Adding Variable Rate Shading support to Godot

Improve GI renderer and add VRS support
Implement render device has_feature and move subgroup settings to limit_get

24 files changed, 816 insertions, 249 deletions

diff --git a/servers/rendering/dummy/storage/texture_storage.h b/servers/rendering/dummy/storage/texture_storage.h
index fe10f6489c..195d378a41 100644
--- a/servers/rendering/dummy/storage/texture_storage.h
+++ b/servers/rendering/dummy/storage/texture_storage.h
@@ -169,6 +169,9 @@ public:
 	virtual void render_target_set_sdf_size_and_scale(RID p_render_target, RS::ViewportSDFOversize p_size, RS::ViewportSDFScale p_scale) override {}
 	virtual Rect2i render_target_get_sdf_rect(RID p_render_target) const override { return Rect2i(); }
 	virtual void render_target_mark_sdf_enabled(RID p_render_target, bool p_enabled) override {}
+
+	virtual void render_target_set_vrs_mode(RID p_render_target, RS::ViewportVRSMode p_mode) override{};
+	virtual void render_target_set_vrs_texture(RID p_render_target, RID p_texture) override{};
 };
 
 } // namespace RendererDummy
diff --git a/servers/rendering/renderer_rd/effects/copy_effects.cpp b/servers/rendering/renderer_rd/effects/copy_effects.cpp
index c30e8ed58f..cbf7046887 100644
--- a/servers/rendering/renderer_rd/effects/copy_effects.cpp
+++ b/servers/rendering/renderer_rd/effects/copy_effects.cpp
@@ -100,11 +100,11 @@ CopyEffects::CopyEffects(bool p_prefer_raster_effects) {
 
 	{
 		Vector<String> copy_modes;
-		copy_modes.push_back("\n");
-		copy_modes.push_back("\n#define MODE_PANORAMA_TO_DP\n");
-		copy_modes.push_back("\n#define MODE_TWO_SOURCES\n");
-		copy_modes.push_back("\n#define MULTIVIEW\n");
-		copy_modes.push_back("\n#define MULTIVIEW\n#define MODE_TWO_SOURCES\n");
+		copy_modes.push_back("\n"); // COPY_TO_FB_COPY
+		copy_modes.push_back("\n#define MODE_PANORAMA_TO_DP\n"); // COPY_TO_FB_COPY_PANORAMA_TO_DP
+		copy_modes.push_back("\n#define MODE_TWO_SOURCES\n"); // COPY_TO_FB_COPY2
+		copy_modes.push_back("\n#define MULTIVIEW\n"); // COPY_TO_FB_MULTIVIEW
+		copy_modes.push_back("\n#define MULTIVIEW\n#define MODE_TWO_SOURCES\n"); // COPY_TO_FB_MULTIVIEW_WITH_DEPTH
 
 		copy_to_fb.shader.initialize(copy_modes);
 
diff --git a/servers/rendering/renderer_rd/effects/vrs.cpp b/servers/rendering/renderer_rd/effects/vrs.cpp
new file mode 100644
index 0000000000..505a35a269
--- /dev/null
+++ b/servers/rendering/renderer_rd/effects/vrs.cpp
@@ -0,0 +1,171 @@
+/*************************************************************************/
+/*  vrs.cpp                                                              */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2022 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2022 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.                */
+/*************************************************************************/
+
+#include "vrs.h"
+#include "../renderer_compositor_rd.h"
+#include "../storage_rd/texture_storage.h"
+#include "../uniform_set_cache_rd.h"
+#include "servers/xr_server.h"
+
+using namespace RendererRD;
+
+VRS::VRS() {
+	{
+		Vector<String> vrs_modes;
+		vrs_modes.push_back("\n"); // VRS_DEFAULT
+		vrs_modes.push_back("\n#define MULTIVIEW\n"); // VRS_MULTIVIEW
+
+		vrs_shader.shader.initialize(vrs_modes);
+
+		if (!RendererCompositorRD::singleton->is_xr_enabled()) {
+			vrs_shader.shader.set_variant_enabled(VRS_MULTIVIEW, false);
+		}
+
+		vrs_shader.shader_version = vrs_shader.shader.version_create();
+
+		//use additive
+
+		for (int i = 0; i < VRS_MAX; i++) {
+			if (vrs_shader.shader.is_variant_enabled(i)) {
+				vrs_shader.pipelines[i].setup(vrs_shader.shader.version_get_shader(vrs_shader.shader_version, i), RD::RENDER_PRIMITIVE_TRIANGLES, RD::PipelineRasterizationState(), RD::PipelineMultisampleState(), RD::PipelineDepthStencilState(), RD::PipelineColorBlendState::create_disabled(), 0);
+			} else {
+				vrs_shader.pipelines[i].clear();
+			}
+		}
+	}
+}
+
+VRS::~VRS() {
+	vrs_shader.shader.version_free(vrs_shader.shader_version);
+}
+
+void VRS::copy_vrs(RID p_source_rd_texture, RID p_dest_framebuffer, bool p_multiview) {
+	UniformSetCacheRD *uniform_set_cache = UniformSetCacheRD::get_singleton();
+	ERR_FAIL_NULL(uniform_set_cache);
+	MaterialStorage *material_storage = MaterialStorage::get_singleton();
+	ERR_FAIL_NULL(material_storage);
+
+	// setup our uniforms
+	RID default_sampler = material_storage->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_NEAREST, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED);
+
+	RD::Uniform u_source_rd_texture(RD::UNIFORM_TYPE_SAMPLER_WITH_TEXTURE, 0, Vector<RID>({ default_sampler, p_source_rd_texture }));
+
+	VRSMode mode = p_multiview ? VRS_MULTIVIEW : VRS_DEFAULT;
+
+	RID shader = vrs_shader.shader.version_get_shader(vrs_shader.shader_version, mode);
+	ERR_FAIL_COND(shader.is_null());
+
+	RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dest_framebuffer, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD, Vector<Color>());
+	RD::get_singleton()->draw_list_bind_render_pipeline(draw_list, vrs_shader.pipelines[mode].get_render_pipeline(RD::INVALID_ID, RD::get_singleton()->framebuffer_get_format(p_dest_framebuffer)));
+	RD::get_singleton()->draw_list_bind_uniform_set(draw_list, uniform_set_cache->get_cache(shader, 0, u_source_rd_texture), 0);
+	RD::get_singleton()->draw_list_bind_index_array(draw_list, material_storage->get_quad_index_array());
+	// RD::get_singleton()->draw_list_set_push_constant(draw_list, &vrs_shader.push_constant, sizeof(VRSPushConstant));
+	RD::get_singleton()->draw_list_draw(draw_list, true);
+	RD::get_singleton()->draw_list_end();
+}
+
+void VRS::create_vrs_texture(const int p_base_width, const int p_base_height, const uint32_t p_view_count, RID &p_vrs_texture, RID &p_vrs_fb) {
+	// TODO find a way to skip this if VRS is not supported, but we don't have access to VulkanContext here, even though we're in vulkan.. hmmm
+
+	// TODO we should find some way to store this properly, we're assuming 16x16 as this seems to be the standard but in our vrs_capacities we
+	// obtain a minimum and maximum size, and we should choose something within this range and then make sure that is consistantly set when creating
+	// our frame buffer. Also it is important that we make the resulting size we calculate down below available to the end user so they know the size
+	// of the VRS buffer to supply.
+	Size2i texel_size = Size2i(16, 16);
+
+	RD::TextureFormat tf;
+	if (p_view_count > 1) {
+		tf.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
+	} else {
+		tf.texture_type = RD::TEXTURE_TYPE_2D;
+	}
+	tf.format = RD::DATA_FORMAT_R8_UINT;
+	tf.width = p_base_width / texel_size.x;
+	if (p_base_width % texel_size.x != 0) {
+		tf.width++;
+	}
+	tf.height = p_base_height / texel_size.y;
+	if (p_base_height % texel_size.y != 0) {
+		tf.height++;
+	}
+	tf.array_layers = p_view_count; // create a layer for every view
+	tf.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_VRS_ATTACHMENT_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
+	tf.samples = RD::TEXTURE_SAMPLES_1;
+
+	p_vrs_texture = RD::get_singleton()->texture_create(tf, RD::TextureView());
+
+	// by default VRS is assumed to be our VRS attachment, but if we need to write into it, we need a bit more control
+	Vector<RID> fb;
+	fb.push_back(p_vrs_texture);
+
+	RD::FramebufferPass pass;
+	pass.color_attachments.push_back(0);
+
+	Vector<RD::FramebufferPass> passes;
+	passes.push_back(pass);
+
+	p_vrs_fb = RD::get_singleton()->framebuffer_create_multipass(fb, passes, RenderingDevice::INVALID_ID, p_view_count);
+}
+
+void VRS::update_vrs_texture(RID p_vrs_fb, RID p_render_target) {
+	TextureStorage *texture_storage = TextureStorage::get_singleton();
+	RS::ViewportVRSMode vrs_mode = texture_storage->render_target_get_vrs_mode(p_render_target);
+
+	if (vrs_mode != RS::VIEWPORT_VRS_DISABLED) {
+		RD::get_singleton()->draw_command_begin_label("VRS Setup");
+
+		// TODO figure out if image has changed since it was last copied so we can save some resources..
+
+		if (vrs_mode == RS::VIEWPORT_VRS_TEXTURE) {
+			RID vrs_texture = texture_storage->render_target_get_vrs_texture(p_render_target);
+			if (vrs_texture.is_valid()) {
+				Texture *texture = texture_storage->get_texture(vrs_texture);
+				if (texture) {
+					// Copy into our density buffer
+					copy_vrs(texture->rd_texture, p_vrs_fb, texture->layers > 1);
+				}
+			}
+		} else if (vrs_mode == RS::VIEWPORT_VRS_XR) {
+			Ref<XRInterface> interface = XRServer::get_singleton()->get_primary_interface();
+			if (interface.is_valid()) {
+				RID vrs_texture = interface->get_vrs_texture();
+				if (vrs_texture.is_valid()) {
+					Texture *texture = texture_storage->get_texture(vrs_texture);
+					if (texture) {
+						// Copy into our density buffer
+						copy_vrs(texture->rd_texture, p_vrs_fb, texture->layers > 1);
+					}
+				}
+			}
+		}
+
+		RD::get_singleton()->draw_command_end_label();
+	}
+}
diff --git a/servers/rendering/renderer_rd/effects/vrs.h b/servers/rendering/renderer_rd/effects/vrs.h
new file mode 100644
index 0000000000..0f2bdd31b6
--- /dev/null
+++ b/servers/rendering/renderer_rd/effects/vrs.h
@@ -0,0 +1,75 @@
+/*************************************************************************/
+/*  vrs.h                                                                */
+/*************************************************************************/
+/*                       This file is part of:                           */
+/*                           GODOT ENGINE                                */
+/*                      https://godotengine.org                          */
+/*************************************************************************/
+/* Copyright (c) 2007-2022 Juan Linietsky, Ariel Manzur.                 */
+/* Copyright (c) 2014-2022 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 VRS_RD_H
+#define VRS_RD_H
+
+#include "servers/rendering/renderer_rd/pipeline_cache_rd.h"
+#include "servers/rendering/renderer_rd/shaders/effects/vrs.glsl.gen.h"
+#include "servers/rendering/renderer_scene_render.h"
+
+#include "servers/rendering_server.h"
+
+namespace RendererRD {
+
+class VRS {
+private:
+	enum VRSMode {
+		VRS_DEFAULT,
+		VRS_MULTIVIEW,
+		VRS_MAX,
+	};
+
+	/* we have no push constant here (yet)
+	struct VRSPushConstant {
+
+	};
+	*/
+
+	struct VRSShader {
+		// VRSPushConstant push_constant;
+		VrsShaderRD shader;
+		RID shader_version;
+		PipelineCacheRD pipelines[VRS_MAX];
+	} vrs_shader;
+
+public:
+	VRS();
+	~VRS();
+
+	void copy_vrs(RID p_source_rd_texture, RID p_dest_framebuffer, bool p_multiview = false);
+
+	void create_vrs_texture(const int p_base_width, const int p_base_height, const uint32_t p_view_count, RID &p_vrs_texture, RID &p_vrs_fb);
+	void update_vrs_texture(RID p_vrs_fb, RID p_render_target);
+};
+
+} // namespace RendererRD
+
+#endif // !VRS_RD_H
diff --git a/servers/rendering/renderer_rd/effects_rd.cpp b/servers/rendering/renderer_rd/effects_rd.cpp
index d45ddbc392..f731a0007a 100644
--- a/servers/rendering/renderer_rd/effects_rd.cpp
+++ b/servers/rendering/renderer_rd/effects_rd.cpp
@@ -1309,7 +1309,7 @@ EffectsRD::EffectsRD(bool p_prefer_raster_effects) {
 		FSR_upscale_modes.push_back("\n#define MODE_FSR_UPSCALE_FALLBACK\n");
 #else
 		// Everyone else can use normal mode when available.
-		if (RD::get_singleton()->get_device_capabilities()->supports_fsr_half_float) {
+		if (RD::get_singleton()->has_feature(RD::SUPPORTS_FSR_HALF_FLOAT)) {
 			FSR_upscale_modes.push_back("\n#define MODE_FSR_UPSCALE_NORMAL\n");
 		} else {
 			FSR_upscale_modes.push_back("\n#define MODE_FSR_UPSCALE_FALLBACK\n");
diff --git a/servers/rendering/renderer_rd/environment/gi.cpp b/servers/rendering/renderer_rd/environment/gi.cpp
index a749e7d5bc..f17d5172b7 100644
--- a/servers/rendering/renderer_rd/environment/gi.cpp
+++ b/servers/rendering/renderer_rd/environment/gi.cpp
@@ -109,6 +109,7 @@ void GI::voxel_gi_allocate_data(RID p_voxel_gi, const Transform3D &p_to_cell_xfo
 			Vector<Vector<uint8_t>> s;
 			s.push_back(p_distance_field);
 			voxel_gi->sdf_texture = RD::get_singleton()->texture_create(tf, RD::TextureView(), s);
+			RD::get_singleton()->set_resource_name(voxel_gi->sdf_texture, "VoxelGI SDF Texture");
 		}
 #if 0
 			{
@@ -122,6 +123,7 @@ void GI::voxel_gi_allocate_data(RID p_voxel_gi, const Transform3D &p_to_cell_xfo
 				tf.shareable_formats.push_back(RD::DATA_FORMAT_R8_UNORM);
 				tf.shareable_formats.push_back(RD::DATA_FORMAT_R8_UINT);
 				voxel_gi->sdf_texture = RD::get_singleton()->texture_create(tf, RD::TextureView());
+				RD::get_singleton()->set_resource_name(voxel_gi->sdf_texture, "VoxelGI SDF Texture");
 			}
 			RID shared_tex;
 			{
@@ -402,29 +404,38 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 		RD::TextureFormat tf_render = tf_sdf;
 		tf_render.format = RD::DATA_FORMAT_R16_UINT;
 		render_albedo = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_albedo, "VoxelGI Render Albedo");
 		tf_render.format = RD::DATA_FORMAT_R32_UINT;
 		render_emission = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_emission, "VoxelGI Render Emission");
 		render_emission_aniso = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_emission_aniso, "VoxelGI Render Emission Aniso");
 
 		tf_render.format = RD::DATA_FORMAT_R8_UNORM; //at least its easy to visualize
 
 		for (int i = 0; i < 8; i++) {
 			render_occlusion[i] = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+			RD::get_singleton()->set_resource_name(render_occlusion[i], String("VoxelGI Render Occlusion ") + itos(i));
 		}
 
 		tf_render.format = RD::DATA_FORMAT_R32_UINT;
 		render_geom_facing = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_geom_facing, "VoxelGI Render Geometry Facing");
 
 		tf_render.format = RD::DATA_FORMAT_R8G8B8A8_UINT;
 		render_sdf[0] = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_sdf[0], "VoxelGI Render SDF 0");
 		render_sdf[1] = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_sdf[1], "VoxelGI Render SDF 1");
 
 		tf_render.width /= 2;
 		tf_render.height /= 2;
 		tf_render.depth /= 2;
 
 		render_sdf_half[0] = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_sdf_half[0], "VoxelGI Render SDF Half 0");
 		render_sdf_half[1] = RD::get_singleton()->texture_create(tf_render, RD::TextureView());
+		RD::get_singleton()->set_resource_name(render_sdf_half[1], "VoxelGI Render SDF Half 1");
 	}
 
 	RD::TextureFormat tf_occlusion = tf_sdf;
@@ -465,7 +476,9 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 	tf_probe_average.texture_type = RD::TEXTURE_TYPE_2D;
 
 	lightprobe_history_scroll = RD::get_singleton()->texture_create(tf_probe_history, RD::TextureView());
+	RD::get_singleton()->set_resource_name(lightprobe_history_scroll, "VoxelGI LightProbe History Scroll");
 	lightprobe_average_scroll = RD::get_singleton()->texture_create(tf_probe_average, RD::TextureView());
+	RD::get_singleton()->set_resource_name(lightprobe_average_scroll, "VoxelGI LightProbe Average Scroll");
 
 	{
 		//octahedral lightprobes
@@ -479,6 +492,7 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 		//lightprobe texture is an octahedral texture
 
 		lightprobe_data = RD::get_singleton()->texture_create(tf_octprobes, RD::TextureView());
+		RD::get_singleton()->set_resource_name(lightprobe_data, "VoxelGI LightProbe Data");
 		RD::TextureView tv;
 		tv.format_override = RD::DATA_FORMAT_E5B9G9R9_UFLOAT_PACK32;
 		lightprobe_texture = RD::get_singleton()->texture_create_shared(tv, lightprobe_data);
@@ -492,11 +506,13 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 		tf_ambient.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
 		//lightprobe texture is an octahedral texture
 		ambient_texture = RD::get_singleton()->texture_create(tf_ambient, RD::TextureView());
+		RD::get_singleton()->set_resource_name(ambient_texture, "VoxelGI Ambient Texture");
 	}
 
 	cascades_ubo = RD::get_singleton()->uniform_buffer_create(sizeof(SDFGI::Cascade::UBO) * SDFGI::MAX_CASCADES);
 
 	occlusion_data = RD::get_singleton()->texture_create(tf_occlusion, RD::TextureView());
+	RD::get_singleton()->set_resource_name(occlusion_data, "VoxelGI Occlusion Data");
 	{
 		RD::TextureView tv;
 		tv.format_override = RD::DATA_FORMAT_R4G4B4A4_UNORM_PACK16;
@@ -509,11 +525,15 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 		/* 3D Textures */
 
 		cascade.sdf_tex = RD::get_singleton()->texture_create(tf_sdf, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.sdf_tex, "VoxelGI Cascade SDF Texture");
 
 		cascade.light_data = RD::get_singleton()->texture_create(tf_light, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.light_data, "VoxelGI Cascade Light Data");
 
 		cascade.light_aniso_0_tex = RD::get_singleton()->texture_create(tf_aniso0, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.light_aniso_0_tex, "VoxelGI Cascade Light Aniso 0 Texture");
 		cascade.light_aniso_1_tex = RD::get_singleton()->texture_create(tf_aniso1, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.light_aniso_1_tex, "VoxelGI Cascade Light Aniso 1 Texture");
 
 		{
 			RD::TextureView tv;
@@ -540,9 +560,11 @@ void GI::SDFGI::create(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world
 		/* Probe History */
 
 		cascade.lightprobe_history_tex = RD::get_singleton()->texture_create(tf_probe_history, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.lightprobe_history_tex, "VoxelGI Cascade LightProbe History Texture");
 		RD::get_singleton()->texture_clear(cascade.lightprobe_history_tex, Color(0, 0, 0, 0), 0, 1, 0, tf_probe_history.array_layers); //needs to be cleared for average to work
 
 		cascade.lightprobe_average_tex = RD::get_singleton()->texture_create(tf_probe_average, RD::TextureView());
+		RD::get_singleton()->set_resource_name(cascade.lightprobe_average_tex, "VoxelGI Cascade LightProbe Average Texture");
 		RD::get_singleton()->texture_clear(cascade.lightprobe_average_tex, Color(0, 0, 0, 0), 0, 1, 0, 1); //needs to be cleared for average to work
 
 		/* Buffers */
@@ -2444,6 +2466,7 @@ void GI::VoxelGIInstance::update(bool p_update_light_instances, const Vector<RID
 			tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_CAN_COPY_TO_BIT;
 
 			texture = RD::get_singleton()->texture_create(tf, RD::TextureView());
+			RD::get_singleton()->set_resource_name(texture, "VoxelGI Instance Texture");
 
 			RD::get_singleton()->texture_clear(texture, Color(0, 0, 0, 0), 0, levels.size(), 0, 1);
 
@@ -2573,6 +2596,7 @@ void GI::VoxelGIInstance::update(bool p_update_light_instances, const Vector<RID
 						dtf.usage_bits |= RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
 					}
 					dmap.texture = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+					RD::get_singleton()->set_resource_name(dmap.texture, "VoxelGI Instance DMap Texture");
 
 					if (dynamic_maps.size() == 0) {
 						// Render depth for first one.
@@ -2580,6 +2604,7 @@ void GI::VoxelGIInstance::update(bool p_update_light_instances, const Vector<RID
 						dtf.format = RD::get_singleton()->texture_is_format_supported_for_usage(RD::DATA_FORMAT_D16_UNORM, RD::TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) ? RD::DATA_FORMAT_D16_UNORM : RD::DATA_FORMAT_X8_D24_UNORM_PACK32;
 						dtf.usage_bits = RD::TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
 						dmap.fb_depth = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+						RD::get_singleton()->set_resource_name(dmap.fb_depth, "VoxelGI Instance DMap FB Depth");
 					}
 
 					//just use depth as-is
@@ -2587,13 +2612,17 @@ void GI::VoxelGIInstance::update(bool p_update_light_instances, const Vector<RID
 					dtf.usage_bits = RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
 
 					dmap.depth = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+					RD::get_singleton()->set_resource_name(dmap.depth, "VoxelGI Instance DMap Depth");
 
 					if (dynamic_maps.size() == 0) {
 						dtf.format = RD::DATA_FORMAT_R8G8B8A8_UNORM;
 						dtf.usage_bits = RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT;
 						dmap.albedo = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+						RD::get_singleton()->set_resource_name(dmap.albedo, "VoxelGI Instance DMap Albedo");
 						dmap.normal = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+						RD::get_singleton()->set_resource_name(dmap.normal, "VoxelGI Instance DMap Normal");
 						dmap.orm = RD::get_singleton()->texture_create(dtf, RD::TextureView());
+						RD::get_singleton()->set_resource_name(dmap.orm, "VoxelGI Instance DMap ORM");
 
 						Vector<RID> fb;
 						fb.push_back(dmap.albedo);
@@ -3342,37 +3371,40 @@ void GI::init(RendererSceneSkyRD *p_sky) {
 		//calculate tables
 		String defines = "\n#define SDFGI_OCT_SIZE " + itos(SDFGI::LIGHTPROBE_OCT_SIZE) + "\n";
 		Vector<String> gi_modes;
+
 		gi_modes.push_back("\n#define USE_VOXEL_GI_INSTANCES\n"); // MODE_VOXEL_GI
 		gi_modes.push_back("\n#define USE_SDFGI\n"); // MODE_SDFGI
 		gi_modes.push_back("\n#define USE_SDFGI\n\n#define USE_VOXEL_GI_INSTANCES\n"); // MODE_COMBINED
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_VOXEL_GI_INSTANCES\n"); // MODE_HALF_RES_VOXEL_GI
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n"); // MODE_HALF_RES_SDFGI
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n\n#define USE_VOXEL_GI_INSTANCES\n"); // MODE_HALF_RES_COMBINED
-
-		gi_modes.push_back("\n#define USE_VOXEL_GI_INSTANCES\n#define USE_MULTIVIEW\n"); // MODE_VOXEL_GI_MULTIVIEW
-		gi_modes.push_back("\n#define USE_SDFGI\n#define USE_MULTIVIEW\n"); // MODE_SDFGI_MULTIVIEW
-		gi_modes.push_back("\n#define USE_SDFGI\n\n#define USE_VOXEL_GI_INSTANCES\n#define USE_MULTIVIEW\n"); // MODE_COMBINED_MULTIVIEW
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_VOXEL_GI_INSTANCES\n#define USE_MULTIVIEW\n"); // MODE_HALF_RES_VOXEL_GI_MULTIVIEW
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n#define USE_MULTIVIEW\n"); // MODE_HALF_RES_SDFGI_MULTIVIEW
-		gi_modes.push_back("\n#define MODE_HALF_RES\n#define USE_SDFGI\n\n#define USE_VOXEL_GI_INSTANCES\n#define USE_MULTIVIEW\n"); // MODE_HALF_RES_COMBINED_MULTIVIEW
 
 		shader.initialize(gi_modes, defines);
+		shader_version = shader.version_create();
+
+		Vector<RD::PipelineSpecializationConstant> specialization_constants;
+
+		{
+			RD::PipelineSpecializationConstant sc;
+			sc.type = RD::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL;
+			sc.constant_id = 0; // SHADER_SPECIALIZATION_HALF_RES
+			sc.bool_value = false;
+			specialization_constants.push_back(sc);
 
-		if (!RendererCompositorRD::singleton->is_xr_enabled()) {
-			shader.set_variant_enabled(MODE_VOXEL_GI_MULTIVIEW, false);
-			shader.set_variant_enabled(MODE_SDFGI_MULTIVIEW, false);
-			shader.set_variant_enabled(MODE_COMBINED_MULTIVIEW, false);
-			shader.set_variant_enabled(MODE_HALF_RES_VOXEL_GI_MULTIVIEW, false);
-			shader.set_variant_enabled(MODE_HALF_RES_SDFGI_MULTIVIEW, false);
-			shader.set_variant_enabled(MODE_HALF_RES_COMBINED_MULTIVIEW, false);
+			sc.type = RD::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL;
+			sc.constant_id = 1; // SHADER_SPECIALIZATION_USE_FULL_PROJECTION_MATRIX
+			sc.bool_value = false;
+			specialization_constants.push_back(sc);
+
+			sc.type = RD::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL;
+			sc.constant_id = 2; // SHADER_SPECIALIZATION_USE_VRS
+			sc.bool_value = false;
+			specialization_constants.push_back(sc);
 		}
 
-		shader_version = shader.version_create();
-		for (int i = 0; i < MODE_MAX; i++) {
-			if (shader.is_variant_enabled(i)) {
-				pipelines[i] = RD::get_singleton()->compute_pipeline_create(shader.version_get_shader(shader_version, i));
-			} else {
-				pipelines[i] = RID();
+		for (int v = 0; v < SHADER_SPECIALIZATION_VARIATIONS; v++) {
+			specialization_constants.ptrw()[0].bool_value = (v & SHADER_SPECIALIZATION_HALF_RES) ? true : false;
+			specialization_constants.ptrw()[1].bool_value = (v & SHADER_SPECIALIZATION_USE_FULL_PROJECTION_MATRIX) ? true : false;
+			specialization_constants.ptrw()[2].bool_value = (v & SHADER_SPECIALIZATION_USE_VRS) ? true : false;
+			for (int i = 0; i < MODE_MAX; i++) {
+				pipelines[v][i] = RD::get_singleton()->compute_pipeline_create(shader.version_get_shader(shader_version, i), specialization_constants);
 			}
 		}
 
@@ -3564,25 +3596,17 @@ void GI::RenderBuffersGI::free() {
 	}
 
 	if (ambient_buffer.is_valid()) {
-		if (view_count == 1) {
-			// Only one view? then these are copies of our main buffers.
-			ambient_view[0] = RID();
-			reflection_view[0] = RID();
-		} else {
-			// Multiple views? free our slices.
-			for (uint32_t v = 0; v < view_count; v++) {
-				RD::get_singleton()->free(ambient_view[v]);
-				RD::get_singleton()->free(reflection_view[v]);
-				ambient_view[v] = RID();
-				reflection_view[v] = RID();
-			}
-		}
-
-		// Now we can free our buffers.
 		RD::get_singleton()->free(ambient_buffer);
 		RD::get_singleton()->free(reflection_buffer);
 		ambient_buffer = RID();
 		reflection_buffer = RID();
+
+		// these are automatically freed when we free the textures, so just reset..
+		for (uint32_t v = 0; v < RendererSceneRender::MAX_RENDER_VIEWS; v++) {
+			ambient_slice[v] = RID();
+			reflection_slice[v] = RID();
+		}
+
 		view_count = 0;
 	}
 
@@ -3592,7 +3616,7 @@ void GI::RenderBuffersGI::free() {
 	}
 }
 
-void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_voxel_gi_buffer, RID p_environment, uint32_t p_view_count, const CameraMatrix *p_projections, const Vector3 *p_eye_offsets, const Transform3D &p_cam_transform, const PagedArray<RID> &p_voxel_gi_instances, RendererSceneRenderRD *p_scene_render) {
+void GI::process_gi(RID p_render_buffers, const RID *p_normal_roughness_slices, RID p_voxel_gi_buffer, const RID *p_vrs_slices, RID p_environment, uint32_t p_view_count, const CameraMatrix *p_projections, const Vector3 *p_eye_offsets, const Transform3D &p_cam_transform, const PagedArray<RID> &p_voxel_gi_instances, RendererSceneRenderRD *p_scene_render) {
 	RendererRD::TextureStorage *texture_storage = RendererRD::TextureStorage::get_singleton();
 	RendererRD::MaterialStorage *material_storage = RendererRD::MaterialStorage::get_singleton();
 
@@ -3606,14 +3630,13 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 	if (rb->rbgi.ambient_buffer.is_null() || rb->rbgi.using_half_size_gi != half_resolution || rb->rbgi.view_count != p_view_count) {
 		// Free our old buffer if applicable
 		if (rb->rbgi.ambient_buffer.is_valid()) {
-			if (rb->rbgi.view_count > 1) {
-				for (uint32_t v = 0; v < rb->rbgi.view_count; v++) {
-					RD::get_singleton()->free(rb->rbgi.ambient_view[v]);
-					RD::get_singleton()->free(rb->rbgi.reflection_view[v]);
-				}
-			}
 			RD::get_singleton()->free(rb->rbgi.ambient_buffer);
 			RD::get_singleton()->free(rb->rbgi.reflection_buffer);
+
+			for (uint32_t v = 0; v < RendererSceneRender::MAX_RENDER_VIEWS; v++) {
+				rb->rbgi.ambient_slice[v] = RID();
+				rb->rbgi.reflection_slice[v] = RID();
+			}
 		}
 
 		// Remember the view count we're using
@@ -3637,18 +3660,19 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 		}
 		tf.usage_bits = RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT;
 		rb->rbgi.ambient_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
+		RD::get_singleton()->set_resource_name(rb->rbgi.ambient_buffer, "GI Ambient Buffer");
 		rb->rbgi.reflection_buffer = RD::get_singleton()->texture_create(tf, RD::TextureView());
+		RD::get_singleton()->set_resource_name(rb->rbgi.reflection_buffer, "GI Reflection Buffer");
 		rb->rbgi.using_half_size_gi = half_resolution;
 
 		if (p_view_count == 1) {
-			// Just one view? Copy our buffers
-			rb->rbgi.ambient_view[0] = rb->rbgi.ambient_buffer;
-			rb->rbgi.reflection_view[0] = rb->rbgi.reflection_buffer;
+			// Just copy, we don't need to create slices
+			rb->rbgi.ambient_slice[0] = rb->rbgi.ambient_buffer;
+			rb->rbgi.reflection_slice[0] = rb->rbgi.reflection_buffer;
 		} else {
-			// More then one view? Create slices for each view
 			for (uint32_t v = 0; v < p_view_count; v++) {
-				rb->rbgi.ambient_view[v] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->rbgi.ambient_buffer, v, 0);
-				rb->rbgi.reflection_view[v] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->rbgi.reflection_buffer, v, 0);
+				rb->rbgi.ambient_slice[v] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->rbgi.ambient_buffer, v, 0);
+				rb->rbgi.reflection_slice[v] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), rb->rbgi.reflection_buffer, v, 0);
 			}
 		}
 	}
@@ -3681,29 +3705,45 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 	// Now compute the contents of our buffers.
 	RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin(true);
 
-	for (uint32_t v = 0; v < p_view_count; v++) {
-		// Render each eye seperately.
-		// We need to look into whether we can make our compute shader use Multiview but not sure that works or makes a difference..
+	// Render each eye seperately.
+	// We need to look into whether we can make our compute shader use Multiview but not sure that works or makes a difference..
 
-		// setup our push constant
+	// setup our push constant
 
-		PushConstant push_constant;
+	PushConstant push_constant;
 
-		push_constant.view_index = v;
-		push_constant.orthogonal = p_projections[v].is_orthogonal();
-		push_constant.max_voxel_gi_instances = MIN((uint64_t)MAX_VOXEL_GI_INSTANCES, p_voxel_gi_instances.size());
-		push_constant.high_quality_vct = voxel_gi_quality == RS::VOXEL_GI_QUALITY_HIGH;
+	push_constant.max_voxel_gi_instances = MIN((uint64_t)MAX_VOXEL_GI_INSTANCES, p_voxel_gi_instances.size());
+	push_constant.high_quality_vct = voxel_gi_quality == RS::VOXEL_GI_QUALITY_HIGH;
 
-		push_constant.z_near = p_projections[v].get_z_near();
-		push_constant.z_far = p_projections[v].get_z_far();
+	// these should be the same for all views
+	push_constant.orthogonal = p_projections[0].is_orthogonal();
+	push_constant.z_near = p_projections[0].get_z_near();
+	push_constant.z_far = p_projections[0].get_z_far();
 
-		push_constant.proj_info[0] = -2.0f / (rb->internal_width * p_projections[v].matrix[0][0]);
-		push_constant.proj_info[1] = -2.0f / (rb->internal_height * p_projections[v].matrix[1][1]);
-		push_constant.proj_info[2] = (1.0f - p_projections[v].matrix[0][2]) / p_projections[v].matrix[0][0];
-		push_constant.proj_info[3] = (1.0f + p_projections[v].matrix[1][2]) / p_projections[v].matrix[1][1];
+	// these are only used if we have 1 view, else we use the projections in our scene data
+	push_constant.proj_info[0] = -2.0f / (rb->internal_width * p_projections[0].matrix[0][0]);
+	push_constant.proj_info[1] = -2.0f / (rb->internal_height * p_projections[0].matrix[1][1]);
+	push_constant.proj_info[2] = (1.0f - p_projections[0].matrix[0][2]) / p_projections[0].matrix[0][0];
+	push_constant.proj_info[3] = (1.0f + p_projections[0].matrix[1][2]) / p_projections[0].matrix[1][1];
 
-		bool use_sdfgi = rb->sdfgi != nullptr;
-		bool use_voxel_gi_instances = push_constant.max_voxel_gi_instances > 0;
+	bool use_sdfgi = rb->sdfgi != nullptr;
+	bool use_voxel_gi_instances = push_constant.max_voxel_gi_instances > 0;
+
+	uint32_t pipeline_specialization = 0;
+	if (rb->rbgi.using_half_size_gi) {
+		pipeline_specialization |= SHADER_SPECIALIZATION_HALF_RES;
+	}
+	if (p_view_count > 1) {
+		pipeline_specialization |= SHADER_SPECIALIZATION_USE_FULL_PROJECTION_MATRIX;
+	}
+	if (p_vrs_slices[0].is_valid()) {
+		pipeline_specialization |= SHADER_SPECIALIZATION_USE_VRS;
+	}
+
+	Mode mode = (use_sdfgi && use_voxel_gi_instances) ? MODE_COMBINED : (use_sdfgi ? MODE_SDFGI : MODE_VOXEL_GI);
+
+	for (uint32_t v = 0; v < p_view_count; v++) {
+		push_constant.view_index = v;
 
 		// setup our uniform set
 		if (rb->rbgi.uniform_set[v].is_null() || !RD::get_singleton()->uniform_set_is_valid(rb->rbgi.uniform_set[v])) {
@@ -3790,7 +3830,7 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 				RD::Uniform u;
 				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
 				u.binding = 9;
-				u.append_id(rb->rbgi.ambient_view[v]);
+				u.append_id(rb->rbgi.ambient_slice[v]);
 				uniforms.push_back(u);
 			}
 
@@ -3798,7 +3838,7 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 				RD::Uniform u;
 				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
 				u.binding = 10;
-				u.append_id(rb->rbgi.reflection_view[v]);
+				u.append_id(rb->rbgi.reflection_slice[v]);
 				uniforms.push_back(u);
 			}
 
@@ -3824,7 +3864,7 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 				RD::Uniform u;
 				u.uniform_type = RD::UNIFORM_TYPE_TEXTURE;
 				u.binding = 13;
-				u.append_id(p_normal_roughness_views[v]);
+				u.append_id(p_normal_roughness_slices[v]);
 				uniforms.push_back(u);
 			}
 			{
@@ -3865,27 +3905,19 @@ void GI::process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_v
 				u.append_id(rb->rbgi.scene_data_ubo);
 				uniforms.push_back(u);
 			}
+			{
+				RD::Uniform u;
+				u.uniform_type = RD::UNIFORM_TYPE_IMAGE;
+				u.binding = 19;
+				RID buffer = p_vrs_slices[v].is_valid() ? p_vrs_slices[v] : texture_storage->texture_rd_get_default(RendererRD::DEFAULT_RD_TEXTURE_VRS);
+				u.append_id(buffer);
+				uniforms.push_back(u);
+			}
 
 			rb->rbgi.uniform_set[v] = RD::get_singleton()->uniform_set_create(uniforms, shader.version_get_shader(shader_version, 0), 0);
 		}
 
-		Mode mode;
-
-		if (p_view_count > 1) {
-			if (rb->rbgi.using_half_size_gi) {
-				mode = (use_sdfgi && use_voxel_gi_instances) ? MODE_HALF_RES_COMBINED_MULTIVIEW : (use_sdfgi ? MODE_HALF_RES_SDFGI_MULTIVIEW : MODE_HALF_RES_VOXEL_GI_MULTIVIEW);
-			} else {
-				mode = (use_sdfgi && use_voxel_gi_instances) ? MODE_COMBINED_MULTIVIEW : (use_sdfgi ? MODE_SDFGI_MULTIVIEW : MODE_VOXEL_GI_MULTIVIEW);
-			}
-		} else {
-			if (rb->rbgi.using_half_size_gi) {
-				mode = (use_sdfgi && use_voxel_gi_instances) ? MODE_HALF_RES_COMBINED : (use_sdfgi ? MODE_HALF_RES_SDFGI : MODE_HALF_RES_VOXEL_GI);
-			} else {
-				mode = (use_sdfgi && use_voxel_gi_instances) ? MODE_COMBINED : (use_sdfgi ? MODE_SDFGI : MODE_VOXEL_GI);
-			}
-		}
-
-		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, pipelines[mode]);
+		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, pipelines[pipeline_specialization][mode]);
 		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, rb->rbgi.uniform_set[v], 0);
 		RD::get_singleton()->compute_list_set_push_constant(compute_list, &push_constant, sizeof(PushConstant));
 
diff --git a/servers/rendering/renderer_rd/environment/gi.h b/servers/rendering/renderer_rd/environment/gi.h
index 294b8d3cfd..d950ff9e86 100644
--- a/servers/rendering/renderer_rd/environment/gi.h
+++ b/servers/rendering/renderer_rd/environment/gi.h
@@ -660,13 +660,13 @@ public:
 
 		/* GI buffers */
 		RID ambient_buffer;
+		RID ambient_slice[RendererSceneRender::MAX_RENDER_VIEWS];
 		RID reflection_buffer;
-		RID ambient_view[RendererSceneRender::MAX_RENDER_VIEWS];
-		RID reflection_view[RendererSceneRender::MAX_RENDER_VIEWS];
-		RID uniform_set[RendererSceneRender::MAX_RENDER_VIEWS];
+		RID reflection_slice[RendererSceneRender::MAX_RENDER_VIEWS];
 		bool using_half_size_gi = false;
 		uint32_t view_count = 1;
 
+		RID uniform_set[RendererSceneRender::MAX_RENDER_VIEWS];
 		RID scene_data_ubo;
 
 		void free();
@@ -729,44 +729,41 @@ public:
 	};
 
 	struct PushConstant {
-		uint32_t view_index;
 		uint32_t max_voxel_gi_instances;
 		uint32_t high_quality_vct;
 		uint32_t orthogonal;
+		uint32_t view_index;
 
 		float proj_info[4];
 
 		float z_near;
 		float z_far;
-		float pad1;
 		float pad2;
+		float pad3;
 	};
 
 	RID sdfgi_ubo;
+
 	enum Mode {
 		MODE_VOXEL_GI,
 		MODE_SDFGI,
 		MODE_COMBINED,
-		MODE_HALF_RES_VOXEL_GI,
-		MODE_HALF_RES_SDFGI,
-		MODE_HALF_RES_COMBINED,
-
-		MODE_VOXEL_GI_MULTIVIEW,
-		MODE_SDFGI_MULTIVIEW,
-		MODE_COMBINED_MULTIVIEW,
-		MODE_HALF_RES_VOXEL_GI_MULTIVIEW,
-		MODE_HALF_RES_SDFGI_MULTIVIEW,
-		MODE_HALF_RES_COMBINED_MULTIVIEW,
-
 		MODE_MAX
 	};
 
+	enum ShaderSpecializations {
+		SHADER_SPECIALIZATION_HALF_RES = 1 << 0,
+		SHADER_SPECIALIZATION_USE_FULL_PROJECTION_MATRIX = 1 << 1,
+		SHADER_SPECIALIZATION_USE_VRS = 1 << 2,
+		SHADER_SPECIALIZATION_VARIATIONS = 0x07,
+	};
+
 	RID default_voxel_gi_buffer;
 
 	bool half_resolution = false;
 	GiShaderRD shader;
 	RID shader_version;
-	RID pipelines[MODE_MAX];
+	RID pipelines[SHADER_SPECIALIZATION_VARIATIONS][MODE_MAX];
 
 	GI();
 	~GI();
@@ -777,7 +774,7 @@ public:
 	SDFGI *create_sdfgi(RendererSceneEnvironmentRD *p_env, const Vector3 &p_world_position, uint32_t p_requested_history_size);
 
 	void setup_voxel_gi_instances(RID p_render_buffers, const Transform3D &p_transform, const PagedArray<RID> &p_voxel_gi_instances, uint32_t &r_voxel_gi_instances_used, RendererSceneRenderRD *p_scene_render);
-	void process_gi(RID p_render_buffers, RID *p_normal_roughness_views, RID p_voxel_gi_buffer, RID p_environment, uint32_t p_view_count, const CameraMatrix *p_projections, const Vector3 *p_eye_offsets, const Transform3D &p_cam_transform, const PagedArray<RID> &p_voxel_gi_instances, RendererSceneRenderRD *p_scene_render);
+	void process_gi(RID p_render_buffers, const RID *p_normal_roughness_slices, RID p_voxel_gi_buffer, const RID *p_vrs_slices, RID p_environment, uint32_t p_view_count, const CameraMatrix *p_projections, const Vector3 *p_eye_offsets, const Transform3D &p_cam_transform, const PagedArray<RID> &p_voxel_gi_instances, RendererSceneRenderRD *p_scene_render);
 
 	RID voxel_gi_instance_create(RID p_base);
 	void voxel_gi_instance_set_transform_to_data(RID p_probe, const Transform3D &p_xform);
diff --git a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp
index f759fa3aa5..85652a041d 100644
--- a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp
+++ b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.cpp
@@ -171,29 +171,24 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::ensure_voxelgi()
 }
 
 void RenderForwardClustered::RenderBufferDataForwardClustered::clear() {
+	// note, slices are freed automatically when the parent texture is freed so we just clear them.
+	for (uint32_t v = 0; v < RendererSceneRender::MAX_RENDER_VIEWS; v++) {
+		color_views[v] = RID();
+		depth_views[v] = RID();
+		color_msaa_views[v] = RID();
+		depth_msaa_views[v] = RID();
+		normal_roughness_views[v] = RID();
+		normal_roughness_msaa_views[v] = RID();
+		voxelgi_views[v] = RID();
+		voxelgi_msaa_views[v] = RID();
+		vrs_views[v] = RID();
+	}
+
 	if (voxelgi_buffer != RID()) {
 		RD::get_singleton()->free(voxelgi_buffer);
 		voxelgi_buffer = RID();
 
-		if (view_count == 1) {
-			voxelgi_views[0] = RID();
-		} else {
-			for (uint32_t v = 0; v < view_count; v++) {
-				RD::get_singleton()->free(voxelgi_views[v]);
-				voxelgi_views[v] = RID();
-			}
-		}
-
 		if (voxelgi_buffer_msaa.is_valid()) {
-			if (view_count == 1) {
-				voxelgi_msaa_views[0] = RID();
-			} else {
-				for (uint32_t v = 0; v < view_count; v++) {
-					RD::get_singleton()->free(voxelgi_msaa_views[v]);
-					voxelgi_msaa_views[v] = RID();
-				}
-			}
-
 			RD::get_singleton()->free(voxelgi_buffer_msaa);
 			voxelgi_buffer_msaa = RID();
 		}
@@ -202,35 +197,11 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::clear() {
 	}
 
 	if (color_msaa.is_valid()) {
-		if (view_count == 1) {
-			color_views[0] = RID();
-			color_msaa_views[0] = RID();
-		} else {
-			for (uint32_t v = 0; v < view_count; v++) {
-				RD::get_singleton()->free(color_views[v]);
-				RD::get_singleton()->free(color_msaa_views[v]);
-				color_views[v] = RID();
-				color_msaa_views[v] = RID();
-			}
-		}
-
 		RD::get_singleton()->free(color_msaa);
 		color_msaa = RID();
 	}
 
 	if (depth_msaa.is_valid()) {
-		if (view_count == 1) {
-			depth_views[0] = RID();
-			depth_msaa_views[0] = RID();
-		} else {
-			for (uint32_t v = 0; v < view_count; v++) {
-				RD::get_singleton()->free(depth_views[v]);
-				RD::get_singleton()->free(depth_msaa_views[v]);
-				depth_views[v] = RID();
-				depth_msaa_views[v] = RID();
-			}
-		}
-
 		RD::get_singleton()->free(depth_msaa);
 		depth_msaa = RID();
 	}
@@ -245,33 +216,17 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::clear() {
 	}
 
 	color = RID();
+	color_only_fb = RID();
 	depth = RID();
 	depth_fb = RID();
 
 	color_framebuffers.clear(); // Color pass framebuffers are freed automatically by their dependency relations
 
 	if (normal_roughness_buffer.is_valid()) {
-		if (view_count == 1) {
-			normal_roughness_views[0] = RID();
-		} else {
-			for (uint32_t v = 0; v < view_count; v++) {
-				RD::get_singleton()->free(normal_roughness_views[v]);
-				normal_roughness_views[v] = RID();
-			}
-		}
-
 		RD::get_singleton()->free(normal_roughness_buffer);
 		normal_roughness_buffer = RID();
 
 		if (normal_roughness_buffer_msaa.is_valid()) {
-			if (view_count == 1) {
-				normal_roughness_msaa_views[0] = RID();
-			} else {
-				for (uint32_t v = 0; v < view_count; v++) {
-					RD::get_singleton()->free(normal_roughness_msaa_views[v]);
-					normal_roughness_msaa_views[v] = RID();
-				}
-			}
 			RD::get_singleton()->free(normal_roughness_buffer_msaa);
 			normal_roughness_buffer_msaa = RID();
 		}
@@ -294,11 +249,12 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::clear() {
 	}
 }
 
-void RenderForwardClustered::RenderBufferDataForwardClustered::configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count) {
+void RenderForwardClustered::RenderBufferDataForwardClustered::configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count, RID p_vrs_texture) {
 	clear();
 
 	msaa = p_msaa;
 	use_taa = p_use_taa;
+	vrs = p_vrs_texture;
 
 	width = p_width;
 	height = p_height;
@@ -307,11 +263,26 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::configure(RID p_c
 	color = p_color_buffer;
 	depth = p_depth_buffer;
 
+	if (vrs.is_valid()) {
+		if (view_count == 1) {
+			// just reuse
+			vrs_views[0] = vrs;
+		} else {
+			// create slices
+			for (uint32_t v = 0; v < view_count; v++) {
+				vrs_views[v] = RD::get_singleton()->texture_create_shared_from_slice(RD::TextureView(), vrs, v, 0);
+			}
+		}
+	}
+
 	if (p_msaa == RS::VIEWPORT_MSAA_DISABLED) {
 		{
 			Vector<RID> fb;
 			fb.push_back(p_color_buffer);
 			fb.push_back(depth);
+			if (vrs.is_valid()) {
+				fb.push_back(vrs);
+			}
 
 			color_only_fb = RD::get_singleton()->framebuffer_create(fb, RenderingDevice::INVALID_ID, view_count);
 		}
@@ -371,6 +342,9 @@ void RenderForwardClustered::RenderBufferDataForwardClustered::configure(RID p_c
 			Vector<RID> fb;
 			fb.push_back(color_msaa);
 			fb.push_back(depth_msaa);
+			if (vrs.is_valid()) {
+				fb.push_back(vrs);
+			}
 
 			color_only_fb = RD::get_singleton()->framebuffer_create(fb, RenderingDevice::INVALID_ID, view_count);
 		}
@@ -409,6 +383,10 @@ RID RenderForwardClustered::RenderBufferDataForwardClustered::get_color_pass_fb(
 
 	fb.push_back(use_msaa ? depth_msaa : depth);
 
+	if (vrs.is_valid()) {
+		fb.push_back(vrs);
+	}
+
 	int v_count = (p_color_pass_flags & COLOR_PASS_FLAG_MULTIVIEW) ? view_count : 1;
 	RID framebuffer = RD::get_singleton()->framebuffer_create(fb, RD::INVALID_ID, v_count);
 	color_framebuffers[p_color_pass_flags] = framebuffer;
@@ -1673,8 +1651,8 @@ void RenderForwardClustered::_render_scene(RenderDataRD *p_render_data, const Co
 		continue_depth = !finish_depth;
 	}
 
-	RID null_rids[2];
-	_pre_opaque_render(p_render_data, using_ssao, using_ssil, using_sdfgi || using_voxelgi, render_buffer ? render_buffer->normal_roughness_views : null_rids, render_buffer ? render_buffer->voxelgi_buffer : RID());
+	RID nullrids[RendererSceneRender::MAX_RENDER_VIEWS];
+	_pre_opaque_render(p_render_data, using_ssao, using_ssil, using_sdfgi || using_voxelgi, render_buffer ? render_buffer->normal_roughness_views : nullrids, render_buffer ? render_buffer->voxelgi_buffer : RID(), render_buffer ? render_buffer->vrs_views : nullrids);
 
 	RD::get_singleton()->draw_command_begin_label("Render Opaque Pass");
 
diff --git a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h
index 9e1f1b9954..ff712a20a1 100644
--- a/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h
+++ b/servers/rendering/renderer_rd/forward_clustered/render_forward_clustered.h
@@ -107,11 +107,14 @@ class RenderForwardClustered : public RendererSceneRenderRD {
 		RID depth_normal_roughness_voxelgi_fb;
 		RID color_only_fb;
 		RID specular_only_fb;
+
+		RID vrs;
+
 		int width, height;
 		HashMap<uint32_t, RID> color_framebuffers;
 
 		// for multiview
-		uint32_t view_count;
+		uint32_t view_count = 1;
 		RID color_views[RendererSceneRender::MAX_RENDER_VIEWS]; // we should rewrite this so we get access to the existing views in our renderer, something we can address when we reorg this
 		RID depth_views[RendererSceneRender::MAX_RENDER_VIEWS]; // we should rewrite this so we get access to the existing views in our renderer, something we can address when we reorg this
 		RID color_msaa_views[RendererSceneRender::MAX_RENDER_VIEWS];
@@ -120,13 +123,14 @@ class RenderForwardClustered : public RendererSceneRenderRD {
 		RID normal_roughness_msaa_views[RendererSceneRender::MAX_RENDER_VIEWS];
 		RID voxelgi_views[RendererSceneRender::MAX_RENDER_VIEWS];
 		RID voxelgi_msaa_views[RendererSceneRender::MAX_RENDER_VIEWS];
+		RID vrs_views[RendererSceneRender::MAX_RENDER_VIEWS];
 
 		RID render_sdfgi_uniform_set;
 		void ensure_specular();
 		void ensure_voxelgi();
 		void ensure_velocity();
 		void clear();
-		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count);
+		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count, RID p_vrs_texture);
 		RID get_color_pass_fb(uint32_t p_color_pass_flags);
 
 		~RenderBufferDataForwardClustered();
diff --git a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp
index e1855ddb36..966621c93e 100644
--- a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp
+++ b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.cpp
@@ -87,10 +87,11 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::clear() {
 	}
 }
 
-void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count) {
+void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count, RID p_vrs_texture) {
 	clear();
 
 	msaa = p_msaa;
+	vrs = p_vrs_texture;
 
 	Size2i target_size = RD::get_singleton()->texture_size(p_target_buffer);
 
@@ -108,6 +109,9 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 		Vector<RID> fb;
 		fb.push_back(p_color_buffer); // 0 - color buffer
 		fb.push_back(depth); // 1 - depth buffer
+		if (vrs.is_valid()) {
+			fb.push_back(vrs); // 2 - vrs texture
+		}
 
 		// Now define our subpasses
 		Vector<RD::FramebufferPass> passes;
@@ -116,6 +120,9 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 		// re-using the same attachments
 		pass.color_attachments.push_back(0);
 		pass.depth_attachment = 1;
+		if (vrs.is_valid()) {
+			pass.vrs_attachment = 2;
+		}
 
 		// - opaque pass
 		passes.push_back(pass);
@@ -131,12 +138,13 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 
 		if (!is_scaled) {
 			// - add blit to 2D pass
-			fb.push_back(p_target_buffer); // 2 - target buffer
+			int target_buffer_id = fb.size();
+			fb.push_back(p_target_buffer); // 2/3 - target buffer
 
 			RD::FramebufferPass blit_pass;
-			blit_pass.color_attachments.push_back(2);
+			blit_pass.color_attachments.push_back(target_buffer_id);
 			blit_pass.input_attachments.push_back(0);
-			passes.push_back(blit_pass);
+			passes.push_back(blit_pass); // this doesn't need VRS
 
 			color_fbs[FB_CONFIG_FOUR_SUBPASSES] = RD::get_singleton()->framebuffer_create_multipass(fb, passes, RenderingDevice::INVALID_ID, view_count);
 		} else {
@@ -179,6 +187,9 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 			Vector<RID> fb;
 			fb.push_back(color_msaa); // 0 - msaa color buffer
 			fb.push_back(depth_msaa); // 1 - msaa depth buffer
+			if (vrs.is_valid()) {
+				fb.push_back(vrs); // 2 - vrs texture
+			}
 
 			// Now define our subpasses
 			Vector<RD::FramebufferPass> passes;
@@ -187,18 +198,22 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 			// re-using the same attachments
 			pass.color_attachments.push_back(0);
 			pass.depth_attachment = 1;
+			if (vrs.is_valid()) {
+				pass.vrs_attachment = 2;
+			}
 
 			// - opaque pass
 			passes.push_back(pass);
 
 			// - add sky pass
-			fb.push_back(color); // 2 - color buffer
+			int color_buffer_id = fb.size();
+			fb.push_back(color); // color buffer
 			passes.push_back(pass); // without resolve for our 3 + 4 subpass config
 			{
 				// but with resolve for our 2 subpass config
 				Vector<RD::FramebufferPass> two_passes;
 				two_passes.push_back(pass); // opaque subpass without resolve
-				pass.resolve_attachments.push_back(2);
+				pass.resolve_attachments.push_back(color_buffer_id);
 				two_passes.push_back(pass); // sky subpass with resolve
 
 				color_fbs[FB_CONFIG_TWO_SUBPASSES] = RD::get_singleton()->framebuffer_create_multipass(fb, two_passes, RenderingDevice::INVALID_ID, view_count);
@@ -217,10 +232,11 @@ void RenderForwardMobile::RenderBufferDataForwardMobile::configure(RID p_color_b
 
 			if (!is_scaled) {
 				// - add blit to 2D pass
-				fb.push_back(p_target_buffer); // 3 - target buffer
+				int target_buffer_id = fb.size();
+				fb.push_back(p_target_buffer); // target buffer
 				RD::FramebufferPass blit_pass;
-				blit_pass.color_attachments.push_back(3);
-				blit_pass.input_attachments.push_back(2);
+				blit_pass.color_attachments.push_back(target_buffer_id);
+				blit_pass.input_attachments.push_back(color_buffer_id);
 				passes.push_back(blit_pass);
 
 				color_fbs[FB_CONFIG_FOUR_SUBPASSES] = RD::get_singleton()->framebuffer_create_multipass(fb, passes, RenderingDevice::INVALID_ID, view_count);
@@ -675,8 +691,8 @@ void RenderForwardMobile::_render_scene(RenderDataRD *p_render_data, const Color
 		RD::get_singleton()->draw_command_end_label(); // Setup Sky resolution buffers
 	}
 
-	RID null_rids[2];
-	_pre_opaque_render(p_render_data, false, false, false, null_rids, RID());
+	RID nullrids[RendererSceneRender::MAX_RENDER_VIEWS];
+	_pre_opaque_render(p_render_data, false, false, false, nullrids, RID(), nullrids);
 
 	uint32_t spec_constant_base_flags = 0;
 
diff --git a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.h b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.h
index 473a58045c..bf4a52d466 100644
--- a/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.h
+++ b/servers/rendering/renderer_rd/forward_mobile/render_forward_mobile.h
@@ -131,12 +131,14 @@ protected:
 		RID depth_msaa;
 		// RID normal_roughness_buffer_msaa;
 
+		RID vrs;
+
 		RID color_fbs[FB_CONFIG_MAX];
 		int width, height;
 		uint32_t view_count;
 
 		void clear();
-		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count);
+		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count, RID p_vrs_texture);
 
 		~RenderBufferDataForwardMobile();
 	};
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
index 120bd9ece3..a2a0538e04 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
+++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
@@ -1827,6 +1827,16 @@ void RendererSceneRenderRD::_free_render_buffer_data(RenderBuffers *rb) {
 		rb->sss_texture = RID();
 	}
 
+	if (rb->vrs_fb.is_valid()) {
+		RD::get_singleton()->free(rb->vrs_fb);
+		rb->vrs_fb = RID();
+	}
+
+	if (rb->vrs_texture.is_valid()) {
+		RD::get_singleton()->free(rb->vrs_texture);
+		rb->vrs_texture = RID();
+	}
+
 	for (int i = 0; i < 2; i++) {
 		for (int l = 0; l < rb->blur[i].layers.size(); l++) {
 			for (int m = 0; m < rb->blur[i].layers[l].mipmaps.size(); m++) {
@@ -3151,8 +3161,13 @@ void RendererSceneRenderRD::render_buffers_configure(RID p_render_buffers, RID p
 		}
 	}
 
+	RS::ViewportVRSMode vrs_mode = texture_storage->render_target_get_vrs_mode(rb->render_target);
+	if (is_vrs_supported() && vrs_mode != RS::VIEWPORT_VRS_DISABLED) {
+		vrs->create_vrs_texture(p_internal_width, p_internal_height, p_view_count, rb->vrs_texture, rb->vrs_fb);
+	}
+
 	RID target_texture = texture_storage->render_target_get_rd_texture(rb->render_target);
-	rb->data->configure(rb->internal_texture, rb->depth_texture, target_texture, p_internal_width, p_internal_height, p_msaa, p_use_taa, p_view_count);
+	rb->data->configure(rb->internal_texture, rb->depth_texture, target_texture, p_internal_width, p_internal_height, p_msaa, p_use_taa, p_view_count, rb->vrs_texture);
 
 	if (is_clustered_enabled()) {
 		rb->cluster_builder->setup(Size2i(p_internal_width, p_internal_height), max_cluster_elements, rb->depth_texture, RendererRD::MaterialStorage::get_singleton()->sampler_rd_get_default(RS::CANVAS_ITEM_TEXTURE_FILTER_NEAREST, RS::CANVAS_ITEM_TEXTURE_REPEAT_DISABLED), rb->internal_texture);
@@ -4929,7 +4944,7 @@ void RendererSceneRenderRD::_pre_resolve_render(RenderDataRD *p_render_data, boo
 	}
 }
 
-void RendererSceneRenderRD::_pre_opaque_render(RenderDataRD *p_render_data, bool p_use_ssao, bool p_use_ssil, bool p_use_gi, RID *p_normal_roughness_views, RID p_voxel_gi_buffer) {
+void RendererSceneRenderRD::_pre_opaque_render(RenderDataRD *p_render_data, bool p_use_ssao, bool p_use_ssil, bool p_use_gi, const RID *p_normal_roughness_slices, RID p_voxel_gi_buffer, const RID *p_vrs_slices) {
 	// Render shadows while GI is rendering, due to how barriers are handled, this should happen at the same time
 	RendererRD::LightStorage *light_storage = RendererRD::LightStorage::get_singleton();
 
@@ -5004,7 +5019,7 @@ void RendererSceneRenderRD::_pre_opaque_render(RenderDataRD *p_render_data, bool
 
 	//start GI
 	if (render_gi) {
-		gi.process_gi(p_render_data->render_buffers, p_normal_roughness_views, p_voxel_gi_buffer, p_render_data->environment, p_render_data->view_count, p_render_data->view_projection, p_render_data->view_eye_offset, p_render_data->cam_transform, *p_render_data->voxel_gi_instances, this);
+		gi.process_gi(p_render_data->render_buffers, p_normal_roughness_slices, p_voxel_gi_buffer, p_vrs_slices, p_render_data->environment, p_render_data->view_count, p_render_data->view_projection, p_render_data->view_eye_offset, p_render_data->cam_transform, *p_render_data->voxel_gi_instances, this);
 	}
 
 	//Do shadow rendering (in parallel with GI)
@@ -5045,13 +5060,13 @@ void RendererSceneRenderRD::_pre_opaque_render(RenderDataRD *p_render_data, bool
 		}
 
 		if (p_use_ssao) {
-			// TODO make these proper stereo and thus use p_normal_roughness_views correctly
-			_process_ssao(p_render_data->render_buffers, p_render_data->environment, p_normal_roughness_views[0], p_render_data->cam_projection);
+			// TODO make these proper stereo
+			_process_ssao(p_render_data->render_buffers, p_render_data->environment, p_normal_roughness_slices[0], p_render_data->cam_projection);
 		}
 
 		if (p_use_ssil) {
-			// TODO make these proper stereo and thus use p_normal_roughness_views correctly
-			_process_ssil(p_render_data->render_buffers, p_render_data->environment, p_normal_roughness_views[0], p_render_data->cam_projection, p_render_data->cam_transform);
+			// TODO make these proper stereo
+			_process_ssil(p_render_data->render_buffers, p_render_data->environment, p_normal_roughness_slices[0], p_render_data->cam_projection, p_render_data->cam_transform);
 		}
 	}
 
@@ -5240,6 +5255,11 @@ void RendererSceneRenderRD::render_scene(RID p_render_buffers, const CameraData
 		render_data.cluster_max_elements = current_cluster_builder->get_max_cluster_elements();
 	}
 
+	if (rb != nullptr && rb->vrs_fb.is_valid()) {
+		// vrs_fb will only be valid if vrs is enabled
+		vrs->update_vrs_texture(rb->vrs_fb, rb->render_target);
+	}
+
 	_render_scene(&render_data, clear_color);
 
 	if (p_render_buffers.is_valid()) {
@@ -5736,6 +5756,10 @@ int RendererSceneRenderRD::get_max_directional_lights() const {
 	return cluster.max_directional_lights;
 }
 
+bool RendererSceneRenderRD::is_vrs_supported() const {
+	return RD::get_singleton()->has_feature(RD::SUPPORTS_ATTACHMENT_VRS);
+}
+
 bool RendererSceneRenderRD::is_dynamic_gi_supported() const {
 	// usable by default (unless low end = true)
 	return true;
@@ -5975,6 +5999,7 @@ void fog() {
 	bokeh_dof = memnew(RendererRD::BokehDOF(!can_use_storage));
 	copy_effects = memnew(RendererRD::CopyEffects(!can_use_storage));
 	tone_mapper = memnew(RendererRD::ToneMapper);
+	vrs = memnew(RendererRD::VRS);
 }
 
 RendererSceneRenderRD::~RendererSceneRenderRD() {
@@ -5989,6 +6014,9 @@ RendererSceneRenderRD::~RendererSceneRenderRD() {
 	if (tone_mapper) {
 		memdelete(tone_mapper);
 	}
+	if (vrs) {
+		memdelete(vrs);
+	}
 
 	for (const KeyValue<int, ShadowCubemap> &E : shadow_cubemaps) {
 		RD::get_singleton()->free(E.value.cubemap);
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.h b/servers/rendering/renderer_rd/renderer_scene_render_rd.h
index 4249e7dbe4..d11bbd183e 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_rd.h
+++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.h
@@ -38,6 +38,7 @@
 #include "servers/rendering/renderer_rd/effects/bokeh_dof.h"
 #include "servers/rendering/renderer_rd/effects/copy_effects.h"
 #include "servers/rendering/renderer_rd/effects/tone_mapper.h"
+#include "servers/rendering/renderer_rd/effects/vrs.h"
 #include "servers/rendering/renderer_rd/environment/gi.h"
 #include "servers/rendering/renderer_rd/renderer_scene_environment_rd.h"
 #include "servers/rendering/renderer_rd/renderer_scene_sky_rd.h"
@@ -104,11 +105,12 @@ protected:
 	RendererRD::BokehDOF *bokeh_dof = nullptr;
 	RendererRD::CopyEffects *copy_effects = nullptr;
 	RendererRD::ToneMapper *tone_mapper = nullptr;
+	RendererRD::VRS *vrs = nullptr;
 	double time = 0.0;
 	double time_step = 0.0;
 
 	struct RenderBufferData {
-		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count) = 0;
+		virtual void configure(RID p_color_buffer, RID p_depth_buffer, RID p_target_buffer, int p_width, int p_height, RS::ViewportMSAA p_msaa, bool p_use_taa, uint32_t p_view_count, RID p_vrs_texture) = 0;
 		virtual ~RenderBufferData() {}
 	};
 	virtual RenderBufferData *_create_render_buffer_data() = 0;
@@ -149,7 +151,7 @@ protected:
 	void _post_prepass_render(RenderDataRD *p_render_data, bool p_use_gi);
 	void _pre_resolve_render(RenderDataRD *p_render_data, bool p_use_gi);
 
-	void _pre_opaque_render(RenderDataRD *p_render_data, bool p_use_ssao, bool p_use_ssil, bool p_use_gi, RID *p_normal_roughness_views, RID p_voxel_gi_buffer);
+	void _pre_opaque_render(RenderDataRD *p_render_data, bool p_use_ssao, bool p_use_ssil, bool p_use_gi, const RID *p_normal_roughness_slices, RID p_voxel_gi_buffer, const RID *p_vrs_slices);
 
 	void _render_buffers_copy_screen_texture(const RenderDataRD *p_render_data);
 	void _render_buffers_copy_depth_texture(const RenderDataRD *p_render_data);
@@ -492,6 +494,8 @@ private:
 		RID depth_texture; //main depth texture
 		RID texture_fb; // framebuffer for the main texture, ONLY USED FOR MOBILE RENDERER POST EFFECTS, DO NOT USE FOR RENDERING 3D!!!
 		RID upscale_texture; //used when upscaling internal_texture (This uses the same resource as internal_texture if there is no upscaling)
+		RID vrs_texture; // texture for vrs.
+		RID vrs_fb; // framebuffer to write to our vrs texture
 
 		// Access to the layers for each of our views (specifically needed for applying post effects on stereoscopic images)
 		struct View {
@@ -1503,6 +1507,7 @@ public:
 
 	virtual void sdfgi_set_debug_probe_select(const Vector3 &p_position, const Vector3 &p_dir) override;
 
+	virtual bool is_vrs_supported() const;
 	virtual bool is_dynamic_gi_supported() const;
 	virtual bool is_clustered_enabled() const;
 	virtual bool is_volumetric_supported() const;
diff --git a/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl b/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl
index 9787c9879d..1c17eabb56 100644
--- a/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl
+++ b/servers/rendering/renderer_rd/shaders/effects/copy_to_fb.glsl
@@ -88,7 +88,7 @@ layout(set = 0, binding = 0) uniform sampler2DArray source_color;
 layout(set = 1, binding = 0) uniform sampler2DArray source_depth;
 layout(location = 1) out float depth;
 #endif /* MODE_TWO_SOURCES */
-#else
+#else /* MULTIVIEW */
 layout(set = 0, binding = 0) uniform sampler2D source_color;
 #ifdef MODE_TWO_SOURCES
 layout(set = 1, binding = 0) uniform sampler2D source_color2;
@@ -139,7 +139,7 @@ void main() {
 		//uv.y = 1.0 - uv.y;
 		uv = 1.0 - uv;
 	}
-#endif
+#endif /* MODE_PANORAMA_TO_DP */
 
 #ifdef MULTIVIEW
 	vec4 color = textureLod(source_color, uv, 0.0);
@@ -148,12 +148,13 @@ void main() {
 	depth = textureLod(source_depth, uv, 0.0).r;
 #endif /* MODE_TWO_SOURCES */
 
-#else
+#else /* MULTIVIEW */
 	vec4 color = textureLod(source_color, uv, 0.0);
 #ifdef MODE_TWO_SOURCES
 	color += textureLod(source_color2, uv, 0.0);
 #endif /* MODE_TWO_SOURCES */
 #endif /* MULTIVIEW */
+
 	if (params.force_luminance) {
 		color.rgb = vec3(max(max(color.r, color.g), color.b));
 	}
@@ -163,5 +164,6 @@ void main() {
 	if (params.srgb) {
 		color.rgb = linear_to_srgb(color.rgb);
 	}
+
 	frag_color = color;
 }
diff --git a/servers/rendering/renderer_rd/shaders/effects/vrs.glsl b/servers/rendering/renderer_rd/shaders/effects/vrs.glsl
new file mode 100644
index 0000000000..5ef83c0b44
--- /dev/null
+++ b/servers/rendering/renderer_rd/shaders/effects/vrs.glsl
@@ -0,0 +1,72 @@
+#[vertex]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef MULTIVIEW
+#ifdef has_VK_KHR_multiview
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#else // has_VK_KHR_multiview
+#define ViewIndex 0
+#endif // has_VK_KHR_multiview
+#endif //MULTIVIEW
+
+#ifdef MULTIVIEW
+layout(location = 0) out vec3 uv_interp;
+#else
+layout(location = 0) out vec2 uv_interp;
+#endif
+
+void main() {
+	vec2 base_arr[4] = vec2[](vec2(0.0, 0.0), vec2(0.0, 1.0), vec2(1.0, 1.0), vec2(1.0, 0.0));
+	uv_interp.xy = base_arr[gl_VertexIndex];
+#ifdef MULTIVIEW
+	uv_interp.z = ViewIndex;
+#endif
+
+	gl_Position = vec4(uv_interp.xy * 2.0 - 1.0, 0.0, 1.0);
+}
+
+#[fragment]
+
+#version 450
+
+#VERSION_DEFINES
+
+#ifdef MULTIVIEW
+#ifdef has_VK_KHR_multiview
+#extension GL_EXT_multiview : enable
+#define ViewIndex gl_ViewIndex
+#else // has_VK_KHR_multiview
+#define ViewIndex 0
+#endif // has_VK_KHR_multiview
+#endif //MULTIVIEW
+
+#ifdef MULTIVIEW
+layout(location = 0) in vec3 uv_interp;
+layout(set = 0, binding = 0) uniform sampler2DArray source_color;
+#else /* MULTIVIEW */
+layout(location = 0) in vec2 uv_interp;
+layout(set = 0, binding = 0) uniform sampler2D source_color;
+#endif /* MULTIVIEW */
+
+layout(location = 0) out uint frag_color;
+
+void main() {
+#ifdef MULTIVIEW
+	vec3 uv = uv_interp;
+#else
+	vec2 uv = uv_interp;
+#endif
+
+#ifdef MULTIVIEW
+	vec4 color = textureLod(source_color, uv, 0.0);
+#else /* MULTIVIEW */
+	vec4 color = textureLod(source_color, uv, 0.0);
+#endif /* MULTIVIEW */
+
+	// See if we can change the sampler to one that returns int...
+	frag_color = uint(color.r * 256.0);
+}
diff --git a/servers/rendering/renderer_rd/shaders/environment/gi.glsl b/servers/rendering/renderer_rd/shaders/environment/gi.glsl
index f687d50a2d..5f34e7112d 100644
--- a/servers/rendering/renderer_rd/shaders/environment/gi.glsl
+++ b/servers/rendering/renderer_rd/shaders/environment/gi.glsl
@@ -8,6 +8,12 @@ layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
 
 #define M_PI 3.141592
 
+/* Specialization Constants (Toggles) */
+
+layout(constant_id = 0) const bool sc_half_res = false;
+layout(constant_id = 1) const bool sc_use_full_projection_matrix = false;
+layout(constant_id = 2) const bool sc_use_vrs = false;
+
 #define SDFGI_MAX_CASCADES 8
 
 //set 0 for SDFGI and render buffers
@@ -97,18 +103,20 @@ layout(set = 0, binding = 18, std140) uniform SceneData {
 }
 scene_data;
 
+layout(r8ui, set = 0, binding = 19) uniform restrict readonly uimage2D vrs_buffer;
+
 layout(push_constant, std430) uniform Params {
-	uint view_index;
 	uint max_voxel_gi_instances;
 	bool high_quality_vct;
 	bool orthogonal;
+	uint view_index;
 
 	vec4 proj_info;
 
 	float z_near;
 	float z_far;
-	float pad1;
 	float pad2;
+	float pad3;
 }
 params;
 
@@ -140,34 +148,34 @@ vec4 blend_color(vec4 src, vec4 dst) {
 }
 
 vec3 reconstruct_position(ivec2 screen_pos) {
-#ifdef USE_MULTIVIEW
-	vec4 pos;
-	pos.xy = (2.0 * vec2(screen_pos) / vec2(scene_data.screen_size)) - 1.0;
-	pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r * 2.0 - 1.0;
-	pos.w = 1.0;
+	if (sc_use_full_projection_matrix) {
+		vec4 pos;
+		pos.xy = (2.0 * vec2(screen_pos) / vec2(scene_data.screen_size)) - 1.0;
+		pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r * 2.0 - 1.0;
+		pos.w = 1.0;
 
-	pos = scene_data.inv_projection[params.view_index] * pos;
+		pos = scene_data.inv_projection[params.view_index] * pos;
 
-	return pos.xyz / pos.w;
-#else
-	vec3 pos;
-	pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r;
-
-	pos.z = pos.z * 2.0 - 1.0;
-	if (params.orthogonal) {
-		pos.z = ((pos.z + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
+		return pos.xyz / pos.w;
 	} else {
-		pos.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - pos.z * (params.z_far - params.z_near));
-	}
-	pos.z = -pos.z;
+		vec3 pos;
+		pos.z = texelFetch(sampler2D(depth_buffer, linear_sampler), screen_pos, 0).r;
+
+		pos.z = pos.z * 2.0 - 1.0;
+		if (params.orthogonal) {
+			pos.z = ((pos.z + (params.z_far + params.z_near) / (params.z_far - params.z_near)) * (params.z_far - params.z_near)) / 2.0;
+		} else {
+			pos.z = 2.0 * params.z_near * params.z_far / (params.z_far + params.z_near - pos.z * (params.z_far - params.z_near));
+		}
+		pos.z = -pos.z;
 
-	pos.xy = vec2(screen_pos) * params.proj_info.xy + params.proj_info.zw;
-	if (!params.orthogonal) {
-		pos.xy *= pos.z;
-	}
+		pos.xy = vec2(screen_pos) * params.proj_info.xy + params.proj_info.zw;
+		if (!params.orthogonal) {
+			pos.xy *= pos.z;
+		}
 
-	return pos;
-#endif
+		return pos;
+	}
 }
 
 void sdfvoxel_gi_process(uint cascade, vec3 cascade_pos, vec3 cam_pos, vec3 cam_normal, vec3 cam_specular_normal, float roughness, out vec3 diffuse_light, out vec3 specular_light) {
@@ -587,7 +595,6 @@ void voxel_gi_compute(uint index, vec3 position, vec3 normal, vec3 ref_vec, mat3
 
 vec4 fetch_normal_and_roughness(ivec2 pos) {
 	vec4 normal_roughness = texelFetch(sampler2D(normal_roughness_buffer, linear_sampler), pos, 0);
-
 	normal_roughness.xyz = normalize(normal_roughness.xyz * 2.0 - 1.0);
 	return normal_roughness;
 }
@@ -600,7 +607,7 @@ void process_gi(ivec2 pos, vec3 vertex, inout vec4 ambient_light, inout vec4 ref
 	if (normal.length() > 0.5) {
 		//valid normal, can do GI
 		float roughness = normal_roughness.w;
-		vec3 view = -normalize(mat3(scene_data.cam_transform) * (vertex - scene_data.eye_offset[params.view_index].xyz));
+		vec3 view = -normalize(mat3(scene_data.cam_transform) * (vertex - scene_data.eye_offset[gl_GlobalInvocationID.z].xyz));
 		vertex = mat3(scene_data.cam_transform) * vertex;
 		normal = normalize(mat3(scene_data.cam_transform) * normal);
 		vec3 reflection = normalize(reflect(-view, normal));
@@ -648,9 +655,35 @@ void process_gi(ivec2 pos, vec3 vertex, inout vec4 ambient_light, inout vec4 ref
 void main() {
 	ivec2 pos = ivec2(gl_GlobalInvocationID.xy);
 
-#ifdef MODE_HALF_RES
-	pos <<= 1;
-#endif
+	uint vrs_x, vrs_y;
+	if (sc_use_vrs) {
+		ivec2 vrs_pos;
+
+		// Currenty we use a 16x16 texel, possibly some day make this configurable.
+		if (sc_half_res) {
+			vrs_pos = pos >> 3;
+		} else {
+			vrs_pos = pos >> 4;
+		}
+
+		uint vrs_texel = imageLoad(vrs_buffer, vrs_pos).r;
+		// note, valid values for vrs_x and vrs_y are 1, 2 and 4.
+		vrs_x = 1 << ((vrs_texel >> 2) & 3);
+		vrs_y = 1 << (vrs_texel & 3);
+
+		if (mod(pos.x, vrs_x) != 0) {
+			return;
+		}
+
+		if (mod(pos.y, vrs_y) != 0) {
+			return;
+		}
+	}
+
+	if (sc_half_res) {
+		pos <<= 1;
+	}
+
 	if (any(greaterThanEqual(pos, scene_data.screen_size))) { //too large, do nothing
 		return;
 	}
@@ -663,10 +696,69 @@ void main() {
 
 	process_gi(pos, vertex, ambient_light, reflection_light);
 
-#ifdef MODE_HALF_RES
-	pos >>= 1;
-#endif
+	if (sc_half_res) {
+		pos >>= 1;
+	}
 
 	imageStore(ambient_buffer, pos, ambient_light);
 	imageStore(reflection_buffer, pos, reflection_light);
+
+	if (sc_use_vrs) {
+		if (vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 0), reflection_light);
+		}
+
+		if (vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 0), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 0), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 0), reflection_light);
+		}
+
+		if (vrs_y > 1) {
+			imageStore(ambient_buffer, pos + ivec2(0, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 1), reflection_light);
+		}
+
+		if (vrs_y > 1 && vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 1), reflection_light);
+		}
+
+		if (vrs_y > 1 && vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 1), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 1), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 1), reflection_light);
+		}
+
+		if (vrs_y > 2) {
+			imageStore(ambient_buffer, pos + ivec2(0, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(0, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(0, 3), reflection_light);
+		}
+
+		if (vrs_y > 2 && vrs_x > 1) {
+			imageStore(ambient_buffer, pos + ivec2(1, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(1, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(1, 3), reflection_light);
+		}
+
+		if (vrs_y > 2 && vrs_x > 2) {
+			imageStore(ambient_buffer, pos + ivec2(2, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(2, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(2, 3), reflection_light);
+
+			imageStore(ambient_buffer, pos + ivec2(3, 2), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 2), reflection_light);
+			imageStore(ambient_buffer, pos + ivec2(3, 3), ambient_light);
+			imageStore(reflection_buffer, pos + ivec2(3, 3), reflection_light);
+		}
+	}
 }
diff --git a/servers/rendering/renderer_rd/storage_rd/texture_storage.cpp b/servers/rendering/renderer_rd/storage_rd/texture_storage.cpp
index 1109357a74..762ad685e8 100644
--- a/servers/rendering/renderer_rd/storage_rd/texture_storage.cpp
+++ b/servers/rendering/renderer_rd/storage_rd/texture_storage.cpp
@@ -349,7 +349,6 @@ TextureStorage::TextureStorage() {
 
 		Vector<uint8_t> pv;
 		pv.resize(16 * 4);
-
 		for (int i = 0; i < 16; i++) {
 			pv.set(i * 4 + 0, 0);
 			pv.set(i * 4 + 1, 0);
@@ -358,7 +357,6 @@ TextureStorage::TextureStorage() {
 		}
 
 		{
-			//take the chance and initialize decal atlas to something
 			Vector<Vector<uint8_t>> vpv;
 			vpv.push_back(pv);
 			decal_atlas.texture = RD::get_singleton()->texture_create(tformat, RD::TextureView(), vpv);
@@ -366,6 +364,29 @@ TextureStorage::TextureStorage() {
 		}
 	}
 
+	{ //create default VRS
+
+		RD::TextureFormat tformat;
+		tformat.format = RD::DATA_FORMAT_R8_UINT;
+		tformat.width = 4;
+		tformat.height = 4;
+		tformat.array_layers = 1;
+		tformat.usage_bits = RD::TEXTURE_USAGE_COLOR_ATTACHMENT_BIT | RD::TEXTURE_USAGE_VRS_ATTACHMENT_BIT | RD::TEXTURE_USAGE_SAMPLING_BIT | RD::TEXTURE_USAGE_STORAGE_BIT | RD::TEXTURE_USAGE_CAN_UPDATE_BIT;
+		tformat.texture_type = RD::TEXTURE_TYPE_2D_ARRAY;
+
+		Vector<uint8_t> pv;
+		pv.resize(4 * 4);
+		for (int i = 0; i < 4 * 4; i++) {
+			pv.set(i, 0);
+		}
+
+		{
+			Vector<Vector<uint8_t>> vpv;
+			vpv.push_back(pv);
+			default_rd_textures[DEFAULT_RD_TEXTURE_VRS] = RD::get_singleton()->texture_create(tformat, RD::TextureView(), vpv);
+		}
+	}
+
 	{
 		Vector<String> sdf_modes;
 		sdf_modes.push_back("\n#define MODE_LOAD\n");
@@ -2751,3 +2772,31 @@ void TextureStorage::render_target_set_backbuffer_uniform_set(RID p_render_targe
 	ERR_FAIL_COND(!rt);
 	rt->backbuffer_uniform_set = p_uniform_set;
 }
+
+void TextureStorage::render_target_set_vrs_mode(RID p_render_target, RS::ViewportVRSMode p_mode) {
+	RenderTarget *rt = render_target_owner.get_or_null(p_render_target);
+	ERR_FAIL_COND(!rt);
+
+	rt->vrs_mode = p_mode;
+}
+
+void TextureStorage::render_target_set_vrs_texture(RID p_render_target, RID p_texture) {
+	RenderTarget *rt = render_target_owner.get_or_null(p_render_target);
+	ERR_FAIL_COND(!rt);
+
+	rt->vrs_texture = p_texture;
+}
+
+RS::ViewportVRSMode TextureStorage::render_target_get_vrs_mode(RID p_render_target) const {
+	RenderTarget *rt = render_target_owner.get_or_null(p_render_target);
+	ERR_FAIL_COND_V(!rt, RS::VIEWPORT_VRS_DISABLED);
+
+	return rt->vrs_mode;
+}
+
+RID TextureStorage::render_target_get_vrs_texture(RID p_render_target) const {
+	RenderTarget *rt = render_target_owner.get_or_null(p_render_target);
+	ERR_FAIL_COND_V(!rt, RID());
+
+	return rt->vrs_texture;
+}
diff --git a/servers/rendering/renderer_rd/storage_rd/texture_storage.h b/servers/rendering/renderer_rd/storage_rd/texture_storage.h
index 7a96e6c6ed..ac95e13604 100644
--- a/servers/rendering/renderer_rd/storage_rd/texture_storage.h
+++ b/servers/rendering/renderer_rd/storage_rd/texture_storage.h
@@ -52,6 +52,7 @@ enum DefaultRDTexture {
 	DEFAULT_RD_TEXTURE_3D_BLACK,
 	DEFAULT_RD_TEXTURE_2D_ARRAY_WHITE,
 	DEFAULT_RD_TEXTURE_2D_UINT,
+	DEFAULT_RD_TEXTURE_VRS,
 	DEFAULT_RD_TEXTURE_MAX
 };
 
@@ -229,6 +230,10 @@ struct RenderTarget {
 	RS::ViewportSDFScale sdf_scale = RS::VIEWPORT_SDF_SCALE_50_PERCENT;
 	Size2i process_size;
 
+	// VRS
+	RS::ViewportVRSMode vrs_mode = RS::VIEWPORT_VRS_DISABLED;
+	RID vrs_texture;
+
 	//texture generated for this owner (nor RD).
 	RID texture;
 	bool was_used;
@@ -549,6 +554,12 @@ public:
 	virtual void render_target_mark_sdf_enabled(RID p_render_target, bool p_enabled) override;
 	bool render_target_is_sdf_enabled(RID p_render_target) const;
 
+	virtual void render_target_set_vrs_mode(RID p_render_target, RS::ViewportVRSMode p_mode) override;
+	virtual void render_target_set_vrs_texture(RID p_render_target, RID p_texture) override;
+
+	RS::ViewportVRSMode render_target_get_vrs_mode(RID p_render_target) const;
+	RID render_target_get_vrs_texture(RID p_render_target) const;
+
 	Size2 render_target_get_size(RID p_render_target);
 	RID render_target_get_rd_framebuffer(RID p_render_target);
 	RID render_target_get_rd_texture(RID p_render_target);
diff --git a/servers/rendering/renderer_viewport.cpp b/servers/rendering/renderer_viewport.cpp
index 303efe50f7..7c9b2567d6 100644
--- a/servers/rendering/renderer_viewport.cpp
+++ b/servers/rendering/renderer_viewport.cpp
@@ -1207,6 +1207,22 @@ RID RendererViewport::viewport_find_from_screen_attachment(DisplayServer::Window
 	return RID();
 }
 
+void RendererViewport::viewport_set_vrs_mode(RID p_viewport, RS::ViewportVRSMode p_mode) {
+	Viewport *viewport = viewport_owner.get_or_null(p_viewport);
+	ERR_FAIL_COND(!viewport);
+
+	RSG::texture_storage->render_target_set_vrs_mode(viewport->render_target, p_mode);
+	_configure_3d_render_buffers(viewport);
+}
+
+void RendererViewport::viewport_set_vrs_texture(RID p_viewport, RID p_texture) {
+	Viewport *viewport = viewport_owner.get_or_null(p_viewport);
+	ERR_FAIL_COND(!viewport);
+
+	RSG::texture_storage->render_target_set_vrs_texture(viewport->render_target, p_texture);
+	_configure_3d_render_buffers(viewport);
+}
+
 bool RendererViewport::free(RID p_rid) {
 	if (viewport_owner.owns(p_rid)) {
 		Viewport *viewport = viewport_owner.get_or_null(p_rid);
diff --git a/servers/rendering/renderer_viewport.h b/servers/rendering/renderer_viewport.h
index 49ee9a6224..027f2dfad6 100644
--- a/servers/rendering/renderer_viewport.h
+++ b/servers/rendering/renderer_viewport.h
@@ -284,6 +284,9 @@ public:
 
 	virtual RID viewport_find_from_screen_attachment(DisplayServer::WindowID p_id = DisplayServer::MAIN_WINDOW_ID) const;
 
+	void viewport_set_vrs_mode(RID p_viewport, RS::ViewportVRSMode p_mode);
+	void viewport_set_vrs_texture(RID p_viewport, RID p_texture);
+
 	void handle_timestamp(String p_timestamp, uint64_t p_cpu_time, uint64_t p_gpu_time);
 
 	void set_default_clear_color(const Color &p_color);
diff --git a/servers/rendering/rendering_device.cpp b/servers/rendering/rendering_device.cpp
index 6fc5d0b3e8..0b76bb3051 100644
--- a/servers/rendering/rendering_device.cpp
+++ b/servers/rendering/rendering_device.cpp
@@ -64,12 +64,12 @@ Vector<uint8_t> RenderingDevice::shader_compile_spirv_from_source(ShaderStage p_
 
 	ERR_FAIL_COND_V(!compile_to_spirv_function, Vector<uint8_t>());
 
-	return compile_to_spirv_function(p_stage, p_source_code, p_language, r_error, &device_capabilities);
+	return compile_to_spirv_function(p_stage, p_source_code, p_language, r_error, this);
 }
 
 String RenderingDevice::shader_get_spirv_cache_key() const {
 	if (get_spirv_cache_key_function) {
-		return get_spirv_cache_key_function(&device_capabilities);
+		return get_spirv_cache_key_function(this);
 	}
 	return String();
 }
@@ -279,6 +279,7 @@ static Vector<RenderingDevice::PipelineSpecializationConstant> _get_spec_constan
 	}
 	return ret;
 }
+
 RID RenderingDevice::_render_pipeline_create(RID p_shader, FramebufferFormatID p_framebuffer_format, VertexFormatID p_vertex_format, RenderPrimitive p_render_primitive, const Ref<RDPipelineRasterizationState> &p_rasterization_state, const Ref<RDPipelineMultisampleState> &p_multisample_state, const Ref<RDPipelineDepthStencilState> &p_depth_stencil_state, const Ref<RDPipelineColorBlendState> &p_blend_state, int p_dynamic_state_flags, uint32_t p_for_render_pass, const TypedArray<RDPipelineSpecializationConstant> &p_specialization_constants) {
 	PipelineRasterizationState rasterization_state;
 	if (p_rasterization_state.is_valid()) {
diff --git a/servers/rendering/rendering_device.h b/servers/rendering/rendering_device.h
index 0973e29974..03aa6f7644 100644
--- a/servers/rendering/rendering_device.h
+++ b/servers/rendering/rendering_device.h
@@ -123,19 +123,10 @@ public:
 		DeviceFamily device_family = DEVICE_UNKNOWN;
 		uint32_t version_major = 1.0;
 		uint32_t version_minor = 0.0;
-
-		// subgroup capabilities
-		uint32_t subgroup_size = 0;
-		uint32_t subgroup_in_shaders = 0; // Set flags using SHADER_STAGE_VERTEX_BIT, SHADER_STAGE_FRAGMENT_BIT, etc.
-		uint32_t subgroup_operations = 0; // Set flags, using SubgroupOperations
-
-		// features
-		bool supports_multiview = false; // If true this device supports multiview options
-		bool supports_fsr_half_float = false; // If true this device supports FSR scaling 3D in half float mode, otherwise use the fallback mode
 	};
 
-	typedef String (*ShaderSPIRVGetCacheKeyFunction)(const Capabilities *p_capabilities);
-	typedef Vector<uint8_t> (*ShaderCompileToSPIRVFunction)(ShaderStage p_stage, const String &p_source_code, ShaderLanguage p_language, String *r_error, const Capabilities *p_capabilities);
+	typedef String (*ShaderSPIRVGetCacheKeyFunction)(const RenderingDevice *p_render_device);
+	typedef Vector<uint8_t> (*ShaderCompileToSPIRVFunction)(ShaderStage p_stage, const String &p_source_code, ShaderLanguage p_language, String *r_error, const RenderingDevice *p_render_device);
 	typedef Vector<uint8_t> (*ShaderCacheFunction)(ShaderStage p_stage, const String &p_source_code, ShaderLanguage p_language);
 
 private:
@@ -444,6 +435,7 @@ public:
 		TEXTURE_USAGE_CAN_COPY_FROM_BIT = (1 << 7),
 		TEXTURE_USAGE_CAN_COPY_TO_BIT = (1 << 8),
 		TEXTURE_USAGE_INPUT_ATTACHMENT_BIT = (1 << 9),
+		TEXTURE_USAGE_VRS_ATTACHMENT_BIT = (1 << 10),
 	};
 
 	enum TextureSwizzle {
@@ -552,6 +544,7 @@ public:
 		Vector<int32_t> resolve_attachments;
 		Vector<int32_t> preserve_attachments;
 		int32_t depth_attachment = ATTACHMENT_UNUSED;
+		int32_t vrs_attachment = ATTACHMENT_UNUSED; // density map for VRS, only used if supported
 	};
 
 	virtual FramebufferFormatID framebuffer_format_create_multipass(const Vector<AttachmentFormat> &p_attachments, Vector<FramebufferPass> &p_passes, uint32_t p_view_count = 1) = 0;
@@ -675,6 +668,13 @@ public:
 
 	const Capabilities *get_device_capabilities() const { return &device_capabilities; };
 
+	enum Features {
+		SUPPORTS_MULTIVIEW,
+		SUPPORTS_FSR_HALF_FLOAT,
+		SUPPORTS_ATTACHMENT_VRS,
+	};
+	virtual bool has_feature(const Features p_feature) const = 0;
+
 	virtual Vector<uint8_t> shader_compile_spirv_from_source(ShaderStage p_stage, const String &p_source_code, ShaderLanguage p_language = SHADER_LANGUAGE_GLSL, String *r_error = nullptr, bool p_allow_cache = true);
 	virtual String shader_get_spirv_cache_key() const;
 
@@ -1221,9 +1221,12 @@ public:
 		LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_X,
 		LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_Y,
 		LIMIT_MAX_COMPUTE_WORKGROUP_SIZE_Z,
+		LIMIT_SUBGROUP_SIZE,
+		LIMIT_SUBGROUP_IN_SHADERS, // Set flags using SHADER_STAGE_VERTEX_BIT, SHADER_STAGE_FRAGMENT_BIT, etc.
+		LIMIT_SUBGROUP_OPERATIONS,
 	};
 
-	virtual uint64_t limit_get(Limit p_limit) = 0;
+	virtual uint64_t limit_get(Limit p_limit) const = 0;
 
 	//methods below not exposed, used by RenderingDeviceRD
 	virtual void prepare_screen_for_drawing() = 0;
@@ -1324,6 +1327,7 @@ VARIANT_ENUM_CAST(RenderingDevice::InitialAction)
 VARIANT_ENUM_CAST(RenderingDevice::FinalAction)
 VARIANT_ENUM_CAST(RenderingDevice::Limit)
 VARIANT_ENUM_CAST(RenderingDevice::MemoryType)
+VARIANT_ENUM_CAST(RenderingDevice::Features)
 
 typedef RenderingDevice RD;
 
diff --git a/servers/rendering/rendering_server_default.h b/servers/rendering/rendering_server_default.h
index 9dfd8ffb94..429b8a06e2 100644
--- a/servers/rendering/rendering_server_default.h
+++ b/servers/rendering/rendering_server_default.h
@@ -637,6 +637,9 @@ public:
 
 	FUNC2(call_set_vsync_mode, DisplayServer::VSyncMode, DisplayServer::WindowID)
 
+	FUNC2(viewport_set_vrs_mode, RID, ViewportVRSMode)
+	FUNC2(viewport_set_vrs_texture, RID, RID)
+
 	/* ENVIRONMENT API */
 
 #undef server_name
diff --git a/servers/rendering/storage/texture_storage.h b/servers/rendering/storage/texture_storage.h
index e90a028713..92238c19ee 100644
--- a/servers/rendering/storage/texture_storage.h
+++ b/servers/rendering/storage/texture_storage.h
@@ -143,6 +143,9 @@ public:
 	virtual void render_target_set_sdf_size_and_scale(RID p_render_target, RS::ViewportSDFOversize p_size, RS::ViewportSDFScale p_scale) = 0;
 	virtual Rect2i render_target_get_sdf_rect(RID p_render_target) const = 0;
 	virtual void render_target_mark_sdf_enabled(RID p_render_target, bool p_enabled) = 0;
+
+	virtual void render_target_set_vrs_mode(RID p_render_target, RS::ViewportVRSMode p_mode) = 0;
+	virtual void render_target_set_vrs_texture(RID p_render_target, RID p_texture) = 0;
 };
 
 #endif // !TEXTURE_STORAGE_H