Improve typing in render.

Makefile

@@ -21,7 +21,7 @@ cpp/math.cpp
 
 OUTPUTFILE = engine.so
 
-CXXFLAGS = -Wall -Werror -O2 -march=x86-64 -mtune=generic -msse2 -ffast-math -fpic -flto -fno-rtti -fno-exceptions
+CXXFLAGS = -Wall -Werror -O2 -msse2 -ffast-math -fpic -flto -fno-rtti -fno-exceptions
 
 .PHONY: all
 all: clean $(OUTPUTFILE)

__init__.py

@@ -380,8 +380,9 @@ _create_vertices.argtypes = (
     ctypes.c_void_p) # meshes
 
 def create_vertices(format, nvertices, vertices, indices, meshes):
+    assert len(meshes) % 2 == 0
     return _create_vertices(format,
-        nvertices, _ubyte_addr(vertices), len(indices), _ushort_addr(indices), len(meshes), _uint_addr(meshes))
+        nvertices, _ubyte_addr(vertices), len(indices), _ushort_addr(indices), len(meshes) // 2, _uint_addr(meshes))
 
 create_batch = _engine.rk_create_batch
 create_batch.restype = _handle

cpp/cmp_memcpy.hpp

@@ -0,0 +1,99 @@
+// Copyright (C) 2023 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/>.
+
+#ifndef RK_ENGINE_CMP_MEMCPY_H
+#define RK_ENGINE_CMP_MEMCPY_H
+
+#include "types.hpp"
+
+template<typename _small>
+bool _rk_cmp_memcpy_small(
+    _small * __restrict dst,
+    _small const * __restrict src,
+    unsigned count) {
+    _small cmp = 0;
+    do {
+        cmp |= *dst ^ *src;
+        *dst++ = *src++;
+    } while(--count > 0);
+    return (cmp != 0);
+}
+
+template<typename _big, typename _small>
+bool _rk_cmp_memcpy_big(
+    _small * const __restrict _dst,
+    _small const * const __restrict _src,
+    unsigned const _count) {
+    unsigned const ratio = sizeof(_big) / sizeof(_small);
+    unsigned big_count = _count / ratio;
+    unsigned const small_count = _count % ratio;
+    _big * dst = reinterpret_cast<_big *>(_dst);
+    _big const * src = reinterpret_cast<_big const *>(_src);
+    _big cmp = 0;
+    do {
+        cmp |= *dst ^ *src;
+        *dst++ = *src++;
+    } while(--big_count > 0);
+    bool modified = (cmp != 0);
+    if (small_count) {
+        modified |= _rk_cmp_memcpy_small<_small>(
+            reinterpret_cast<_small *>(dst), reinterpret_cast<_small const *>(src), small_count);
+    }
+    return modified;
+}
+
+#ifdef RK_CMP_MEMCPY_UNALIGNED
+#define _rk_count_and_alignment(_t) (count >= (sizeof(_t) / sizeof(_small)))
+#else
+#define _rk_count_and_alignment(_t) ((count >= (sizeof(_t) / sizeof(_small))) && !(alignment % sizeof(_t)))
+#endif
+
+template<typename _small>
+bool rk_cmp_memcpy(
+    _small * const __restrict _dst,
+    _small const * const __restrict _src,
+    unsigned const count) {
+#ifndef RK_CMP_MEMCPY_UNALIGNED
+    unsigned const alignment = reinterpret_cast<uintptr_t>(_dst) | reinterpret_cast<uintptr_t const>(_src);
+#endif
+    if (sizeof(_small) < sizeof(rk_ullong)) {
+        if (_rk_count_and_alignment(rk_ullong)) {
+            return _rk_cmp_memcpy_big<rk_ullong, _small>(_dst, _src, count);
+        }
+    }
+    if (sizeof(_small) < sizeof(rk_ulong)) {
+        if (_rk_count_and_alignment(rk_ulong)) {
+            return _rk_cmp_memcpy_big<rk_ulong, _small>(_dst, _src, count);
+        }
+    }
+    if (sizeof(_small) < sizeof(rk_uint)) {
+        if (_rk_count_and_alignment(rk_uint)) {
+            return _rk_cmp_memcpy_big<rk_uint, _small>(_dst, _src, count);
+        }
+    }
+    if (sizeof(_small) < sizeof(rk_ushort)) {
+        if (_rk_count_and_alignment(rk_ushort)) {
+            return _rk_cmp_memcpy_big<rk_ushort, _small>(_dst, _src, count);
+        }
+    }
+    if (count) {
+        return _rk_cmp_memcpy_small<_small>(_dst, _src, count);
+    }
+    return false;
+}
+
+#undef _rk_count_and_alignment
+
+#endif // RK_ENGINE_CMP_MEMCPY_H

cpp/math.hpp

@@ -26,6 +26,14 @@ typedef glm::vec4 rk_vec4;
 typedef glm::mat3 rk_mat3;
 typedef glm::mat4 rk_mat4;
 
+#define RK_CHECK_MATH_TYPE(_t, _e, _c) static_assert(sizeof(_t) == sizeof(_e) * (_c))
+
+RK_CHECK_MATH_TYPE(rk_vec2, float, 2);
+RK_CHECK_MATH_TYPE(rk_vec3, float, 3);
+RK_CHECK_MATH_TYPE(rk_vec4, float, 4);
+RK_CHECK_MATH_TYPE(rk_mat3, rk_vec3, 3);
+RK_CHECK_MATH_TYPE(rk_mat4, rk_vec4, 4);
+
 #define vec3_right   (rk_vec3(1.f, 0.f, 0.f))
 #define vec3_forward (rk_vec3(0.f, 1.f, 0.f))
 #define vec3_up      (rk_vec3(0.f, 0.f, 1.f))

cpp/render.hpp

@@ -27,14 +27,18 @@ typedef rk_handle_t rk_triangles_t;
 typedef rk_handle_t rk_vertices_t;
 typedef rk_handle_t rk_batch_t;
 
-enum rk_texture_format : rk_uint {
+typedef rk_uint rk_texture_format;
+
+enum : rk_uint {
     RK_TEXTURE_FORMAT_SRGB8_A8 = 0,
     RK_TEXTURE_FORMAT_RGBA8 = 1,
     RK_TEXTURE_FORMAT_RGB10_A2 = 2,
     RK_TEXTURE_FORMAT_FLOAT_32 = 3
 };
 
-enum rk_texture_flags : rk_uint {
+typedef rk_uint rk_texture_flags;
+
+enum : rk_uint {
     RK_TEXTURE_FLAG_3D = RK_FLAG(0),
     RK_TEXTURE_FLAG_MIPMAPS = RK_FLAG(1),
     RK_TEXTURE_FLAG_MIN_NEAREST = 0,
@@ -43,41 +47,53 @@ enum rk_texture_flags : rk_uint {
     RK_TEXTURE_FLAG_MAG_LINEAR = RK_FLAG(3),
 };
 
-enum rk_vertex_format : rk_ubyte {
+typedef rk_ubyte rk_vertex_format;
+
+enum : rk_ubyte {
     RK_VERTEX_FORMAT_VEC3_FLOAT = 1,
     RK_VERTEX_FORMAT_VEC3_INT10 = 2,
-    RK_VERTEX_FORMAT_VEC3_UINT10 = 3
+    RK_VERTEX_FORMAT_VEC3_UINT10 = 3,
+    RK_VERTEX_FORMAT_NORMALIZE = RK_FLAG(7),
+    RK_VERTEX_FORMAT_MASK = RK_VERTEX_FORMAT_NORMALIZE - 1
 };
 
-enum : rk_ubyte { RK_VERTEX_FORMAT_NORMALIZE = RK_FLAG(7) };
-enum : rk_ubyte { RK_VERTEX_FORMAT_MASK = RK_VERTEX_FORMAT_NORMALIZE - 1 };
+typedef rk_ubyte rk_param_format;
 
-enum rk_param_format : rk_ubyte {
+enum : rk_ubyte {
     RK_PARAM_FORMAT_VEC3_FLOAT = 1,
     RK_PARAM_FORMAT_VEC3_SHORT = 2,
     RK_PARAM_FORMAT_VEC3_INT10 = 3,
     RK_PARAM_FORMAT_MAT3_FLOAT = 4,
-    RK_PARAM_FORMAT_MAT3_INT10 = 5
+    RK_PARAM_FORMAT_MAT3_INT10 = 5,
+    RK_PARAM_FORMAT_NORMALIZE = RK_FLAG(7),
+    RK_PARAM_FORMAT_MASK = RK_PARAM_FORMAT_NORMALIZE - 1
 };
 
-enum : rk_ubyte { RK_PARAM_FORMAT_NORMALIZE = RK_FLAG(7) };
-enum : rk_ubyte { RK_PARAM_FORMAT_MASK = RK_PARAM_FORMAT_NORMALIZE - 1 };
+typedef rk_ubyte rk_instance_flags;
 
-enum rk_instance_flags : rk_ubyte {
+enum : rk_ubyte {
     RK_INSTANCE_FLAG_SPAWNED = RK_FLAG(0),
-    RK_INSTANCE_FLAG_VISIBLE = RK_FLAG(1)
+    RK_INSTANCE_FLAG_VISIBLE = RK_FLAG(1),
+    RK_INSTANCE_FLAGS_SPAWNED_VISIBLE = RK_INSTANCE_FLAG_SPAWNED | RK_INSTANCE_FLAG_VISIBLE
 };
 
-enum : rk_ubyte { RK_INSTANCE_FLAGS_SPAWNED_VISIBLE = RK_INSTANCE_FLAG_SPAWNED | RK_INSTANCE_FLAG_VISIBLE };
+typedef rk_ushort rk_instance_index;
 
-enum : rk_uint { RK_BATCH_MAX_SIZE = 65536 };
+enum : rk_uint {
+    RK_BATCH_MAX_SIZE = 1 << (sizeof(rk_instance_index) * 8)
+};
 
-union rk_mesh {
-    rk_uint packed;
-    struct {
-        rk_ushort base_index;
-        rk_ushort ntriangles;
-    };
+typedef rk_ushort rk_vertex_index;
+typedef rk_ushort rk_mesh_index;
+
+// param input types must be size compatible with an array of rk_param_input
+typedef rk_uint rk_param_input;
+
+#define RK_CHECK_PARAM_INPUT_TYPE(_t) static_assert(!(sizeof(_t) % sizeof(rk_param_input)))
+
+struct rk_mesh {
+    rk_uint base_index;
+    rk_uint ntriangles;
 };
 
 RK_EXPORT void rk_render_initialize(
@@ -116,7 +132,7 @@ RK_EXPORT rk_vertices_t rk_create_vertices(
     rk_uint nvertices,
     rk_ubyte const * vertices,
     rk_uint nindices,
-    rk_ushort const * indices,
+    rk_vertex_index const * indices,
     rk_uint nmeshes,
     rk_mesh const * meshes);
 
@@ -129,8 +145,8 @@ RK_EXPORT 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_mesh_index const * meshes,
+    rk_param_input const * const * params);
 
 RK_EXPORT void rk_clear_buffer(
     rk_bool pixels,

cpp/render/render_opengles.cpp

@@ -16,6 +16,8 @@
 #include "../render.hpp"
 #include "render_opengles.hpp"
 #include "../display/display_glx.hpp"
+#include "../cmp_memcpy.hpp"
+#include <cstdlib>
 #include <cstdio>
 #include <cstring>
 
@@ -25,6 +27,9 @@ typedef void (*rk_MultiDrawElementsIndirectFunc)(rk_uint, rk_uint, const void *,
 static rk_DrawElementsInstancedBaseInstanceFunc rk_DrawElementsInstancedBaseInstance = nullptr;
 static rk_MultiDrawElementsIndirectFunc rk_MultiDrawElementsIndirect = nullptr;
 
+static unsigned rk_nbuckets = 0;
+static rk_bucket * rk_buckets = nullptr;
+
 static void rk_gl_printf(char const * message) {
     printf("[GL] %s\n", message);
 }
@@ -314,7 +319,7 @@ rk_vertices_t rk_create_vertices(
     rk_uint nvertices,
     rk_ubyte const * _vertices,
     rk_uint nindices,
-    rk_ushort const * indices,
+    rk_vertex_index const * indices,
     rk_uint nmeshes,
     rk_mesh const * meshes) {
     if (!format || !nvertices || !_vertices || !nindices || !indices) {
@@ -352,8 +357,8 @@ rk_vertices_t rk_create_vertices(
     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->indices = new rk_vertex_index[nindices];
+    memcpy(vertices->indices, indices, nindices * sizeof(rk_vertex_index));
     vertices->meshes = new rk_mesh[nmeshes];
     memcpy(vertices->meshes, meshes, nmeshes * sizeof(rk_mesh));
     vertices->vertices_buffer = 0;
@@ -361,28 +366,83 @@ rk_vertices_t rk_create_vertices(
     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_instance_index *>(malloc(size * sizeof(rk_instance_index)));
+        }
+        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_instance_index *>(
+                    realloc(bucket.indices, size * sizeof(rk_instance_index)));
+                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_instance_index *>(
+                    realloc(bucket.indices, size * sizeof(rk_instance_index)));
+            }
+        }
+        for (unsigned index = rk_nbuckets; index < count; ++index) {
+            rk_bucket & bucket = rk_buckets[index];
+            bucket.size = size;
+            bucket.indices = reinterpret_cast<rk_instance_index *>(
+                malloc(size * sizeof(rk_instance_index)));
+        }
+        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_instance_index);
+        }
+        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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
@@ -393,14 +453,14 @@ static void rk_pack_vec3_short(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
@@ -412,14 +472,14 @@ static void rk_pack_vec3_short_norm(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
     }
@@ -428,14 +488,14 @@ static void rk_pack_vec3_int10(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
     }
@@ -444,13 +504,13 @@ static void rk_pack_vec3_int10_norm(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index const * __restrict index = indices; index < last_index; ++index, ++dst) {
         *dst = src[*index];
     }
     #undef _convert
@@ -458,14 +518,14 @@ static void rk_pack_mat3_float(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
@@ -476,14 +536,14 @@ static void rk_pack_mat3_int10(
 
 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_instance_index const * const __restrict indices,
+    rk_param_output * __restrict _dst,
+    rk_param_input const * const __restrict _src) {
+    rk_instance_index 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) {
+    for (rk_instance_index 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);
@@ -551,9 +611,13 @@ rk_batch_t rk_create_batch(
     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->flags, 0xFF, max_size * sizeof(rk_instance_flags));
+    batch->meshes = new rk_mesh_index[max_size];
+    memset(batch->meshes, 0xFF, max_size * sizeof(rk_mesh_index));
+    batch->indices = new rk_instance_index[max_size];
+    memset(batch->indices, 0, max_size * sizeof(rk_instance_index));
     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 {
@@ -572,7 +636,8 @@ rk_batch_t rk_create_batch(
     } 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);
+        glBufferData(GL_ELEMENT_ARRAY_BUFFER,
+            vertices->nindices * sizeof(rk_vertex_index), vertices->indices, GL_STATIC_DRAW);
     }
     if (rk_MultiDrawElementsIndirect) {
         glGenBuffers(1, &batch->commands_buffer);
@@ -682,52 +747,50 @@ rk_batch_t rk_create_batch(
                     break;
             }
             glVertexBindingDivisor(binding, 1);
-            param->source = new rk_ubyte[max_size * param->src_size];
+            param->src_len = param->src_size / sizeof(rk_param_input);
+            param->dst_len = param->dst_size / sizeof(rk_param_output);
+            param->source = new rk_param_input[max_size * param->src_len];
+            memset(param->source, 0xFF, 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_instance_flags const * flags = batch.flags;
-    rk_ushort * indices = batch.indices;
-    for (unsigned index = 0; index < batch.count; ++index, ++flags) {
+    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_mesh_index 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) {
-            *indices++ = index;
+            rk_bucket & __restrict bucket = rk_buckets[*mesh_index];
+            bucket.indices[bucket.count++] = index;
+        }
+    }
+    rk_instance_index * __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) {
+            memcpy(indices, bucket->indices, bucket->count * sizeof(rk_instance_index));
+            command->nvertices = mesh->ntriangles * 3;
+            command->ninstances = bucket->count;
+            command->base_index = mesh->base_index;
+            command->base_instance = indices - batch.indices;
+            indices += bucket->count;
+            ++command;
         }
     }
     batch.ninstances = indices - batch.indices;
-    batch.ncommands = 0;
-    if (batch.ninstances) {
-        rk_command * const last_command = batch.commands + batch.vertices->nmeshes;
-        rk_command * command = batch.commands;
-        rk_ushort * base = batch.indices;
-        rk_ushort * const last = batch.indices + batch.ninstances;
-        for (rk_ushort * first = batch.indices; first < last && command < last_command; base = first, ++command) {
-            unsigned const mesh_index = batch.meshes[*first++];
-            for ( ; first < last && mesh_index == batch.meshes[*first]; ++first) {
-            }
-            for (rk_ushort * second = first; second < last; ++second) {
-                unsigned const index = *second;
-                if (mesh_index == batch.meshes[index]) {
-                    *second = *first;
-                    *first++ = index;
-                }
-            }
-            rk_mesh const & mesh = batch.vertices->meshes[mesh_index];
-            command->nvertices = static_cast<GLuint>(mesh.ntriangles) * 3;
-            command->ninstances = first - base;
-            command->base_index = mesh.base_index;
-            command->base_vertex = 0;
-            command->base_instance = base - batch.indices;
-        }
-        batch.ncommands = command - batch.commands;
-        if (rk_MultiDrawElementsIndirect && batch.ncommands) {
-            glBufferSubData(GL_DRAW_INDIRECT_BUFFER, 0, batch.ncommands * sizeof(rk_command), batch.commands);
-        }
+    batch.ncommands = command - batch.commands;
+    if (rk_MultiDrawElementsIndirect) {
+        glBufferSubData(GL_DRAW_INDIRECT_BUFFER, 0, batch.ncommands * sizeof(rk_command), batch.commands);
     }
     batch.state = RK_BATCH_STATE_SORTED;
 }
@@ -740,7 +803,7 @@ static void rk_pack_batch(
             if (param->dirty) {
                 param->dirty = false;
                 if (batch.ninstances) {
-                    rk_ubyte * const dst = reinterpret_cast<rk_ubyte *>(
+                    rk_param_output * const dst = reinterpret_cast<rk_param_output *>(
                         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) {
@@ -755,38 +818,12 @@ static void rk_pack_batch(
     batch.state = RK_BATCH_STATE_PACKED;
 }
 
-static bool rk_compare_replace(
-    void * __restrict _dst,
-    void const * __restrict _src,
-    unsigned const size) {
-    rk_ulong hash = 0;
-    rk_ulong * dst = reinterpret_cast<rk_ulong *>(_dst);
-    rk_ulong const * src = reinterpret_cast<rk_ulong const *>(_src);
-    unsigned count = size / sizeof(rk_ulong);
-    unsigned remain = (size - count * sizeof(rk_ulong));
-    if (count) {
-        do {
-            hash |= *dst ^ *src;
-            *dst++ = *src++;
-        } while(--count > 0);
-    }
-    rk_ubyte * rdst = reinterpret_cast<rk_ubyte *>(dst);
-    rk_ubyte const * rsrc = reinterpret_cast<rk_ubyte const *>(src);
-    if (remain) {
-        do {
-            hash |= *rdst ^ *rsrc;
-            *rdst++ = *rsrc++;
-        } while(--remain > 0);
-    }
-    return (hash != 0);
-}
-
 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_mesh_index const * meshes,
+    rk_param_input 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.");
@@ -797,7 +834,7 @@ void rk_fill_batch(
     if (batch->nparams) {
         got_all_params = (params != nullptr);
         if (params) {
-            for (rk_ubyte const * const * param = params; param < params + batch->nparams; ++param) {
+            for (rk_param_input 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;
@@ -815,22 +852,20 @@ void rk_fill_batch(
         return;
     }
     batch->count = count;
-    bool const cmp_flags =
-        (flags && rk_compare_replace(batch->flags, flags, batch->count * sizeof(rk_instance_flags)));
-    bool const cmp_meshes =
-        (meshes && rk_compare_replace(batch->meshes, meshes, batch->count * sizeof(rk_mesh)));
+    bool const cmp_flags = (flags && rk_cmp_memcpy(batch->flags, flags, batch->count));
+    bool const cmp_meshes = (meshes && rk_cmp_memcpy(batch->meshes, meshes, batch->count));
     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_compare_replace(dst->source, *src, batch->count * dst->src_size)) || need_sorting;
+            rk_param_input const * const * src = params;
+            for (rk_parameter const * param = batch->params; param < last_param; ++param, ++src) {
+                param->dirty =
+                    ((*src && rk_cmp_memcpy(param->source, *src, batch->count * param->src_len)) || need_sorting);
             }
         } else if (need_sorting) {
-            for (rk_parameter const * dst = batch->params; dst < last_param; ++dst) {
-                dst->dirty = true;
+            for (rk_parameter const * param = batch->params; param < last_param; ++param) {
+                param->dirty = true;
             }
         }
     }

cpp/render/render_opengles.hpp

@@ -17,6 +17,7 @@
 #define _RK_ENGINE_RENDER_OPENGLES_H
 
 #include "../types.hpp"
+#include "../math.hpp"
 #include <GLES3/gl32.h>
 #include <GLES3/gl3ext.h>
 #include <GLES3/gl3platform.h>
@@ -44,7 +45,7 @@ struct rk_vertices {
     unsigned nmeshes;
     rk_vertex_format * format;
     rk_ubyte * vertices;
-    rk_ushort * indices;
+    rk_vertex_index * indices;
     rk_mesh * meshes;
     GLuint vertices_buffer;
     GLuint indices_buffer;
@@ -58,12 +59,19 @@ struct rk_command {
     GLuint base_instance;
 };
 
+// param output types must be size compatible with an array of rk_param_output
+typedef rk_uint rk_param_output;
+
+#define RK_CHECK_PARAM_OUTPUT_TYPE(_t) static_assert(!(sizeof(_t) % sizeof(rk_param_output)))
+
 struct rk_vec3_float {
     float x;
     float y;
     float z;
 };
 
+static_assert(sizeof(rk_vec3_float) == sizeof(rk_vec3));
+
 struct rk_vec3_short {
     rk_short x;
     rk_short y;
@@ -80,17 +88,29 @@ struct rk_mat3_float {
     rk_vec3_float z;
 };
 
+static_assert(sizeof(rk_mat3_float) == sizeof(rk_mat3));
+
 struct rk_mat3_int10 {
     rk_vec3_int10 x;
     rk_vec3_int10 y;
     rk_vec3_int10 z;
 };
 
+RK_CHECK_PARAM_INPUT_TYPE(rk_vec3_float);
+RK_CHECK_PARAM_INPUT_TYPE(rk_mat3_float);
+
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_vec3_float);
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_vec3_short);
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_vec3_int10);
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_vec3_uint10);
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_mat3_float);
+RK_CHECK_PARAM_OUTPUT_TYPE(rk_mat3_int10);
+
 typedef void (*rk_packer)(
     unsigned const, // count
-    rk_ushort const * const, // indices
-    rk_ubyte *, // dst
-    rk_ubyte const * const); // src
+    rk_instance_index const * const, // indices
+    rk_param_output *, // dst
+    rk_param_input const * const); // src
 
 struct rk_parameter {
     mutable bool dirty;
@@ -98,10 +118,18 @@ struct rk_parameter {
     unsigned offset;
     unsigned src_size;
     unsigned dst_size;
-    rk_ubyte * source;
+    unsigned src_len;
+    unsigned dst_len;
+    rk_param_input * source;
     rk_packer packer;
 };
 
+struct rk_bucket {
+    unsigned size;
+    unsigned count;
+    rk_instance_index * indices;
+};
+
 enum rk_batch_state {
     RK_BATCH_STATE_EMPTY = 0,
     RK_BATCH_STATE_FILLED = 1,
@@ -118,8 +146,8 @@ struct rk_batch {
     unsigned nparams;
     rk_vertices const * vertices;
     rk_instance_flags * flags;
-    rk_ushort * meshes;
-    rk_ushort * indices;
+    rk_mesh_index * meshes;
+    rk_instance_index * indices;
     rk_command * commands;
     rk_parameter * params;
     GLuint vertex_array;

cpp/types.hpp

@@ -33,6 +33,8 @@ typedef int32_t rk_int;
 typedef uint32_t rk_uint;
 typedef int64_t rk_long;
 typedef uint64_t rk_ulong;
+typedef __int128 rk_llong;
+typedef unsigned __int128 rk_ullong;
 typedef float rk_float;
 typedef void * rk_handle_t;