LLVM Lands APX JMPABS Support, More Advanced Performance Extensions Work Landing

Intel compiler engineers remain quite busy working not only on AVX10 family support but also plumbing in the Advanced Performance Extensions (APX) to be found with future Intel processors.

Intel's open-source compiler experts have been posting a number of GCC patches around APX support while more APX code has also been trickling into the LLVM/Clang compiler codebase.

Most recently merged this week is the support for encoding/decoding for JMPABS. APX JMPABS allows for jumping to 64-bit absolute addresses. APX JMPABS will be beneficial for just-in-time (JIT) compiled code and a variety of other possible optimizations to come with JMPABS usage.

There's also been APX EGPR merged two weeks ago along with other Intel APX code pending.

These recent patches upstreamed into LLVM will in turn premiere with the LLVM 18.0 release that typically happens for their H1 release in the March~April timeframe. With GCC 14 now under a feature freeze and not all of that APX code upstreamed yet, LLVM is in better shape for seeing complete APX support first in a released compiler thanks to its six month release cycle compared to the GNU Compiler Collection pushing out major releases on an annual basis. In any event it's great as usual seeing Intel's timely contributions to GCC and LLVM/Clang for new CPU ISA features.