Skip to content

API.md

Latest commit

 

History

History
81 lines (50 loc) · 2.33 KB

API.md

File metadata and controls

81 lines (50 loc) · 2.33 KB
Raw

Library API (commercial DLL)

Today: evaluation DLL in Releases. Full CUDA source — commercial license.

Header: include/mldsa_gpu_verify.h

In short

  1. Create a context once per public key (key material is cached on GPU).
  2. Call verify on batches of signatures — get OK/FAIL per signature.
  3. Destroy the context when finished.
MldsaGpuVerifyCtx* ctx = mldsa_gpu_verify_ctx_create(2, public_key);

int ok[1000];
mldsa_gpu_verify_batch(ctx, ok, 1000, signatures, messages, msg_length);

mldsa_gpu_verify_ctx_destroy(ctx);

Functions

mldsa_gpu_verify_version

const char* mldsa_gpu_verify_version(void);

Returns version string (e.g. "0.2.0-eval").

mldsa_gpu_verify_ctx_create

MldsaGpuVerifyCtx* mldsa_gpu_verify_ctx_create(int mode, const uint8_t* pk);
  • mode = 2 → Dilithium2 (available). Modes 3 and 5 planned.
  • pk → 1312 bytes (mode 2).
  • Returns NULL on error.

Unpacks the key on CPU, uploads matrix / t1 / tr to GPU — done once, not per signature.

mldsa_gpu_verify_batch

int mldsa_gpu_verify_batch(ctx, results, count, sigs, msgs, msglen);
  • count — batch size (512–4096 recommended for throughput).
  • sigs[i] — 2420 bytes each (mode 2).
  • results[i] — 1 valid, 0 invalid.
  • Returns 0 on success, -1 on GPU/memory error.

All signatures must use the same public key as the context.

mldsa_gpu_verify_batch_replay

mldsa_gpu_verify_batch_replay(ctx, count);

Re-runs the last uploaded batch without host repack — used in our throughput benchmark.

Your responsibility before calling verify

The library checks raw message bytes. If you are on a blockchain or custom protocol, you must serialize/hash the transaction exactly as the network expects, then pass those bytes in.

We do not ship Bitcoin / Ethereum / NEAR parsers out of the box — chain adapters are a separate integration (available under commercial license).

Evaluation vs production

The evaluation DLL is free to download from Releases for testing on your hardware. Production deployment and CUDA source require a commercial license — contact @MaxVip124.