Switch x86_64 asm to use "i" instead of "n" for immediate values.

The constraint "n" is a more obscure setting which is needed for
 special cases that don't apply to our use.

Use of "n" exposed some bugs in pre-release versions of clang.
This fixes bitcoin#594

src/scalar_4x64_impl.h

@@ -376,7 +376,7 @@ static void secp256k1_scalar_reduce_512(secp256k1_scalar *r, const uint64_t *l)
     /* extract m6 */
     "movq %%r8, %q6\n"
     : "=g"(m0), "=g"(m1), "=g"(m2), "=g"(m3), "=g"(m4), "=g"(m5), "=g"(m6)
-    : "S"(l), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
+    : "S"(l), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
     : "rax", "rdx", "r8", "r9", "r10", "r11", "r12", "r13", "r14", "cc");
 
     /* Reduce 385 bits into 258. */
@@ -455,7 +455,7 @@ static void secp256k1_scalar_reduce_512(secp256k1_scalar *r, const uint64_t *l)
     /* extract p4 */
     "movq %%r9, %q4\n"
     : "=&g"(p0), "=&g"(p1), "=&g"(p2), "=g"(p3), "=g"(p4)
-    : "g"(m0), "g"(m1), "g"(m2), "g"(m3), "g"(m4), "g"(m5), "g"(m6), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
+    : "g"(m0), "g"(m1), "g"(m2), "g"(m3), "g"(m4), "g"(m5), "g"(m6), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
     : "rax", "rdx", "r8", "r9", "r10", "r11", "r12", "r13", "cc");
 
     /* Reduce 258 bits into 256. */
@@ -501,7 +501,7 @@ static void secp256k1_scalar_reduce_512(secp256k1_scalar *r, const uint64_t *l)
     /* Extract c */
     "movq %%r9, %q0\n"
     : "=g"(c)
-    : "g"(p0), "g"(p1), "g"(p2), "g"(p3), "g"(p4), "D"(r), "n"(SECP256K1_N_C_0), "n"(SECP256K1_N_C_1)
+    : "g"(p0), "g"(p1), "g"(p2), "g"(p3), "g"(p4), "D"(r), "i"(SECP256K1_N_C_0), "i"(SECP256K1_N_C_1)
     : "rax", "rdx", "r8", "r9", "r10", "cc", "memory");
 #else
     uint128_t c;