From 5c45a50daf03b9c21fa83d4e38134679aaf0344b Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Thu, 30 Aug 2018 15:24:47 +0200
Subject: [PATCH] [d3d11] Use D3D11CounterBuffer for UAV counters

---
 src/d3d11/d3d11_device.cpp | 33 +++++++++++++++++++++++++++++----
 src/d3d11/d3d11_device.h   |  6 +++++-
 2 files changed, 34 insertions(+), 5 deletions(-)

diff --git a/src/d3d11/d3d11_device.cpp b/src/d3d11/d3d11_device.cpp
index 73ed8f5d5..48610eeb4 100644
--- a/src/d3d11/d3d11_device.cpp
+++ b/src/d3d11/d3d11_device.cpp
@@ -113,16 +113,16 @@ namespace dxvk {
       throw DxvkError("D3D11Device: Failed to query adapter");
     
     m_initializer = new D3D11Initializer(m_dxvkDevice);
-    m_uavCounters = new D3D11UavCounterAllocator(this);
     m_context     = new D3D11ImmediateContext(this, m_dxvkDevice);
     m_d3d10Device = new D3D10Device(this, m_context);
+
+    m_uavCounters = CreateUAVCounterBuffer();
   }
   
   
   D3D11Device::~D3D11Device() {
     delete m_d3d10Device;
     delete m_context;
-    delete m_uavCounters;
     delete m_initializer;
   }
   
@@ -1368,6 +1368,31 @@ namespace dxvk {
   }
   
   
+  Rc<D3D11CounterBuffer> D3D11Device::CreateUAVCounterBuffer() {
+    // UAV counters are going to be used as raw storage buffers, so
+    // we need to align them to the minimum SSBO offset alignment
+    const auto& devInfo = m_dxvkAdapter->deviceProperties();
+
+    VkDeviceSize uavCounterSliceLength = align<VkDeviceSize>(
+      sizeof(D3D11UavCounter), devInfo.limits.minStorageBufferOffsetAlignment);
+
+    DxvkBufferCreateInfo uavCounterInfo;
+    uavCounterInfo.size   = 4096 * uavCounterSliceLength;
+    uavCounterInfo.usage  = VK_BUFFER_USAGE_TRANSFER_DST_BIT
+                          | VK_BUFFER_USAGE_TRANSFER_SRC_BIT
+                          | VK_BUFFER_USAGE_STORAGE_BUFFER_BIT;
+    uavCounterInfo.stages = VK_PIPELINE_STAGE_TRANSFER_BIT
+                          | GetEnabledShaderStages();
+    uavCounterInfo.access = VK_ACCESS_TRANSFER_READ_BIT
+                          | VK_ACCESS_TRANSFER_WRITE_BIT
+                          | VK_ACCESS_SHADER_READ_BIT
+                          | VK_ACCESS_SHADER_WRITE_BIT;
+    
+    return new D3D11CounterBuffer(m_dxvkDevice,
+      uavCounterInfo, uavCounterSliceLength);
+  }
+  
+  
   HRESULT D3D11Device::CreateShaderModule(
           D3D11CommonShader*      pShaderModule,
     const void*                   pShaderBytecode,
@@ -1387,8 +1412,8 @@ namespace dxvk {
       return E_FAIL;
     }
   }
-  
-  
+
+
   HRESULT D3D11Device::GetFormatSupportFlags(DXGI_FORMAT Format, UINT* pFlags1, UINT* pFlags2) const {
     // Query some general information from DXGI, DXVK and Vulkan about the format
     const DXGI_VK_FORMAT_INFO fmtMapping = m_dxgiAdapter->LookupFormat(Format, DXGI_VK_FORMAT_MODE_ANY);
diff --git a/src/d3d11/d3d11_device.h b/src/d3d11/d3d11_device.h
index 2e6bc98df..817b3658d 100644
--- a/src/d3d11/d3d11_device.h
+++ b/src/d3d11/d3d11_device.h
@@ -13,6 +13,7 @@
 
 #include "../util/com/com_private_data.h"
 
+#include "d3d11_counter_buffer.h"
 #include "d3d11_initializer.h"
 #include "d3d11_interfaces.h"
 #include "d3d11_options.h"
@@ -379,9 +380,10 @@ namespace dxvk {
     DxvkCsChunkPool                 m_csChunkPool;
     
     D3D11Initializer*               m_initializer = nullptr;
-    D3D11UavCounterAllocator*       m_uavCounters = nullptr;
     D3D11ImmediateContext*          m_context     = nullptr;
     D3D10Device*                    m_d3d10Device = nullptr;
+
+    Rc<D3D11CounterBuffer>          m_uavCounters;
     
     D3D11StateObjectSet<D3D11BlendState>        m_bsStateObjects;
     D3D11StateObjectSet<D3D11DepthStencilState> m_dsStateObjects;
@@ -389,6 +391,8 @@ namespace dxvk {
     D3D11StateObjectSet<D3D11SamplerState>      m_samplerObjects;
     D3D11ShaderModuleSet                        m_shaderModules;
     
+    Rc<D3D11CounterBuffer> CreateUAVCounterBuffer();
+
     HRESULT CreateShaderModule(
             D3D11CommonShader*      pShaderModule,
       const void*                   pShaderBytecode,