Fix usages of mesh simplification functions in float=64 builds

5 files changed, 54 insertions, 7 deletions

diff --git a/core/io/marshalls.cpp b/core/io/marshalls.cpp
index 9ba653e1a9..9f89f5d8c9 100644
--- a/core/io/marshalls.cpp
+++ b/core/io/marshalls.cpp
@@ -1813,3 +1813,24 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 
 	return OK;
 }
+
+Vector<float> vector3_to_float32_array(const Vector3 *vecs, size_t count) {
+	// We always allocate a new array, and we don't memcpy.
+	// We also don't consider returning a pointer to the passed vectors when sizeof(real_t) == 4.
+	// One reason is that we could decide to put a 4th component in Vector3 for SIMD/mobile performance,
+	// which would cause trouble with these optimizations.
+	Vector<float> floats;
+	if (count == 0) {
+		return floats;
+	}
+	floats.resize(count * 3);
+	float *floats_w = floats.ptrw();
+	for (size_t i = 0; i < count; ++i) {
+		const Vector3 v = vecs[i];
+		floats_w[0] = v.x;
+		floats_w[1] = v.y;
+		floats_w[2] = v.z;
+		floats_w += 3;
+	}
+	return floats;
+}
diff --git a/core/io/marshalls.h b/core/io/marshalls.h
index fef3a1c2c1..66e2571066 100644
--- a/core/io/marshalls.h
+++ b/core/io/marshalls.h
@@ -215,4 +215,6 @@ public:
 Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int *r_len = nullptr, bool p_allow_objects = false, int p_depth = 0);
 Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bool p_full_objects = false, int p_depth = 0);
 
+Vector<float> vector3_to_float32_array(const Vector3 *vecs, size_t count);
+
 #endif // MARSHALLS_H
diff --git a/scene/3d/occluder_instance_3d.cpp b/scene/3d/occluder_instance_3d.cpp
index 4e1ed5654a..f150a602b3 100644
--- a/scene/3d/occluder_instance_3d.cpp
+++ b/scene/3d/occluder_instance_3d.cpp
@@ -32,6 +32,7 @@
 
 #include "core/config/project_settings.h"
 #include "core/core_string_names.h"
+#include "core/io/marshalls.h"
 #include "core/math/geometry_2d.h"
 #include "core/math/triangulate.h"
 #include "scene/3d/importer_mesh_instance_3d.h"
@@ -533,11 +534,20 @@ void OccluderInstance3D::_bake_surface(const Transform3D &p_transform, Array p_s
 	}
 
 	if (!Math::is_zero_approx(p_simplification_dist) && SurfaceTool::simplify_func) {
-		float error_scale = SurfaceTool::simplify_scale_func((float *)vertices.ptr(), vertices.size(), sizeof(Vector3));
+		Vector<float> vertices_f32 = vector3_to_float32_array(vertices.ptr(), vertices.size());
+
+		float error_scale = SurfaceTool::simplify_scale_func(vertices_f32.ptr(), vertices.size(), sizeof(float) * 3);
 		float target_error = p_simplification_dist / error_scale;
 		float error = -1.0f;
 		int target_index_count = MIN(indices.size(), 36);
-		uint32_t index_count = SurfaceTool::simplify_func((unsigned int *)indices.ptrw(), (unsigned int *)indices.ptr(), indices.size(), (float *)vertices.ptr(), vertices.size(), sizeof(Vector3), target_index_count, target_error, &error);
+
+		uint32_t index_count = SurfaceTool::simplify_func(
+				(unsigned int *)indices.ptrw(),
+				(unsigned int *)indices.ptr(),
+				indices.size(),
+				vertices_f32.ptr(), vertices.size(), sizeof(float) * 3,
+				target_index_count, target_error, &error);
+
 		indices.resize(index_count);
 	}
 
diff --git a/scene/resources/importer_mesh.cpp b/scene/resources/importer_mesh.cpp
index b728c24e0d..cd08103187 100644
--- a/scene/resources/importer_mesh.cpp
+++ b/scene/resources/importer_mesh.cpp
@@ -30,6 +30,7 @@
 
 #include "importer_mesh.h"
 
+#include "core/io/marshalls.h"
 #include "core/math/random_pcg.h"
 #include "core/math/static_raycaster.h"
 #include "scene/resources/surface_tool.h"
@@ -424,9 +425,8 @@ void ImporterMesh::generate_lods(float p_normal_merge_angle, float p_normal_spli
 			normal_weights[j] = 2.0; // Give some weight to normal preservation, may be worth exposing as an import setting
 		}
 
-		const float max_mesh_error = FLT_MAX; // We don't want to limit by error, just by index target
-		float scale = SurfaceTool::simplify_scale_func((const float *)merged_vertices_ptr, merged_vertex_count, sizeof(Vector3));
-		float mesh_error = 0.0f;
+		Vector<float> merged_vertices_f32 = vector3_to_float32_array(merged_vertices_ptr, merged_vertex_count);
+		float scale = SurfaceTool::simplify_scale_func(merged_vertices_f32.ptr(), merged_vertex_count, sizeof(float) * 3);
 
 		unsigned int index_target = 12; // Start with the smallest target, 4 triangles
 		unsigned int last_index_count = 0;
@@ -446,11 +446,25 @@ void ImporterMesh::generate_lods(float p_normal_merge_angle, float p_normal_spli
 			raycaster->commit();
 		}
 
+		const float max_mesh_error = FLT_MAX; // We don't want to limit by error, just by index target
+		float mesh_error = 0.0f;
+
 		while (index_target < index_count) {
 			PackedInt32Array new_indices;
 			new_indices.resize(index_count);
 
-			size_t new_index_count = SurfaceTool::simplify_with_attrib_func((unsigned int *)new_indices.ptrw(), (const uint32_t *)merged_indices_ptr, index_count, (const float *)merged_vertices_ptr, merged_vertex_count, sizeof(Vector3), index_target, max_mesh_error, &mesh_error, (float *)merged_normals.ptr(), normal_weights.ptr(), 3);
+			Vector<float> merged_normals_f32 = vector3_to_float32_array(merged_normals.ptr(), merged_normals.size());
+
+			size_t new_index_count = SurfaceTool::simplify_with_attrib_func(
+					(unsigned int *)new_indices.ptrw(),
+					(const uint32_t *)merged_indices_ptr, index_count,
+					merged_vertices_f32.ptr(), merged_vertex_count,
+					sizeof(float) * 3, // Vertex stride
+					index_target,
+					max_mesh_error,
+					&mesh_error,
+					merged_normals_f32.ptr(),
+					normal_weights.ptr(), 3);
 
 			if (new_index_count < last_index_count * 1.5f) {
 				index_target = index_target * 1.5f;
diff --git a/scene/resources/surface_tool.cpp b/scene/resources/surface_tool.cpp
index 94967352c8..185cbdc6bf 100644
--- a/scene/resources/surface_tool.cpp
+++ b/scene/resources/surface_tool.cpp
@@ -46,7 +46,7 @@ void SurfaceTool::strip_mesh_arrays(PackedVector3Array &r_vertices, PackedInt32A
 
 	Vector<uint32_t> remap;
 	remap.resize(r_vertices.size());
-	uint32_t new_vertex_count = generate_remap_func(remap.ptrw(), (unsigned int *)r_indices.ptr(), r_indices.size(), (float *)r_vertices.ptr(), r_vertices.size(), sizeof(Vector3));
+	uint32_t new_vertex_count = generate_remap_func(remap.ptrw(), (unsigned int *)r_indices.ptr(), r_indices.size(), r_vertices.ptr(), r_vertices.size(), sizeof(Vector3));
 	remap_vertex_func(r_vertices.ptrw(), r_vertices.ptr(), r_vertices.size(), sizeof(Vector3), remap.ptr());
 	r_vertices.resize(new_vertex_count);
 	remap_index_func((unsigned int *)r_indices.ptrw(), (unsigned int *)r_indices.ptr(), r_indices.size(), remap.ptr());