Run benchmarks of all functions, not just some

1 files changed, 184 insertions, 90 deletions

diff --git a/bench.cc b/bench.cc
index 4c58911..b4b30d2 100644
--- a/bench.cc
+++ b/bench.cc
@@ -87,33 +87,129 @@ void save_result(realvec_t result)
   for (int i=0; i<realvec_t::size; ++i) {
     global_result += result[i];
   }
+  // Check global accumulator to prevent optimisation
+  if (! vml_std::isfinite(global_result)) {
+    cout << "\n"
+         << "WARNING: Global accumulator is not finite\n";
+  }
 }
 
 
 
 template<typename T> inline T nop(T x) { return x; }
 
+template<typename T> inline T fneg(T x) { return -x; }
+
+template<typename T> inline T fadd(T x, T y) { return x+y; }
+template<typename T> inline T fsub(T x, T y) { return x-y; }
+template<typename T> inline T fmul(T x, T y) { return x*y; }
+template<typename T> inline T fdiv(T x, T y) { return x/y; }
+
+template<typename T> inline T frexp0(T x)
+{
+  typename T::intvec_t ir;
+  return frexp(x, &ir);
+}
+template<typename T> inline typename T::intvec_t frexp1(T x)
+{
+  typename T::intvec_t ir;
+  frexp(x, &ir);
+  return ir;
+}
+
+template<typename T> inline T ldexps(T x, T y)
+{
+  typename T::intvec_t iy = convert_int(y);
+  return ldexp(x, iy[0]);
+}
+template<typename T> inline T ldexpv(T x, T y)
+{
+  typename T::intvec_t iy = convert_int(y);
+  return ldexp(x, iy);
+}
+
+
+
 #define DECLARE_FUNCTOR(FUNC, XMIN, XMAX)                       \
   template<typename T>                                          \
   struct functor_##FUNC {                                       \
     static typename T::real_t get_xmin() { return XMIN; }       \
     static typename T::real_t get_xmax() { return XMAX; }       \
     static const char* name() { return #FUNC; }                 \
-    T operator()(T x) { return FUNC(x); }                       \
+    T operator()(T x) {                                         \
+      return FUNC(x);                                           \
+    }                                                           \
+  }
+
+#define DECLARE_BFUNCTOR(FUNC, XMIN, XMAX)                      \
+  template<typename T>                                          \
+  struct functor_##FUNC {                                       \
+    static typename T::real_t get_xmin() { return XMIN; }       \
+    static typename T::real_t get_xmax() { return XMAX; }       \
+    static const char* name() { return #FUNC; }                 \
+    T operator()(T x) {                                         \
+      typename T::boolvec_t res = FUNC(x);                      \
+      return convert_float(convert_int(res));                   \
+    }                                                           \
+  }
+
+#define DECLARE_IFUNCTOR(FUNC, XMIN, XMAX)                      \
+  template<typename T>                                          \
+  struct functor_##FUNC {                                       \
+    static typename T::real_t get_xmin() { return XMIN; }       \
+    static typename T::real_t get_xmax() { return XMAX; }       \
+    static const char* name() { return #FUNC; }                 \
+    T operator()(T x) {                                         \
+      typename T::intvec_t res = FUNC(x);                       \
+      return convert_float(res);                                \
+    }                                                           \
   }
 
+#define DECLARE_FUNCTOR2(FUNC, XMIN, XMAX, YOFFSET)             \
+  template<typename T>                                          \
+  struct functor_##FUNC {                                       \
+    static typename T::real_t get_xmin() { return XMIN; }       \
+    static typename T::real_t get_xmax() { return XMAX; }       \
+    static const char* name() { return #FUNC; }                 \
+    T operator()(T x) {                                         \
+      const typename T::real_t yoffset = YOFFSET;               \
+      return FUNC(x, x + T(yoffset));                           \
+    }                                                           \
+  }
+
+#define DECLARE_FUNCTOR3(FUNC, XMIN, XMAX, YOFFSET, ZOFFSET)    \
+  template<typename T>                                          \
+  struct functor_##FUNC {                                       \
+    static typename T::real_t get_xmin() { return XMIN; }       \
+    static typename T::real_t get_xmax() { return XMAX; }       \
+    static const char* name() { return #FUNC; }                 \
+    T operator()(T x) {                                         \
+      const typename T::real_t yoffset = YOFFSET;               \
+      const typename T::real_t zoffset = ZOFFSET;               \
+      return FUNC(x, x + T(yoffset), x + T(zoffset));           \
+    }                                                           \
+  }
+
+
+
 DECLARE_FUNCTOR(nop, 0.0, 1.0);
 
+DECLARE_FUNCTOR(fneg, 0.0, 1.0);
+DECLARE_FUNCTOR2(fadd, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR2(fsub, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR2(fmul, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR2(fdiv, 0.0, 1.0, 2.0);
+
 DECLARE_FUNCTOR(acos, -0.5, +0.5);
-DECLARE_FUNCTOR(acosh, 0.0, 1.0);
+DECLARE_FUNCTOR(acosh, 1.0, 2.0);
 DECLARE_FUNCTOR(asin, -0.5, +0.5);
 DECLARE_FUNCTOR(asinh, -1.0, +1.0);
 DECLARE_FUNCTOR(atan, -1.0, +1.0);
-// DECLARE_FUNCTOR(atan2, 0.0);
+DECLARE_FUNCTOR2(atan2, 0.0, 1.0, 2.0);
 DECLARE_FUNCTOR(atanh, -0.5, +0.5);
 DECLARE_FUNCTOR(cbrt, -1.0, 1.0);
 DECLARE_FUNCTOR(ceil, -1.0, +1.0);
-// DECLARE_FUNCTOR(copysign, 0.0);
+DECLARE_FUNCTOR2(copysign, -1.0, +1.0, 2.0);
 DECLARE_FUNCTOR(cos, 0.0, 1.0);
 DECLARE_FUNCTOR(cosh, 0.0, 1.0);
 DECLARE_FUNCTOR(exp, 0.0, 1.0);
@@ -122,32 +218,33 @@ DECLARE_FUNCTOR(exp2, 0.0, 1.0);
 DECLARE_FUNCTOR(expm1, 0.0, 1.0);
 DECLARE_FUNCTOR(fabs, -1.0, 1.0);
 DECLARE_FUNCTOR(floor, -1.0, +1.0);
-// DECLARE_FUNCTOR(fdim, 0.0);
-// DECLARE_FUNCTOR(fma, 0.0);
-// DECLARE_FUNCTOR(fmax, 0.0);
-// DECLARE_FUNCTOR(fmin, 0.0);
-// DECLARE_FUNCTOR(fmod, 0.0);
-// DECLARE_FUNCTOR(frexp, 0.0);
-// DECLARE_FUNCTOR(hypot, 0.0);
-// DECLARE_FUNCTOR(ilogb, 0.0);
-// DECLARE_FUNCTOR(isfinite, 0.0);
-// DECLARE_FUNCTOR(isinf, 0.0);
-// DECLARE_FUNCTOR(isnan, 0.0);
-// DECLARE_FUNCTOR(isnormal, 0.0);
-// DECLARE_FUNCTOR(ldexp, 0.0);
-// DECLARE_FUNCTOR(ldexp, 0.0);
+DECLARE_FUNCTOR2(fdim, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR3(fma, 0.0, 1.0, 2.0, 3.0);
+DECLARE_FUNCTOR2(fmax, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR2(fmin, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR2(fmod, 0.0, 1.0, 2.0);
+DECLARE_FUNCTOR(frexp0, 1.0, 100.0);
+DECLARE_IFUNCTOR(frexp1, 1.0, 100.0);
+DECLARE_FUNCTOR2(hypot, 0.0, 1.0, 2.0);
+DECLARE_IFUNCTOR(ilogb, 1.0, 100.0);
+DECLARE_BFUNCTOR(isfinite, 0.0, 1.0);
+DECLARE_BFUNCTOR(isinf, 0.0, 1.0);
+DECLARE_BFUNCTOR(isnan, 0.0, 1.0);
+DECLARE_BFUNCTOR(isnormal, 0.0, 1.0);
+DECLARE_FUNCTOR2(ldexps, 1.0, 20.0, -10.0);
+DECLARE_FUNCTOR2(ldexpv, 1.0, 20.0, -10.0);
 DECLARE_FUNCTOR(log, 1.0, 2.0);
 DECLARE_FUNCTOR(log10, 1.0, 2.0);
 DECLARE_FUNCTOR(log1p, 0.0, 1.0);
 DECLARE_FUNCTOR(log2, 1.0, 2.0);
-// DECLARE_FUNCTOR(nextafter, 0.0);
-// DECLARE_FUNCTOR(pow, 0.0);
+DECLARE_FUNCTOR2(nextafter, -1.0, +1.0, 0.0);
+DECLARE_FUNCTOR2(pow, 0.0, 1.0, 2.0);
 DECLARE_FUNCTOR(rcp, 1.0, 2.0);
-// DECLARE_FUNCTOR(remainder, 0.0);
+DECLARE_FUNCTOR2(remainder, 0.0, 1.0, 2.0);
 DECLARE_FUNCTOR(rint, -1.0, +1.0);
 DECLARE_FUNCTOR(round, -1.0, +1.0);
-DECLARE_FUNCTOR(rsqrt, 0.0, 1.0);
-// DECLARE_FUNCTOR(signbit, 0.0);
+DECLARE_FUNCTOR(rsqrt, 1.0, 2.0);
+DECLARE_BFUNCTOR(signbit, -1.0, +1.0);
 DECLARE_FUNCTOR(sin, 0.0, 1.0);
 DECLARE_FUNCTOR(sinh, -1.0, +1.0);
 DECLARE_FUNCTOR(sqrt, 0.0, 1.0);
@@ -201,45 +298,45 @@ template<template<typename> class func_t>
 void bench_func()
 {
   cout << "\n"
-       << "Benchmarking " << func_t<float_vec>().name() << ":\n";
+       << "Benchmarking " << func_t<float32_vec>().name() << ":\n";
   
-//   bench_type_func<realpseudovec<float,1>, func_t>();
-//   // bench_type_func<realbuiltinvec<float,1>, func_t>();
-//   bench_type_func<realtestvec<float,1>, func_t>();
-// #ifdef VECMATHLIB_HAVE_VEC_FLOAT_1
-//   bench_type_func<realvec<float,1>, func_t>();
-// #endif
-// #ifdef VECMATHLIB_HAVE_VEC_FLOAT_2
-//   bench_type_func<realpseudovec<float,2>, func_t>();
-//   // bench_type_func<realbuiltinvec<float,2>, func_t>();
-//   // bench_type_func<realtestvec<float,2>, func_t>();
-//   bench_type_func<realvec<float,2>, func_t>();
-// #endif
-// #ifdef VECMATHLIB_HAVE_VEC_FLOAT_4
-//   bench_type_func<realpseudovec<float,4>, func_t>();
-//   // bench_type_func<realbuiltinvec<float,4>, func_t>();
-//   // bench_type_func<realtestvec<float,4>, func_t>();
-//   bench_type_func<realvec<float,4>, func_t>();
-// #endif
-// #ifdef VECMATHLIB_HAVE_VEC_FLOAT_8
-//   bench_type_func<realpseudovec<float,8>, func_t>();
-//   // bench_type_func<realbuiltinvec<float,8>, func_t>();
-//   // bench_type_func<realtestvec<float,8>, func_t>();
-//   bench_type_func<realvec<float,8>, func_t>();
-// #endif
+  bench_type_func<realpseudovec<float,1>, func_t>();
+  // bench_type_func<realbuiltinvec<float,1>, func_t>();
+  bench_type_func<realtestvec<float,1>, func_t>();
+#ifdef VECMATHLIB_HAVE_VEC_FLOAT_1
+  bench_type_func<realvec<float,1>, func_t>();
+#endif
+#ifdef VECMATHLIB_HAVE_VEC_FLOAT_2
+  bench_type_func<realpseudovec<float,2>, func_t>();
+  // bench_type_func<realbuiltinvec<float,2>, func_t>();
+  // bench_type_func<realtestvec<float,2>, func_t>();
+  bench_type_func<realvec<float,2>, func_t>();
+#endif
+#ifdef VECMATHLIB_HAVE_VEC_FLOAT_4
+  bench_type_func<realpseudovec<float,4>, func_t>();
+  // bench_type_func<realbuiltinvec<float,4>, func_t>();
+  // bench_type_func<realtestvec<float,4>, func_t>();
+  bench_type_func<realvec<float,4>, func_t>();
+#endif
+#ifdef VECMATHLIB_HAVE_VEC_FLOAT_8
+  bench_type_func<realpseudovec<float,8>, func_t>();
+  // bench_type_func<realbuiltinvec<float,8>, func_t>();
+  // bench_type_func<realtestvec<float,8>, func_t>();
+  bench_type_func<realvec<float,8>, func_t>();
+#endif
   
-//   bench_type_func<realpseudovec<double,1>, func_t>();
-//   // bench_type_func<realbuiltinvec<double,1>, func_t>();
-//   bench_type_func<realtestvec<double,1>, func_t>();
-// #ifdef VECMATHLIB_HAVE_VEC_DOUBLE_1
-//   bench_type_func<realvec<double,1>, func_t>();
-// #endif
-// #ifdef VECMATHLIB_HAVE_VEC_DOUBLE_2
-//   bench_type_func<realpseudovec<double,2>, func_t>();
-//   // bench_type_func<realbuiltinvec<double,2>, func_t>();
-//   // bench_type_func<realtestvec<double,2>, func_t>();
-//   bench_type_func<realvec<double,2>, func_t>();
-// #endif
+  bench_type_func<realpseudovec<double,1>, func_t>();
+  // bench_type_func<realbuiltinvec<double,1>, func_t>();
+  bench_type_func<realtestvec<double,1>, func_t>();
+#ifdef VECMATHLIB_HAVE_VEC_DOUBLE_1
+  bench_type_func<realvec<double,1>, func_t>();
+#endif
+#ifdef VECMATHLIB_HAVE_VEC_DOUBLE_2
+  bench_type_func<realpseudovec<double,2>, func_t>();
+  // bench_type_func<realbuiltinvec<double,2>, func_t>();
+  // bench_type_func<realtestvec<double,2>, func_t>();
+  bench_type_func<realvec<double,2>, func_t>();
+#endif
 #ifdef VECMATHLIB_HAVE_VEC_DOUBLE_4
   bench_type_func<realpseudovec<double,4>, func_t>();
   // bench_type_func<realbuiltinvec<double,4>, func_t>();
@@ -252,16 +349,22 @@ void bench()
 {
   bench_func<functor_nop>();
   
+  bench_func<functor_fneg>();
+  bench_func<functor_fadd>();
+  bench_func<functor_fsub>();
+  bench_func<functor_fmul>();
+  bench_func<functor_fdiv>();
+  
   bench_func<functor_acos>();
   bench_func<functor_acosh>();
   bench_func<functor_asin>();
   bench_func<functor_asinh>();
   bench_func<functor_atan>();
-  // bench_func<functor_atan2>();
+  bench_func<functor_atan2>();
   bench_func<functor_atanh>();
   bench_func<functor_cbrt>();
   bench_func<functor_ceil>();
-  // bench_func<functor_copysign>();
+  bench_func<functor_copysign>();
   bench_func<functor_cos>();
   bench_func<functor_cosh>();
   bench_func<functor_exp>();
@@ -270,32 +373,33 @@ void bench()
   bench_func<functor_expm1>();
   bench_func<functor_fabs>();
   bench_func<functor_floor>();
-  // bench_func<functor_fdim>();
-  // bench_func<functor_fma>();
-  // bench_func<functor_fmax>();
-  // bench_func<functor_fmin>();
-  // bench_func<functor_fmod>();
-  // bench_func<functor_frexp>();
-  // bench_func<functor_hypot>();
-  // bench_func<functor_ilogb>();
-  // bench_func<functor_isfinite>();
-  // bench_func<functor_isinf>();
-  // bench_func<functor_isnan>();
-  // bench_func<functor_isnormal>();
-  // bench_func<functor_ldexp>();
-  // bench_func<functor_ldexp>();
+  bench_func<functor_fdim>();
+  bench_func<functor_fma>();
+  bench_func<functor_fmax>();
+  bench_func<functor_fmin>();
+  bench_func<functor_fmod>();
+  bench_func<functor_frexp0>();
+  bench_func<functor_frexp1>();
+  bench_func<functor_hypot>();
+  bench_func<functor_ilogb>();
+  bench_func<functor_isfinite>();
+  bench_func<functor_isinf>();
+  bench_func<functor_isnan>();
+  bench_func<functor_isnormal>();
+  bench_func<functor_ldexps>();
+  bench_func<functor_ldexpv>();
   bench_func<functor_log>();
   bench_func<functor_log10>();
   bench_func<functor_log1p>();
   bench_func<functor_log2>();
-  // bench_func<functor_nextafter>();
-  // bench_func<functor_pow>();
+  bench_func<functor_nextafter>();
+  bench_func<functor_pow>();
   bench_func<functor_rcp>();
-  // bench_func<functor_remainder>();
+  bench_func<functor_remainder>();
   bench_func<functor_rint>();
   bench_func<functor_round>();
   bench_func<functor_rsqrt>();
-  // bench_func<functor_signbit>();
+  bench_func<functor_signbit>();
   bench_func<functor_sin>();
   bench_func<functor_sinh>();
   bench_func<functor_sqrt>();
@@ -308,17 +412,7 @@ void bench()
 
 int main(int argc, char** argv)
 {
-  using namespace vecmathlib;
-
   cout << "Benchmarking math functions:\n";
-  
   bench();
-  
-  // Checking global accumulator to prevent optimisation
-  if (! std::isfinite(global_result)) {
-    cout << "\n"
-         << "WARNING: Global accumulator is not finite\n";
-  }
-  
   return 0;
 }