Flaky `crypto.kangarootwelve.test.KT256 sequential and parallel produce same output for large inputs` failure on LoongArch CI

[alexrp]

First time seeing this:

mq test-modules
   mq test-std
      mq run test std-native-la664-ReleaseSmall-libc 2968 pass, 33 skip, 1 fail (3002 total)
error: 'crypto.kangarootwelve.test.KT256 sequential and parallel produce same output for large inputs' failed:
       slices differ. first difference occurs at index 0 (0x0)
       
       ============ expected this output: =============  len: 64 (0x40)
       
       7B 4B 6E 1C B2 EF 50 16  CE A6 54 36 C4 3A 7B A6  {Kn...P...T6.:{.
       1C 84 93 39 65 B1 48 55  36 78 9D B3 91 F3 38 75  ...9e.HU6x....8u
       C3 C4 C7 32 98 89 49 C0  A1 67 F4 01 85 FA D7 90  ...2..I..g......
       2F 6E 8C E2 B5 12 6C 2E  5B 04 2D A8 DE D4 08 52  /n....l.[.-....R
       
       ============= instead found this: ==============  len: 64 (0x40)
       
       0B 58 7C B6 EA BF 2E 81  E9 27 5A 97 53 6C 5D 8F  .X|......'Z.Sl].
       27 66 D2 D9 B4 A3 F1 CB  A1 FE 1E 46 89 76 00 CB  'f.........F.v..
       CE DD 8C E2 2D 2E 39 24  DC 55 CD 37 E6 AB CB 70  ....-.9$.U.7...p
       F2 03 48 14 74 95 CD 9A  01 A0 6D 05 43 A1 9E 24  ..H.t.....m.C..$
       
       ================================================
failed command: /home/ci/.cache/act/b21a761523900bc9/hostexecutor/zig-local-cache/o/7644d75574707a8ca5893fa9ef4e230e/test --cache-dir=/home/ci/.cache/act/b21a761523900bc9/hostexecutor/zig-local-cache --seed=0x55ed4980 --listen=-
]9;4
test transitive failure
mq test-modules transitive failure
   mq test-std transitive failure
      mq run test std-native-la664-ReleaseSmall-libc 2968 pass, 33 skip, 1 fail (3002 total)
error: the following build command failed with exit code 1:
/home/ci/.cache/act/b21a761523900bc9/hostexecutor/zig-local-cache/o/e31e9c0d64bab1c851db42a3f6385e69/build /home/ci/.cache/act/b21a761523900bc9/hostexecutor/build-release/stage3-release/bin/zig /home/ci/.cache/act/b21a761523900bc9/hostexecutor/lib /home/ci/.cache/act/b21a761523900bc9/hostexecutor /home/ci/.cache/act/b21a761523900bc9/hostexecutor/zig-local-cache /home/ci/.cache/act/b21a761523900bc9/hostexecutor/zig-global-cache --seed 0x55ed4980 -Z329e1ae7a334aa22 test docs --maxrss 60129542144 -Dstatic-llvm -Dskip-non-native -Dtarget=native-native-musl --search-prefix /home/ci/deps/zig+llvm+lld+clang-loongarch64-linux-musl-0.16.0-dev.157+7fdd60df1 --test-timeout 4m

cc @jedisct1

Given this is testing a parallel algorithm, it seems likely that this is a legitimate bug and not some LoongArch compiler issue or similar.

[jacobly0]

Also happened on mips.

[jedisct1]

I'm having a hard time reproducing it.

But Loongarch and Mips have weak memory ordering, so it's possible that some CVs read right after the last thread exits don't match what was just written. Atomic writes of the CVs should have minimal overhead and make this more reliable on these architectures.