AmpereOne Sees Last Minute Compiler Tuning Ahead Of GCC 13

Going back to late 2021 was the initial GCC compiler patch for "Ampere-1" for that next-gen AArch64 server processor while last year this successor to Ampere Altra (Max) was formally announced under the AmpereOne brand. That initial compiler support appeared in GCC 12 while ahead of the GCC 13 release in the coming weeks has been some last minute tuning for the AmpereOne cost table.

Philipp Tomsich landed an updated AmpereOne (Ampere1) vectorization cost information today for the GCC 13 compiler. He explained in that commit:

The original submission of AmpereOne (-mcpu=ampere1) costs occurred prior to exhaustive testing of vectorizable workloads against hardware.

Adjust the vector costs to achieve the best results and more closely match the underlying hardware.

The updated vectorization costs information for AmpereOne halve the store cost, drop the align/unaligned load costs from 5 to 4, and other reductions but increases the floating point scalar cost from 1 to 3.

Long story short, some last minute tuning for AmpereOne has made it into GCC 13 ahead of the upcoming GCC 13.1 stable release in the next few weeks for benefiting those that may be targeting -march=ampere1 for optimized performance on these Ampere Computing processors. GCC 13 also introduces a new Ampere-1A variant among the many changes in this annual open-source compiler feature release.

AmpereOne is Ampere Computing's AArch64 competition for the likes of AMD 4th Gen EPYC "Genoa" and Intel Sapphire Rapids with 5nm TSMC manufacturing, DDR5 system memory, PCIe 5.0, and many other improvements in the Arm server chip space. AmpereOne was originally expected to launch in 2022 but so far hasn't been officially unveiled.