Intel Fully Embracing LLVM For Their C/C++ Compilers
Intel's longtime compiler expert James Reinders wrote a blog post today outlining their use of LLVM for their specialized C/C++ compilers moving forward. "LLVM helps us with our goal to offer the best C/C++ compilers for Intel architecture. The latest Intel C/C++ compilers, using LLVM, deliver faster compiler times, better optimizations, enhanced standards support, and support for GPU and FPGA offloading...The benefits of adopting LLVM are numerous...We are committed to making this as seamless as possible while yielding numerous benefits for developers who use the Intel compilers."
Intel engineers are reporting their new oneAPI DPC++/C++ Compiler 2021.3 that is based on LLVM has 14% lower build times than their classic C++ compiler. While their classic Intel C++ Compiler has been faster than upstream LLVM and GCC, their oneAPI DPC++/C++ Compiler 2021.3 is said to offer 41% faster performance than GCC 11.1 and similar margins to LLVM 12. The floating-point performance of their new LLVM-based compiler is about 19% higher than ICC.
Intel is recommending all new projects make use of their LLVM-based Intel C/C++ compilers and existing projects should plan to migrate to the new compiler this year. Their "classic" Intel C/C++ compilers will transition to legacy mode at the end of regular updates. James added, "The new LLVM-based Intel C/C++ has reached parity with the classic version, and the LLVM-based C/C++ offers the best optimization technology we have. We suggest all users should try the new C/C++ compiler now, enjoy the benefits, and provide feedback."
While the C/C++ support is still in great shape, Intel engineers are still working on transitioning their Fortran compiler to be LLVM-based. At the moment they have a beta version of an LLVM-based Fortran compiler.
More details on this exciting milestone for Intel's compilers via software.intel.com. Looks like it's time to fire up some new Intel compiler benchmarks on Linux.