3 files changed, 64 insertions, 1 deletions

diff --git a/modules/lightmapper_rd/lightmapper_rd.cpp b/modules/lightmapper_rd/lightmapper_rd.cpp
index 89495e2c83..8ba6f9e2ba 100644
--- a/modules/lightmapper_rd/lightmapper_rd.cpp
+++ b/modules/lightmapper_rd/lightmapper_rd.cpp
@@ -793,6 +793,35 @@ LightmapperRD::BakeError LightmapperRD::_dilate(RenderingDevice *rd, Ref<RDShade
 	return BAKE_OK;
 }
 
+LightmapperRD::BakeError LightmapperRD::_pack_l1(RenderingDevice *rd, Ref<RDShaderFile> &compute_shader, RID &compute_base_uniform_set, PushConstant &push_constant, RID &source_light_tex, RID &dest_light_tex, const Size2i &atlas_size, int atlas_slices) {
+	Vector<RD::Uniform> uniforms = dilate_or_denoise_common_uniforms(source_light_tex, dest_light_tex);
+
+	RID compute_shader_pack = rd->shader_create_from_spirv(compute_shader->get_spirv_stages("pack_coeffs"));
+	ERR_FAIL_COND_V(compute_shader_pack.is_null(), BAKE_ERROR_LIGHTMAP_CANT_PRE_BAKE_MESHES); //internal check, should not happen
+	RID compute_shader_pack_pipeline = rd->compute_pipeline_create(compute_shader_pack);
+
+	RID dilate_uniform_set = rd->uniform_set_create(uniforms, compute_shader_pack, 1);
+
+	RD::ComputeListID compute_list = rd->compute_list_begin();
+	rd->compute_list_bind_compute_pipeline(compute_list, compute_shader_pack_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;
+	Vector3i group_size(Math::division_round_up(atlas_size.x, 8), Math::division_round_up(atlas_size.y, 8), 1); //restore group size
+
+	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();
+	rd->free(compute_shader_pack);
+
+	return BAKE_OK;
+}
+
 Error LightmapperRD::_store_pfm(RenderingDevice *p_rd, RID p_atlas_tex, int p_index, const Size2i &p_atlas_size, const String &p_name) {
 	Vector<uint8_t> data = p_rd->texture_get_data(p_atlas_tex, p_index);
 	Ref<Image> img = Image::create_from_data(p_atlas_size.width, p_atlas_size.height, false, Image::FORMAT_RGBAH, data);
@@ -2002,6 +2031,14 @@ LightmapperRD::BakeError LightmapperRD::bake(BakeQuality p_quality, bool p_use_d
 		}
 	}
 
+	if (p_bake_sh) {
+		SWAP(light_accum_tex, light_accum_tex2);
+		BakeError error = _pack_l1(rd, compute_shader, compute_base_uniform_set, push_constant, light_accum_tex2, light_accum_tex, atlas_size, atlas_slices);
+		if (unlikely(error != BAKE_OK)) {
+			return error;
+		}
+	}
+
 #ifdef DEBUG_TEXTURES
 
 	for (int i = 0; i < atlas_slices * (p_bake_sh ? 4 : 1); i++) {
diff --git a/modules/lightmapper_rd/lightmapper_rd.h b/modules/lightmapper_rd/lightmapper_rd.h
index 59c2d52e69..f43da39670 100644
--- a/modules/lightmapper_rd/lightmapper_rd.h
+++ b/modules/lightmapper_rd/lightmapper_rd.h
@@ -273,6 +273,7 @@ class LightmapperRD : public Lightmapper {
 
 	BakeError _dilate(RenderingDevice *rd, Ref<RDShaderFile> &compute_shader, RID &compute_base_uniform_set, PushConstant &push_constant, RID &source_light_tex, RID &dest_light_tex, const Size2i &atlas_size, int atlas_slices);
 	BakeError _denoise(RenderingDevice *p_rd, Ref<RDShaderFile> &p_compute_shader, const RID &p_compute_base_uniform_set, PushConstant &p_push_constant, RID p_source_light_tex, RID p_source_normal_tex, RID p_dest_light_tex, float p_denoiser_strength, int p_denoiser_range, const Size2i &p_atlas_size, int p_atlas_slices, bool p_bake_sh, BakeStepFunc p_step_function, void *p_bake_userdata);
+	BakeError _pack_l1(RenderingDevice *rd, Ref<RDShaderFile> &compute_shader, RID &compute_base_uniform_set, PushConstant &push_constant, RID &source_light_tex, RID &dest_light_tex, const Size2i &atlas_size, int atlas_slices);
 
 	Error _store_pfm(RenderingDevice *p_rd, RID p_atlas_tex, int p_index, const Size2i &p_atlas_size, const String &p_name);
 	Ref<Image> _read_pfm(const String &p_name);
diff --git a/modules/lightmapper_rd/lm_compute.glsl b/modules/lightmapper_rd/lm_compute.glsl
index 88fc316679..2c85fff6f3 100644
--- a/modules/lightmapper_rd/lm_compute.glsl
+++ b/modules/lightmapper_rd/lm_compute.glsl
@@ -6,6 +6,7 @@ dilate = "#define MODE_DILATE";
 unocclude = "#define MODE_UNOCCLUDE";
 light_probes = "#define MODE_LIGHT_PROBES";
 denoise = "#define MODE_DENOISE";
+pack_coeffs = "#define MODE_PACK_L1_COEFFS";
 
 #[compute]
 
@@ -63,7 +64,7 @@ layout(rgba16f, set = 1, binding = 4) uniform restrict image2DArray accum_light;
 layout(set = 1, binding = 5) uniform texture2D environment;
 #endif
 
-#if defined(MODE_DILATE) || defined(MODE_DENOISE)
+#if defined(MODE_DILATE) || defined(MODE_DENOISE) || defined(MODE_PACK_L1_COEFFS)
 layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2DArray dest_light;
 layout(set = 1, binding = 1) uniform texture2DArray source_light;
 #endif
@@ -1037,4 +1038,28 @@ void main() {
 		imageStore(dest_light, ivec3(atlas_pos, lightmap_slice), vec4(denoised_rgb, input_light.a));
 	}
 #endif
+
+#ifdef MODE_PACK_L1_COEFFS
+	vec4 base_coeff = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice * 4), 0);
+
+	for (int i = 1; i < 4; i++) {
+		vec4 c = texelFetch(sampler2DArray(source_light, linear_sampler), ivec3(atlas_pos, params.atlas_slice * 4 + i), 0);
+
+		if (abs(base_coeff.r) > 0.0) {
+			c.r /= (base_coeff.r * 8);
+		}
+
+		if (abs(base_coeff.g) > 0.0) {
+			c.g /= (base_coeff.g * 8);
+		}
+
+		if (abs(base_coeff.b) > 0.0) {
+			c.b /= (base_coeff.b * 8);
+		}
+
+		c.rgb += vec3(0.5);
+		c.rgb = clamp(c.rgb, vec3(0.0), vec3(1.0));
+		imageStore(dest_light, ivec3(atlas_pos, params.atlas_slice * 4 + i), c);
+	}
+#endif
 }