Revert "Revert "Synchronization validation fix patch set 3 (inclusive of all previous patches)""

2 files changed, 149 insertions, 67 deletions

diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp
index 9fe3f7e6c0..a01f63caae 100644
--- a/drivers/vulkan/rendering_device_vulkan.cpp
+++ b/drivers/vulkan/rendering_device_vulkan.cpp
@@ -38,7 +38,58 @@
 
 #include "thirdparty/spirv-reflect/spirv_reflect.h"
 
-#define FORCE_FULL_BARRIER
+//#define FORCE_FULL_BARRIER
+
+// Get the Vulkan object information and possible stage access types (bitwise OR'd with incoming values)
+RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &stage_mask, VkAccessFlags &access_mask) {
+	Buffer *buffer = nullptr;
+	if (vertex_buffer_owner.owns(p_buffer)) {
+		stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
+		access_mask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
+		buffer = vertex_buffer_owner.getornull(p_buffer);
+	} else if (index_buffer_owner.owns(p_buffer)) {
+		stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
+		access_mask |= VK_ACCESS_INDEX_READ_BIT;
+		buffer = index_buffer_owner.getornull(p_buffer);
+	} else if (uniform_buffer_owner.owns(p_buffer)) {
+		stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+		access_mask |= VK_ACCESS_UNIFORM_READ_BIT;
+		buffer = uniform_buffer_owner.getornull(p_buffer);
+	} else if (texture_buffer_owner.owns(p_buffer)) {
+		stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+		access_mask |= VK_ACCESS_SHADER_READ_BIT;
+		buffer = &texture_buffer_owner.getornull(p_buffer)->buffer;
+	} else if (storage_buffer_owner.owns(p_buffer)) {
+		stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+		access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
+		buffer = storage_buffer_owner.getornull(p_buffer);
+	}
+	return buffer;
+}
+
+static void update_external_dependency_for_store(VkSubpassDependency &dependency, bool is_sampled, bool is_storage, bool is_depth) {
+	// Transitioning from write to read, protect the shaders that may use this next
+	// Allow for copies/image layout transitions
+	dependency.dstStageMask |= VK_PIPELINE_STAGE_TRANSFER_BIT;
+	dependency.dstAccessMask |= VK_ACCESS_TRANSFER_READ_BIT;
+
+	if (is_sampled) {
+		dependency.dstStageMask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+		dependency.dstAccessMask |= VK_ACCESS_SHADER_READ_BIT;
+	} else if (is_storage) {
+		dependency.dstStageMask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+		dependency.dstAccessMask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
+	} else {
+		dependency.dstStageMask |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT;
+		dependency.dstAccessMask |= VK_ACCESS_COLOR_ATTACHMENT_READ_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
+	}
+
+	if (is_depth) {
+		// Depth resources have addtional stages that may be interested in them
+		dependency.dstStageMask |= VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT;
+		dependency.dstAccessMask |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT | VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
+	}
+}
 
 void RenderingDeviceVulkan::_add_dependency(RID p_id, RID p_depends_on) {
 	if (!dependency_map.has(p_depends_on)) {
@@ -1932,7 +1983,7 @@ RID RenderingDeviceVulkan::texture_create(const TextureFormat &p_format, const T
 		image_memory_barrier.subresourceRange.baseArrayLayer = 0;
 		image_memory_barrier.subresourceRange.layerCount = image_create_info.arrayLayers;
 
-		vkCmdPipelineBarrier(frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 	}
 
 	RID id = texture_owner.make_rid(texture);
@@ -2189,7 +2240,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con
 		image_memory_barrier.subresourceRange.baseArrayLayer = p_layer;
 		image_memory_barrier.subresourceRange.layerCount = 1;
 
-		vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 	}
 
 	uint32_t mipmap_offset = 0;
@@ -2322,7 +2373,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con
 		image_memory_barrier.subresourceRange.baseArrayLayer = p_layer;
 		image_memory_barrier.subresourceRange.layerCount = 1;
 
-		vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 	}
 
 	return OK;
@@ -2485,6 +2536,9 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t
 			image_memory_barrier.pNext = nullptr;
 			image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
 			image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
+			if (tex->usage_flags & TEXTURE_USAGE_STORAGE_BIT) {
+				image_memory_barrier.dstAccessMask |= VK_ACCESS_SHADER_WRITE_BIT;
+			}
 			image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
 			image_memory_barrier.newLayout = tex->layout;
 			image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -2496,7 +2550,7 @@ Vector<uint8_t> RenderingDeviceVulkan::texture_get_data(RID p_texture, uint32_t
 			image_memory_barrier.subresourceRange.baseArrayLayer = p_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 
 		_flush(true);
@@ -2672,7 +2726,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture,
 			image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 
 		{ //make dst readable
@@ -2694,7 +2748,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture,
 			image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 	}
 
@@ -2755,7 +2809,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID
 			image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 		{ //Dest
 			VkImageMemoryBarrier image_memory_barrier;
@@ -2824,7 +2878,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID
 			image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 
 		{ //make dst readable
@@ -2846,7 +2900,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID
 			image_memory_barrier.subresourceRange.baseArrayLayer = dst_tex->base_layer;
 			image_memory_barrier.subresourceRange.layerCount = 1;
 
-			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 		}
 	}
 
@@ -2878,16 +2932,22 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color,
 
 	VkCommandBuffer command_buffer = p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer;
 
-	VkImageLayout layout = src_tex->layout;
+	VkImageLayout clear_layout = (src_tex->layout == VK_IMAGE_LAYOUT_GENERAL) ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
 
-	if (src_tex->layout != VK_IMAGE_LAYOUT_GENERAL) { //storage may be in general state
+	// NOTE: Perhaps the valid stages/accesses for a given onwner should be a property of the owner. (Here and places like _get_buffer_from_owner)
+	const VkPipelineStageFlags valid_texture_stages = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
+	constexpr VkAccessFlags read_access = VK_ACCESS_SHADER_READ_BIT;
+	constexpr VkAccessFlags read_write_access = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
+	const VkAccessFlags valid_texture_access = (src_tex->usage_flags & TEXTURE_USAGE_STORAGE_BIT) ? read_write_access : read_access;
+
+	{ // Barrier from previous access with optional layout change (see clear_layout logic above)
 		VkImageMemoryBarrier image_memory_barrier;
 		image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
 		image_memory_barrier.pNext = nullptr;
-		image_memory_barrier.srcAccessMask = 0;
+		image_memory_barrier.srcAccessMask = valid_texture_access;
 		image_memory_barrier.dstAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
 		image_memory_barrier.oldLayout = src_tex->layout;
-		image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
+		image_memory_barrier.oldLayout = clear_layout;
 
 		image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
 		image_memory_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -2898,8 +2958,7 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color,
 		image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer + p_base_layer;
 		image_memory_barrier.subresourceRange.layerCount = p_layers;
 
-		layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
-		vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(command_buffer, valid_texture_stages, VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 	}
 
 	VkClearColorValue clear_color;
@@ -2915,16 +2974,15 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color,
 	range.baseMipLevel = src_tex->base_mipmap + p_base_mipmap;
 	range.levelCount = p_mipmaps;
 
-	vkCmdClearColorImage(command_buffer, src_tex->image, layout, &clear_color, 1, &range);
-
-	if (src_tex->layout != VK_IMAGE_LAYOUT_GENERAL) { //storage may be in general state
+	vkCmdClearColorImage(command_buffer, src_tex->image, clear_layout, &clear_color, 1, &range);
 
+	{ // Barrier to post clear accesses (changing back the layout if needed)
 		VkImageMemoryBarrier image_memory_barrier;
 		image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
 		image_memory_barrier.pNext = nullptr;
 		image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
-		image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
-		image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
+		image_memory_barrier.dstAccessMask = valid_texture_access;
+		image_memory_barrier.oldLayout = clear_layout;
 		image_memory_barrier.newLayout = src_tex->layout;
 
 		image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED;
@@ -2936,7 +2994,7 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color,
 		image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer + p_base_layer;
 		image_memory_barrier.subresourceRange.layerCount = p_layers;
 
-		vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, valid_texture_stages, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 	}
 
 	return OK;
@@ -2991,6 +3049,19 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 	Vector<VkAttachmentReference> depth_stencil_references;
 	Vector<VkAttachmentReference> resolve_references;
 
+	// Set up a dependencies from/to external equivalent to the default (implicit) one, and then amend them
+	const VkPipelineStageFlags default_access_mask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT |
+													 VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
+													 VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+													 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
+													 VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT; // From Section 7.1 of Vulkan API Spec v1.1.148
+
+	VkPipelineStageFlags reading_stages = VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT;
+	VkSubpassDependency dependencies[2] = { { VK_SUBPASS_EXTERNAL, 0, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, 0, default_access_mask, 0 },
+		{ 0, VK_SUBPASS_EXTERNAL, VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, default_access_mask, 0, 0 } };
+	VkSubpassDependency &dependency_from_external = dependencies[0];
+	VkSubpassDependency &dependency_to_external = dependencies[1];
+
 	for (int i = 0; i < p_format.size(); i++) {
 		ERR_FAIL_INDEX_V(p_format[i].format, DATA_FORMAT_MAX, VK_NULL_HANDLE);
 		ERR_FAIL_INDEX_V(p_format[i].samples, TEXTURE_SAMPLES_MAX, VK_NULL_HANDLE);
@@ -3006,11 +3077,16 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 		bool is_sampled = p_format[i].usage_flags & TEXTURE_USAGE_SAMPLING_BIT;
 		bool is_storage = p_format[i].usage_flags & TEXTURE_USAGE_STORAGE_BIT;
 
+		// For each UNDEFINED, assume the prior use was a *read*, as we'd be discarding the output of a write
+		// Also, each UNDEFINED will do an immediate layout transition (write), s.t. we must ensure execution syncronization vs.
+		// the read.  If this is a performance issue, one could track the actual last accessor of each resource, adding only that
+		// stage
 		switch (is_depth_stencil ? p_initial_depth_action : p_initial_color_action) {
 			case INITIAL_ACTION_CLEAR: {
 				description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
 				description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
 				description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
+				dependency_from_external.srcStageMask |= reading_stages;
 			} break;
 			case INITIAL_ACTION_KEEP: {
 				if (p_format[i].usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) {
@@ -3021,10 +3097,12 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_CLEAR;
+					dependency_from_external.srcStageMask |= reading_stages;
 				} else {
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
+					dependency_from_external.srcStageMask |= reading_stages;
 				}
 			} break;
 			case INITIAL_ACTION_DROP: {
@@ -3036,10 +3114,12 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
+					dependency_from_external.srcStageMask |= reading_stages;
 				} else {
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
+					dependency_from_external.srcStageMask |= reading_stages;
 				}
 			} break;
 			case INITIAL_ACTION_CONTINUE: {
@@ -3055,6 +3135,7 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
+					dependency_from_external.srcStageMask |= reading_stages;
 				}
 			} break;
 			default: {
@@ -3068,14 +3149,17 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 					description.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
 					description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
 					description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL);
+					update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, false);
 				} else if (p_format[i].usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) {
 					description.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
 					description.stencilStoreOp = VK_ATTACHMENT_STORE_OP_STORE;
 					description.finalLayout = is_sampled ? VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL : (is_storage ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL);
+					update_external_dependency_for_store(dependency_to_external, is_sampled, is_storage, true);
 				} else {
 					description.loadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
 					description.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //don't care what is there
+					// TODO: What does this mean about the next usage (and thus appropriate dependency masks
 				}
 			} break;
 			case FINAL_ACTION_DISCARD: {
@@ -3128,9 +3212,24 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 		} else if (p_format[i].usage_flags & TEXTURE_USAGE_RESOLVE_ATTACHMENT_BIT) {
 			reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
 			resolve_references.push_back(reference);
+			// if resolves are done, we need to ensure the copy is safe
+			dependency_to_external.dstStageMask |= VK_PIPELINE_STAGE_TRANSFER_BIT;
+			dependency_to_external.dstAccessMask |= VK_ACCESS_TRANSFER_READ_BIT;
 		} else {
 			ERR_FAIL_V_MSG(VK_NULL_HANDLE, "Texture index " + itos(i) + " is neither color, depth stencil or resolve so it can't be used as attachment.");
 		}
+
+		// NOTE: Big Mallet Approach -- any layout transition causes a full barrier
+		if (reference.layout != description.initialLayout) {
+			// NOTE: this should be smarter based on the textures knowledge of it's previous role
+			dependency_from_external.srcStageMask |= VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
+			dependency_from_external.srcAccessMask |= VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
+		}
+		if (reference.layout != description.finalLayout) {
+			// NOTE: this should be smarter based on the textures knowledge of it's subsequent role
+			dependency_from_external.dstStageMask |= VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
+			dependency_from_external.dstAccessMask |= VK_ACCESS_MEMORY_READ_BIT | VK_ACCESS_MEMORY_WRITE_BIT;
+		}
 	}
 
 	ERR_FAIL_COND_V_MSG(depth_stencil_references.size() > 1, VK_NULL_HANDLE,
@@ -3159,8 +3258,8 @@ VkRenderPass RenderingDeviceVulkan::_render_pass_create(const Vector<AttachmentF
 	render_pass_create_info.pAttachments = attachments.ptr();
 	render_pass_create_info.subpassCount = 1;
 	render_pass_create_info.pSubpasses = &subpass;
-	render_pass_create_info.dependencyCount = 0;
-	render_pass_create_info.pDependencies = nullptr;
+	render_pass_create_info.dependencyCount = 2;
+	render_pass_create_info.pDependencies = dependencies;
 
 	VkRenderPass render_pass;
 	VkResult res = vkCreateRenderPass(device, &render_pass_create_info, nullptr, &render_pass);
@@ -3185,7 +3284,7 @@ RenderingDevice::FramebufferFormatID RenderingDeviceVulkan::framebuffer_format_c
 	}
 
 	int color_references;
-	VkRenderPass render_pass = _render_pass_create(p_format, INITIAL_ACTION_CLEAR, FINAL_ACTION_DISCARD, INITIAL_ACTION_CLEAR, FINAL_ACTION_DISCARD, &color_references); //actions don't matter for this use case
+	VkRenderPass render_pass = _render_pass_create(p_format, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, INITIAL_ACTION_CLEAR, FINAL_ACTION_READ, &color_references); //actions don't matter for this use case
 
 	if (render_pass == VK_NULL_HANDLE) { //was likely invalid
 		return INVALID_ID;
@@ -4886,43 +4985,24 @@ Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint
 	ERR_FAIL_COND_V_MSG(draw_list && p_sync_with_draw, ERR_INVALID_PARAMETER,
 			"Updating buffers in 'sync to draw' mode is forbidden during creation of a draw list");
 
-	VkPipelineStageFlags dst_stage_mask;
-	VkAccessFlags dst_access;
+	// Protect subsequent updates...
+	VkPipelineStageFlags dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT;
+	VkAccessFlags dst_access = VK_ACCESS_TRANSFER_WRITE_BIT;
 
-	Buffer *buffer = nullptr;
-	if (vertex_buffer_owner.owns(p_buffer)) {
-		dst_stage_mask = VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
-		dst_access = VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT;
-		buffer = vertex_buffer_owner.getornull(p_buffer);
-	} else if (index_buffer_owner.owns(p_buffer)) {
-		dst_stage_mask = VK_PIPELINE_STAGE_VERTEX_INPUT_BIT;
-		dst_access = VK_ACCESS_INDEX_READ_BIT;
-		buffer = index_buffer_owner.getornull(p_buffer);
-	} else if (uniform_buffer_owner.owns(p_buffer)) {
-		dst_stage_mask = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
-		dst_access = VK_ACCESS_UNIFORM_READ_BIT;
-		buffer = uniform_buffer_owner.getornull(p_buffer);
-	} else if (texture_buffer_owner.owns(p_buffer)) {
-		dst_stage_mask = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
-		dst_access = VK_ACCESS_SHADER_READ_BIT;
-		buffer = &texture_buffer_owner.getornull(p_buffer)->buffer;
-	} else if (storage_buffer_owner.owns(p_buffer)) {
-		dst_stage_mask = VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT;
-		dst_access = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
-		buffer = storage_buffer_owner.getornull(p_buffer);
-	} else {
+	Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access);
+	if (!buffer) {
 		ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Buffer argument is not a valid buffer of any type.");
 	}
 
 	ERR_FAIL_COND_V_MSG(p_offset + p_size > buffer->size, ERR_INVALID_PARAMETER,
 			"Attempted to write buffer (" + itos((p_offset + p_size) - buffer->size) + " bytes) past the end.");
 
+	_buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, p_sync_with_draw);
 	Error err = _buffer_update(buffer, p_offset, (uint8_t *)p_data, p_size, p_sync_with_draw);
 	if (err) {
 		return err;
 	}
 
-	_buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, p_sync_with_draw);
 #ifdef FORCE_FULL_BARRIER
 	_full_barrier(p_sync_with_draw);
 #else
@@ -4934,20 +5014,20 @@ Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint
 Vector<uint8_t> RenderingDeviceVulkan::buffer_get_data(RID p_buffer) {
 	_THREAD_SAFE_METHOD_
 
-	Buffer *buffer = nullptr;
-	if (vertex_buffer_owner.owns(p_buffer)) {
-		buffer = vertex_buffer_owner.getornull(p_buffer);
-	} else if (index_buffer_owner.owns(p_buffer)) {
-		buffer = index_buffer_owner.getornull(p_buffer);
-	} else if (texture_buffer_owner.owns(p_buffer)) {
-		buffer = &texture_buffer_owner.getornull(p_buffer)->buffer;
-	} else if (storage_buffer_owner.owns(p_buffer)) {
-		buffer = storage_buffer_owner.getornull(p_buffer);
-	} else {
+	// It could be this buffer was just created
+	VkPipelineShaderStageCreateFlags src_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT;
+	VkAccessFlags src_access_mask = VK_ACCESS_TRANSFER_WRITE_BIT;
+	// Get the vulkan buffer and the potential stage/access possible
+	Buffer *buffer = _get_buffer_from_owner(p_buffer, src_stage_mask, src_access_mask);
+	if (!buffer) {
 		ERR_FAIL_V_MSG(Vector<uint8_t>(), "Buffer is either invalid or this type of buffer can't be retrieved. Only Index and Vertex buffers allow retrieving.");
 	}
 
+	// Make sure  no one is using the buffer -- the "false" gets us to the same command buffer as below.
+	_buffer_memory_barrier(buffer->buffer, 0, buffer->size, src_stage_mask, src_access_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_TRANSFER_READ_BIT, false);
+
 	VkCommandBuffer command_buffer = frames[frame].setup_command_buffer;
+
 	Buffer tmp_buffer;
 	_buffer_allocate(&tmp_buffer, buffer->size, VK_BUFFER_USAGE_TRANSFER_DST_BIT, VMA_MEMORY_USAGE_CPU_ONLY);
 	VkBufferCopy region;
@@ -5614,7 +5694,7 @@ Error RenderingDeviceVulkan::_draw_list_render_pass_begin(Framebuffer *framebuff
 			image_memory_barrier.subresourceRange.baseArrayLayer = texture->base_layer;
 			image_memory_barrier.subresourceRange.layerCount = texture->layers;
 
-			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 
 			texture->layout = VK_IMAGE_LAYOUT_GENERAL;
 
@@ -6352,7 +6432,7 @@ void RenderingDeviceVulkan::draw_list_end() {
 		image_memory_barrier.subresourceRange.baseArrayLayer = texture->base_layer;
 		image_memory_barrier.subresourceRange.layerCount = texture->layers;
 
-		vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 
 		texture->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
 	}
@@ -6494,7 +6574,7 @@ void RenderingDeviceVulkan::compute_list_bind_uniform_set(ComputeListID p_list,
 			image_memory_barrier.subresourceRange.baseArrayLayer = textures_to_sampled[i]->base_layer;
 			image_memory_barrier.subresourceRange.layerCount = textures_to_sampled[i]->layers;
 
-			vkCmdPipelineBarrier(cl->command_buffer, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+			vkCmdPipelineBarrier(cl->command_buffer, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 
 			textures_to_sampled[i]->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
 
@@ -6691,13 +6771,13 @@ void RenderingDeviceVulkan::compute_list_add_barrier(ComputeListID p_list) {
 
 void RenderingDeviceVulkan::compute_list_end() {
 	ERR_FAIL_COND(!compute_list);
-
+	const VkPipelineStageFlags dest_stage_mask = VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT;
 	for (Set<Texture *>::Element *E = compute_list->state.textures_to_sampled_layout.front(); E; E = E->next()) {
 		VkImageMemoryBarrier image_memory_barrier;
 		image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER;
 		image_memory_barrier.pNext = nullptr;
 		image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT;
-		image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
+		image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT;
 		image_memory_barrier.oldLayout = E->get()->layout;
 		image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
 
@@ -6710,7 +6790,8 @@ void RenderingDeviceVulkan::compute_list_end() {
 		image_memory_barrier.subresourceRange.baseArrayLayer = E->get()->base_layer;
 		image_memory_barrier.subresourceRange.layerCount = E->get()->layers;
 
-		vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
+		// TODO: Look at the usages in the compute list and determine tighter dst stage and access masks based on some "final" usage equivalent
+		vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, dest_stage_mask, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier);
 
 		E->get()->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL;
 	}
@@ -6720,7 +6801,7 @@ void RenderingDeviceVulkan::compute_list_end() {
 #ifdef FORCE_FULL_BARRIER
 	_full_barrier(true);
 #else
-	_memory_barrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_SHADER_WRITE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT, true);
+	_memory_barrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, dest_stage_mask, VK_ACCESS_SHADER_WRITE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT, true);
 #endif
 }
 
diff --git a/drivers/vulkan/rendering_device_vulkan.h b/drivers/vulkan/rendering_device_vulkan.h
index 6f8bbc9c64..e6cbf2e01d 100644
--- a/drivers/vulkan/rendering_device_vulkan.h
+++ b/drivers/vulkan/rendering_device_vulkan.h
@@ -793,6 +793,7 @@ class RenderingDeviceVulkan : public RenderingDevice {
 	Error _draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass);
 	Error _draw_list_render_pass_begin(Framebuffer *framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector<Color> &p_clear_colors, float p_clear_depth, uint32_t p_clear_stencil, Point2i viewport_offset, Point2i viewport_size, VkFramebuffer vkframebuffer, VkRenderPass render_pass, VkCommandBuffer command_buffer, VkSubpassContents subpass_contents, const Vector<RID> &p_storage_textures);
 	_FORCE_INLINE_ DrawList *_get_draw_list_ptr(DrawListID p_id);
+	Buffer *_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &dst_stage_mask, VkAccessFlags &dst_access);
 
 	/**********************/
 	/**** COMPUTE LIST ****/