Optimize TCipherKernelRegistry for allocation-free reads and EC multipliers - #169
Merged
Conversation
The read path (Snapshot/GetSnapshot/TryAcquire*) allocated and refcounted a fresh factory array on every call. Register/Unregister now rebuild an immutable copy-on-write snapshot under the write lock; the read takes a single reference load of it - no per-call allocation. The lock is held only for that reference load (not the copy, not the factory TryCreate), so it stays uncontended and, unlike a lock-free read, keeps runtime registration concurrent with acquisition thread-safe - matching the guarantee the external-registration API implies.
Replace the array-based constant-time EC point multiplier with an allocation-free, value-type implementation over stack records, wired as the default multiplier for P-256, secp256k1, P-384 and P-521. Core layer (Math/EC/Multiplier): - ClpCTFieldValue: TFe/TFeExt/TFePoint fixed-size stack records (no heap). - ClpCTFieldOps: TCTFieldOpsBase, virtual/abstract class-method field ops overridden per curve — the one dispatch form that is refcount-free on FPC x86_64/i386 and Delphi. - ClpCTLadder: generic TCTLadder<TOps> — RCB2016 complete add/double (Algorithms 1 & 3) over TFePoint. - ClpFpCTMultiplier: generic TFpCTMultiplier<TOps> (IECMultiplier) — the hot loop (fixed 4-bit window, masked table lookup, one unconditional add per window, scalar blinding, randomized projective coords). Curve context (order, affine conversion, inverse) via IFpFieldOps. SIMD Fp multiply/square kernel (Simd/Facade + Simd/Backend + Include/Simd): - Width-general 64-bit-limb schoolbook mul/sqr for x86_64, i386, aarch64, behind an arch-neutral TFpKernelSimd.TryMul/TrySqr facade; scalar bridge in the caller when unavailable. Per-curve field ops gained TFe overloads over PUInt32 cores (kernel path alloc-free); TNat/TNat256 gained PUInt32 overloads (Add/Sub/Gte/Inc/Dec/ Eq/AddTo/ShiftDownBits/Mul33*/IncAt/DecAt/Add33To/Sub33From) with every array form delegating to the pointer core. P-521 mixed-width: no new asm — the 17 real limbs are padded to 18 uint32 (= 9 uint64) with a zeroed top limb so the even-width kernel applies; W[17] kept zero on every field-op output. Cleanup: retire the now-dead array multiplier and its math. - Delete ClpFixedWindowCTMultiplier and ClpHomogeneousPoint (TCTHomogeneous- Math); remove from all FPC and Delphi manifests. - Slim IFpFieldOps: drop Square/Add/MulByA/MulByB3 (unused after the move). - Retarget the exceptional-formula test to the live TCTLadder and the blind-bits test to TFpCTMultiplier. - Extend the dudect harness to all four prime curves.
…ayer refactor Continues the committed value-type constant-time scalar multiplier (44a951a) for the Fp prime curves. Fixed-base comb (ECDSA signing) - Add TFpCombMultiplier<TOps>: a value-type Lim-Lee comb over stack TFePoint via the RCB complete formulas, reusing TFixedPointUtilities' generator table and caching the value-type table on the precomp-info object. Constant-time: masked table lookup, branchless recoding, fixed loop length. - Wire it in through a base-point hook rather than the variable-base default: IECCurve.GetBasePointMultiplier + TECCurve.CreateBasePointMultiplier (default TFixedPointCombMultiplier), overridden per custom curve; TECDsaSigner consumes it. Beats OpenSSL's generic [k]G on secp256k1/384/521. a = -3 / a = 0 field specialization - Fold the curve coefficient into MulByA: -(x+x+x) via field adds on P-256/384/521, zero on secp256k1 (a=0), instead of a full field multiply by a constant. ~1.2-1.4x on [k]G; the RCB add/double formulas are unchanged. Drop the now-dead FFa constant from all four curves. Constant-time point-layer refactor - Merge the group law and the point-representation helpers into a single TCTPoint<TOps> (ClpCTLadder -> ClpCTPoint); "Ladder" was a misnomer for the RCB group law and clashed with the real TLopezDahabLadder on F2m. - Rename the value-type arithmetic ClpCTFieldOps/TCTFieldOpsBase -> ClpCTFieldArith/TCTFieldArithBase and per-curve TSecXXXOps -> TSecXXXFieldArith to end the name collision with the array-based IFpFieldOps adapter. The core is now a clean Value -> Arith -> Point trio.
- ClpFpCombMultiplier: hold the fixed-point precomp info as a TWeakRef to break the curve <-> comb reference cycle that leaked the curve/comb/adapter/table. - ClpPkixPolicyNode: use the implicit TWeakRef operators, matching convention. - ClpX509Certificate: thread-safe lazy caches (SigAlgName/PublicKey/ CachedEncoding) + Equals fast-path via a class-level critical section with double-checked locking.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.