Use multiple threads to import.

- For now everything imports multithreaded by default (should work I guess, let's test).
- Controllable per importer

Early test benchmark. 64 large textures (importing as lossless, _not_ as vram) on a mobile i7, 12 threads:
Importing goes down from 46 to 7 seconds.

For VRAM I will change the logic to use a compressing thread in a subsequent PR, as well as implementing Betsy.

3 files changed, 42 insertions, 2 deletions

diff --git a/core/io/resource_importer.cpp b/core/io/resource_importer.cpp
index 5ca0eb884a..b503655edd 100644
--- a/core/io/resource_importer.cpp
+++ b/core/io/resource_importer.cpp
@@ -192,6 +192,34 @@ bool ResourceFormatImporter::recognize_path(const String &p_path, const String &
 	return FileAccess::exists(p_path + ".import");
 }
 
+Error ResourceFormatImporter::get_import_order_threads_and_importer(const String &p_path, int &r_order, bool &r_can_threads, String &r_importer) const {
+	r_order = 0;
+	r_importer = "";
+
+	r_can_threads = false;
+	Ref<ResourceImporter> importer;
+
+	if (FileAccess::exists(p_path + ".import")) {
+		PathAndType pat;
+		Error err = _get_path_and_type(p_path, pat);
+
+		if (err == OK) {
+			importer = get_importer_by_name(pat.importer);
+		}
+	} else {
+		importer = get_importer_by_extension(p_path.get_extension().to_lower());
+	}
+
+	if (importer.is_valid()) {
+		r_order = importer->get_import_order();
+		r_importer = importer->get_importer_name();
+		r_can_threads = importer->can_import_threaded();
+		return OK;
+	} else {
+		return ERR_INVALID_PARAMETER;
+	}
+}
+
 int ResourceFormatImporter::get_import_order(const String &p_path) const {
 	Ref<ResourceImporter> importer;
 
diff --git a/core/io/resource_importer.h b/core/io/resource_importer.h
index eeb486073e..a14d6ba52c 100644
--- a/core/io/resource_importer.h
+++ b/core/io/resource_importer.h
@@ -72,6 +72,8 @@ public:
 
 	virtual int get_import_order(const String &p_path) const;
 
+	Error get_import_order_threads_and_importer(const String &p_path, int &r_order, bool &r_can_threads, String &r_importer) const;
+
 	String get_internal_resource_path(const String &p_path) const;
 	void get_internal_resource_path_list(const String &p_path, List<String> *r_paths);
 
@@ -126,6 +128,9 @@ public:
 	virtual String get_option_group_file() const { return String(); }
 
 	virtual Error import(const String &p_source_file, const String &p_save_path, const Map<StringName, Variant> &p_options, List<String> *r_platform_variants, List<String> *r_gen_files = nullptr, Variant *r_metadata = nullptr) = 0;
+	virtual bool can_import_threaded() const { return true; }
+	virtual void import_threaded_begin() {}
+	virtual void import_threaded_end() {}
 
 	virtual Error import_group_file(const String &p_group_file, const Map<String, Map<StringName, Variant>> &p_source_file_options, const Map<String, String> &p_base_paths) { return ERR_UNAVAILABLE; }
 	virtual bool are_import_settings_valid(const String &p_path) const { return true; }
diff --git a/core/templates/thread_work_pool.h b/core/templates/thread_work_pool.h
index 19ab1dda3a..9f7a692cc5 100644
--- a/core/templates/thread_work_pool.h
+++ b/core/templates/thread_work_pool.h
@@ -83,7 +83,7 @@ public:
 		ERR_FAIL_COND(!threads); //never initialized
 		ERR_FAIL_COND(current_work != nullptr);
 
-		index.store(0);
+		index.store(0, std::memory_order_release);
 
 		Work<C, M, U> *w = memnew((Work<C, M, U>));
 		w->instance = p_instance;
@@ -104,8 +104,15 @@ public:
 		return current_work != nullptr;
 	}
 
+	bool is_done_dispatching() const {
+		ERR_FAIL_COND_V(current_work == nullptr, false);
+		return index.load(std::memory_order_acquire) >= current_work->max_elements;
+	}
+
 	uint32_t get_work_index() const {
-		return index;
+		ERR_FAIL_COND_V(current_work == nullptr, 0);
+		uint32_t idx = index.load(std::memory_order_acquire);
+		return MIN(idx, current_work->max_elements);
 	}
 
 	void end_work() {