Looking at trending [0], there was a small progression in January (cause not known yet), notable progression in February (I bet related to change [1] in optimization flags) and regression in June when CSIT changed [2] the compilation to use platform cmake files. The CSIT change was needed for ipsec performance, so I guess the VPP ip6 code should be improved. For example, be more mindful of instruction cache considering some optimization flags can lead to longer code.
[0] ip6 2n-spr
[1] 44866: build: move from x86 vendor specific to generic variants | https://gerrit.fd.io/r/c/vpp/+/44866
[2] 44870: feat(gather): Compile VPP, platform if supported | https://gerrit.fd.io/r/c/csit/+/44870
Looking at trending [0], there was a small progression in January (cause not known yet), notable progression in February (I bet related to change [1] in optimization flags) and regression in June when CSIT changed [2] the compilation to use platform cmake files. The CSIT change was needed for ipsec performance, so I guess the VPP ip6 code should be improved. For example, be more mindful of instruction cache considering some optimization flags can lead to longer code.
[0] ip6 2n-spr
[1] 44866: build: move from x86 vendor specific to generic variants | https://gerrit.fd.io/r/c/vpp/+/44866
[2] 44870: feat(gather): Compile VPP, platform if supported | https://gerrit.fd.io/r/c/csit/+/44870