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/future.h>\r
  35 #include <common/array.h>\r
  36 #include <common/gl/gl_check.h>\r
  37 #include <common/os/windows/windows.h>\r
  38 \r
  39 #include <boost/foreach.hpp>\r
  40 \r
  41 #include <gl/glew.h>\r
  42 \r
  43 #include <SFML/Window/Context.hpp>\r
  44 \r
  45 #include <tbb/concurrent_unordered_map.h>\r
  46 #include <tbb/concurrent_hash_map.h>\r
  47 #include <tbb/concurrent_queue.h>\r
  48 \r
  49 #include <boost/utility/declval.hpp>\r
  50 \r
  51 #include <array>\r
  52 #include <unordered_map>\r
  53 \r
  54 #include <asmlib.h>\r
  55 #include <tbb/parallel_for.h>\r
  56 \r
  57 tbb::atomic<int> g_count = tbb::atomic<int>();\r
  58 \r
  59 namespace caspar { namespace accelerator { namespace ogl {\r
  60                 \r
  61 struct device::impl : public std::enable_shared_from_this<impl>\r
  62 {       \r
  63         static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");\r
  64 \r
  65         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_cache_;\r
  66 \r
  67         std::unique_ptr<sf::Context> device_;\r
  68         std::unique_ptr<sf::Context> host_alloc_device_;\r
  69         \r
  70         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 4>      device_pools_;\r
  71         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2>       host_pools_;\r
  72         \r
  73         GLuint fbo_;\r
  74 \r
  75         executor& render_executor_;\r
  76         executor  alloc_executor_;\r
  77                                 \r
  78         impl(executor& executor) \r
  79                 : render_executor_(executor)\r
  80                 , alloc_executor_(L"OpenGL allocation context.")\r
  81         {\r
  82                 if(g_count++ > 1)\r
  83                         CASPAR_LOG(warning) << L"Multiple OGL devices.";\r
  84 \r
  85                 CASPAR_LOG(info) << L"Initializing OpenGL Device.";\r
  86                 \r
  87                 auto ctx1 = render_executor_.invoke([=]() -> HGLRC \r
  88                 {\r
  89                         device_.reset(new sf::Context());\r
  90                         device_->SetActive(true);               \r
  91                                                 \r
  92                         if (glewInit() != GLEW_OK)\r
  93                                 BOOST_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));\r
  94                 \r
  95                         if(!GLEW_VERSION_3_0)\r
  96                                 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
  97         \r
  98                         glGenFramebuffers(1, &fbo_);                            \r
  99                         glBindFramebuffer(GL_FRAMEBUFFER, fbo_);\r
 100                         \r
 101                         auto ctx1 = wglGetCurrentContext();\r
 102                         \r
 103                         device_->SetActive(false);\r
 104 \r
 105                         return ctx1;\r
 106                 });\r
 107 \r
 108                 alloc_executor_.invoke([=]\r
 109                 {\r
 110                         host_alloc_device_.reset(new sf::Context());\r
 111                         host_alloc_device_->SetActive(true);    \r
 112                         auto ctx2 = wglGetCurrentContext();\r
 113 \r
 114                         if(!wglShareLists(ctx1, ctx2))\r
 115                                 BOOST_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to share OpenGL devices."));\r
 116                 });\r
 117 \r
 118                 render_executor_.invoke([=]\r
 119                 {               \r
 120                         device_->SetActive(true);\r
 121                 });\r
 122                 \r
 123                 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();\r
 124         }\r
 125 \r
 126         ~impl()\r
 127         {\r
 128                 alloc_executor_.invoke([=]\r
 129                 {\r
 130                         host_alloc_device_.reset();\r
 131                         BOOST_FOREACH(auto& pool, host_pools_)\r
 132                                 pool.clear();\r
 133                 });\r
 134 \r
 135                 render_executor_.invoke([=]\r
 136                 {\r
 137                         BOOST_FOREACH(auto& pool, device_pools_)\r
 138                                 pool.clear();\r
 139                         glDeleteFramebuffers(1, &fbo_);\r
 140 \r
 141                         device_.reset();\r
 142                 });\r
 143         }\r
 144                 \r
 145         std::wstring version()\r
 146         {       \r
 147                 try\r
 148                 {\r
 149                         return alloc_executor_.invoke([]\r
 150                         {\r
 151                                 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));\r
 152                         });     \r
 153                 }\r
 154                 catch(...)\r
 155                 {\r
 156                         return L"Not found";;\r
 157                 }\r
 158         }\r
 159                                                         \r
 160         spl::shared_ptr<texture> create_texture(int width, int height, int stride, bool clear = false)\r
 161         {\r
 162                 CASPAR_VERIFY(stride > 0 && stride < 5);\r
 163                 CASPAR_VERIFY(width > 0 && height > 0);\r
 164 \r
 165                 if(!render_executor_.is_current())\r
 166                         BOOST_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));\r
 167                                         \r
 168                 auto pool = &device_pools_[stride-1][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];\r
 169                 \r
 170                 std::shared_ptr<texture> buffer;\r
 171                 if(!pool->try_pop(buffer))              \r
 172                         buffer = spl::make_shared<texture>(width, height, stride);\r
 173         \r
 174                 if(clear)\r
 175                         buffer->clear();\r
 176 \r
 177                 return spl::shared_ptr<texture>(buffer.get(), [buffer, pool](texture*) mutable\r
 178                 {               \r
 179                         pool->push(buffer);     \r
 180                 });\r
 181         }\r
 182                 \r
 183         spl::shared_ptr<buffer> create_buffer(std::size_t size, buffer::usage usage)\r
 184         {\r
 185                 CASPAR_VERIFY(size > 0);\r
 186                 \r
 187                 auto pool = &host_pools_[usage.value()][size];\r
 188                 \r
 189                 std::shared_ptr<buffer> buf;\r
 190                 if(!pool->try_pop(buf)) \r
 191                 {\r
 192                         buf = alloc_executor_.invoke([&]\r
 193                         {\r
 194                                 return spl::make_shared<buffer>(size, usage);\r
 195                         });\r
 196                 }\r
 197                 \r
 198                 auto self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.\r
 199                 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable\r
 200                 {       \r
 201                         texture_cache_.erase(buf.get());\r
 202                         pool->push(buf);\r
 203                 });\r
 204         }\r
 205 \r
 206         array<std::uint8_t> create_array(std::size_t size)\r
 207         {               \r
 208                 auto buf = create_buffer(size, buffer::usage::write_only);\r
 209                 return array<std::uint8_t>(buf->data(), buf->size(), false, buf);\r
 210         }\r
 211 \r
 212         template<typename T>\r
 213         std::shared_ptr<buffer> copy_to_buf(const T& source)\r
 214         {\r
 215                 std::shared_ptr<buffer> buf;\r
 216 \r
 217                 auto tmp = source.storage<spl::shared_ptr<buffer>>();\r
 218                 if(tmp)\r
 219                         buf = *tmp;\r
 220                 else\r
 221                 {                       \r
 222                         buf = create_buffer(source.size(), buffer::usage::write_only);\r
 223                         tbb::parallel_for(tbb::blocked_range<std::size_t>(0, source.size()), [&](const tbb::blocked_range<std::size_t>& r)\r
 224                         {\r
 225                                 A_memcpy(buf->data() + r.begin(), source.data() + r.begin(), r.size());\r
 226                         });\r
 227                 }\r
 228 \r
 229                 return buf;\r
 230         }\r
 231 \r
 232         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<const std::uint8_t>& source, int width, int height, int stride)\r
 233         {\r
 234                 std::shared_ptr<buffer> buf = copy_to_buf(source);\r
 235                                 \r
 236                 return render_executor_.begin_invoke([=]() -> spl::shared_ptr<texture>\r
 237                 {\r
 238                         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;\r
 239                         if(texture_cache_.find(a, buf.get()))\r
 240                                 return spl::make_shared_ptr(a->second);\r
 241 \r
 242                         auto texture = create_texture(width, height, stride);\r
 243                         texture->copy_from(*buf);       \r
 244 \r
 245                         texture_cache_.insert(std::make_pair(buf.get(), texture));\r
 246 \r
 247                         return texture;\r
 248                 }, task_priority::high_priority);\r
 249         }\r
 250         \r
 251         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<std::uint8_t>& source, int width, int height, int stride)\r
 252         {\r
 253                 std::shared_ptr<buffer> buf = copy_to_buf(source);\r
 254 \r
 255                 return render_executor_.begin_invoke([=]() -> spl::shared_ptr<texture>\r
 256                 {\r
 257                         auto texture = create_texture(width, height, stride, false);\r
 258                         texture->copy_from(*buf);                               \r
 259                         return texture;\r
 260                 }, task_priority::high_priority);\r
 261         }\r
 262 \r
 263         boost::unique_future<array<const std::uint8_t>> copy_async(const spl::shared_ptr<texture>& source)\r
 264         {\r
 265                 if(!render_executor_.is_current())\r
 266                         BOOST_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));\r
 267 \r
 268                 auto buffer = create_buffer(source->size(), buffer::usage::read_only); \r
 269                 source->copy_to(*buffer);       \r
 270 \r
 271                 auto self = shared_from_this();\r
 272                 return async(launch::deferred, [self, buffer]() mutable -> array<const std::uint8_t>\r
 273                 {\r
 274                         self->alloc_executor_.invoke(std::bind(&buffer::map, std::ref(buffer))); // Defer blocking "map" call until data is needed.\r
 275                         return array<const std::uint8_t>(buffer->data(), buffer->size(), true, buffer);\r
 276                 });\r
 277         }\r
 278 };\r
 279 \r
 280 device::device() \r
 281         : executor_(L"OpenGL Rendering Context.")\r
 282         , impl_(new impl(executor_)){}\r
 283 device::~device(){}\r
 284 spl::shared_ptr<texture>                                                        device::create_texture(int width, int height, int stride){return impl_->create_texture(width, height, stride, true);}\r
 285 array<std::uint8_t>                                                                     device::create_array(int size){return impl_->create_array(size);}\r
 286 boost::unique_future<spl::shared_ptr<texture>>          device::copy_async(const array<const std::uint8_t>& source, int width, int height, int stride){return impl_->copy_async(source, width, height, stride);}\r
 287 boost::unique_future<spl::shared_ptr<texture>>          device::copy_async(const array<std::uint8_t>& source, int width, int height, int stride){return impl_->copy_async(source, width, height, stride);}\r
 288 boost::unique_future<array<const std::uint8_t>>         device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}\r
 289 std::wstring                                                                            device::version() const{return impl_->version();}\r
 290 \r
 291 \r
 292 }}}\r
 293 \r
 294 \r