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 namespace caspar { namespace accelerator { namespace ogl {\r
  58                 \r
  59 struct device::impl : public std::enable_shared_from_this<impl>\r
  60 {       \r
  61         static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");\r
  62 \r
  63         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_cache_;\r
  64 \r
  65         std::unique_ptr<sf::Context> device_;\r
  66         \r
  67         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 4>      device_pools_;\r
  68         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2>       host_pools_;\r
  69         \r
  70         GLuint fbo_;\r
  71 \r
  72         executor& executor_;\r
  73                                 \r
  74         impl(executor& executor) \r
  75                 : executor_(executor)\r
  76         {\r
  77                 executor_.set_capacity(256);\r
  78 \r
  79                 CASPAR_LOG(info) << L"Initializing OpenGL Device.";\r
  80                 \r
  81                 executor_.invoke([=]\r
  82                 {\r
  83                         device_.reset(new sf::Context());\r
  84                         device_->SetActive(true);               \r
  85                                                 \r
  86                         if (glewInit() != GLEW_OK)\r
  87                                 CASPAR_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));\r
  88                 \r
  89                         if(!GLEW_VERSION_3_0)\r
  90                                 CASPAR_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
  91         \r
  92                         glGenFramebuffers(1, &fbo_);                            \r
  93                         glBindFramebuffer(GL_FRAMEBUFFER, fbo_);\r
  94                 });\r
  95                                 \r
  96                 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();\r
  97         }\r
  98 \r
  99         ~impl()\r
 100         {\r
 101                 executor_.invoke([=]\r
 102                 {\r
 103                         BOOST_FOREACH(auto& pool, device_pools_)\r
 104                                 pool.clear();\r
 105                         glDeleteFramebuffers(1, &fbo_);\r
 106 \r
 107                         device_.reset();\r
 108                 });\r
 109         }\r
 110                 \r
 111         std::wstring version()\r
 112         {       \r
 113                 try\r
 114                 {\r
 115                         return executor_.invoke([]\r
 116                         {\r
 117                                 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));\r
 118                         });     \r
 119                 }\r
 120                 catch(...)\r
 121                 {\r
 122                         return L"Not found";;\r
 123                 }\r
 124         }\r
 125                                                         \r
 126         spl::shared_ptr<texture> create_texture(int width, int height, int stride, bool clear = false)\r
 127         {\r
 128                 CASPAR_VERIFY(stride > 0 && stride < 5);\r
 129                 CASPAR_VERIFY(width > 0 && height > 0);\r
 130 \r
 131                 if(!executor_.is_current())\r
 132                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));\r
 133                                         \r
 134                 auto pool = &device_pools_[stride-1][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];\r
 135                 \r
 136                 std::shared_ptr<texture> tex;\r
 137                 if(!pool->try_pop(tex))         \r
 138                         tex = spl::make_shared<texture>(width, height, stride);\r
 139         \r
 140                 if(clear)\r
 141                         tex->clear();\r
 142 \r
 143                 return spl::shared_ptr<texture>(tex.get(), [tex, pool](texture*) mutable\r
 144                 {               \r
 145                         pool->push(tex);        \r
 146                 });\r
 147         }\r
 148                 \r
 149         spl::shared_ptr<buffer> create_buffer(std::size_t size, buffer::usage usage)\r
 150         {\r
 151                 CASPAR_VERIFY(size > 0);\r
 152                 \r
 153                 auto pool = &host_pools_[usage.value()][size];\r
 154                 \r
 155                 std::shared_ptr<buffer> buf;\r
 156                 if(!pool->try_pop(buf)) \r
 157                 {\r
 158                         buf = executor_.invoke([&]\r
 159                         {\r
 160                                 return spl::make_shared<buffer>(size, usage);\r
 161                         }, task_priority::high_priority);\r
 162                 }\r
 163                 \r
 164                 auto self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.\r
 165                 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable\r
 166                 {       \r
 167                         texture_cache_.erase(buf.get());\r
 168                         pool->push(buf);\r
 169                 });\r
 170         }\r
 171 \r
 172         array<std::uint8_t> create_array(std::size_t size)\r
 173         {               \r
 174                 auto buf = create_buffer(size, buffer::usage::write_only);\r
 175                 return array<std::uint8_t>(buf->data(), buf->size(), false, buf);\r
 176         }\r
 177 \r
 178         template<typename T>\r
 179         std::shared_ptr<buffer> copy_to_buf(const T& source)\r
 180         {\r
 181                 std::shared_ptr<buffer> buf;\r
 182 \r
 183                 auto tmp = source.storage<spl::shared_ptr<buffer>>();\r
 184                 if(tmp)\r
 185                         buf = *tmp;\r
 186                 else\r
 187                 {                       \r
 188                         buf = create_buffer(source.size(), buffer::usage::write_only);\r
 189                         tbb::parallel_for(tbb::blocked_range<std::size_t>(0, source.size()), [&](const tbb::blocked_range<std::size_t>& r)\r
 190                         {\r
 191                                 A_memcpy(buf->data() + r.begin(), source.data() + r.begin(), r.size());\r
 192                         });\r
 193                 }\r
 194 \r
 195                 return buf;\r
 196         }\r
 197 \r
 198         // TODO: Since the returned texture is cached it SHOULD NOT be modified.\r
 199         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<const std::uint8_t>& source, int width, int height, int stride)\r
 200         {\r
 201                 std::shared_ptr<buffer> buf = copy_to_buf(source);\r
 202                                 \r
 203                 return executor_.begin_invoke([=]() -> spl::shared_ptr<texture>\r
 204                 {\r
 205                         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;\r
 206                         if(texture_cache_.find(a, buf.get()))\r
 207                                 return spl::make_shared_ptr(a->second);\r
 208 \r
 209                         auto texture = create_texture(width, height, stride);\r
 210                         texture->copy_from(*buf);       \r
 211 \r
 212                         texture_cache_.insert(std::make_pair(buf.get(), texture));\r
 213 \r
 214                         return texture;\r
 215                 }, task_priority::high_priority);\r
 216         }\r
 217         \r
 218         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<std::uint8_t>& source, int width, int height, int stride)\r
 219         {\r
 220                 std::shared_ptr<buffer> buf = copy_to_buf(source);\r
 221 \r
 222                 return executor_.begin_invoke([=]() -> spl::shared_ptr<texture>\r
 223                 {\r
 224                         auto texture = create_texture(width, height, stride, false);\r
 225                         texture->copy_from(*buf);                               \r
 226                         return texture;\r
 227                 }, task_priority::high_priority);\r
 228         }\r
 229 \r
 230         boost::unique_future<array<const std::uint8_t>> copy_async(const spl::shared_ptr<texture>& source)\r
 231         {\r
 232                 if(!executor_.is_current())\r
 233                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));\r
 234 \r
 235                 auto buffer = create_buffer(source->size(), buffer::usage::read_only); \r
 236                 source->copy_to(*buffer);       \r
 237 \r
 238                 auto self = shared_from_this();\r
 239                 return async(launch::deferred, [self, buffer]() mutable -> array<const std::uint8_t>\r
 240                 {\r
 241                         self->executor_.invoke(std::bind(&buffer::map, std::ref(buffer))); // Defer blocking "map" call until data is needed.\r
 242                         return array<const std::uint8_t>(buffer->data(), buffer->size(), true, buffer);\r
 243                 });\r
 244         }\r
 245 };\r
 246 \r
 247 device::device() \r
 248         : executor_(L"OpenGL Rendering Context")\r
 249         , impl_(new impl(executor_)){}\r
 250 device::~device(){}\r
 251 spl::shared_ptr<texture>                                                        device::create_texture(int width, int height, int stride){return impl_->create_texture(width, height, stride, true);}\r
 252 array<std::uint8_t>                                                                     device::create_array(int size){return impl_->create_array(size);}\r
 253 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
 254 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
 255 boost::unique_future<array<const std::uint8_t>>         device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}\r
 256 std::wstring                                                                            device::version() const{return impl_->version();}\r
 257 \r
 258 \r
 259 }}}\r
 260 \r
 261 \r