Unified GLES2 / GLES3 Batching

Batching is mostly separated into a common template which can be used with multiple backends (GLES2 and GLES3 here). Only necessary specifics are in the backend files.

Batching is extended to cover more primitives.

doc/classes/ProjectSettings.xml

@@ -1062,6 +1062,10 @@
 		<member name="rendering/limits/time/time_rollover_secs" type="float" setter="" getter="" default="3600">
 			Shaders have a time variable that constantly increases. At some point, it needs to be rolled back to zero to avoid precision errors on shader animations. This setting specifies when (in seconds).
 		</member>
+		<member name="rendering/quality/2d/ninepatch_mode" type="int" setter="" getter="" default="0">
+			Choose between default mode where corner scalings are preserved matching the artwork, and scaling mode.
+			Not available in GLES3 when [member rendering/batching/options/use_batching] is off.
+		</member>
 		<member name="rendering/quality/2d/use_nvidia_rect_flicker_workaround" type="bool" setter="" getter="" default="false">
 			Some NVIDIA GPU drivers have a bug which produces flickering issues for the [code]draw_rect[/code] method, especially as used in [TileMap]. Refer to [url=https://github.com/godotengine/godot/issues/9913]GitHub issue 9913[/url] for details.
 			If [code]true[/code], this option enables a "safe" code path for such NVIDIA GPUs at the cost of performance. This option affects GLES2 and GLES3 rendering, but only on desktop platforms.
@@ -1070,6 +1074,10 @@
 			If [code]true[/code], forces snapping of polygons to pixels in 2D rendering. May help in some pixel art styles.
 			Consider using the project setting [member rendering/batching/precision/uv_contract] to prevent artifacts.
 		</member>
+		<member name="rendering/quality/2d/use_software_skinning" type="bool" setter="" getter="" default="true">
+			If [code]true[/code], performs 2d skinning on the CPU rather than the GPU. This provides greater compatibility with a wide range of hardware, and also may be faster in some circumstances.
+			Currently only available when [member rendering/batching/options/use_batching] is active.
+		</member>
 		<member name="rendering/quality/depth/hdr" type="bool" setter="" getter="" default="true">
 			If [code]true[/code], allocates the main framebuffer with high dynamic range. High dynamic range allows the use of [Color] values greater than 1.
 			[b]Note:[/b] Only available on the GLES3 backend.

drivers/SCsub

@@ -26,6 +26,7 @@ SConscript("winmidi/SCsub")
 if env["platform"] != "server":
     SConscript("gles3/SCsub")
     SConscript("gles2/SCsub")
+    SConscript("gles_common/SCsub")
     SConscript("gl_context/SCsub")
 else:
     SConscript("dummy/SCsub")

drivers/gles2/rasterizer_canvas_base_gles2.cpp

@@ -56,7 +56,12 @@ void RasterizerCanvasBaseGLES2::canvas_begin() {
 
 	// always start with light_angle unset
 	state.using_light_angle = false;
-	state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_LIGHT_ANGLE, false);
+	state.using_large_vertex = false;
+	state.using_modulate = false;
+
+	state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_LIGHT_ANGLE, false);
+	state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_MODULATE, false);
+	state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_LARGE_VERTEX, false);
 	state.canvas_shader.bind();
 
 	int viewport_x, viewport_y, viewport_width, viewport_height;
@@ -160,13 +165,23 @@ void RasterizerCanvasBaseGLES2::draw_generic_textured_rect(const Rect2 &p_rect,
 	glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
 }
 
-void RasterizerCanvasBaseGLES2::_set_texture_rect_mode(bool p_texture_rect, bool p_light_angle) {
+void RasterizerCanvasBaseGLES2::_set_texture_rect_mode(bool p_texture_rect, bool p_light_angle, bool p_modulate, bool p_large_vertex) {
 	// always set this directly (this could be state checked)
 	state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_TEXTURE_RECT, p_texture_rect);
 
 	if (state.using_light_angle != p_light_angle) {
 		state.using_light_angle = p_light_angle;
-		state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_LIGHT_ANGLE, p_light_angle);
+		state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_LIGHT_ANGLE, p_light_angle);
+	}
+
+	if (state.using_modulate != p_modulate) {
+		state.using_modulate = p_modulate;
+		state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_MODULATE, p_modulate);
+	}
+
+	if (state.using_large_vertex != p_large_vertex) {
+		state.using_large_vertex = p_large_vertex;
+		state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_ATTRIB_LARGE_VERTEX, p_large_vertex);
 	}
 }
 
@@ -1024,10 +1039,4 @@ void RasterizerCanvasBaseGLES2::finalize() {
 }
 
 RasterizerCanvasBaseGLES2::RasterizerCanvasBaseGLES2() {
-#ifdef GLES_OVER_GL
-	use_nvidia_rect_workaround = GLOBAL_GET("rendering/quality/2d/use_nvidia_rect_flicker_workaround");
-#else
-	// Not needed (a priori) on GLES devices
-	use_nvidia_rect_workaround = false;
-#endif
 }

drivers/gles2/rasterizer_canvas_base_gles2.h

@@ -31,13 +31,14 @@
 #ifndef RASTERIZERCANVASBASEGLES2_H
 #define RASTERIZERCANVASBASEGLES2_H
 
-#include "rasterizer_array_gles2.h"
+#include "drivers/gles_common/rasterizer_array.h"
 #include "rasterizer_storage_gles2.h"
 #include "servers/visual/rasterizer.h"
 
 #include "shaders/canvas.glsl.gen.h"
 #include "shaders/lens_distorted.glsl.gen.h"
 
+#include "drivers/gles_common/rasterizer_storage_common.h"
 #include "shaders/canvas_shadow.glsl.gen.h"
 
 class RasterizerCanvasBaseGLES2 : public RasterizerCanvas {
@@ -77,7 +78,11 @@ public:
 		LensDistortedShaderGLES2 lens_shader;
 
 		bool using_texture_rect;
+
 		bool using_light_angle;
+		bool using_modulate;
+		bool using_large_vertex;
+
 		bool using_ninepatch;
 		bool using_skeleton;
 
@@ -102,8 +107,6 @@ public:
 
 	RasterizerStorageGLES2 *storage;
 
-	bool use_nvidia_rect_workaround;
-
 	void _set_uniforms();
 
 	virtual RID light_internal_create();
@@ -131,7 +134,7 @@ public:
 	virtual void canvas_debug_viewport_shadows(Light *p_lights_with_shadow);
 
 	RasterizerStorageGLES2::Texture *_bind_canvas_texture(const RID &p_texture, const RID &p_normal_map);
-	void _set_texture_rect_mode(bool p_texture_rect, bool p_light_angle = false);
+	void _set_texture_rect_mode(bool p_texture_rect, bool p_light_angle = false, bool p_modulate = false, bool p_large_vertex = false);
 
 	void initialize();
 	void finalize();

drivers/gles2/rasterizer_canvas_gles2.h

@@ -31,297 +31,14 @@
 #ifndef RASTERIZERCANVASGLES2_H
 #define RASTERIZERCANVASGLES2_H
 
+#include "drivers/gles_common/rasterizer_canvas_batcher.h"
 #include "rasterizer_canvas_base_gles2.h"
 
 class RasterizerSceneGLES2;
 
-class RasterizerCanvasGLES2 : public RasterizerCanvasBaseGLES2 {
+class RasterizerCanvasGLES2 : public RasterizerCanvasBaseGLES2, public RasterizerCanvasBatcher<RasterizerCanvasGLES2, RasterizerStorageGLES2> {
 
-	// used to determine whether we use hardware transform (none)
+	friend class RasterizerCanvasBatcher<RasterizerCanvasGLES2, RasterizerStorageGLES2>;
-	// software transform all verts, or software transform just a translate
-	// (no rotate or scale)
-	enum TransformMode {
-		TM_NONE,
-		TM_ALL,
-		TM_TRANSLATE,
-	};
-
-	// pod versions of vector and color and RID, need to be 32 bit for vertex format
-	struct BatchVector2 {
-		float x, y;
-		void set(const Vector2 &p_o) {
-			x = p_o.x;
-			y = p_o.y;
-		}
-		void to(Vector2 &r_o) const {
-			r_o.x = x;
-			r_o.y = y;
-		}
-	};
-
-	struct BatchColor {
-		float r, g, b, a;
-		void set(const Color &p_c) {
-			r = p_c.r;
-			g = p_c.g;
-			b = p_c.b;
-			a = p_c.a;
-		}
-		bool operator==(const BatchColor &p_c) const {
-			return (r == p_c.r) && (g == p_c.g) && (b == p_c.b) && (a == p_c.a);
-		}
-		bool operator!=(const BatchColor &p_c) const { return (*this == p_c) == false; }
-		bool equals(const Color &p_c) const {
-			return (r == p_c.r) && (g == p_c.g) && (b == p_c.b) && (a == p_c.a);
-		}
-		const float *get_data() const { return &r; }
-		String to_string() const;
-	};
-
-	struct BatchVertex {
-		// must be 32 bit pod
-		BatchVector2 pos;
-		BatchVector2 uv;
-	};
-
-	struct BatchVertexColored : public BatchVertex {
-		// must be 32 bit pod
-		BatchColor col;
-	};
-
-	struct BatchVertexLightAngled : public BatchVertexColored {
-		// must be pod
-		float light_angle;
-	};
-
-	struct Batch {
-		enum CommandType : uint32_t {
-			BT_DEFAULT,
-			BT_RECT,
-		};
-
-		CommandType type;
-		uint32_t first_command; // also item reference number
-		uint32_t num_commands;
-		uint32_t first_quad;
-		uint32_t batch_texture_id;
-		BatchColor color;
-	};
-
-	struct BatchTex {
-		enum TileMode : uint32_t {
-			TILE_OFF,
-			TILE_NORMAL,
-			TILE_FORCE_REPEAT,
-		};
-		RID RID_texture;
-		RID RID_normal;
-		TileMode tile_mode;
-		BatchVector2 tex_pixel_size;
-		uint32_t flags;
-	};
-
-	// items in a list to be sorted prior to joining
-	struct BSortItem {
-		// have a function to keep as pod, rather than operator
-		void assign(const BSortItem &o) {
-			item = o.item;
-			z_index = o.z_index;
-		}
-		Item *item;
-		int z_index;
-	};
-
-	// batch item may represent 1 or more items
-	struct BItemJoined {
-		uint32_t first_item_ref;
-		uint32_t num_item_refs;
-
-		Rect2 bounding_rect;
-
-		// note the z_index  may only be correct for the first of the joined item references
-		// this has implications for light culling with z ranged lights.
-		int16_t z_index;
-
-		// these are defined in RasterizerStorageGLES2::Shader::CanvasItem::BatchFlags
-		uint16_t flags;
-
-		// we are always splitting items with lots of commands,
-		// and items with unhandled primitives (default)
-		bool use_hardware_transform() const { return num_item_refs == 1; }
-	};
-
-	struct BItemRef {
-		Item *item;
-		Color final_modulate;
-	};
-
-	struct BLightRegion {
-		void reset() {
-			light_bitfield = 0;
-			shadow_bitfield = 0;
-			too_many_lights = false;
-		}
-		uint64_t light_bitfield;
-		uint64_t shadow_bitfield;
-		bool too_many_lights; // we can only do light region optimization if there are 64 or less lights
-	};
-
-	struct BatchData {
-		BatchData();
-		void reset_flush() {
-			batches.reset();
-			batch_textures.reset();
-
-			vertices.reset();
-			light_angles.reset();
-
-			total_quads = 0;
-			total_color_changes = 0;
-			use_light_angles = false;
-		}
-
-		GLuint gl_vertex_buffer;
-		GLuint gl_index_buffer;
-
-		uint32_t max_quads;
-		uint32_t vertex_buffer_size_units;
-		uint32_t vertex_buffer_size_bytes;
-		uint32_t index_buffer_size_units;
-		uint32_t index_buffer_size_bytes;
-
-		// small vertex FVF type - pos and UV.
-		// This will always be written to initially, but can be translated
-		// to larger FVFs if necessary.
-		RasterizerArrayGLES2<BatchVertex> vertices;
-
-		// extra data which can be stored during prefilling, for later translation to larger FVFs
-		RasterizerArrayGLES2<float> light_angles;
-
-		// instead of having a different buffer for each vertex FVF type
-		// we have a special array big enough for the biggest FVF
-		// which can have a changeable unit size, and reuse it.
-		RasterizerUnitArrayGLES2 unit_vertices;
-
-		RasterizerArrayGLES2<Batch> batches;
-		RasterizerArrayGLES2<Batch> batches_temp; // used for translating to colored vertex batches
-		RasterizerArray_non_pod_GLES2<BatchTex> batch_textures; // the only reason this is non-POD is because of RIDs
-
-		// flexible vertex format.
-		// all verts have pos and UV.
-		// some have color, some light angles etc.
-		bool use_colored_vertices;
-		bool use_light_angles;
-
-		RasterizerArrayGLES2<BItemJoined> items_joined;
-		RasterizerArrayGLES2<BItemRef> item_refs;
-
-		// items are sorted prior to joining
-		RasterizerArrayGLES2<BSortItem> sort_items;
-
-		// counts
-		int total_quads;
-
-		// we keep a record of how many color changes caused new batches
-		// if the colors are causing an excessive number of batches, we switch
-		// to alternate batching method and add color to the vertex format.
-		int total_color_changes;
-
-		// if the shader is using MODULATE, we prevent baking color so the final_modulate can
-		// be read in the shader.
-		// if the shader is reading VERTEX, we prevent baking vertex positions with extra matrices etc
-		// to prevent the read position being incorrect.
-		// These flags are defined in RasterizerStorageGLES2::Shader::CanvasItem::BatchFlags
-		uint32_t joined_item_batch_flags;
-
-		// measured in pixels, recalculated each frame
-		float scissor_threshold_area;
-
-		// diagnose this frame, every nTh frame when settings_diagnose_frame is on
-		bool diagnose_frame;
-		String frame_string;
-		uint32_t next_diagnose_tick;
-		uint64_t diagnose_frame_number;
-
-		// whether to join items across z_indices - this can interfere with z ranged lights,
-		// so has to be disabled in some circumstances
-		bool join_across_z_indices;
-
-		// global settings
-		bool settings_use_batching; // the current use_batching (affected by flash)
-		bool settings_use_batching_original_choice; // the choice entered in project settings
-		bool settings_flash_batching; // for regression testing, flash between non-batched and batched renderer
-		bool settings_diagnose_frame; // print out batches to help optimize / regression test
-		int settings_max_join_item_commands;
-		float settings_colored_vertex_format_threshold;
-		int settings_batch_buffer_num_verts;
-		bool settings_scissor_lights;
-		float settings_scissor_threshold; // 0.0 to 1.0
-		int settings_item_reordering_lookahead;
-		bool settings_use_single_rect_fallback;
-		int settings_light_max_join_items;
-
-		// uv contraction
-		bool settings_uv_contract;
-		float settings_uv_contract_amount;
-
-		// only done on diagnose frame
-		void reset_stats() {
-			stats_items_sorted = 0;
-			stats_light_items_joined = 0;
-		}
-
-		// frame stats (just for monitoring and debugging)
-		int stats_items_sorted;
-		int stats_light_items_joined;
-	} bdata;
-
-	struct RenderItemState {
-		RenderItemState() { reset(); }
-		void reset();
-		Item *current_clip;
-		RasterizerStorageGLES2::Shader *shader_cache;
-		bool rebind_shader;
-		bool prev_use_skeleton;
-		int last_blend_mode;
-		RID canvas_last_material;
-		Color final_modulate;
-
-		// used for joining items only
-		BItemJoined *joined_item;
-		bool join_batch_break;
-		BLightRegion light_region;
-
-		// 'item group' is data over a single call to canvas_render_items
-		int item_group_z;
-		Color item_group_modulate;
-		Light *item_group_light;
-		Transform2D item_group_base_transform;
-	} _render_item_state;
-
-	struct FillState {
-		void reset() {
-			// don't reset members that need to be preserved after flushing
-			// half way through a list of commands
-			curr_batch = 0;
-			batch_tex_id = -1;
-			texpixel_size = Vector2(1, 1);
-			contract_uvs = false;
-		}
-		Batch *curr_batch;
-		int batch_tex_id;
-		bool use_hardware_transform;
-		bool contract_uvs;
-		Vector2 texpixel_size;
-		Color final_modulate;
-		TransformMode transform_mode;
-		TransformMode orig_transform_mode;
-
-		// support for extra matrices
-		bool extra_matrix_sent; // whether sent on this item (in which case software transform can't be used untl end of item)
-		int transform_extra_command_number_p1; // plus one to allow fast checking against zero
-		Transform2D transform_combined; // final * extra
-	};
 
 public:
 	virtual void canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
@@ -332,598 +49,27 @@ public:
 
 private:
 	// legacy codepath .. to remove after testing
-	void _canvas_render_item(Item *p_ci, RenderItemState &r_ris);
+	void _legacy_canvas_render_item(Item *p_ci, RenderItemState &r_ris);
-	void _canvas_item_render_commands(Item *p_item, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material);
 
 	// high level batch funcs
 	void canvas_render_items_implementation(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
 	void render_joined_item(const BItemJoined &p_bij, RenderItemState &r_ris);
-	void record_items(Item *p_item_list, int p_z);
-	void join_items(Item *p_item_list, int p_z);
-	void join_sorted_items();
 	bool try_join_item(Item *p_ci, RenderItemState &r_ris, bool &r_batch_break);
-	void render_joined_item_commands(const BItemJoined &p_bij, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material, bool p_lit);
 	void render_batches(Item::Command *const *p_commands, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material);
 
-	bool prefill_joined_item(FillState &r_fill_state, int &r_command_start, Item *p_item, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material);
-
-	void flush_render_batches(Item *p_first_item, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES2::Material *p_material);
-
 	// low level batch funcs
-	int _batch_find_or_create_tex(const RID &p_texture, const RID &p_normal, bool p_tile, int p_previous_match);
-	RasterizerStorageGLES2::Texture *_get_canvas_texture(const RID &p_texture) const;
 	void _batch_upload_buffers();
 	void _batch_render_rects(const Batch &p_batch, RasterizerStorageGLES2::Material *p_material);
-	BatchVertex *_batch_vertex_request_new() { return bdata.vertices.request(); }
+	void _batch_render_polys(const Batch &p_batch, RasterizerStorageGLES2::Material *p_material);
-	Batch *_batch_request_new(bool p_blank = true);
+	void _batch_render_lines(const Batch &p_batch, RasterizerStorageGLES2::Material *p_material, bool p_anti_alias);
 
-	bool _detect_batch_break(Item *p_ci);
+	// funcs used from rasterizer_canvas_batcher template
-	void _software_transform_vertex(BatchVector2 &r_v, const Transform2D &p_tr) const;
+	void gl_enable_scissor(int p_x, int p_y, int p_width, int p_height) const;
-	void _software_transform_vertex(Vector2 &r_v, const Transform2D &p_tr) const;
+	void gl_disable_scissor() const;
-	TransformMode _find_transform_mode(const Transform2D &p_tr) const;
-	void _prefill_default_batch(FillState &r_fill_state, int p_command_num, const Item &p_item);
-
-	// sorting
-	void sort_items();
-	bool sort_items_from(int p_start);
-	bool _sort_items_match(const BSortItem &p_a, const BSortItem &p_b) const;
-
-	// light scissoring
-	bool _light_find_intersection(const Rect2 &p_item_rect, const Transform2D &p_light_xform, const Rect2 &p_light_rect, Rect2 &r_cliprect) const;
-	bool _light_scissor_begin(const Rect2 &p_item_rect, const Transform2D &p_light_xform, const Rect2 &p_light_rect) const;
-	void _calculate_scissor_threshold_area();
-
-	// no need to compile these in in release, they are unneeded outside the editor and only add to executable size
-#ifdef DEBUG_ENABLED
-	void diagnose_batches(Item::Command *const *p_commands);
-	String get_command_type_string(const Item::Command &p_command) const;
-#endif
 
 public:
 	void initialize();
 	RasterizerCanvasGLES2();
-
-private:
-	template <bool SEND_LIGHT_ANGLES>
-	bool prefill_rect(Item::CommandRect *rect, FillState &r_fill_state, int &r_command_start, int command_num, int command_count, Item::Command *const *commands, Item *p_item, bool multiply_final_modulate);
-
-	template <class BATCH_VERTEX_TYPE, bool INCLUDE_LIGHT_ANGLES>
-	void _translate_batches_to_larger_FVF();
 };
 
-//////////////////////////////////////////////////////////////
-
-// Default batches will not occur in software transform only items
-// EXCEPT IN THE CASE OF SINGLE RECTS (and this may well not occur, check the logic in prefill_join_item TYPE_RECT)
-// but can occur where transform commands have been sent during hardware batch
-inline void RasterizerCanvasGLES2::_prefill_default_batch(FillState &r_fill_state, int p_command_num, const Item &p_item) {
-	if (r_fill_state.curr_batch->type == Batch::BT_DEFAULT) {
-		// don't need to flush an extra transform command?
-		if (!r_fill_state.transform_extra_command_number_p1) {
-			// another default command, just add to the existing batch
-			r_fill_state.curr_batch->num_commands++;
-		} else {
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-			if (r_fill_state.transform_extra_command_number_p1 != p_command_num) {
-				WARN_PRINT_ONCE("_prefill_default_batch : transform_extra_command_number_p1 != p_command_num");
-			}
-#endif
-			// if the first member of the batch is a transform we have to be careful
-			if (!r_fill_state.curr_batch->num_commands) {
-				// there can be leading useless extra transforms (sometimes happens with debug collision polys)
-				// we need to rejig the first_command for the first useful transform
-				r_fill_state.curr_batch->first_command += r_fill_state.transform_extra_command_number_p1 - 1;
-			}
-
-			// we do have a pending extra transform command to flush
-			// either the extra transform is in the prior command, or not, in which case we need 2 batches
-			r_fill_state.curr_batch->num_commands += 2;
-
-			r_fill_state.transform_extra_command_number_p1 = 0; // mark as sent
-			r_fill_state.extra_matrix_sent = true;
-
-			// the original mode should always be hardware transform ..
-			// test this assumption
-			//CRASH_COND(r_fill_state.orig_transform_mode != TM_NONE);
-			r_fill_state.transform_mode = r_fill_state.orig_transform_mode;
-
-			// do we need to restore anything else?
-		}
-	} else {
-		// end of previous different type batch, so start new default batch
-
-		// first consider whether there is a dirty extra matrix to send
-		if (r_fill_state.transform_extra_command_number_p1) {
-			// get which command the extra is in, and blank all the records as it no longer is stored CPU side
-			int extra_command = r_fill_state.transform_extra_command_number_p1 - 1; // plus 1 based
-			r_fill_state.transform_extra_command_number_p1 = 0;
-			r_fill_state.extra_matrix_sent = true;
-
-			// send the extra to the GPU in a batch
-			r_fill_state.curr_batch = _batch_request_new();
-			r_fill_state.curr_batch->type = Batch::BT_DEFAULT;
-			r_fill_state.curr_batch->first_command = extra_command;
-			r_fill_state.curr_batch->num_commands = 1;
-
-			// revert to the original transform mode
-			// e.g. go back to NONE if we were in hardware transform mode
-			r_fill_state.transform_mode = r_fill_state.orig_transform_mode;
-
-			// reset the original transform if we are going back to software mode,
-			// because the extra is now done on the GPU...
-			// (any subsequent extras are sent directly to the GPU, no deferring)
-			if (r_fill_state.orig_transform_mode != TM_NONE) {
-				r_fill_state.transform_combined = p_item.final_transform;
-			}
-
-			// can possibly combine batch with the next one in some cases
-			// this is more efficient than having an extra batch especially for the extra
-			if ((extra_command + 1) == p_command_num) {
-				r_fill_state.curr_batch->num_commands = 2;
-				return;
-			}
-		}
-
-		// start default batch
-		r_fill_state.curr_batch = _batch_request_new();
-		r_fill_state.curr_batch->type = Batch::BT_DEFAULT;
-		r_fill_state.curr_batch->first_command = p_command_num;
-		r_fill_state.curr_batch->num_commands = 1;
-	}
-}
-
-inline void RasterizerCanvasGLES2::_software_transform_vertex(BatchVector2 &r_v, const Transform2D &p_tr) const {
-	Vector2 vc(r_v.x, r_v.y);
-	vc = p_tr.xform(vc);
-	r_v.set(vc);
-}
-
-inline void RasterizerCanvasGLES2::_software_transform_vertex(Vector2 &r_v, const Transform2D &p_tr) const {
-	r_v = p_tr.xform(r_v);
-}
-
-inline RasterizerCanvasGLES2::TransformMode RasterizerCanvasGLES2::_find_transform_mode(const Transform2D &p_tr) const {
-	// decided whether to do translate only for software transform
-	if ((p_tr.elements[0].x == 1.0) &&
-			(p_tr.elements[0].y == 0.0) &&
-			(p_tr.elements[1].x == 0.0) &&
-			(p_tr.elements[1].y == 1.0)) {
-		return TM_TRANSLATE;
-	}
-
-	return TM_ALL;
-}
-
-inline bool RasterizerCanvasGLES2::_sort_items_match(const BSortItem &p_a, const BSortItem &p_b) const {
-	const Item *a = p_a.item;
-	const Item *b = p_b.item;
-
-	if (b->commands.size() != 1)
-		return false;
-
-	// tested outside function
-	//	if (a->commands.size() != 1)
-	//		return false;
-
-	const Item::Command &cb = *b->commands[0];
-	if (cb.type != Item::Command::TYPE_RECT)
-		return false;
-
-	const Item::Command &ca = *a->commands[0];
-	// tested outside function
-	//	if (ca.type != Item::Command::TYPE_RECT)
-	//		return false;
-
-	const Item::CommandRect *rect_a = static_cast<const Item::CommandRect *>(&ca);
-	const Item::CommandRect *rect_b = static_cast<const Item::CommandRect *>(&cb);
-
-	if (rect_a->texture != rect_b->texture)
-		return false;
-
-	return true;
-}
-
-//////////////////////////////////////////////////////////////
-// TEMPLATE FUNCS
-
-// Translation always involved adding color to the FVF, which enables
-// joining of batches that have different colors.
-// There is a trade off. Non colored verts are smaller so work faster, but
-// there comes a point where it is better to just use colored verts to avoid lots of
-// batches.
-// In addition this can optionally add light angles to the FVF, necessary for normal mapping.
-template <class BATCH_VERTEX_TYPE, bool INCLUDE_LIGHT_ANGLES>
-void RasterizerCanvasGLES2::_translate_batches_to_larger_FVF() {
-
-	// zeros the size and sets up how big each unit is
-	bdata.unit_vertices.prepare(sizeof(BATCH_VERTEX_TYPE));
-	bdata.batches_temp.reset();
-
-	// As the vertices_colored and batches_temp are 'mirrors' of the non-colored version,
-	// the sizes should be equal, and allocations should never fail. Hence the use of debug
-	// asserts to check program flow, these should not occur at runtime unless the allocation
-	// code has been altered.
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-	CRASH_COND(bdata.unit_vertices.max_size() != bdata.vertices.max_size());
-	CRASH_COND(bdata.batches_temp.max_size() != bdata.batches.max_size());
-#endif
-
-	Color curr_col(-1.0, -1.0, -1.0, -1.0);
-
-	Batch *dest_batch = 0;
-
-	const float *source_light_angles = &bdata.light_angles[0];
-
-	// translate the batches into vertex colored batches
-	for (int n = 0; n < bdata.batches.size(); n++) {
-		const Batch &source_batch = bdata.batches[n];
-
-		// does source batch use light angles?
-		const BatchTex &btex = bdata.batch_textures[source_batch.batch_texture_id];
-		bool source_batch_uses_light_angles = btex.RID_normal != RID();
-
-		bool needs_new_batch = true;
-
-		if (dest_batch) {
-			if (dest_batch->type == source_batch.type) {
-				if (source_batch.type == Batch::BT_RECT) {
-					if (dest_batch->batch_texture_id == source_batch.batch_texture_id) {
-						// add to previous batch
-						dest_batch->num_commands += source_batch.num_commands;
-						needs_new_batch = false;
-
-						// create the colored verts (only if not default)
-						int first_vert = source_batch.first_quad * 4;
-						int end_vert = 4 * (source_batch.first_quad + source_batch.num_commands);
-
-						for (int v = first_vert; v < end_vert; v++) {
-							const BatchVertex &bv = bdata.vertices[v];
-							BATCH_VERTEX_TYPE *cv = (BatchVertexLightAngled *)bdata.unit_vertices.request();
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-							CRASH_COND(!cv);
-#endif
-							cv->pos = bv.pos;
-							cv->uv = bv.uv;
-							cv->col = source_batch.color;
-
-							if (INCLUDE_LIGHT_ANGLES) {
-								// this is required to allow compilation with non light angle vertex.
-								// it should be compiled out.
-								BatchVertexLightAngled *lv = (BatchVertexLightAngled *)cv;
-								if (source_batch_uses_light_angles)
-									lv->light_angle = *source_light_angles++;
-								else
-									lv->light_angle = 0.0f; // dummy, unused in vertex shader (could possibly be left uninitialized, but probably bad idea)
-							}
-						}
-					} // textures match
-				} else {
-					// default
-					// we can still join, but only under special circumstances
-					// does this ever happen? not sure at this stage, but left for future expansion
-					uint32_t source_last_command = source_batch.first_command + source_batch.num_commands;
-					if (source_last_command == dest_batch->first_command) {
-						dest_batch->num_commands += source_batch.num_commands;
-						needs_new_batch = false;
-					} // if the commands line up exactly
-				}
-			} // if both batches are the same type
-
-		} // if dest batch is valid
-
-		if (needs_new_batch) {
-			dest_batch = bdata.batches_temp.request();
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-			CRASH_COND(!dest_batch);
-#endif
-
-			*dest_batch = source_batch;
-
-			// create the colored verts (only if not default)
-			if (source_batch.type != Batch::BT_DEFAULT) {
-				int first_vert = source_batch.first_quad * 4;
-				int end_vert = 4 * (source_batch.first_quad + source_batch.num_commands);
-
-				for (int v = first_vert; v < end_vert; v++) {
-					const BatchVertex &bv = bdata.vertices[v];
-					BATCH_VERTEX_TYPE *cv = (BatchVertexLightAngled *)bdata.unit_vertices.request();
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-					CRASH_COND(!cv);
-#endif
-					cv->pos = bv.pos;
-					cv->uv = bv.uv;
-					cv->col = source_batch.color;
-
-					if (INCLUDE_LIGHT_ANGLES) {
-						// this is required to allow compilation with non light angle vertex.
-						// it should be compiled out.
-						BatchVertexLightAngled *lv = (BatchVertexLightAngled *)cv;
-						if (source_batch_uses_light_angles)
-							lv->light_angle = *source_light_angles++;
-						else
-							lv->light_angle = 0.0f; // dummy, unused in vertex shader (could possibly be left uninitialized, but probably bad idea)
-					} // if using light angles
-				}
-			}
-		}
-	}
-
-	// copy the temporary batches to the master batch list (this could be avoided but it makes the code cleaner)
-	bdata.batches.copy_from(bdata.batches_temp);
-}
-
-// return true if buffer full up, else return false
-template <bool SEND_LIGHT_ANGLES>
-bool RasterizerCanvasGLES2::prefill_rect(Item::CommandRect *rect, FillState &r_fill_state, int &r_command_start, int command_num, int command_count, Item::Command *const *commands, Item *p_item, bool multiply_final_modulate) {
-	bool change_batch = false;
-
-	// conditions for creating a new batch
-	if (r_fill_state.curr_batch->type != Batch::BT_RECT) {
-		change_batch = true;
-
-		// check for special case if there is only a single or small number of rects,
-		// in which case we will use the legacy default rect renderer
-		// because it is faster for single rects
-
-		// we only want to do this if not a joined item with more than 1 item,
-		// because joined items with more than 1, the command * will be incorrect
-		// NOTE - this is assuming that use_hardware_transform means that it is a non-joined item!!
-		// If that assumption is incorrect this will go horribly wrong.
-		if (bdata.settings_use_single_rect_fallback && r_fill_state.use_hardware_transform) {
-			bool is_single_rect = false;
-			int command_num_next = command_num + 1;
-			if (command_num_next < command_count) {
-				Item::Command *command_next = commands[command_num_next];
-				if ((command_next->type != Item::Command::TYPE_RECT) && (command_next->type != Item::Command::TYPE_TRANSFORM)) {
-					is_single_rect = true;
-				}
-			} else {
-				is_single_rect = true;
-			}
-			// if it is a rect on its own, do exactly the same as the default routine
-			if (is_single_rect) {
-				_prefill_default_batch(r_fill_state, command_num, *p_item);
-				return false;
-			}
-		} // if use hardware transform
-	}
-
-	Color col = rect->modulate;
-	if (multiply_final_modulate) {
-		col *= r_fill_state.final_modulate;
-	}
-
-	// instead of doing all the texture preparation for EVERY rect,
-	// we build a list of texture combinations and do this once off.
-	// This means we have a potentially rather slow step to identify which texture combo
-	// using the RIDs.
-	int old_batch_tex_id = r_fill_state.batch_tex_id;
-	r_fill_state.batch_tex_id = _batch_find_or_create_tex(rect->texture, rect->normal_map, rect->flags & CANVAS_RECT_TILE, old_batch_tex_id);
-
-	//r_fill_state.use_light_angles = send_light_angles;
-	if (SEND_LIGHT_ANGLES)
-		bdata.use_light_angles = true;
-
-	// try to create vertices BEFORE creating a batch,
-	// because if the vertex buffer is full, we need to finish this
-	// function, draw what we have so far, and then start a new set of batches
-
-	// request FOUR vertices at a time, this is more efficient
-	BatchVertex *bvs = bdata.vertices.request(4);
-	if (!bvs) {
-		// run out of space in the vertex buffer .. finish this function and draw what we have so far
-		// return where we got to
-		r_command_start = command_num;
-		return true;
-	}
-
-	// conditions for creating a new batch
-	if (old_batch_tex_id != r_fill_state.batch_tex_id) {
-		change_batch = true;
-	}
-
-	// we need to treat color change separately because we need to count these
-	// to decide whether to switch on the fly to colored vertices.
-	if (!r_fill_state.curr_batch->color.equals(col)) {
-		change_batch = true;
-		bdata.total_color_changes++;
-	}
-
-	if (change_batch) {
-		// put the tex pixel size  in a local (less verbose and can be a register)
-		const BatchTex &batchtex = bdata.batch_textures[r_fill_state.batch_tex_id];
-		batchtex.tex_pixel_size.to(r_fill_state.texpixel_size);
-
-		if (bdata.settings_uv_contract) {
-			r_fill_state.contract_uvs = (batchtex.flags & VS::TEXTURE_FLAG_FILTER) == 0;
-		}
-
-		// need to preserve texpixel_size between items
-		r_fill_state.texpixel_size = r_fill_state.texpixel_size;
-
-		// open new batch (this should never fail, it dynamically grows)
-		r_fill_state.curr_batch = _batch_request_new(false);
-
-		r_fill_state.curr_batch->type = Batch::BT_RECT;
-		r_fill_state.curr_batch->color.set(col);
-		r_fill_state.curr_batch->batch_texture_id = r_fill_state.batch_tex_id;
-		r_fill_state.curr_batch->first_command = command_num;
-		r_fill_state.curr_batch->num_commands = 1;
-		r_fill_state.curr_batch->first_quad = bdata.total_quads;
-	} else {
-		// we could alternatively do the count when closing a batch .. perhaps more efficient
-		r_fill_state.curr_batch->num_commands++;
-	}
-
-	// fill the quad geometry
-	Vector2 mins = rect->rect.position;
-
-	if (r_fill_state.transform_mode == TM_TRANSLATE) {
-		_software_transform_vertex(mins, r_fill_state.transform_combined);
-	}
-
-	Vector2 maxs = mins + rect->rect.size;
-
-	// just aliases
-	BatchVertex *bA = &bvs[0];
-	BatchVertex *bB = &bvs[1];
-	BatchVertex *bC = &bvs[2];
-	BatchVertex *bD = &bvs[3];
-
-	bA->pos.x = mins.x;
-	bA->pos.y = mins.y;
-
-	bB->pos.x = maxs.x;
-	bB->pos.y = mins.y;
-
-	bC->pos.x = maxs.x;
-	bC->pos.y = maxs.y;
-
-	bD->pos.x = mins.x;
-	bD->pos.y = maxs.y;
-
-	// possibility of applying flips here for normal mapping .. but they don't seem to be used
-	if (rect->rect.size.x < 0) {
-		SWAP(bA->pos, bB->pos);
-		SWAP(bC->pos, bD->pos);
-	}
-	if (rect->rect.size.y < 0) {
-		SWAP(bA->pos, bD->pos);
-		SWAP(bB->pos, bC->pos);
-	}
-
-	if (r_fill_state.transform_mode == TM_ALL) {
-		_software_transform_vertex(bA->pos, r_fill_state.transform_combined);
-		_software_transform_vertex(bB->pos, r_fill_state.transform_combined);
-		_software_transform_vertex(bC->pos, r_fill_state.transform_combined);
-		_software_transform_vertex(bD->pos, r_fill_state.transform_combined);
-	}
-
-	// uvs
-	Vector2 src_min;
-	Vector2 src_max;
-	if (rect->flags & CANVAS_RECT_REGION) {
-		src_min = rect->source.position;
-		src_max = src_min + rect->source.size;
-
-		src_min *= r_fill_state.texpixel_size;
-		src_max *= r_fill_state.texpixel_size;
-
-		const float uv_epsilon = bdata.settings_uv_contract_amount;
-
-		// nudge offset for the maximum to prevent precision error on GPU reading into line outside the source rect
-		// this is very difficult to get right.
-		if (r_fill_state.contract_uvs) {
-			src_min.x += uv_epsilon;
-			src_min.y += uv_epsilon;
-			src_max.x -= uv_epsilon;
-			src_max.y -= uv_epsilon;
-		}
-	} else {
-		src_min = Vector2(0, 0);
-		src_max = Vector2(1, 1);
-	}
-
-	// 10% faster calculating the max first
-	Vector2 uvs[4] = {
-		src_min,
-		Vector2(src_max.x, src_min.y),
-		src_max,
-		Vector2(src_min.x, src_max.y),
-	};
-
-	// for encoding in light angle
-	// flips should be optimized out when not being used for light angle.
-	bool flip_h = false;
-	bool flip_v = false;
-
-	if (rect->flags & CANVAS_RECT_TRANSPOSE) {
-		SWAP(uvs[1], uvs[3]);
-	}
-
-	if (rect->flags & CANVAS_RECT_FLIP_H) {
-		SWAP(uvs[0], uvs[1]);
-		SWAP(uvs[2], uvs[3]);
-		flip_h = !flip_h;
-		flip_v = !flip_v;
-	}
-	if (rect->flags & CANVAS_RECT_FLIP_V) {
-		SWAP(uvs[0], uvs[3]);
-		SWAP(uvs[1], uvs[2]);
-		flip_v = !flip_v;
-	}
-
-	bA->uv.set(uvs[0]);
-	bB->uv.set(uvs[1]);
-	bC->uv.set(uvs[2]);
-	bD->uv.set(uvs[3]);
-
-	if (SEND_LIGHT_ANGLES) {
-		// we can either keep the light angles in sync with the verts when writing,
-		// or sync them up during translation. We are syncing in translation.
-		// N.B. There may be batches that don't require light_angles between batches that do.
-		float *angles = bdata.light_angles.request(4);
-#if defined(TOOLS_ENABLED) && defined(DEBUG_ENABLED)
-		CRASH_COND(angles == nullptr);
-#endif
-
-		float angle = 0.0f;
-		const float TWO_PI = Math_PI * 2;
-
-		if (r_fill_state.transform_mode != TM_NONE) {
-
-			const Transform2D &tr = r_fill_state.transform_combined;
-
-			// apply to an x axis
-			// the x axis and y axis can be taken directly from the transform (no need to xform identity vectors)
-			Vector2 x_axis(tr.elements[0][0], tr.elements[1][0]);
-
-			// have to do a y axis to check for scaling flips
-			// this is hassle and extra slowness. We could only allow flips via the flags.
-			Vector2 y_axis(tr.elements[0][1], tr.elements[1][1]);
-
-			// has the x / y axis flipped due to scaling?
-			float cross = x_axis.cross(y_axis);
-			if (cross < 0.0f) {
-				flip_v = !flip_v;
-			}
-
-			// passing an angle is smaller than a vector, it can be reconstructed in the shader
-			angle = x_axis.angle();
-
-			// we don't want negative angles, as negative is used to encode flips.
-			// This moves range from -PI to PI to 0 to TWO_PI
-			if (angle < 0.0f)
-				angle += TWO_PI;
-
-		} // if transform needed
-
-		// if horizontal flip, angle is shifted by 180 degrees
-		if (flip_h) {
-			angle += Math_PI;
-
-			// mod to get back to 0 to TWO_PI range
-			angle = fmodf(angle, TWO_PI);
-		}
-
-		// add 1 (to take care of zero floating point error with sign)
-		angle += 1.0f;
-
-		// flip if necessary to indicate a vertical flip in the shader
-		if (flip_v)
-			angle *= -1.0f;
-
-		// light angle must be sent for each vert, instead as a single uniform in the uniform draw method
-		// this has the benefit of enabling batching with light angles.
-		for (int n = 0; n < 4; n++) {
-			angles[n] = angle;
-		}
-	}
-
-	// increment quad count
-	bdata.total_quads++;
-
-	return false;
-}
-
 #endif // RASTERIZERCANVASGLES2_H

drivers/gles2/rasterizer_gles2.cpp

@@ -483,6 +483,42 @@ void RasterizerGLES2::make_current() {
 void RasterizerGLES2::register_config() {
 }
 
+// returns NULL if no error, or an error string
+const char *RasterizerGLES2::gl_check_for_error(bool p_print_error) {
+	GLenum err = glGetError();
+
+	const char *err_string = nullptr;
+
+	switch (err) {
+		default: {
+			// not recognised
+		} break;
+		case GL_NO_ERROR: {
+		} break;
+		case GL_INVALID_ENUM: {
+			err_string = "GL_INVALID_ENUM";
+		} break;
+		case GL_INVALID_VALUE: {
+			err_string = "GL_INVALID_VALUE";
+		} break;
+		case GL_INVALID_OPERATION: {
+			err_string = "GL_INVALID_OPERATION";
+		} break;
+		case GL_INVALID_FRAMEBUFFER_OPERATION: {
+			err_string = "GL_INVALID_FRAMEBUFFER_OPERATION";
+		} break;
+		case GL_OUT_OF_MEMORY: {
+			err_string = "GL_OUT_OF_MEMORY";
+		} break;
+	}
+
+	if (p_print_error && err_string) {
+		print_line(err_string);
+	}
+
+	return err_string;
+}
+
 RasterizerGLES2::RasterizerGLES2() {
 
 	storage = memnew(RasterizerStorageGLES2);

drivers/gles2/rasterizer_gles2.h

@@ -71,6 +71,8 @@ public:
 
 	virtual bool is_low_end() const { return true; }
 
+	virtual const char *gl_check_for_error(bool p_print_error = true);
+
 	RasterizerGLES2();
 	~RasterizerGLES2();
 };

drivers/gles2/rasterizer_storage_gles2.cpp

@@ -1452,6 +1452,7 @@ void RasterizerStorageGLES2::_update_shader(Shader *p_shader) const {
 			shaders.actions_canvas.usage_flag_pointers["TIME"] = &p_shader->canvas_item.uses_time;
 			shaders.actions_canvas.usage_flag_pointers["MODULATE"] = &p_shader->canvas_item.uses_modulate;
 			shaders.actions_canvas.usage_flag_pointers["COLOR"] = &p_shader->canvas_item.uses_color;
+
 			shaders.actions_canvas.usage_flag_pointers["VERTEX"] = &p_shader->canvas_item.uses_vertex;
 
 			actions = &shaders.actions_canvas;
@@ -1552,10 +1553,10 @@ void RasterizerStorageGLES2::_update_shader(Shader *p_shader) const {
 	// some logic for batching
 	if (p_shader->mode == VS::SHADER_CANVAS_ITEM) {
 		if (p_shader->canvas_item.uses_modulate | p_shader->canvas_item.uses_color) {
-			p_shader->canvas_item.batch_flags |= Shader::CanvasItem::PREVENT_COLOR_BAKING;
+			p_shader->canvas_item.batch_flags |= RasterizerStorageCommon::PREVENT_COLOR_BAKING;
 		}
 		if (p_shader->canvas_item.uses_vertex) {
-			p_shader->canvas_item.batch_flags |= Shader::CanvasItem::PREVENT_VERTEX_BAKING;
+			p_shader->canvas_item.batch_flags |= RasterizerStorageCommon::PREVENT_VERTEX_BAKING;
 		}
 	}
 

drivers/gles2/rasterizer_storage_gles2.h

@@ -40,11 +40,6 @@
 
 #include "shaders/copy.glsl.gen.h"
 #include "shaders/cubemap_filter.glsl.gen.h"
-/*
-#include "shaders/blend_shape.glsl.gen.h"
-#include "shaders/canvas.glsl.gen.h"
-#include "shaders/particles.glsl.gen.h"
-*/
 
 class RasterizerCanvasGLES2;
 class RasterizerSceneGLES2;
@@ -450,10 +445,7 @@ public:
 			// these flags are specifically for batching
 			// some of the logic is thus in rasterizer_storage.cpp
 			// we could alternatively set bitflags for each 'uses' and test on the fly
-			enum BatchFlags {
+			// defined in RasterizerStorageCommon::BatchFlags
-				PREVENT_COLOR_BAKING = 1 << 0,
-				PREVENT_VERTEX_BAKING = 1 << 1,
-			};
 			unsigned int batch_flags;
 
 			bool uses_screen_texture;

drivers/gles2/shaders/canvas.glsl

@@ -19,7 +19,7 @@ uniform highp mat4 modelview_matrix;
 uniform highp mat4 extra_matrix;
 attribute highp vec2 vertex; // attrib:0
 
-#ifdef USE_LIGHT_ANGLE
+#ifdef USE_ATTRIB_LIGHT_ANGLE
 // shared with tangent, not used in canvas shader
 attribute highp float light_angle; // attrib:2
 #endif
@@ -27,6 +27,16 @@ attribute highp float light_angle; // attrib:2
 attribute vec4 color_attrib; // attrib:3
 attribute vec2 uv_attrib; // attrib:4
 
+#ifdef USE_ATTRIB_MODULATE
+attribute highp vec4 modulate_attrib; // attrib:5
+#endif
+
+#ifdef USE_ATTRIB_LARGE_VERTEX
+// shared with skeleton attributes, not used in batched shader
+attribute highp vec2 translate_attrib; // attrib:6
+attribute highp vec4 basis_attrib; // attrib:7
+#endif
+
 #ifdef USE_SKELETON
 attribute highp vec4 bone_indices; // attrib:6
 attribute highp vec4 bone_weights; // attrib:7
@@ -54,6 +64,12 @@ uniform highp mat4 skeleton_transform_inverse;
 
 varying vec2 uv_interp;
 varying vec4 color_interp;
+
+#ifdef USE_ATTRIB_MODULATE
+// modulate doesn't need interpolating but we need to send it to the fragment shader
+varying vec4 modulate_interp;
+#endif
+
 #ifdef MODULATE_USED
 uniform vec4 final_modulate;
 #endif
@@ -171,6 +187,24 @@ VERTEX_SHADER_CODE
 
 	gl_PointSize = point_size;
 
+#ifdef USE_ATTRIB_MODULATE
+	// modulate doesn't need interpolating but we need to send it to the fragment shader
+	modulate_interp = modulate_attrib;
+#endif
+
+#ifdef USE_ATTRIB_LARGE_VERTEX
+	// transform is in attributes
+	vec2 temp;
+
+	temp = outvec.xy;
+	temp.x = (outvec.x * basis_attrib.x) + (outvec.y * basis_attrib.z);
+	temp.y = (outvec.x * basis_attrib.y) + (outvec.y * basis_attrib.w);
+
+	temp += translate_attrib;
+	outvec.xy = temp;
+
+#endif
+
 #if !defined(SKIP_TRANSFORM_USED)
 	outvec = extra_matrix_instance * outvec;
 	outvec = modelview_matrix * outvec;
@@ -225,7 +259,7 @@ VERTEX_SHADER_CODE
 	pos = outvec.xy;
 #endif
 
-#ifdef USE_LIGHT_ANGLE
+#ifdef USE_ATTRIB_LIGHT_ANGLE
 	// we add a fixed offset because we are using the sign later,
 	// and don't want floating point error around 0.0
 	float la = abs(light_angle) - 1.0;
@@ -303,6 +337,10 @@ uniform mediump sampler2D normal_texture; // texunit:-2
 varying mediump vec2 uv_interp;
 varying mediump vec4 color_interp;
 
+#ifdef USE_ATTRIB_MODULATE
+varying mediump vec4 modulate_interp;
+#endif
+
 uniform highp float time;
 
 uniform vec4 final_modulate;
@@ -442,6 +480,11 @@ FRAGMENT_SHADER_CODE
 	color *= final_modulate;
 #endif
 
+#ifdef USE_ATTRIB_MODULATE
+	// todo .. this won't be used at the same time as MODULATE_USED
+	color *= modulate_interp;
+#endif
+
 #ifdef USE_LIGHTING
 
 	vec2 light_vec = transformed_light_uv;

drivers/gles3/rasterizer_canvas_base_gles3.h

@@ -0,0 +1,163 @@
+/*************************************************************************/
+/*  rasterizer_canvas_base_gles3.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 RASTERIZERCANVASBASEGLES3_H
+#define RASTERIZERCANVASBASEGLES3_H
+
+#include "rasterizer_storage_gles3.h"
+#include "servers/visual/rasterizer.h"
+
+#include "shaders/canvas_shadow.glsl.gen.h"
+#include "shaders/lens_distorted.glsl.gen.h"
+
+class RasterizerSceneGLES3;
+
+class RasterizerCanvasBaseGLES3 : public RasterizerCanvas {
+public:
+	struct CanvasItemUBO {
+
+		float projection_matrix[16];
+		float time;
+		uint8_t padding[12];
+	};
+
+	RasterizerSceneGLES3 *scene_render;
+
+	struct Data {
+
+		enum { NUM_QUAD_ARRAY_VARIATIONS = 8 };
+
+		GLuint canvas_quad_vertices;
+		GLuint canvas_quad_array;
+
+		GLuint polygon_buffer;
+		GLuint polygon_buffer_quad_arrays[NUM_QUAD_ARRAY_VARIATIONS];
+		GLuint polygon_buffer_pointer_array;
+		GLuint polygon_index_buffer;
+
+		GLuint particle_quad_vertices;
+		GLuint particle_quad_array;
+
+		uint32_t polygon_buffer_size;
+		uint32_t polygon_index_buffer_size;
+
+	} data;
+
+	struct State {
+		CanvasItemUBO canvas_item_ubo_data;
+		GLuint canvas_item_ubo;
+		bool canvas_texscreen_used;
+		CanvasShaderGLES3 canvas_shader;
+		CanvasShadowShaderGLES3 canvas_shadow_shader;
+		LensDistortedShaderGLES3 lens_shader;
+
+		bool using_texture_rect;
+		bool using_ninepatch;
+
+		bool using_light_angle;
+		bool using_modulate;
+		bool using_large_vertex;
+
+		RID current_tex;
+		RID current_normal;
+		RasterizerStorageGLES3::Texture *current_tex_ptr;
+
+		Transform vp;
+
+		Color canvas_item_modulate;
+		Transform2D extra_matrix;
+		Transform2D final_transform;
+		bool using_skeleton;
+		Transform2D skeleton_transform;
+		Transform2D skeleton_transform_inverse;
+
+	} state;
+
+	RasterizerStorageGLES3 *storage;
+
+	struct LightInternal : public RID_Data {
+
+		struct UBOData {
+
+			float light_matrix[16];
+			float local_matrix[16];
+			float shadow_matrix[16];
+			float color[4];
+			float shadow_color[4];
+			float light_pos[2];
+			float shadowpixel_size;
+			float shadow_gradient;
+			float light_height;
+			float light_outside_alpha;
+			float shadow_distance_mult;
+			uint8_t padding[4];
+		} ubo_data;
+
+		GLuint ubo;
+	};
+
+	RID_Owner<LightInternal> light_internal_owner;
+
+	virtual RID light_internal_create();
+	virtual void light_internal_update(RID p_rid, Light *p_light);
+	virtual void light_internal_free(RID p_rid);
+
+	virtual void canvas_begin();
+	virtual void canvas_end();
+
+	void _set_texture_rect_mode(bool p_enable, bool p_ninepatch = false, bool p_light_angle = false, bool p_modulate = false, bool p_large_vertex = false);
+	RasterizerStorageGLES3::Texture *_bind_canvas_texture(const RID &p_texture, const RID &p_normal_map, bool p_force = false);
+
+	void _draw_gui_primitive(int p_points, const Vector2 *p_vertices, const Color *p_colors, const Vector2 *p_uvs, const float *p_light_angles = nullptr);
+	void _draw_polygon(const int *p_indices, int p_index_count, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor, const int *p_bones, const float *p_weights);
+	void _draw_generic(GLuint p_primitive, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor);
+	void _draw_generic_indices(GLuint p_primitive, const int *p_indices, int p_index_count, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor);
+
+	void _copy_texscreen(const Rect2 &p_rect);
+
+	virtual void canvas_debug_viewport_shadows(Light *p_lights_with_shadow);
+
+	virtual void canvas_light_shadow_buffer_update(RID p_buffer, const Transform2D &p_light_xform, int p_light_mask, float p_near, float p_far, LightOccluderInstance *p_occluders, CameraMatrix *p_xform_cache);
+
+	virtual void reset_canvas();
+
+	void draw_generic_textured_rect(const Rect2 &p_rect, const Rect2 &p_src);
+	void draw_lens_distortion_rect(const Rect2 &p_rect, float p_k1, float p_k2, const Vector2 &p_eye_center, float p_oversample);
+	void render_rect_nvidia_workaround(const Item::CommandRect *p_rect, const RasterizerStorageGLES3::Texture *p_texture);
+
+	void initialize();
+	void finalize();
+
+	virtual void draw_window_margins(int *black_margin, RID *black_image);
+
+	RasterizerCanvasBaseGLES3();
+};
+
+#endif // RASTERIZERCANVASBASEGLES3_H

drivers/gles3/rasterizer_canvas_gles3.h

@@ -31,132 +31,49 @@
 #ifndef RASTERIZERCANVASGLES3_H
 #define RASTERIZERCANVASGLES3_H
 
-#include "rasterizer_storage_gles3.h"
+#include "drivers/gles_common/rasterizer_canvas_batcher.h"
-#include "servers/visual/rasterizer.h"
+#include "rasterizer_canvas_base_gles3.h"
 
-#include "shaders/canvas_shadow.glsl.gen.h"
+class RasterizerCanvasGLES3 : public RasterizerCanvasBaseGLES3, public RasterizerCanvasBatcher<RasterizerCanvasGLES3, RasterizerStorageGLES3> {
-#include "shaders/lens_distorted.glsl.gen.h"
+	friend class RasterizerCanvasBatcher<RasterizerCanvasGLES3, RasterizerStorageGLES3>;
 
-class RasterizerSceneGLES3;
+private:
+	struct BatchGLData {
+		// for batching
+		GLuint batch_vertex_array[5];
+	} batch_gl_data;
 
-class RasterizerCanvasGLES3 : public RasterizerCanvas {
 public:
-	struct CanvasItemUBO {
+	virtual void canvas_render_items_begin(const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
-
+	virtual void canvas_render_items_end();
-		float projection_matrix[16];
+	virtual void canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
-		float time;
-		uint8_t padding[12];
-	};
-
-	RasterizerSceneGLES3 *scene_render;
-
-	struct Data {
-
-		enum { NUM_QUAD_ARRAY_VARIATIONS = 8 };
-
-		GLuint canvas_quad_vertices;
-		GLuint canvas_quad_array;
-
-		GLuint polygon_buffer;
-		GLuint polygon_buffer_quad_arrays[NUM_QUAD_ARRAY_VARIATIONS];
-		GLuint polygon_buffer_pointer_array;
-		GLuint polygon_index_buffer;
-
-		GLuint particle_quad_vertices;
-		GLuint particle_quad_array;
-
-		uint32_t polygon_buffer_size;
-		uint32_t polygon_index_buffer_size;
-
-	} data;
-
-	struct State {
-		CanvasItemUBO canvas_item_ubo_data;
-		GLuint canvas_item_ubo;
-		bool canvas_texscreen_used;
-		CanvasShaderGLES3 canvas_shader;
-		CanvasShadowShaderGLES3 canvas_shadow_shader;
-		LensDistortedShaderGLES3 lens_shader;
-
-		bool using_texture_rect;
-		bool using_ninepatch;
-		bool using_light_angle;
-
-		RID current_tex;
-		RID current_normal;
-		RasterizerStorageGLES3::Texture *current_tex_ptr;
-
-		Transform vp;
-
-		Color canvas_item_modulate;
-		Transform2D extra_matrix;
-		Transform2D final_transform;
-		bool using_skeleton;
-		Transform2D skeleton_transform;
-		Transform2D skeleton_transform_inverse;
-
-	} state;
-
-	RasterizerStorageGLES3 *storage;
-	bool use_nvidia_rect_workaround;
-
-	struct LightInternal : public RID_Data {
-
-		struct UBOData {
-
-			float light_matrix[16];
-			float local_matrix[16];
-			float shadow_matrix[16];
-			float color[4];
-			float shadow_color[4];
-			float light_pos[2];
-			float shadowpixel_size;
-			float shadow_gradient;
-			float light_height;
-			float light_outside_alpha;
-			float shadow_distance_mult;
-			uint8_t padding[4];
-		} ubo_data;
-
-		GLuint ubo;
-	};
-
-	RID_Owner<LightInternal> light_internal_owner;
-
-	virtual RID light_internal_create();
-	virtual void light_internal_update(RID p_rid, Light *p_light);
-	virtual void light_internal_free(RID p_rid);
-
 	virtual void canvas_begin();
 	virtual void canvas_end();
 
-	_FORCE_INLINE_ void _set_texture_rect_mode(bool p_enable, bool p_ninepatch = false, bool p_light_angle = false);
+private:
-	_FORCE_INLINE_ RasterizerStorageGLES3::Texture *_bind_canvas_texture(const RID &p_texture, const RID &p_normal_map, bool p_force = false);
+	// legacy codepath .. to remove after testing
+	void _legacy_canvas_render_item(Item *p_ci, RenderItemState &r_ris);
 
-	_FORCE_INLINE_ void _draw_gui_primitive(int p_points, const Vector2 *p_vertices, const Color *p_colors, const Vector2 *p_uvs, const float *p_light_angles = nullptr);
+	// high level batch funcs
-	_FORCE_INLINE_ void _draw_polygon(const int *p_indices, int p_index_count, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor, const int *p_bones, const float *p_weights);
+	void canvas_render_items_implementation(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_base_transform);
-	_FORCE_INLINE_ void _draw_generic(GLuint p_primitive, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor);
+	void render_joined_item(const BItemJoined &p_bij, RenderItemState &r_ris);
-	_FORCE_INLINE_ void _draw_generic_indices(GLuint p_primitive, const int *p_indices, int p_index_count, int p_vertex_count, const Vector2 *p_vertices, const Vector2 *p_uvs, const Color *p_colors, bool p_singlecolor);
+	bool try_join_item(Item *p_ci, RenderItemState &r_ris, bool &r_batch_break);
+	void render_batches(Item::Command *const *p_commands, Item *p_current_clip, bool &r_reclip, RasterizerStorageGLES3::Material *p_material);
 
-	_FORCE_INLINE_ void _canvas_item_render_commands(Item *p_item, Item *current_clip, bool &reclip);
+	// low level batch funcs
-	_FORCE_INLINE_ void _copy_texscreen(const Rect2 &p_rect);
+	void _batch_upload_buffers();
+	void _batch_render_rects(const Batch &p_batch, RasterizerStorageGLES3::Material *p_material);
+	void _batch_render_polys(const Batch &p_batch, RasterizerStorageGLES3::Material *p_material);
+	void _batch_render_lines(const Batch &p_batch, RasterizerStorageGLES3::Material *p_material, bool p_anti_alias);
 
-	virtual void canvas_render_items(Item *p_item_list, int p_z, const Color &p_modulate, Light *p_light, const Transform2D &p_transform);
+	// funcs used from rasterizer_canvas_batcher template
-	virtual void canvas_debug_viewport_shadows(Light *p_lights_with_shadow);
+	void gl_enable_scissor(int p_x, int p_y, int p_width, int p_height) const;
+	void gl_disable_scissor() const;
 
-	virtual void canvas_light_shadow_buffer_update(RID p_buffer, const Transform2D &p_light_xform, int p_light_mask, float p_near, float p_far, LightOccluderInstance *p_occluders, CameraMatrix *p_xform_cache);
+	void gl_checkerror();
-
-	virtual void reset_canvas();
-
-	void draw_generic_textured_rect(const Rect2 &p_rect, const Rect2 &p_src);
-	void draw_lens_distortion_rect(const Rect2 &p_rect, float p_k1, float p_k2, const Vector2 &p_eye_center, float p_oversample);
-	void render_rect_nvidia_workaround(const Item::CommandRect *p_rect, const RasterizerStorageGLES3::Texture *p_texture);
 
+public:
 	void initialize();
-	void finalize();
-
-	virtual void draw_window_margins(int *black_margin, RID *black_image);
-
 	RasterizerCanvasGLES3();
 };
 

drivers/gles3/rasterizer_gles3.h

@@ -71,7 +71,7 @@ public:
 
 	virtual bool is_low_end() const { return false; }
 
-	const char *gl_check_for_error(bool p_print_error = true);
+	virtual const char *gl_check_for_error(bool p_print_error = true);
 
 	RasterizerGLES3();
 	~RasterizerGLES3();

drivers/gles3/rasterizer_storage_gles3.cpp

@@ -2305,6 +2305,10 @@ void RasterizerStorageGLES3::_update_shader(Shader *p_shader) const {
 			p_shader->canvas_item.uses_screen_texture = false;
 			p_shader->canvas_item.uses_screen_uv = false;
 			p_shader->canvas_item.uses_time = false;
+			p_shader->canvas_item.uses_modulate = false;
+			p_shader->canvas_item.uses_color = false;
+			p_shader->canvas_item.uses_vertex = false;
+			p_shader->canvas_item.batch_flags = 0;
 
 			shaders.actions_canvas.render_mode_values["blend_add"] = Pair<int *, int>(&p_shader->canvas_item.blend_mode, Shader::CanvasItem::BLEND_MODE_ADD);
 			shaders.actions_canvas.render_mode_values["blend_mix"] = Pair<int *, int>(&p_shader->canvas_item.blend_mode, Shader::CanvasItem::BLEND_MODE_MIX);
@@ -2321,6 +2325,10 @@ void RasterizerStorageGLES3::_update_shader(Shader *p_shader) const {
 			shaders.actions_canvas.usage_flag_pointers["SCREEN_TEXTURE"] = &p_shader->canvas_item.uses_screen_texture;
 			shaders.actions_canvas.usage_flag_pointers["TIME"] = &p_shader->canvas_item.uses_time;
 
+			shaders.actions_canvas.usage_flag_pointers["MODULATE"] = &p_shader->canvas_item.uses_modulate;
+			shaders.actions_canvas.usage_flag_pointers["COLOR"] = &p_shader->canvas_item.uses_color;
+			shaders.actions_canvas.usage_flag_pointers["VERTEX"] = &p_shader->canvas_item.uses_vertex;
+
 			actions = &shaders.actions_canvas;
 			actions->uniforms = &p_shader->uniforms;
 
@@ -2417,6 +2425,16 @@ void RasterizerStorageGLES3::_update_shader(Shader *p_shader) const {
 	p_shader->uses_vertex_time = gen_code.uses_vertex_time;
 	p_shader->uses_fragment_time = gen_code.uses_fragment_time;
 
+	// some logic for batching
+	if (p_shader->mode == VS::SHADER_CANVAS_ITEM) {
+		if (p_shader->canvas_item.uses_modulate | p_shader->canvas_item.uses_color) {
+			p_shader->canvas_item.batch_flags |= RasterizerStorageCommon::PREVENT_COLOR_BAKING;
+		}
+		if (p_shader->canvas_item.uses_vertex) {
+			p_shader->canvas_item.batch_flags |= RasterizerStorageCommon::PREVENT_VERTEX_BAKING;
+		}
+	}
+
 	//all materials using this shader will have to be invalidated, unfortunately
 
 	for (SelfList<Material> *E = p_shader->materials.first(); E; E = E->next()) {
@@ -8313,6 +8331,9 @@ void RasterizerStorageGLES3::initialize() {
 
 #endif
 
+	// not yet detected on GLES3 (is this mandated?)
+	config.support_npot_repeat_mipmap = true;
+
 	config.pvrtc_supported = config.extensions.has("GL_IMG_texture_compression_pvrtc");
 	config.srgb_decode_supported = config.extensions.has("GL_EXT_texture_sRGB_decode");
 

drivers/gles3/rasterizer_storage_gles3.h

@@ -84,6 +84,7 @@ public:
 
 		bool srgb_decode_supported;
 
+		bool support_npot_repeat_mipmap;
 		bool texture_float_linear_supported;
 		bool framebuffer_float_supported;
 		bool framebuffer_half_float_supported;
@@ -455,9 +456,18 @@ public:
 
 			int light_mode;
 
+			// these flags are specifically for batching
+			// some of the logic is thus in rasterizer_storage.cpp
+			// we could alternatively set bitflags for each 'uses' and test on the fly
+			// defined in RasterizerStorageCommon::BatchFlags
+			unsigned int batch_flags;
+
 			bool uses_screen_texture;
 			bool uses_screen_uv;
 			bool uses_time;
+			bool uses_modulate;
+			bool uses_color;
+			bool uses_vertex;
 
 		} canvas_item;
 

drivers/gles3/shaders/canvas.glsl

@@ -3,13 +3,23 @@
 
 layout(location = 0) in highp vec2 vertex;
 
-#ifdef USE_LIGHT_ANGLE
+#ifdef USE_ATTRIB_LIGHT_ANGLE
 layout(location = 2) in highp float light_angle;
 #endif
 
 /* clang-format on */
 layout(location = 3) in vec4 color_attrib;
 
+#ifdef USE_ATTRIB_MODULATE
+layout(location = 5) in vec4 modulate_attrib; // attrib:5
+#endif
+
+#ifdef USE_ATTRIB_LARGE_VERTEX
+// shared with skeleton attributes, not used in batched shader
+layout(location = 6) in vec2 translate_attrib; // attrib:6
+layout(location = 7) in vec4 basis_attrib; // attrib:7
+#endif
+
 #ifdef USE_SKELETON
 layout(location = 6) in uvec4 bone_indices; // attrib:6
 layout(location = 7) in vec4 bone_weights; // attrib:7
@@ -53,6 +63,12 @@ uniform highp mat4 extra_matrix;
 
 out highp vec2 uv_interp;
 out mediump vec4 color_interp;
+
+#ifdef USE_ATTRIB_MODULATE
+// modulate doesn't need interpolating but we need to send it to the fragment shader
+out mediump vec4 modulate_interp;
+#endif
+
 #ifdef MODULATE_USED
 uniform mediump vec4 final_modulate;
 #endif
@@ -177,6 +193,23 @@ VERTEX_SHADER_CODE
 	pixel_size_interp = abs(dst_rect.zw) * vertex;
 #endif
 
+#ifdef USE_ATTRIB_MODULATE
+	// modulate doesn't need interpolating but we need to send it to the fragment shader
+	modulate_interp = modulate_attrib;
+#endif
+
+#ifdef USE_ATTRIB_LARGE_VERTEX
+	// transform is in attributes
+	vec2 temp;
+
+	temp = outvec.xy;
+	temp.x = (outvec.x * basis_attrib.x) + (outvec.y * basis_attrib.z);
+	temp.y = (outvec.x * basis_attrib.y) + (outvec.y * basis_attrib.w);
+
+	temp += translate_attrib;
+	outvec.xy = temp;
+#endif
+
 #if !defined(SKIP_TRANSFORM_USED)
 	outvec = extra_matrix * outvec;
 	outvec = modelview_matrix * outvec;
@@ -253,7 +286,7 @@ VERTEX_SHADER_CODE
 	pos = outvec.xy;
 #endif
 
-#ifdef USE_LIGHT_ANGLE
+#ifdef USE_ATTRIB_LIGHT_ANGLE
 	// we add a fixed offset because we are using the sign later,
 	// and don't want floating point error around 0.0
 	float la = abs(light_angle) - 1.0;
@@ -294,6 +327,10 @@ uniform mediump sampler2D normal_texture; // texunit:1
 in highp vec2 uv_interp;
 in mediump vec4 color_interp;
 
+#ifdef USE_ATTRIB_MODULATE
+in mediump vec4 modulate_interp;
+#endif
+
 #if defined(SCREEN_TEXTURE_USED)
 
 uniform sampler2D screen_texture; // texunit:-3
@@ -410,16 +447,14 @@ uniform bool np_draw_center;
 // left top right bottom in pixel coordinates
 uniform vec4 np_margins;
 
-float map_ninepatch_axis(float pixel, float draw_size, float tex_pixel_size, float margin_begin, float margin_end, float s_ratio, int np_repeat, inout int draw_center) {
+float map_ninepatch_axis(float pixel, float draw_size, float tex_pixel_size, float margin_begin, float margin_end, int np_repeat, inout int draw_center) {
 
 	float tex_size = 1.0 / tex_pixel_size;
 
-	float screen_margin_begin = margin_begin / s_ratio;
+	if (pixel < margin_begin) {
-	float screen_margin_end = margin_end / s_ratio;
+		return pixel * tex_pixel_size;
-	if (pixel < screen_margin_begin) {
+	} else if (pixel >= draw_size - margin_end) {
-		return pixel * s_ratio * tex_pixel_size;
+		return (tex_size - (draw_size - pixel)) * tex_pixel_size;
-	} else if (pixel >= draw_size - screen_margin_end) {
-		return (tex_size - (draw_size - pixel) * s_ratio) * tex_pixel_size;
 	} else {
 		if (!np_draw_center) {
 			draw_center--;
@@ -428,21 +463,21 @@ float map_ninepatch_axis(float pixel, float draw_size, float tex_pixel_size, flo
 		// np_repeat is passed as uniform using NinePatchRect::AxisStretchMode enum.
 		if (np_repeat == 0) { // Stretch.
 			// Convert to ratio.
-			float ratio = (pixel - screen_margin_begin) / (draw_size - screen_margin_begin - screen_margin_end);
+			float ratio = (pixel - margin_begin) / (draw_size - margin_begin - margin_end);
 			// Scale to source texture.
 			return (margin_begin + ratio * (tex_size - margin_begin - margin_end)) * tex_pixel_size;
 		} else if (np_repeat == 1) { // Tile.
 			// Convert to offset.
-			float ofs = mod((pixel - screen_margin_begin), tex_size - margin_begin - margin_end);
+			float ofs = mod((pixel - margin_begin), tex_size - margin_begin - margin_end);
 			// Scale to source texture.
 			return (margin_begin + ofs) * tex_pixel_size;
 		} else if (np_repeat == 2) { // Tile Fit.
 			// Calculate scale.
-			float src_area = draw_size - screen_margin_begin - screen_margin_end;
+			float src_area = draw_size - margin_begin - margin_end;
 			float dst_area = tex_size - margin_begin - margin_end;
 			float scale = max(1.0, floor(src_area / max(dst_area, 0.0000001) + 0.5));
 			// Convert to ratio.
-			float ratio = (pixel - screen_margin_begin) / src_area;
+			float ratio = (pixel - margin_begin) / src_area;
 			ratio = mod(ratio * scale, 1.0);
 			// Scale to source texture.
 			return (margin_begin + ratio * dst_area) * tex_pixel_size;
@@ -467,11 +502,9 @@ void main() {
 #ifdef USE_NINEPATCH
 
 	int draw_center = 2;
-	float s_ratio = max((1.0 / color_texpixel_size.x) / abs(dst_rect.z), (1.0 / color_texpixel_size.y) / abs(dst_rect.w));
-	s_ratio = max(1.0, s_ratio);
 	uv = vec2(
-			map_ninepatch_axis(pixel_size_interp.x, abs(dst_rect.z), color_texpixel_size.x, np_margins.x, np_margins.z, s_ratio, np_repeat_h, draw_center),
+			map_ninepatch_axis(pixel_size_interp.x, abs(dst_rect.z), color_texpixel_size.x, np_margins.x, np_margins.z, np_repeat_h, draw_center),
-			map_ninepatch_axis(pixel_size_interp.y, abs(dst_rect.w), color_texpixel_size.y, np_margins.y, np_margins.w, s_ratio, np_repeat_v, draw_center));
+			map_ninepatch_axis(pixel_size_interp.y, abs(dst_rect.w), color_texpixel_size.y, np_margins.y, np_margins.w, np_repeat_v, draw_center));
 
 	if (draw_center == 0) {
 		color.a = 0.0;
@@ -549,6 +582,11 @@ FRAGMENT_SHADER_CODE
 	color *= final_modulate;
 #endif
 
+#ifdef USE_ATTRIB_MODULATE
+	// todo .. this won't be used at the same time as MODULATE_USED
+	color *= modulate_interp;
+#endif
+
 #ifdef USE_LIGHTING
 
 	vec2 light_vec = transformed_light_uv;

drivers/gles_common/SCsub

@@ -0,0 +1,5 @@
+#!/usr/bin/env python
+
+Import("env")
+
+env.add_source_files(env.drivers_sources, "*.cpp")

drivers/gles_common/batch_diagnose.inc

@@ -0,0 +1,148 @@
+String get_command_type_string(const RasterizerCanvas::Item::Command &p_command) const {
+	String sz = "";
+
+	switch (p_command.type) {
+		default:
+			break;
+		case RasterizerCanvas::Item::Command::TYPE_LINE: {
+			sz = "l";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_POLYLINE: {
+			sz = "PL";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_RECT: {
+			sz = "r";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_NINEPATCH: {
+			sz = "n";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_PRIMITIVE: {
+			sz = "PR";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_POLYGON: {
+			sz = "p";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_MESH: {
+			sz = "m";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_MULTIMESH: {
+			sz = "MM";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_PARTICLES: {
+			sz = "PA";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_CIRCLE: {
+			sz = "c";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_TRANSFORM: {
+			sz = "t";
+
+			// add a bit more info in debug build
+			const RasterizerCanvas::Item::CommandTransform *transform = static_cast<const RasterizerCanvas::Item::CommandTransform *>(&p_command);
+			const Transform2D &mat = transform->xform;
+
+			sz += " ";
+			sz += String(Variant(mat.elements[2]));
+			sz += " ";
+		} break;
+		case RasterizerCanvas::Item::Command::TYPE_CLIP_IGNORE: {
+			sz = "CI";
+		} break;
+	} // switch
+
+	return sz;
+}
+
+void diagnose_batches(RasterizerCanvas::Item::Command *const *p_commands) {
+	int num_batches = bdata.batches.size();
+
+	BatchColor curr_color;
+	curr_color.set(Color(-1, -1, -1, -1));
+	bool first_color_change = true;
+
+	for (int batch_num = 0; batch_num < num_batches; batch_num++) {
+		const Batch &batch = bdata.batches[batch_num];
+		bdata.frame_string += "\t\t\tbatch ";
+
+		switch (batch.type) {
+
+			case RasterizerStorageCommon::BT_POLY: {
+				bdata.frame_string += "P ";
+				bdata.frame_string += itos(batch.first_command) + "-";
+				bdata.frame_string += itos(batch.num_commands);
+
+				bdata.frame_string += " " + batch.color.to_string();
+
+				if (batch.num_commands > 1) {
+					bdata.frame_string += " MULTI";
+				}
+				if (curr_color != batch.color) {
+					curr_color = batch.color;
+					if (!first_color_change) {
+						bdata.frame_string += " color";
+					} else {
+						first_color_change = false;
+					}
+				}
+				bdata.frame_string += "\n";
+			} break;
+			case RasterizerStorageCommon::BT_LINE:
+			case RasterizerStorageCommon::BT_LINE_AA: {
+				bdata.frame_string += "L ";
+				bdata.frame_string += itos(batch.first_command) + "-";
+				bdata.frame_string += itos(batch.num_commands);
+
+				bdata.frame_string += " " + batch.color.to_string();
+
+				if (batch.num_commands > 1) {
+					bdata.frame_string += " MULTI";
+				}
+				if (curr_color != batch.color) {
+					curr_color = batch.color;
+					if (!first_color_change) {
+						bdata.frame_string += " color";
+					} else {
+						first_color_change = false;
+					}
+				}
+				bdata.frame_string += "\n";
+			} break;
+			case RasterizerStorageCommon::BT_RECT: {
+				bdata.frame_string += "R ";
+				bdata.frame_string += itos(batch.first_command) + "-";
+				bdata.frame_string += itos(batch.num_commands);
+
+				int tex_id = (int)bdata.batch_textures[batch.batch_texture_id].RID_texture.get_id();
+				bdata.frame_string += " [" + itos(batch.batch_texture_id) + " - " + itos(tex_id) + "]";
+
+				bdata.frame_string += " " + batch.color.to_string();
+
+				if (batch.num_commands > 1) {
+					bdata.frame_string += " MULTI";
+				}
+				if (curr_color != batch.color) {
+					curr_color = batch.color;
+					if (!first_color_change) {
+						bdata.frame_string += " color";
+					} else {
+						first_color_change = false;
+					}
+				}
+				bdata.frame_string += "\n";
+			} break;
+			default: {
+				bdata.frame_string += "D ";
+				bdata.frame_string += itos(batch.first_command) + "-";
+				bdata.frame_string += itos(batch.num_commands) + " ";
+
+				int num_show = MIN(batch.num_commands, 16);
+				for (int n = 0; n < num_show; n++) {
+					const RasterizerCanvas::Item::Command &comm = *p_commands[batch.first_command + n];
+					bdata.frame_string += get_command_type_string(comm) + " ";
+				}
+
+				bdata.frame_string += "\n";
+			} break;
+		}
+	}
+}

drivers/gles_common/rasterizer_array.h

@@ -1,5 +1,5 @@
 /*************************************************************************/
-/*  rasterizer_array_gles2.h                                             */
+/*  rasterizer_array.h                                                   */
 /*************************************************************************/
 /*                       This file is part of:                           */
 /*                           GODOT ENGINE                                */
@@ -30,36 +30,6 @@
 
 #pragma once
 
-/*************************************************************************/
-/*  rasterizer_array_gles2.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.                */
-/*************************************************************************/
-
 /**
  * Fast single-threaded growable array for POD types.
  * For use in render drivers, not for general use.
@@ -141,14 +111,14 @@ private:
 };
 
 template <class T>
-class RasterizerArrayGLES2 {
+class RasterizerArray {
 public:
-	RasterizerArrayGLES2() {
+	RasterizerArray() {
 		_list = 0;
 		_size = 0;
 		_max_size = 0;
 	}
-	~RasterizerArrayGLES2() { free(); }
+	~RasterizerArray() { free(); }
 
 	T &operator[](unsigned int ui) { return _list[ui]; }
 	const T &operator[](unsigned int ui) const { return _list[ui]; }
@@ -208,7 +178,7 @@ public:
 	int max_size() const { return _max_size; }
 	const T *get_data() const { return _list; }
 
-	bool copy_from(const RasterizerArrayGLES2<T> &o) {
+	bool copy_from(const RasterizerArray<T> &o) {
 		// no resizing done here, it should be done manually
 		if (o.size() > _max_size)
 			return false;
@@ -247,9 +217,9 @@ private:
 };
 
 template <class T>
-class RasterizerArray_non_pod_GLES2 {
+class RasterizerArray_non_pod {
 public:
-	RasterizerArray_non_pod_GLES2() {
+	RasterizerArray_non_pod() {
 		_size = 0;
 	}
 
@@ -287,3 +257,72 @@ private:
 	Vector<T> _list;
 	int _size;
 };
+
+// very simple non-growable array, that keeps track of the size of a 'unit'
+// which can be cast to whatever vertex format FVF required, and is initially
+// created with enough memory to hold the biggest FVF.
+// This allows multiple FVFs to use the same array.
+class RasterizerUnitArray {
+public:
+	RasterizerUnitArray() {
+		_list = nullptr;
+		free();
+	}
+	~RasterizerUnitArray() { free(); }
+
+	uint8_t *get_unit(unsigned int ui) { return &_list[ui * _unit_size_bytes]; }
+	const uint8_t *get_unit(unsigned int ui) const { return &_list[ui * _unit_size_bytes]; }
+
+	int size() const { return _size; }
+	int max_size() const { return _max_size; }
+
+	void free() {
+		if (_list) {
+			memdelete_arr(_list);
+			_list = 0;
+		}
+		_size = 0;
+		_max_size = 0;
+		_max_size_bytes = 0;
+		_unit_size_bytes = 0;
+	}
+
+	void create(int p_max_size_units, int p_max_unit_size_bytes) {
+		free();
+
+		_max_unit_size_bytes = p_max_unit_size_bytes;
+		_max_size = p_max_size_units;
+		_max_size_bytes = p_max_size_units * p_max_unit_size_bytes;
+
+		if (_max_size_bytes) {
+			_list = memnew_arr(uint8_t, _max_size_bytes);
+		}
+	}
+
+	void prepare(int p_unit_size_bytes) {
+		_unit_size_bytes = p_unit_size_bytes;
+		_size = 0;
+	}
+
+	// several items at a time
+	uint8_t *request(int p_num_items = 1) {
+		int old_size = _size;
+		_size += p_num_items;
+
+		if (_size <= _max_size) {
+			return get_unit(old_size);
+		}
+
+		// revert
+		_size = old_size;
+		return nullptr;
+	}
+
+private:
+	uint8_t *_list;
+	int _size; // in units
+	int _max_size; // in units
+	int _max_size_bytes;
+	int _unit_size_bytes;
+	int _max_unit_size_bytes;
+};

drivers/gles_common/rasterizer_storage_common.h

@@ -0,0 +1,71 @@
+/*************************************************************************/
+/*  rasterizer_storage_common.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.                */
+/*************************************************************************/
+
+#pragma once
+
+class RasterizerStorageCommon {
+public:
+	enum FVF {
+		FVF_UNBATCHED,
+		FVF_REGULAR,
+		FVF_COLOR,
+		FVF_LIGHT_ANGLE,
+		FVF_MODULATED,
+		FVF_LARGE,
+	};
+
+	// these flags are specifically for batching
+	// some of the logic is thus in rasterizer_storage.cpp
+	// we could alternatively set bitflags for each 'uses' and test on the fly
+	enum BatchFlags {
+		PREVENT_COLOR_BAKING = 1 << 0,
+		PREVENT_VERTEX_BAKING = 1 << 1,
+
+		USE_MODULATE_FVF = 1 << 2,
+		USE_LARGE_FVF = 1 << 3,
+	};
+
+	enum BatchType : uint16_t {
+		BT_DEFAULT = 0,
+		BT_RECT = 1,
+		BT_LINE = 2,
+		BT_LINE_AA = 3,
+		BT_POLY = 4,
+		BT_DUMMY = 5, // dummy batch is just used to keep the batch creation loop simple
+	};
+
+	enum BatchTypeFlags {
+		BTF_DEFAULT = 1 << BT_DEFAULT,
+		BTF_RECT = 1 << BT_RECT,
+		BTF_LINE = 1 << BT_LINE,
+		BTF_LINE_AA = 1 << BT_LINE_AA,
+		BTF_POLY = 1 << BT_POLY,
+	};
+};

editor/plugins/canvas_item_editor_plugin.cpp

@@ -53,6 +53,8 @@
 #include "scene/main/viewport.h"
 #include "scene/resources/packed_scene.h"
 
+#include <stdlib.h>
+
 // Min and Max are power of two in order to play nicely with successive increment.
 // That way, we can naturally reach a 100% zoom from boundaries.
 #define MIN_ZOOM 1. / 128

platform/x11/detect.py

@@ -116,6 +116,7 @@ def configure(env):
             env.Prepend(CCFLAGS=["-g2"])
 
     elif env["target"] == "debug":
+        env.Prepend(CCFLAGS=["-ggdb"])
         env.Prepend(CCFLAGS=["-g3"])
         env.Prepend(CPPDEFINES=["DEBUG_ENABLED"])
         env.Append(LINKFLAGS=["-rdynamic"])

servers/visual/rasterizer.h

@@ -1180,6 +1180,8 @@ public:
 
 	virtual bool is_low_end() const = 0;
 
+	virtual const char *gl_check_for_error(bool p_print_error = true) = 0;
+
 	virtual ~Rasterizer() {}
 };
 

servers/visual_server.cpp

@@ -2439,6 +2439,10 @@ VisualServer::VisualServer() {
 	GLOBAL_DEF(sz_balance_render_tree, 0.17f);
 	ProjectSettings::get_singleton()->set_custom_property_info(sz_balance_render_tree, PropertyInfo(Variant::REAL, sz_balance_render_tree, PROPERTY_HINT_RANGE, "0.0,1.0,0.01"));
 
+	GLOBAL_DEF("rendering/quality/2d/use_software_skinning", true);
+	GLOBAL_DEF("rendering/quality/2d/ninepatch_mode", 0);
+	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/2d/ninepatch_mode", PropertyInfo(Variant::INT, "rendering/quality/2d/ninepatch_mode", PROPERTY_HINT_ENUM, "Default,Scaling"));
+
 	GLOBAL_DEF("rendering/batching/options/use_batching", true);
 	GLOBAL_DEF_RST("rendering/batching/options/use_batching_in_editor", true);
 	GLOBAL_DEF("rendering/batching/options/single_rect_fallback", false);