Remove conflicting declaration for _mulx_u64.

CHANGELOG.md

@@ -1,6 +1,6 @@
 # v0.6.0
 - Added automatic cryptonight self test.
-- Added support for software AES `--av=4`. Will be automatically selected if cpu not support AES-NI.
+- New software AES algorithm variation `--av=4`. Will be automatically selected if cpu not support AES-NI.
 - Added 32 bit builds.
 - Documented [algorithm variations](https://github.com/xmrig/xmrig#algorithm-variations).
 

README.md

@@ -15,9 +15,9 @@ Based on cpuminer-multi with heavy optimizations/rewrites and removing a lot of
 * [Donations](#Donations)
 
 ## Features
-* High performance, faster than others (290+ H/s on i7 6700).
+* High performance (290+ H/s on i7 6700).
 * Official Windows support.
-* Small Windows executable, only 350 KB without dependencies.
+* Small Windows executable, only 430 KB without dependencies.
 * Support for backup (failover) mining server.
 * keepalived support.
 * Command line options compatible with cpuminer.

algo/cryptonight/cryptonight_av1_aesni.c

@@ -30,16 +30,6 @@
 #include "crypto/c_keccak.h"
 
 
-#ifndef __BMI2__
-static inline uint64_t _mulx_u64(uint64_t a, uint64_t b, uint64_t* hi)
-{
-    unsigned __int128 r = (unsigned __int128) a * (unsigned __int128) b;
-    *hi = r >> 64;
-    return (uint64_t) r;
-}
-#endif
-
-
 void cryptonight_av1_aesni(const void *restrict input, size_t size, void *restrict output, struct cryptonight_ctx *restrict ctx)
 {
     keccak((const uint8_t *) input, size, ctx->state, 200);
@@ -67,7 +57,7 @@ void cryptonight_av1_aesni(const void *restrict input, size_t size, void *restri
         uint64_t hi, lo, cl, ch;
         cl = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[0];
         ch = ((uint64_t*) &l0[idx0 & 0x1FFFF0])[1];
-        lo = _mulx_u64(idx0, cl, &hi);
+        lo = _umul128(idx0, cl, &hi);
 
         al0 += hi;
         ah0 += lo;

algo/cryptonight/cryptonight_av2_aesni_stak.c

@@ -30,16 +30,6 @@
 #include "crypto/c_keccak.h"
 
 
-#ifndef __BMI2__
-static inline uint64_t _mulx_u64(uint64_t a, uint64_t b, uint64_t* hi)
-{
-    unsigned __int128 r = (unsigned __int128) a * (unsigned __int128) b;
-    *hi = r >> 64;
-    return (uint64_t) r;
-}
-#endif
-
-
 void cryptonight_av2_aesni_stak(const void *restrict input, size_t size, void *restrict output, struct cryptonight_ctx *restrict ctx)
 {
     keccak((const uint8_t *) input, size, ctx->state, 200);
@@ -64,12 +54,10 @@ void cryptonight_av2_aesni_stak(const void *restrict input, size_t size, void *r
         idx0 = _mm_cvtsi128_si64(cx);
         bx0 = cx;
 
-         _mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
-
         uint64_t hi, lo, cl, ch;
         cl = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0];
         ch = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1];
-        lo = _mulx_u64(idx0, cl, &hi);
+        lo = _umul128(idx0, cl, &hi);
 
         al0 += hi;
         ah0 += lo;
@@ -80,8 +68,6 @@ void cryptonight_av2_aesni_stak(const void *restrict input, size_t size, void *r
         ah0 ^= ch;
         al0 ^= cl;
         idx0 = al0;
-
-        _mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
     }
 
     cn_implode_scratchpad((__m128i*) ctx->memory, (__m128i*) ctx->state);

algo/cryptonight/cryptonight_p.h

@@ -212,4 +212,12 @@ inline void cn_implode_scratchpad(const __m128i* input, __m128i* output)
 }
 
 
+inline uint64_t _umul128(uint64_t a, uint64_t b, uint64_t* hi)
+{
+    unsigned __int128 r = (unsigned __int128) a * (unsigned __int128) b;
+    *hi = r >> 64;
+    return (uint64_t) r;
+}
+
+
 #endif /* __CRYPTONIGHT_P_H__ */