Follow-up to PR #1977. Enables the disabled (#if 0) adler32_copy_avx2_vnni copy variant by
rebuilding it on the same structure as the shipping checksum-only path.
The author left adler32_copy_avx2_vnni under #if 0 with a note about a "data dependency
bubble," faster on Gracemont E-cores but slower on Raptor Cove P-cores, and a conversation
comment: "I think there may need to be more unrolling."
Follow-up idea spun off from PR #1977 (AVX2-VNNI adler32).
adler32_avx512_vnni uses only 2 VPDPBUSD accumulators (vs2, vs2_1). VPDPBUSD
has ~5-cycle latency, so two in-flight dot products can't cover it and the 128-byte loop is
latency-bound at ~5c/128B — the same ceiling as the AVX2-VNNI path.
| name: nathan | |
| description: Configuration files synced by config-sync | |
| url: https://github.com/nmoinvaz/gcs | |
| files: | |
| - path: .claude/skills/security-evidence/SKILL.md | |
| gist: nathan_.claude_skills_security-evidence_SKILL.md | |
| updated_at: 2026-06-12T01:39:34.319047220Z | |
| - path: .claude/skills/security-evidence/scripts/ss-capture.sh | |
| gist: nathan_.claude_skills_security-evidence_scripts_ss-capture.sh | |
| updated_at: 2026-06-12T01:38:43.526802792Z |
Follow-up to PR #2291 (nv/develop/deflate-strategy-locals-hoist). After the scalar-local hoist landed, tried two variants in deflate_quick.c to see whether packing the two uint32_t fields into a struct could coax the compiler into 64-bit load/store transfers.
CMAKE_OSX_ARCHITECTURES=x86_64-D BUILD_SHARED_LIBS=OFF
Branch: nv/develop/deflate-struct-hoist
Commit: c435d01e — "Hoist deflate bit-emit state into deflate_emit_hot local"
Base: upstream/develop (48087450)
Introduces a deflate_emit_hot struct (bi_buf, bi_valid, pending_buf, pending) and DEFLATE_EMIT_HOT_LOAD/STORE macros that cache bit-emit state in a local at the top of hot loops and write it back once on exit. Converts put_byte/put_short/put_short_msb/put_uint32/put_uint32_msb/put_uint64 from macros to static inline functions taking a deflate_emit_hot *. Cold-path callers in deflate.c (zlib/gzip header and trailer writers, name/comment loops) and deflate_stored.c bracket their put_* clusters with LOAD/STORE.
| /* benchmark_chunkmemset.cc -- benchmark chunkmemset_safe variants | |
| * Copyright (C) 2026 Nathan Moinvaziri | |
| * For conditions of distribution and use, see copyright notice in zlib.h | |
| */ | |
| #include <benchmark/benchmark.h> | |
| extern "C" { | |
| # include "zbuild.h" | |
| # include "zutil_p.h" |
| /* benchmark_dist1.cc -- compare strategies for the dist=1 path in CHUNKMEMSET */ | |
| #include <benchmark/benchmark.h> | |
| extern "C" { | |
| # include "zbuild.h" | |
| # include "zutil.h" | |
| } | |
| #if defined(__ARM_NEON) || defined(__ARM_NEON__) |
| diff --git a/deflate_quick.c b/deflate_quick.c | |
| index 6b84388e..41086525 100644 | |
| --- a/deflate_quick.c | |
| +++ b/deflate_quick.c | |
| @@ -49,13 +49,19 @@ Z_INTERNAL block_state deflate_quick(deflate_state *s, int flush) { | |
| unsigned char *window; | |
| unsigned last = (flush == Z_FINISH) ? 1 : 0; | |
| + /* Hold strstart and lookahead in registers across the inner loop. | |
| + * Sync to s before fill_window (which mutates them) and quick_*_block |
Inspecting AArch64 disassembly of zlib-ng's level-9 deflate path showed that the rolling-hash insert loop emitted a redundant store of the running hash to memory on every byte processed. The cause was visible in the C source and the fix is small and self-contained.
PR: zlib-ng/zlib-ng#2281 — [CI] Cache Ubuntu .deb packages to speed up installing dependencies.
Follow-up to https://gist.github.com/nmoinvaz/978f248ea7d528c954e3d52fb8dc99c0 — measuring the v2 design after the author refactored into composite actions and adopted the matrix.packages gate.
v2 is a clean win across the board. Total apt-install time across Ubuntu jobs is now −71% vs. develop, up from −47% in v1. Default-only jobs no longer pay cache overhead, and warm cache hits skip the write-back step entirely.