Open Projects

Add initial integration of Clad with Enzyme

Description

In mathematics and computer algebra, automatic differentiation (AD) is a set of techniques to numerically evaluate the derivative of a function specified by a computer program. Automatic differentiation is an alternative technique to Symbolic differentiation and Numerical differentiation (the method of finite differences). Clad is based on Clang which provides the necessary facilities for code transformation. The AD library is able to differentiate non-trivial functions, to find a partial derivative for trivial cases and has good unit test coverage. Enzyme is a prominent autodiff framework which works on LLVM IR.

Clad and Enzyme can be considered as a C++ frontend and a backend automatic differentiation framework. In many cases, when clad needs to fall back to numeric differentiation it can try configuring and using Enzyme to perform the automatic differentiation on lower level.

Task ideas and expected results

Understand how both systems work. Define the Enzyme configuration requirements and enable Clad to communicate efficiently with Enzyme. That may require several steps: start building and using the optimization pass of Enzyme as part of the Clad toolchain; use Enzyme for cross-validation derivative results; etc.

Add numerical differentiation support in clad

Description

In mathematics and computer algebra, automatic differentiation (AD) is a set of techniques to numerically evaluate the derivative of a function specified by a computer program. Automatic differentiation is an alternative technique to Symbolic differentiation and Numerical differentiation (the method of finite differences). Clad is based on Clang which provides the necessary facilities for code transformation. The AD library is able to differentiate non-trivial functions, to find a partial derivative for trivial cases and has good unit test coverage. In a number of cases, due different limitations, it is either inefficient or impossible to differentiate a function.

Currently, clad cannot differentiate declared-but-not-defined functions. In that case it issues an error. Instead, clad should fall back to its future numerical differentiation facilities.

Task ideas and expected results

Implement numerical differentiation support in clad. It should be available through a dedicated interface (for example clad::num_differentiate). The new functionality should be connected to the forward mode automatic differentiation. If time permits, a prototype of configurable error estimation for the numerical differentiation should be implemented.

Add support for functor objects in clad

Description

Many computations are modelled using functor objects. Usually, a functor object is a lightweight C++ object which has state stored as members and has overridden call operator (operator()). For example:

struct Functor {
  double x;
  double operator()() { return x * x ;}
};

int main () {
  Functor Fn;
  Fn.x = 2;
  double pow2 = Fn();
  auto dpow2_dx = clad::differentiate(Fn, /*wrt*/ 0); // unsupported
  return pow2;
}

The goal of this project is to modify Clad to handle such cases.

Task ideas and expected results

Implement functor object differentiation in both forward and reverse mode. The candidate should be ready to investigate performance bottlenecks, add test and benchmarking coverage and improve documentation for various parts of clad not only limited to the functor object differentiation support.

Utilize second order derivatives from Clad in ROOT

Description

ROOT is a framework for data processing, born at CERN, at the heart of the research on high-energy physics. Every day, thousands of physicists use ROOT applications to analyze their data or to perform simulations. ROOT has a clang-based C++ interpreter Cling and integrates with Clad to enable flexible automatic differentiation facility.

Task ideas and expected results

TFormula is a ROOT class which bridges compiled and interpreted code. Teach TFormula to use second order derivatives (using clad::hessian). The implementation should be very similar to what was done in this pull request. The produced code should be well tested and documented. If time permits, we should pursue converting the C++ gradient function to CUDA device kernels. The integration of that feature should be comparable in terms of complexity to integrating clad::hessians.

Improve Cling’s Development Lifecycle

Description

Cling is an interactive C++ interpreter, built on top of Clang and LLVM compiler infrastructure. Cling realizes the read-eval-print loop (REPL) concept, in order to leverage rapid application development. Implemented as a small extension to LLVM and Clang, the interpreter reuses their strengths such as the praised concise and expressive compiler diagnostics.

Task ideas and expected results

The project foresees to enhance the Github Actions infrastructure by adding development process automation tools:

Code Coverage information (codecov)
Static code analysis (clang-tidy)
Coding conventions checks (clang-format)
Release binary upload automation

Allow redefinition of CUDA functions in Cling

Description

Since the development of Cling started, it got some new features to enable new workflows. One of the features is CUDA mode, which allows you to interactively develop and run CUDA code on Nvidia GPUs. Another feature is the redefinition of functions, variable classes and more, bypassing the one-definition rule of C++. This feature enables comfortable rapid prototyping in C++. Currently, the two features cannot be used together because parsing and executing CUDA code behaves differently compared to pure C++.

Task ideas and expected results

The task is to adapt the redefinitions feature of the pure C++ mode for the CUDA mode. To do this, the student must develop solutions to known and unknown problems that parsing and executing CUDA code causes.

Improving Cling Reflection for Scripting Languages

Description

Cling has basic facilities to make queries about the C++ code that it has seen/collected so far. These lookups assume, however, that the caller knows what it is looking for and the information returned, although exact, usually only makes sense within C++ and is thus often too specific to be used as-is.

A scripting language, such as Python, that wants to make use of such lookups by name, is forced to loop over all possible entities (classes, functions, templates, enums, …) to find a match. This is inefficient. Furthermore, many lookups will be multi-stage: a function, but which overload? A template, but which instantiation? A typedef, of what? The current mechanism forces the scripting language to provide a type-based match, even where C++ makes distinctions (e.g. pointer v.s. reference) that do not exist in the scripting language. This, too, makes lookups very inefficient.

The returned information, once a match is found, is exact, but because of its specificity, requires the caller to figure out C++ concepts that have no meaning in the scripting language. E.g., there is no reason for Python to consider an implicitly instantiated function template different from an explicitly instantiated one.

Task ideas and expected results

The project should start with a design C++ entities grouping that make sense for scripting languages and design a query API on top of these. Special consideration should be given to the various name aliasing mechanisms in C++ (e.g., using and typedef), especially as relates to C++ access rules. This design can likely start from the existing Cling wrapper API from Cppyy and modify it to improve consistency, remove redundancy, and enable usage patterns that minimize lookups into Cling.

Next is a redesign of Cling’s lookup facilities to support this new API efficiently, with particular care to use cases that require multiple, consecutive/related, lookups, such as finding a specific function template instantiation, or step-wise resolution of a typedef.

This design is then to be implemented, using test-driven development. As a stretch goal, the cppyy-backend should be modified to use the new API.

Developing C++ modules support in CMSSW and Boost

Description

The LHC smashes groups of protons together at close to the speed of light: 40 million times per second and with seven times the energy of the most powerful accelerators built up to now. Many of these will just be glancing blows but some will be head on collisions and very energetic. When this happens some of the energy of the collision is turned into mass and previously unobserved, short-lived particles – which could give clues about how Nature behaves at a fundamental level - fly out and into the detector. Our work includes the experimental discovery of the Higgs boson, which leads to the award of a Nobel prize for the underlying theory that predicted the Higgs boson as an important piece of the standard model theory of particle physics.

CMS is a particle detector that is designed to see a wide range of particles and phenomena produced in high-energy collisions in the LHC. Like a cylindrical onion, different layers of detectors measure the different particles, and use this key data to build up a picture of events at the heart of the collision.

Last year, thanks to Lucas Calmolezi and GSoC, the usage of boost in CMSSW was modernized. It improved the C++ modules support of local boost fork.

Task ideas and expected results

Many of the accumulated local patches add missing includes to the relevant boost header files. The candidate should start by proposing the existing patches to the boost community. Try to compile more boost-specific modules which is mostly a mechanical task. The student should be ready to work towards making the C++ module files more efficient containing less duplications. The student should be prepared to write a progress report and present the results.

Implement a shared-memory based JITLinkMemoryManager for out-of-process JITting

Description

Write a shared-memory based JITLinkMemoryManager.

LLVM’s JIT uses the JITLinkMemoryManager interface to allocate both working memory (where the JIT fixes up the relocatable objects produced by the compiler) and target memory (where the JIT’d code will reside in the target). JITLinkMemoryManager instances are also responsible for transporting fixed-up code from working memory to target memory. LLVM has an existing cross-process allocator that uses remote procedure calls (RPC) to allocate and copy bytes to the target process, however a more attractive solution (when the JIT and target process share the same physical memory) would be to use shared memory pages to avoid copies between processes.

Task ideas and expected results:

Implement a shared-memory based JITLinkMemoryManager:

Write generic LLVM APIs for shared memory allocation.
Write a JITLinkMemoryManager that uses these generic APIs to allocate shared working-and-target memory.
Make an extensive performance study of the approach.

Modernize the LLVM “Building A JIT” tutorial series

Description

The LLVM BuildingAJIT tutorial series teaches readers to build their own JIT class from scratch using LLVM’s ORC APIs, however the tutorial chapters have not kept pace with recent API improvements. Bring the existing tutorial chapters up to speed, write up a new chapter on lazy compilation (chapter code already available) or write a new chapter from scratch.

Task ideas and expected results:

Update chapter text for Chapters 1-3 – Easy, but offers a chance to get up-to-speed on the APIs.
Write chapter text for Chapter 4 – Chapter code is already available, but no chapter text exists yet.
Write a new chapter from scratch – E.g. How to write an out-of-process JIT, or how to directly manipulate the JIT’d instruction stream using the ObjectLinkingLayer::Plugin API.

Write JITLink support for a new format/architecture

Description

JITLink is LLVM’s new JIT linker API – the low-level API that transforms compiler output (relocatable object files) into ready-to-execute bytes in memory. To do this JITLink’s generic linker algorithm needs to be specialized to support the target object format (COFF, ELF, MachO), and architecture (arm, arm64, i386, x86-64). LLVM already has mature implementations of JITLink for MachO/arm64 and MachO/x86-64, and a relatively new implementation for ELF/x86-64. Write a JITLink implementation for a missing target that interests you. If you choose to implement support for a new architecture using the ELF or MachO formats then you will be able to re-use the existing generic code for these formats. If you want to implement support for a new target using the COFF format then you will need to write both the generic COFF support code and the architecture support code for your chosen architecture.

Task ideas and expected results:

Write a JITLink specialization for a not-yet-supported format/architecture.

Extend clang AST to provide information for the type as written in template instantiations.

Description

When instantiating a template, the template arguments are canonicalized before being substituted into the template pattern. Clang does not preserve type sugar when subsequently accessing members of the instantiation.

std::vector<std::string> vs;
int n = vs.front(); // bad diagnostic: [...] aka 'std::basic_string<char>' [...]

template<typename T> struct Id { typedef T type; };
Id<size_t>::type // just 'unsigned long', 'size_t' sugar has been lost

Clang should “re-sugar” the type when performing member access on a class template specialization, based on the type sugar of the accessed specialization. The type of vs.front() should be std::string, not std::basic_string<char, […]>.

Suggested design approach: add a new type node to represent template argument sugar, and implicitly create an instance of this node whenever a member of a class template specialization is accessed. When performing a single-step desugar of this node, lazily create the desugared representation by propagating the sugared template arguments onto inner type nodes (and in particular, replacing Subst*Parm nodes with the corresponding sugar). When printing the type for diagnostic purposes, use the annotated type sugar to print the type as originally written.

For good results, template argument deduction will also need to be able to deduce type sugar (and reconcile cases where the same type is deduced twice with different sugar).

Task ideas and expected results:

Diagnostics preserve type sugar even when accessing members of a template specialization. T<unsigned long> and T<size_t> are still the same type and the same template instantiation, but T<unsigned long>::type single-step desugars to ‘unsigned long’ and T<size_t>::type single-step desugars to ‘size_t’.

Infrastructure

Improve Cling’s packaging system cpt

Cling has a flexible tool which can build and package binaries. It is implemented in python.

Currently it has an issue deb package creation. The tool calls shell commands such as mv and wget which should be replaced with their proper python versions.