Memory Model C++11 – C++23About Me: alex.dathskovsky@speedata.io www.linkedin.com/in/alexdathskovsky https://www.cppnext.comAlex Dathskovsky | alex.dathskovsky@speedata.io | www.linkedin.com/in/a

Agenda • Understand the motivation behind Taro • Learn to use the Taro C++ programming model • Dive into the Taro’s coroutine-aware scheduling algorithm • Evaluate Taro on microbenchmarks and a real-world Conclusion 2Agenda • Understand the motivation behind Taro • Learn to use the Taro C++ programming model • Dive into the Taro’s coroutine-aware scheduling algorithm • Evaluate Taro on microbenchmarks and details Taro!Agenda • Understand the motivation behind Taro • Learn to use the Taro C++ programming model • Dive into the Taro’s coroutine-aware scheduling algorithm • Evaluate Taro on microbenchmarks and

3RUNTIME POLYMORPHISM struct object_t { template object_t(T t) : self_(make_unique<model>(move(t)) {} // + move, copy & assignment 1 2 3 4 5 friend void draw(object_t const& o) virtual void draw_() = 0; 11 }; 12 13 template 14 struct model final : concept_t { 15 model(T t) : data_(move(t)) {} 16 void draw_() final { 17 draw(data_); 18 } 19 55RUNTIME POLYMORPHISM struct object_t { template object_t(T t) : self_(make_unique<model>(move(t)) {} // + move, copy & assignment 1 2 3 4 5 friend void draw(object_t const& o)

has_leaves = true;}; struct car {int model_year = 1982; }; void prune(tree* t) { t->has_leaves = false; } void drive(const car* c ) { printf("Driving %d\r\n", c->model_year); } int main() { const tree to float f = mustang.model_year; //implicit conversion from int to float drive((car*)&f); // driving a float return 0; }struct tree { bool has_leaves = true;}; struct car {int model_year = 1982; }; void prune(tree* t) { t->has_leaves = false; } void drive(const car* c ) { printf("Driving %d\r\n", c->model_year); } int main() { const tree oak; car mustang; drive(&mustang); //normal function call prune((tree*)&oak);

)> { public: template< typename F > function( F&& f ) : pimpl_( std::make_unique<Model>( std::forward(f) ) ) {} 74 8. …namespace std { R operator()( )> { public: template< typename F > function( F&& f ) : pimpl_( std::make_unique<Model>( std::forward(f) ) ) {} 75 8. …namespace std { R operator()( )> { public: template< typename F > function( F&& f ) : pimpl_( std::make_unique<Model>( std::forward(f) ) ) {} 76 8. …function& operator=( const function&

Memory Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 Atomic Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 Consistency Model . . . Target: x86_64-apple-darwin18.6.0 9 1.1 Deprecated Features CHAPTER 01: TOWARDS MODERN C++ Thread model: posix InstalledDir: /Library/Developer/CommandLineTools/usr/bin 1.1 Deprecated Features Before a thread; notify_all() is to notify all threads. Below is an example of a producer and consumer model: #include #include #include #include #include #include

pre-trained model ◮ CNN architecture created by Google ◮ Trained on 1000 classes ◮ Accepts 224x224-pixel images, with 3 color channels per pixel (RGB) ◮ Labels are stored separate from the model: ◮ labels (11KB) ◮ mobilenet v1 1.0 224 quant.tflite (4.1MB) 17 / 50Image classification algorithm 1. Load model and labels 2. Build interpreter 3. Allocate input and output tensors 4. Read image 5. Resize image / 501. Load model and labels 1 // defined and properly initialized elsewhere: 2 // const char *model_path; 3 // const char *labels_path; 4 5 std::unique_ptr model{ 6 tflite::

management models; buffers and USM - here we will focus on buffers The buffers memory management model separates the storage and access of data and automatically handles data dependencies USM provides /* handle SYCL exception */ } } 29 SYCL 2020 Hello WorldIn the buffers memory management model there are buffers and accessors A buffer manage data across the host and one or more devices be run by that vendor’s SYCL implementation 49Memory Spaces 50SYCL Memory Model ● Initially based on OpenCL Memory Model ● Higher level of abstraction ● Buffers & Accessors ○ Buffers ■ Storage

TMP ( - Value-based TMP ( *, *, *)*) - - (Circle-lang meta model*) (Circle-lang meta model*) - - (comptime) (comptime) - ... - ... * - this_talk * - this_talk C++ C++ 58Circle-lang - Circle-lang - Member packs - Member packs - Circle-lang meta model (Metaprogramming) Circle-lang meta model (Metaprogramming) Reflection Reflection ...... https://www.circle-lang.org Benchmarks Circle-lang meta model is the fastest to compile all around Circle-lang meta model is the fastest to compile all around 56 / 58Benchmarks Benchmarks Circle-lang meta model is the fastest to compile