batch_t rk_create_batch(
    rk_vertices_t _vertices,
    rk_uint max_size,
    rk_param_format const * params_format) {
    rk_vertices * const vertices = reinterpret_cast<rk_vertices *>(_vertices);
    if (!vertices || !max_size || max_size > RK_BATCH_MAX_SIZE) {
        rk_printf("rk_create_batch(): invalid params.");
        return RK_INVALID_HANDLE;
    }
    unsigned vertex_size = 0;
    for (rk_vertex_format const * f = vertices->format; *f; ++f) {
        switch (*f & RK_VERTEX_FORMAT_MASK) {
            case RK_VERTEX_FORMAT_VEC3_FLOAT:
                vertex_size += sizeof(rk_vec3_float);
                break;
            case RK_VERTEX_FORMAT_VEC3_INT10:
                vertex_size += sizeof(rk_vec3_int10);
                break;
            case RK_VERTEX_FORMAT_VEC3_UINT10:
                vertex_size += sizeof(rk_vec3_uint10);
                break;
        }
    }
    unsigned nparams = 0;
    unsigned params_size = 0;
    if (params_format) {
        for (rk_param_format const * f = params_format; *f; ++f, ++nparams) {
            switch (*f & RK_PARAM_FORMAT_MASK) {
                case RK_PARAM_FORMAT_VEC3_FLOAT:
                    params_size += sizeof(rk_vec3_float);
                    break;
                case RK_PARAM_FORMAT_VEC3_SHORT:
                    params_size += sizeof(rk_vec3_short);
                    break;
                case RK_PARAM_FORMAT_VEC3_INT10:
                    params_size += sizeof(rk_vec3_int10);
                    break;
                case RK_PARAM_FORMAT_MAT3_FLOAT:
                    params_size += sizeof(rk_mat3_float);
                    break;
                case RK_PARAM_FORMAT_MAT3_INT10:
                    params_size += sizeof(rk_mat3_int10);
                    break;
                default:
                    rk_printf("rk_create_batch(): invalid param format.");
                    return RK_INVALID_HANDLE;
                    break;
            }
        }
    }
    rk_batch * batch = new rk_batch;
    batch->state = RK_BATCH_STATE_EMPTY;
    batch->count = 0;
    batch->ncommands = 0;
    batch->ninstances = 0;
    batch->max_size = max_size;
    batch->nparams = nparams;
    batch->vertices = vertices;
    batch->flags = new rk_instance_flags[max_size];
    batch->meshes = new rk_ushort[max_size];
    batch->indices = new rk_ushort[max_size];
    memset(batch->indices, 0xFF, max_size * sizeof(rk_ushort));
    batch->commands = new rk_command[vertices->nmeshes];
    memset(batch->commands, 0, vertices->nmeshes * sizeof(rk_command));
    if (nparams) {
        batch->params = new rk_parameter[nparams];
    } else {
        batch->params = nullptr;
    }
    glGenVertexArrays(1, &batch->vertex_array);
    glBindVertexArray(batch->vertex_array);
    if (!vertices->vertices_buffer) {
        glGenBuffers(1, &vertices->vertices_buffer);
        glBindBuffer(GL_ARRAY_BUFFER, vertices->vertices_buffer);
        glBufferData(GL_ARRAY_BUFFER, vertices->nvertices * vertex_size, vertices->vertices, GL_STATIC_DRAW);
        glBindBuffer(GL_ARRAY_BUFFER, 0);
    }
    if (vertices->indices_buffer) {
        glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, vertices->indices_buffer);
    } else {
        glGenBuffers(1, &vertices->indices_buffer);
        glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, vertices->indices_buffer);
        glBufferData(GL_ELEMENT_ARRAY_BUFFER, vertices->nindices * sizeof(rk_ushort), vertices->indices, GL_STATIC_DRAW);
    }
    if (rk_MultiDrawElementsIndirect) {
        glGenBuffers(1, &batch->commands_buffer);
        glBindBuffer(GL_DRAW_INDIRECT_BUFFER, batch->commands_buffer);
        glBufferData(GL_DRAW_INDIRECT_BUFFER, vertices->nmeshes * sizeof(rk_command), nullptr, GL_DYNAMIC_DRAW);
        glBindBuffer(GL_DRAW_INDIRECT_BUFFER, 0);
    }
    if (nparams) {
        glGenBuffers(1, &batch->params_buffer);
        glBindBuffer(GL_ARRAY_BUFFER, batch->params_buffer);
        glBufferData(GL_ARRAY_BUFFER, max_size * params_size, nullptr, GL_DYNAMIC_DRAW);
        glBindBuffer(GL_ARRAY_BUFFER, 0);
    }
    unsigned binding = 0;
    unsigned attrib = 0;
    unsigned offset = 0;
    glBindVertexBuffer(binding, vertices->vertices_buffer, 0, vertex_size);
    for (rk_vertex_format const * f = vertices->format; *f; ++f) {
        GLboolean const norm = (*f & RK_VERTEX_FORMAT_NORMALIZE) != 0;
        switch (*f & RK_VERTEX_FORMAT_MASK) {
            case RK_VERTEX_FORMAT_VEC3_FLOAT:
                glEnableVertexAttribArray(attrib);
                glVertexAttribFormat(attrib, 3, GL_FLOAT, GL_FALSE, offset);
                glVertexAttribBinding(attrib++, binding);
                offset += sizeof(rk_vec3_float);
                break;
            case RK_VERTEX_FORMAT_VEC3_INT10:
                glEnableVertexAttribArray(attrib);
                glVertexAttribFormat(attrib, 4, GL_INT_2_10_10_10_REV, norm, offset);
                glVertexAttribBinding(attrib++, binding);
                offset += sizeof(rk_vec3_int10);
                break;
            case RK_VERTEX_FORMAT_VEC3_UINT10:
                glEnableVertexAttribArray(attrib);
                glVertexAttribFormat(attrib, 4, GL_UNSIGNED_INT_2_10_10_10_REV, norm, offset);
                glVertexAttribBinding(attrib++, binding);
                offset += sizeof(rk_vec3_uint10);
                break;
        }
    }
    binding += 1;
    if (nparams) {
        offset = 0;
        rk_parameter * param = batch->params;
        for (rk_param_format const * f = params_format; *f; ++f, ++param, ++binding) {
            GLboolean const norm = (*f & RK_PARAM_FORMAT_NORMALIZE) != 0;
            param->dirty = false;
            param->binding = binding;
            param->offset = offset;
            switch (*f & RK_PARAM_FORMAT_MASK) {
                case RK_PARAM_FORMAT_VEC3_FLOAT:
                    param->src_size = sizeof(rk_vec3);
                    param->dst_size = sizeof(rk_vec3_float);
                    param->packer = rk_pack_vec3_float;
                    glBindVertexBuffer(binding, batch->params_buffer, param->offset, param->dst_size);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 3, GL_FLOAT, GL_FALSE, 0);
                    glVertexAttribBinding(attrib++, binding);
                    break;
                case RK_PARAM_FORMAT_VEC3_SHORT:
                    param->src_size = sizeof(rk_vec3);
                    param->dst_size = sizeof(rk_vec3_short);
                    param->packer = norm ? rk_pack_vec3_short_norm : rk_pack_vec3_short;
                    glBindVertexBuffer(binding, batch->params_buffer, param->offset, param->dst_size);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 3, GL_SHORT, norm, 0);
                    glVertexAttribBinding(attrib++, binding);
                    break;
                case RK_PARAM_FORMAT_VEC3_INT10:
                    param->src_size = sizeof(rk_vec3);
                    param->dst_size = sizeof(rk_vec3_int10);
                    param->packer = norm ? rk_pack_vec3_int10_norm : rk_pack_vec3_int10;
                    glBindVertexBuffer(binding, batch->params_buffer, param->offset, param->dst_size);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 4, GL_INT_2_10_10_10_REV, norm, 0);
                    glVertexAttribBinding(attrib++, binding);
                    break;
                case RK_PARAM_FORMAT_MAT3_FLOAT:
                    param->src_size = sizeof(rk_mat3);
                    param->dst_size = sizeof(rk_mat3_float);
                    param->packer = rk_pack_mat3_float;
                    glBindVertexBuffer(binding, batch->params_buffer, param->offset, param->dst_size);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 3, GL_FLOAT, GL_FALSE, offsetof(rk_mat3_float, x));
                    glVertexAttribBinding(attrib++, binding);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 3, GL_FLOAT, GL_FALSE, offsetof(rk_mat3_float, y));
                    glVertexAttribBinding(attrib++, binding);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 3, GL_FLOAT, GL_FALSE, offsetof(rk_mat3_float, z));
                    glVertexAttribBinding(attrib++, binding);
                    break;
                case RK_PARAM_FORMAT_MAT3_INT10:
                    param->src_size = sizeof(rk_mat3);
                    param->dst_size = sizeof(rk_mat3_int10);
                    param->packer = norm ? rk_pack_mat3_int10_norm : rk_pack_mat3_int10;
                    glBindVertexBuffer(binding, batch->params_buffer, param->offset, param->dst_size);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 4, GL_INT_2_10_10_10_REV, norm, offsetof(rk_mat3_int10, x));
                    glVertexAttribBinding(attrib++, binding);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 4, GL_INT_2_10_10_10_REV, norm, offsetof(rk_mat3_int10, y));
                    glVertexAttribBinding(attrib++, binding);
                    glEnableVertexAttribArray(attrib);
                    glVertexAttribFormat(attrib, 4, GL_INT_2_10_10_10_REV, norm, offsetof(rk_mat3_int10, z));
                    glVertexAttribBinding(attrib++, binding);
                    break;
            }
            glVertexBindingDivisor(binding, 1);
            param->source = new rk_ubyte[max_size * param->src_size];
            offset += max_size * param->dst_size;
        }
    }
    glBindVertexArray(0);
    rk_buckets_alloc(*batch);
    return reinterpret_cast<rk_batch_t>(batch);
}

static void rk_sort_batch(
    rk_batch const & batch) {
    rk_bucket const * const last_bucket = rk_buckets + batch.vertices->nmeshes;
    for (rk_bucket * __restrict bucket = rk_buckets; bucket < last_bucket; ++bucket) {
        bucket->count = 0;
    }
    rk_instance_flags const * __restrict flags = batch.flags;
    rk_ushort const * __restrict mesh_index = batch.meshes;
    for (unsigned index = 0; index < batch.count; ++index, ++flags, ++mesh_index) {
        if ((*flags & RK_INSTANCE_FLAGS_SPAWNED_VISIBLE) == RK_INSTANCE_FLAGS_SPAWNED_VISIBLE) {
            rk_bucket & __restrict bucket = rk_buckets[*mesh_index];
            bucket.indices[bucket.count++] = index;
        }
    }
    bool modified = false;
    rk_ushort * __restrict indices = batch.indices;
    rk_command * __restrict command = batch.commands;
    rk_mesh const * __restrict mesh = batch.vertices->meshes;
    for (rk_bucket const * __restrict bucket = rk_buckets; bucket < last_bucket; ++bucket, ++mesh) {
        if (bucket->count) {
            command->nvertices = static_cast<GLuint>(mesh->ntriangles) * 3;
            command->ninstances = bucket->count;
            command->base_index = mesh->base_index;
            command->base_instance = indices - batch.indices;
            modified |= rk_cmp_memcpy<sizeof(rk_ushort)>(indices, bucket->indices, bucket->count * sizeof(rk_ushort));
            indices += bucket->count;
            ++command;
        }
    }
    unsigned const ninstances = indices - batch.indices;
    modified |= (ninstances != batch.ninstances);
    batch.ninstances = ninstances;
    batch.ncommands = command - batch.commands;
    if (modified && rk_MultiDrawElementsIndirect) {
        glBufferSubData(GL_DRAW_INDIRECT_BUFFER, 0, batch.ncommands * sizeof(rk_command), batch.commands);
    }
    batch.state = RK_BATCH_STATE_SORTED;
}

static void rk_pack_batch(
    rk_batch const & batch) {
    if (batch.nparams) {
        glBindBuffer(GL_ARRAY_BUFFER, batch.params_buffer);
        for (rk_parameter const * param = batch.params; param < batch.params + batch.nparams; ++param) {
            if (param->dirty) {
                param->dirty = false;
                if (batch.ninstances) {
                    rk_ubyte * const dst = reinterpret_cast<rk_ubyte *>(
                        glMapBufferRange(GL_ARRAY_BUFFER, param->offset, batch.ninstances * param->dst_size,
                            GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_RANGE_BIT | GL_MAP_UNSYNCHRONIZED_BIT));
                    if (dst) {
                        param->packer(batch.ninstances, batch.indices, dst, param->source);
                        glUnmapBuffer(GL_ARRAY_BUFFER);
                    }
                }
            }
        }
        glBindBuffer(GL_ARRAY_BUFFER, 0);
    }
    batch.state = RK_BATCH_STATE_PACKED;
}

void rk_fill_batch(
    rk_batch_t _batch,
    rk_uint count,
    rk_instance_flags const * flags,
    rk_ushort const * meshes,
    rk_ubyte const * const * params) {
    rk_batch const * const batch = reinterpret_cast<rk_batch const *>(_batch);
    if (!batch || !count || count > batch->max_size) {
        rk_printf("rk_fill_batch(): invalid params.");
        return;
    }
    bool got_any_params = false;
    bool got_all_params = !batch->nparams;
    if (batch->nparams) {
        got_all_params = (params != nullptr);
        if (params) {
            for (rk_ubyte const * const * param = params; param < params + batch->nparams; ++param) {
                bool const got_param = (*param != nullptr);
                got_any_params |= got_param;
                got_all_params &= got_param;
            }
        }
    }
    bool const is_empty = (batch->state < RK_BATCH_STATE_FILLED);
    bool const resized = (count != batch->count);
    bool const got_everything = (flags && meshes && got_all_params);
    if (is_empty && !got_everything) {
        rk_printf("rk_fill_batch(): cannot freeze and empty batch.");
        return;
    } else if (count > batch->count && !got_everything) {
        rk_printf("rk_fill_batch(): cannot grow a frozen batch.");
        return;
    }
    batch->count = count;
    bool const cmp_flags = (flags &&
        rk_cmp_memcpy<sizeof(rk_ubyte)>(batch->flags, flags, batch->count * sizeof(rk_instance_flags)));
    bool const cmp_meshes = (meshes &&
        rk_cmp_memcpy<sizeof(rk_ushort)>(batch->meshes, meshes, batch->count * sizeof(rk_mesh)));
    bool const need_sorting = (cmp_flags || cmp_meshes || resized);
    if (batch->nparams) {
        rk_parameter const * const last_param = batch->params + batch->nparams;
        if (got_any_params) {
            rk_ubyte const * const * src = params;
            for (rk_parameter const * dst = batch->params; dst < last_param; ++dst, ++src) {
                dst->dirty = ((*src &&
                    rk_cmp_memcpy<sizeof(rk_uint)>(dst->source, *src, batch->count * dst->src_size))
                    || need_sorting);
            }
        } else if (need_sorting) {
            for (rk_parameter const * dst = batch->params; dst < last_param; ++dst) {
                dst->dirty = true;
            }
        }
    }
    if (is_empty) {
        glBindVertexArray(batch->vertex_array);
        if (rk_MultiDrawElementsIndirect) {
            glBindBuffer(GL_DRAW_INDIRECT_BUFFER, batch->commands_buffer);
        }
        rk_sort_batch(*batch);
        rk_pack_batch(*batch);
        if (rk_MultiDrawElementsIndirect) {
            glBindBuffer(GL_DRAW_INDIRECT_BUFFER, 0);
        }
        glBindVertexArray(0);
    } else if (need_sorting) {
        batch->state = RK_BATCH_STATE_FILLED;
    } else {
        batch->state = RK_BATCH_STATE_SORTED;
    }
}

void rk_clear_buffer(
    rk_bool pixels,
    rk_bool depth,
    rk_bool stencil) {
    glClear((GL_COLOR_BUFFER_BIT * pixels) | (GL_DEPTH_BUFFER_BIT * depth) | (GL_STENCIL_BUFFER_BIT * stencil));
}

void rk_select_shader(
    rk_shader_t _shader) {
    rk_shader * const shader = reinterpret_cast<rk_shader *>(_shader);
    if (shader) {
        glUseProgram(shader->program);
    }
}

void rk_set_input_float(
    rk_input_t _input,
    float value) {
    GLint const input = reinterpret_cast<intptr_t>(_input) - 1;
    if (input > -1) {
        glUniform1f(input, value);
    }
}

void rk_set_input_vec3(
    rk_input_t _input,
    rk_vec3 const & value) {
    GLint const input = reinterpret_cast<intptr_t>(_input) - 1;
    if (input > -1) {
        glUniform3fv(input, 1, glm::value_ptr(value));
    }
}

void rk_set_input_mat3(
    rk_input_t _input,
    rk_mat3 const & value) {
    GLint const input = reinterpret_cast<intptr_t>(_input) - 1;
    if (input > -1) {
        glUniformMatrix3fv(input, 1, GL_FALSE, glm::value_ptr(value));
    }
}

void rk_set_input_mat4(
    rk_input_t _input,
    rk_mat4 const & value) {
    GLint const input = reinterpret_cast<intptr_t>(_input) - 1;
    if (input > -1) {
        glUniformMatrix4fv(input, 1, GL_FALSE, glm::value_ptr(value));
    }
}

void rk_set_param_vec3(
    rk_param_t _param,
    rk_vec3 const & value) {
    GLint const param = reinterpret_cast<intptr_t>(_param) - 1;
    if (param > -1) {
        glVertexAttrib3fv(param, glm::value_ptr(value));
    }
}

void rk_set_param_mat3(
    rk_param_t _param,
    rk_mat3 const & value) {
    GLint const param = reinterpret_cast<intptr_t>(_param) - 1;
    if (param > -1) {
        glVertexAttrib3fv(param + 0, glm::value_ptr(value[0]));
        glVertexAttrib3fv(param + 1, glm::value_ptr(value[1]));
        glVertexAttrib3fv(param + 2, glm::value_ptr(value[2]));
    }
}

void rk_select_texture(
    rk_uint slot,
    rk_texture_t _texture) {
    rk_texture const * const texture = reinterpret_cast<rk_texture const *>(_texture);
    if (texture) {
        glActiveTexture(GL_TEXTURE0 + slot);
        if (texture->nlevels) {
            glBindTexture(GL_TEXTURE_2D_ARRAY, texture->texture);
        } else {
            glBindTexture(GL_TEXTURE_2D, texture->texture);
        }
    }
}

RK_EXPORT void rk_draw_triangles(
    rk_triangles_t _triangles) {
    rk_triangles const * const triangles = reinterpret_cast<rk_triangles const *>(_triangles);
    if (triangles) {
        glBindVertexArray(triangles->array);
        glDrawArrays(GL_TRIANGLES, 0, triangles->size);
        glBindVertexArray(0);
    }
}

void rk_draw_batch(
    rk_batch_t _batch) {
    rk_batch const * const batch = reinterpret_cast<rk_batch const *>(_batch);
    if (!batch) {
        rk_printf("rk_draw_batch(): invalid params.");
        return;
    }
    if (batch->state < RK_BATCH_STATE_FILLED) {
        rk_printf("rk_draw_batch(): invalid state.");
        return;
    }
    glBindVertexArray(batch->vertex_array);
    if (rk_MultiDrawElementsIndirect) {
        glBindBuffer(GL_DRAW_INDIRECT_BUFFER, batch->commands_buffer);
    }
    if (batch->state < RK_BATCH_STATE_SORTED) {
        rk_sort_batch(*batch);
    }
    if (batch->state < RK_BATCH_STATE_PACKED && batch->nparams) {
        rk_pack_batch(*batch);
    }
    if (batch->ncommands) {
        if (rk_DrawElementsInstancedBaseInstance) {
            if (rk_MultiDrawElementsIndirect) {
                rk_MultiDrawElementsIndirect(
                    GL_TRIANGLES, GL_UNSIGNED_SHORT, nullptr, batch->ncommands, sizeof(rk_command));
            } else {
                rk_command const * const last_command = batch->commands + batch->ncommands;
                for (rk_command const * command = batch->commands; command < last_command; ++command) {
                    rk_DrawElementsInstancedBaseInstance(
                        GL_TRIANGLES, command->nvertices, GL_UNSIGNED_SHORT,
                        reinterpret_cast<void const *>(command->base_index << 1),
                        command->ninstances, command->base_instance);
                }
            }
        } else {
            rk_command const * const last_command = batch->commands + batch->ncommands;
            rk_parameter const * const last_param = batch->params + batch->nparams;
            unsigned param_index = 0;
            for (rk_command const * command = batch->commands; command < last_command; ++command) {
                for (rk_parameter const * param = batch->params; param < last_param; ++param) {
                    glBindVertexBuffer(param->binding, batch->params_buffer,
                        param->offset + param_index * param->dst_size, param->dst_size);
                }
                glDrawElementsInstanced(
                    GL_TRIANGLES, command->nvertices, GL_UNSIGNED_SHORT,
                    reinterpret_cast<void const *>(command->base_index << 1),
                    command->ninstances);
                param_index += command->ninstances;
            }
        }
    }
    if (rk_MultiDrawElementsIndirect) {
        glBindBuffer(GL_DRAW_INDIRECT_BUFFER, 0);
    }
    glBindVertexArray(0);
}

void rk_unselect_texture(
    rk_uint slot,
    rk_texture_t _texture) {
    rk_texture const * const texture = reinterpret_cast<rk_texture const *>(_texture);
    if (texture) {
        glActiveTexture(GL_TEXTURE0 + slot);
        if (texture->nlevels) {
            glBindTexture(GL_TEXTURE_2D, 0);
        } else {
            glBindTexture(GL_TEXTURE_2D_ARRAY, 0);
        }
    }
}

void rk_unselect_shader(
    rk_shader_t _shader) {
    glUseProgram(0);
}

void rk_destroy_batch(
    rk_batch_t _batch) {
    rk_batch * const batch = reinterpret_cast<rk_batch *>(_batch);
    if (batch) {
        delete[] batch->indices;
        delete[] batch->commands;
        delete[] batch->flags;
        delete[] batch->meshes;
        if (rk_MultiDrawElementsIndirect) {
            glDeleteBuffers(1, &batch->commands_buffer);
        }
        if (batch->nparams) {
            for (rk_parameter * param = batch->params; param < batch->params + batch->nparams; ++param) {
                delete[] param->source;
            }
            delete[] batch->params;
            glDeleteBuffers(1, &batch->params_buffer);
        }
        glDeleteVertexArrays(1, &batch->vertex_array);
        delete batch;
    }
}

void rk_destroy_triangles(
    rk_triangles_t _triangles) {
    rk_triangles * const triangles = reinterpret_cast<rk_triangles *>(_triangles);
    if (triangles) {
        glDeleteBuffers(1, &triangles->vertices);
        glDeleteVertexArrays(1, &triangles->array);
        delete triangles;
    }
}

void rk_destroy_vertices(
    rk_vertices_t _vertices) {
    rk_vertices * const vertices = reinterpret_cast<rk_vertices *>(_vertices);
    if (vertices) {
        delete[] vertices->format;
        delete[] vertices->vertices;
        delete[] vertices->indices;
        if (vertices->vertices_buffer) {
            glDeleteBuffers(1, &vertices->vertices_buffer);
        }
        if (vertices->indices_buffer) {
            glDeleteBuffers(1, &vertices->indices_buffer);
        }
        delete vertices;
    }
}

void rk_destroy_texture(
    rk_texture_t _texture) {
    rk_texture * const texture = reinterpret_cast<rk_texture *>(_texture);
    if (texture) {
        glDeleteTextures(1, &texture->texture);
        delete texture;
    }
}

void rk_destroy_shader(
    rk_shader_t _shader) {
    rk_shader * const shader = reinterpret_cast<rk_shader *>(_shader);
    if (shader) {
        glDeleteShader(shader->vertex);
        glDeleteShader(shader->fragment);
        glDeleteProgram(shader->program);
        delete shader;
    }
}