path: root/vec_float_sse2_scalar.h

// -*-C++-*-

#ifndef VEC_FLOAT_SSE2_SCALAR_H
#define VEC_FLOAT_SSE2_SCALAR_H

#include "floatprops.h"
#include "mathfuncs.h"
#include "vec_base.h"

#include <cmath>

// SSE2 intrinsics
#include <emmintrin.h>
#ifdef __SSE3__                 // Intel's SSE 3
#  include <pmmintrin.h>
#endif
#ifdef __SSE4_1__               // Intel's SSE 4.1
#  include <smmintrin.h>
#endif
#ifdef __SSE4A__                // AMD's SSE 4a
#  include <ammintrin.h>
#endif
#if defined __AVX__             // Intel's AVX
#  include <immintrin.h>
#endif



namespace vecmathlib {
  
#define VECMATHLIB_HAVE_VEC_FLOAT_1
  template<> struct boolvec<float,1>;
  template<> struct intvec<float,1>;
  template<> struct realvec<float,1>;
  
  
  
  template<>
  struct boolvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef bool scalar_t;
    typedef uint_t bvector_t;
    
    static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                  "vector size is wrong");
    
    // true values are non-zero, false values are zero
    
    typedef boolvec boolvec_t;
    typedef intvec<real_t, size> intvec_t;
    typedef realvec<real_t, size> realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    bvector_t v;
    
    boolvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // boolvec(boolvec const& x): v(x.v) {}
    // boolvec& operator=(boolvec const& x) { return v=x.v, *this; }
    boolvec(bvector_t x): v(x) {}
    boolvec(bool a): v(a) {}
    boolvec(bool const* as): v(as[0]) {}
    
    operator bvector_t() const { return v; }
    bool operator[](int n) const { return v; }
    boolvec& set_elt(int n, bool a) { return v=a, *this; }
    
    
    
    intvec_t as_int() const;      // defined after intvec
    intvec_t convert_int() const; // defined after intvec
    
    
    
    boolvec operator!() const { return !v; }
    
    boolvec operator&&(boolvec x) const { return v && x.v; }
    boolvec operator||(boolvec x) const { return v || x.v; }
    boolvec operator==(boolvec x) const { return bool(v) == bool(x.v); }
    boolvec operator!=(boolvec x) const { return bool(v) != bool(x.v); }
    
    bool all() const { return v; }
    bool any() const { return v; }
    
    
    
    // ifthen(condition, then-value, else-value)
    intvec_t ifthen(intvec_t x, intvec_t y) const; // defined after intvec
    realvec_t ifthen(realvec_t x, realvec_t y) const; // defined after realvec
  };
  
  
  
  template<>
  struct intvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef int_t scalar_t;
    typedef int_t ivector_t;
    
    static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                  "vector size is wrong");
    
    typedef boolvec<real_t, size> boolvec_t;
    typedef intvec intvec_t;
    typedef realvec<real_t, size> realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    ivector_t v;
    
    intvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // intvec(intvec const& x): v(x.v) {}
    // intvec& operator=(intvec const& x) { return v=x.v, *this; }
    intvec(int_t a): v(a) {}
    intvec(int_t const* as): v(as[0]) {}
    static intvec iota() { return intvec(I(0)); }
    
    operator ivector_t() const { return v; }
    int_t operator[](int n) const { return v; }
    intvec& set_elt(int n, int_t a) { return v=a, *this; }
    
    
    
    boolvec_t as_bool() const { return U(v); }
    boolvec_t convert_bool() const { return bool(v); }
    realvec_t as_float() const;      // defined after realvec
    realvec_t convert_float() const; // defined after realvec
    
    
    
    intvec operator+() const { return +v; }
    intvec operator-() const { return -v; }
    
    intvec operator+(intvec x) const { return v+x.v; }
    intvec operator-(intvec x) const { return v-x.v; }
    intvec operator*(intvec x) const { return v*x.v; }
    intvec operator/(intvec x) const { return v/x.v; }
    intvec operator%(intvec x) const { return v%x.v; }
    
    intvec& operator+=(intvec const& x) { return *this=*this+x; }
    intvec& operator-=(intvec const& x) { return *this=*this-x; }
    intvec& operator*=(intvec const& x) { return *this=*this*x; }
    intvec& operator/=(intvec const& x) { return *this=*this/x; }
    intvec& operator%=(intvec const& x) { return *this=*this%x; }
    
    
    
    intvec operator~() const { return ~v; }
    
    intvec operator&(intvec x) const { return v&x.v; }
    intvec operator|(intvec x) const { return v|x.v; }
    intvec operator^(intvec x) const { return v^x.v; }
    
    intvec& operator&=(intvec const& x) { return *this=*this&x; }
    intvec& operator|=(intvec const& x) { return *this=*this|x; }
    intvec& operator^=(intvec const& x) { return *this=*this^x; }
    
    
    
    intvec lsr(int_t n) const { return U(v) >> U(n); }
    intvec operator>>(int_t n) const { return v>>n; }
    intvec operator<<(int_t n) const { return v<<n; }
    
    intvec& operator>>=(int_t n) { return *this=*this>>n; }
    intvec& operator<<=(int_t n) { return *this=*this<<n; }
    
    intvec lsr(intvec n) const { return U(v) >> U(n); }
    intvec operator>>(intvec n) const { return v>>n; }
    intvec operator<<(intvec n) const { return v<<n; }
    
    intvec& operator>>=(intvec n) { return *this=*this>>n; }
    intvec& operator<<=(intvec n) { return *this=*this<<n; }
    
    
    
    boolvec_t operator==(intvec const& x) const { return v==x.v; }
    boolvec_t operator!=(intvec const& x) const { return v!=x.v; }
    boolvec_t operator<(intvec const& x) const { return v<x.v; }
    boolvec_t operator<=(intvec const& x) const { return v<=x.v; }
    boolvec_t operator>(intvec const& x) const { return v>x.v; }
    boolvec_t operator>=(intvec const& x) const { return v>=x.v; }
  };
  
  
  
  template<>
  struct realvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef real_t scalar_t;
    typedef float vector_t;
    
    static char const* name() { return "<SSE2:1*float>"; }
    inline void barrier() { asm("": "+x" (v)); }
    
    static_assert(size * sizeof(real_t) == sizeof(vector_t),
                  "vector size is wrong");
    
  private:
    static __m128 from_float(float a) { return _mm_set_ss(a); }
    static float to_float(__m128 a) { return _mm_cvtss_f32(a); }
  public:
    
    typedef boolvec<real_t, size> boolvec_t;
    typedef intvec<real_t, size> intvec_t;
    typedef realvec realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    vector_t v;
    
    realvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // realvec(realvec const& x): v(x.v) {}
    // realvec& operator=(realvec const& x) { return v=x.v, *this; }
    realvec(real_t a): v(a) {}
    realvec(real_t const* as): v(as[0]) {}
    
    operator vector_t() const { return v; }
    real_t operator[](int n) const { return v; }
    realvec& set_elt(int n, real_t a) { return v=a, *this; }
    
    
    
    typedef vecmathlib::mask_t<realvec_t> mask_t;
    
    static realvec_t loada(real_t const* p)
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      return *p;
    }
    static realvec_t loadu(real_t const* p)
    {
      return *p;
    }
    static realvec_t loadu(real_t const* p, size_t ioff)
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      return loada(p);
    }
    realvec_t loada(real_t const* p, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      if (__builtin_expect(all(m.m), true)) {
        return loada(p);
      } else {
        return *this;
      }
    }
    realvec_t loadu(real_t const* p, mask_t const& m) const
    {
      if (__builtin_expect(m.all_m, true)) {
        return loadu(p);
      } else {
        return *this;
      }
    }
    realvec_t loadu(real_t const* p, size_t ioff, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      return loada(p, m);
    }
    
    void storea(real_t* p) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      *p = v;
    }
    void storeu(real_t* p) const
    {
      *p = v;
    }
    void storeu(real_t* p, size_t ioff) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      *p = v;
    }
    void storea(real_t* p, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      if (__builtin_expect(m.all_m, true)) {
        storea(p);
      }
    }
    void storeu(real_t* p, mask_t const& m) const
    {
      if (__builtin_expect(m.all_m, true)) {
        storeu(p);
      }
    }
    void storeu(real_t* p, size_t ioff, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % sizeof(realvec_t) == 0);
      return storea(p, m);
    }
    
    
    
    intvec_t as_int() const { return floatprops::as_int(v); }
    intvec_t convert_int() const { return floatprops::convert_int(v); }
    
    
    
    realvec operator+() const { return +v; }
    realvec operator-() const { return -v; }
    
    realvec operator+(realvec x) const { return v+x.v; }
    realvec operator-(realvec x) const { return v-x.v; }
    realvec operator*(realvec x) const { return v*x.v; }
    realvec operator/(realvec x) const { return v/x.v; }
    
    realvec& operator+=(realvec const& x) { return *this=*this+x; }
    realvec& operator-=(realvec const& x) { return *this=*this-x; }
    realvec& operator*=(realvec const& x) { return *this=*this*x; }
    realvec& operator/=(realvec const& x) { return *this=*this/x; }
    
    real_t prod() const { return v; }
    real_t sum() const { return v; }
    
    
    
    boolvec_t operator==(realvec const& x) const { return v==x.v; }
    boolvec_t operator!=(realvec const& x) const { return v!=x.v; }
    boolvec_t operator<(realvec const& x) const { return v<x.v; }
    boolvec_t operator<=(realvec const& x) const { return v<=x.v; }
    boolvec_t operator>(realvec const& x) const { return v>x.v; }
    boolvec_t operator>=(realvec const& x) const { return v>=x.v; }
    
    
    
    realvec acos() const { return MF::vml_acos(*this); }
    realvec acosh() const { return MF::vml_acosh(*this); }
    realvec asin() const { return MF::vml_asin(*this); }
    realvec asinh() const { return MF::vml_asinh(*this); }
    realvec atan() const { return MF::vml_atan(*this); }
    realvec atanh() const { return MF::vml_atanh(*this); }
    realvec ceil() const { return std::ceil(v); }
    realvec copysign(realvec y) const { return std::copysign(v, y.v); }
    realvec cos() const { return MF::vml_cos(*this); }
    realvec cosh() const { return MF::vml_cosh(*this); }
    realvec exp() const { return MF::vml_exp(*this); }
    realvec exp10() const { return MF::vml_exp10(*this); }
    realvec exp2() const { return MF::vml_exp2(*this); }
    realvec expm1() const { return MF::vml_expm1(*this); }
    realvec fabs() const { return std::fabs(v); }
    realvec fdim(realvec y) const { return MF::vml_fdim(*this, y); }
    realvec floor() const { return std::floor(v); }
    realvec fma(realvec y, realvec z) const { return MF::vml_fma(*this, y, z); }
    realvec fmax(realvec y) const
    {
      return to_float(_mm_max_ss(from_float(v), from_float(y.v)));
    }
    realvec fmin(realvec y) const
    {
      return to_float(_mm_min_ss(from_float(v), from_float(y.v)));
    }
    realvec fmod(realvec y) const { return std::fmod(v, y.v); }
    intvec_t ilogb() const { return std::ilogb(v); }
    boolvec_t isfinite() const { return std::isfinite(v); }
    boolvec_t isinf() const { return std::isinf(v); }
    boolvec_t isnan() const { return std::isnan(v); }
    boolvec_t isnormal() const { return std::isnormal(v); }
    realvec log() const { return MF::vml_log(*this); }
    realvec log10() const { return MF::vml_log10(*this); }
    realvec log1p() const { return MF::vml_log1p(*this); }
    realvec log2() const { return MF::vml_log2(*this); }
    realvec pow(realvec y) const { return MF::vml_pow(*this, y); }
    realvec rcp() const { return R(1.0)/v; }
    realvec remainder(realvec y) const { return std::remainder(v, y.v); }
    realvec round1() const { return std::round(v); }
    realvec round() const
    {
#ifdef __SSE4_1__
      return to_float(_mm_round_ss(from_float(v), from_float(v),
                                   _MM_FROUND_NINT));
#else
      return MF::vml_round(*this);
#endif
    }
    realvec rsqrt() const { return MF::vml_rsqrt(*this); }
    realvec scalbn(int_t n) const { return std::scalbn(v, n); }
    realvec scalbn(intvec_t n) const { return std::scalbn(v, n); }
    boolvec_t signbit() const { return std::signbit(v); }
    realvec sin() const { return MF::vml_sin(*this); }
    realvec sinh() const { return MF::vml_sinh(*this); }
    realvec sqrt1() const { return std::sqrt(v); }
    realvec sqrt() const { return to_float(_mm_sqrt_ss(from_float(v))); }
    realvec tan() const { return MF::vml_tan(*this); }
    realvec tanh() const { return MF::vml_tanh(*this); }
  };
  
  
  
  // boolvec definitions
  
  inline
  auto boolvec<float,1>::as_int() const -> intvec_t
  {
    return I(v);
  }
  
  inline
  auto boolvec<float,1>::convert_int() const -> intvec_t
  {
    return v;
  }
  
  inline
  auto boolvec<float,1>::ifthen(intvec_t x, intvec_t y) const -> intvec_t
  {
    return v ? x : y;
  }
  
  inline
  auto boolvec<float,1>::ifthen(realvec_t x, realvec_t y) const -> realvec_t
  {
    return v ? x : y;
  }
  
  
  
  // intvec definitions
  
  inline auto intvec<float,1>::as_float() const -> realvec_t
  {
    return FP::as_float(v);
  }
  
  inline auto intvec<float,1>::convert_float() const -> realvec_t
  {
    return _mm_cvtss_f32(_mm_cvtsi32_ss(_mm_setzero_ps(), v));
  }
  
} // namespace vecmathlib

#endif  // #ifndef VEC_FLOAT_SSE2_SCALAR_H