News › ai-infrastructure
By Zayden R., September 2, 2026
The GCC 17 compiler has merged initial support for AI Compute Extensions (ACEv1) from Intel and AMD, setting the stage for enhanced AI/ML performance. This collaboration aims to unify matrix acceleration architecture in future CPUs.
In a significant development for AI and machine learning workloads, the GNU Compiler Collection (GCC) has integrated initial support for the AI Compute Extensions (ACEv1). This update comes as part of a collaborative effort between Intel and AMD to create a unified matrix acceleration architecture. Engineers should take note, as this enhancement is poised to improve performance on future CPUs.
The ACEv1 extension is set to supersede Intel's Advanced Matrix Extensions (AMX) with additional enhancements, and it will also support upcoming AMD processors. The documentation for ACEv1 was published in June, and since July, patches have been circulating on the GCC mailing list. With the initial ACE enablement patches landing in the GCC Git codebase, GCC 17 is now equipped to support the first generation of AI Compute Extensions.
Introduced with this update is the new -macev1 compiler flag, which enables ACEv1 built-in functions and code generation. This flag also implies support for AVX10.1, AVX/AVX2, and SSE/SSE4 extensions, which are crucial for optimizing AI/ML tasks. The full debut of GCC 17.1 is anticipated between March and April of next year, and it is expected to include further ACE enablement work.
This move is not isolated to GCC; similar efforts are underway in the LLVM/Clang compiler, where ACEv1 support is in progress. For engineers, the introduction of ACEv1 means that they should prepare for enhanced AI/ML processing capabilities in their development workflows, especially when working with future Intel and AMD processors.
Overall, this is a solid improvement in the compiler landscape, reflecting the industry's ongoing commitment to advancing AI and machine learning technologies.