Use fence

Author: floppyhammer

pathfinder/core/d3d11/renderer.cpp

@@ -315,7 +315,7 @@ void RendererD3D11::set_dest_texture(const std::shared_ptr<Texture> &new_texture
 void RendererD3D11::upload_scene(SegmentsD3D11 &draw_segments, SegmentsD3D11 &clip_segments) {
     auto encoder = device->create_command_encoder("upload scene");
     scene_buffers.upload(draw_segments, clip_segments, allocator, device, encoder);
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 void RendererD3D11::prepare_and_draw_tiles(DrawTileBatchD3D11 &batch) {
@@ -425,7 +425,7 @@ void RendererD3D11::draw_tiles(uint64_t tiles_d3d11_buffer_id,
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 Vec2I RendererD3D11::tile_size() const {
@@ -475,7 +475,7 @@ PropagateMetadataBufferIDsD3D11 RendererD3D11::upload_propagate_metadata(
                           0,
                           propagate_metadata.size() * sizeof(PropagateMetadataD3D11),
                           propagate_metadata.data());
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     return {propagate_metadata_storage_id, backdrops_storage_id};
 }
@@ -485,7 +485,7 @@ void RendererD3D11::upload_initial_backdrops(uint64_t backdrops_buffer_id, std::
 
     auto encoder = device->create_command_encoder("upload initial backdrops");
     encoder->write_buffer(backdrops_buffer, 0, backdrops.size() * sizeof(BackdropInfoD3D11), backdrops.data());
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 void RendererD3D11::prepare_tiles(TileBatchDataD3D11 &batch) {
@@ -684,7 +684,7 @@ std::shared_ptr<MicrolinesBufferIDsD3D11> RendererD3D11::dice_segments(std::vect
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     // Read indirect draw params back to CPU memory.
     indirect_draw_params_buffer->download_via_mapping(FILL_INDIRECT_DRAW_PARAMS_SIZE * sizeof(uint32_t),
@@ -751,7 +751,7 @@ void RendererD3D11::bound(uint64_t tiles_d3d11_buffer_id,
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     allocator->free_buffer(path_info_buffer_id);
 }
@@ -811,7 +811,7 @@ std::shared_ptr<FillBufferInfoD3D11> RendererD3D11::bin_segments(
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     // Read buffer.
     z_buffer->download_via_mapping(FILL_INDIRECT_DRAW_PARAMS_SIZE * sizeof(uint32_t), 0, indirect_draw_params);
@@ -919,7 +919,7 @@ PropagateTilesInfoD3D11 RendererD3D11::propagate_tiles(uint32_t column_count,
 
     uint32_t fill_indirect_draw_params[FILL_INDIRECT_DRAW_PARAMS_SIZE];
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     // Read buffer.
     z_buffer->download_via_mapping(FILL_INDIRECT_DRAW_PARAMS_SIZE * sizeof(uint32_t), 0, fill_indirect_draw_params);
@@ -980,7 +980,7 @@ void RendererD3D11::draw_fills(FillBufferInfoD3D11 &fill_storage_info,
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 void RendererD3D11::sort_tiles(uint64_t tiles_d3d11_buffer_id,
@@ -1017,7 +1017,7 @@ void RendererD3D11::sort_tiles(uint64_t tiles_d3d11_buffer_id,
 
     encoder->end_compute_pass();
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 void RendererD3D11::free_tile_batch_buffers() {

pathfinder/core/d3d9/renderer.cpp

@@ -68,7 +68,7 @@ RendererD3D9::RendererD3D9(const std::shared_ptr<Device> &_device, const std::sh
                           quad_vertex_data_size,
                           QUAD_VERTEX_POSITIONS);
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 void RendererD3D9::set_dest_texture(const std::shared_ptr<Texture> &texture) {
@@ -363,7 +363,7 @@ void RendererD3D9::draw(const std::shared_ptr<SceneBuilder> &_scene_builder, boo
         // We can do fill drawing as soon as the fill vertex buffer is ready.
         draw_fills(fill_vertex_buffer_id, scene_builder->pending_fills.size(), encoder);
 
-        queue->submit_and_wait(encoder);
+        queue->submit(encoder, fence);
 
         allocator->free_buffer(fill_vertex_buffer_id);
     }
@@ -442,7 +442,7 @@ void RendererD3D9::upload_and_draw_tiles(const std::vector<DrawTileBatchD3D9> &t
                    z_buffer_texture_id,
                    encoder);
 
-        queue->submit_and_wait(encoder);
+        queue->submit(encoder, fence);
 
         allocator->free_texture(z_buffer_texture_id);
         allocator->free_buffer(tile_vertex_buffer_id);

pathfinder/core/renderer.cpp

@@ -27,11 +27,13 @@ Renderer::Renderer(const std::shared_ptr<Device> &_device, const std::shared_ptr
                                                       TextureFormat::Rgba16Float,
                                                       "metadata texture");
 
+    fence = device->create_fence("renderer fence");
+
     auto encoder = device->create_command_encoder("upload common renderer data");
 
     encoder->write_texture(allocator->get_texture(area_lut_texture_id), {}, image_buffer->get_data());
 
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 Renderer::~Renderer() {
@@ -102,7 +104,7 @@ void Renderer::upload_texel_data(std::vector<ColorU> &texels, TextureLocation lo
 
     auto encoder = device->create_command_encoder("upload data of the pattern texture pages");
     encoder->write_texture(texture, location.rect, texels.data());
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 
     texture_page->must_preserve_contents_ = true;
 }
@@ -249,7 +251,7 @@ void Renderer::upload_texture_metadata(const std::vector<TextureMetadataEntry> &
     auto encoder = device->create_command_encoder("upload to metadata texture");
     encoder->add_callback(callback);
     encoder->write_texture(allocator->get_texture(metadata_texture_id), region_rect, raw_texels);
-    queue->submit_and_wait(encoder);
+    queue->submit(encoder, fence);
 }
 
 } // namespace Pathfinder

pathfinder/core/renderer.h

@@ -90,6 +90,8 @@ class Renderer {
 
     std::shared_ptr<Queue> queue;
 
+    std::shared_ptr<Fence> fence;
+
 protected:
     virtual TextureFormat mask_texture_format() const = 0;
 

pathfinder/gpu/gl/queue.h

@@ -22,8 +22,10 @@ class QueueGl : public Queue {
 
         encoder->finish();
 
-        auto fence_gl = (FenceGl *)fence.get();
-        fence_gl->wait();
+        if (fence) {
+            auto fence_gl = (FenceGl *)fence.get();
+            fence_gl->wait();
+        }
     }
 
     void submit_and_wait(const std::shared_ptr<CommandEncoder> &encoder) override {

pathfinder/gpu/vk/queue.cpp

@@ -25,11 +25,15 @@ void QueueVk::submit(const std::shared_ptr<CommandEncoder> &encoder, const std::
     submit_info.commandBufferCount = 1;
     submit_info.pCommandBuffers = &encoder_vk->vk_command_buffer_;
 
-    auto fence_vk = (FenceVk *)fence.get();
+    if (fence) {
+        auto fence_vk = (FenceVk *)fence.get();
 
-    vkQueueSubmit(vk_graphics_queue_, 1, &submit_info, fence_vk->fence);
+        vkQueueSubmit(vk_graphics_queue_, 1, &submit_info, fence_vk->fence);
 
-    fence_vk->wait();
+        fence_vk->wait();
+    } else {
+        vkQueueSubmit(vk_graphics_queue_, 1, &submit_info, VK_NULL_HANDLE);
+    }
 }
 
 void QueueVk::submit_and_wait(const std::shared_ptr<CommandEncoder> &encoder) {