From 191c9644af94371871712ed091d33a111de02e3e Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Sat, 26 Oct 2019 15:40:58 +0200
Subject: [PATCH] [d3d11] Allocate counter buffer per UAV

Might slightly increase memory overhead, however this allows
the backend to execute UAV counter updates more efficiently.
---
 src/d3d11/d3d11_device.cpp   | 26 --------------------------
 src/d3d11/d3d11_device.h     |  4 ----
 src/d3d11/d3d11_view_uav.cpp | 23 +++++++++++++++++++----
 src/d3d11/d3d11_view_uav.h   |  8 ++++++--
 4 files changed, 25 insertions(+), 36 deletions(-)

diff --git a/src/d3d11/d3d11_device.cpp b/src/d3d11/d3d11_device.cpp
index aadbe1187..2083a8a72 100644
--- a/src/d3d11/d3d11_device.cpp
+++ b/src/d3d11/d3d11_device.cpp
@@ -45,7 +45,6 @@ namespace dxvk {
     m_context     = new D3D11ImmediateContext(this, m_dxvkDevice);
     m_d3d10Device = new D3D10Device(this, m_context);
 
-    m_uavCounters = CreateUAVCounterBuffer();
     m_xfbCounters = CreateXFBCounterBuffer();
     m_predicates  = CreatePredicateBuffer();
   }
@@ -1970,31 +1969,6 @@ 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(uint32_t), 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);
-  }
-
-
   Rc<D3D11CounterBuffer> D3D11Device::CreateXFBCounterBuffer() {
     DxvkBufferCreateInfo xfbCounterInfo;
     xfbCounterInfo.size   = 4096 * sizeof(D3D11SOCounter);
diff --git a/src/d3d11/d3d11_device.h b/src/d3d11/d3d11_device.h
index c17f2ab82..8d24e2a0a 100644
--- a/src/d3d11/d3d11_device.h
+++ b/src/d3d11/d3d11_device.h
@@ -425,11 +425,9 @@ namespace dxvk {
       return m_d3d10Device;
     }
     
-    DxvkBufferSlice AllocUavCounterSlice() { return m_uavCounters->AllocSlice(); }
     DxvkBufferSlice AllocXfbCounterSlice() { return m_xfbCounters->AllocSlice(); }
     DxvkBufferSlice AllocPredicateSlice () { return m_predicates ->AllocSlice(); }
     
-    void FreeUavCounterSlice(const DxvkBufferSlice& Slice) { m_uavCounters->FreeSlice(Slice); }
     void FreeXfbCounterSlice(const DxvkBufferSlice& Slice) { m_xfbCounters->FreeSlice(Slice); }
     void FreePredicateSlice (const DxvkBufferSlice& Slice) { m_predicates ->FreeSlice(Slice); }
     
@@ -461,7 +459,6 @@ namespace dxvk {
     D3D11ImmediateContext*          m_context     = nullptr;
     D3D10Device*                    m_d3d10Device = nullptr;
 
-    Rc<D3D11CounterBuffer>          m_uavCounters;
     Rc<D3D11CounterBuffer>          m_xfbCounters;
     Rc<D3D11CounterBuffer>          m_predicates;
     
@@ -472,7 +469,6 @@ namespace dxvk {
     D3D11ShaderModuleSet                        m_shaderModules;
     D3D11CommandListAllocator                   m_commandListAllocator;
     
-    Rc<D3D11CounterBuffer> CreateUAVCounterBuffer();
     Rc<D3D11CounterBuffer> CreateXFBCounterBuffer();
     Rc<D3D11CounterBuffer> CreatePredicateBuffer();
 
diff --git a/src/d3d11/d3d11_view_uav.cpp b/src/d3d11/d3d11_view_uav.cpp
index 123d1e3b1..f392b5f48 100644
--- a/src/d3d11/d3d11_view_uav.cpp
+++ b/src/d3d11/d3d11_view_uav.cpp
@@ -43,7 +43,7 @@ namespace dxvk {
       }
       
       if (pDesc->Buffer.Flags & (D3D11_BUFFER_UAV_FLAG_APPEND | D3D11_BUFFER_UAV_FLAG_COUNTER))
-        m_counterSlice = pDevice->AllocUavCounterSlice();
+        m_counterBuffer = CreateCounterBuffer();
       
       // Populate view info struct
       m_info.Buffer.Offset = viewInfo.rangeOffset;
@@ -123,9 +123,6 @@ namespace dxvk {
   
   D3D11UnorderedAccessView::~D3D11UnorderedAccessView() {
     ResourceReleasePrivate(m_resource);
-
-    if (m_counterSlice.defined())
-      m_device->FreeUavCounterSlice(m_counterSlice);
   }
   
   
@@ -420,5 +417,23 @@ namespace dxvk {
     
     return S_OK;
   }
+
+
+  Rc<DxvkBuffer> D3D11UnorderedAccessView::CreateCounterBuffer() {
+    Rc<DxvkDevice> device = m_device->GetDXVKDevice();
+
+    DxvkBufferCreateInfo info;
+    info.size   = sizeof(uint32_t);
+    info.usage  = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT
+                | VK_BUFFER_USAGE_TRANSFER_DST_BIT
+                | VK_BUFFER_USAGE_TRANSFER_SRC_BIT;
+    info.stages = VK_PIPELINE_STAGE_TRANSFER_BIT
+                | device->getShaderPipelineStages();
+    info.access = VK_ACCESS_TRANSFER_WRITE_BIT
+                | VK_ACCESS_TRANSFER_READ_BIT
+                | VK_ACCESS_SHADER_WRITE_BIT
+                | VK_ACCESS_SHADER_READ_BIT;
+    return device->createBuffer(info, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
+  }
   
 }
diff --git a/src/d3d11/d3d11_view_uav.h b/src/d3d11/d3d11_view_uav.h
index bd2532e88..c54c35650 100644
--- a/src/d3d11/d3d11_view_uav.h
+++ b/src/d3d11/d3d11_view_uav.h
@@ -60,7 +60,9 @@ namespace dxvk {
     }
     
     DxvkBufferSlice GetCounterSlice() const {
-      return m_counterSlice;
+      return m_counterBuffer != nullptr
+        ? DxvkBufferSlice(m_counterBuffer)
+        : DxvkBufferSlice();
     }
     
     static HRESULT GetDescFromResource(
@@ -82,7 +84,9 @@ namespace dxvk {
     D3D11_VK_VIEW_INFO                m_info;
     Rc<DxvkBufferView>                m_bufferView;
     Rc<DxvkImageView>                 m_imageView;
-    DxvkBufferSlice                   m_counterSlice;
+    Rc<DxvkBuffer>                    m_counterBuffer;
+
+    Rc<DxvkBuffer> CreateCounterBuffer();
     
   };