Intel Compiler Engineer Shares Latest Project Gains For APX

Last week at the GNU Tools Cauldron 2026 conference in Prague there was an x86 BOF with presentations both by AMD and Intel compiler engineers. There are a few interesting nuggets out of that with both Advanced Performance Extensions (APX) and AI Compute Extensions (ACE) talked about as part of their ongoing enablement in the GCC compiler.

As covered in many Phoronix articles in recent months, both APX and AI Compute Extensions have been seeing a lot of enablement work happening to the GNU Compiler Collection as well as other open-source compiler infrastructure. Hongtao Liu of Intel presented on the APX work with its additional registers, new instructions, and other capabilities. He reaffirmed that the GCC support is in good shape as of GCC 15.1 when paired with Binutils 2.44+, Glibc 2.39+, and Linux 6.16+. LLVM 20.1 and newer is also good on the LLVM side.

Projected gains out of APX were also shared during the presentation. Using SPEC CPU 2017 simulations, making use of APX is projected to result in more than 10% fewer loads, more than 20% fewer stores, and more than 10% fewer instructions. It will be fun to measure the impact of APX in real-world applications once APX-enabled processors are available with Diamond Rapids and Nova Lake.

Venkataramanan Kumar and Dipesh Sharma, both of AMD, presented on ACE V1 during the GCC x86 session. That work has been hitting GCC Git and will hopefully be in good shape for GCC 17.1 next year.

Those wanting to go through all the session material from the GCC x86 presentation at the GNU Tools Cauldron 2026 can find the material here.