Added Rasterizer API for display color management.

Use VisualServer::set_screen_lut to set a 3D texture for color management.
Its X, Y, Z axes correspond to R, G, B channels.

LUT transformation happens in Rasterizer, ensuring that all content blitted
to screen are processed exactly once.

This implements part of godotengine#26826.

doc/classes/VisualServer.xml

@@ -3165,6 +3165,19 @@
 				Sets a boot image. The color defines the background color. If [code]scale[/code] is [code]true[/code], the image will be scaled to fit the screen size. If [code]use_filter[/code] is [code]true[/code], the image will be scaled with linear interpolation. If [code]use_filter[/code] is [code]false[/code], the image will be scaled with nearest-neighbor interpolation.
 			</description>
 		</method>
+		<method name="set_screen_lut">
+			<return type="void">
+			</return>
+			<argument index="0" name="lut" type="Image">
+			</argument>
+			<argument index="1" name="h_slices" type="int">
+			</argument>
+			<argument index="2" name="v_slices" type="int">
+			</argument>
+			<description>
+				Sets a 3D output LUT. It will be used to transform all screen output. If lut is [code]null[/code] the current LUT will be removed.
+			</description>
+		</method>
 		<method name="set_debug_generate_wireframes">
 			<return type="void">
 			</return>

drivers/dummy/rasterizer_dummy.h

@@ -783,6 +783,7 @@ class RasterizerDummy : public Rasterizer {
 	RasterizerScene *get_scene() { return &scene; }
 
 	void set_boot_image(const Ref<Image> &p_image, const Color &p_color, bool p_scale, bool p_use_filter = true) {}
+	void set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices) {}
 
 	void initialize() {}
 	void begin_frame(double frame_step) {}

drivers/gles2/rasterizer_gles2.cpp

@@ -268,6 +268,7 @@ void RasterizerGLES2::initialize() {
 	storage->initialize();
 	canvas->initialize();
 	scene->initialize();
+	state.lut_shader.init();
 }
 
 void RasterizerGLES2::begin_frame(double frame_step) {
@@ -397,34 +398,71 @@ void RasterizerGLES2::set_boot_image(const Ref<Image> &p_image, const Color &p_c
 	end_frame(true);
 }
 
+void RasterizerGLES2::set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices) {
+
+	if (state.screen_lut.is_valid()) {
+		storage->free(state.screen_lut);
+		state.screen_lut = RID();
+	}
+
+	if (p_lut.is_null() || p_lut->empty())
+		return;
+
+	RID texture = storage->texture_create();
+	storage->texture_allocate(texture, p_lut->get_width(), p_lut->get_height(), 0, p_lut->get_format(), VS::TEXTURE_TYPE_2D, VS::TEXTURE_FLAG_FILTER);
+	storage->texture_set_data(texture, p_lut);
+
+	state.screen_lut = texture;
+	state.lut_texel_count = Vector2(p_lut->get_width(), p_lut->get_height());
+	state.lut_chunk_count = Vector2(p_h_slices, p_v_slices);
+}
+
 void RasterizerGLES2::blit_render_target_to_screen(RID p_render_target, const Rect2 &p_screen_rect, int p_screen) {
 
 	ERR_FAIL_COND(storage->frame.current_rt);
 
 	RasterizerStorageGLES2::RenderTarget *rt = storage->render_target_owner.getornull(p_render_target);
 	ERR_FAIL_COND(!rt);
 
-	canvas->state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_TEXTURE_RECT, true);
+	if (!state.screen_lut.is_valid()) {
+		// draw normally
+		canvas->state.canvas_shader.set_conditional(CanvasShaderGLES2::USE_TEXTURE_RECT, true);
 
-	canvas->state.canvas_shader.set_custom_shader(0);
-	canvas->state.canvas_shader.bind();
+		canvas->state.canvas_shader.set_custom_shader(0);
+		canvas->state.canvas_shader.bind();
 
-	canvas->canvas_begin();
-	glDisable(GL_BLEND);
-	glBindFramebuffer(GL_FRAMEBUFFER, RasterizerStorageGLES2::system_fbo);
-	glActiveTexture(GL_TEXTURE0 + storage->config.max_texture_image_units - 1);
-	if (rt->external.fbo != 0) {
-		glBindTexture(GL_TEXTURE_2D, rt->external.color);
+		canvas->canvas_begin();
+		glDisable(GL_BLEND);
+		glBindFramebuffer(GL_FRAMEBUFFER, RasterizerStorageGLES2::system_fbo);
+		glActiveTexture(GL_TEXTURE0 + storage->config.max_texture_image_units - 1);
+		if (rt->external.fbo != 0) {
+			glBindTexture(GL_TEXTURE_2D, rt->external.color);
+		} else {
+			glBindTexture(GL_TEXTURE_2D, rt->color);
+		}
+		canvas->canvas_end();
 	} else {
-		glBindTexture(GL_TEXTURE_2D, rt->color);
+		// draw with lut
+		Size2 win_size = OS::get_singleton()->get_window_size();
+		glDisable(GL_BLEND);
+		glBindFramebuffer(GL_FRAMEBUFFER, RasterizerStorageGLES2::system_fbo);
+		glActiveTexture(GL_TEXTURE0);
+		if (rt->external.fbo != 0) {
+			glBindTexture(GL_TEXTURE_2D, rt->external.color);
+		} else {
+			glBindTexture(GL_TEXTURE_2D, rt->color);
+		}
+		glActiveTexture(GL_TEXTURE1);
+		glBindTexture(GL_TEXTURE_2D, storage->texture_owner.get(state.screen_lut)->tex_id);
+		state.lut_shader.bind();
+		state.lut_shader.set_uniform(LutTransformShaderGLES2::SCREEN_RECT, Color(p_screen_rect.position.x / win_size.width, 1.0 - (p_screen_rect.position.y + p_screen_rect.size.height) / win_size.height, p_screen_rect.size.width / win_size.width, p_screen_rect.size.height / win_size.height));
+		state.lut_shader.set_uniform(LutTransformShaderGLES2::TEXEL_COUNT, state.lut_texel_count);
+		state.lut_shader.set_uniform(LutTransformShaderGLES2::CHUNK_COUNT, state.lut_chunk_count);
+		storage->bind_quad_array();
+		glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
+		glBindBuffer(GL_ARRAY_BUFFER, 0);
+		glBindTexture(GL_TEXTURE_2D, 0);
 	}
-
-	// TODO normals
-
-	canvas->draw_generic_textured_rect(p_screen_rect, Rect2(0, 0, 1, -1));
-
-	glBindTexture(GL_TEXTURE_2D, 0);
-	canvas->canvas_end();
 }
 
 void RasterizerGLES2::output_lens_distorted_to_screen(RID p_render_target, const Rect2 &p_screen_rect, float p_k1, float p_k2, const Vector2 &p_eye_center, float p_oversample) {
@@ -477,6 +515,10 @@ void RasterizerGLES2::end_frame(bool p_swap_buffers) {
 }
 
 void RasterizerGLES2::finalize() {
+
+	if (state.screen_lut.is_valid()) {
+		storage->free(state.screen_lut);
+	}
 }
 
 Rasterizer *RasterizerGLES2::_create_current() {

drivers/gles2/rasterizer_gles2.h

@@ -36,6 +36,8 @@
 #include "rasterizer_storage_gles2.h"
 #include "servers/visual/rasterizer.h"
 
+#include "shaders/lut_transform.glsl.gen.h"
+
 class RasterizerGLES2 : public Rasterizer {
 
 	static Rasterizer *_create_current();
@@ -46,12 +48,20 @@ class RasterizerGLES2 : public Rasterizer {
 
 	double time_total;
 
+	struct State {
+		RID screen_lut;
+		Vector2 lut_texel_count;
+		Vector2 lut_chunk_count;
+		LutTransformShaderGLES2 lut_shader;
+	} state;
+
 public:
 	virtual RasterizerStorage *get_storage();
 	virtual RasterizerCanvas *get_canvas();
 	virtual RasterizerScene *get_scene();
 
 	virtual void set_boot_image(const Ref<Image> &p_image, const Color &p_color, bool p_scale, bool p_use_filter = true);
+	virtual void set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices);
 
 	virtual void initialize();
 	virtual void begin_frame(double frame_step);

drivers/gles2/shaders/SCsub

@@ -21,3 +21,4 @@ if 'GLES2_GLSL' in env['BUILDERS']:
 #    env.GLES2_GLSL('tonemap.glsl');
 #    env.GLES2_GLSL('particles.glsl');
     env.GLES2_GLSL('lens_distorted.glsl');
+    env.GLES2_GLSL('lut_transform.glsl');

drivers/gles2/shaders/lut_transform.glsl

@@ -0,0 +1,48 @@
+/* clang-format off */
+[vertex]
+
+attribute highp vec4 vertex_attrib; // attrib:0
+/* clang-format on */
+
+attribute vec2 uv_in; // attrib:4
+
+uniform highp vec4 screen_rect;
+
+varying vec2 uv_interp;
+
+void main() {
+	uv_interp = uv_in;
+	gl_Position = vec4(screen_rect.xy + vertex_attrib.xy * screen_rect.zw, vertex_attrib.zw);
+}
+
+/* clang-format off */
+[fragment]
+
+uniform vec2 texel_count;
+uniform vec2 chunk_count;
+
+uniform sampler2D source; //texunit:0
+uniform sampler2D lut; //texunit:1
+/* clang-format on */
+
+varying vec2 uv_interp;
+
+vec2 flatten_3d(vec2 chunk_size, vec2 texel_size, vec2 xy, float z_scaled) {
+	float row = floor(z_scaled / chunk_count.x);
+	float col = mod(z_scaled, chunk_count.x);
+	return chunk_size * vec2(col, row) + (chunk_size - texel_size) * xy + texel_size * 0.5;
+}
+
+void main() {
+	vec4 color = texture2D(source, uv_interp);
+	vec3 clamped_rgb = clamp(color.rgb, 0.0, 1.0);
+
+	vec2 texel_size = 1.0 / texel_count;
+	vec2 chunk_size = 1.0 / chunk_count;
+
+	float z_scaled = clamped_rgb.b * (chunk_count.x * chunk_count.y - 1.0);
+	vec4 low = texture2D(lut, flatten_3d(chunk_size, texel_size, clamped_rgb.rg, floor(z_scaled)));
+	vec4 high = texture2D(lut, flatten_3d(chunk_size, texel_size, clamped_rgb.rg, ceil(z_scaled)));
+
+	gl_FragColor = vec4(mix(low.rgb, high.rgb, mod(z_scaled, 1.0)), color.a);
+}

drivers/gles3/rasterizer_gles3.cpp

@@ -191,6 +191,7 @@ void RasterizerGLES3::initialize() {
 	storage->initialize();
 	canvas->initialize();
 	scene->initialize();
+	state.lut_shader.init();
 }
 
 void RasterizerGLES3::begin_frame(double frame_step) {
@@ -333,6 +334,34 @@ void RasterizerGLES3::set_boot_image(const Ref<Image> &p_image, const Color &p_c
 	end_frame(true);
 }
 
+void RasterizerGLES3::set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices) {
+
+	if (state.screen_lut.is_valid()) {
+		storage->free(state.screen_lut);
+		state.screen_lut = RID();
+	}
+
+	if (p_lut.is_null() || p_lut->empty())
+		return;
+
+	int slice_w = p_lut->get_width() / p_h_slices;
+	int slice_h = p_lut->get_height() / p_v_slices;
+
+	RID texture = storage->texture_create();
+	storage->texture_allocate(texture, slice_w, slice_h, p_h_slices * p_v_slices, p_lut->get_format(), VS::TEXTURE_TYPE_3D, VS::TEXTURE_FLAG_FILTER);
+
+	for (int i = 0; i < p_v_slices; i++) {
+		for (int j = 0; j < p_h_slices; j++) {
+			int x = slice_w * j;
+			int y = slice_h * i;
+			storage->texture_set_data(texture, p_lut->get_rect(Rect2(x, y, slice_w, slice_h)), i * p_h_slices + j);
+		}
+	}
+
+	state.screen_lut = texture;
+	state.lut_texel_count = Vector3(slice_w, slice_h, p_h_slices * p_v_slices);
+}
+
 void RasterizerGLES3::blit_render_target_to_screen(RID p_render_target, const Rect2 &p_screen_rect, int p_screen) {
 
 	ERR_FAIL_COND(storage->frame.current_rt);
@@ -341,17 +370,39 @@ void RasterizerGLES3::blit_render_target_to_screen(RID p_render_target, const Re
 	ERR_FAIL_COND(!rt);
 
 #if 1
-
 	Size2 win_size = OS::get_singleton()->get_window_size();
-	if (rt->external.fbo != 0) {
-		glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->external.fbo);
+
+	if (!state.screen_lut.is_valid()) {
+		// no lut, just blit buffer
+		if (rt->external.fbo != 0) {
+			glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->external.fbo);
+		} else {
+			glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->fbo);
+		}
+		glReadBuffer(GL_COLOR_ATTACHMENT0);
+		glBindFramebuffer(GL_DRAW_FRAMEBUFFER, RasterizerStorageGLES3::system_fbo);
+		glBlitFramebuffer(0, 0, rt->width, rt->height, p_screen_rect.position.x, win_size.height - p_screen_rect.position.y - p_screen_rect.size.height, p_screen_rect.position.x + p_screen_rect.size.width, win_size.height - p_screen_rect.position.y, GL_COLOR_BUFFER_BIT, GL_NEAREST);
 	} else {
-		glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->fbo);
+		// draw with lut
+		glDisable(GL_BLEND);
+		if (rt->external.fbo != 0) {
+			glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->external.fbo);
+		} else {
+			glBindFramebuffer(GL_READ_FRAMEBUFFER, rt->fbo);
+		}
+		glBindFramebuffer(GL_DRAW_FRAMEBUFFER, RasterizerStorageGLES3::system_fbo);
+		glActiveTexture(GL_TEXTURE0);
+		glBindTexture(GL_TEXTURE_2D, rt->color);
+		glActiveTexture(GL_TEXTURE1);
+		glBindTexture(GL_TEXTURE_3D, storage->texture_owner.get(state.screen_lut)->tex_id);
+		state.lut_shader.bind();
+		state.lut_shader.set_uniform(LutTransformShaderGLES3::SCREEN_RECT, Color(p_screen_rect.position.x / win_size.width, 1.0 - (p_screen_rect.position.y + p_screen_rect.size.height) / win_size.height, p_screen_rect.size.width / win_size.width, p_screen_rect.size.height / win_size.height));
+		state.lut_shader.set_uniform(LutTransformShaderGLES3::TEXEL_COUNT, state.lut_texel_count);
+		glBindVertexArray(storage->resources.quadie_array);
+		glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
+		glBindVertexArray(0);
+		glBindTexture(GL_TEXTURE_2D, 0);
 	}
-	glReadBuffer(GL_COLOR_ATTACHMENT0);
-	glBindFramebuffer(GL_DRAW_FRAMEBUFFER, RasterizerStorageGLES3::system_fbo);
-	glBlitFramebuffer(0, 0, rt->width, rt->height, p_screen_rect.position.x, win_size.height - p_screen_rect.position.y - p_screen_rect.size.height, p_screen_rect.position.x + p_screen_rect.size.width, win_size.height - p_screen_rect.position.y, GL_COLOR_BUFFER_BIT, GL_NEAREST);
-
 #else
 	canvas->canvas_begin();
 	glDisable(GL_BLEND);
@@ -418,6 +469,9 @@ void RasterizerGLES3::end_frame(bool p_swap_buffers) {
 }
 
 void RasterizerGLES3::finalize() {
+	if (state.screen_lut.is_valid()) {
+		storage->free(state.screen_lut);
+	}
 
 	storage->finalize();
 	canvas->finalize();

drivers/gles3/rasterizer_gles3.h

@@ -36,6 +36,8 @@
 #include "rasterizer_storage_gles3.h"
 #include "servers/visual/rasterizer.h"
 
+#include "shaders/lut_transform.glsl.gen.h"
+
 class RasterizerGLES3 : public Rasterizer {
 
 	static Rasterizer *_create_current();
@@ -46,12 +48,19 @@ class RasterizerGLES3 : public Rasterizer {
 
 	double time_total;
 
+	struct State {
+		RID screen_lut;
+		Vector3 lut_texel_count;
+		LutTransformShaderGLES3 lut_shader;
+	} state;
+
 public:
 	virtual RasterizerStorage *get_storage();
 	virtual RasterizerCanvas *get_canvas();
 	virtual RasterizerScene *get_scene();
 
 	virtual void set_boot_image(const Ref<Image> &p_image, const Color &p_color, bool p_scale, bool p_use_filter = true);
+	virtual void set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices);
 
 	virtual void initialize();
 	virtual void begin_frame(double frame_step);

drivers/gles3/shaders/SCsub

@@ -21,3 +21,4 @@ if 'GLES3_GLSL' in env['BUILDERS']:
     env.GLES3_GLSL('tonemap.glsl');
     env.GLES3_GLSL('particles.glsl');
     env.GLES3_GLSL('lens_distorted.glsl');
+    env.GLES3_GLSL('lut_transform.glsl');

drivers/gles3/shaders/lut_transform.glsl

@@ -0,0 +1,35 @@
+/* clang-format off */
+[vertex]
+
+layout(location = 0) in highp vec4 vertex_attrib;
+/* clang-format on */
+
+layout(location = 4) in vec2 uv_in;
+
+uniform highp vec4 screen_rect;
+
+out vec2 uv_interp;
+
+void main() {
+	uv_interp = uv_in;
+	gl_Position = vec4(screen_rect.xy + vertex_attrib.xy * screen_rect.zw, vertex_attrib.zw);
+}
+
+/* clang-format off */
+[fragment]
+
+uniform vec3 texel_count;
+
+uniform sampler2D source; //texunit:0
+uniform sampler3D lut; //texunit:1
+/* clang-format on */
+
+in vec2 uv_interp;
+
+layout(location = 0) out vec4 frag_color;
+
+void main() {
+	vec4 color = texture(source, uv_interp);
+	vec3 coords = clamp(color.rgb, 0.0, 1.0) * (1.0 - 1.0 / texel_count) + 0.5 / texel_count;
+	frag_color = texture(lut, coords);
+}

servers/visual/rasterizer.h

@@ -1106,6 +1106,7 @@ class Rasterizer {
 	virtual RasterizerScene *get_scene() = 0;
 
 	virtual void set_boot_image(const Ref<Image> &p_image, const Color &p_color, bool p_scale, bool p_use_filter = true) = 0;
+	virtual void set_screen_lut(const Ref<Image> &p_lut, int p_h_slices, int p_v_slices) = 0;
 
 	virtual void initialize() = 0;
 	virtual void begin_frame(double frame_step) = 0;