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

   1 /*
   2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
   3 *
   4 * This file is part of CasparCG (www.casparcg.com).
   5 *
   6 * CasparCG is free software: you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License as published by
   8 * the Free Software Foundation, either version 3 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * CasparCG is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with CasparCG. If not, see <http://www.gnu.org/licenses/>.
  18 *
  19 * Author: Robert Nagy, ronag89@gmail.com
  20 */
  21
  22 // TODO: Smart GC
  23
  24 #include "../../stdafx.h"
  25
  26 #include "device.h"
  27
  28 #include "buffer.h"
  29 #include "texture.h"
  30 #include "shader.h"
  31
  32 #include <common/assert.h>
  33 #include <common/except.h>
  34 #include <common/future.h>
  35 #include <common/array.h>
  36 #include <common/gl/gl_check.h>
  37 #include <common/os/windows/windows.h>
  38
  39 #include <boost/foreach.hpp>
  40
  41 #include <gl/glew.h>
  42
  43 #include <SFML/Window/Context.hpp>
  44
  45 #include <tbb/concurrent_unordered_map.h>
  46 #include <tbb/concurrent_hash_map.h>
  47 #include <tbb/concurrent_queue.h>
  48
  49 #include <boost/utility/declval.hpp>
  50
  51 #include <array>
  52 #include <unordered_map>
  53
  54 #include <asmlib.h>
  55 #include <tbb/parallel_for.h>
  56
  57 namespace caspar { namespace accelerator { namespace ogl {
  58
  59 struct device::impl : public std::enable_shared_from_this<impl>
  60 {
  61         static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");
  62
  63         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_cache_;
  64
  65         std::unique_ptr<sf::Context> device_;
  66
  67         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 4>      device_pools_;
  68         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2>       host_pools_;
  69
  70         GLuint fbo_;
  71
  72         executor& executor_;
  73
  74         impl(executor& executor)
  75                 : executor_(executor)
  76         {
  77                 executor_.set_capacity(256);
  78
  79                 CASPAR_LOG(info) << L"Initializing OpenGL Device.";
  80
  81                 executor_.invoke([=]
  82                 {
  83                         device_.reset(new sf::Context());
  84                         device_->SetActive(true);
  85
  86                         if (glewInit() != GLEW_OK)
  87                                 CASPAR_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));
  88
  89                         if(!GLEW_VERSION_3_0)
  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."));
  91
  92                         glGenFramebuffers(1, &fbo_);
  93                         glBindFramebuffer(GL_FRAMEBUFFER, fbo_);
  94                 });
  95
  96                 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();
  97         }
  98
  99         ~impl()
 100         {
 101                 executor_.invoke([=]
 102                 {
 103                         BOOST_FOREACH(auto& pool, device_pools_)
 104                                 pool.clear();
 105                         glDeleteFramebuffers(1, &fbo_);
 106
 107                         device_.reset();
 108                 });
 109         }
 110
 111         std::wstring version()
 112         {
 113                 try
 114                 {
 115                         return executor_.invoke([]
 116                         {
 117                                 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));
 118                         });
 119                 }
 120                 catch(...)
 121                 {
 122                         return L"Not found";;
 123                 }
 124         }
 125
 126         spl::shared_ptr<texture> create_texture(int width, int height, int stride, bool clear = false)
 127         {
 128                 CASPAR_VERIFY(stride > 0 && stride < 5);
 129                 CASPAR_VERIFY(width > 0 && height > 0);
 130
 131                 if(!executor_.is_current())
 132                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
 133
 134                 auto pool = &device_pools_[stride-1][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];
 135
 136                 std::shared_ptr<texture> tex;
 137                 if(!pool->try_pop(tex))
 138                         tex = spl::make_shared<texture>(width, height, stride);
 139
 140                 if(clear)
 141                         tex->clear();
 142
 143                 return spl::shared_ptr<texture>(tex.get(), [tex, pool](texture*) mutable
 144                 {
 145                         pool->push(tex);
 146                 });
 147         }
 148
 149         spl::shared_ptr<buffer> create_buffer(std::size_t size, buffer::usage usage)
 150         {
 151                 CASPAR_VERIFY(size > 0);
 152
 153                 auto pool = &host_pools_[usage.value()][size];
 154
 155                 std::shared_ptr<buffer> buf;
 156                 if(!pool->try_pop(buf))
 157                 {
 158                         boost::timer timer;
 159
 160                         buf = executor_.invoke([&]
 161                         {
 162                                 return spl::make_shared<buffer>(size, usage);
 163                         }, task_priority::high_priority);
 164
 165                         if(timer.elapsed() > 0.02)
 166                                 CASPAR_LOG(debug) << L"[ogl-device] Performance warning. Buffer allocation blocked: " << timer.elapsed();
 167                 }
 168
 169                 auto self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.
 170                 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable
 171                 {
 172                         texture_cache_.erase(buf.get());
 173                         pool->push(buf);
 174                 });
 175         }
 176
 177         array<std::uint8_t> create_array(std::size_t size)
 178         {
 179                 auto buf = create_buffer(size, buffer::usage::write_only);
 180                 return array<std::uint8_t>(buf->data(), buf->size(), false, buf);
 181         }
 182
 183         template<typename T>
 184         std::shared_ptr<buffer> copy_to_buf(const T& source)
 185         {
 186                 std::shared_ptr<buffer> buf;
 187
 188                 auto tmp = source.storage<spl::shared_ptr<buffer>>();
 189                 if(tmp)
 190                         buf = *tmp;
 191                 else
 192                 {
 193                         buf = create_buffer(source.size(), buffer::usage::write_only);
 194                         tbb::parallel_for(tbb::blocked_range<std::size_t>(0, source.size()), [&](const tbb::blocked_range<std::size_t>& r)
 195                         {
 196                                 A_memcpy(buf->data() + r.begin(), source.data() + r.begin(), r.size());
 197                         });
 198                 }
 199
 200                 return buf;
 201         }
 202
 203         // TODO: Since the returned texture is cached it SHOULD NOT be modified.
 204         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<const std::uint8_t>& source, int width, int height, int stride)
 205         {
 206                 std::shared_ptr<buffer> buf = copy_to_buf(source);
 207
 208                 return executor_.begin_invoke([=]() -> spl::shared_ptr<texture>
 209                 {
 210                         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;
 211                         if(texture_cache_.find(a, buf.get()))
 212                                 return spl::make_shared_ptr(a->second);
 213
 214                         auto texture = create_texture(width, height, stride);
 215                         texture->copy_from(*buf);
 216
 217                         texture_cache_.insert(std::make_pair(buf.get(), texture));
 218
 219                         return texture;
 220                 }, task_priority::high_priority);
 221         }
 222
 223         boost::unique_future<spl::shared_ptr<texture>> copy_async(const array<std::uint8_t>& source, int width, int height, int stride)
 224         {
 225                 std::shared_ptr<buffer> buf = copy_to_buf(source);
 226
 227                 return executor_.begin_invoke([=]() -> spl::shared_ptr<texture>
 228                 {
 229                         auto texture = create_texture(width, height, stride, false);
 230                         texture->copy_from(*buf);
 231
 232                         return texture;
 233                 }, task_priority::high_priority);
 234         }
 235
 236         boost::unique_future<array<const std::uint8_t>> copy_async(const spl::shared_ptr<texture>& source)
 237         {
 238                 if(!executor_.is_current())
 239                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
 240
 241                 auto buffer = create_buffer(source->size(), buffer::usage::read_only);
 242                 source->copy_to(*buffer);
 243
 244                 auto self = shared_from_this();
 245                 return async(launch::deferred, [self, buffer]() mutable -> array<const std::uint8_t>
 246                 {
 247                         self->executor_.invoke(std::bind(&buffer::map, std::ref(buffer))); // Defer blocking "map" call until data is needed.
 248                         return array<const std::uint8_t>(buffer->data(), buffer->size(), true, buffer);
 249                 });
 250         }
 251 };
 252
 253 device::device()
 254         : executor_(L"OpenGL Rendering Context")
 255         , impl_(new impl(executor_)){}
 256 device::~device(){}
 257 spl::shared_ptr<texture>                                                        device::create_texture(int width, int height, int stride){return impl_->create_texture(width, height, stride, true);}
 258 array<std::uint8_t>                                                                     device::create_array(int size){return impl_->create_array(size);}
 259 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);}
 260 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);}
 261 boost::unique_future<array<const std::uint8_t>>         device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}
 262 std::wstring                                                                            device::version() const{return impl_->version();}
 263
 264
 265 }}}
 266
 267