a655ec8a9d
* Add state destroy to SHA2 API * Include optimized SPHINCS+ implementations I've generated new implementations from the sphincsplus repository. * Don't destroy sha256ctx after finalize * Attempt to shut up MSVC * Make sure to drop errors in rmtree
27 lines
1009 B
C
27 lines
1009 B
C
#ifndef PQCLEAN_SPHINCSHARAKA128FROBUST_AESNI_HASH_STATE_H
|
||
#define PQCLEAN_SPHINCSHARAKA128FROBUST_AESNI_HASH_STATE_H
|
||
|
||
/**
|
||
* Defines the type of the hash function state.
|
||
*
|
||
* Don't be fooled into thinking this instance of SPHINCS+ isn't stateless!
|
||
*
|
||
* From Section 7.2.2 from the SPHINCS+ round-2 specification:
|
||
*
|
||
* Each of the instances of the tweakable hash function take PK.seed as its
|
||
* first input, which is constant for a given key pair – and, thus, across
|
||
* a single signature. This leads to a lot of redundant computation. To remedy
|
||
* this, we pad PK.seed to the length of a full 64-byte SHA-256 input block.
|
||
* Because of the Merkle-Damgård construction that underlies SHA-256, this
|
||
* allows for reuse of the intermediate SHA-256 state after the initial call to
|
||
* the compression function which improves performance.
|
||
*
|
||
* We pass this hash state around in functions, because otherwise we need to
|
||
* have a global variable.
|
||
*/
|
||
|
||
#include "haraka.h"
|
||
#define hash_state harakactx
|
||
|
||
#endif
|