(/LIBRARY) IS THE COMPILER... (/LIBRARY) IS THE COMPILER... JUST-IN-TIME (JIT) IS COMPILATION AT THE POINT OF NEED JUST-IN-TIME (JIT) IS COMPILATION AT THE POINT OF NEED 3 . 2WHAT IS JITTING? WHAT IS JITTING (/LIBRARY) IS THE COMPILER... (/LIBRARY) IS THE COMPILER... JUST-IN-TIME (JIT) IS COMPILATION AT THE POINT OF NEED JUST-IN-TIME (JIT) IS COMPILATION AT THE POINT OF NEED IT'S LIKE THE APPLICATION (/LIBRARY) IS THE COMPILER... 3 . 2WHY JIT-COMPILE? WHY JIT-COMPILE? 3 . 3WHY JIT-COMPILE? WHY JIT-COMPILE? AOT IS NOT POSSIBLE AOT IS NOT POSSIBLE 3 . 3WHY JIT-COMPILE? WHY JIT-COMPILE? AOT IS NOT POSSIBLE

still today, are actually a continuum.Papers in this talk github.com/jfbastien/jit-talk CppCon Slack channel #sig_jit I’ll cover 20-some papers in this talk, and have collected them on a GitHub repo years. We’ve also got a CppCon Slack channel: SIG_JIT (Special Interest Group). In a way this talk isn’t my usual talk because it’s more of a lecture on JiT compilers, where I’ll outline the papers that spoke than usual, from these papers. First, some definitions.JiT JiT Just-in-Time compilation With some artistic liberty, folks usually think of JIT as: The executable code changes after the program is loaded

Profiler Bar example(Foo&) example+0 pushq example+2 pushq example+4 ...JIT Code jit+0 movabs jit+7 ... jit+n TRAP 1. Inject OI into the target process 2. Wait for a thread to hit our example+0 pushq example+2 pushq example+4 ... Data Buffer TRAP TRAPThread A JIT Code jit+0 movabs jit+7 ... jit+n TRAP 1. Inject OI into the target process 2. Wait for a thread to hit our example+0 pushq example+2 pushq example+4 ... Data Buffer TRAP TRAPThread A JIT Code jit+0 movabs jit+7 ... jit+n TRAP 1. Inject OI into the target process 2. Wait for a thread to hit our

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1922 105.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1924 105.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 167 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1922 105.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1924 105.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 167 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1922 105.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1924 105.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 167 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Julia + MMTk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1976 106.29 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1978 106.30 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 168 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1972 106.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1974 106.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 168 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1972 106.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1974 106.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 168 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment

Garbage Collection in Julia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1972 106.28 JIT Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1974 106.29 Core typing, multiple dispatch, and good performance, achieved using type inference and just-in-time (JIT) compilation (and optional ahead-of-time compilation), implemented using LLVM. It is multi-paradigm ability to statically infer and compile code to run fast, without theCHAPTER 13. METHODS 168 usual JIT tricks and overhead. Indeed, any new method definition won't be visible to the current runtime environment