Bring back global lithium pools

why?
cause sorting around the global driver pool would basically use more cpu power then just copy data sequences into the right order
And sorting data references costs 0 performance at all

CMakeLists.txt

@@ -52,6 +52,7 @@ set(PD_SOURCES
     source/lithium/drawlist.cpp
     source/lithium/font.cpp
     source/lithium/math.cpp
+    source/lithium/pools.cpp
 
     # UI7
     source/ui7/io.cpp

include/pd/drivers/gfx.hpp

@@ -2,7 +2,8 @@
 
 #include <pd/core/mat.hpp>
 #include <pd/drivers/interface.hpp>
-#include <pd/lithium/command.hpp>
+#include <pd/lithium/drawlist.hpp>
+#include <pd/lithium/pools.hpp>
 #include <pd/lithium/texture.hpp>
 
 using PDGfxBackendFlags = PD::u32;
@@ -37,7 +38,7 @@ class PD_API GfxDriver : public DriverInterface {
     return Li::Texture();
   }
   virtual void DeleteTexture(const Li::Texture& tex) {}
-  virtual void Draw(const Pool<Li::Command>& commands) {}
+  virtual void Draw(const Li::Drawlist& commands) {}
   Li::Texture::Ptr GetWhiteTexture() { return &pWhite; }
   PDGfxBackendFlags GetFlags() { return Flags; }
 
@@ -110,28 +111,67 @@ class GfxDriverBase : public GfxDriver {
     pWhite = LoadTexture(img, 16, 16);
   }
 
-  void Draw(const Pool<Li::Command>& commands) override {
+  void Draw(const Li::Drawlist& dl) override {
+    const auto& commands = dl.Data();
+    if (commands.size() == 0) return;
     pCountCommands += commands.size();
-    size_t index = 0;
+
+    size_t vtotal = dl.GetNumVertices();
+    size_t itotal = dl.GetNumIndices();
+
+    size_t start_vtx = pVtxPool.size();
+    pVtxPool.Allocate(vtotal);
+    CurrentVertex += vtotal;
+
+    size_t start_idx = pIdxPool.size();
+    pIdxPool.Allocate(itotal);
+    CurrentIndex += itotal;
+
+    size_t current_vtx = start_vtx;
+    size_t current_idx = start_idx;
+
+    const auto& vpool = Li::GetVertexPool();
+    const auto& ipool = Li::GetIndexPool();
+    /** Build Pools */
+    for (size_t i = 0; i < commands.size(); i++) {
+      const auto& cmd = commands[i];
+      if (cmd.VertexCount > 0) {
+        std::memcpy(pVtxPool.begin() + current_vtx,
+                    vpool.begin() + cmd.FirstVertex,
+                    cmd.VertexCount * sizeof(Li::Vertex));
+      }
+      for (size_t idx = 0; idx < cmd.IndexCount; idx++) {
+        u16 local_idx = ipool[cmd.FirstIndex + idx];
+        pIdxPool[current_idx + idx] = static_cast<u16>(current_vtx + local_idx);
+      }
+      current_vtx += cmd.VertexCount;
+      current_idx += cmd.IndexCount;
+    }
+
     UploadPools();
+
+    size_t index = 0;
+    size_t ioff = start_idx;
+
     while (index < commands.size()) {
       CurrentTex = commands[index].Tex;
       CurrentTexIsSDF = commands[index].SDF;
       if (!CurrentTex) {
         CurrentTex = pWhite.GetID();
       }
-      size_t startidx = commands[index].FirstIndex;
       size_t num_indices = 0;
       while (index < commands.size() &&
              CurrentTexIsSDF == commands[index].SDF &&
              (CurrentTex == commands[index].Tex ||
-              (CurrentTex == pWhite.GetID() && commands[index].Tex == 0)) &&
+              (CurrentTex == pWhite.GetID() && commands[index].Tex == 0))) {
-             commands[index].FirstIndex == startidx + num_indices) {
         num_indices += commands[index].IndexCount;
         index++;
       }
-      Submit(num_indices, startidx);
+      if (num_indices > 0) {
-      pCountDrawcalls++;
+        Submit(num_indices, ioff);
+        ioff += num_indices;
+        pCountDrawcalls++;
+      }
     }
   }
 
@@ -211,9 +251,7 @@ class PD_API Gfx {
   static void SetViewPort(const ivec2& vp) { driver->SetViewPort(vp); }
   static void SetViewPort(int w, int h) { driver->SetViewPort(w, h); }
   static void Reset() { driver->Reset(); }
-  static void Draw(const Pool<Li::Command>& commands) {
+  static void Draw(const Li::Drawlist& dl) { driver->Draw(dl); }
-    driver->Draw(commands);
-  }
   static Li::Texture LoadTexture(const std::vector<PD::u8>& pixels, int w,
                                  int h,
                                  TextureFormat type = TextureFormat::RGBA32,

include/pd/lithium/lithium.hpp

@@ -4,4 +4,5 @@
 #include <pd/lithium/drawlist.hpp>
 #include <pd/lithium/font.hpp>
 #include <pd/lithium/math.hpp>
+#include <pd/lithium/pools.hpp>
 #include <pd/lithium/vertex.hpp>

include/pd/lithium/pools.hpp

@@ -0,0 +1,12 @@
+#pragma once
+
+#include <pd/core/core.hpp>
+#include <pd/lithium/vertex.hpp>
+
+namespace PD {
+namespace Li {
+PD_API PD::Pool<Vertex>& GetVertexPool();
+PD_API PD::Pool<u16>& GetIndexPool();
+PD_API void ResetPools();
+}  // namespace Li
+}  // namespace PD

source/lithium/command.cpp

@@ -1,23 +1,48 @@
 #include <pd/drivers/gfx.hpp>
 #include <pd/lithium/command.hpp>
+#include <pd/lithium/pools.hpp>
 
 namespace PD {
 namespace Li {
 
 void Command::Reserve(size_t vtx, size_t idx) {
+  auto& vpool = GetVertexPool();
+  auto& ipool = GetIndexPool();
   if (VertexCountMax == 0) {
-    FirstVertex = Gfx::AllocateVertices(vtx, (PD::ptr)this);
+    FirstVertex = vpool.size();
+    vpool.Allocate(vtx);
     VertexCountMax = vtx;
   } else {
-    Gfx::ExpandVertices(vtx, (PD::ptr)this);
+    if (vpool.size() == FirstVertex + VertexCountMax) {
-    VertexCountMax += vtx;
+      vpool.Allocate(vtx);
+      VertexCountMax += vtx;
+    } else {
+      size_t tmp = FirstVertex;
+      FirstVertex = vpool.size();
+      vpool.Allocate(VertexCountMax + vtx);
+      for (size_t i = 0; i < VertexCount; i++) {
+        vpool[FirstVertex + i] = vpool[tmp + i];
+      }
+      VertexCountMax += vtx;
+    }
   }
   if (IndexCountMax == 0) {
-    FirstIndex = Gfx::AllocateIndices(idx, (PD::ptr)this);
+    FirstIndex = ipool.size();
+    ipool.Allocate(idx);
     IndexCountMax = idx;
   } else {
-    Gfx::ExpandIndices(idx, (PD::ptr)this);
+    if (ipool.size() == FirstIndex + IndexCountMax) {
-    IndexCountMax += idx;
+      ipool.Allocate(idx);
+      IndexCountMax += idx;
+    } else {
+      size_t tmp = FirstIndex;
+      FirstIndex = ipool.size();
+      ipool.Allocate(IndexCountMax + idx);
+      for (size_t i = 0; i < IndexCount; i++) {
+        ipool[FirstIndex + i] = ipool[tmp + i];
+      }
+      IndexCountMax += idx;
+    }
   }
 }
 
@@ -34,22 +59,25 @@ void Command::Reset() {
 }
 
 Command& Command::Add(const Vertex& vtx) {
-  if (VertexCount <= VertexCountMax)
+  if (VertexCount < VertexCountMax) {
-    Gfx::PutVertex(FirstVertex + VertexCount++, vtx, (PD::ptr)this);
+    GetVertexPool()[FirstVertex + VertexCount++] = vtx;
+  }
   return *this;
 }
 Command& Command::Add(u16 idx) {
-  if (IndexCount <= IndexCountMax)
+  if (IndexCount < IndexCountMax) {
-    Gfx::PutIndex(FirstIndex + IndexCount++, FirstVertex + VertexCount + idx,
+    GetIndexPool()[FirstIndex + IndexCount++] =
-                  (PD::ptr)this);
+        static_cast<u16>(VertexCount + idx);
+  }
   return *this;
 }
 Command& Command::Add(u16 a, u16 b, u16 c) {
   if (IndexCount + 3 <= IndexCountMax) {
+    auto& ip = GetIndexPool();
     size_t idx = FirstIndex + IndexCount;
-    Gfx::PutIndex(idx + 0, FirstVertex + VertexCount + a, (PD::ptr)this);
+    ip[idx + 0] = static_cast<u16>(VertexCount + a);
-    Gfx::PutIndex(idx + 1, FirstVertex + VertexCount + b, (PD::ptr)this);
+    ip[idx + 1] = static_cast<u16>(VertexCount + b);
-    Gfx::PutIndex(idx + 2, FirstVertex + VertexCount + c, (PD::ptr)this);
+    ip[idx + 2] = static_cast<u16>(VertexCount + c);
     IndexCount += 3;
   }
   return *this;

source/lithium/drawlist.cpp

@@ -1,3 +1,4 @@
+#include <algorithm>
 #include <iostream>
 #include <pd/drivers/gfx.hpp>
 #include <pd/lithium/drawlist.hpp>
@@ -19,19 +20,26 @@ PD_API void Drawlist::Copy(Drawlist& other) {
   pCommands.AppendCopy(other.pCommands);
 }
 
-PD_API void Drawlist::Optimize() {}
+PD_API void Drawlist::Optimize() {
+  if (pCommands.size() <= 1) return;
+  std::stable_sort(pCommands.begin(), pCommands.end(),
+                   [](const Command& a, const Command& b) {
+                     if (a.Layer != b.Layer) return a.Layer < b.Layer;
+                     if (a.SDF != b.SDF) return a.SDF < b.SDF;
+                     return a.Tex < b.Tex;
+                   });
+}
 
 PD_API void Drawlist::Clear() {
   UnbindTexture();
   pPath.ResetFast();
-  pVertices.ResetFast();
+  pCommands.NoReset();
-  pIndices.ResetFast();
-  pCommands.ResetFast();
 }
 
 /** Command Allocation */
 PD_API Command& Drawlist::NewCommand() {
   auto cmd = pCommands.Allocate(1);
+  cmd->Reset();
   cmd->Tex = pCurrentTexture.GetID();
   return *cmd;
 }

source/lithium/pools.cpp

@@ -0,0 +1,16 @@
+#include <pd/lithium/pools.hpp>
+
+namespace PD {
+namespace Li {
+static PD::Pool<Vertex> gVertexPool;
+static PD::Pool<u16> gIndexPool;
+
+PD_API PD::Pool<Vertex>& GetVertexPool() { return gVertexPool; }
+PD_API PD::Pool<u16>& GetIndexPool() { return gIndexPool; }
+
+PD_API void ResetPools() {
+  gVertexPool.NoReset();
+  gIndexPool.NoReset();
+}
+}  // namespace Li
+}  // namespace PD

tests/gfx/source/main.cpp

@@ -217,6 +217,7 @@ int main(int argc, char** argv) {
     PD::TT::Beg("MainRaw");
     PD::Hid::Update();
     PD::Gfx::NewFrame();
+    PD::Li::ResetPools();
     // app.Update(pOs->GetViewport(), pList);
     pList.SetFontscale(0.7);
     if (PD::Hid::IsEvent(PD::Hid::Event::Down, PD::Hid::Gamepad::CPLeft |
@@ -297,7 +298,6 @@ int main(int argc, char** argv) {
           "#ff00ff");
     }
     pList.SetFont(&font);
-#ifndef __3DS__
     PD::TT::Beg("BuildUI7Menus");
     if (auto m = ui7.BeginMenu("Test")) {
       m->Label("Hello World!");
@@ -318,11 +318,10 @@ int main(int argc, char** argv) {
     PD::TT::Beg("UI7::Context::Update");
     ui7.Update();
     PD::TT::End("UI7::Context::Update");
-#endif
     PD::Gfx::Reset();
     PD::TT::Beg("PD::Gfx::Draw");
-    PD::Gfx::Draw(pList);
     PD::Gfx::Draw(ui7.GetDrawData());
+    PD::Gfx::Draw(pList);
     PD::TT::End("PD::Gfx::Draw");
     pList.Clear();
     PD::TT::End("MainRaw");

tests/gfx/source/os/horizon-ctr.cpp

@@ -33,7 +33,7 @@ void HorizonCtr::Init() {
   romfsInit();
   osSetSpeedupEnable(true);
   gfxInitDefault();
-  consoleInit(GFX_BOTTOM, nullptr);
+  // consoleInit(GFX_BOTTOM, nullptr);
   C3D_Init(C3D_DEFAULT_CMDBUF_SIZE);
   impl->Top =
       C3D_RenderTargetCreate(240, 400, GPU_RB_RGBA8, GPU_RB_DEPTH24_STENCIL8);
@@ -63,12 +63,12 @@ bool HorizonCtr::Mainloop() {
   pViewPort = ivec2(400, 240);
   hidScanInput();
   bool _kill = hidKeysUp() & KEY_START;
+  C3D_FrameBegin(C3D_FRAME_SYNCDRAW);
   return aptMainLoop() && !_kill;
 }
 
 void HorizonCtr::ClearViewPort() {
   PD::Gfx::SetViewPort(pViewPort);
-  C3D_FrameBegin(C3D_FRAME_SYNCDRAW);
   C3D_FrameDrawOn(impl->Top);
   C3D_RenderTargetClear(impl->Top, C3D_CLEAR_ALL, PD::Color(25, 25, 25, 25), 0);
 }