New lightmapper

-Added LocalVector (needed it)
-Added stb_rect_pack (It's pretty cool, we could probably use it for other stuff too)
-Fixes and changes all around the place
-Added library for 128 bits fixed point (required for Delaunay3D)

21 files changed, 3581 insertions, 4 deletions

diff --git a/modules/denoise/SCsub b/modules/denoise/SCsub
new file mode 100644
index 0000000000..8cf91b7dbd
--- /dev/null
+++ b/modules/denoise/SCsub
@@ -0,0 +1,119 @@
+#!/usr/bin/env python
+
+import resource_to_cpp
+from platform_methods import run_in_subprocess
+
+Import("env")
+Import("env_modules")
+
+env_oidn = env_modules.Clone()
+
+# Thirdparty source files
+thirdparty_dir = "#thirdparty/oidn/"
+thirdparty_sources = [
+    "core/api.cpp",
+    "core/device.cpp",
+    "core/filter.cpp",
+    "core/network.cpp",
+    "core/autoencoder.cpp",
+    "core/transfer_function.cpp",
+    "weights/rtlightmap_hdr.cpp",
+    "mkl-dnn/src/common/batch_normalization.cpp",
+    "mkl-dnn/src/common/concat.cpp",
+    "mkl-dnn/src/common/convolution.cpp",
+    "mkl-dnn/src/common/convolution_pd.cpp",
+    "mkl-dnn/src/common/deconvolution.cpp",
+    "mkl-dnn/src/common/eltwise.cpp",
+    "mkl-dnn/src/common/engine.cpp",
+    "mkl-dnn/src/common/inner_product.cpp",
+    "mkl-dnn/src/common/inner_product_pd.cpp",
+    "mkl-dnn/src/common/lrn.cpp",
+    "mkl-dnn/src/common/memory.cpp",
+    "mkl-dnn/src/common/memory_desc_wrapper.cpp",
+    "mkl-dnn/src/common/mkldnn_debug.cpp",
+    "mkl-dnn/src/common/mkldnn_debug_autogenerated.cpp",
+    "mkl-dnn/src/common/pooling.cpp",
+    "mkl-dnn/src/common/primitive.cpp",
+    "mkl-dnn/src/common/primitive_attr.cpp",
+    "mkl-dnn/src/common/primitive_desc.cpp",
+    "mkl-dnn/src/common/primitive_exec_types.cpp",
+    "mkl-dnn/src/common/primitive_iterator.cpp",
+    "mkl-dnn/src/common/query.cpp",
+    "mkl-dnn/src/common/reorder.cpp",
+    "mkl-dnn/src/common/rnn.cpp",
+    "mkl-dnn/src/common/scratchpad.cpp",
+    "mkl-dnn/src/common/shuffle.cpp",
+    "mkl-dnn/src/common/softmax.cpp",
+    "mkl-dnn/src/common/stream.cpp",
+    "mkl-dnn/src/common/sum.cpp",
+    "mkl-dnn/src/common/utils.cpp",
+    "mkl-dnn/src/common/verbose.cpp",
+    "mkl-dnn/src/cpu/cpu_barrier.cpp",
+    "mkl-dnn/src/cpu/cpu_concat.cpp",
+    "mkl-dnn/src/cpu/cpu_engine.cpp",
+    "mkl-dnn/src/cpu/cpu_memory.cpp",
+    "mkl-dnn/src/cpu/cpu_reducer.cpp",
+    "mkl-dnn/src/cpu/cpu_reorder.cpp",
+    "mkl-dnn/src/cpu/cpu_sum.cpp",
+    "mkl-dnn/src/cpu/jit_avx2_conv_kernel_f32.cpp",
+    "mkl-dnn/src/cpu/jit_avx2_convolution.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_common_conv_kernel.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_common_conv_winograd_kernel_f32.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_common_convolution.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_common_convolution_winograd.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_core_fp32_wino_conv_2x3.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_core_fp32_wino_conv_4x3.cpp",
+    "mkl-dnn/src/cpu/jit_avx512_core_fp32_wino_conv_4x3_kernel.cpp",
+    "mkl-dnn/src/cpu/jit_sse42_conv_kernel_f32.cpp",
+    "mkl-dnn/src/cpu/jit_sse42_convolution.cpp",
+    "mkl-dnn/src/cpu/jit_transpose_src_utils.cpp",
+    "mkl-dnn/src/cpu/jit_uni_eltwise.cpp",
+    "mkl-dnn/src/cpu/jit_uni_pool_kernel_f32.cpp",
+    "mkl-dnn/src/cpu/jit_uni_pooling.cpp",
+    "mkl-dnn/src/cpu/jit_uni_reorder.cpp",
+    "mkl-dnn/src/cpu/jit_uni_reorder_utils.cpp",
+    "mkl-dnn/src/cpu/jit_utils/jit_utils.cpp",
+    "mkl-dnn/src/cpu/jit_utils/jitprofiling/jitprofiling.c",
+    "common/platform.cpp",
+    "common/thread.cpp",
+    "common/tensor.cpp",
+]
+thirdparty_sources = [thirdparty_dir + file for file in thirdparty_sources]
+
+thirdparty_include_dirs = [
+    "",
+    "include",
+    "mkl-dnn/include",
+    "mkl-dnn/src",
+    "mkl-dnn/src/common",
+    "mkl-dnn/src/cpu/xbyak",
+    "mkl-dnn/src/cpu",
+]
+thirdparty_include_dirs = [thirdparty_dir + file for file in thirdparty_include_dirs]
+
+
+env_oidn.Prepend(CPPPATH=thirdparty_include_dirs)
+env_oidn.Append(
+    CPPDEFINES=[
+        "MKLDNN_THR=MKLDNN_THR_SEQ",
+        "OIDN_STATIC_LIB",
+        "__STDC_CONSTANT_MACROS",
+        "__STDC_LIMIT_MACROS",
+        "DISABLE_VERBOSE",
+        "MKLDNN_ENABLE_CONCURRENT_EXEC",
+        "NDEBUG",
+    ]
+)
+
+env_thirdparty = env_oidn.Clone()
+env_thirdparty.disable_warnings()
+env_thirdparty.add_source_files(env.modules_sources, thirdparty_sources)
+
+weights_in_path = thirdparty_dir + "weights/rtlightmap_hdr.tza"
+weights_out_path = thirdparty_dir + "weights/rtlightmap_hdr.cpp"
+
+env_thirdparty.Depends(weights_out_path, weights_in_path)
+env_thirdparty.CommandNoCache(weights_out_path, weights_in_path, resource_to_cpp.tza_to_cpp)
+
+env_oidn.add_source_files(env.modules_sources, "denoise_wrapper.cpp")
+env_modules.add_source_files(env.modules_sources, ["register_types.cpp", "lightmap_denoiser.cpp"])
diff --git a/modules/denoise/config.py b/modules/denoise/config.py
new file mode 100644
index 0000000000..53b8f2f2e3
--- /dev/null
+++ b/modules/denoise/config.py
@@ -0,0 +1,6 @@
+def can_build(env, platform):
+    return env["tools"]
+
+
+def configure(env):
+    pass
diff --git a/modules/denoise/denoise_wrapper.cpp b/modules/denoise/denoise_wrapper.cpp
new file mode 100644
index 0000000000..feeeaef507
--- /dev/null
+++ b/modules/denoise/denoise_wrapper.cpp
@@ -0,0 +1,34 @@
+#include "denoise_wrapper.h"
+#include "thirdparty/oidn/include/OpenImageDenoise/oidn.h"
+#include <stdio.h>
+
+void *oidn_denoiser_init() {
+	OIDNDeviceImpl *device = oidnNewDevice(OIDN_DEVICE_TYPE_CPU);
+	oidnCommitDevice(device);
+	return device;
+}
+
+bool oidn_denoise(void *deviceptr, float *p_floats, int p_width, int p_height) {
+	OIDNDeviceImpl *device = (OIDNDeviceImpl *)deviceptr;
+	OIDNFilter filter = oidnNewFilter(device, "RTLightmap");
+	oidnSetSharedFilterImage(filter, "color", (void *)p_floats, OIDN_FORMAT_FLOAT3, p_width, p_height, 0, 0, 0);
+	oidnSetSharedFilterImage(filter, "output", (void *)p_floats, OIDN_FORMAT_FLOAT3, p_width, p_height, 0, 0, 0);
+	oidnSetFilter1b(filter, "hdr", true);
+	//oidnSetFilter1f(filter, "hdrScale", 1.0f);
+	oidnCommitFilter(filter);
+	oidnExecuteFilter(filter);
+
+	const char *msg;
+	bool success = true;
+	if (oidnGetDeviceError(device, &msg) != OIDN_ERROR_NONE) {
+		printf("LightmapDenoiser: %s\n", msg);
+		success = false;
+	}
+
+	oidnReleaseFilter(filter);
+	return success;
+}
+
+void oidn_denoiser_finish(void *device) {
+	oidnReleaseDevice((OIDNDeviceImpl *)device);
+}
diff --git a/modules/denoise/denoise_wrapper.h b/modules/denoise/denoise_wrapper.h
new file mode 100644
index 0000000000..3aef326e22
--- /dev/null
+++ b/modules/denoise/denoise_wrapper.h
@@ -0,0 +1,8 @@
+#ifndef DENOISE_WRAPPER_H
+#define DENOISE_WRAPPER_H
+
+void *oidn_denoiser_init();
+bool oidn_denoise(void *device, float *p_floats, int p_width, int p_height);
+void oidn_denoiser_finish(void *device);
+
+#endif // DENOISE_WRAPPER_H
diff --git a/modules/denoise/lightmap_denoiser.cpp b/modules/denoise/lightmap_denoiser.cpp
new file mode 100644
index 0000000000..c821b22d85
--- /dev/null
+++ b/modules/denoise/lightmap_denoiser.cpp
@@ -0,0 +1,63 @@
+/*************************************************************************/
+/*  lightmap_denoiser.cpp                                                */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+#include "lightmap_denoiser.h"
+#include "denoise_wrapper.h"
+
+LightmapDenoiser *LightmapDenoiserOIDN::create_oidn_denoiser() {
+	return memnew(LightmapDenoiserOIDN);
+}
+
+void LightmapDenoiserOIDN::make_default_denoiser() {
+	create_function = create_oidn_denoiser;
+}
+
+Ref<Image> LightmapDenoiserOIDN::denoise_image(const Ref<Image> &p_image) {
+
+	Ref<Image> img = p_image->duplicate();
+
+	img->convert(Image::FORMAT_RGBF);
+
+	Vector<uint8_t> data = img->get_data();
+	if (!oidn_denoise(device, (float *)data.ptrw(), img->get_width(), img->get_height())) {
+		return p_image;
+	}
+
+	img->create(img->get_width(), img->get_height(), false, img->get_format(), data);
+	return img;
+}
+
+LightmapDenoiserOIDN::LightmapDenoiserOIDN() {
+	device = oidn_denoiser_init();
+}
+
+LightmapDenoiserOIDN::~LightmapDenoiserOIDN() {
+	oidn_denoiser_finish(device);
+}
diff --git a/modules/denoise/lightmap_denoiser.h b/modules/denoise/lightmap_denoiser.h
new file mode 100644
index 0000000000..ac0cc8b9db
--- /dev/null
+++ b/modules/denoise/lightmap_denoiser.h
@@ -0,0 +1,57 @@
+/*************************************************************************/
+/*  lightmap_denoiser.h                                                  */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+#ifndef LIGHTMAP_DENOISER_H
+#define LIGHTMAP_DENOISER_H
+
+#include "core/object.h"
+#include "scene/3d/lightmapper.h"
+
+struct OIDNDeviceImpl;
+
+class LightmapDenoiserOIDN : public LightmapDenoiser {
+
+	GDCLASS(LightmapDenoiserOIDN, LightmapDenoiser);
+
+protected:
+	void *device = nullptr;
+
+public:
+	static LightmapDenoiser *create_oidn_denoiser();
+
+	Ref<Image> denoise_image(const Ref<Image> &p_image);
+
+	static void make_default_denoiser();
+
+	LightmapDenoiserOIDN();
+	~LightmapDenoiserOIDN();
+};
+
+#endif // LIGHTMAP_DENOISER_H
diff --git a/modules/denoise/register_types.cpp b/modules/denoise/register_types.cpp
new file mode 100644
index 0000000000..b6b92701c8
--- /dev/null
+++ b/modules/denoise/register_types.cpp
@@ -0,0 +1,41 @@
+/*************************************************************************/
+/*  register_types.cpp                                                   */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+#include "register_types.h"
+#include "core/engine.h"
+#include "lightmap_denoiser.h"
+
+void register_denoise_types() {
+
+	LightmapDenoiserOIDN::make_default_denoiser();
+}
+
+void unregister_denoise_types() {
+}
diff --git a/modules/denoise/register_types.h b/modules/denoise/register_types.h
new file mode 100644
index 0000000000..2ffc36ee2c
--- /dev/null
+++ b/modules/denoise/register_types.h
@@ -0,0 +1,32 @@
+/*************************************************************************/
+/*  register_types.h                                                     */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+void register_denoise_types();
+void unregister_denoise_types();
diff --git a/modules/denoise/resource_to_cpp.py b/modules/denoise/resource_to_cpp.py
new file mode 100644
index 0000000000..4c0b67f701
--- /dev/null
+++ b/modules/denoise/resource_to_cpp.py
@@ -0,0 +1,70 @@
+#!/usr/bin/env python
+
+## ======================================================================== ##
+## Copyright 2009-2019 Intel Corporation                                    ##
+##                                                                          ##
+## Licensed under the Apache License, Version 2.0 (the "License");          ##
+## you may not use this file except in compliance with the License.         ##
+## You may obtain a copy of the License at                                  ##
+##                                                                          ##
+##     http://www.apache.org/licenses/LICENSE-2.0                           ##
+##                                                                          ##
+## Unless required by applicable law or agreed to in writing, software      ##
+## distributed under the License is distributed on an "AS IS" BASIS,        ##
+## WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. ##
+## See the License for the specific language governing permissions and      ##
+## limitations under the License.                                           ##
+## ======================================================================== ##
+
+import os
+import sys
+import argparse
+from array import array
+
+# Generates a C++ file from the specified binary resource file
+def generate(in_path, out_path):
+
+    namespace = "oidn::weights"
+    scopes = namespace.split("::")
+
+    file_name = os.path.basename(in_path)
+    var_name = os.path.splitext(file_name)[0]
+
+    with open(in_path, "rb") as in_file, open(out_path, "w") as out_file:
+        # Header
+        out_file.write("// Generated from: %s\n" % file_name)
+        out_file.write("#include <cstddef>\n\n")
+
+        # Open the namespaces
+        for s in scopes:
+            out_file.write("namespace %s {\n" % s)
+        if scopes:
+            out_file.write("\n")
+
+        # Read the file
+        in_data = array("B", in_file.read())
+
+        # Write the size
+        out_file.write("//const size_t %s_size = %d;\n\n" % (var_name, len(in_data)))
+
+        # Write the data
+        out_file.write("unsigned char %s[] = {" % var_name)
+        for i in range(len(in_data)):
+            c = in_data[i]
+            if i > 0:
+                out_file.write(",")
+            if (i + 1) % 20 == 1:
+                out_file.write("\n")
+            out_file.write("%d" % c)
+        out_file.write("\n};\n")
+
+        # Close the namespaces
+        if scopes:
+            out_file.write("\n")
+        for scope in reversed(scopes):
+            out_file.write("} // namespace %s\n" % scope)
+
+
+def tza_to_cpp(target, source, env):
+    for x in zip(source, target):
+        generate(str(x[0]), str(x[1]))
diff --git a/modules/lightmapper_rd/SCsub b/modules/lightmapper_rd/SCsub
new file mode 100644
index 0000000000..2f04f1833e
--- /dev/null
+++ b/modules/lightmapper_rd/SCsub
@@ -0,0 +1,12 @@
+#!/usr/bin/env python
+
+Import("env")
+Import("env_modules")
+
+env_lightmapper_rd = env_modules.Clone()
+env_lightmapper_rd.GLSL_HEADER("lm_raster.glsl")
+env_lightmapper_rd.GLSL_HEADER("lm_compute.glsl")
+env_lightmapper_rd.GLSL_HEADER("lm_blendseams.glsl")
+
+# Godot source files
+env_lightmapper_rd.add_source_files(env.modules_sources, "*.cpp")
diff --git a/modules/lightmapper_rd/config.py b/modules/lightmapper_rd/config.py
new file mode 100644
index 0000000000..d22f9454ed
--- /dev/null
+++ b/modules/lightmapper_rd/config.py
@@ -0,0 +1,6 @@
+def can_build(env, platform):
+    return True
+
+
+def configure(env):
+    pass
diff --git a/modules/lightmapper_rd/lightmapper_rd.cpp b/modules/lightmapper_rd/lightmapper_rd.cpp
new file mode 100644
index 0000000000..6983c222c0
--- /dev/null
+++ b/modules/lightmapper_rd/lightmapper_rd.cpp
@@ -0,0 +1,1754 @@
+#include "lightmapper_rd.h"
+#include "core/math/geometry.h"
+#include "core/project_settings.h"
+#include "lm_blendseams.glsl.gen.h"
+#include "lm_compute.glsl.gen.h"
+#include "lm_raster.glsl.gen.h"
+#include "servers/rendering/rendering_device_binds.h"
+
+//uncomment this if you want to see textures from all the process saved
+//#define DEBUG_TEXTURES
+
+void LightmapperRD::add_mesh(const MeshData &p_mesh) {
+	ERR_FAIL_COND(p_mesh.albedo_on_uv2.is_null() || p_mesh.albedo_on_uv2->empty());
+	ERR_FAIL_COND(p_mesh.emission_on_uv2.is_null() || p_mesh.emission_on_uv2->empty());
+	ERR_FAIL_COND(p_mesh.albedo_on_uv2->get_width() != p_mesh.emission_on_uv2->get_width());
+	ERR_FAIL_COND(p_mesh.albedo_on_uv2->get_height() != p_mesh.emission_on_uv2->get_height());
+	ERR_FAIL_COND(p_mesh.points.size() == 0);
+	MeshInstance mi;
+	mi.data = p_mesh;
+	mesh_instances.push_back(mi);
+}
+
+void LightmapperRD::add_directional_light(bool p_static, const Vector3 &p_direction, const Color &p_color, float p_energy, float p_angular_distance) {
+	Light l;
+	l.type = LIGHT_TYPE_DIRECTIONAL;
+	l.direction[0] = p_direction.x;
+	l.direction[1] = p_direction.y;
+	l.direction[2] = p_direction.z;
+	l.color[0] = p_color.r;
+	l.color[1] = p_color.g;
+	l.color[2] = p_color.b;
+	l.energy = p_energy;
+	l.static_bake = p_static;
+	l.size = p_angular_distance;
+	lights.push_back(l);
+}
+void LightmapperRD::add_omni_light(bool p_static, const Vector3 &p_position, const Color &p_color, float p_energy, float p_range, float p_attenuation, float p_size) {
+	Light l;
+	l.type = LIGHT_TYPE_OMNI;
+	l.position[0] = p_position.x;
+	l.position[1] = p_position.y;
+	l.position[2] = p_position.z;
+	l.range = p_range;
+	l.attenuation = p_attenuation;
+	l.color[0] = p_color.r;
+	l.color[1] = p_color.g;
+	l.color[2] = p_color.b;
+	l.energy = p_energy;
+	l.static_bake = p_static;
+	l.size = p_size;
+	lights.push_back(l);
+}
+void LightmapperRD::add_spot_light(bool p_static, const Vector3 &p_position, const Vector3 p_direction, const Color &p_color, float p_energy, float p_range, float p_attenuation, float p_spot_angle, float p_spot_attenuation, float p_size) {
+
+	Light l;
+	l.type = LIGHT_TYPE_SPOT;
+	l.position[0] = p_position.x;
+	l.position[1] = p_position.y;
+	l.position[2] = p_position.z;
+	l.direction[0] = p_direction.x;
+	l.direction[1] = p_direction.y;
+	l.direction[2] = p_direction.z;
+	l.range = p_range;
+	l.attenuation = p_attenuation;
+	l.spot_angle = Math::deg2rad(p_spot_angle);
+	l.spot_attenuation = p_spot_attenuation;
+	l.color[0] = p_color.r;
+	l.color[1] = p_color.g;
+	l.color[2] = p_color.b;
+	l.energy = p_energy;
+	l.static_bake = p_static;
+	l.size = p_size;
+	lights.push_back(l);
+}
+
+void LightmapperRD::add_probe(const Vector3 &p_position) {
+	Probe probe;
+	probe.position[0] = p_position.x;
+	probe.position[1] = p_position.y;
+	probe.position[2] = p_position.z;
+	probe.position[3] = 0;
+	probe_positions.push_back(probe);
+}
+
+void LightmapperRD::_plot_triangle_into_triangle_index_list(int p_size, const Vector3i &p_ofs, const AABB &p_bounds, const Vector3 p_points[3], uint32_t p_triangle_index, LocalVector<TriangleSort> &triangles, uint32_t p_grid_size) {
+
+	int half_size = p_size / 2;
+
+	for (int i = 0; i < 8; i++) {
+
+		AABB aabb = p_bounds;
+		aabb.size *= 0.5;
+		Vector3i n = p_ofs;
+
+		if (i & 1) {
+			aabb.position.x += aabb.size.x;
+			n.x += half_size;
+		}
+		if (i & 2) {
+			aabb.position.y += aabb.size.y;
+			n.y += half_size;
+		}
+		if (i & 4) {
+			aabb.position.z += aabb.size.z;
+			n.z += half_size;
+		}
+
+		{
+			Vector3 qsize = aabb.size * 0.5; //quarter size, for fast aabb test
+
+			if (!Geometry::triangle_box_overlap(aabb.position + qsize, qsize, p_points)) {
+				//does not fit in child, go on
+				continue;
+			}
+		}
+
+		if (half_size == 1) {
+			//got to the end
+			TriangleSort ts;
+			ts.cell_index = n.x + (n.y * p_grid_size) + (n.z * p_grid_size * p_grid_size);
+			ts.triangle_index = p_triangle_index;
+			triangles.push_back(ts);
+		} else {
+			_plot_triangle_into_triangle_index_list(half_size, n, aabb, p_points, p_triangle_index, triangles, p_grid_size);
+		}
+	}
+}
+
+Lightmapper::BakeError LightmapperRD::_blit_meshes_into_atlas(int p_max_texture_size, Vector<Ref<Image>> &albedo_images, Vector<Ref<Image>> &emission_images, AABB &bounds, Size2i &atlas_size, int &atlas_slices, BakeStepFunc p_step_function, void *p_bake_userdata) {
+
+	Vector<Size2i> sizes;
+
+	for (int m_i = 0; m_i < mesh_instances.size(); m_i++) {
+
+		MeshInstance &mi = mesh_instances.write[m_i];
+		Size2i s = Size2i(mi.data.albedo_on_uv2->get_width(), mi.data.albedo_on_uv2->get_height());
+		sizes.push_back(s);
+		atlas_size.width = MAX(atlas_size.width, s.width);
+		atlas_size.height = MAX(atlas_size.height, s.height);
+	}
+
+	int max = nearest_power_of_2_templated(atlas_size.width);
+	max = MAX(max, nearest_power_of_2_templated(atlas_size.height));
+
+	if (max > p_max_texture_size) {
+		return BAKE_ERROR_LIGHTMAP_TOO_SMALL;
+	}
+
+	if (p_step_function) {
+		p_step_function(0.1, TTR("Determining optimal atlas size"), p_bake_userdata, true);
+	}
+
+	atlas_size = Size2i(max, max);
+
+	Size2i best_atlas_size;
+	int best_atlas_slices = 0;
+	int best_atlas_memory = 0x7FFFFFFF;
+	Vector<Vector3i> best_atlas_offsets;
+
+	//determine best texture array atlas size by bruteforce fitting
+	while (atlas_size.x <= p_max_texture_size && atlas_size.y <= p_max_texture_size) {
+
+		Vector<Vector2i> source_sizes = sizes;
+		Vector<int> source_indices;
+		source_indices.resize(source_sizes.size());
+		for (int i = 0; i < source_indices.size(); i++) {
+			source_indices.write[i] = i;
+		}
+		Vector<Vector3i> atlas_offsets;
+		atlas_offsets.resize(source_sizes.size());
+
+		int slices = 0;
+
+		while (source_sizes.size() > 0) {
+
+			Vector<Vector3i> offsets = Geometry::partial_pack_rects(source_sizes, atlas_size);
+			Vector<int> new_indices;
+			Vector<Vector2i> new_sources;
+			for (int i = 0; i < offsets.size(); i++) {
+				Vector3i ofs = offsets[i];
+				int sidx = source_indices[i];
+				if (ofs.z > 0) {
+					//valid
+					ofs.z = slices;
+					atlas_offsets.write[sidx] = ofs;
+				} else {
+					new_indices.push_back(sidx);
+					new_sources.push_back(source_sizes[i]);
+				}
+			}
+
+			source_sizes = new_sources;
+			source_indices = new_indices;
+			slices++;
+		}
+
+		int mem_used = atlas_size.x * atlas_size.y * slices;
+		if (mem_used < best_atlas_memory) {
+			best_atlas_size = atlas_size;
+			best_atlas_offsets = atlas_offsets;
+			best_atlas_slices = slices;
+			best_atlas_memory = mem_used;
+		}
+
+		if (atlas_size.width == atlas_size.height) {
+			atlas_size.width *= 2;
+		} else {
+			atlas_size.height *= 2;
+		}
+	}
+	atlas_size = best_atlas_size;
+	atlas_slices = best_atlas_slices;
+
+	// apply the offsets and slice to all images, and also blit albedo and emission
+	albedo_images.resize(atlas_slices);
+	emission_images.resize(atlas_slices);
+
+	if (p_step_function) {
+		p_step_function(0.2, TTR("Blitting albedo and emission"), p_bake_userdata, true);
+	}
+
+	for (int i = 0; i < atlas_slices; i++) {
+		Ref<Image> albedo;
+		albedo.instance();
+		albedo->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBA8);
+		albedo->set_as_black();
+		albedo_images.write[i] = albedo;
+
+		Ref<Image> emission;
+		emission.instance();
+		emission->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH);
+		emission->set_as_black();
+		emission_images.write[i] = emission;
+	}
+
+	//assign uv positions
+
+	for (int m_i = 0; m_i < mesh_instances.size(); m_i++) {
+
+		MeshInstance &mi = mesh_instances.write[m_i];
+		mi.offset.x = best_atlas_offsets[m_i].x;
+		mi.offset.y = best_atlas_offsets[m_i].y;
+		mi.slice = best_atlas_offsets[m_i].z;
+		albedo_images.write[mi.slice]->blit_rect(mi.data.albedo_on_uv2, Rect2(Vector2(), Size2i(mi.data.albedo_on_uv2->get_width(), mi.data.albedo_on_uv2->get_height())), mi.offset);
+		emission_images.write[mi.slice]->blit_rect(mi.data.emission_on_uv2, Rect2(Vector2(), Size2i(mi.data.emission_on_uv2->get_width(), mi.data.emission_on_uv2->get_height())), mi.offset);
+	}
+
+	return BAKE_OK;
+}
+
+void LightmapperRD::_create_acceleration_structures(RenderingDevice *rd, Size2i atlas_size, int atlas_slices, AABB &bounds, int grid_size, Vector<Probe> &probe_positions, GenerateProbes p_generate_probes, Vector<int> &slice_triangle_count, Vector<int> &slice_seam_count, RID &vertex_buffer, RID &triangle_buffer, RID &box_buffer, RID &lights_buffer, RID &triangle_cell_indices_buffer, RID &probe_positions_buffer, RID &grid_texture, RID &grid_texture_sdf, RID &seams_buffer, BakeStepFunc p_step_function, void *p_bake_userdata) {
+
+	HashMap<Vertex, uint32_t, VertexHash> vertex_map;
+
+	//fill triangles array and vertex array
+	LocalVector<Triangle> triangles;
+	LocalVector<Vertex> vertex_array;
+	LocalVector<Box> box_array;
+	LocalVector<Seam> seams;
+
+	slice_triangle_count.resize(atlas_slices);
+	slice_seam_count.resize(atlas_slices);
+
+	for (int i = 0; i < atlas_slices; i++) {
+		slice_triangle_count.write[i] = 0;
+		slice_seam_count.write[i] = 0;
+	}
+
+	bounds = AABB();
+
+	for (int m_i = 0; m_i < mesh_instances.size(); m_i++) {
+
+		if (p_step_function) {
+			float p = float(m_i + 1) / mesh_instances.size() * 0.1;
+			p_step_function(0.3 + p, vformat(TTR("Plotting mesh into acceleration structure %d/%d"), m_i + 1, mesh_instances.size()), p_bake_userdata, false);
+		}
+
+		HashMap<Edge, EdgeUV2, EdgeHash> edges;
+
+		MeshInstance &mi = mesh_instances.write[m_i];
+
+		Vector2 uv_scale = Vector2(mi.data.albedo_on_uv2->get_width(), mi.data.albedo_on_uv2->get_height()) / Vector2(atlas_size);
+		Vector2 uv_offset = Vector2(mi.offset) / Vector2(atlas_size);
+		if (m_i == 0) {
+			bounds.position = mi.data.points[0];
+		}
+
+		for (int i = 0; i < mi.data.points.size(); i += 3) {
+
+			Vector3 vtxs[3] = { mi.data.points[i + 0], mi.data.points[i + 1], mi.data.points[i + 2] };
+			Vector2 uvs[3] = { mi.data.uv2[i + 0] * uv_scale + uv_offset, mi.data.uv2[i + 1] * uv_scale + uv_offset, mi.data.uv2[i + 2] * uv_scale + uv_offset };
+			Vector3 normal[3] = { mi.data.normal[i + 0], mi.data.normal[i + 1], mi.data.normal[i + 2] };
+
+			AABB taabb;
+			Triangle t;
+			t.slice = mi.slice;
+			for (int k = 0; k < 3; k++) {
+
+				bounds.expand_to(vtxs[k]);
+
+				Vertex v;
+				v.position[0] = vtxs[k].x;
+				v.position[1] = vtxs[k].y;
+				v.position[2] = vtxs[k].z;
+				v.uv[0] = uvs[k].x;
+				v.uv[1] = uvs[k].y;
+				v.normal_xy[0] = normal[k].x;
+				v.normal_xy[1] = normal[k].y;
+				v.normal_z = normal[k].z;
+
+				uint32_t *indexptr = vertex_map.getptr(v);
+
+				if (indexptr) {
+					t.indices[k] = *indexptr;
+				} else {
+					uint32_t new_index = vertex_map.size();
+					t.indices[k] = new_index;
+					vertex_map[v] = new_index;
+					vertex_array.push_back(v);
+				}
+
+				if (k == 0) {
+					taabb.position = vtxs[k];
+				} else {
+					taabb.expand_to(vtxs[k]);
+				}
+			}
+
+			//compute seams that will need to be blended later
+			for (int k = 0; k < 3; k++) {
+				int n = (k + 1) % 3;
+
+				Edge edge(vtxs[k], vtxs[n], normal[k], normal[n]);
+				Vector2i edge_indices(t.indices[k], t.indices[n]);
+				EdgeUV2 uv2(uvs[k], uvs[n], edge_indices);
+
+				if (edge.b == edge.a) {
+					continue; //degenerate, somehow
+				}
+				if (edge.b < edge.a) {
+					SWAP(edge.a, edge.b);
+					SWAP(edge.na, edge.nb);
+					SWAP(uv2.a, uv2.b);
+					SWAP(edge_indices.x, edge_indices.y);
+				}
+
+				EdgeUV2 *euv2 = edges.getptr(edge);
+				if (!euv2) {
+					edges[edge] = uv2;
+				} else {
+					if (*euv2 == uv2) {
+						continue; // seam shared UV space, no need to blend
+					}
+					if (euv2->seam_found) {
+						continue; //bad geometry
+					}
+
+					Seam seam;
+					seam.a = edge_indices;
+					seam.b = euv2->indices;
+					seam.slice = mi.slice;
+					seams.push_back(seam);
+					slice_seam_count.write[mi.slice]++;
+					euv2->seam_found = true;
+				}
+			}
+
+			Box box;
+			box.min_bounds[0] = taabb.position.x;
+			box.min_bounds[1] = taabb.position.y;
+			box.min_bounds[2] = taabb.position.z;
+			box.max_bounds[0] = taabb.position.x + MAX(taabb.size.x, 0.0001);
+			box.max_bounds[1] = taabb.position.y + MAX(taabb.size.y, 0.0001);
+			box.max_bounds[2] = taabb.position.z + MAX(taabb.size.z, 0.0001);
+			box.pad0 = box.pad1 = 0; //make valgrind not complain
+			box_array.push_back(box);
+
+			triangles.push_back(t);
+			slice_triangle_count.write[t.slice]++;
+		}
+	}
+
+	//also consider probe positions for bounds
+	for (int i = 0; i < probe_positions.size(); i++) {
+		Vector3 pp(probe_positions[i].position[0], probe_positions[i].position[1], probe_positions[i].position[2]);
+		bounds.expand_to(pp);
+	}
+	bounds.grow_by(0.1); //grow a bit to avoid numerical error
+
+	triangles.sort(); //sort by slice
+	seams.sort();
+
+	if (p_step_function) {
+		p_step_function(0.4, TTR("Optimizing acceleration structure"), p_bake_userdata, true);
+	}
+
+	//fill list of triangles in grid
+	LocalVector<TriangleSort> triangle_sort;
+	for (uint32_t i = 0; i < triangles.size(); i++) {
+
+		const Triangle &t = triangles[i];
+		Vector3 face[3] = {
+			Vector3(vertex_array[t.indices[0]].position[0], vertex_array[t.indices[0]].position[1], vertex_array[t.indices[0]].position[2]),
+			Vector3(vertex_array[t.indices[1]].position[0], vertex_array[t.indices[1]].position[1], vertex_array[t.indices[1]].position[2]),
+			Vector3(vertex_array[t.indices[2]].position[0], vertex_array[t.indices[2]].position[1], vertex_array[t.indices[2]].position[2])
+		};
+		_plot_triangle_into_triangle_index_list(grid_size, Vector3i(), bounds, face, i, triangle_sort, grid_size);
+	}
+	//sort it
+	triangle_sort.sort();
+
+	Vector<uint32_t> triangle_indices;
+	triangle_indices.resize(triangle_sort.size());
+	Vector<uint32_t> grid_indices;
+	grid_indices.resize(grid_size * grid_size * grid_size * 2);
+	zeromem(grid_indices.ptrw(), grid_indices.size() * sizeof(uint32_t));
+	Vector<bool> solid;
+	solid.resize(grid_size * grid_size * grid_size);
+	zeromem(solid.ptrw(), solid.size() * sizeof(bool));
+
+	{
+		uint32_t *tiw = triangle_indices.ptrw();
+		uint32_t last_cell = 0xFFFFFFFF;
+		uint32_t *giw = grid_indices.ptrw();
+		bool *solidw = solid.ptrw();
+		for (uint32_t i = 0; i < triangle_sort.size(); i++) {
+			uint32_t cell = triangle_sort[i].cell_index;
+			if (cell != last_cell) {
+				//cell changed, update pointer to indices
+				giw[cell * 2 + 1] = i;
+				last_cell = cell;
+				solidw[cell] = true;
+			}
+			tiw[i] = triangle_sort[i].triangle_index;
+			giw[cell * 2]++; //update counter
+			last_cell = cell;
+		}
+	}
+#if 0
+	for (int i = 0; i < grid_size; i++) {
+		for (int j = 0; j < grid_size; j++) {
+			for (int k = 0; k < grid_size; k++) {
+				uint32_t index = i * (grid_size * grid_size) + j * grid_size + k;
+				grid_indices.write[index * 2] = float(i) / grid_size * 255;
+				grid_indices.write[index * 2 + 1] = float(j) / grid_size * 255;
+			}
+		}
+	}
+#endif
+
+#if 0
+	for (int i = 0; i < grid_size; i++) {
+		Vector<uint8_t> grid_usage;
+		grid_usage.resize(grid_size * grid_size);
+		for (int j = 0; j < grid_usage.size(); j++) {
+			uint32_t ofs = i * grid_size * grid_size + j;
+			uint32_t count = grid_indices[ofs * 2];
+			grid_usage.write[j] = count > 0 ? 255 : 0;
+		}
+
+		Ref<Image> img;
+		img.instance();
+		img->create(grid_size, grid_size, false, Image::FORMAT_L8, grid_usage);
+		img->save_png("res://grid_layer_" + itos(1000 + i).substr(1, 3) + ".png");
+	}
+#endif
+	if (p_step_function) {
+		p_step_function(0.45, TTR("Generating Signed Distance Field"), p_bake_userdata, true);
+	}
+
+	//generate SDF for raytracing
+	Vector<uint32_t> euclidean_pos = Geometry::generate_edf(solid, Vector3i(grid_size, grid_size, grid_size), false);
+	Vector<uint32_t> euclidean_neg = Geometry::generate_edf(solid, Vector3i(grid_size, grid_size, grid_size), true);
+	Vector<int8_t> sdf8 = Geometry::generate_sdf8(euclidean_pos, euclidean_neg);
+
+	/*****************************/
+	/*** CREATE GPU STRUCTURES ***/
+	/*****************************/
+
+	lights.sort();
+
+	Vector<Vector2i> seam_buffer_vec;
+	seam_buffer_vec.resize(seams.size() * 2);
+	for (uint32_t i = 0; i < seams.size(); i++) {
+		seam_buffer_vec.write[i * 2 + 0] = seams[i].a;
+		seam_buffer_vec.write[i * 2 + 1] = seams[i].b;
+	}
+
+	{ //buffers
+		Vector<uint8_t> vb = vertex_array.to_byte_array();
+		vertex_buffer = rd->storage_buffer_create(vb.size(), vb);
+
+		Vector<uint8_t> tb = triangles.to_byte_array();
+		triangle_buffer = rd->storage_buffer_create(tb.size(), tb);
+
+		Vector<uint8_t> bb = box_array.to_byte_array();
+		box_buffer = rd->storage_buffer_create(bb.size(), bb);
+
+		Vector<uint8_t> tib = triangle_indices.to_byte_array();
+		triangle_cell_indices_buffer = rd->storage_buffer_create(tib.size(), tib);
+
+		Vector<uint8_t> lb = lights.to_byte_array();
+		if (lb.size() == 0) {
+			lb.resize(sizeof(Light)); //even if no lights, the buffer must exist
+		}
+		lights_buffer = rd->storage_buffer_create(lb.size(), lb);
+
+		Vector<uint8_t> sb = seam_buffer_vec.to_byte_array();
+		if (sb.size() == 0) {
+			sb.resize(sizeof(Vector2i) * 2); //even if no seams, the buffer must exist
+		}
+		seams_buffer = rd->storage_buffer_create(sb.size(), sb);
+
+		Vector<uint8_t> pb = probe_positions.to_byte_array();
+		if (pb.size() == 0) {
+			pb.resize(sizeof(Probe));
+		}
+		probe_positions_buffer = rd->storage_buffer_create(pb.size(), pb);
+	}
+
+	{ //grid
+
+		RD::TextureFormat tf;
+		tf.width = grid_size;
+		tf.height = grid_size;
+		tf.depth = grid_size;
+		tf.type = RD::TEXTURE_TYPE_3D;
+		tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+
+		Vector<Vector<uint8_t>> texdata;
+		texdata.resize(1);
+		//grid and indices
+		tf.format = RD::DATA_FORMAT_R32G32_UINT;
+		texdata.write[0] = grid_indices.to_byte_array();
+		grid_texture = rd->texture_create(tf, RD::TextureView(), texdata);
+		//sdf
+		tf.format = RD::DATA_FORMAT_R8_SNORM;
+		texdata.write[0] = sdf8.to_byte_array();
+		grid_texture_sdf = rd->texture_create(tf, RD::TextureView(), texdata);
+	}
+}
+
+void LightmapperRD::_raster_geometry(RenderingDevice *rd, Size2i atlas_size, int atlas_slices, int grid_size, AABB bounds, float p_bias, Vector<int> slice_triangle_count, RID position_tex, RID unocclude_tex, RID normal_tex, RID raster_depth_buffer, RID rasterize_shader, RID raster_base_uniform) {
+
+	Vector<RID> framebuffers;
+
+	for (int i = 0; i < atlas_slices; i++) {
+		RID slice_pos_tex = rd->texture_create_shared_from_slice(RD::TextureView(), position_tex, i, 0);
+		RID slice_unoc_tex = rd->texture_create_shared_from_slice(RD::TextureView(), unocclude_tex, i, 0);
+		RID slice_norm_tex = rd->texture_create_shared_from_slice(RD::TextureView(), normal_tex, i, 0);
+		Vector<RID> fb;
+		fb.push_back(slice_pos_tex);
+		fb.push_back(slice_norm_tex);
+		fb.push_back(slice_unoc_tex);
+		fb.push_back(raster_depth_buffer);
+		framebuffers.push_back(rd->framebuffer_create(fb));
+	}
+
+	RD::PipelineDepthStencilState ds;
+	ds.enable_depth_test = true;
+	ds.enable_depth_write = true;
+	ds.depth_compare_operator = RD::COMPARE_OP_LESS; //so it does render same pixel twice
+
+	RID raster_pipeline = rd->render_pipeline_create(rasterize_shader, rd->framebuffer_get_format(framebuffers[0]), RD::INVALID_FORMAT_ID, RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), ds, RD::PipelineColorBlendState::create_disabled(3), 0);
+	RID raster_pipeline_wire;
+	{
+
+		RD::PipelineRasterizationState rw;
+		rw.wireframe = true;
+		raster_pipeline_wire = rd->render_pipeline_create(rasterize_shader, rd->framebuffer_get_format(framebuffers[0]), RD::INVALID_FORMAT_ID, RD::RENDER_PRIMITIVE_TRIANGLES, rw, RD::PipelineMultisampleState(), ds, RD::PipelineColorBlendState::create_disabled(3), 0);
+	}
+
+	uint32_t triangle_offset = 0;
+	Vector<Color> clear_colors;
+	clear_colors.push_back(Color(0, 0, 0, 0));
+	clear_colors.push_back(Color(0, 0, 0, 0));
+	clear_colors.push_back(Color(0, 0, 0, 0));
+
+	for (int i = 0; i < atlas_slices; i++) {
+
+		RasterPushConstant raster_push_constant;
+		raster_push_constant.atlas_size[0] = atlas_size.x;
+		raster_push_constant.atlas_size[1] = atlas_size.y;
+		raster_push_constant.base_triangle = triangle_offset;
+		raster_push_constant.to_cell_offset[0] = bounds.position.x;
+		raster_push_constant.to_cell_offset[1] = bounds.position.y;
+		raster_push_constant.to_cell_offset[2] = bounds.position.z;
+		raster_push_constant.bias = p_bias;
+		raster_push_constant.to_cell_size[0] = (1.0 / bounds.size.x) * float(grid_size);
+		raster_push_constant.to_cell_size[1] = (1.0 / bounds.size.y) * float(grid_size);
+		raster_push_constant.to_cell_size[2] = (1.0 / bounds.size.z) * float(grid_size);
+		raster_push_constant.grid_size[0] = grid_size;
+		raster_push_constant.grid_size[1] = grid_size;
+		raster_push_constant.grid_size[2] = grid_size;
+		raster_push_constant.uv_offset[0] = 0;
+		raster_push_constant.uv_offset[1] = 0;
+
+		RD::DrawListID draw_list = rd->draw_list_begin(framebuffers[i], RD::INITIAL_ACTION_CLEAR, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CLEAR, RD::FINAL_ACTION_DISCARD, clear_colors);
+		//draw opaque
+		rd->draw_list_bind_render_pipeline(draw_list, raster_pipeline);
+		rd->draw_list_bind_uniform_set(draw_list, raster_base_uniform, 0);
+		rd->draw_list_set_push_constant(draw_list, &raster_push_constant, sizeof(RasterPushConstant));
+		rd->draw_list_draw(draw_list, false, 1, slice_triangle_count[i] * 3);
+		//draw wire
+		rd->draw_list_bind_render_pipeline(draw_list, raster_pipeline_wire);
+		rd->draw_list_bind_uniform_set(draw_list, raster_base_uniform, 0);
+		rd->draw_list_set_push_constant(draw_list, &raster_push_constant, sizeof(RasterPushConstant));
+		rd->draw_list_draw(draw_list, false, 1, slice_triangle_count[i] * 3);
+
+		rd->draw_list_end();
+
+		triangle_offset += slice_triangle_count[i];
+	}
+}
+
+LightmapperRD::BakeError LightmapperRD::bake(BakeQuality p_quality, bool p_use_denoiser, int p_bounces, float p_bias, int p_max_texture_size, bool p_bake_sh, GenerateProbes p_generate_probes, const Ref<Image> &p_environment_panorama, const Basis &p_environment_transform, BakeStepFunc p_step_function, void *p_bake_userdata) {
+
+	if (p_step_function) {
+		p_step_function(0.0, TTR("Begin Bake"), p_bake_userdata, true);
+	}
+	bake_textures.clear();
+	int grid_size = 128;
+
+	/* STEP 1: Fetch material textures and compute the bounds */
+
+	AABB bounds;
+	Size2i atlas_size;
+	int atlas_slices;
+	Vector<Ref<Image>> albedo_images;
+	Vector<Ref<Image>> emission_images;
+
+	BakeError bake_error = _blit_meshes_into_atlas(p_max_texture_size, albedo_images, emission_images, bounds, atlas_size, atlas_slices, p_step_function, p_bake_userdata);
+	if (bake_error != BAKE_OK) {
+		return bake_error;
+	}
+
+#ifdef DEBUG_TEXTURES
+	for (int i = 0; i < atlas_slices; i++) {
+		albedo_images[i]->save_png("res://0_albedo_" + itos(i) + ".png");
+		emission_images[i]->save_png("res://0_emission_" + itos(i) + ".png");
+	}
+#endif
+
+	RenderingDevice *rd = RenderingDevice::get_singleton()->create_local_device();
+
+	RID albedo_array_tex;
+	RID emission_array_tex;
+	RID normal_tex;
+	RID position_tex;
+	RID unocclude_tex;
+	RID light_source_tex;
+	RID light_dest_tex;
+	RID light_accum_tex;
+	RID light_accum_tex2;
+	RID light_primary_dynamic_tex;
+	RID light_environment_tex;
+
+#define FREE_TEXTURES                    \
+	rd->free(albedo_array_tex);          \
+	rd->free(emission_array_tex);        \
+	rd->free(normal_tex);                \
+	rd->free(position_tex);              \
+	rd->free(unocclude_tex);             \
+	rd->free(light_source_tex);          \
+	rd->free(light_accum_tex2);          \
+	rd->free(light_accum_tex);           \
+	rd->free(light_primary_dynamic_tex); \
+	rd->free(light_environment_tex);
+
+	{ // create all textures
+
+		Vector<Vector<uint8_t>> albedo_data;
+		Vector<Vector<uint8_t>> emission_data;
+		for (int i = 0; i < atlas_slices; i++) {
+			albedo_data.push_back(albedo_images[i]->get_data());
+			emission_data.push_back(emission_images[i]->get_data());
+		}
+
+		RD::TextureFormat tf;
+		tf.width = atlas_size.width;
+		tf.height = atlas_size.height;
+		tf.array_layers = atlas_slices;
+		tf.type = RD::TEXTURE_TYPE_2D_ARRAY;
+		tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+		tf.format = RD::DATA_FORMAT_R8G8B8A8_UNORM;
+
+		albedo_array_tex = rd->texture_create(tf, RD::TextureView(), albedo_data);
+
+		tf.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
+
+		emission_array_tex = rd->texture_create(tf, RD::TextureView(), emission_data);
+
+		//this will be rastered to
+		tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
+		normal_tex = rd->texture_create(tf, RD::TextureView());
+		tf.format = RD::DATA_FORMAT_R32G32B32A32_SFLOAT;
+		position_tex = rd->texture_create(tf, RD::TextureView());
+		unocclude_tex = rd->texture_create(tf, RD::TextureView());
+
+		tf.format = RD::DATA_FORMAT_R16G16B16A16_SFLOAT;
+		tf.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_CAN_COPY_FROM_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+
+		light_source_tex = rd->texture_create(tf, RD::TextureView());
+		rd->texture_clear(light_source_tex, Color(0, 0, 0, 0), 0, 1, 0, atlas_slices);
+		light_primary_dynamic_tex = rd->texture_create(tf, RD::TextureView());
+		rd->texture_clear(light_primary_dynamic_tex, Color(0, 0, 0, 0), 0, 1, 0, atlas_slices);
+
+		if (p_bake_sh) {
+			tf.array_layers *= 4;
+		}
+		light_accum_tex = rd->texture_create(tf, RD::TextureView());
+		rd->texture_clear(light_accum_tex, Color(0, 0, 0, 0), 0, 1, 0, tf.array_layers);
+		light_dest_tex = rd->texture_create(tf, RD::TextureView());
+		rd->texture_clear(light_dest_tex, Color(0, 0, 0, 0), 0, 1, 0, tf.array_layers);
+		light_accum_tex2 = light_dest_tex;
+
+		//env
+		{
+			Ref<Image> panorama_tex;
+			if (p_environment_panorama.is_valid()) {
+				panorama_tex = p_environment_panorama;
+				panorama_tex->convert(Image::FORMAT_RGBAF);
+			} else {
+				panorama_tex.instance();
+				panorama_tex->create(8, 8, false, Image::FORMAT_RGBAF);
+				for (int i = 0; i < 8; i++) {
+					for (int j = 0; j < 8; j++) {
+						panorama_tex->set_pixel(i, j, Color(0, 0, 0, 1));
+					}
+				}
+			}
+
+			RD::TextureFormat tfp;
+			tfp.width = panorama_tex->get_width();
+			tfp.height = panorama_tex->get_height();
+			tfp.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+			tfp.format = RD::DATA_FORMAT_R32G32B32A32_SFLOAT;
+
+			Vector<Vector<uint8_t>> tdata;
+			tdata.push_back(panorama_tex->get_data());
+			light_environment_tex = rd->texture_create(tfp, RD::TextureView(), tdata);
+
+#ifdef DEBUG_TEXTURES
+			panorama_tex->convert(Image::FORMAT_RGB8);
+			panorama_tex->save_png("res://0_panorama.png");
+#endif
+		}
+	}
+
+	/* STEP 2: create the acceleration structure for the GPU*/
+
+	Vector<int> slice_triangle_count;
+	RID vertex_buffer;
+	RID triangle_buffer;
+	RID box_buffer;
+	RID lights_buffer;
+	RID triangle_cell_indices_buffer;
+	RID grid_texture;
+	RID grid_texture_sdf;
+	RID seams_buffer;
+	RID probe_positions_buffer;
+
+	Vector<int> slice_seam_count;
+
+#define FREE_BUFFERS                        \
+	rd->free(vertex_buffer);                \
+	rd->free(triangle_buffer);              \
+	rd->free(box_buffer);                   \
+	rd->free(lights_buffer);                \
+	rd->free(triangle_cell_indices_buffer); \
+	rd->free(grid_texture);                 \
+	rd->free(grid_texture_sdf);             \
+	rd->free(seams_buffer);                 \
+	rd->free(probe_positions_buffer);
+
+	_create_acceleration_structures(rd, atlas_size, atlas_slices, bounds, grid_size, probe_positions, p_generate_probes, slice_triangle_count, slice_seam_count, vertex_buffer, triangle_buffer, box_buffer, lights_buffer, triangle_cell_indices_buffer, probe_positions_buffer, grid_texture, grid_texture_sdf, seams_buffer, p_step_function, p_bake_userdata);
+
+	if (p_step_function) {
+		p_step_function(0.47, TTR("Preparing shaders"), p_bake_userdata, true);
+	}
+
+	//shaders
+	Ref<RDShaderFile> raster_shader;
+	raster_shader.instance();
+	Error err = raster_shader->parse_versions_from_text(lm_raster_shader_glsl);
+	if (err != OK) {
+		raster_shader->print_errors("raster_shader");
+
+		FREE_TEXTURES
+		FREE_BUFFERS
+
+		memdelete(rd);
+	}
+	ERR_FAIL_COND_V(err != OK, BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES);
+
+	RID rasterize_shader = rd->shader_create_from_bytecode(raster_shader->get_bytecode());
+
+	ERR_FAIL_COND_V(rasterize_shader.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); //this is a bug check, though, should not happen
+
+	RID sampler;
+	{
+		RD::SamplerState s;
+		s.mag_filter = RD::SAMPLER_FILTER_LINEAR;
+		s.min_filter = RD::SAMPLER_FILTER_LINEAR;
+		s.max_lod = 0;
+
+		sampler = rd->sampler_create(s);
+	}
+
+	Vector<RD::Uniform> base_uniforms;
+	{
+		{
+
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 1;
+			u.ids.push_back(vertex_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 2;
+			u.ids.push_back(triangle_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 3;
+			u.ids.push_back(box_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 4;
+			u.ids.push_back(triangle_cell_indices_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 5;
+			u.ids.push_back(lights_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 6;
+			u.ids.push_back(seams_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+			u.binding = 7;
+			u.ids.push_back(probe_positions_buffer);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_TEXTURE;
+			u.binding = 8;
+			u.ids.push_back(grid_texture);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_TEXTURE;
+			u.binding = 9;
+			u.ids.push_back(grid_texture_sdf);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_TEXTURE;
+			u.binding = 10;
+			u.ids.push_back(albedo_array_tex);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_TEXTURE;
+			u.binding = 11;
+			u.ids.push_back(emission_array_tex);
+			base_uniforms.push_back(u);
+		}
+		{
+			RD::Uniform u;
+			u.type = RD::UNIFORM_TYPE_SAMPLER;
+			u.binding = 12;
+			u.ids.push_back(sampler);
+			base_uniforms.push_back(u);
+		}
+	}
+
+	RID raster_base_uniform = rd->uniform_set_create(base_uniforms, rasterize_shader, 0);
+	RID raster_depth_buffer;
+	{
+		RD::TextureFormat tf;
+		tf.width = atlas_size.width;
+		tf.height = atlas_size.height;
+		tf.depth = 1;
+		tf.type = RD::TEXTURE_TYPE_2D;
+		tf.usage_bits = RD::TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
+		tf.format = RD::DATA_FORMAT_D32_SFLOAT;
+
+		raster_depth_buffer = rd->texture_create(tf, RD::TextureView());
+	}
+
+	rd->submit();
+	rd->sync();
+
+	/* STEP 3: Raster the geometry to UV2 coords in the atlas textures GPU*/
+
+	_raster_geometry(rd, atlas_size, atlas_slices, grid_size, bounds, p_bias, slice_triangle_count, position_tex, unocclude_tex, normal_tex, raster_depth_buffer, rasterize_shader, raster_base_uniform);
+
+#ifdef DEBUG_TEXTURES
+
+	for (int i = 0; i < atlas_slices; i++) {
+		Vector<uint8_t> s = rd->texture_get_data(position_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAF, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://1_position_" + itos(i) + ".png");
+
+		s = rd->texture_get_data(normal_tex, i);
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://1_normal_" + itos(i) + ".png");
+	}
+#endif
+
+#define FREE_RASTER_RESOURCES   \
+	rd->free(rasterize_shader); \
+	rd->free(sampler);          \
+	rd->free(raster_depth_buffer);
+
+	/* Plot direct light */
+
+	Ref<RDShaderFile> compute_shader;
+	compute_shader.instance();
+	err = compute_shader->parse_versions_from_text(lm_compute_shader_glsl, p_bake_sh ? "\n#define USE_SH_LIGHTMAPS\n" : "");
+	if (err != OK) {
+
+		FREE_TEXTURES
+		FREE_BUFFERS
+		FREE_RASTER_RESOURCES
+		memdelete(rd);
+		compute_shader->print_errors("compute_shader");
+	}
+	ERR_FAIL_COND_V(err != OK, BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES);
+
+	//unoccluder
+	RID compute_shader_unocclude = rd->shader_create_from_bytecode(compute_shader->get_bytecode("unocclude"));
+	ERR_FAIL_COND_V(compute_shader_unocclude.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); // internal check, should not happen
+	RID compute_shader_unocclude_pipeline = rd->compute_pipeline_create(compute_shader_unocclude);
+
+	//direct light
+	RID compute_shader_primary = rd->shader_create_from_bytecode(compute_shader->get_bytecode("primary"));
+	ERR_FAIL_COND_V(compute_shader_primary.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); // internal check, should not happen
+	RID compute_shader_primary_pipeline = rd->compute_pipeline_create(compute_shader_primary);
+
+	//indirect light
+	RID compute_shader_secondary = rd->shader_create_from_bytecode(compute_shader->get_bytecode("secondary"));
+	ERR_FAIL_COND_V(compute_shader_secondary.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); //internal check, should not happen
+	RID compute_shader_secondary_pipeline = rd->compute_pipeline_create(compute_shader_secondary);
+
+	//dilate
+	RID compute_shader_dilate = rd->shader_create_from_bytecode(compute_shader->get_bytecode("dilate"));
+	ERR_FAIL_COND_V(compute_shader_dilate.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); //internal check, should not happen
+	RID compute_shader_dilate_pipeline = rd->compute_pipeline_create(compute_shader_dilate);
+
+	//dilate
+	RID compute_shader_light_probes = rd->shader_create_from_bytecode(compute_shader->get_bytecode("light_probes"));
+	ERR_FAIL_COND_V(compute_shader_light_probes.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); //internal check, should not happen
+	RID compute_shader_light_probes_pipeline = rd->compute_pipeline_create(compute_shader_light_probes);
+
+	RID compute_base_uniform_set = rd->uniform_set_create(base_uniforms, compute_shader_primary, 0);
+
+#define FREE_COMPUTE_RESOURCES          \
+	rd->free(compute_shader_unocclude); \
+	rd->free(compute_shader_primary);   \
+	rd->free(compute_shader_secondary); \
+	rd->free(compute_shader_dilate);    \
+	rd->free(compute_shader_light_probes);
+
+	PushConstant push_constant;
+	{
+		//set defaults
+		push_constant.atlas_size[0] = atlas_size.width;
+		push_constant.atlas_size[1] = atlas_size.height;
+		push_constant.world_size[0] = bounds.size.x;
+		push_constant.world_size[1] = bounds.size.y;
+		push_constant.world_size[2] = bounds.size.z;
+		push_constant.to_cell_offset[0] = bounds.position.x;
+		push_constant.to_cell_offset[1] = bounds.position.y;
+		push_constant.to_cell_offset[2] = bounds.position.z;
+		push_constant.bias = p_bias;
+		push_constant.to_cell_size[0] = (1.0 / bounds.size.x) * float(grid_size);
+		push_constant.to_cell_size[1] = (1.0 / bounds.size.y) * float(grid_size);
+		push_constant.to_cell_size[2] = (1.0 / bounds.size.z) * float(grid_size);
+		push_constant.light_count = lights.size();
+		push_constant.grid_size = grid_size;
+		push_constant.atlas_slice = 0;
+		push_constant.region_ofs[0] = 0;
+		push_constant.region_ofs[1] = 0;
+		push_constant.environment_xform[0] = p_environment_transform.elements[0][0];
+		push_constant.environment_xform[1] = p_environment_transform.elements[1][0];
+		push_constant.environment_xform[2] = p_environment_transform.elements[2][0];
+		push_constant.environment_xform[3] = 0;
+		push_constant.environment_xform[4] = p_environment_transform.elements[0][1];
+		push_constant.environment_xform[5] = p_environment_transform.elements[1][1];
+		push_constant.environment_xform[6] = p_environment_transform.elements[2][1];
+		push_constant.environment_xform[7] = 0;
+		push_constant.environment_xform[8] = p_environment_transform.elements[0][2];
+		push_constant.environment_xform[9] = p_environment_transform.elements[1][2];
+		push_constant.environment_xform[10] = p_environment_transform.elements[2][2];
+		push_constant.environment_xform[11] = 0;
+	}
+
+	Vector3i group_size((atlas_size.x - 1) / 8 + 1, (atlas_size.y - 1) / 8 + 1, 1);
+	rd->submit();
+	rd->sync();
+
+	if (p_step_function) {
+		p_step_function(0.49, TTR("Un-occluding geometry"), p_bake_userdata, true);
+	}
+
+	/* UNOCCLUDE */
+	{
+
+		Vector<RD::Uniform> uniforms;
+		{
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(position_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 1;
+				u.ids.push_back(unocclude_tex); //will be unused
+				uniforms.push_back(u);
+			}
+		}
+
+		RID unocclude_uniform_set = rd->uniform_set_create(uniforms, compute_shader_unocclude, 1);
+
+		RD::ComputeListID compute_list = rd->compute_list_begin();
+		rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_unocclude_pipeline);
+		rd->compute_list_bind_uniform_set(compute_list, compute_base_uniform_set, 0);
+		rd->compute_list_bind_uniform_set(compute_list, unocclude_uniform_set, 1);
+
+		for (int i = 0; i < atlas_slices; i++) {
+			push_constant.atlas_slice = i;
+			rd->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
+			rd->compute_list_dispatch(compute_list, group_size.x, group_size.y, group_size.z);
+			//no barrier, let them run all together
+		}
+		rd->compute_list_end(); //done
+	}
+
+	if (p_step_function) {
+		p_step_function(0.5, TTR("Plot direct lighting"), p_bake_userdata, true);
+	}
+
+	/* PRIMARY (direct) LIGHT PASS */
+	{
+
+		Vector<RD::Uniform> uniforms;
+		{
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(light_source_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 1;
+				u.ids.push_back(light_dest_tex); //will be unused
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 2;
+				u.ids.push_back(position_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 3;
+				u.ids.push_back(normal_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 4;
+				u.ids.push_back(light_accum_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 5;
+				u.ids.push_back(light_primary_dynamic_tex);
+				uniforms.push_back(u);
+			}
+		}
+
+		RID light_uniform_set = rd->uniform_set_create(uniforms, compute_shader_primary, 1);
+
+		RD::ComputeListID compute_list = rd->compute_list_begin();
+		rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_primary_pipeline);
+		rd->compute_list_bind_uniform_set(compute_list, compute_base_uniform_set, 0);
+		rd->compute_list_bind_uniform_set(compute_list, light_uniform_set, 1);
+
+		for (int i = 0; i < atlas_slices; i++) {
+			push_constant.atlas_slice = i;
+			rd->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
+			rd->compute_list_dispatch(compute_list, group_size.x, group_size.y, group_size.z);
+			//no barrier, let them run all together
+		}
+		rd->compute_list_end(); //done
+	}
+
+#ifdef DEBUG_TEXTURES
+
+	for (int i = 0; i < atlas_slices; i++) {
+		Vector<uint8_t> s = rd->texture_get_data(light_source_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://2_light_primary_" + itos(i) + ".png");
+	}
+#endif
+
+	/* SECONDARY (indirect) LIGHT PASS(ES) */
+	if (p_step_function) {
+		p_step_function(0.6, TTR("Integrate indirect lighting"), p_bake_userdata, true);
+	}
+
+	if (p_bounces > 0) {
+
+		Vector<RD::Uniform> uniforms;
+		{
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(light_dest_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 1;
+				u.ids.push_back(light_source_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 2;
+				u.ids.push_back(position_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 3;
+				u.ids.push_back(normal_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 4;
+				u.ids.push_back(light_accum_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 5;
+				u.ids.push_back(unocclude_tex); //reuse unocclude tex
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 6;
+				u.ids.push_back(light_environment_tex); //reuse unocclude tex
+				uniforms.push_back(u);
+			}
+		}
+
+		RID secondary_uniform_set[2];
+		secondary_uniform_set[0] = rd->uniform_set_create(uniforms, compute_shader_secondary, 1);
+		uniforms.write[0].ids.write[0] = light_source_tex;
+		uniforms.write[1].ids.write[0] = light_dest_tex;
+		secondary_uniform_set[1] = rd->uniform_set_create(uniforms, compute_shader_secondary, 1);
+
+		switch (p_quality) {
+			case BAKE_QUALITY_LOW: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/low_quality_ray_count");
+			} break;
+			case BAKE_QUALITY_MEDIUM: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/medium_quality_ray_count");
+			} break;
+			case BAKE_QUALITY_HIGH: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/high_quality_ray_count");
+			} break;
+			case BAKE_QUALITY_ULTRA: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/ultra_quality_ray_count");
+			} break;
+		}
+
+		push_constant.ray_count = CLAMP(push_constant.ray_count, 16, 8192);
+
+		int max_region_size = nearest_power_of_2_templated(int(GLOBAL_GET("rendering/gpu_lightmapper/performance/region_size")));
+		int max_rays = GLOBAL_GET("rendering/gpu_lightmapper/performance/max_rays_per_pass");
+
+		int x_regions = (atlas_size.width - 1) / max_region_size + 1;
+		int y_regions = (atlas_size.height - 1) / max_region_size + 1;
+		int ray_iterations = (push_constant.ray_count - 1) / max_rays + 1;
+
+		rd->submit();
+		rd->sync();
+
+		for (int b = 0; b < p_bounces; b++) {
+			int count = 0;
+			if (b > 0) {
+				SWAP(light_source_tex, light_dest_tex);
+				SWAP(secondary_uniform_set[0], secondary_uniform_set[1]);
+			}
+
+			for (int s = 0; s < atlas_slices; s++) {
+				push_constant.atlas_slice = s;
+
+				for (int i = 0; i < x_regions; i++) {
+					for (int j = 0; j < y_regions; j++) {
+
+						int x = i * max_region_size;
+						int y = j * max_region_size;
+						int w = MIN((i + 1) * max_region_size, atlas_size.width) - x;
+						int h = MIN((j + 1) * max_region_size, atlas_size.height) - y;
+
+						push_constant.region_ofs[0] = x;
+						push_constant.region_ofs[1] = y;
+
+						group_size = Vector3i((w - 1) / 8 + 1, (h - 1) / 8 + 1, 1);
+
+						for (int k = 0; k < ray_iterations; k++) {
+
+							RD::ComputeListID compute_list = rd->compute_list_begin();
+							rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_secondary_pipeline);
+							rd->compute_list_bind_uniform_set(compute_list, compute_base_uniform_set, 0);
+							rd->compute_list_bind_uniform_set(compute_list, secondary_uniform_set[0], 1);
+
+							push_constant.ray_from = k * max_rays;
+							push_constant.ray_to = MIN((k + 1) * max_rays, int32_t(push_constant.ray_count));
+							rd->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
+							rd->compute_list_dispatch(compute_list, group_size.x, group_size.y, group_size.z);
+
+							rd->compute_list_end(); //done
+							rd->submit();
+							rd->sync();
+
+							count++;
+							if (p_step_function) {
+								int total = (atlas_slices * x_regions * y_regions * ray_iterations);
+								int percent = count * 100 / total;
+								float p = float(count) / total * 0.1;
+								p_step_function(0.6 + p, vformat(TTR("Bounce %d/%d: Integrate indirect lighting %d%%"), b + 1, p_bounces, percent), p_bake_userdata, false);
+							}
+						}
+					}
+				}
+			}
+		}
+	}
+
+	/* LIGHPROBES */
+
+	RID light_probe_buffer;
+
+	if (probe_positions.size()) {
+
+		light_probe_buffer = rd->storage_buffer_create(sizeof(float) * 4 * 9 * probe_positions.size());
+
+		if (p_step_function) {
+			p_step_function(0.7, TTR("Baking lightprobes"), p_bake_userdata, true);
+		}
+
+		Vector<RD::Uniform> uniforms;
+		{
+
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_STORAGE_BUFFER;
+				u.binding = 0;
+				u.ids.push_back(light_probe_buffer);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 1;
+				u.ids.push_back(light_dest_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 2;
+				u.ids.push_back(light_primary_dynamic_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 3;
+				u.ids.push_back(light_environment_tex);
+				uniforms.push_back(u);
+			}
+		}
+		RID light_probe_uniform_set = rd->uniform_set_create(uniforms, compute_shader_light_probes, 1);
+
+		switch (p_quality) {
+			case BAKE_QUALITY_LOW: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/low_quality_probe_ray_count");
+			} break;
+			case BAKE_QUALITY_MEDIUM: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/medium_quality_probe_ray_count");
+			} break;
+			case BAKE_QUALITY_HIGH: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/high_quality_probe_ray_count");
+			} break;
+			case BAKE_QUALITY_ULTRA: {
+				push_constant.ray_count = GLOBAL_GET("rendering/gpu_lightmapper/quality/ultra_quality_probe_ray_count");
+			} break;
+		}
+
+		push_constant.atlas_size[0] = probe_positions.size();
+		push_constant.ray_count = CLAMP(push_constant.ray_count, 16, 8192);
+
+		int max_rays = GLOBAL_GET("rendering/gpu_lightmapper/performance/max_rays_per_probe_pass");
+		int ray_iterations = (push_constant.ray_count - 1) / max_rays + 1;
+
+		for (int i = 0; i < ray_iterations; i++) {
+
+			RD::ComputeListID compute_list = rd->compute_list_begin();
+			rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_light_probes_pipeline);
+			rd->compute_list_bind_uniform_set(compute_list, compute_base_uniform_set, 0);
+			rd->compute_list_bind_uniform_set(compute_list, light_probe_uniform_set, 1);
+
+			push_constant.ray_from = i * max_rays;
+			push_constant.ray_to = MIN((i + 1) * max_rays, int32_t(push_constant.ray_count));
+			rd->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
+			rd->compute_list_dispatch(compute_list, (probe_positions.size() - 1) / 64 + 1, 1, 1);
+
+			rd->compute_list_end(); //done
+			rd->submit();
+			rd->sync();
+
+			if (p_step_function) {
+				int percent = i * 100 / ray_iterations;
+				float p = float(i) / ray_iterations * 0.1;
+				p_step_function(0.7 + p, vformat(TTR("Integrating light probes %d%%"), percent), p_bake_userdata, false);
+			}
+		}
+
+		push_constant.atlas_size[0] = atlas_size.x; //restore
+	}
+
+#if 0
+	for (int i = 0; i < probe_positions.size(); i++) {
+		Ref<Image> img;
+		img.instance();
+		img->create(6, 4, false, Image::FORMAT_RGB8);
+		for (int j = 0; j < 6; j++) {
+			Vector<uint8_t> s = rd->texture_get_data(lightprobe_tex, i * 6 + j);
+			Ref<Image> img2;
+			img2.instance();
+			img2->create(2, 2, false, Image::FORMAT_RGBAF, s);
+			img2->convert(Image::FORMAT_RGB8);
+			img->blit_rect(img2, Rect2(0, 0, 2, 2), Point2((j % 3) * 2, (j / 3) * 2));
+		}
+		img->save_png("res://3_light_probe_" + itos(i) + ".png");
+	}
+#endif
+
+	/* DENOISE */
+
+	if (p_use_denoiser) {
+		if (p_step_function) {
+			p_step_function(0.8, TTR("Denoising"), p_bake_userdata, true);
+		}
+
+		Ref<LightmapDenoiser> denoiser = LightmapDenoiser::create();
+		if (denoiser.is_valid()) {
+			for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+				Vector<uint8_t> s = rd->texture_get_data(light_accum_tex, i);
+				Ref<Image> img;
+				img.instance();
+				img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+
+				Ref<Image> denoised = denoiser->denoise_image(img);
+				if (denoised != img) {
+					denoised->convert(Image::FORMAT_RGBAH);
+					Vector<uint8_t> ds = denoised->get_data();
+					denoised.unref(); //avoid copy on write
+					{ //restore alpha
+						uint32_t count = s.size() / 2; //uint16s
+						const uint16_t *src = (const uint16_t *)s.ptr();
+						uint16_t *dst = (uint16_t *)ds.ptrw();
+						for (uint32_t j = 0; j < count; j += 4) {
+							dst[j + 3] = src[j + 3];
+						}
+					}
+					rd->texture_update(light_accum_tex, i, ds, true);
+				}
+			}
+		}
+	}
+
+#ifdef DEBUG_TEXTURES
+
+	for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+		Vector<uint8_t> s = rd->texture_get_data(light_accum_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://4_light_secondary_" + itos(i) + ".png");
+	}
+#endif
+
+	/* DILATE LIGHTMAP */
+	{
+
+		SWAP(light_accum_tex, light_accum_tex2);
+
+		Vector<RD::Uniform> uniforms;
+		{
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 0;
+				u.ids.push_back(light_accum_tex);
+				uniforms.push_back(u);
+			}
+			{
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 1;
+				u.ids.push_back(light_accum_tex2);
+				uniforms.push_back(u);
+			}
+		}
+
+		RID dilate_uniform_set = rd->uniform_set_create(uniforms, compute_shader_dilate, 1);
+
+		RD::ComputeListID compute_list = rd->compute_list_begin();
+		rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_dilate_pipeline);
+		rd->compute_list_bind_uniform_set(compute_list, compute_base_uniform_set, 0);
+		rd->compute_list_bind_uniform_set(compute_list, dilate_uniform_set, 1);
+		push_constant.region_ofs[0] = 0;
+		push_constant.region_ofs[1] = 0;
+		group_size = Vector3i((atlas_size.x - 1) / 8 + 1, (atlas_size.y - 1) / 8 + 1, 1); //restore group size
+
+		for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+			push_constant.atlas_slice = i;
+			rd->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
+			rd->compute_list_dispatch(compute_list, group_size.x, group_size.y, group_size.z);
+			//no barrier, let them run all together
+		}
+		rd->compute_list_end();
+	}
+
+#ifdef DEBUG_TEXTURES
+
+	for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+		Vector<uint8_t> s = rd->texture_get_data(light_accum_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://5_dilated_" + itos(i) + ".png");
+	}
+#endif
+
+	/* BLEND SEAMS */
+	//shaders
+	Ref<RDShaderFile> blendseams_shader;
+	blendseams_shader.instance();
+	err = blendseams_shader->parse_versions_from_text(lm_blendseams_shader_glsl);
+	if (err != OK) {
+		FREE_TEXTURES
+		FREE_BUFFERS
+		FREE_RASTER_RESOURCES
+		FREE_COMPUTE_RESOURCES
+		memdelete(rd);
+		blendseams_shader->print_errors("blendseams_shader");
+	}
+	ERR_FAIL_COND_V(err != OK, BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES);
+
+	RID blendseams_line_raster_shader = rd->shader_create_from_bytecode(blendseams_shader->get_bytecode("lines"));
+
+	ERR_FAIL_COND_V(blendseams_line_raster_shader.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES);
+
+	RID blendseams_triangle_raster_shader = rd->shader_create_from_bytecode(blendseams_shader->get_bytecode("triangles"));
+
+	ERR_FAIL_COND_V(blendseams_triangle_raster_shader.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES);
+
+#define FREE_BLENDSEAMS_RESOURCES            \
+	rd->free(blendseams_line_raster_shader); \
+	rd->free(blendseams_triangle_raster_shader);
+
+	{
+
+		//pre copy
+		for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+			rd->texture_copy(light_accum_tex, light_accum_tex2, Vector3(), Vector3(), Vector3(atlas_size.width, atlas_size.height, 1), 0, 0, i, i, true);
+		}
+
+		Vector<RID> framebuffers;
+		for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+			RID slice_tex = rd->texture_create_shared_from_slice(RD::TextureView(), light_accum_tex, i, 0);
+			Vector<RID> fb;
+			fb.push_back(slice_tex);
+			fb.push_back(raster_depth_buffer);
+			framebuffers.push_back(rd->framebuffer_create(fb));
+		}
+
+		Vector<RD::Uniform> uniforms;
+		{
+			{
+
+				RD::Uniform u;
+				u.type = RD::UNIFORM_TYPE_TEXTURE;
+				u.binding = 0;
+				u.ids.push_back(light_accum_tex2);
+				uniforms.push_back(u);
+			}
+		}
+
+		RID blendseams_raster_uniform = rd->uniform_set_create(uniforms, blendseams_line_raster_shader, 1);
+
+		bool debug = false;
+		RD::PipelineColorBlendState bs = RD::PipelineColorBlendState::create_blend(1);
+		bs.attachments.write[0].src_alpha_blend_factor = RD::BLEND_FACTOR_ZERO;
+		bs.attachments.write[0].dst_alpha_blend_factor = RD::BLEND_FACTOR_ONE;
+
+		RD::PipelineDepthStencilState ds;
+		ds.enable_depth_test = true;
+		ds.enable_depth_write = true;
+		ds.depth_compare_operator = RD::COMPARE_OP_LESS; //so it does not render same pixel twice, this avoids wrong blending
+
+		RID blendseams_line_raster_pipeline = rd->render_pipeline_create(blendseams_line_raster_shader, rd->framebuffer_get_format(framebuffers[0]), RD::INVALID_FORMAT_ID, RD::RENDER_PRIMITIVE_LINES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), ds, bs, 0);
+		RID blendseams_triangle_raster_pipeline = rd->render_pipeline_create(blendseams_triangle_raster_shader, rd->framebuffer_get_format(framebuffers[0]), RD::INVALID_FORMAT_ID, RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), ds, bs, 0);
+
+		uint32_t seam_offset = 0;
+		uint32_t triangle_offset = 0;
+
+		Vector<Color> clear_colors;
+		clear_colors.push_back(Color(0, 0, 0, 1));
+		for (int i = 0; i < atlas_slices; i++) {
+
+			int subslices = (p_bake_sh ? 4 : 1);
+			for (int k = 0; k < subslices; k++) {
+
+				RasterSeamsPushConstant seams_push_constant;
+				seams_push_constant.slice = uint32_t(i * subslices + k);
+				seams_push_constant.debug = debug;
+
+				RD::DrawListID draw_list = rd->draw_list_begin(framebuffers[i], RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_CLEAR, RD::FINAL_ACTION_DISCARD, clear_colors);
+
+				rd->draw_list_bind_uniform_set(draw_list, raster_base_uniform, 0);
+				rd->draw_list_bind_uniform_set(draw_list, blendseams_raster_uniform, 1);
+
+				const int uv_offset_count = 9;
+				static const Vector3 uv_offsets[uv_offset_count] = {
+					Vector3(0, 0, 0.5), //using zbuffer, so go inwards-outwards
+					Vector3(0, 1, 0.2),
+					Vector3(0, -1, 0.2),
+					Vector3(1, 0, 0.2),
+					Vector3(-1, 0, 0.2),
+					Vector3(-1, -1, 0.1),
+					Vector3(1, -1, 0.1),
+					Vector3(1, 1, 0.1),
+					Vector3(-1, 1, 0.1),
+				};
+
+				/* step 1 use lines to blend the edges */
+				{
+					seams_push_constant.base_index = seam_offset;
+					rd->draw_list_bind_render_pipeline(draw_list, blendseams_line_raster_pipeline);
+					seams_push_constant.uv_offset[0] = uv_offsets[0].x / float(atlas_size.width);
+					seams_push_constant.uv_offset[1] = uv_offsets[0].y / float(atlas_size.height);
+					seams_push_constant.blend = uv_offsets[0].z;
+
+					rd->draw_list_set_push_constant(draw_list, &seams_push_constant, sizeof(RasterSeamsPushConstant));
+					rd->draw_list_draw(draw_list, false, 1, slice_seam_count[i] * 4);
+				}
+
+				/* step 2 use triangles to mask the interior */
+
+				{
+					seams_push_constant.base_index = triangle_offset;
+					rd->draw_list_bind_render_pipeline(draw_list, blendseams_triangle_raster_pipeline);
+					seams_push_constant.blend = 0; //do not draw them, just fill the z-buffer so its used as a mask
+
+					rd->draw_list_set_push_constant(draw_list, &seams_push_constant, sizeof(RasterSeamsPushConstant));
+					rd->draw_list_draw(draw_list, false, 1, slice_triangle_count[i] * 3);
+				}
+				/* step 3 blend around the triangle */
+
+				rd->draw_list_bind_render_pipeline(draw_list, blendseams_line_raster_pipeline);
+
+				for (int j = 1; j < uv_offset_count; j++) {
+
+					seams_push_constant.base_index = seam_offset;
+					seams_push_constant.uv_offset[0] = uv_offsets[j].x / float(atlas_size.width);
+					seams_push_constant.uv_offset[1] = uv_offsets[j].y / float(atlas_size.height);
+					seams_push_constant.blend = uv_offsets[0].z;
+
+					rd->draw_list_set_push_constant(draw_list, &seams_push_constant, sizeof(RasterSeamsPushConstant));
+					rd->draw_list_draw(draw_list, false, 1, slice_seam_count[i] * 4);
+				}
+				rd->draw_list_end();
+			}
+			seam_offset += slice_seam_count[i];
+			triangle_offset += slice_triangle_count[i];
+		}
+	}
+
+#ifdef DEBUG_TEXTURES
+
+	for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+		Vector<uint8_t> s = rd->texture_get_data(light_accum_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBA8);
+		img->save_png("res://5_blendseams" + itos(i) + ".png");
+	}
+#endif
+	if (p_step_function) {
+		p_step_function(0.9, TTR("Retrieving textures"), p_bake_userdata, true);
+	}
+
+	for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
+		Vector<uint8_t> s = rd->texture_get_data(light_accum_tex, i);
+		Ref<Image> img;
+		img.instance();
+		img->create(atlas_size.width, atlas_size.height, false, Image::FORMAT_RGBAH, s);
+		img->convert(Image::FORMAT_RGBH); //remove alpha
+		bake_textures.push_back(img);
+	}
+
+	if (probe_positions.size() > 0) {
+		probe_values.resize(probe_positions.size() * 9);
+		Vector<uint8_t> probe_data = rd->buffer_get_data(light_probe_buffer);
+		copymem(probe_values.ptrw(), probe_data.ptr(), probe_data.size());
+		rd->free(light_probe_buffer);
+
+#ifdef DEBUG_TEXTURES
+		{
+			Ref<Image> img2;
+			img2.instance();
+			img2->create(probe_values.size(), 1, false, Image::FORMAT_RGBAF, probe_data);
+			img2->save_png("res://6_lightprobes.png");
+		}
+#endif
+	}
+
+	FREE_TEXTURES
+	FREE_BUFFERS
+	FREE_RASTER_RESOURCES
+	FREE_COMPUTE_RESOURCES
+	FREE_BLENDSEAMS_RESOURCES
+
+	memdelete(rd);
+
+	return BAKE_OK;
+}
+
+int LightmapperRD::get_bake_texture_count() const {
+	return bake_textures.size();
+}
+Ref<Image> LightmapperRD::get_bake_texture(int p_index) const {
+	ERR_FAIL_INDEX_V(p_index, bake_textures.size(), Ref<Image>());
+	return bake_textures[p_index];
+}
+int LightmapperRD::get_bake_mesh_count() const {
+	return mesh_instances.size();
+}
+Variant LightmapperRD::get_bake_mesh_userdata(int p_index) const {
+	ERR_FAIL_INDEX_V(p_index, mesh_instances.size(), Variant());
+	return mesh_instances[p_index].data.userdata;
+}
+Rect2 LightmapperRD::get_bake_mesh_uv_scale(int p_index) const {
+
+	ERR_FAIL_COND_V(bake_textures.size() == 0, Rect2());
+	Rect2 uv_ofs;
+	Vector2 atlas_size = Vector2(bake_textures[0]->get_width(), bake_textures[0]->get_height());
+	uv_ofs.position = Vector2(mesh_instances[p_index].offset) / atlas_size;
+	uv_ofs.size = Vector2(mesh_instances[p_index].data.albedo_on_uv2->get_width(), mesh_instances[p_index].data.albedo_on_uv2->get_height()) / atlas_size;
+	return uv_ofs;
+}
+int LightmapperRD::get_bake_mesh_texture_slice(int p_index) const {
+	ERR_FAIL_INDEX_V(p_index, mesh_instances.size(), Variant());
+	return mesh_instances[p_index].slice;
+}
+
+int LightmapperRD::get_bake_probe_count() const {
+	return probe_positions.size();
+}
+
+Vector3 LightmapperRD::get_bake_probe_point(int p_probe) const {
+	ERR_FAIL_INDEX_V(p_probe, probe_positions.size(), Variant());
+	return Vector3(probe_positions[p_probe].position[0], probe_positions[p_probe].position[1], probe_positions[p_probe].position[2]);
+}
+
+Vector<Color> LightmapperRD::get_bake_probe_sh(int p_probe) const {
+	ERR_FAIL_INDEX_V(p_probe, probe_positions.size(), Vector<Color>());
+	Vector<Color> ret;
+	ret.resize(9);
+	copymem(ret.ptrw(), &probe_values[p_probe * 9], sizeof(Color) * 9);
+	return ret;
+}
+
+LightmapperRD::LightmapperRD() {
+}
diff --git a/modules/lightmapper_rd/lightmapper_rd.h b/modules/lightmapper_rd/lightmapper_rd.h
new file mode 100644
index 0000000000..cb98efbeaa
--- /dev/null
+++ b/modules/lightmapper_rd/lightmapper_rd.h
@@ -0,0 +1,229 @@
+#ifndef LIGHTMAPPER_RD_H
+#define LIGHTMAPPER_RD_H
+
+#include "core/local_vector.h"
+#include "scene/3d/lightmapper.h"
+#include "scene/resources/mesh.h"
+#include "servers/rendering/rendering_device.h"
+
+class LightmapperRD : public Lightmapper {
+	GDCLASS(LightmapperRD, Lightmapper)
+
+	struct MeshInstance {
+		MeshData data;
+		int slice = 0;
+		Vector2i offset;
+	};
+
+	struct Light {
+		float position[3];
+		uint32_t type = LIGHT_TYPE_DIRECTIONAL;
+		float direction[3];
+		float energy;
+		float color[3];
+		float size;
+		float range;
+		float attenuation;
+		float spot_angle;
+		float spot_attenuation;
+		uint32_t static_bake;
+		uint32_t pad[3];
+
+		bool operator<(const Light &p_light) const {
+			return type < p_light.type;
+		}
+	};
+
+	struct Vertex {
+		float position[3];
+		float normal_z;
+		float uv[2];
+		float normal_xy[2];
+
+		bool operator==(const Vertex &p_vtx) const {
+			return (position[0] == p_vtx.position[0]) &&
+				   (position[1] == p_vtx.position[1]) &&
+				   (position[2] == p_vtx.position[2]) &&
+				   (uv[0] == p_vtx.uv[0]) &&
+				   (uv[1] == p_vtx.uv[1]) &&
+				   (normal_xy[0] == p_vtx.normal_xy[0]) &&
+				   (normal_xy[1] == p_vtx.normal_xy[1]) &&
+				   (normal_z == p_vtx.normal_z);
+		}
+	};
+
+	struct Edge {
+		Vector3 a;
+		Vector3 b;
+		Vector3 na;
+		Vector3 nb;
+		bool operator==(const Edge &p_seam) const {
+			return a == p_seam.a && b == p_seam.b && na == p_seam.na && nb == p_seam.nb;
+		}
+		Edge() {
+		}
+
+		Edge(const Vector3 &p_a, const Vector3 &p_b, const Vector3 &p_na, const Vector3 &p_nb) {
+			a = p_a;
+			b = p_b;
+			na = p_na;
+			nb = p_nb;
+		}
+	};
+
+	struct Probe {
+		float position[4];
+	};
+
+	Vector<Probe> probe_positions;
+
+	struct EdgeHash {
+		_FORCE_INLINE_ static uint32_t hash(const Edge &p_edge) {
+			uint32_t h = hash_djb2_one_float(p_edge.a.x);
+			h = hash_djb2_one_float(p_edge.a.y, h);
+			h = hash_djb2_one_float(p_edge.a.z, h);
+			h = hash_djb2_one_float(p_edge.b.x, h);
+			h = hash_djb2_one_float(p_edge.b.y, h);
+			h = hash_djb2_one_float(p_edge.b.z, h);
+			return h;
+		}
+	};
+	struct EdgeUV2 {
+		Vector2 a;
+		Vector2 b;
+		Vector2i indices;
+		bool operator==(const EdgeUV2 &p_uv2) const {
+			return a == p_uv2.a && b == p_uv2.b;
+		}
+		bool seam_found = false;
+		EdgeUV2(Vector2 p_a, Vector2 p_b, Vector2i p_indices) {
+			a = p_a;
+			b = p_b;
+			indices = p_indices;
+		}
+		EdgeUV2() {}
+	};
+
+	struct Seam {
+		Vector2i a;
+		Vector2i b;
+		uint32_t slice;
+		bool operator<(const Seam &p_seam) const {
+			return slice < p_seam.slice;
+		}
+	};
+
+	struct VertexHash {
+		_FORCE_INLINE_ static uint32_t hash(const Vertex &p_vtx) {
+			uint32_t h = hash_djb2_one_float(p_vtx.position[0]);
+			h = hash_djb2_one_float(p_vtx.position[1], h);
+			h = hash_djb2_one_float(p_vtx.position[2], h);
+			h = hash_djb2_one_float(p_vtx.uv[0], h);
+			h = hash_djb2_one_float(p_vtx.uv[1], h);
+			h = hash_djb2_one_float(p_vtx.normal_xy[0], h);
+			h = hash_djb2_one_float(p_vtx.normal_xy[1], h);
+			h = hash_djb2_one_float(p_vtx.normal_z, h);
+			return h;
+		}
+	};
+
+	struct Box {
+		float min_bounds[3];
+		float pad0;
+		float max_bounds[3];
+		float pad1;
+	};
+
+	struct Triangle {
+		uint32_t indices[3];
+		uint32_t slice;
+		bool operator<(const Triangle &p_triangle) const {
+			return slice < p_triangle.slice;
+		}
+	};
+
+	Vector<MeshInstance> mesh_instances;
+
+	Vector<Light> lights;
+
+	struct TriangleSort {
+		uint32_t cell_index;
+		uint32_t triangle_index;
+		bool operator<(const TriangleSort &p_triangle_sort) const {
+			return cell_index < p_triangle_sort.cell_index; //sorting by triangle index in this case makes no sense
+		}
+	};
+
+	void _plot_triangle_into_triangle_index_list(int p_size, const Vector3i &p_ofs, const AABB &p_bounds, const Vector3 p_points[], uint32_t p_triangle_index, LocalVector<TriangleSort> &triangles, uint32_t p_grid_size);
+
+	struct RasterPushConstant {
+		float atlas_size[2];
+		float uv_offset[2];
+		float to_cell_size[3];
+		uint32_t base_triangle;
+		float to_cell_offset[3];
+		float bias;
+		int32_t grid_size[3];
+		uint32_t pad2;
+	};
+
+	struct RasterSeamsPushConstant {
+
+		uint32_t base_index;
+		uint32_t slice;
+		float uv_offset[2];
+		uint32_t debug;
+		float blend;
+		uint32_t pad[2];
+	};
+
+	struct PushConstant {
+		int32_t atlas_size[2];
+		uint32_t ray_count;
+		uint32_t ray_to;
+
+		float world_size[3];
+		float bias;
+
+		float to_cell_offset[3];
+		uint32_t ray_from;
+
+		float to_cell_size[3];
+		uint32_t light_count;
+
+		int32_t grid_size;
+		int32_t atlas_slice;
+		int32_t region_ofs[2];
+
+		float environment_xform[12];
+	};
+
+	Vector<Ref<Image>> bake_textures;
+	Vector<Color> probe_values;
+
+	BakeError _blit_meshes_into_atlas(int p_max_texture_size, Vector<Ref<Image>> &albedo_images, Vector<Ref<Image>> &emission_images, AABB &bounds, Size2i &atlas_size, int &atlas_slices, BakeStepFunc p_step_function, void *p_bake_userdata);
+	void _create_acceleration_structures(RenderingDevice *rd, Size2i atlas_size, int atlas_slices, AABB &bounds, int grid_size, Vector<Probe> &probe_positions, GenerateProbes p_generate_probes, Vector<int> &slice_triangle_count, Vector<int> &slice_seam_count, RID &vertex_buffer, RID &triangle_buffer, RID &box_buffer, RID &lights_buffer, RID &triangle_cell_indices_buffer, RID &probe_positions_buffer, RID &grid_texture, RID &grid_texture_sdf, RID &seams_buffer, BakeStepFunc p_step_function, void *p_bake_userdata);
+	void _raster_geometry(RenderingDevice *rd, Size2i atlas_size, int atlas_slices, int grid_size, AABB bounds, float p_bias, Vector<int> slice_triangle_count, RID position_tex, RID unocclude_tex, RID normal_tex, RID raster_depth_buffer, RID rasterize_shader, RID raster_base_uniform);
+
+public:
+	virtual void add_mesh(const MeshData &p_mesh);
+	virtual void add_directional_light(bool p_static, const Vector3 &p_direction, const Color &p_color, float p_energy, float p_angular_distance);
+	virtual void add_omni_light(bool p_static, const Vector3 &p_position, const Color &p_color, float p_energy, float p_range, float p_attenuation, float p_size);
+	virtual void add_spot_light(bool p_static, const Vector3 &p_position, const Vector3 p_direction, const Color &p_color, float p_energy, float p_range, float p_attenuation, float p_spot_angle, float p_spot_attenuation, float p_size);
+	virtual void add_probe(const Vector3 &p_position);
+	virtual BakeError bake(BakeQuality p_quality, bool p_use_denoiser, int p_bounces, float p_bias, int p_max_texture_size, bool p_bake_sh, GenerateProbes p_generate_probes, const Ref<Image> &p_environment_panorama, const Basis &p_environment_transform, BakeStepFunc p_step_function = nullptr, void *p_bake_userdata = nullptr);
+
+	int get_bake_texture_count() const;
+	Ref<Image> get_bake_texture(int p_index) const;
+	int get_bake_mesh_count() const;
+	Variant get_bake_mesh_userdata(int p_index) const;
+	Rect2 get_bake_mesh_uv_scale(int p_index) const;
+	int get_bake_mesh_texture_slice(int p_index) const;
+	int get_bake_probe_count() const;
+	Vector3 get_bake_probe_point(int p_probe) const;
+	Vector<Color> get_bake_probe_sh(int p_probe) const;
+
+	LightmapperRD();
+};
+
+#endif // LIGHTMAPPER_H
diff --git a/modules/lightmapper_rd/lm_blendseams.glsl b/modules/lightmapper_rd/lm_blendseams.glsl
new file mode 100644
index 0000000000..ef1ece8ea1
--- /dev/null
+++ b/modules/lightmapper_rd/lm_blendseams.glsl
@@ -0,0 +1,117 @@
+/* clang-format off */
+[versions]
+
+lines = "#define MODE_LINES"
+triangles = "#define MODE_TRIANGLES"
+
+[vertex]
+
+#version 450
+
+VERSION_DEFINES
+
+#include "lm_common_inc.glsl"
+
+				/* clang-format on */
+
+				layout(push_constant, binding = 0, std430) uniform Params {
+					uint base_index;
+					uint slice;
+					vec2 uv_offset;
+					bool debug;
+					float blend;
+					uint pad[2];
+				} params;
+
+layout(location = 0) out vec3 uv_interp;
+
+void main() {
+
+#ifdef MODE_TRIANGLES
+
+	uint triangle_idx = params.base_index + gl_VertexIndex / 3;
+	uint triangle_subidx = gl_VertexIndex % 3;
+
+	vec2 uv;
+	if (triangle_subidx == 0) {
+		uv = vertices.data[triangles.data[triangle_idx].indices.x].uv;
+	} else if (triangle_subidx == 1) {
+		uv = vertices.data[triangles.data[triangle_idx].indices.y].uv;
+	} else {
+		uv = vertices.data[triangles.data[triangle_idx].indices.z].uv;
+	}
+
+	uv_interp = vec3(uv, float(params.slice));
+	gl_Position = vec4((uv + params.uv_offset) * 2.0 - 1.0, 0.0001, 1.0);
+
+#endif
+
+#ifdef MODE_LINES
+	uint seam_idx = params.base_index + gl_VertexIndex / 4;
+	uint seam_subidx = gl_VertexIndex % 4;
+
+	uint src_idx;
+	uint dst_idx;
+
+	if (seam_subidx == 0) {
+		src_idx = seams.data[seam_idx].b.x;
+		dst_idx = seams.data[seam_idx].a.x;
+	} else if (seam_subidx == 1) {
+		src_idx = seams.data[seam_idx].b.y;
+		dst_idx = seams.data[seam_idx].a.y;
+	} else if (seam_subidx == 2) {
+		src_idx = seams.data[seam_idx].a.x;
+		dst_idx = seams.data[seam_idx].b.x;
+	} else if (seam_subidx == 3) {
+		src_idx = seams.data[seam_idx].a.y;
+		dst_idx = seams.data[seam_idx].b.y;
+	}
+
+	vec2 src_uv = vertices.data[src_idx].uv;
+	vec2 dst_uv = vertices.data[dst_idx].uv + params.uv_offset;
+
+	uv_interp = vec3(src_uv, float(params.slice));
+	gl_Position = vec4(dst_uv * 2.0 - 1.0, 0.0001, 1.0);
+	;
+#endif
+}
+
+/* clang-format off */
+[fragment]
+
+#version 450
+
+VERSION_DEFINES
+
+#include "lm_common_inc.glsl"
+
+				/* clang-format on */
+
+				layout(push_constant, binding = 0, std430) uniform Params {
+					uint base_index;
+					uint slice;
+					vec2 uv_offset;
+					bool debug;
+					float blend;
+					uint pad[2];
+				} params;
+
+layout(location = 0) in vec3 uv_interp;
+
+layout(location = 0) out vec4 dst_color;
+
+layout(set = 1, binding = 0) uniform texture2DArray src_color_tex;
+
+void main() {
+
+	if (params.debug) {
+#ifdef MODE_TRIANGLES
+		dst_color = vec4(1, 0, 1, 1);
+#else
+		dst_color = vec4(1, 1, 0, 1);
+#endif
+	} else {
+		vec4 src_color = textureLod(sampler2DArray(src_color_tex, linear_sampler), uv_interp, 0.0);
+		dst_color = vec4(src_color.rgb, params.blend); //mix
+	}
+}
diff --git a/modules/lightmapper_rd/lm_common_inc.glsl b/modules/lightmapper_rd/lm_common_inc.glsl
new file mode 100644
index 0000000000..0ff455936e
--- /dev/null
+++ b/modules/lightmapper_rd/lm_common_inc.glsl
@@ -0,0 +1,92 @@
+
+/* SET 0, static data that does not change between any call */
+
+struct Vertex {
+	vec3 position;
+	float normal_z;
+	vec2 uv;
+	vec2 normal_xy;
+};
+
+layout(set = 0, binding = 1, std430) restrict readonly buffer Vertices {
+	Vertex data[];
+}
+vertices;
+
+struct Triangle {
+	uvec3 indices;
+	uint slice;
+};
+
+layout(set = 0, binding = 2, std430) restrict readonly buffer Triangles {
+	Triangle data[];
+}
+triangles;
+
+struct Box {
+	vec3 min_bounds;
+	uint pad0;
+	vec3 max_bounds;
+	uint pad1;
+};
+
+layout(set = 0, binding = 3, std430) restrict readonly buffer Boxes {
+	Box data[];
+}
+boxes;
+
+layout(set = 0, binding = 4, std430) restrict readonly buffer GridIndices {
+	uint data[];
+}
+grid_indices;
+
+#define LIGHT_TYPE_DIRECTIONAL 0
+#define LIGHT_TYPE_OMNI 1
+#define LIGHT_TYPE_SPOT 2
+
+struct Light {
+	vec3 position;
+	uint type;
+
+	vec3 direction;
+	float energy;
+
+	vec3 color;
+	float size;
+
+	float range;
+	float attenuation;
+	float spot_angle;
+	float spot_attenuation;
+
+	bool static_bake;
+	uint pad[3];
+};
+
+layout(set = 0, binding = 5, std430) restrict readonly buffer Lights {
+	Light data[];
+}
+lights;
+
+struct Seam {
+	uvec2 a;
+	uvec2 b;
+};
+
+layout(set = 0, binding = 6, std430) restrict readonly buffer Seams {
+	Seam data[];
+}
+seams;
+
+layout(set = 0, binding = 7, std430) restrict readonly buffer Probes {
+	vec4 data[];
+}
+probe_positions;
+
+layout(set = 0, binding = 8) uniform utexture3D grid;
+layout(set = 0, binding = 9) uniform texture3D grid_sdf;
+
+layout(set = 0, binding = 10) uniform texture2DArray albedo_tex;
+layout(set = 0, binding = 11) uniform texture2DArray emission_tex;
+
+layout(set = 0, binding = 12) uniform sampler linear_sampler;
diff --git a/modules/lightmapper_rd/lm_compute.glsl b/modules/lightmapper_rd/lm_compute.glsl
new file mode 100644
index 0000000000..a178bd9b2e
--- /dev/null
+++ b/modules/lightmapper_rd/lm_compute.glsl
@@ -0,0 +1,657 @@
+/* clang-format off */
+[versions]
+
+primary = "#define MODE_DIRECT_LIGHT"
+secondary = "#define MODE_BOUNCE_LIGHT"
+dilate = "#define MODE_DILATE"
+unocclude = "#define MODE_UNOCCLUDE"
+light_probes = "#define MODE_LIGHT_PROBES"
+
+[compute]
+
+#version 450
+
+VERSION_DEFINES
+
+// One 2D local group focusing in one layer at a time, though all
+// in parallel (no barriers) makes more sense than a 3D local group
+// as this can take more advantage of the cache for each group.
+
+#ifdef MODE_LIGHT_PROBES
+
+layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
+
+#else
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+#endif
+
+#include "lm_common_inc.glsl"
+
+/* clang-format on */
+
+#ifdef MODE_LIGHT_PROBES
+
+layout(set = 1, binding = 0, std430) restrict buffer LightProbeData {
+	vec4 data[];
+}
+light_probes;
+
+layout(set = 1, binding = 1) uniform texture2DArray source_light;
+layout(set = 1, binding = 2) uniform texture2DArray source_direct_light; //also need the direct light, which was omitted
+layout(set = 1, binding = 3) uniform texture2D environment;
+#endif
+
+#ifdef MODE_UNOCCLUDE
+
+layout(rgba32f, set = 1, binding = 0) uniform restrict image2DArray position;
+layout(rgba32f, set = 1, binding = 1) uniform restrict readonly image2DArray unocclude;
+
+#endif
+
+#if defined(MODE_DIRECT_LIGHT) || defined(MODE_BOUNCE_LIGHT)
+
+layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_light;
+layout(set = 1, binding = 1) uniform texture2DArray source_light;
+layout(set = 1, binding = 2) uniform texture2DArray source_position;
+layout(set = 1, binding = 3) uniform texture2DArray source_normal;
+layout(rgba16f, set = 1, binding = 4) uniform restrict image2DArray accum_light;
+
+#endif
+
+#ifdef MODE_BOUNCE_LIGHT
+layout(rgba32f, set = 1, binding = 5) uniform restrict image2DArray bounce_accum;
+layout(set = 1, binding = 6) uniform texture2D environment;
+#endif
+#ifdef MODE_DIRECT_LIGHT
+layout(rgba32f, set = 1, binding = 5) uniform restrict writeonly image2DArray primary_dynamic;
+#endif
+
+#ifdef MODE_DILATE
+layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_light;
+layout(set = 1, binding = 1) uniform texture2DArray source_light;
+#endif
+
+layout(push_constant, binding = 0, std430) uniform Params {
+	ivec2 atlas_size; // x used for light probe mode total probes
+	uint ray_count;
+	uint ray_to;
+
+	vec3 world_size;
+	float bias;
+
+	vec3 to_cell_offset;
+	uint ray_from;
+
+	vec3 to_cell_size;
+	uint light_count;
+
+	int grid_size;
+	int atlas_slice;
+	ivec2 region_ofs;
+
+	mat3x4 env_transform;
+}
+params;
+
+//check it, but also return distance and barycentric coords (for uv lookup)
+bool ray_hits_triangle(vec3 from, vec3 dir, float max_dist, vec3 p0, vec3 p1, vec3 p2, out float r_distance, out vec3 r_barycentric) {
+
+	const vec3 e0 = p1 - p0;
+	const vec3 e1 = p0 - p2;
+	vec3 triangleNormal = cross(e1, e0);
+
+	const vec3 e2 = (1.0 / dot(triangleNormal, dir)) * (p0 - from);
+	const vec3 i = cross(dir, e2);
+
+	r_barycentric.y = dot(i, e1);
+	r_barycentric.z = dot(i, e0);
+	r_barycentric.x = 1.0 - (r_barycentric.z + r_barycentric.y);
+	r_distance = dot(triangleNormal, e2);
+	return (r_distance > params.bias) && (r_distance < max_dist) && all(greaterThanEqual(r_barycentric, vec3(0.0)));
+}
+
+bool trace_ray(vec3 p_from, vec3 p_to
+#if defined(MODE_BOUNCE_LIGHT) || defined(MODE_LIGHT_PROBES)
+		,
+		out uint r_triangle, out vec3 r_barycentric
+#endif
+#if defined(MODE_UNOCCLUDE)
+		,
+		out float r_distance, out vec3 r_normal
+#endif
+) {
+
+	/* world coords */
+
+	vec3 rel = p_to - p_from;
+	float rel_len = length(rel);
+	vec3 dir = normalize(rel);
+	vec3 inv_dir = 1.0 / dir;
+
+	/* cell coords */
+
+	vec3 from_cell = (p_from - params.to_cell_offset) * params.to_cell_size;
+	vec3 to_cell = (p_to - params.to_cell_offset) * params.to_cell_size;
+
+	//prepare DDA
+	vec3 rel_cell = to_cell - from_cell;
+	ivec3 icell = ivec3(from_cell);
+	ivec3 iendcell = ivec3(to_cell);
+	vec3 dir_cell = normalize(rel_cell);
+	vec3 delta = abs(1.0 / dir_cell); //vec3(length(rel_cell)) / rel_cell);
+	ivec3 step = ivec3(sign(rel_cell));
+	vec3 side = (sign(rel_cell) * (vec3(icell) - from_cell) + (sign(rel_cell) * 0.5) + 0.5) * delta;
+
+	uint iters = 0;
+	while (all(greaterThanEqual(icell, ivec3(0))) && all(lessThan(icell, ivec3(params.grid_size))) && iters < 1000) {
+
+		uvec2 cell_data = texelFetch(usampler3D(grid, linear_sampler), icell, 0).xy;
+		if (cell_data.x > 0) { //triangles here
+
+			bool hit = false;
+#if defined(MODE_UNOCCLUDE)
+			bool hit_backface = false;
+#endif
+			float best_distance = 1e20;
+
+			for (uint i = 0; i < cell_data.x; i++) {
+				uint tidx = grid_indices.data[cell_data.y + i];
+
+				//Ray-Box test
+				vec3 t0 = (boxes.data[tidx].min_bounds - p_from) * inv_dir;
+				vec3 t1 = (boxes.data[tidx].max_bounds - p_from) * inv_dir;
+				vec3 tmin = min(t0, t1), tmax = max(t0, t1);
+
+				if (max(tmin.x, max(tmin.y, tmin.z)) <= min(tmax.x, min(tmax.y, tmax.z))) {
+					continue; //ray box failed
+				}
+
+				//prepare triangle vertices
+				vec3 vtx0 = vertices.data[triangles.data[tidx].indices.x].position;
+				vec3 vtx1 = vertices.data[triangles.data[tidx].indices.y].position;
+				vec3 vtx2 = vertices.data[triangles.data[tidx].indices.z].position;
+#if defined(MODE_UNOCCLUDE)
+				vec3 normal = -normalize(cross((vtx0 - vtx1), (vtx0 - vtx2)));
+
+				bool backface = dot(normal, dir) >= 0.0;
+#endif
+				float distance;
+				vec3 barycentric;
+
+				if (ray_hits_triangle(p_from, dir, rel_len, vtx0, vtx1, vtx2, distance, barycentric)) {
+#ifdef MODE_DIRECT_LIGHT
+					return true; //any hit good
+#endif
+
+#if defined(MODE_UNOCCLUDE)
+					if (!backface) {
+						// the case of meshes having both a front and back face in the same plane is more common than
+						// expected, so if this is a front-face, bias it closer to the ray origin, so it always wins over the back-face
+						distance = max(params.bias, distance - params.bias);
+					}
+
+					hit = true;
+
+					if (distance < best_distance) {
+						hit_backface = backface;
+						best_distance = distance;
+						r_distance = distance;
+						r_normal = normal;
+					}
+
+#endif
+
+#if defined(MODE_BOUNCE_LIGHT) || defined(MODE_LIGHT_PROBES)
+
+					hit = true;
+					if (distance < best_distance) {
+						best_distance = distance;
+						r_triangle = tidx;
+						r_barycentric = barycentric;
+					}
+
+#endif
+				}
+			}
+#if defined(MODE_UNOCCLUDE)
+
+			if (hit) {
+				return hit_backface;
+			}
+#endif
+#if defined(MODE_BOUNCE_LIGHT) || defined(MODE_LIGHT_PROBES)
+			if (hit) {
+				return true;
+			}
+#endif
+		}
+
+		if (icell == iendcell) {
+			break;
+		}
+
+		bvec3 mask = lessThanEqual(side.xyz, min(side.yzx, side.zxy));
+		side += vec3(mask) * delta;
+		icell += ivec3(vec3(mask)) * step;
+
+		iters++;
+	}
+
+	return false;
+}
+
+const float PI = 3.14159265f;
+const float GOLDEN_ANGLE = PI * (3.0 - sqrt(5.0));
+
+vec3 vogel_hemisphere(uint p_index, uint p_count, float p_offset) {
+	float r = sqrt(float(p_index) + 0.5f) / sqrt(float(p_count));
+	float theta = float(p_index) * GOLDEN_ANGLE + p_offset;
+	float y = cos(r * PI * 0.5);
+	float l = sin(r * PI * 0.5);
+	return vec3(l * cos(theta), l * sin(theta), y);
+}
+
+float quick_hash(vec2 pos) {
+	return fract(sin(dot(pos * 19.19, vec2(49.5791, 97.413))) * 49831.189237);
+}
+
+void main() {
+
+#ifdef MODE_LIGHT_PROBES
+	int probe_index = int(gl_GlobalInvocationID.x);
+	if (probe_index >= params.atlas_size.x) { //too large, do nothing
+		return;
+	}
+
+#else
+	ivec2 atlas_pos = ivec2(gl_GlobalInvocationID.xy) + params.region_ofs;
+	if (any(greaterThanEqual(atlas_pos, params.atlas_size))) { //too large, do nothing
+		return;
+	}
+#endif
+
+#ifdef MODE_DIRECT_LIGHT
+
+	vec3 normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
+	if (length(normal) < 0.5) {
+		return; //empty texel, no process
+	}
+	vec3 position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
+
+	//go through all lights
+	//start by own light (emissive)
+	vec3 static_light = vec3(0.0);
+	vec3 dynamic_light = vec3(0.0);
+
+#ifdef USE_SH_LIGHTMAPS
+	vec4 sh_accum[4] = vec4[](
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0));
+#endif
+
+	for (uint i = 0; i < params.light_count; i++) {
+
+		vec3 light_pos;
+		float attenuation;
+		if (lights.data[i].type == LIGHT_TYPE_DIRECTIONAL) {
+			vec3 light_vec = lights.data[i].direction;
+			light_pos = position - light_vec * length(params.world_size);
+			attenuation = 1.0;
+		} else {
+			light_pos = lights.data[i].position;
+			float d = distance(position, light_pos);
+			if (d > lights.data[i].range) {
+				continue;
+			}
+
+			d /= lights.data[i].range;
+
+			attenuation = pow(max(1.0 - d, 0.0), lights.data[i].attenuation);
+
+			if (lights.data[i].type == LIGHT_TYPE_SPOT) {
+
+				vec3 rel = normalize(position - light_pos);
+				float angle = acos(dot(rel, lights.data[i].direction));
+				if (angle > lights.data[i].spot_angle) {
+					continue; //invisible, dont try
+				}
+
+				float d = clamp(angle / lights.data[i].spot_angle, 0, 1);
+				attenuation *= pow(1.0 - d, lights.data[i].spot_attenuation);
+			}
+		}
+
+		vec3 light_dir = normalize(light_pos - position);
+		attenuation *= max(0.0, dot(normal, light_dir));
+
+		if (attenuation <= 0.0001) {
+			continue; //no need to do anything
+		}
+
+		if (!trace_ray(position + light_dir * params.bias, light_pos)) {
+			vec3 light = lights.data[i].color * lights.data[i].energy * attenuation;
+			if (lights.data[i].static_bake) {
+				static_light += light;
+#ifdef USE_SH_LIGHTMAPS
+
+				float c[4] = float[](
+						0.282095, //l0
+						0.488603 * light_dir.y, //l1n1
+						0.488603 * light_dir.z, //l1n0
+						0.488603 * light_dir.x //l1p1
+				);
+
+				for (uint j = 0; j < 4; j++) {
+					sh_accum[j].rgb += light * c[j] * (1.0 / 3.0);
+				}
+#endif
+
+			} else {
+				dynamic_light += light;
+			}
+		}
+	}
+
+	vec3 albedo = texelFetch(sampler2DArray(albedo_tex, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).rgb;
+	vec3 emissive = texelFetch(sampler2DArray(emission_tex, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).rgb;
+
+	dynamic_light *= albedo; //if it will bounce, must multiply by albedo
+	dynamic_light += emissive;
+
+	//keep for lightprobes
+	imageStore(primary_dynamic, ivec3(atlas_pos, params.atlas_slice), vec4(dynamic_light, 1.0));
+
+	dynamic_light += static_light * albedo; //send for bounces
+	imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice), vec4(dynamic_light, 1.0));
+
+#ifdef USE_SH_LIGHTMAPS
+	//keep for adding at the end
+	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 0), sh_accum[0]);
+	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 1), sh_accum[1]);
+	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 2), sh_accum[2]);
+	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + 3), sh_accum[3]);
+
+#else
+	imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice), vec4(static_light, 1.0));
+#endif
+
+#endif
+
+#ifdef MODE_BOUNCE_LIGHT
+
+	vec3 normal = texelFetch(sampler2DArray(source_normal, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
+	if (length(normal) < 0.5) {
+		return; //empty texel, no process
+	}
+
+	vec3 position = texelFetch(sampler2DArray(source_position, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0).xyz;
+
+	vec3 v0 = abs(normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
+	vec3 tangent = normalize(cross(v0, normal));
+	vec3 bitangent = normalize(cross(tangent, normal));
+	mat3 normal_mat = mat3(tangent, bitangent, normal);
+
+#ifdef USE_SH_LIGHTMAPS
+	vec4 sh_accum[4] = vec4[](
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0),
+			vec4(0.0, 0.0, 0.0, 1.0));
+#endif
+	vec3 light_average = vec3(0.0);
+	for (uint i = params.ray_from; i < params.ray_to; i++) {
+		vec3 ray_dir = normal_mat * vogel_hemisphere(i, params.ray_count, quick_hash(vec2(atlas_pos)));
+
+		uint tidx;
+		vec3 barycentric;
+
+		vec3 light;
+		if (trace_ray(position + ray_dir * params.bias, position + ray_dir * length(params.world_size), tidx, barycentric)) {
+			//hit a triangle
+			vec2 uv0 = vertices.data[triangles.data[tidx].indices.x].uv;
+			vec2 uv1 = vertices.data[triangles.data[tidx].indices.y].uv;
+			vec2 uv2 = vertices.data[triangles.data[tidx].indices.z].uv;
+			vec3 uvw = vec3(barycentric.x * uv0 + barycentric.y * uv1 + barycentric.z * uv2, float(triangles.data[tidx].slice));
+
+			light = textureLod(sampler2DArray(source_light, linear_sampler), uvw, 0.0).rgb;
+		} else {
+			//did not hit a triangle, reach out for the sky
+			vec3 sky_dir = normalize(mat3(params.env_transform) * ray_dir);
+
+			vec2 st = vec2(
+					atan(sky_dir.x, sky_dir.z),
+					acos(sky_dir.y));
+
+			if (st.x < 0.0)
+				st.x += PI * 2.0;
+
+			st /= vec2(PI * 2.0, PI);
+
+			light = textureLod(sampler2D(environment, linear_sampler), st, 0.0).rgb;
+		}
+
+		light_average += light;
+
+#ifdef USE_SH_LIGHTMAPS
+
+		float c[4] = float[](
+				0.282095, //l0
+				0.488603 * ray_dir.y, //l1n1
+				0.488603 * ray_dir.z, //l1n0
+				0.488603 * ray_dir.x //l1p1
+		);
+
+		for (uint j = 0; j < 4; j++) {
+			sh_accum[j].rgb += light * c[j] * (8.0 / float(params.ray_count));
+		}
+#endif
+	}
+
+	vec3 light_total;
+	if (params.ray_from == 0) {
+		light_total = vec3(0.0);
+	} else {
+		light_total = imageLoad(bounce_accum, ivec3(atlas_pos, params.atlas_slice)).rgb;
+	}
+
+	light_total += light_average;
+
+#ifdef USE_SH_LIGHTMAPS
+
+	for (int i = 0; i < 4; i++) {
+		vec4 accum = imageLoad(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + i));
+		accum.rgb += sh_accum[i].rgb;
+		imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice * 4 + i), accum);
+	}
+
+#endif
+	if (params.ray_to == params.ray_count) {
+		light_total /= float(params.ray_count);
+		imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice), vec4(light_total, 1.0));
+#ifndef USE_SH_LIGHTMAPS
+		vec4 accum = imageLoad(accum_light, ivec3(atlas_pos, params.atlas_slice));
+		accum.rgb += light_total;
+		imageStore(accum_light, ivec3(atlas_pos, params.atlas_slice), accum);
+#endif
+	} else {
+		imageStore(bounce_accum, ivec3(atlas_pos, params.atlas_slice), vec4(light_total, 1.0));
+	}
+
+#endif
+
+#ifdef MODE_UNOCCLUDE
+
+	//texel_size = 0.5;
+	//compute tangents
+
+	vec4 position_alpha = imageLoad(position, ivec3(atlas_pos, params.atlas_slice));
+	if (position_alpha.a < 0.5) {
+		return;
+	}
+
+	vec3 vertex_pos = position_alpha.xyz;
+	vec4 normal_tsize = imageLoad(unocclude, ivec3(atlas_pos, params.atlas_slice));
+
+	vec3 face_normal = normal_tsize.xyz;
+	float texel_size = normal_tsize.w;
+
+	vec3 v0 = abs(face_normal.z) < 0.999 ? vec3(0.0, 0.0, 1.0) : vec3(0.0, 1.0, 0.0);
+	vec3 tangent = normalize(cross(v0, face_normal));
+	vec3 bitangent = normalize(cross(tangent, face_normal));
+	vec3 base_pos = vertex_pos + face_normal * params.bias; //raise a bit
+
+	vec3 rays[4] = vec3[](tangent, bitangent, -tangent, -bitangent);
+	float min_d = 1e20;
+	for (int i = 0; i < 4; i++) {
+		vec3 ray_to = base_pos + rays[i] * texel_size;
+		float d;
+		vec3 norm;
+
+		if (trace_ray(base_pos, ray_to, d, norm)) {
+
+			if (d < min_d) {
+				vertex_pos = base_pos + rays[i] * d + norm * params.bias * 10.0; //this bias needs to be greater than the regular bias, because otherwise later, rays will go the other side when pointing back.
+				min_d = d;
+			}
+		}
+	}
+
+	position_alpha.xyz = vertex_pos;
+
+	imageStore(position, ivec3(atlas_pos, params.atlas_slice), position_alpha);
+
+#endif
+
+#ifdef MODE_LIGHT_PROBES
+
+	vec3 position = probe_positions.data[probe_index].xyz;
+
+	vec4 probe_sh_accum[9] = vec4[](
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0),
+			vec4(0.0));
+
+	for (uint i = params.ray_from; i < params.ray_to; i++) {
+		vec3 ray_dir = vogel_hemisphere(i, params.ray_count, quick_hash(vec2(float(probe_index), 0.0)));
+		if (bool(i & 1)) {
+			//throw to both sides, so alternate them
+			ray_dir.z *= -1.0;
+		}
+
+		uint tidx;
+		vec3 barycentric;
+		vec3 light;
+
+		if (trace_ray(position + ray_dir * params.bias, position + ray_dir * length(params.world_size), tidx, barycentric)) {
+			vec2 uv0 = vertices.data[triangles.data[tidx].indices.x].uv;
+			vec2 uv1 = vertices.data[triangles.data[tidx].indices.y].uv;
+			vec2 uv2 = vertices.data[triangles.data[tidx].indices.z].uv;
+			vec3 uvw = vec3(barycentric.x * uv0 + barycentric.y * uv1 + barycentric.z * uv2, float(triangles.data[tidx].slice));
+
+			light = textureLod(sampler2DArray(source_light, linear_sampler), uvw, 0.0).rgb;
+			light += textureLod(sampler2DArray(source_direct_light, linear_sampler), uvw, 0.0).rgb;
+		} else {
+
+			//did not hit a triangle, reach out for the sky
+			vec3 sky_dir = normalize(mat3(params.env_transform) * ray_dir);
+
+			vec2 st = vec2(
+					atan(sky_dir.x, sky_dir.z),
+					acos(sky_dir.y));
+
+			if (st.x < 0.0)
+				st.x += PI * 2.0;
+
+			st /= vec2(PI * 2.0, PI);
+
+			light = textureLod(sampler2D(environment, linear_sampler), st, 0.0).rgb;
+		}
+
+		{
+			float c[9] = float[](
+					0.282095, //l0
+					0.488603 * ray_dir.y, //l1n1
+					0.488603 * ray_dir.z, //l1n0
+					0.488603 * ray_dir.x, //l1p1
+					1.092548 * ray_dir.x * ray_dir.y, //l2n2
+					1.092548 * ray_dir.y * ray_dir.z, //l2n1
+					//0.315392 * (ray_dir.x * ray_dir.x + ray_dir.y * ray_dir.y + 2.0 * ray_dir.z * ray_dir.z), //l20
+					0.315392 * (3.0 * ray_dir.z * ray_dir.z - 1.0), //l20
+					1.092548 * ray_dir.x * ray_dir.z, //l2p1
+					0.546274 * (ray_dir.x * ray_dir.x - ray_dir.y * ray_dir.y) //l2p2
+			);
+
+			for (uint j = 0; j < 9; j++) {
+				probe_sh_accum[j].rgb += light * c[j];
+			}
+		}
+	}
+
+	if (params.ray_from > 0) {
+		for (uint j = 0; j < 9; j++) { //accum from existing
+			probe_sh_accum[j] += light_probes.data[probe_index * 9 + j];
+		}
+	}
+
+	if (params.ray_to == params.ray_count) {
+		for (uint j = 0; j < 9; j++) { //accum from existing
+			probe_sh_accum[j] *= 4.0 / float(params.ray_count);
+		}
+	}
+
+	for (uint j = 0; j < 9; j++) { //accum from existing
+		light_probes.data[probe_index * 9 + j] = probe_sh_accum[j];
+	}
+
+#endif
+
+#ifdef MODE_DILATE
+
+	vec4 c = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice), 0);
+	//sides first, as they are closer
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 0), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, 1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 0), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, -1), params.atlas_slice), 0);
+	//endpoints second
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, -1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, -1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 1), params.atlas_slice), 0);
+
+	//far sides third
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 0), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, 2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 0), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(0, -2), params.atlas_slice), 0);
+
+	//far-mid endpoints
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, -1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, -1), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 1), params.atlas_slice), 0);
+
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, -2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-1, 2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, -2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(1, 2), params.atlas_slice), 0);
+	//far endpoints
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, -2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(-2, 2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, -2), params.atlas_slice), 0);
+	c = c.a > 0.5 ? c : texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos + ivec2(2, 2), params.atlas_slice), 0);
+
+	imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice), c);
+
+#endif
+}
diff --git a/modules/lightmapper_rd/lm_raster.glsl b/modules/lightmapper_rd/lm_raster.glsl
new file mode 100644
index 0000000000..ae3038aead
--- /dev/null
+++ b/modules/lightmapper_rd/lm_raster.glsl
@@ -0,0 +1,170 @@
+/* clang-format off */
+[vertex]
+
+#version 450
+
+VERSION_DEFINES
+
+#include "lm_common_inc.glsl"
+
+				/* clang-format on */
+
+				layout(location = 0) out vec3 vertex_interp;
+layout(location = 1) out vec3 normal_interp;
+layout(location = 2) out vec2 uv_interp;
+layout(location = 3) out vec3 barycentric;
+layout(location = 4) flat out uvec3 vertex_indices;
+layout(location = 5) flat out vec3 face_normal;
+
+layout(push_constant, binding = 0, std430) uniform Params {
+	vec2 atlas_size;
+	vec2 uv_offset;
+	vec3 to_cell_size;
+	uint base_triangle;
+	vec3 to_cell_offset;
+	float bias;
+	ivec3 grid_size;
+	uint pad2;
+}
+params;
+
+/* clang-format on */
+
+void main() {
+
+	uint triangle_idx = params.base_triangle + gl_VertexIndex / 3;
+	uint triangle_subidx = gl_VertexIndex % 3;
+
+	vertex_indices = triangles.data[triangle_idx].indices;
+
+	uint vertex_idx;
+	if (triangle_subidx == 0) {
+		vertex_idx = vertex_indices.x;
+		barycentric = vec3(1, 0, 0);
+	} else if (triangle_subidx == 1) {
+		vertex_idx = vertex_indices.y;
+		barycentric = vec3(0, 1, 0);
+	} else {
+		vertex_idx = vertex_indices.z;
+		barycentric = vec3(0, 0, 1);
+	}
+
+	vertex_interp = vertices.data[vertex_idx].position;
+	uv_interp = vertices.data[vertex_idx].uv;
+	normal_interp = vec3(vertices.data[vertex_idx].normal_xy, vertices.data[vertex_idx].normal_z);
+
+	face_normal = -normalize(cross((vertices.data[vertex_indices.x].position - vertices.data[vertex_indices.y].position), (vertices.data[vertex_indices.x].position - vertices.data[vertex_indices.z].position)));
+
+	gl_Position = vec4((uv_interp + params.uv_offset) * 2.0 - 1.0, 0.0001, 1.0);
+	;
+}
+
+/* clang-format off */
+
+[fragment]
+
+#version 450
+
+VERSION_DEFINES
+
+#include "lm_common_inc.glsl"
+
+
+layout(push_constant, binding = 0, std430) uniform Params {
+	vec2 atlas_size;
+	vec2 uv_offset;
+	vec3 to_cell_size;
+	uint base_triangle;
+	vec3 to_cell_offset;
+	float bias;
+	ivec3 grid_size;
+	uint pad2;
+} params;
+
+/* clang-format on */
+
+layout(location = 0) in vec3 vertex_interp;
+layout(location = 1) in vec3 normal_interp;
+layout(location = 2) in vec2 uv_interp;
+layout(location = 3) in vec3 barycentric;
+layout(location = 4) in flat uvec3 vertex_indices;
+layout(location = 5) in flat vec3 face_normal;
+
+layout(location = 0) out vec4 position;
+layout(location = 1) out vec4 normal;
+layout(location = 2) out vec4 unocclude;
+
+void main() {
+
+	vec3 vertex_pos = vertex_interp;
+
+	{
+		// smooth out vertex position by interpolating its projection in the 3 normal planes (normal plane is created by vertex pos and normal)
+		// because we don't want to interpolate inwards, normals found pointing inwards are pushed out.
+
+		vec3 pos_a = vertices.data[vertex_indices.x].position;
+		vec3 pos_b = vertices.data[vertex_indices.y].position;
+		vec3 pos_c = vertices.data[vertex_indices.z].position;
+		vec3 center = (pos_a + pos_b + pos_c) * 0.3333333;
+		vec3 norm_a = vec3(vertices.data[vertex_indices.x].normal_xy, vertices.data[vertex_indices.x].normal_z);
+		vec3 norm_b = vec3(vertices.data[vertex_indices.y].normal_xy, vertices.data[vertex_indices.y].normal_z);
+		vec3 norm_c = vec3(vertices.data[vertex_indices.z].normal_xy, vertices.data[vertex_indices.z].normal_z);
+
+		{
+			vec3 dir_a = normalize(pos_a - center);
+			float d_a = dot(dir_a, norm_a);
+			if (d_a < 0) {
+				//pointing inwards
+				norm_a = normalize(norm_a - dir_a * d_a);
+			}
+		}
+		{
+			vec3 dir_b = normalize(pos_b - center);
+			float d_b = dot(dir_b, norm_b);
+			if (d_b < 0) {
+				//pointing inwards
+				norm_b = normalize(norm_b - dir_b * d_b);
+			}
+		}
+		{
+			vec3 dir_c = normalize(pos_c - center);
+			float d_c = dot(dir_c, norm_c);
+			if (d_c < 0) {
+				//pointing inwards
+				norm_c = normalize(norm_c - dir_c * d_c);
+			}
+		}
+
+		float d_a = dot(norm_a, pos_a);
+		float d_b = dot(norm_b, pos_b);
+		float d_c = dot(norm_c, pos_c);
+
+		vec3 proj_a = vertex_pos - norm_a * (dot(norm_a, vertex_pos) - d_a);
+		vec3 proj_b = vertex_pos - norm_b * (dot(norm_b, vertex_pos) - d_b);
+		vec3 proj_c = vertex_pos - norm_c * (dot(norm_c, vertex_pos) - d_c);
+
+		vec3 smooth_position = proj_a * barycentric.x + proj_b * barycentric.y + proj_c * barycentric.z;
+
+		if (dot(face_normal, smooth_position) > dot(face_normal, vertex_pos)) { //only project outwards
+			vertex_pos = smooth_position;
+		}
+	}
+
+	{
+		// unocclusion technique based on:
+		// https://ndotl.wordpress.com/2018/08/29/baking-artifact-free-lightmaps/
+
+		/* compute texel size */
+		vec3 delta_uv = max(abs(dFdx(vertex_interp)), abs(dFdy(vertex_interp)));
+		float texel_size = max(delta_uv.x, max(delta_uv.y, delta_uv.z));
+		texel_size *= sqrt(2.0); //expand to unit box edge length (again, worst case)
+
+		unocclude.xyz = face_normal;
+		unocclude.w = texel_size;
+
+		//continued on lm_compute.glsl
+	}
+
+	position = vec4(vertex_pos, 1.0);
+	normal = vec4(normalize(normal_interp), 1.0);
+}
diff --git a/modules/lightmapper_rd/register_types.cpp b/modules/lightmapper_rd/register_types.cpp
new file mode 100644
index 0000000000..f3938f3190
--- /dev/null
+++ b/modules/lightmapper_rd/register_types.cpp
@@ -0,0 +1,64 @@
+/*************************************************************************/
+/*  register_types.cpp                                                   */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+#include "register_types.h"
+
+#include "core/project_settings.h"
+#include "lightmapper_rd.h"
+#include "scene/3d/lightmapper.h"
+
+#ifndef _3D_DISABLED
+static Lightmapper *create_lightmapper_rd() {
+	return memnew(LightmapperRD);
+}
+#endif
+
+void register_lightmapper_rd_types() {
+
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/low_quality_ray_count", 16);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/medium_quality_ray_count", 64);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/high_quality_ray_count", 256);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/ultra_quality_ray_count", 1024);
+	GLOBAL_DEF("rendering/gpu_lightmapper/performance/max_rays_per_pass", 32);
+	GLOBAL_DEF("rendering/gpu_lightmapper/performance/region_size", 512);
+
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/low_quality_probe_ray_count", 64);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/medium_quality_probe_ray_count", 256);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/high_quality_probe_ray_count", 512);
+	GLOBAL_DEF("rendering/gpu_lightmapper/quality/ultra_quality_probe_ray_count", 2048);
+	GLOBAL_DEF("rendering/gpu_lightmapper/performance/max_rays_per_probe_pass", 64);
+#ifndef _3D_DISABLED
+	ClassDB::register_class<LightmapperRD>();
+	Lightmapper::create_gpu = create_lightmapper_rd;
+#endif
+}
+
+void unregister_lightmapper_rd_types() {
+}
diff --git a/modules/lightmapper_rd/register_types.h b/modules/lightmapper_rd/register_types.h
new file mode 100644
index 0000000000..b0e15a927f
--- /dev/null
+++ b/modules/lightmapper_rd/register_types.h
@@ -0,0 +1,37 @@
+/*************************************************************************/
+/*  register_types.h                                                     */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2020 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2020 Godot Engine contributors (cf. AUTHORS.md).   */
+/*                                                                       */
+/* Permission is hereby granted, free of charge, to any person obtaining */
+/* a copy of this software and associated documentation files (the       */
+/* "Software"), to deal in the Software without restriction, including   */
+/* without limitation the rights to use, copy, modify, merge, publish,   */
+/* distribute, sublicense, and/or sell copies of the Software, and to    */
+/* permit persons to whom the Software is furnished to do so, subject to */
+/* the following conditions:                                             */
+/*                                                                       */
+/* The above copyright notice and this permission notice shall be        */
+/* included in all copies or substantial portions of the Software.       */
+/*                                                                       */
+/* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,       */
+/* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF    */
+/* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.*/
+/* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY  */
+/* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,  */
+/* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE     */
+/* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.                */
+/*************************************************************************/
+
+#ifndef LIGHTMAPPER_RD_REGISTER_TYPES_H
+#define LIGHTMAPPER_RD_REGISTER_TYPES_H
+
+void register_lightmapper_rd_types();
+void unregister_lightmapper_rd_types();
+
+#endif // XATLAS_UNWRAP_REGISTER_TYPES_H
diff --git a/modules/tinyexr/image_saver_tinyexr.cpp b/modules/tinyexr/image_saver_tinyexr.cpp
index 05080289bd..bc30f4e4fd 100644
--- a/modules/tinyexr/image_saver_tinyexr.cpp
+++ b/modules/tinyexr/image_saver_tinyexr.cpp
@@ -267,13 +267,21 @@ Error save_exr(const String &p_path, const Ref<Image> &p_img, bool p_grayscale)
 	header.channels = channel_infos;
 	header.pixel_types = pixel_types;
 	header.requested_pixel_types = requested_pixel_types;
+	header.compression_type = TINYEXR_COMPRESSIONTYPE_PIZ;
 
-	CharString utf8_filename = p_path.utf8();
-	const char *err;
-	int ret = SaveEXRImageToFile(&image, &header, utf8_filename.ptr(), &err);
-	if (ret != TINYEXR_SUCCESS) {
+	unsigned char *mem = nullptr;
+	const char *err = nullptr;
+
+	size_t bytes = SaveEXRImageToMemory(&image, &header, &mem, &err);
+
+	if (bytes == 0) {
 		print_error(String("Saving EXR failed. Error: {0}").format(varray(err)));
 		return ERR_FILE_CANT_WRITE;
+	} else {
+		FileAccessRef ref = FileAccess::open(p_path, FileAccess::WRITE);
+		ERR_FAIL_COND_V(!ref, ERR_FILE_CANT_WRITE);
+		ref->store_buffer(mem, bytes);
+		free(mem);
 	}
 
 	return OK;
diff --git a/modules/xatlas_unwrap/register_types.cpp b/modules/xatlas_unwrap/register_types.cpp
index 8c5525bed3..f77646ce28 100644
--- a/modules/xatlas_unwrap/register_types.cpp
+++ b/modules/xatlas_unwrap/register_types.cpp
@@ -137,6 +137,7 @@ bool xatlas_mesh_lightmap_unwrap_callback(float p_texel_size, const float *p_ver
 
 	pack_options.maxChartSize = 4096;
 	pack_options.blockAlign = true;
+	pack_options.padding = 1;
 	pack_options.texelsPerUnit = 1.0 / p_texel_size;
 
 	xatlas::Atlas *atlas = xatlas::Create();