1 files changed, 147 insertions, 10 deletions

diff --git a/servers/rendering/rendering_device.cpp b/servers/rendering/rendering_device.cpp
index 942d5631e4..e9fa38475e 100644
--- a/servers/rendering/rendering_device.cpp
+++ b/servers/rendering/rendering_device.cpp
@@ -946,6 +946,7 @@ RID RenderingDevice::texture_create_shared(const TextureView &p_view, RID p_with
 
 		RDG::ResourceTracker *tracker = RDG::resource_tracker_create();
 		tracker->texture_driver_id = texture.shared_fallback->texture;
+		tracker->texture_size = Size2i(texture.width, texture.height);
 		tracker->texture_subresources = texture.barrier_range();
 		tracker->texture_usage = alias_format.usage_bits;
 		tracker->reference_count = 1;
@@ -1125,6 +1126,7 @@ RID RenderingDevice::texture_create_shared_from_slice(const TextureView &p_view,
 
 		RDG::ResourceTracker *tracker = RDG::resource_tracker_create();
 		tracker->texture_driver_id = texture.shared_fallback->texture;
+		tracker->texture_size = Size2i(texture.width, texture.height);
 		tracker->texture_subresources = slice_range;
 		tracker->texture_usage = slice_format.usage_bits;
 		tracker->reference_count = 1;
@@ -1260,8 +1262,6 @@ Error RenderingDevice::_texture_initialize(RID p_texture, uint32_t p_layer, cons
 			write_ptr = driver->buffer_map(transfer_worker->staging_buffer);
 			ERR_FAIL_NULL_V(write_ptr, ERR_CANT_CREATE);
 
-			write_ptr += staging_worker_offset;
-
 			if (driver->api_trait_get(RDD::API_TRAIT_HONORS_PIPELINE_BARRIERS)) {
 				// Transition the texture to the optimal layout.
 				RDD::TextureBarrier tb;
@@ -1303,11 +1303,12 @@ Error RenderingDevice::_texture_initialize(RID p_texture, uint32_t p_layer, cons
 
 				if (copy_pass) {
 					const uint8_t *read_ptr_mipmap_layer = read_ptr_mipmap + (tight_mip_size / depth) * z;
-					_copy_region_block_or_regular(read_ptr_mipmap_layer, write_ptr, 0, 0, width, width, height, block_w, block_h, pitch, pixel_size, block_size);
-					write_ptr += to_allocate;
+					uint64_t staging_buffer_offset = staging_worker_offset + staging_local_offset;
+					uint8_t *write_ptr_mipmap_layer = write_ptr + staging_buffer_offset;
+					_copy_region_block_or_regular(read_ptr_mipmap_layer, write_ptr_mipmap_layer, 0, 0, width, width, height, block_w, block_h, pitch, pixel_size, block_size);
 
 					RDD::BufferTextureCopyRegion copy_region;
-					copy_region.buffer_offset = staging_worker_offset + staging_local_offset;
+					copy_region.buffer_offset = staging_buffer_offset;
 					copy_region.texture_subresources.aspect = texture->read_aspect_flags;
 					copy_region.texture_subresources.mipmap = mm_i;
 					copy_region.texture_subresources.base_layer = p_layer;
@@ -3758,6 +3759,15 @@ int RenderingDevice::screen_get_height(DisplayServer::WindowID p_screen) const {
 	return context->surface_get_height(surface);
 }
 
+int RenderingDevice::screen_get_pre_rotation_degrees(DisplayServer::WindowID p_screen) const {
+	_THREAD_SAFE_METHOD_
+
+	HashMap<DisplayServer::WindowID, RDD::SwapChainID>::ConstIterator it = screen_swap_chains.find(p_screen);
+	ERR_FAIL_COND_V_MSG(it == screen_swap_chains.end(), ERR_CANT_CREATE, "A swap chain was not created for the screen.");
+
+	return driver->swap_chain_get_pre_rotation_degrees(it->value);
+}
+
 RenderingDevice::FramebufferFormatID RenderingDevice::screen_get_framebuffer_format(DisplayServer::WindowID p_screen) const {
 	_THREAD_SAFE_METHOD_
 
@@ -4441,6 +4451,117 @@ void RenderingDevice::draw_list_draw(DrawListID p_list, bool p_use_indices, uint
 	dl->state.draw_count++;
 }
 
+void RenderingDevice::draw_list_draw_indirect(DrawListID p_list, bool p_use_indices, RID p_buffer, uint32_t p_offset, uint32_t p_draw_count, uint32_t p_stride) {
+	ERR_RENDER_THREAD_GUARD();
+
+	DrawList *dl = _get_draw_list_ptr(p_list);
+	ERR_FAIL_NULL(dl);
+
+	Buffer *buffer = storage_buffer_owner.get_or_null(p_buffer);
+	ERR_FAIL_NULL(buffer);
+
+	ERR_FAIL_COND_MSG(!buffer->usage.has_flag(RDD::BUFFER_USAGE_INDIRECT_BIT), "Buffer provided was not created to do indirect dispatch.");
+
+#ifdef DEBUG_ENABLED
+	ERR_FAIL_COND_MSG(!dl->validation.active, "Submitted Draw Lists can no longer be modified.");
+#endif
+
+#ifdef DEBUG_ENABLED
+	ERR_FAIL_COND_MSG(!dl->validation.pipeline_active,
+			"No render pipeline was set before attempting to draw.");
+	if (dl->validation.pipeline_vertex_format != INVALID_ID) {
+		// Pipeline uses vertices, validate format.
+		ERR_FAIL_COND_MSG(dl->validation.vertex_format == INVALID_ID,
+				"No vertex array was bound, and render pipeline expects vertices.");
+		// Make sure format is right.
+		ERR_FAIL_COND_MSG(dl->validation.pipeline_vertex_format != dl->validation.vertex_format,
+				"The vertex format used to create the pipeline does not match the vertex format bound.");
+	}
+
+	if (dl->validation.pipeline_push_constant_size > 0) {
+		// Using push constants, check that they were supplied.
+		ERR_FAIL_COND_MSG(!dl->validation.pipeline_push_constant_supplied,
+				"The shader in this pipeline requires a push constant to be set before drawing, but it's not present.");
+	}
+#endif
+
+#ifdef DEBUG_ENABLED
+	for (uint32_t i = 0; i < dl->state.set_count; i++) {
+		if (dl->state.sets[i].pipeline_expected_format == 0) {
+			// Nothing expected by this pipeline.
+			continue;
+		}
+
+		if (dl->state.sets[i].pipeline_expected_format != dl->state.sets[i].uniform_set_format) {
+			if (dl->state.sets[i].uniform_set_format == 0) {
+				ERR_FAIL_MSG(vformat("Uniforms were never supplied for set (%d) at the time of drawing, which are required by the pipeline.", i));
+			} else if (uniform_set_owner.owns(dl->state.sets[i].uniform_set)) {
+				UniformSet *us = uniform_set_owner.get_or_null(dl->state.sets[i].uniform_set);
+				ERR_FAIL_MSG(vformat("Uniforms supplied for set (%d):\n%s\nare not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n%s", i, _shader_uniform_debug(us->shader_id, us->shader_set), _shader_uniform_debug(dl->state.pipeline_shader)));
+			} else {
+				ERR_FAIL_MSG(vformat("Uniforms supplied for set (%s, which was just freed) are not the same format as required by the pipeline shader. Pipeline shader requires the following bindings:\n%s", i, _shader_uniform_debug(dl->state.pipeline_shader)));
+			}
+		}
+	}
+#endif
+
+	// Prepare descriptor sets if the API doesn't use pipeline barriers.
+	if (!driver->api_trait_get(RDD::API_TRAIT_HONORS_PIPELINE_BARRIERS)) {
+		for (uint32_t i = 0; i < dl->state.set_count; i++) {
+			if (dl->state.sets[i].pipeline_expected_format == 0) {
+				// Nothing expected by this pipeline.
+				continue;
+			}
+
+			draw_graph.add_draw_list_uniform_set_prepare_for_use(dl->state.pipeline_shader_driver_id, dl->state.sets[i].uniform_set_driver_id, i);
+		}
+	}
+
+	// Bind descriptor sets.
+	for (uint32_t i = 0; i < dl->state.set_count; i++) {
+		if (dl->state.sets[i].pipeline_expected_format == 0) {
+			continue; // Nothing expected by this pipeline.
+		}
+		if (!dl->state.sets[i].bound) {
+			// All good, see if this requires re-binding.
+			draw_graph.add_draw_list_bind_uniform_set(dl->state.pipeline_shader_driver_id, dl->state.sets[i].uniform_set_driver_id, i);
+
+			UniformSet *uniform_set = uniform_set_owner.get_or_null(dl->state.sets[i].uniform_set);
+			_uniform_set_update_shared(uniform_set);
+
+			draw_graph.add_draw_list_usages(uniform_set->draw_trackers, uniform_set->draw_trackers_usage);
+
+			dl->state.sets[i].bound = true;
+		}
+	}
+
+	if (p_use_indices) {
+#ifdef DEBUG_ENABLED
+		ERR_FAIL_COND_MSG(!dl->validation.index_array_count,
+				"Draw command requested indices, but no index buffer was set.");
+
+		ERR_FAIL_COND_MSG(dl->validation.pipeline_uses_restart_indices != dl->validation.index_buffer_uses_restart_indices,
+				"The usage of restart indices in index buffer does not match the render primitive in the pipeline.");
+#endif
+
+		ERR_FAIL_COND_MSG(p_offset + 20 > buffer->size, "Offset provided (+20) is past the end of buffer.");
+
+		draw_graph.add_draw_list_draw_indexed_indirect(buffer->driver_id, p_offset, p_draw_count, p_stride);
+	} else {
+		ERR_FAIL_COND_MSG(p_offset + 16 > buffer->size, "Offset provided (+16) is past the end of buffer.");
+
+		draw_graph.add_draw_list_draw_indirect(buffer->driver_id, p_offset, p_draw_count, p_stride);
+	}
+
+	dl->state.draw_count++;
+
+	if (buffer->draw_tracker != nullptr) {
+		draw_graph.add_draw_list_usage(buffer->draw_tracker, RDG::RESOURCE_USAGE_INDIRECT_BUFFER_READ);
+	}
+
+	_check_transfer_worker_buffer(buffer);
+}
+
 void RenderingDevice::draw_list_enable_scissor(DrawListID p_list, const Rect2 &p_rect) {
 	ERR_RENDER_THREAD_GUARD();
 
@@ -5153,14 +5274,13 @@ void RenderingDevice::_wait_for_transfer_worker(TransferWorker *p_transfer_worke
 		p_transfer_worker->operations_processed = p_transfer_worker->operations_submitted;
 	}
 
-	if (!p_transfer_worker->texture_barriers.is_empty()) {
-		MutexLock transfer_worker_lock(transfer_worker_pool_mutex);
-		_flush_barriers_for_transfer_worker(p_transfer_worker);
-	}
+	_flush_barriers_for_transfer_worker(p_transfer_worker);
 }
 
 void RenderingDevice::_flush_barriers_for_transfer_worker(TransferWorker *p_transfer_worker) {
+	// Caller must have already acquired the mutex for the worker.
 	if (!p_transfer_worker->texture_barriers.is_empty()) {
+		MutexLock transfer_worker_lock(transfer_worker_pool_texture_barriers_mutex);
 		for (uint32_t i = 0; i < p_transfer_worker->texture_barriers.size(); i++) {
 			transfer_worker_pool_texture_barriers.push_back(p_transfer_worker->texture_barriers[i]);
 		}
@@ -5233,8 +5353,11 @@ void RenderingDevice::_submit_transfer_workers(RDD::CommandBufferID p_draw_comma
 			}
 		}
 	}
+}
 
-	if (p_draw_command_buffer && !transfer_worker_pool_texture_barriers.is_empty()) {
+void RenderingDevice::_submit_transfer_barriers(RDD::CommandBufferID p_draw_command_buffer) {
+	MutexLock transfer_worker_lock(transfer_worker_pool_texture_barriers_mutex);
+	if (!transfer_worker_pool_texture_barriers.is_empty()) {
 		driver->command_pipeline_barrier(p_draw_command_buffer, RDD::PIPELINE_STAGE_COPY_BIT, RDD::PIPELINE_STAGE_ALL_COMMANDS_BIT, {}, {}, transfer_worker_pool_texture_barriers);
 		transfer_worker_pool_texture_barriers.clear();
 	}
@@ -5294,6 +5417,7 @@ bool RenderingDevice::_texture_make_mutable(Texture *p_texture, RID p_texture_id
 						draw_tracker = RDG::resource_tracker_create();
 						draw_tracker->parent = owner_texture->draw_tracker;
 						draw_tracker->texture_driver_id = p_texture->driver_id;
+						draw_tracker->texture_size = Size2i(p_texture->width, p_texture->height);
 						draw_tracker->texture_subresources = p_texture->barrier_range();
 						draw_tracker->texture_usage = p_texture->usage_flags;
 						draw_tracker->texture_slice_or_dirty_rect = p_texture->slice_rect;
@@ -5316,6 +5440,7 @@ bool RenderingDevice::_texture_make_mutable(Texture *p_texture, RID p_texture_id
 			// Regular texture.
 			p_texture->draw_tracker = RDG::resource_tracker_create();
 			p_texture->draw_tracker->texture_driver_id = p_texture->driver_id;
+			p_texture->draw_tracker->texture_size = Size2i(p_texture->width, p_texture->height);
 			p_texture->draw_tracker->texture_subresources = p_texture->barrier_range();
 			p_texture->draw_tracker->texture_usage = p_texture->usage_flags;
 			p_texture->draw_tracker->reference_count = 1;
@@ -5834,6 +5959,7 @@ void RenderingDevice::_end_frame() {
 	// The command buffer must be copied into a stack variable as the driver workarounds can change the command buffer in use.
 	RDD::CommandBufferID command_buffer = frames[frame].command_buffer;
 	_submit_transfer_workers(command_buffer);
+	_submit_transfer_barriers(command_buffer);
 
 	draw_graph.end(RENDER_GRAPH_REORDER, RENDER_GRAPH_FULL_BARRIERS, command_buffer, frames[frame].command_buffer_pool);
 	driver->command_buffer_end(command_buffer);
@@ -6553,6 +6679,12 @@ void RenderingDevice::finalize() {
 	ERR_FAIL_COND(reverse_dependency_map.size());
 }
 
+void RenderingDevice::_set_max_fps(int p_max_fps) {
+	for (const KeyValue<DisplayServer::WindowID, RDD::SwapChainID> &it : screen_swap_chains) {
+		driver->swap_chain_set_max_fps(it.value, p_max_fps);
+	}
+}
+
 RenderingDevice *RenderingDevice::create_local_device() {
 	RenderingDevice *rd = memnew(RenderingDevice);
 	rd->initialize(context);
@@ -6651,6 +6783,7 @@ void RenderingDevice::_bind_methods() {
 	ClassDB::bind_method(D_METHOD("draw_list_set_push_constant", "draw_list", "buffer", "size_bytes"), &RenderingDevice::_draw_list_set_push_constant);
 
 	ClassDB::bind_method(D_METHOD("draw_list_draw", "draw_list", "use_indices", "instances", "procedural_vertex_count"), &RenderingDevice::draw_list_draw, DEFVAL(0));
+	ClassDB::bind_method(D_METHOD("draw_list_draw_indirect", "draw_list", "use_indices", "buffer", "offset", "draw_count", "stride"), &RenderingDevice::draw_list_draw_indirect, DEFVAL(0), DEFVAL(1), DEFVAL(0));
 
 	ClassDB::bind_method(D_METHOD("draw_list_enable_scissor", "draw_list", "rect"), &RenderingDevice::draw_list_enable_scissor, DEFVAL(Rect2()));
 	ClassDB::bind_method(D_METHOD("draw_list_disable_scissor", "draw_list"), &RenderingDevice::draw_list_disable_scissor);
@@ -7260,6 +7393,10 @@ void RenderingDevice::_bind_methods() {
 	BIND_ENUM_CONSTANT(DEBUG_PASS);
 }
 
+void RenderingDevice::make_current() {
+	render_thread_id = Thread::get_caller_id();
+}
+
 RenderingDevice::~RenderingDevice() {
 	finalize();