libera #monero-pow

20 Oct 2023

Monero's Proof of Work algorithm (RandomX) and related topics. Use #p2pool-log and #p2pool-mini for P2Pool discussions. Anti-spam active: PM hyc for voice.

• 07:16
  sech1
  tevador/RandomX #282
• 07:17
  sech1
  Not an issue right now, but better to fix it before it bites us in the future
• 12:47
  selsta
  should we tag a new randomx release, or just update the submodule to the newest commit?
• 12:50
  sech1
  JIT compiler for RISC-V, macOS Apple silicon fix. Yes, it's worth enough to make a release
• 12:51
  sech1
  One more small PR tevador/RandomX #284 and we can tag
• 14:40
  tevador
  tevador/RandomX #285
• 14:40
  tevador
  I will tag after this
• 14:47
  tevador
  tagged v1.2.0
• 15:06
  tevador
  Strange, Monero RISC-V build fails
• 15:11
  sech1
  fails where?
• 15:12
  selsta
  github.com/monero-project/monero/ac…/6589422571/job/17903873454?pr=9026
• 15:12
  selsta
  linking error
• 15:13
  tevador
  does monero set ARCH to something other than riscv64?
• 15:13
  tevador
  because the jit compiler files are not added to the build
• 15:22
  m-relay
  <t​obtoht:monero.social> depends toolchain.cmake sets it to rv64imafdc
• 15:23
  tevador
  that would explain it
• 15:24
  tevador
  but it's dumb, the architecture is riscv64, i.e. 64-bit RISC-V
• 15:26
  tevador
  github.com/monero-project/monero/blob/master/CMakeLists.txt#L328
• 15:28
  m-relay
  <t​obtoht:monero.social> try changing it here: github.com/monero-project/monero/bl…rib/depends/toolchain.cmake.in#L148
• 15:30
  tevador
  set(NO_AES ON) <-- is this needed?
• 15:31
  m-relay
  <t​obtoht:monero.social> does riscv64 have hardware aes? if yes, then no
• 15:31
  tevador
  the architecture has it, but it's an optional extension
• 15:32
  m-relay
  <t​obtoht:monero.social> binaries won't start on board that don't have it
• 15:33
  tevador
  that's the same for x86 or ARM
• 15:34
  tevador
  but they don't set NO_AES there
• 15:37
  tevador
  NO_AES is only used in crypto/slow-hash.c together with x86 and arm-specific defines, so it should be safe to remove it
• 15:37
  m-relay
  <t​obtoht:monero.social> oh it's always disabled github.com/monero-project/monero/blob/master/CMakeLists.txt#L747
• 15:37
  tevador
  it seems to force no AES even if the code supports it
• 15:38
  m-relay
  <t​obtoht:monero.social> "so it should be safe to remove it" <- yeah
• 15:52
  tevador
  after changing the ARCH to "riscv64", the RISC-V build is taking forever, seems to be building boost, openssl, sodium etc. from source
• 15:53
  selsta
  it's because it doesn't use ccache this time due to changing ARCH
• 15:56
  tevador
  Failed. Seems to be setting -march=${ARCH_ID}, where can we disable this?
• 15:59
  tevador
  and the randomx build still failed to include jit_compiler_rv64.cpp
• 16:36
  m-relay
  <t​obtoht:monero.social> hmm github.com/tobtoht/monero/actions/r…133669/job/17906111979#step:9:13888
• 16:41
  tevador
  must be some ancient assembler version
• 16:43
  tevador
  github.com/tevador/RandomX/actions/…589233240/job/17903232776#step:5:59
• 16:45
  tevador
  Which gcc version is it using? The RandomX CI has gcc 13.1.1.
• 16:46
  m-relay
  <t​obtoht:monero.social> 9.4.0
• 17:02
  sech1
  GCC 9 is ancient and received the latest update in May 2022
• 17:03
  sech1
  It's better to use at least GCC 11 which is still being supported: github.com/gcc-mirror/gcc/tree/releases/gcc-11 (last commit 7 hours ago)
• 17:03
  tevador
  I can fix zext.b in RandomX if needed.
• 17:03
  tevador
  If we really want to support ancient gcc versions...
• 17:04
  sech1
  Not for RISC-V probably
• 17:04
  sech1
  because all future RISC-V end users will install fres OS on their boards
• 17:04
  sech1
  *fresh
• 17:04
  sech1
  I'd say require GCC 11 for RISC-V
• 17:05
  tevador
  GCC 9.4 is from June 2021
• 17:06
  sech1
  hmm, GCC 9.5 exists: gcc-mirror/gcc 7a15b50
• 17:12
  tevador
  godbolt.org/z/Yb56cv8WW
• 17:13
  sech1
  huh
• 17:13
  tevador
  works even with gcc 8.2, which is the oldest one they have here
• 17:15
  sech1
  then it must be something wrong with compiler flags
• 17:15
  m-relay
  <t​obtoht:monero.social> i'll try changing the isa string to match that
• 17:15
  m-relay
  <t​obtoht:monero.social> missing extension maybe
• 17:17
  tevador
  zext.b is an alias
• 17:17
  tevador
  for andi r, r, 255
• 17:33
  tevador
  tobtoht: tevador/RandomX #286
• 17:34
  m-relay
  <t​obtoht:monero.social> ok will pick that
• 17:34
  tevador
  try just set(ARCH_ID "riscv64") in toolchain.cmake
• 17:35
  m-relay
  <t​obtoht:monero.social> remove the set(ARCH)?
• 17:35
  tevador
  is it needed?
• 17:35
  m-relay
  <t​obtoht:monero.social> lets see
• 17:36
  tevador
  RandomX has this: github.com/tevador/RandomX/blob/master/CMakeLists.txt#L71-L73
• 17:36
  tevador
  not sure if Monero needs it
• 17:41
  tevador
  if that PR helps, we'll have to tag 1.2.1
• 17:44
  m-relay
  <t​obtoht:monero.social> (1.2.0 builds fine when cherry-picked on top of #8929, uses gcc 10.3 - not important for now)
• 17:44
  m-relay
  <t​obtoht:monero.social> trying 286
• 18:19
  m-relay
  <t​obtoht:monero.social> works github.com/tobtoht/monero/actions/r…988466/job/17908674745#step:9:13863
• 18:19
  m-relay
  <t​obtoht:monero.social> gcc docs seem to suggest that default is system dependent for riscv, which could introduce non-determinism
• 18:20
  m-relay
  <t​obtoht:monero.social> so better to set ARCH explicitely
• 18:20
  m-relay
  <t​obtoht:monero.social> gcc.gnu.org/onlinedocs/gcc/RISC-V-Options.html#index-march-14
• 18:25
  tevador
  I thought the default was rv64gc
• 18:26
  m-relay
  <t​obtoht:monero.social> actually i'm not sure if it matters, since we're cross-compiling anyway
• 18:26
  tevador
  At least RandomX assumes rv64gc
• 18:28
  m-relay
  <t​obtoht:monero.social> ok i have this tobtoht/monero e2fd69a
• 18:57
  m-relay
  <t​obtoht:monero.social> don't forget release
• 18:58
  tevador
  I'm waiting for CI to success on master
• 18:58
  tevador
  succeed*
• 19:10
  tevador
  that risc-v build took forever
• 19:12
  m-relay
  <t​obtoht:monero.social> like watching paint dry
• 19:14
  sech1
  It runs risc-v build in qemu (as well as other non-x86 builds)
• 19:15
  m-relay
  <t​obtoht:monero.social> ci cross-compiles
• 19:15
  sech1
  Oh, I though you were talking about RandomX CI
• 19:16
  m-relay
  <t​obtoht:monero.social> wish those runners had more than 2 vcpus
• 19:17
  sech1
  Self-hosted runners can
• 19:17
  sech1
  Someone can CCS that
• 19:17
  m-relay
  <t​obtoht:monero.social> didn't ginger have CCS for a speccy machine?
• 19:18
  m-relay
  <t​obtoht:monero.social> is that thing still running
• 19:19
  m-relay
  <p​lowsof:matrix.org> last i heard from gingeropolous is that its available for devs .. Rucknium has used it in the past for things
• 20:10
  gingeropolous
  yep all my mining rigs are available for monero R&D
• 21:13
  tevador
  Now it's time to set up gitian for RISC-V so Monero can release riscv64 binaries next time.
• 21:49
  selsta
  working on it
• 22:20
  pauliouk
  happy 3000000
• 23:22
  m-relay
  <r​ecanman:agoradesk.com> Yay