Issues

• Mac: servil mt streams of 128 B to 64 KiB 4-41% slower under sustained load (bench 0.19) than in 1 ms bursts

  johnservil/BLAKE3#212 · Filter by author johnservil opened on Nov 1, 2026

• Core-to-core latency by address: could the pool place its hot synchronization lines on fast addresses?

  johnservil/BLAKE3#211 · Filter by author johnservil opened on Nov 1, 2026

• Mac: one-thread batches of 1-6 messages about 2 ns slower a call since the Batch change (5-11%, borderline)

  johnservil/BLAKE3#208 · Filter by author johnservil opened on Oct 29, 2026

• Mac: mt ManyMessages 4 and 6 slower at 0.8 than at base-0.8 (+9.6%, +6.9%, noisy)

  johnservil/BLAKE3#200 · Filter by author johnservil opened on Oct 27, 2026

• Ice Lake: mt ManyMessages 16384 14–18% slower at 0.8 than at base-0.8

  johnservil/BLAKE3#198 · Filter by author johnservil opened on Oct 27, 2026
  3

• Zen 4: mt IdleManyMessages 2 8–10% slower at 0.8 than at base-0.8

  johnservil/BLAKE3#199 · Filter by author johnservil opened on Oct 27, 2026
  1

• mt ContinuousMessages 256 B runs at one of two speeds a process (0.19 or 0.30 ns/B, EPYC 9255)

  johnservil/BLAKE3#196 · Filter by author johnservil opened on Oct 27, 2026
  15

• Two small-call cells on the VM, after 0.8: LentMessages 64 B one-thread +4.8%, LentBatches 16 +31% before a restart

  johnservil/BLAKE3#191 · Filter by author johnservil opened on Oct 27, 2026
  3

• After 0.8: the wide AVX-512 kernel costs 2-11% on large inputs hashed with every thread (EPYC)

  johnservil/BLAKE3#185 · Filter by author samshackleton-tech opened on Oct 26, 2026

• Submit the Lean specification to C2SP (c2sp/pr)

  johnservil/BLAKE3#167 · Filter by author johnservil opened on Oct 25, 2026

• Fil-C compatibility

  johnservil/BLAKE3#162 · Filter by author devonjonte opened on Oct 25, 2026
  1

• Windows: unsupported in 0.8; full support needs dedicated hardware

  johnservil/BLAKE3#160 · Filter by author devonjonte opened on Oct 25, 2026
  2

• Release C ABI review: output initialization, aliasing and callback contracts

  johnservil/BLAKE3#155 · Filter by author samshackleton-tech opened on Oct 25, 2026
  1

• Bob's check: the kernels' proofs with the least to trust

  johnservil/BLAKE3#130 · Filter by author johnservil opened on Oct 22, 2026

• Release 0.8: what it takes, step by step

  johnservil/BLAKE3#125 · Filter by author johnservil opened on Oct 21, 2026
  124

• A Merkle tree API, for callers like WHIR

  johnservil/BLAKE3#121 · Filter by author johnservil opened on Oct 21, 2026

• After 0.8: what the proofs could reach next

  johnservil/BLAKE3#120 · Filter by author johnservil opened on Oct 21, 2026

• After 0.8: speed ideas, each with its evidence

  johnservil/BLAKE3#118 · Filter by author johnservil opened on Oct 21, 2026
  7

• Revisit support for 32-bit targets (Arm Cortex-M and Armv7, 32-bit RISC-V)

  johnservil/BLAKE3#116 · Filter by author johnservil opened on Oct 21, 2026

• After 0.8: speed on AVX-512 and on Apple M6

  johnservil/BLAKE3#91 · Filter by author johnservil opened on Oct 19, 2026