2 * Copyright (c) 2011 Sveriges Television AB <info@casparcg.com>
4 * This file is part of CasparCG (www.casparcg.com).
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.
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.
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/>.
19 * Author: Robert Nagy, ronag89@gmail.com
24 #include "../../StdAfx.h"
32 #include <common/assert.h>
33 #include <common/except.h>
34 #include <common/future.h>
35 #include <common/array.h>
36 #include <common/memory.h>
37 #include <common/gl/gl_check.h>
38 #include <common/timer.h>
42 #include <SFML/Window/Context.hpp>
44 #include <tbb/concurrent_unordered_map.h>
45 #include <tbb/concurrent_hash_map.h>
46 #include <tbb/concurrent_queue.h>
48 #include <boost/utility/declval.hpp>
49 #include <boost/property_tree/ptree.hpp>
52 #include <unordered_map>
55 #include <tbb/parallel_for.h>
57 namespace caspar { namespace accelerator { namespace ogl {
59 struct device::impl : public std::enable_shared_from_this<impl>
61 static_assert(std::is_same<decltype(boost::declval<device>().impl_), spl::shared_ptr<impl>>::value, "impl_ must be shared_ptr");
63 tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>> texture_cache_;
65 std::unique_ptr<sf::Context> device_;
67 std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<texture>>>, 8> device_pools_;
68 std::array<tbb::concurrent_unordered_map<std::size_t, tbb::concurrent_bounded_queue<std::shared_ptr<buffer>>>, 2> host_pools_;
74 impl(executor& executor)
77 executor_.set_capacity(256);
79 CASPAR_LOG(info) << L"Initializing OpenGL Device.";
83 device_.reset(new sf::Context());
84 device_->setActive(true);
86 if (glewInit() != GLEW_OK)
87 CASPAR_THROW_EXCEPTION(gl::ogl_exception() << msg_info("Failed to initialize GLEW."));
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."));
92 glGenFramebuffers(1, &fbo_);
93 glBindFramebuffer(GL_FRAMEBUFFER, fbo_);
96 CASPAR_LOG(info) << L"Successfully initialized OpenGL " << version();
101 auto context = executor_.is_current() ? std::string() : get_context();
105 CASPAR_SCOPED_CONTEXT_MSG(context);
106 texture_cache_.clear();
108 for (auto& pool : host_pools_)
111 for (auto& pool : device_pools_)
114 glDeleteFramebuffers(1, &fbo_);
120 boost::property_tree::wptree info() const
122 boost::property_tree::wptree info;
124 boost::property_tree::wptree pooled_device_buffers;
125 size_t total_pooled_device_buffer_size = 0;
126 size_t total_pooled_device_buffer_count = 0;
128 for (size_t i = 0; i < device_pools_.size(); ++i)
130 auto& pools = device_pools_.at(i);
131 bool mipmapping = i > 3;
132 auto stride = mipmapping ? i - 3 : i + 1;
134 for (auto& pool : pools)
136 auto width = pool.first >> 16;
137 auto height = pool.first & 0x0000FFFF;
138 auto size = width * height * stride;
139 auto count = pool.second.size();
144 boost::property_tree::wptree pool_info;
146 pool_info.add(L"stride", stride);
147 pool_info.add(L"mipmapping", mipmapping);
148 pool_info.add(L"width", width);
149 pool_info.add(L"height", height);
150 pool_info.add(L"size", size);
151 pool_info.add(L"count", count);
153 total_pooled_device_buffer_size += size * count;
154 total_pooled_device_buffer_count += count;
156 pooled_device_buffers.add_child(L"device_buffer_pool", pool_info);
160 info.add_child(L"gl.details.pooled_device_buffers", pooled_device_buffers);
162 boost::property_tree::wptree pooled_host_buffers;
163 size_t total_read_size = 0;
164 size_t total_write_size = 0;
165 size_t total_read_count = 0;
166 size_t total_write_count = 0;
168 for (size_t i = 0; i < host_pools_.size(); ++i)
170 auto& pools = host_pools_.at(i);
171 auto usage = static_cast<buffer::usage>(i);
173 for (auto& pool : pools)
175 auto size = pool.first;
176 auto count = pool.second.size();
181 boost::property_tree::wptree pool_info;
183 pool_info.add(L"usage", usage == buffer::usage::read_only ? L"read_only" : L"write_only");
184 pool_info.add(L"size", size);
185 pool_info.add(L"count", count);
187 pooled_host_buffers.add_child(L"host_buffer_pool", pool_info);
189 (usage == buffer::usage::read_only ? total_read_count : total_write_count) += count;
190 (usage == buffer::usage::read_only ? total_read_size : total_write_size) += size * count;
194 info.add_child(L"gl.details.pooled_host_buffers", pooled_host_buffers);
195 info.add(L"gl.summary.pooled_device_buffers.total_count", total_pooled_device_buffer_count);
196 info.add(L"gl.summary.pooled_device_buffers.total_size", total_pooled_device_buffer_size);
197 info.add_child(L"gl.summary.all_device_buffers", texture::info());
198 info.add(L"gl.summary.pooled_host_buffers.total_read_count", total_read_count);
199 info.add(L"gl.summary.pooled_host_buffers.total_write_count", total_write_count);
200 info.add(L"gl.summary.pooled_host_buffers.total_read_size", total_read_size);
201 info.add(L"gl.summary.pooled_host_buffers.total_write_size", total_write_size);
202 info.add_child(L"gl.summary.all_host_buffers", buffer::info());
207 std::wstring version()
211 return executor_.invoke([]
213 return u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VERSION)))) + L" " + u16(reinterpret_cast<const char*>(GL2(glGetString(GL_VENDOR))));
218 return L"Not found";;
222 spl::shared_ptr<texture> create_texture(int width, int height, int stride, bool mipmapped, bool clear)
224 CASPAR_VERIFY(stride > 0 && stride < 5);
225 CASPAR_VERIFY(width > 0 && height > 0);
227 if(!executor_.is_current())
228 CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
230 auto pool = &device_pools_[stride - 1 + (mipmapped ? 4 : 0)][((width << 16) & 0xFFFF0000) | (height & 0x0000FFFF)];
232 std::shared_ptr<texture> tex;
233 if(!pool->try_pop(tex))
234 tex = spl::make_shared<texture>(width, height, stride, mipmapped);
239 return spl::shared_ptr<texture>(tex.get(), [tex, pool](texture*) mutable
245 spl::shared_ptr<buffer> create_buffer(std::size_t size, buffer::usage usage)
247 CASPAR_VERIFY(size > 0);
249 auto pool = &host_pools_[static_cast<int>(usage)][size];
251 std::shared_ptr<buffer> buf;
252 if(!pool->try_pop(buf))
256 auto context = executor_.is_current() ? std::string() : get_context();
258 buf = executor_.invoke([&]
260 CASPAR_SCOPED_CONTEXT_MSG(context);
261 return std::make_shared<buffer>(size, usage);
262 }, task_priority::high_priority);
264 if(timer.elapsed() > 0.02)
265 CASPAR_LOG(warning) << L"[ogl-device] Performance warning. Buffer allocation blocked: " << timer.elapsed();
268 std::weak_ptr<impl> self = shared_from_this(); // buffers can leave the device context, take a hold on life-time.
269 return spl::shared_ptr<buffer>(buf.get(), [=](buffer*) mutable
271 auto strong = self.lock();
275 auto context = executor_.is_current() ? std::string() : get_context();
277 strong->executor_.invoke([&]
279 CASPAR_SCOPED_CONTEXT_MSG(context);
280 strong->texture_cache_.erase(buf.get());
281 }, task_priority::high_priority);
287 CASPAR_LOG(info) << L"Buffer outlived ogl device";
292 array<std::uint8_t> create_array(std::size_t size)
294 auto buf = create_buffer(size, buffer::usage::write_only);
295 return array<std::uint8_t>(buf->data(), buf->size(), false, buf);
299 std::shared_ptr<buffer> copy_to_buf(const T& source)
301 std::shared_ptr<buffer> buf;
303 auto tmp = source.template storage<spl::shared_ptr<buffer>>();
308 buf = create_buffer(source.size(), buffer::usage::write_only);
309 tbb::parallel_for(tbb::blocked_range<std::size_t>(0, source.size()), [&](const tbb::blocked_range<std::size_t>& r)
311 A_memcpy(buf->data() + r.begin(), source.data() + r.begin(), r.size());
318 // TODO: Since the returned texture is cached it SHOULD NOT be modified.
319 std::future<std::shared_ptr<texture>> copy_async(const array<const std::uint8_t>& source, int width, int height, int stride, bool mipmapped)
321 std::shared_ptr<buffer> buf = copy_to_buf(source);
322 auto context = executor_.is_current() ? std::string() : get_context();
324 return executor_.begin_invoke([=]() -> std::shared_ptr<texture>
326 CASPAR_SCOPED_CONTEXT_MSG(context);
327 tbb::concurrent_hash_map<buffer*, std::shared_ptr<texture>>::const_accessor a;
328 if(texture_cache_.find(a, buf.get()))
329 return spl::make_shared_ptr(a->second);
331 auto texture = create_texture(width, height, stride, mipmapped, false);
332 texture->copy_from(*buf);
334 texture_cache_.insert(std::make_pair(buf.get(), texture));
337 }, task_priority::high_priority);
340 std::future<std::shared_ptr<texture>> copy_async(const array<std::uint8_t>& source, int width, int height, int stride, bool mipmapped)
342 std::shared_ptr<buffer> buf = copy_to_buf(source);
343 auto context = executor_.is_current() ? std::string() : get_context();
345 return executor_.begin_invoke([=]() -> std::shared_ptr<texture>
347 CASPAR_SCOPED_CONTEXT_MSG(context);
348 auto texture = create_texture(width, height, stride, mipmapped, false);
349 texture->copy_from(*buf);
352 }, task_priority::high_priority);
355 std::future<array<const std::uint8_t>> copy_async(const spl::shared_ptr<texture>& source)
357 if(!executor_.is_current())
358 CASPAR_THROW_EXCEPTION(invalid_operation() << msg_info("Operation only valid in an OpenGL Context."));
360 auto buffer = create_buffer(source->size(), buffer::usage::read_only);
361 source->copy_to(*buffer);
363 auto self = shared_from_this();
364 auto context = get_context();
365 auto cmd = [self, buffer, context]() mutable -> array<const std::uint8_t>
367 self->executor_.invoke([&buffer, &context] // Defer blocking "map" call until data is needed.
369 CASPAR_LOG_CALL(trace) << "Readback <- " << context;
372 return array<const std::uint8_t>(buffer->data(), buffer->size(), true, buffer);
374 return std::async(std::launch::deferred, std::move(cmd));
377 std::future<void> gc()
379 return executor_.begin_invoke([=]
381 CASPAR_LOG(info) << " ogl: Running GC.";
385 for (auto& pools : device_pools_)
387 for (auto& pool : pools)
390 for (auto& pools : host_pools_)
392 for (auto& pool : pools)
398 CASPAR_LOG_CURRENT_EXCEPTION();
400 }, task_priority::high_priority);
405 : executor_(L"OpenGL Rendering Context")
406 , impl_(new impl(executor_)){}
408 spl::shared_ptr<texture> device::create_texture(int width, int height, int stride, bool mipmapped){ return impl_->create_texture(width, height, stride, mipmapped, true); }
409 array<std::uint8_t> device::create_array(int size){return impl_->create_array(size);}
410 std::future<std::shared_ptr<texture>> device::copy_async(const array<const std::uint8_t>& source, int width, int height, int stride, bool mipmapped){return impl_->copy_async(source, width, height, stride, mipmapped);}
411 std::future<std::shared_ptr<texture>> device::copy_async(const array<std::uint8_t>& source, int width, int height, int stride, bool mipmapped){ return impl_->copy_async(source, width, height, stride, mipmapped); }
412 std::future<array<const std::uint8_t>> device::copy_async(const spl::shared_ptr<texture>& source){return impl_->copy_async(source);}
413 std::future<void> device::gc() { return impl_->gc(); }
414 boost::property_tree::wptree device::info() const { return impl_->info(); }
415 std::wstring device::version() const{return impl_->version();}