void ScryptCore(UInt32 x[32], UInt32 alignedScratch[1024*32+32]) {
DECLSPEC_ALIGN(128) UInt32 w[32];
ShuffleForSalsa(w, x);
#if UCFG_USE_MASM && !defined(_M_X64)
if (s_bHasSse2)
ScryptCore_x86x64(w, alignedScratch);
else
#endif
{
for (int i=0; i<1024; ++i) {
memcpy(alignedScratch+i*32, w, 32*sizeof(UInt32));
Salsa20Core(w, w+16, 8);
Salsa20Core(w+16, w, 8);
}
for (int i=0; i<1024; ++i) {
int j = w[16] & 1023; // w[16] is fixed point
UInt32 *p = alignedScratch+j*32;
for (int k=0; k<32; ++k)
w[k] ^= p[k];
Salsa20Core(w, w+16, 8);
Salsa20Core(w+16, w, 8);
}
}
UnShuffleForSalsa(x, w);
}
void ScryptCore(uint32_t x[32], uint32_t alignedScratch[1024*32+32]) noexcept {
#if UCFG_USE_MASM
if (s_bHasSse2) {
DECLSPEC_ALIGN(128) uint32_t w[32];
ShuffleForSalsa(w, x);
ScryptCore_x86x64(w, alignedScratch);
UnShuffleForSalsa(x, w);
} else
#endif
{
uint32_t tmp[2][16];
NeoScryptCore((SalsaBlockPtr)x, tmp, alignedScratch, 1, 8, 1024, false);
/*!!!
memcpy(w, x, 32 * sizeof(uint32_t));
for (int i=0; i<1024; ++i) {
memcpy(alignedScratch+i*32, w, 32*sizeof(uint32_t));
VectorXor(w, w+16, 16);
Salsa20Core(w, 8);
VectorXor(w+16, w, 16);
Salsa20Core(w+16, 8);
}
for (int i=0; i<1024; ++i) {
int j = w[16] & 1023; // w[16] is fixed point
uint32_t *p = alignedScratch+j*32;
for (int k=0; k<32; ++k)
w[k] ^= p[k];
VectorXor(w, w+16, 16);
Salsa20Core(w, 8);
VectorXor(w+16, w, 16);
Salsa20Core(w+16, 8);
}
memcpy(x, w, 32 * sizeof(uint32_t));
*/
}
}