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

RandomX: optimized program generation

Browse source
This commit is contained in:
SChernykh 2023-05-21 00:39:32 +02:00
parent 7b51e23aa0
commit 5dcbab7e3a
4 changed files with 113 additions and 73 deletions
Download patch file Download diff file Expand all files Collapse all files

14
src/crypto/randomx/intrin_portable.h
View file

@ -126,6 +126,7 @@ FORCE_INLINE rx_vec_f128 rx_set1_vec_f128(uint64_t x) {
#define rx_xor_vec_f128 _mm_xor_pd
#define rx_and_vec_f128 _mm_and_pd
#define rx_and_vec_i128 _mm_and_si128
#define rx_or_vec_f128 _mm_or_pd
#ifdef __AES__
@ -278,6 +279,10 @@ FORCE_INLINE rx_vec_f128 rx_and_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
return (rx_vec_f128)vec_and(a,b);
}
FORCE_INLINE rx_vec_i128 rx_and_vec_i128(rx_vec_i128 a, rx_vec_i128 b) {
return (rx_vec_i128)vec_and(a, b);
}
FORCE_INLINE rx_vec_f128 rx_or_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
return (rx_vec_f128)vec_or(a,b);
}
@ -444,6 +449,8 @@ FORCE_INLINE rx_vec_f128 rx_and_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
return vreinterpretq_f64_u8(vandq_u8(vreinterpretq_u8_f64(a), vreinterpretq_u8_f64(b)));
}
#define rx_and_vec_i128 vandq_u8
FORCE_INLINE rx_vec_f128 rx_or_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
return vreinterpretq_f64_u8(vorrq_u8(vreinterpretq_u8_f64(a), vreinterpretq_u8_f64(b)));
}
@ -635,6 +642,13 @@ FORCE_INLINE rx_vec_f128 rx_and_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
return x;
}
FORCE_INLINE rx_vec_i128 rx_and_vec_i128(rx_vec_i128 a, rx_vec_i128 b) {
rx_vec_i128 x;
x.u64[0] = a.u64[0] & b.u64[0];
x.u64[1] = a.u64[1] & b.u64[1];
return x;
}
FORCE_INLINE rx_vec_f128 rx_or_vec_f128(rx_vec_f128 a, rx_vec_f128 b) {
rx_vec_f128 x;
x.i.u64[0] = a.i.u64[0] | b.i.u64[0];