From 600dddf88db0a786b945c65d27da05105410bfe6 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow
Date: Sat, 8 Jun 2019 17:04:41 -0400
Subject: Async GPU: do invalidate as synced operation

Async GPU: Always invalidate synced.
---
 src/video_core/gpu_thread.cpp | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

(limited to 'src/video_core/gpu_thread.cpp')

diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 3f0939ec9..692655395 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -89,12 +89,7 @@ void ThreadManager::FlushRegion(CacheAddr addr, u64 size) {
 }
 
 void ThreadManager::InvalidateRegion(CacheAddr addr, u64 size) {
-    if (state.queue.Empty()) {
-        // It's quicker to invalidate a single region on the CPU if the queue is already empty
-        system.Renderer().Rasterizer().InvalidateRegion(addr, size);
-    } else {
-        PushCommand(InvalidateRegionCommand(addr, size));
-    }
+    system.Renderer().Rasterizer().InvalidateRegion(addr, size);
 }
 
 void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
-- 
cgit v1.2.3


From f2e026a1d8fb2384c1ece24e6dd32062b4f390a2 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow
Date: Mon, 10 Jun 2019 08:36:22 -0400
Subject: gpu_asynch: Simplify synchronization to a simpler consumer->producer
 scheme.

---
 src/video_core/gpu_thread.cpp | 18 +++---------------
 1 file changed, 3 insertions(+), 15 deletions(-)

(limited to 'src/video_core/gpu_thread.cpp')

diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index 692655395..b87938fdd 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -21,7 +21,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
     MicroProfileOnThreadCreate("GpuThread");
 
     // Wait for first GPU command before acquiring the window context
-    state.WaitForCommands();
+    while (state.queue.Empty());
 
     // If emulation was stopped during disk shader loading, abort before trying to acquire context
     if (!state.is_running) {
@@ -32,7 +32,6 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
 
     CommandDataContainer next;
     while (state.is_running) {
-        state.WaitForCommands();
         while (!state.queue.Empty()) {
             state.queue.Pop(next);
             if (const auto submit_list = std::get_if<SubmitListCommand>(&next.data)) {
@@ -49,8 +48,7 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
             } else {
                 UNREACHABLE();
             }
-            state.signaled_fence = next.fence;
-            state.TrySynchronize();
+            state.signaled_fence.store(next.fence);
         }
     }
 }
@@ -100,22 +98,12 @@ void ThreadManager::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
-    state.SignalCommands();
     return fence;
 }
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 void SynchState::WaitForSynchronization(u64 fence) {
-    if (signaled_fence >= fence) {
-        return;
-    }
-
-    // Wait for the GPU to be idle (all commands to be executed)
-    {
-        MICROPROFILE_SCOPE(GPU_wait);
-        std::unique_lock lock{synchronization_mutex};
-        synchronization_condition.wait(lock, [this, fence] { return signaled_fence >= fence; });
-    }
+    while (signaled_fence.load() < fence);
 }
 
 } // namespace VideoCommon::GPUThread
-- 
cgit v1.2.3


From b391e5f6386eecf6170b544245e3e4e31427913c Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow
Date: Tue, 18 Jun 2019 16:58:29 -0400
Subject: NVFlinger: Correct GCC compile error

---
 src/video_core/gpu_thread.cpp | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

(limited to 'src/video_core/gpu_thread.cpp')

diff --git a/src/video_core/gpu_thread.cpp b/src/video_core/gpu_thread.cpp
index b87938fdd..b441e92b0 100644
--- a/src/video_core/gpu_thread.cpp
+++ b/src/video_core/gpu_thread.cpp
@@ -21,7 +21,8 @@ static void RunThread(VideoCore::RendererBase& renderer, Tegra::DmaPusher& dma_p
     MicroProfileOnThreadCreate("GpuThread");
 
     // Wait for first GPU command before acquiring the window context
-    while (state.queue.Empty());
+    while (state.queue.Empty())
+        ;
 
     // If emulation was stopped during disk shader loading, abort before trying to acquire context
     if (!state.is_running) {
@@ -103,7 +104,8 @@ u64 ThreadManager::PushCommand(CommandData&& command_data) {
 
 MICROPROFILE_DEFINE(GPU_wait, "GPU", "Wait for the GPU", MP_RGB(128, 128, 192));
 void SynchState::WaitForSynchronization(u64 fence) {
-    while (signaled_fence.load() < fence);
+    while (signaled_fence.load() < fence)
+        ;
 }
 
 } // namespace VideoCommon::GPUThread
-- 
cgit v1.2.3