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 <gl/glew.h>
  40
  41 #include <SFML/Window/Context.hpp>
  42
  43 #include <tbb/concurrent_unordered_map.h>
  44 #include <tbb/concurrent_hash_map.h>
  45 #include <tbb/concurrent_queue.h>
  46
  47 #include <boost/utility/declval.hpp>
  48
  49 #include <array>
  50 #include <unordered_map>
  51
  52 #include <asmlib.h>
  53 #include <tbb/parallel_for.h>
  54
  55 namespace caspar { namespace accelerator { namespace ogl {
  56
  57 struct device::impl : public std::enable_shared_from_this<impl>
  58 {
  59         static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");
  60
  61         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_cache_;
  62
  63         std::unique_ptr<sf::Context> device_;
  64
  65         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 4>      device_pools_;
  66         std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2>       host_pools_;
  67
  68         GLuint fbo_;
  69
  70         executor& executor_;
  71
  72         impl(executor& executor)
  73                 : executor_(executor)
  74         {
  75                 executor_.set_capacity(256);
  76
  77                 CASPAR_LOG(info) << L"Initializing OpenGL Device.";
  78
  79                 executor_.invoke([=]
  80                 {
  81                         device_.reset(new sf::Context());
  82                         device_->setActive(true);
  83
  84                         if (glewInit() != GLEW_OK)
  85                                 CASPAR_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));
  86
  87                         if(!GLEW_VERSION_3_0)
  88                                 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."));
  89
  90                         glGenFramebuffers(1, &fbo_);
  91                         glBindFramebuffer(GL_FRAMEBUFFER, fbo_);
  92                 });
  93
  94                 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();
  95         }
  96
  97         ~impl()
  98         {
  99                 executor_.invoke([=]
 100                 {
 101                         texture_cache_.clear();
 102
 103                         for (auto& pool : host_pools_)
 104                                 pool.clear();
 105
 106                         for (auto& pool : device_pools_)
 107                                 pool.clear();
 108
 109                         glDeleteFramebuffers(1, &fbo_);
 110
 111                         device_.reset();
 112                 });
 113         }
 114
 115         std::wstring version()
 116         {
 117                 try
 118                 {
 119                         return executor_.invoke([]
 120                         {
 121                                 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));
 122                         });
 123                 }
 124                 catch(...)
 125                 {
 126                         return L"Not found";;
 127                 }
 128         }
 129
 130         spl::shared_ptr<texture> create_texture(int width, int height, int stride, bool clear = false)
 131         {
 132                 CASPAR_VERIFY(stride > 0 && stride < 5);
 133                 CASPAR_VERIFY(width > 0 && height > 0);
 134
 135                 if(!executor_.is_current())
 136                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
 137
 138                 auto pool = &device_pools_[stride-1][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];
 139
 140                 std::shared_ptr<texture> tex;
 141                 if(!pool->try_pop(tex))
 142                         tex = spl::make_shared<texture>(width, height, stride);
 143
 144                 if(clear)
 145                         tex->clear();
 146
 147                 return spl::shared_ptr<texture>(tex.get(), [tex, pool](texture*) mutable
 148                 {
 149                         pool->push(tex);
 150                 });
 151         }
 152
 153         spl::shared_ptr<buffer> create_buffer(std::size_t size, buffer::usage usage)
 154         {
 155                 CASPAR_VERIFY(size > 0);
 156
 157                 auto pool = &host_pools_[usage.value()][size];
 158
 159                 std::shared_ptr<buffer> buf;
 160                 if(!pool->try_pop(buf))
 161                 {
 162                         boost::timer timer;
 163
 164                         buf = executor_.invoke([&]
 165                         {
 166                                 return std::make_shared<buffer>(size, usage);
 167                         }, task_priority::high_priority);
 168
 169                         if(timer.elapsed() > 0.02)
 170                                 CASPAR_LOG(debug) << L"[ogl-device] Performance warning. Buffer allocation blocked: " << timer.elapsed();
 171                 }
 172
 173                 auto self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.
 174                 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable
 175                 {
 176                         texture_cache_.erase(buf.get());
 177                         pool->push(buf);
 178                 });
 179         }
 180
 181         array<std::uint8_t> create_array(std::size_t size)
 182         {
 183                 auto buf = create_buffer(size, buffer::usage::write_only);
 184                 return array<std::uint8_t>(buf->data(), buf->size(), false, buf);
 185         }
 186
 187         template<typename T>
 188         std::shared_ptr<buffer> copy_to_buf(const T& source)
 189         {
 190                 std::shared_ptr<buffer> buf;
 191
 192                 auto tmp = source.storage<spl::shared_ptr<buffer>>();
 193                 if(tmp)
 194                         buf = *tmp;
 195                 else
 196                 {
 197                         buf = create_buffer(source.size(), buffer::usage::write_only);
 198                         tbb::parallel_for(tbb::blocked_range<std::size_t>(0, source.size()), [&](const tbb::blocked_range<std::size_t>& r)
 199                         {
 200                                 A_memcpy(buf->data() + r.begin(), source.data() + r.begin(), r.size());
 201                         });
 202                 }
 203
 204                 return buf;
 205         }
 206
 207         // TODO: Since the returned texture is cached it SHOULD NOT be modified.
 208         std::future<std::shared_ptr<texture>> copy_async(const array<const std::uint8_t>& source, int width, int height, int stride)
 209         {
 210                 std::shared_ptr<buffer> buf = copy_to_buf(source);
 211
 212                 return executor_.begin_invoke([=]() -> std::shared_ptr<texture>
 213                 {
 214                         tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;
 215                         if(texture_cache_.find(a, buf.get()))
 216                                 return spl::make_shared_ptr(a->second);
 217
 218                         auto texture = create_texture(width, height, stride);
 219                         texture->copy_from(*buf);
 220
 221                         texture_cache_.insert(std::make_pair(buf.get(), texture));
 222
 223                         return texture;
 224                 }, task_priority::high_priority);
 225         }
 226
 227         std::future<std::shared_ptr<texture>> copy_async(const array<std::uint8_t>& source, int width, int height, int stride)
 228         {
 229                 std::shared_ptr<buffer> buf = copy_to_buf(source);
 230
 231                 return executor_.begin_invoke([=]() -> std::shared_ptr<texture>
 232                 {
 233                         auto texture = create_texture(width, height, stride, false);
 234                         texture->copy_from(*buf);
 235
 236                         return texture;
 237                 }, task_priority::high_priority);
 238         }
 239
 240         std::future<array<const std::uint8_t>> copy_async(const spl::shared_ptr<texture>& source)
 241         {
 242                 if(!executor_.is_current())
 243                         CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
 244
 245                 auto buffer = create_buffer(source->size(), buffer::usage::read_only);
 246                 source->copy_to(*buffer);
 247
 248                 auto self = shared_from_this();
 249                 auto cmd = [self, buffer]() mutable -> array<const std::uint8_t>
 250                 {
 251                         self->executor_.invoke(std::bind(&buffer::map, std::ref(buffer))); // Defer blocking "map" call until data is needed.
 252                         return array<const std::uint8_t>(buffer->data(), buffer->size(), true, buffer);
 253                 };
 254                 return std::async(std::launch::deferred, std::move(cmd));
 255         }
 256 };
 257
 258 device::device()
 259         : executor_(L"OpenGL Rendering Context")
 260         , impl_(new impl(executor_)){}
 261 device::~device(){}
 262 spl::shared_ptr<texture>                                        device::create_texture(int width, int height, int stride){return impl_->create_texture(width, height, stride, true);}
 263 array<std::uint8_t>                                                     device::create_array(int size){return impl_->create_array(size);}
 264 std::future<std::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);}
 265 std::future<std::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);}
 266 std::future<array<const std::uint8_t>>          device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}
 267 std::wstring                                                            device::version() const{return impl_->version();}
 268
 269
 270 }}}
 271
 272