From e52c8955595c5e3e9e88180847c8eff71356a0c2 Mon Sep 17 00:00:00 2001
From: Fernando Sahmkow <fsahmkow27@gmail.com>
Date: Fri, 26 Jul 2019 14:20:43 -0400
Subject: GPU: Flush commands on every dma pusher step.

This commit ensures that the host gpu is constantly fed with commands to
work with, while the guest gpu keeps producing the rest of the commands.
This reduces syncing time between host and guest gpu.
---
 src/video_core/dma_pusher.cpp                    | 1 +
 src/video_core/gpu.cpp                           | 4 ++++
 src/video_core/gpu.h                             | 2 ++
 src/video_core/rasterizer_interface.h            | 3 +++
 src/video_core/renderer_opengl/gl_rasterizer.cpp | 4 ++++
 src/video_core/renderer_opengl/gl_rasterizer.h   | 1 +
 6 files changed, 15 insertions(+)

(limited to 'src')

diff --git a/src/video_core/dma_pusher.cpp b/src/video_core/dma_pusher.cpp
index bd036cbe8..0094fd715 100644
--- a/src/video_core/dma_pusher.cpp
+++ b/src/video_core/dma_pusher.cpp
@@ -31,6 +31,7 @@ void DmaPusher::DispatchCalls() {
             break;
         }
     }
+    gpu.FlushCommands();
 }
 
 bool DmaPusher::Step() {
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index 1622332a4..c409af194 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -120,6 +120,10 @@ bool GPU::CancelSyncptInterrupt(const u32 syncpoint_id, const u32 value) {
     return true;
 }
 
+void GPU::FlushCommands() {
+    renderer.Rasterizer().FlushCommands();
+}
+
 u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
     ASSERT(format != RenderTargetFormat::NONE);
 
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index 87c96f46b..11857ff99 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -153,6 +153,8 @@ public:
     /// Calls a GPU method.
     void CallMethod(const MethodCall& method_call);
 
+    void FlushCommands();
+
     /// Returns a reference to the Maxwell3D GPU engine.
     Engines::Maxwell3D& Maxwell3D();
 
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 9881df0d5..6e44d51cf 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -50,6 +50,9 @@ public:
     /// and invalidated
     virtual void FlushAndInvalidateRegion(CacheAddr addr, u64 size) = 0;
 
+    // Notify the rasterizer to send all written commands to the host GPU.
+    virtual void FlushCommands() = 0;
+
     /// Notify rasterizer that a frame is about to finish
     virtual void TickFrame() = 0;
 
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index c28ae795c..80cfda7e4 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -863,6 +863,10 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
     InvalidateRegion(addr, size);
 }
 
+void RasterizerOpenGL::FlushCommands() {
+    glFlush();
+}
+
 void RasterizerOpenGL::TickFrame() {
     buffer_cache.TickFrame();
 }
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 8b123c48d..9d20a4fbf 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -63,6 +63,7 @@ public:
     void FlushRegion(CacheAddr addr, u64 size) override;
     void InvalidateRegion(CacheAddr addr, u64 size) override;
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
+    void FlushCommands() override;
     void TickFrame() override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-- 
cgit v1.2.3