vk_scheduler: Delegate commands to a worker thread and state track

Introduce a worker thread approach for delegating Vulkan work derived
from dxvk's approach. https://github.com/doitsujin/dxvk

Now that the scheduler is what handles all Vulkan work related to
command streaming, store state tracking in itself. This way we can know
when to reupload Vulkan dynamic state to the queue (since this one is
invalidated between command buffers unlike NVN). We can also store the
renderpass state and graphics pipeline bound to avoid redundant binds
and renderpass begins/ends.

1 files changed, 138 insertions, 12 deletions

diff --git a/src/video_core/renderer_vulkan/vk_scheduler.cpp b/src/video_core/renderer_vulkan/vk_scheduler.cpp
index 0f8116458..d66133ad1 100644
--- a/src/video_core/renderer_vulkan/vk_scheduler.cpp
+++ b/src/video_core/renderer_vulkan/vk_scheduler.cpp
@@ -3,7 +3,7 @@
 // Refer to the license.txt file included.
 
 #include "common/assert.h"
-#include "common/logging/log.h"
+#include "common/microprofile.h"
 #include "video_core/renderer_vulkan/declarations.h"
 #include "video_core/renderer_vulkan/vk_device.h"
 #include "video_core/renderer_vulkan/vk_resource_manager.h"
@@ -11,46 +11,172 @@
 
 namespace Vulkan {
 
+MICROPROFILE_DECLARE(Vulkan_WaitForWorker);
+
+void VKScheduler::CommandChunk::ExecuteAll(vk::CommandBuffer cmdbuf,
+                                           const vk::DispatchLoaderDynamic& dld) {
+    auto command = first;
+    while (command != nullptr) {
+        auto next = command->GetNext();
+        command->Execute(cmdbuf, dld);
+        command->~Command();
+        command = next;
+    }
+
+    command_offset = 0;
+    first = nullptr;
+    last = nullptr;
+}
+
 VKScheduler::VKScheduler(const VKDevice& device, VKResourceManager& resource_manager)
-    : device{device}, resource_manager{resource_manager} {
-    next_fence = &resource_manager.CommitFence();
+    : device{device}, resource_manager{resource_manager}, next_fence{
+                                                              &resource_manager.CommitFence()} {
+    AcquireNewChunk();
     AllocateNewContext();
+    worker_thread = std::thread(&VKScheduler::WorkerThread, this);
 }
 
-VKScheduler::~VKScheduler() = default;
+VKScheduler::~VKScheduler() {
+    quit = true;
+    cv.notify_all();
+    worker_thread.join();
+}
 
 void VKScheduler::Flush(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
-    if (release_fence)
+    if (release_fence) {
         current_fence->Release();
+    }
     AllocateNewContext();
 }
 
 void VKScheduler::Finish(bool release_fence, vk::Semaphore semaphore) {
     SubmitExecution(semaphore);
     current_fence->Wait();
-    if (release_fence)
+    if (release_fence) {
         current_fence->Release();
+    }
     AllocateNewContext();
 }
 
+void VKScheduler::WaitWorker() {
+    MICROPROFILE_SCOPE(Vulkan_WaitForWorker);
+    DispatchWork();
+
+    bool finished = false;
+    do {
+        cv.notify_all();
+        std::unique_lock lock{mutex};
+        finished = chunk_queue.Empty();
+    } while (!finished);
+}
+
+void VKScheduler::DispatchWork() {
+    if (chunk->Empty()) {
+        return;
+    }
+    chunk_queue.Push(std::move(chunk));
+    cv.notify_all();
+    AcquireNewChunk();
+}
+
+void VKScheduler::RequestRenderpass(const vk::RenderPassBeginInfo& renderpass_bi) {
+    if (state.renderpass && renderpass_bi == *state.renderpass) {
+        return;
+    }
+    const bool end_renderpass = state.renderpass.has_value();
+    state.renderpass = renderpass_bi;
+    Record([renderpass_bi, end_renderpass](auto cmdbuf, auto& dld) {
+        if (end_renderpass) {
+            cmdbuf.endRenderPass(dld);
+        }
+        cmdbuf.beginRenderPass(renderpass_bi, vk::SubpassContents::eInline, dld);
+    });
+}
+
+void VKScheduler::RequestOutsideRenderPassOperationContext() {
+    EndRenderPass();
+}
+
+void VKScheduler::BindGraphicsPipeline(vk::Pipeline pipeline) {
+    if (state.graphics_pipeline == pipeline) {
+        return;
+    }
+    state.graphics_pipeline = pipeline;
+    Record([pipeline](auto cmdbuf, auto& dld) {
+        cmdbuf.bindPipeline(vk::PipelineBindPoint::eGraphics, pipeline, dld);
+    });
+}
+
+void VKScheduler::WorkerThread() {
+    std::unique_lock lock{mutex};
+    do {
+        cv.wait(lock, [this] { return !chunk_queue.Empty() || quit; });
+        if (quit) {
+            continue;
+        }
+        auto extracted_chunk = std::move(chunk_queue.Front());
+        chunk_queue.Pop();
+        extracted_chunk->ExecuteAll(current_cmdbuf, device.GetDispatchLoader());
+        chunk_reserve.Push(std::move(extracted_chunk));
+    } while (!quit);
+}
+
 void VKScheduler::SubmitExecution(vk::Semaphore semaphore) {
+    EndPendingOperations();
+    InvalidateState();
+    WaitWorker();
+
+    std::unique_lock lock{mutex};
+
+    const auto queue = device.GetGraphicsQueue();
     const auto& dld = device.GetDispatchLoader();
     current_cmdbuf.end(dld);
 
-    const auto queue = device.GetGraphicsQueue();
-    const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1u : 0u,
+    const vk::SubmitInfo submit_info(0, nullptr, nullptr, 1, &current_cmdbuf, semaphore ? 1U : 0U,
                                      &semaphore);
-    queue.submit({submit_info}, *current_fence, dld);
+    queue.submit({submit_info}, static_cast<vk::Fence>(*current_fence), dld);
 }
 
 void VKScheduler::AllocateNewContext() {
+    std::unique_lock lock{mutex};
     current_fence = next_fence;
-    current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
     next_fence = &resource_manager.CommitFence();
 
-    const auto& dld = device.GetDispatchLoader();
-    current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit}, dld);
+    current_cmdbuf = resource_manager.CommitCommandBuffer(*current_fence);
+    current_cmdbuf.begin({vk::CommandBufferUsageFlagBits::eOneTimeSubmit},
+                         device.GetDispatchLoader());
+}
+
+void VKScheduler::InvalidateState() {
+    state.graphics_pipeline = nullptr;
+    state.viewports = false;
+    state.scissors = false;
+    state.depth_bias = false;
+    state.blend_constants = false;
+    state.depth_bounds = false;
+    state.stencil_values = false;
+}
+
+void VKScheduler::EndPendingOperations() {
+    EndRenderPass();
+}
+
+void VKScheduler::EndRenderPass() {
+    if (!state.renderpass) {
+        return;
+    }
+    state.renderpass = std::nullopt;
+    Record([](auto cmdbuf, auto& dld) { cmdbuf.endRenderPass(dld); });
+}
+
+void VKScheduler::AcquireNewChunk() {
+    if (chunk_reserve.Empty()) {
+        chunk = std::make_unique<CommandChunk>();
+        return;
+    }
+    chunk = std::move(chunk_reserve.Front());
+    chunk_reserve.Pop();
 }
 
 } // namespace Vulkan