icx-valgrind

Automated Valgrind build script with advanced compiler optimizations using Intel oneAPI compilers. Designed for users who want a more performant Valgrind installation.

Overview

This project provides an optimized build script for Valgrind that leverages Intel's oneAPI compilers (ICX/ICPX) to produce a more performant binary through:

• Profile-Guided Optimization (PGO) - Uses runtime profiling to optimize hot paths
• Link-Time Optimization (LTO) - Enables whole-program optimization across translation units
• Intel-specific flags - Utilizes -march=native, -ffast-math, vectorization, and other Intel compiler optimizations

Quick Install

curl -fsSL https://raw.githubusercontent.com/TheValiant/icx-valgrind/main/install.sh | bash

Then restart your terminal or source your shell configuration:

source ~/.bashrc   # or ~/.zshrc

Requirements

• Intel oneAPI Base Toolkit (includes ICX compiler)
• Standard build tools (make, autoconf)
• LLVM tools (llvm-ar, llvm-ranlib, lld)
• Internet connection (downloads Valgrind source)

Installing Intel oneAPI

On Linux:

# Add Intel repository
wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | sudo apt-key add -
sudo add-apt-repository "deb https://apt.repos.intel.com/oneapi all main"

# Install base toolkit
sudo apt install intel-basekit

On macOS (Intel):

# Install via installer from Intel's website
# https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html

What the Script Does

1. Pre-flight checks - Verifies Intel compilers and required tools are available
2. Downloads Valgrind - Fetches the latest stable Valgrind release (3.25.1)
3. Stage 1 Build - Compiles with basic optimizations
4. PGO Training - Runs benchmark suite to generate optimization profile
5. Stage 2 Build - Recompiles with PGO data and full LTO enabled
6. Installation - Installs to ~/.local/bin/valgrind
7. PATH Setup - Automatically updates shell configuration files

Build Flags

Stage 1 (Initial Build)

-O3 -pipe -fp-model=fast -march=native

Stage 2 (Optimized Build)

-O3 -pipe -fp-model=fast -march=native -static -ffast-math 
-funroll-loops -fvectorize -vec -flto=full 
-fprofile-instr-use=<profile_data>

Manual Build

If you prefer to run the build manually or need to customize:

git clone https://github.com/TheValiant/icx-valgrind.git
cd icx-valgrind
chmod +x install.sh
./install.sh --verbose

Options

• --verbose, -v - Enable verbose build output
• --help, -h - Show usage information

Verification

After installation, verify the optimized build:

which valgrind
# Should show: ~/.local/bin/valgrind

valgrind --version
# Shows installed version

# Check for LTO in binary
readelf -p .comment ~/.local/bin/valgrind

Alternative: GCC Build

For systems without Intel compilers, a GCC fallback script is provided:

./gcc.sh

This uses GCC with similar optimization flags but without Intel-specific optimizations.

Notes

• Build takes approximately 15-30 minutes depending on system
• Requires ~500MB of temporary disk space during build
• PGO training phase runs Valgrind against a comprehensive test suite
• The script is idempotent and safe to re-run

Troubleshooting

Intel compiler not found

Ensure you've sourced the oneAPI environment:

source /opt/intel/oneapi/setvars.sh

LTO verification fails

LTO requires matching versions of compiler tools. Ensure llvm-ar, llvm-ranlib, and lld are from the Intel oneAPI installation.

Build fails with OOM

Reduce parallel jobs by editing the script to use make -j4 instead of make -j$(nproc).

License

This project is provided as-is. Valgrind itself is licensed under GPLv2.