Change FDIV to FMUL

Some small performance improvement from safely changing a
couple floating point divides to floating point multiplies.

src/lib/alp/sls_basic.hpp

@@ -52,6 +52,7 @@ Contents: Some basic functions and types
 namespace Sls { 
 
 	const double pi=3.1415926535897932384626433832795;
+	const double invsqrttwo=0.707106769084930419921875;
 	const double const_val=1/sqrt(2.0*pi);
 	const long int quick_tests_trials_number=100;
 
@@ -189,7 +190,7 @@ namespace Sls {
 
 		static double normal_probability(double x_)
 		{
-			return 0.5*erfc(-sqrt(0.5)*x_);
+			return 0.5*erfc(-invsqrttwo*x_);
 		}
 
 		static double normal_probability(

src/masking/tantan.cpp

@@ -98,12 +98,13 @@ Mask::Ranges mask(Letter *seq,
 	}
 
 	const float z = b * b2b + f.sum() * p_repeat_end;
+	const float zinv = 1.0/z;
 	b = b2b;
 	f = p_repeat_end;
 	Mask::Ranges ranges;
 
 	for (int i = len - 1; i >= 0; --i) {
-		const float pf = 1 - (pb[i] * b / z);
+		const float pf = 1 - (pb[i] * b * zinv);
 
 		if ((i & 15) == 15) {
 			const float s = scale[i / 16];