From e0fbfdf0e23c87d82ca84ddb864fb328ea871295 Mon Sep 17 00:00:00 2001
From: Philip Rebohle <philip.rebohle@tu-dortmund.de>
Date: Fri, 2 Mar 2018 10:31:08 +0100
Subject: [PATCH] [dxvk] Use linked list for CS chunks

Improves memory efficiency and CPU overhead of the CSMT implementation
when the average number of bytes per command entry is less than 64 bytes.
---
 src/dxvk/dxvk_cs.cpp | 22 +++++++++++++++++-----
 src/dxvk/dxvk_cs.h   | 44 ++++++++++++++++++++++++++++++++++++--------
 2 files changed, 53 insertions(+), 13 deletions(-)

diff --git a/src/dxvk/dxvk_cs.cpp b/src/dxvk/dxvk_cs.cpp
index 955c2d48..24ad6a85 100644
--- a/src/dxvk/dxvk_cs.cpp
+++ b/src/dxvk/dxvk_cs.cpp
@@ -8,19 +8,31 @@ namespace dxvk {
   
   
   DxvkCsChunk::~DxvkCsChunk() {
-    for (size_t i = 0; i < m_commandCount; i++)
-      m_commandList[i]->~DxvkCsCmd();
+    auto cmd = m_head;
+    
+    while (cmd != nullptr) {
+      auto next = cmd->next();
+      cmd->~DxvkCsCmd();
+      cmd = next;
+    }
   }
   
   
   void DxvkCsChunk::executeAll(DxvkContext* ctx) {
-    for (size_t i = 0; i < m_commandCount; i++) {
-      m_commandList[i]->exec(ctx);
-      m_commandList[i]->~DxvkCsCmd();
+    auto cmd = m_head;
+    
+    while (cmd != nullptr) {
+      auto next = cmd->next();
+      cmd->exec(ctx);
+      cmd->~DxvkCsCmd();
+      cmd = next;
     }
     
     m_commandCount  = 0;
     m_commandOffset = 0;
+    
+    m_head = nullptr;
+    m_tail = nullptr;
   }
   
   
diff --git a/src/dxvk/dxvk_cs.h b/src/dxvk/dxvk_cs.h
index fec6194b..13c800ad 100644
--- a/src/dxvk/dxvk_cs.h
+++ b/src/dxvk/dxvk_cs.h
@@ -22,12 +22,35 @@ namespace dxvk {
     
     virtual ~DxvkCsCmd() { }
     
+    /**
+     * \brief Retrieves next command in a command chain
+     * 
+     * This can be used to quickly iterate
+     * over commands within a chunk.
+     * \returns Pointer the next command
+     */
+    DxvkCsCmd* next() const {
+      return m_next;
+    }
+    
+    /**
+     * \brief Sets next command in a command chain
+     * \param [in] next Next command
+     */
+    void setNext(DxvkCsCmd* next) {
+      m_next = next;
+    }
+    
     /**
      * \brief Executes embedded commands
      * \param [in] ctx The target context
      */
     virtual void exec(DxvkContext* ctx) const = 0;
     
+  private:
+    
+    DxvkCsCmd* m_next = nullptr;
+    
   };
   
   
@@ -65,8 +88,7 @@ namespace dxvk {
    * Stores a list of commands.
    */
   class DxvkCsChunk : public RcObject {
-    constexpr static size_t MaxCommands  = 1024;
-    constexpr static size_t MaxBlockSize = 64 * MaxCommands;
+    constexpr static size_t MaxBlockSize = 65536;
   public:
     
     DxvkCsChunk();
@@ -96,13 +118,18 @@ namespace dxvk {
     bool push(T& command) {
       using FuncType = DxvkCsTypedCmd<T>;
       
-      if (m_commandCount >= MaxCommands
-       || m_commandOffset + sizeof(FuncType) > MaxBlockSize)
+      if (m_commandOffset + sizeof(FuncType) > MaxBlockSize)
         return false;
       
-      m_commandList[m_commandCount] =
-        new (m_data + m_commandOffset)
-          FuncType(std::move(command));
+      DxvkCsCmd* tail = m_tail;
+      
+      m_tail = new (m_data + m_commandOffset)
+        FuncType(std::move(command));
+      
+      if (tail != nullptr)
+        tail->setNext(m_tail);
+      else
+        m_head = m_tail;
       
       m_commandCount  += 1;
       m_commandOffset += sizeof(FuncType);
@@ -123,7 +150,8 @@ namespace dxvk {
     size_t m_commandCount  = 0;
     size_t m_commandOffset = 0;
     
-    std::array<DxvkCsCmd*, MaxCommands> m_commandList;
+    DxvkCsCmd* m_head = nullptr;
+    DxvkCsCmd* m_tail = nullptr;
     
     alignas(64)
     char m_data[MaxBlockSize];