2026-03-01 00:48:14 i added some debugging code to my pipeline... on the (siLj_NXx) shared-runner nor-ci-1, which is crashing with SIGILL on a vector instruction, cpuinfo does claim vector support: model name: Spacemit(R) X60 isa: rv64imafdcv_zicbom_zicboz_zicntr_zicond_zicsr_zifencei_zihintpause_zihpm_zfh_zfhmin_zca_zcd_zba_zbb_zbc_zbs_zkt_zve32f_zve32x_zve64d_zve64f_zve64x_zvfh_zvfhmin_zvkt_sscofpmf_sstc_svinval_svnapot_svpbmt 2026-03-01 00:50:35 and, even if go is compiled with GORISCV64=rva20u64 (which is also the default), it will check at runtime whether the CPU supports RVV 1.0 (vector extensions). and it seems this CPU does, but regardless crashes with SIGILL at internal/bytealg/indexbyte_riscv64.s:83 2026-03-01 00:54:21 i am thinking, maybe we patch out the runtime check and always use the non-vector impls (which is what go1.25 does) and report this to the go people to let them sort it out? i'm not familiar with riscv64 cpu extensions 2026-03-01 06:54:50 the weird thing here is that the commits which implement the vector support in go specifically mention Spacemit X60, which is ostensibly the same CPU as is failing here 2026-03-01 07:29:33 ikke: you around? i could really use a rv64 shell for this instead of trying to hope my CI jobs get scheduled on the correct box :p 2026-03-01 07:32:22 Only on milk-v pioneer hw. nor-ci-1 is different 2026-03-01 07:32:53 that doesn't help unfortunately, shared-runner pioneer1 cpu does not support the vector insns and thus does not exhibit the issue 2026-03-01 07:34:56 is nor-ci-1 ncopa's perchance? 2026-03-01 07:37:13 Yes 2026-03-01 07:38:02 so i guess i need to talk to him unless i get lucky with CI scheduling 2026-03-01 07:38:46 ncopa: would you mind giving me a temp shell on nor-ci-1 for debugging a vector instruction issue on that specific cpu? 2026-03-01 22:45:11 the s390x runner is idle... but has not picked up any of the spirv rebuilds yet. maybe it's broken?