From b78bc4cb66d29019119349ef31cb4cbadf870cfe Mon Sep 17 00:00:00 2001
From: RSDuck <RSDuck@users.noreply.github.com>
Date: Tue, 26 Jan 2021 16:42:27 +0100
Subject: [PATCH] fixes to the threadedness of the sw rasteriser also fix #639
 and fix #880

---
 src/GPU.cpp        |  5 +++++
 src/GPU.h          |  1 +
 src/GPU2D_Soft.cpp | 28 ++++++++++++++--------------
 src/GPU3D.cpp      |  5 +++++
 src/GPU3D.h        |  2 ++
 src/GPU3D_Soft.cpp |  4 ++++
 src/GPU_OpenGL.cpp |  6 ++++++
 src/NDS.cpp        |  5 +++--
 8 files changed, 40 insertions(+), 16 deletions(-)

diff --git a/src/GPU.cpp b/src/GPU.cpp
index e2ece10d..40334de7 100644
--- a/src/GPU.cpp
+++ b/src/GPU.cpp
@@ -289,6 +289,11 @@ void Stop()
     memset(Framebuffer[0][1], 0, fbsize*4);
     memset(Framebuffer[1][0], 0, fbsize*4);
     memset(Framebuffer[1][1], 0, fbsize*4);
+
+#ifdef OGLRENDERER_ENABLED
+    if (Accelerated)
+        GLCompositor::Stop(); 
+#endif  
 }
 
 void DoSavestate(Savestate* file)
diff --git a/src/GPU.h b/src/GPU.h
index cf2fe7d9..3a254dfb 100644
--- a/src/GPU.h
+++ b/src/GPU.h
@@ -561,6 +561,7 @@ void Reset();
 
 void SetRenderSettings(RenderSettings& settings);
 
+void Stop();
 void RenderFrame();
 void BindOutputTexture();
 
diff --git a/src/GPU2D_Soft.cpp b/src/GPU2D_Soft.cpp
index f9c27685..53e7b733 100644
--- a/src/GPU2D_Soft.cpp
+++ b/src/GPU2D_Soft.cpp
@@ -187,6 +187,20 @@ void GPU2D_Soft::DrawScanline(u32 line)
     // oddly that's not the case for GPU A
     if (Num && !Enabled) forceblank = true;
 
+    if (line == 0 && CaptureCnt & (1 << 31) && !forceblank)
+        CaptureLatch = true;
+
+    if (Num == 0)
+    {
+        if (!Accelerated)
+            _3DLine = GPU3D::GetLine(n3dline);
+        else if (CaptureLatch && (((CaptureCnt >> 29) & 0x3) != 1))
+        {
+            _3DLine = GPU3D::GetLine(n3dline);
+            //GPU3D::GLRenderer::PrepareCaptureFrame();
+        }
+    }
+
     if (forceblank)
     {
         for (int i = 0; i < 256; i++)
@@ -202,20 +216,6 @@ void GPU2D_Soft::DrawScanline(u32 line)
     u32 dispmode = DispCnt >> 16;
     dispmode &= (Num ? 0x1 : 0x3);
 
-    if (Num == 0)
-    {
-        if (!Accelerated)
-            _3DLine = GPU3D::GetLine(n3dline);
-        else if ((CaptureCnt & (1<<31)) && (((CaptureCnt >> 29) & 0x3) != 1))
-        {
-            _3DLine = GPU3D::GetLine(n3dline);
-            //GPU3D::GLRenderer::PrepareCaptureFrame();
-        }
-    }
-
-    if (line == 0 && CaptureCnt & (1 << 31))
-        CaptureLatch = true;
-
     // always render regular graphics
     DrawScanline_BGOBJ(line);
     UpdateMosaicCounters(line);
diff --git a/src/GPU3D.cpp b/src/GPU3D.cpp
index e199d9a6..9b418300 100644
--- a/src/GPU3D.cpp
+++ b/src/GPU3D.cpp
@@ -2500,6 +2500,11 @@ void VCount144()
     if (GPU::Renderer == 0) SoftRenderer::VCount144();
 }
 
+void RestartFrame()
+{
+    if (GPU::Renderer == 0) SoftRenderer::SetupRenderThread();
+}
+
 
 bool YSort(Polygon* a, Polygon* b)
 {
diff --git a/src/GPU3D.h b/src/GPU3D.h
index 62fe92c9..e4629b04 100644
--- a/src/GPU3D.h
+++ b/src/GPU3D.h
@@ -117,6 +117,8 @@ void VCount144();
 void VBlank();
 void VCount215();
 
+void RestartFrame();
+
 void SetRenderXPos(u16 xpos);
 u32* GetLine(int line);
 
diff --git a/src/GPU3D_Soft.cpp b/src/GPU3D_Soft.cpp
index 7fcafb18..3d6ace6c 100644
--- a/src/GPU3D_Soft.cpp
+++ b/src/GPU3D_Soft.cpp
@@ -94,9 +94,13 @@ void SetupRenderThread()
             RenderThread = Platform::Thread_Create(RenderThreadFunc);
         }
 
+        // otherwise more than one frame can be queued up at once
+        Platform::Semaphore_Reset(Sema_RenderStart);
+
         if (RenderThreadRendering)
             Platform::Semaphore_Wait(Sema_RenderDone);
 
+        Platform::Semaphore_Reset(Sema_RenderDone);
         Platform::Semaphore_Reset(Sema_RenderStart);
         Platform::Semaphore_Reset(Sema_ScanlineCount);
 
diff --git a/src/GPU_OpenGL.cpp b/src/GPU_OpenGL.cpp
index fb04ca01..dd28bcdc 100644
--- a/src/GPU_OpenGL.cpp
+++ b/src/GPU_OpenGL.cpp
@@ -182,6 +182,10 @@ void SetRenderSettings(RenderSettings& settings)
     glBindFramebuffer(GL_FRAMEBUFFER, 0);
 }
 
+void Stop()
+{
+    RenderFrame();
+}
 
 void RenderFrame()
 {
@@ -195,6 +199,8 @@ void RenderFrame()
 
     glViewport(0, 0, ScreenW, ScreenH);
 
+    glClear(GL_COLOR_BUFFER_BIT);
+
     // TODO: select more shaders (filtering, etc)
     OpenGL::UseShaderProgram(CompShader[0]);
     glUniform1ui(CompScaleLoc[0], Scale);
diff --git a/src/NDS.cpp b/src/NDS.cpp
index d0346a52..7c0ecea1 100644
--- a/src/NDS.cpp
+++ b/src/NDS.cpp
@@ -1018,8 +1018,8 @@ u32 RunFrame()
             if (CPUStop & 0x40000000)
             {
                 // checkme: when is sleep mode effective?
-                //CancelEvent(Event_LCD);
-                //GPU::TotalScanlines = 263;
+                CancelEvent(Event_LCD);
+                GPU::TotalScanlines = 263;
                 break;
             }
         }
@@ -1163,6 +1163,7 @@ void SetLidClosed(bool closed)
         KeyInput &= ~(1<<23);
         SetIRQ(1, IRQ_LidOpen);
         CPUStop &= ~0x40000000;
+        GPU3D::RestartFrame();
     }
 }