git.sesse.net Git - casparcg/blob - accelerator/ogl/util/device.cpp

   1 /*\r
   2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>\r
   3 *\r
   4 * This file is part of CasparCG (www.casparcg.com).\r
   5 *\r
   6 * CasparCG is free software: you can redistribute it and/or modify\r
   7 * it under the terms of the GNU General Public License as published by\r
   8 * the Free Software Foundation, either version 3 of the License, or\r
   9 * (at your option) any later version.\r
  10 *\r
  11 * CasparCG is distributed in the hope that it will be useful,\r
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of\r
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the\r
  14 * GNU General Public License for more details.\r
  15 *\r
  16 * You should have received a copy of the GNU General Public License\r
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.\r
  18 *\r
  19 * Author: Robert Nagy, ronag89@gmail.com\r
  20 */\r
  21 \r
  22 // TODO: Smart GC\r
  23 \r
  24 #include "../../stdafx.h"\r
  25 \r
  26 #include "device.h"\r
  27 \r
  28 #include "buffer.h"\r
  29 #include "texture.h"\r
  30 #include "shader.h"\r
  31 \r
  32 #include <common/assert.h>\r
  33 #include <common/except.h>\r
  34 #include <common/concurrency/async.h>\r
  35 #include <common/gl/gl_check.h>\r
  36 #include <common/os/windows/windows.h>\r
  37 \r
  38 #include <boost/foreach.hpp>\r
  39 \r
  40 #include <gl/glew.h>\r
  41 \r
  42 #include <SFML/Window/Context.hpp>\r
  43 \r
  44 #include <tbb/concurrent_unordered_map.h>\r
  45 #include <tbb/concurrent_hash_map.h>\r
  46 #include <tbb/concurrent_queue.h>\r
  47 \r
  48 #include <boost/utility/declval.hpp>\r
  49 \r
  50 #include <array>\r
  51 #include <unordered_map>\r
  52 \r
  53 tbb::atomic<int> g_count = tbb::atomic<int>();\r
  54 \r
  55 namespace caspar { namespace accelerator { namespace ogl {\r
  56                 \r
  57 struct device::impl : public std::enable_shared_from_this<impl>\r
  58 {       \r
  59         static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");\r
  60 \r
  61         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_mapping_;\r
  62 \r
  63         std::unique_ptr<sf::Context> device_;\r
  64         std::unique_ptr<sf::Context> host_alloc_device_;\r
  65         \r
  66         std::array<tbb::concurrent_unordered_map<int, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 4>      device_pools_;\r
  67         std::array<tbb::concurrent_unordered_map<int, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2>       host_pools_;\r
  68         \r
  69         GLuint fbo_;\r
  70 \r
  71         executor& render_executor_;\r
  72         executor  alloc_executor_;\r
  73                                 \r
  74         impl(executor& executor) \r
  75                 : render_executor_(executor)\r
  76                 , alloc_executor_(L"OpenGL allocation context.")\r
  77         {\r
  78                 if(g_count++ > 1)\r
  79                         CASPAR_LOG(warning) << L"Multiple OGL devices.";\r
  80 \r
  81                 CASPAR_LOG(info) << L"Initializing OpenGL Device.";\r
  82                 \r
  83                 auto ctx1 = render_executor_.invoke([=]() -> HGLRC \r
  84                 {\r
  85                         device_.reset(new sf::Context());\r
  86                         device_->SetActive(true);               \r
  87                                                 \r
  88                         if (glewInit() != GLEW_OK)\r
  89                                 BOOST_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));\r
  90                 \r
  91                         if(!GLEW_VERSION_3_0)\r
  92                                 BOOST_THROW_EXCEPTION(not_supported() << msg_info("Your graphics card does not meet the minimum hardware requirements since it does not support OpenGL 3.0 or higher."));\r
  93         \r
  94                         glGenFramebuffers(1, &fbo_);                            \r
  95                         glBindFramebuffer(GL_FRAMEBUFFER, fbo_);\r
  96                         \r
  97                         auto ctx1 = wglGetCurrentContext();\r
  98                         \r
  99                         device_->SetActive(false);\r
 100 \r
 101                         return ctx1;\r
 102                 });\r
 103 \r
 104                 alloc_executor_.invoke([=]\r
 105                 {\r
 106                         host_alloc_device_.reset(new sf::Context());\r
 107                         host_alloc_device_->SetActive(true);    \r
 108                         auto ctx2 = wglGetCurrentContext();\r
 109 \r
 110                         if(!wglShareLists(ctx1, ctx2))\r
 111                                 BOOST_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to share OpenGL devices."));\r
 112                 });\r
 113 \r
 114                 render_executor_.invoke([=]\r
 115                 {               \r
 116                         device_->SetActive(true);\r
 117                 });\r
 118                 \r
 119                 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();\r
 120         }\r
 121 \r
 122         ~impl()\r
 123         {\r
 124                 alloc_executor_.invoke([=]\r
 125                 {\r
 126                         host_alloc_device_.reset();\r
 127                         BOOST_FOREACH(auto& pool, host_pools_)\r
 128                                 pool.clear();\r
 129                 });\r
 130 \r
 131                 render_executor_.invoke([=]\r
 132                 {\r
 133                         BOOST_FOREACH(auto& pool, device_pools_)\r
 134                                 pool.clear();\r
 135                         glDeleteFramebuffers(1, &fbo_);\r
 136 \r
 137                         device_.reset();\r
 138                 });\r
 139         }\r
 140                 \r
 141         std::wstring version()\r
 142         {       \r
 143                 try\r
 144                 {\r
 145                         return alloc_executor_.invoke([]\r
 146                         {\r
 147                                 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));\r
 148                         });     \r
 149                 }\r
 150                 catch(...)\r
 151                 {\r
 152                         return L"Not found";;\r
 153                 }\r
 154         }\r
 155                                                         \r
 156         spl::shared_ptr<texture> create_texture(int width, int height, int stride)\r
 157         {\r
 158                 CASPAR_VERIFY(stride > 0 && stride < 5);\r
 159                 CASPAR_VERIFY(width > 0 && height > 0);\r
 160                 \r
 161                 auto pool = &device_pools_[stride-1][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];\r
 162                 \r
 163                 std::shared_ptr<texture> buffer;\r
 164                 if(!pool->try_pop(buffer))              \r
 165                         buffer = spl::make_shared<texture>(width, height, stride);\r
 166         \r
 167                 return spl::shared_ptr<texture>(buffer.get(), [buffer, pool](texture*) mutable\r
 168                 {               \r
 169                         pool->push(buffer);     \r
 170                 });\r
 171         }\r
 172                 \r
 173         spl::shared_ptr<buffer> create_buffer(int size, buffer::usage usage)\r
 174         {\r
 175                 CASPAR_VERIFY(size > 0);\r
 176                 \r
 177                 auto pool = &host_pools_[usage.value()][size];\r
 178                 \r
 179                 std::shared_ptr<buffer> buf;\r
 180                 if(!pool->try_pop(buf)) \r
 181                 {\r
 182                         buf = alloc_executor_.invoke([&]\r
 183                         {\r
 184                                 return spl::make_shared<buffer>(size, usage);\r
 185                         });\r
 186                 }\r
 187                                 \r
 188                 auto ptr = buf->data();\r
 189                 auto self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.\r
 190 \r
 191                 auto on_release = [self, buf, ptr, usage, pool]() mutable\r
 192                 {               \r
 193                         if(usage == buffer::usage::write_only)                                  \r
 194                                 buf->map();                                     \r
 195                         else\r
 196                                 buf->unmap();\r
 197 \r
 198                         self->texture_mapping_.erase(buf.get());\r
 199 \r
 200                         pool->push(buf);\r
 201                 };\r
 202                 \r
 203                 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable\r
 204                 {\r
 205                         self->alloc_executor_.begin_invoke(on_release); \r
 206                 });\r
 207         }\r
 208 \r
 209         core::mutable_array create_array(int size)\r
 210         {               \r
 211                 auto buf = create_buffer(size, buffer::usage::write_only);\r
 212                 return core::mutable_array(buf->data(), buf->size(), buf);\r
 213         }\r
 214 \r
 215         boost::unique_future<spl::shared_ptr<texture>> copy_async(const core::mutable_array& source, int width, int height, int stride)\r
 216         {\r
 217                 auto buf = boost::any_cast<spl::shared_ptr<buffer>>(source.storage());\r
 218                                 \r
 219                 return render_executor_.begin_invoke([=]() -> spl::shared_ptr<texture>\r
 220                 {\r
 221                         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;\r
 222                         if(texture_mapping_.find(a, buf.get()))\r
 223                                 return spl::make_shared_ptr(a->second);\r
 224 \r
 225                         auto texture = create_texture(width, height, stride);\r
 226                         texture->copy_from(*buf);       \r
 227 \r
 228                         texture_mapping_.insert(std::make_pair(buf.get(), texture));\r
 229 \r
 230                         return texture;\r
 231 \r
 232                 }, task_priority::high_priority);\r
 233         }\r
 234 \r
 235         boost::unique_future<core::const_array> copy_async(const spl::shared_ptr<texture>& source)\r
 236         {\r
 237                 return fold(render_executor_.begin_invoke([=]() -> boost::shared_future<core::const_array>\r
 238                 {\r
 239                         auto buffer = create_buffer(source->size(), buffer::usage::read_only); \r
 240                         source->copy_to(*buffer);       \r
 241 \r
 242                         return make_shared(async(launch_policy::deferred, [=]() mutable -> core::const_array\r
 243                         {\r
 244                                 const auto& buf = buffer.get();\r
 245                                 if(!buf->data())\r
 246                                         alloc_executor_.invoke(std::bind(&buffer::map, std::ref(buf))); // Defer blocking "map" call until data is needed.\r
 247 \r
 248                                 return core::const_array(buf->data(), buf->size(), buffer);\r
 249                         }));\r
 250                 }, task_priority::high_priority));\r
 251         }\r
 252 };\r
 253 \r
 254 device::device() \r
 255         : executor_(L"OpenGL Rendering Context.")\r
 256         , impl_(new impl(executor_))\r
 257 {\r
 258 }\r
 259 device::~device(){}     \r
 260 spl::shared_ptr<texture>                                                        device::create_texture(int width, int height, int stride){return impl_->create_texture(width, height, stride);}\r
 261 core::mutable_array                                                                     device::create_array(int size){return impl_->create_array(size);}\r
 262 boost::unique_future<spl::shared_ptr<texture>>          device::copy_async(const core::mutable_array& source, int width, int height, int stride){return impl_->copy_async(source, width, height, stride);}\r
 263 boost::unique_future<core::const_array>                         device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}\r
 264 std::wstring                                                                            device::version(){return impl_->version();}\r
 265 \r
 266 \r
 267 }}}\r
 268 \r
 269 \r