CelerisLab

Files

T

ClaudeandClaude Opus 5.5 781333407e kernels: one shared unforced MRT collision for blocked, channel and patch paths

BlockStep, BlockChannelStep and refinement.cu patch_collide each carried
the same compute_rho_u -> omega clamp -> collide_mrt_no_force sequence.
collide_mrt_unforced (collision_mrt.cuh, __forceinline__) now holds it
once; the three sites call it. Same operations in the same order: no
numerical, ABI or SRT change. The host source contract now requires one
helper definition and exactly one call per file, and forbids the old
inline sequence.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

2026-10-21 01:15:47 +00:00

audit

feat(compat): FP16S and ddf_shifting compatibility, EsoPull curved closure

2026-06-03 10:48:42 +08:00

integration

test: cover TRT restart and actual-rate refinement phase

2026-10-18 07:54:54 -07:00

postproc

fix(esopull): correct init layout and pre-streaming semantics (v0.5.1)

2026-06-27 22:32:01 +08:00

screening

fix(esopull): correct init layout and pre-streaming semantics (v0.5.1)