fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

soft_aes: fix previous optimization

Browse source 
Previously removed unrolled variant is faster on some CPUs
Some CPUs are faster with added unrolled variant
The best variant depends on number of threads on some CPUs
This commit is contained in:
cohcho 2020-10-04 14:47:58 +00:00
parent 31e896feef
commit 5f0f2506e8
6 changed files with 116 additions and 72 deletions
Download patch file Download diff file Expand all files Collapse all files

13
src/crypto/randomx/aes_hash.hpp
View file

@ -30,6 +30,17 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#include <cstddef>
typedef void (hashAndFillAes1Rx4_impl)(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state);
extern hashAndFillAes1Rx4_impl* softAESImpl;
inline hashAndFillAes1Rx4_impl* GetSoftAESImpl()
{
return softAESImpl;
}
void SelectSoftAESImpl(size_t threadsCount);
template<int softAes>
void hashAes1Rx4(const void *input, size_t inputSize, void *hash);
@ -39,5 +50,5 @@ void fillAes1Rx4(void *state, size_t outputSize, void *buffer);
template<int softAes>
void fillAes4Rx4(void *state, size_t outputSize, void *buffer);
template<int softAes>
template<int softAes, int unroll>
void hashAndFillAes1Rx4(void *scratchpad, size_t scratchpadSize, void *hash, void* fill_state);