Changes in / [639c4e3:ed246ab] in sasmodels

Files:

• gen_so.py (modified) (1 diff)
• sasmodels/kernel_template.c (modified) (2 diffs)
• sasmodels/models/lib/sas_gamma.c (modified) (2 diffs)

gen_so.py

@@ -9 +9 @@
 
     # Convert ../sasmodels/models/name.py to name
-    model_loc = os.path.join(sasmodels, "sasmodels", "models") 
-    for i in os.listdir(model_loc):
+    for model_name in core.list_models():
         # Choose only relevant python files
-        if i.endswith(".py") and not i.startswith("_"):
-            model_name = os.path.basename(i)[:-3]
-            model_info = core.load_model_info(model_name)
+            # model_info = core.load_model_info(model_name)
             # Run the conversion but don't delete the .so
-            model = core.build_model(model_info)
+            model = core.precompile_dll(model_name)
 
 if __name__ == "__main__":

sasmodels/kernel_template.c

@@ -24 +24 @@
          inline double isnan(double x) { return _isnan(x); }
          #define NAN (std::numeric_limits<double>::quiet_NaN()) // non-signalling NaN
-         static double cephes_expm1(double x) {
-            // Adapted from the cephes math library.
-            // Copyright 1984 - 1992 by Stephen L. Moshier
-            if (x != x || x == 0.0) {
-               return x; // NaN and +/- 0
-            } else if (x < -0.5 || x > 0.5) {
-               return exp(x) - 1.0;
-            } else {
-               const double xsq = x*x;
-               const double p = (((
-                  +1.2617719307481059087798E-4)*xsq
-                  +3.0299440770744196129956E-2)*xsq
-                  +9.9999999999999999991025E-1);
-               const double q = ((((
-                  +3.0019850513866445504159E-6)*xsq
-                  +2.5244834034968410419224E-3)*xsq
-                  +2.2726554820815502876593E-1)*xsq
-                  +2.0000000000000000000897E0);
-               double r = x * p;
-               r =  r / (q - r);
-               return r+r;
-             }
-         }
-         #define expm1 cephes_expm1
+         #define NEED_EXPM1
+         #define NEED_TGAMMA
      #else
          #define kernel extern "C"
@@ -74 +52 @@
 #endif
 
+#if defined(NEED_EXPM1)
+   static double expm1(double x) {
+      // Adapted from the cephes math library.
+      // Copyright 1984 - 1992 by Stephen L. Moshier
+      if (x != x || x == 0.0) {
+         return x; // NaN and +/- 0
+      } else if (x < -0.5 || x > 0.5) {
+         return exp(x) - 1.0;
+      } else {
+         const double xsq = x*x;
+         const double p = (((
+            +1.2617719307481059087798E-4)*xsq
+            +3.0299440770744196129956E-2)*xsq
+            +9.9999999999999999991025E-1);
+         const double q = ((((
+            +3.0019850513866445504159E-6)*xsq
+            +2.5244834034968410419224E-3)*xsq
+            +2.2726554820815502876593E-1)*xsq
+            +2.0000000000000000000897E0);
+         double r = x * p;
+         r =  r / (q - r);
+         return r+r;
+       }
+   }
+#endif
+
 // Standard mathematical constants:
 //   M_E, M_LOG2E, M_LOG10E, M_LN2, M_LN10, M_PI, M_PI_2=pi/2, M_PI_4=pi/4,

sasmodels/models/lib/sas_gamma.c

@@ -1 +1 @@
 /*
 The wrapper for gamma function from OpenCL and standard libraries
-The OpenCL gamma function fails misserably on values lower than 1.0
+The OpenCL gamma function fails miserably on values lower than 1.0
 while works fine on larger values.
 We use gamma definition Gamma(t + 1) = t * Gamma(t) to compute
@@ -7 +7 @@
 */
 
+#if defined(NEED_TGAMMA)
+static double cephes_stirf(double x)
+{
+        const double MAXSTIR=143.01608;
+        const double SQTPI=2.50662827463100050242E0;
+        double y, w, v;
+
+        w = 1.0 / x;
+
+        w = ((((
+                7.87311395793093628397E-4*w
+                - 2.29549961613378126380E-4)*w
+                - 2.68132617805781232825E-3)*w
+                + 3.47222221605458667310E-3)*w
+                + 8.33333333333482257126E-2)*w
+                + 1.0;
+        y = exp(x);
+        if (x > MAXSTIR)
+        { /* Avoid overflow in pow() */
+                v = pow(x, 0.5 * x - 0.25);
+                y = v * (v / y);
+        }
+        else
+        {
+                y = pow(x, x - 0.5) / y;
+        }
+        y = SQTPI * y * w;
+        return(y);
+}
+
+static double tgamma(double x) {
+        double p, q, z;
+        int sgngam;
+        int i;
+
+        sgngam = 1;
+        if (isnan(x))
+                return(x);
+        q = fabs(x);
+
+        if (q > 33.0)
+        {
+                if (x < 0.0)
+                {
+                        p = floor(q);
+                        if (p == q)
+                        {
+                                return (NAN);
+                        }
+                        i = p;
+                        if ((i & 1) == 0)
+                                sgngam = -1;
+                        z = q - p;
+                        if (z > 0.5)
+                        {
+                                p += 1.0;
+                                z = q - p;
+                        }
+                        z = q * sin(M_PI * z);
+                        if (z == 0.0)
+                        {
+                                return(NAN);
+                        }
+                        z = fabs(z);
+                        z = M_PI / (z * cephes_stirf(q));
+                }
+                else
+                {
+                        z = cephes_stirf(x);
+                }
+                return(sgngam * z);
+        }
+
+        z = 1.0;
+        while (x >= 3.0)
+        {
+                x -= 1.0;
+                z *= x;
+        }
+
+        while (x < 0.0)
+        {
+                if (x > -1.E-9)
+                        goto small;
+                z /= x;
+                x += 1.0;
+        }
+
+        while (x < 2.0)
+        {
+                if (x < 1.e-9)
+                        goto small;
+                z /= x;
+                x += 1.0;
+        }
+
+        if (x == 2.0)
+                return(z);
+
+        x -= 2.0;
+        p = (((((
+                +1.60119522476751861407E-4*x
+                + 1.19135147006586384913E-3)*x
+                + 1.04213797561761569935E-2)*x
+                + 4.76367800457137231464E-2)*x
+                + 2.07448227648435975150E-1)*x
+                + 4.94214826801497100753E-1)*x
+                + 9.99999999999999996796E-1;
+        q = ((((((
+                -2.31581873324120129819E-5*x
+                + 5.39605580493303397842E-4)*x
+                - 4.45641913851797240494E-3)*x
+                + 1.18139785222060435552E-2)*x
+                + 3.58236398605498653373E-2)*x
+                - 2.34591795718243348568E-1)*x
+                + 7.14304917030273074085E-2)*x
+                + 1.00000000000000000320E0;
+        return(z * p / q);
+
+small:
+        if (x == 0.0)
+        {
+                return (NAN);
+        }
+        else
+                return(z / ((1.0 + 0.5772156649015329 * x) * x));
+}
+#endif
+
 
 inline double sas_gamma( double x) { return tgamma(x+1)/x; }