New Intel Lunar Lake / Arrow Lake / Arrow Lake S Patches For GCC
Yesterday Intel engineers sent out early compiler patches for Lunar Lake and Arrow Lake with adding the new instructions of AVX-VNNI-INT16, SM3, SHA512, and SM4. Today that new instruction support was complemented by a new patch out of Intel for actually adding the new Lunar Lake, Arrow Lake, and Arrow Lake S targets to GCC.
The patch adds in the arrowlake / arrowlake-s / lunarlake targets so that this initial support will be ready for the GCC 14 compiler release next year. This patch doesn't provide any new cost table / tuning model for these future generation processors that will succeed Meteor Lake.
At least from the documentation added, Arrow Lake S will support AVX-VNNI-INT16, SHA512, SM3, and SM4 but apparently not other Arrow Lake models.
Today's patch can be found on gcc-patches and should quite quickly work its way to GCC Git. It's great as always seeing all of Intel's early open-source and upstream-focused enablement work for their future generation products -- for both client and server hardware. Meteor Lake support has been present upstream since GCC 13 while on the other side of the table so far we haven't seen any AMD Zen 5 (znver5) compiler patches. But if AMD's past traditions there hold true, we won't see any znver5 patches for GCC or LLVM/Clang until those processors are actually announced -- a very different approach from how Intel is focused on getting their compiler support upstreamed early and into released compiler versions ahead of launch.
The patch adds in the arrowlake / arrowlake-s / lunarlake targets so that this initial support will be ready for the GCC 14 compiler release next year. This patch doesn't provide any new cost table / tuning model for these future generation processors that will succeed Meteor Lake.
At least from the documentation added, Arrow Lake S will support AVX-VNNI-INT16, SHA512, SM3, and SM4 but apparently not other Arrow Lake models.
Today's patch can be found on gcc-patches and should quite quickly work its way to GCC Git. It's great as always seeing all of Intel's early open-source and upstream-focused enablement work for their future generation products -- for both client and server hardware. Meteor Lake support has been present upstream since GCC 13 while on the other side of the table so far we haven't seen any AMD Zen 5 (znver5) compiler patches. But if AMD's past traditions there hold true, we won't see any znver5 patches for GCC or LLVM/Clang until those processors are actually announced -- a very different approach from how Intel is focused on getting their compiler support upstreamed early and into released compiler versions ahead of launch.
1 Comment