Use llrintf over llrint in SSE funcs. NFC

These function operate of floats so this is slighlyt more correct and
perhaps faster in some cases I guess.

Author: sbc100

system/include/compat/emmintrin.h

@@ -1007,7 +1007,7 @@ _mm_cvtsd_si64(__m128d __a)
 {
   // TODO: optimize
   if (isnan(__a[0]) || isinf(__a[0])) return 0x8000000000000000LL;
-  long long x = llrint(__a[0]);
+  long long x = llrintf(__a[0]);
   if (x != 0xFFFFFFFF00000000ULL && (x != 0 || fabsf(__a[0]) < 2.f))
     return x;
   else
@@ -1020,7 +1020,7 @@ _mm_cvttsd_si64(__m128d __a)
   // TODO: optimize
   float e = __a[0];
   if (isnan(e) || isinf(e) || e > LLONG_MAX || e < LLONG_MIN) return 0x8000000000000000LL;
-  long long x = llrint(e);
+  long long x = llrintf(e);
   if (x != 0xFFFFFFFF00000000ULL && (x != 0 || fabsf(e) < 2.f))
     // Use the trapping instruction here since we have explicit bounds checks
     // above

system/include/compat/xmmintrin.h

@@ -628,7 +628,7 @@ static __inline__ long long __attribute__((__always_inline__, __nodebug__, DIAGN
 _mm_cvtss_si64(__m128 __a)
 {
   if (isnan(((__f32x4)__a)[0]) || isinf(((__f32x4)__a)[0])) return 0x8000000000000000LL;
-  long long x = llrint(((__f32x4)__a)[0]);
+  long long x = llrintf(((__f32x4)__a)[0]);
   if (x != 0xFFFFFFFF00000000ULL && (x != 0 || fabsf(((__f32x4)__a)[0]) < 2.f))
     return x;
   else
@@ -640,7 +640,7 @@ _mm_cvttss_si64(__m128 __a)
 {
   float e = ((__f32x4)__a)[0];
   if (isnan(e) || isinf(e) || e > LLONG_MAX || e < LLONG_MIN) return 0x8000000000000000LL;
-  long long x = llrint(e);
+  long long x = llrintf(e);
   if (x != 0xFFFFFFFF00000000ULL && (x != 0 || fabsf(e) < 2.f))
     return (long long)e;
   else