Intel Prepares AMX-COMPLEX Support For GCC & LLVM Compilers

Intel compiler engineers have sent out the initial GCC and LLVM/Clang compiler patches for enabling the newly-disclosed AMX-COMPLEX extension with next year's Xeon Scalable "Granite Rapids" processors.

Building off the AMX (Advanced Matrix Extensions) support introduced with current Xeon Scalable "Sapphire Rapids" processors, after Emerald Rapids is then Granite Rapids and is introducing AMX-COMPLEX.

AMX-COMPLEX is for the matrix multiplication of complex tiles accumulated into a packed single precision tile. Intel recently made public the AMX-COMPLEX instruction set details via the latest programming reference manual.

Sent out this Monday morning were the GCC patches for enabling the AMX-COMPLEX TCMMIMFP16PS/TCMMRLFP16PS instructions and exposing them with the -march=graniterapids tuning. While GCC 13 is to be released in the coming weeks, this new AMX-COMPLEX support shouldn't risk existing GCC usage and for future hardware, so Intel is hoping still to get this squeezed into the upcoming GCC 13.1 stable release.

On Sunday meanwhile was the LLVM patch for enabling Intel AMX-COMPLEX with that open-source compiler stack. That LLVM AMX-COMPLEX support is still under review.