We Say Pack and Unpack intrinsics in the TPC document, but we don’t know what does it do.
Is there any introduction or example about this instructions?
We have a few intrinsic instructions for PACK and UNPACK, check the doc Pack/Unpack — Gaudi Documentation for more details.
We also have the custom TPC kernel example in our github repo, cast_bf16_to_f32.c in gaudi kernel, you can get an idea how pack and unpack work.
Habana_Custom_Kernel/cast_bf16_to_f32.c at main · HabanaAI/Habana_Custom_Kernel · GitHub