The code on this branch is the outcome of a Master thesis project of the Compiler Design Lab at Saarland University.
Based on LLVM release 12.0.1, we implemented an explorative approach to the problem of finding the best vectorization and interleaving factor for innermost loops.
Roughly, the tool works as follows:
-
A pass is inserted immediately before LoopVectorize that retrieves all innermost loops from the function and copies each of them, to insert them into a new function in a new module
-
For every possible combination of vectorization and interleaving factor within the ranges hard-coded in ExplorativeLV.cpp, the copied loop is annotated such that LoopVectorize is forced to choose the factors and runs through the compilation pipeline up until an assembly or object file would be generated
-
A pass at the very end of the backend pipeline calculates a cost estimate for the generated machine code, which is used by the exploration pass to determine the best combination of factors
-
Finally, the combination that has been selected is forced onto the original loop with annotations, such that the loop vectorizer will chose these factors.
The implementation adds the following hidden command line arguments to enable and fine-tune compilation with the exploration tool:
-
enable-explorative-lv
: Set to true in order to activate the tool -
explore-plain
: By default, the tool's cost calculation only takes into account machine code of blocks that are part of the loop, i.e., blocks whose names contain the strings "vector.
", "while.
" or "for.
". Set this option to true if you want to base your result on the complete machine code output. -
explore-divide-by-vf
: Enable this option in order to make the exploration pass divide the cost results by the vectorization factor when comparing them. This shadows LoopVectorize's behaviour when computing vectorization costs and increases the chance of higher factors being used. -
explore-with-mca
: Set this argument to true in order to usellvm-mca
for the machine code cost estimation instead of the machine instruction count. Note: This only works on architectures for which llvm-mca is available and can analyse machine code that may contain control structures
More information about the project as well as an evaluation on an embedded Arm and an Intel© server architecture can be found here: Thesis link to follow soon
From here on follows the original README of the LLVM project
This directory and its sub-directories contain source code for LLVM, a toolkit for the construction of highly optimized compilers, optimizers, and run-time environments.
The README briefly describes how to get started with building LLVM. For more information on how to contribute to the LLVM project, please take a look at the Contributing to LLVM guide.
Taken from https://llvm.org/docs/GettingStarted.html.
Welcome to the LLVM project!
The LLVM project has multiple components. The core of the project is itself called "LLVM". This contains all of the tools, libraries, and header files needed to process intermediate representations and converts it into object files. Tools include an assembler, disassembler, bitcode analyzer, and bitcode optimizer. It also contains basic regression tests.
C-like languages use the Clang front end. This component compiles C, C++, Objective-C, and Objective-C++ code into LLVM bitcode -- and from there into object files, using LLVM.
Other components include: the libc++ C++ standard library, the LLD linker, and more.
The LLVM Getting Started documentation may be out of date. The Clang Getting Started page might have more accurate information.
This is an example work-flow and configuration to get and build the LLVM source:
-
Checkout LLVM (including related sub-projects like Clang):
-
git clone https://github.com/llvm/llvm-project.git
-
Or, on windows,
git clone --config core.autocrlf=false https://github.com/llvm/llvm-project.git
-
-
Configure and build LLVM and Clang:
-
cd llvm-project
-
mkdir build
-
cd build
-
cmake -G <generator> [options] ../llvm
Some common build system generators are:
Ninja
--- for generating Ninja build files. Most llvm developers use Ninja.Unix Makefiles
--- for generating make-compatible parallel makefiles.Visual Studio
--- for generating Visual Studio projects and solutions.Xcode
--- for generating Xcode projects.
Some Common options:
-
-DLLVM_ENABLE_PROJECTS='...'
--- semicolon-separated list of the LLVM sub-projects you'd like to additionally build. Can include any of: clang, clang-tools-extra, libcxx, libcxxabi, libunwind, lldb, compiler-rt, lld, polly, or debuginfo-tests.For example, to build LLVM, Clang, libcxx, and libcxxabi, use
-DLLVM_ENABLE_PROJECTS="clang;libcxx;libcxxabi"
. -
-DCMAKE_INSTALL_PREFIX=directory
--- Specify for directory the full path name of where you want the LLVM tools and libraries to be installed (default/usr/local
). -
-DCMAKE_BUILD_TYPE=type
--- Valid options for type are Debug, Release, RelWithDebInfo, and MinSizeRel. Default is Debug. -
-DLLVM_ENABLE_ASSERTIONS=On
--- Compile with assertion checks enabled (default is Yes for Debug builds, No for all other build types).
-
cmake --build . [-- [options] <target>]
or your build system specified above directly.-
The default target (i.e.
ninja
ormake
) will build all of LLVM. -
The
check-all
target (i.e.ninja check-all
) will run the regression tests to ensure everything is in working order. -
CMake will generate targets for each tool and library, and most LLVM sub-projects generate their own
check-<project>
target. -
Running a serial build will be slow. To improve speed, try running a parallel build. That's done by default in Ninja; for
make
, use the option-j NNN
, whereNNN
is the number of parallel jobs, e.g. the number of CPUs you have.
-
-
For more information see CMake
-
Consult the Getting Started with LLVM page for detailed information on configuring and compiling LLVM. You can visit Directory Layout to learn about the layout of the source code tree.