Merge pull request #45722 from reduz/intel-fixed

Fixes to get Godot running again on Intel IGP

4 files changed, 25 insertions, 16 deletions

diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp
index ba80ca3514..9584dd3f67 100644
--- a/drivers/vulkan/rendering_device_vulkan.cpp
+++ b/drivers/vulkan/rendering_device_vulkan.cpp
@@ -74,11 +74,13 @@ RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID
 	} else if (texture_buffer_owner.owns(p_buffer)) {
 		if (p_post_barrier & BARRIER_MASK_RASTER) {
 			r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT;
+			r_access_mask |= VK_ACCESS_SHADER_READ_BIT;
 		}
 		if (p_post_barrier & BARRIER_MASK_COMPUTE) {
 			r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+			r_access_mask |= VK_ACCESS_SHADER_READ_BIT;
 		}
-		r_access_mask |= VK_ACCESS_SHADER_READ_BIT;
+
 		buffer = &texture_buffer_owner.getornull(p_buffer)->buffer;
 	} else if (storage_buffer_owner.owns(p_buffer)) {
 		buffer = storage_buffer_owner.getornull(p_buffer);
@@ -5271,7 +5273,7 @@ Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint
 	}
 
 	if (p_post_barrier != RD::BARRIER_MASK_NO_BARRIER) {
-		_buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, dst_stage_mask);
+		_buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, true);
 	}
 
 #endif
@@ -7324,6 +7326,10 @@ void RenderingDeviceVulkan::barrier(uint32_t p_from, uint32_t p_to) {
 		src_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT;
 	}
 
+	if (p_from == 0) {
+		src_barrier_flags = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
+	}
+
 	uint32_t dst_barrier_flags = 0;
 	uint32_t dst_access_flags = 0;
 	if (p_to & BARRIER_MASK_COMPUTE) {
@@ -7339,6 +7345,10 @@ void RenderingDeviceVulkan::barrier(uint32_t p_from, uint32_t p_to) {
 		dst_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT;
 	}
 
+	if (p_to == 0) {
+		dst_barrier_flags = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
+	}
+
 	_memory_barrier(src_barrier_flags, dst_barrier_flags, src_access_flags, dst_access_flags, true);
 }
 
diff --git a/servers/rendering/renderer_rd/cluster_builder_rd.cpp b/servers/rendering/renderer_rd/cluster_builder_rd.cpp
index e052a2b0ce..0fdd864d47 100644
--- a/servers/rendering/renderer_rd/cluster_builder_rd.cpp
+++ b/servers/rendering/renderer_rd/cluster_builder_rd.cpp
@@ -403,11 +403,11 @@ void ClusterBuilderRD::bake_cluster() {
 	RD::get_singleton()->draw_command_begin_label("Bake Light Cluster");
 
 	//clear cluster buffer
-	RD::get_singleton()->buffer_clear(cluster_buffer, 0, cluster_buffer_size, 0);
+	RD::get_singleton()->buffer_clear(cluster_buffer, 0, cluster_buffer_size, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 
 	if (render_element_count > 0) {
 		//clear render buffer
-		RD::get_singleton()->buffer_clear(cluster_render_buffer, 0, cluster_render_buffer_size, 0);
+		RD::get_singleton()->buffer_clear(cluster_render_buffer, 0, cluster_render_buffer_size, RD::BARRIER_MASK_RASTER);
 
 		{ //fill state uniform
 
@@ -422,16 +422,15 @@ void ClusterBuilderRD::bake_cluster() {
 			state.cluster_depth_offset = (render_element_max / 32);
 			state.cluster_data_size = state.cluster_depth_offset + render_element_max;
 
-			RD::get_singleton()->buffer_update(state_uniform, 0, sizeof(StateUniform), &state, 0);
+			RD::get_singleton()->buffer_update(state_uniform, 0, sizeof(StateUniform), &state, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 		}
 
 		//update instances
 
-		RD::get_singleton()->buffer_update(element_buffer, 0, sizeof(RenderElementData) * render_element_count, render_elements, 0);
+		RD::get_singleton()->buffer_update(element_buffer, 0, sizeof(RenderElementData) * render_element_count, render_elements, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 
 		RENDER_TIMESTAMP("Render Elements");
 
-		RD::get_singleton()->barrier(RD::BARRIER_MASK_TRANSFER, RD::BARRIER_MASK_RASTER);
 		//render elements
 		{
 			RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(framebuffer, RD::INITIAL_ACTION_DROP, RD::FINAL_ACTION_DISCARD, RD::INITIAL_ACTION_DROP, RD::FINAL_ACTION_DISCARD);
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
index f4afa8fb12..3035124cca 100644
--- a/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
+++ b/servers/rendering/renderer_rd/renderer_scene_render_rd.cpp
@@ -6226,7 +6226,7 @@ void RendererSceneRenderRD::_setup_reflections(const PagedArray<RID> &p_reflecti
 	}
 
 	if (cluster.reflection_count) {
-		RD::get_singleton()->buffer_update(cluster.reflection_buffer, 0, cluster.reflection_count * sizeof(ReflectionData), cluster.reflections, 0); //no barriers for now
+		RD::get_singleton()->buffer_update(cluster.reflection_buffer, 0, cluster.reflection_count * sizeof(ReflectionData), cluster.reflections, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 	}
 }
 
@@ -6620,15 +6620,15 @@ void RendererSceneRenderRD::_setup_lights(const PagedArray<RID> &p_lights, const
 
 	//update without barriers
 	if (cluster.omni_light_count) {
-		RD::get_singleton()->buffer_update(cluster.omni_light_buffer, 0, sizeof(Cluster::LightData) * cluster.omni_light_count, cluster.omni_lights, 0);
+		RD::get_singleton()->buffer_update(cluster.omni_light_buffer, 0, sizeof(Cluster::LightData) * cluster.omni_light_count, cluster.omni_lights, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 	}
 
 	if (cluster.spot_light_count) {
-		RD::get_singleton()->buffer_update(cluster.spot_light_buffer, 0, sizeof(Cluster::LightData) * cluster.spot_light_count, cluster.spot_lights, 0);
+		RD::get_singleton()->buffer_update(cluster.spot_light_buffer, 0, sizeof(Cluster::LightData) * cluster.spot_light_count, cluster.spot_lights, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 	}
 
 	if (r_directional_light_count) {
-		RD::get_singleton()->buffer_update(cluster.directional_light_buffer, 0, sizeof(Cluster::DirectionalLightData) * r_directional_light_count, cluster.directional_lights, 0);
+		RD::get_singleton()->buffer_update(cluster.directional_light_buffer, 0, sizeof(Cluster::DirectionalLightData) * r_directional_light_count, cluster.directional_lights, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 	}
 }
 
@@ -6789,7 +6789,7 @@ void RendererSceneRenderRD::_setup_decals(const PagedArray<RID> &p_decals, const
 	}
 
 	if (cluster.decal_count > 0) {
-		RD::get_singleton()->buffer_update(cluster.decal_buffer, 0, sizeof(Cluster::DecalData) * cluster.decal_count, cluster.decals);
+		RD::get_singleton()->buffer_update(cluster.decal_buffer, 0, sizeof(Cluster::DecalData) * cluster.decal_count, cluster.decals, RD::BARRIER_MASK_RASTER | RD::BARRIER_MASK_COMPUTE);
 	}
 }
 
@@ -7513,6 +7513,9 @@ void RendererSceneRenderRD::_pre_opaque_render(bool p_use_ssao, bool p_use_gi, R
 		}
 	}
 
+	//full barrier here, we need raster, transfer and compute and it depends from the previous work
+	RD::get_singleton()->barrier(RD::BARRIER_MASK_ALL, RD::BARRIER_MASK_ALL);
+
 	if (current_cluster_builder) {
 		current_cluster_builder->begin(render_state.cam_transform, render_state.cam_projection, !render_state.reflection_probe.is_valid());
 	}
@@ -7535,9 +7538,6 @@ void RendererSceneRenderRD::_pre_opaque_render(bool p_use_ssao, bool p_use_gi, R
 
 	render_state.directional_light_count = directional_light_count;
 
-	//full barrier here, we need raster, transfer and compute and it depends from the previous work
-	RD::get_singleton()->barrier(RD::BARRIER_MASK_ALL, RD::BARRIER_MASK_ALL);
-
 	if (current_cluster_builder) {
 		current_cluster_builder->bake_cluster();
 	}
diff --git a/servers/rendering/renderer_rd/shaders/scene_forward.glsl b/servers/rendering/renderer_rd/shaders/scene_forward.glsl
index 4286ef462c..83e2ae9cbd 100644
--- a/servers/rendering/renderer_rd/shaders/scene_forward.glsl
+++ b/servers/rendering/renderer_rd/shaders/scene_forward.glsl
@@ -111,7 +111,7 @@ void main() {
 	color_interp = color_attrib;
 #endif
 
-	uint instance_index = draw_call.instance_index;
+	instance_index = draw_call.instance_index;
 
 	bool is_multimesh = bool(instances.data[instance_index].flags & INSTANCE_FLAGS_MULTIMESH);
 	if (!is_multimesh) {