Move x86 mul128 implementation

Makefile.am

@@ -51,7 +51,7 @@ minerd_SOURCES	= elist.h \
 		  crypto/hash.c \
 		  crypto/aesb.c
 if ARCH_x86
-minerd_SOURCES += sha2-x86.S scrypt-x86.S aesb-x86.S
+minerd_SOURCES += sha2-x86.S scrypt-x86.S aesb-x86.S crypto/aesb-x86-impl.c
 endif
 if ARCH_x86_64
 minerd_SOURCES += sha2-x64.S scrypt-x64.S aesb-x64.S

aesb-x86.S

@@ -19,20 +19,3 @@ _fast_aesb_single_round:
 fast_aesb_pseudo_round_mut:
 _fast_aesb_pseudo_round_mut:
     ret
-
-    .text
-    .globl mul128
-    .globl _mul128
-mul128:
-_mul128:
-#if defined(_WIN32) || defined(__CYGWIN__)
-	mov %rcx, %rax
-	mul %rdx
-	mov %r8, (%r8)
-#else
-	mov %rdx, %r8
-	mov %rdi, %rax
-	mul %rsi
-	mov %rdx, (%r8)
-#endif
-	ret

crypto/aesb-x86-impl.c

@@ -0,0 +1,27 @@
+#pragma once
+
+uint64_t mul128(uint64_t multiplier, uint64_t multiplicand, uint64_t* product_hi) {
+  // multiplier   = ab = a * 2^32 + b
+  // multiplicand = cd = c * 2^32 + d
+  // ab * cd = a * c * 2^64 + (a * d + b * c) * 2^32 + b * d
+  uint64_t a = hi_dword(multiplier);
+  uint64_t b = lo_dword(multiplier);
+  uint64_t c = hi_dword(multiplicand);
+  uint64_t d = lo_dword(multiplicand);
+
+  uint64_t ac = a * c;
+  uint64_t ad = a * d;
+  uint64_t bc = b * c;
+  uint64_t bd = b * d;
+
+  uint64_t adbc = ad + bc;
+  uint64_t adbc_carry = adbc < ad ? 1 : 0;
+
+  // multiplier * multiplicand = product_hi * 2^64 + product_lo
+  uint64_t product_lo = bd + (adbc << 32);
+  uint64_t product_lo_carry = product_lo < bd ? 1 : 0;
+  *product_hi = ac + (adbc >> 32) + (adbc_carry << 32) + product_lo_carry;
+  assert(ac <= *product_hi);
+
+  return product_lo;
+}