util_avxi.h

Go to the documentation of this file.

/*
 * Copyright 2009-2013 Intel Corporation
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */
 
#ifndef __UTIL_AVXI_H__
#define __UTIL_AVXI_H__
 
CCL_NAMESPACE_BEGIN
 
struct avxb;
 
struct avxi {
  typedef avxb Mask;  // mask type for us
  enum { size = 8 };  // number of SIMD elements
  union {             // data
    __m256i m256;
#if !defined(__KERNEL_AVX2__)
    struct {
      __m128i l, h;
    };
#endif
    int32_t v[8];
  };
 
 
  __forceinline avxi()
  {
  }
  __forceinline avxi(const avxi &a)
  {
    m256 = a.m256;
  }
  __forceinline avxi &operator=(const avxi &a)
  {
    m256 = a.m256;
    return *this;
  }
 
  __forceinline avxi(const __m256i a) : m256(a)
  {
  }
  __forceinline operator const __m256i &(void)const
  {
    return m256;
  }
  __forceinline operator __m256i &(void)
  {
    return m256;
  }
 
  __forceinline explicit avxi(const ssei &a)
      : m256(_mm256_insertf128_si256(_mm256_castsi128_si256(a), a, 1))
  {
  }
  __forceinline avxi(const ssei &a, const ssei &b)
      : m256(_mm256_insertf128_si256(_mm256_castsi128_si256(a), b, 1))
  {
  }
#if defined(__KERNEL_AVX2__)
  __forceinline avxi(const __m128i &a, const __m128i &b)
      : m256(_mm256_insertf128_si256(_mm256_castsi128_si256(a), b, 1))
  {
  }
#else
  __forceinline avxi(const __m128i &a, const __m128i &b) : l(a), h(b)
  {
  }
#endif
  __forceinline explicit avxi(const int32_t *const a)
      : m256(_mm256_castps_si256(_mm256_loadu_ps((const float *)a)))
  {
  }
  __forceinline avxi(int32_t a) : m256(_mm256_set1_epi32(a))
  {
  }
  __forceinline avxi(int32_t a, int32_t b) : m256(_mm256_set_epi32(b, a, b, a, b, a, b, a))
  {
  }
  __forceinline avxi(int32_t a, int32_t b, int32_t c, int32_t d)
      : m256(_mm256_set_epi32(d, c, b, a, d, c, b, a))
  {
  }
  __forceinline avxi(
      int32_t a, int32_t b, int32_t c, int32_t d, int32_t e, int32_t f, int32_t g, int32_t h)
      : m256(_mm256_set_epi32(h, g, f, e, d, c, b, a))
  {
  }
 
  __forceinline explicit avxi(const __m256 a) : m256(_mm256_cvtps_epi32(a))
  {
  }
 
 
  __forceinline avxi(ZeroTy) : m256(_mm256_setzero_si256())
  {
  }
#if defined(__KERNEL_AVX2__)
  __forceinline avxi(OneTy) : m256(_mm256_set1_epi32(1))
  {
  }
  __forceinline avxi(PosInfTy) : m256(_mm256_set1_epi32(pos_inf))
  {
  }
  __forceinline avxi(NegInfTy) : m256(_mm256_set1_epi32(neg_inf))
  {
  }
#else
  __forceinline avxi(OneTy) : m256(_mm256_set_epi32(1, 1, 1, 1, 1, 1, 1, 1))
  {
  }
  __forceinline avxi(PosInfTy)
      : m256(_mm256_set_epi32(
            pos_inf, pos_inf, pos_inf, pos_inf, pos_inf, pos_inf, pos_inf, pos_inf))
  {
  }
  __forceinline avxi(NegInfTy)
      : m256(_mm256_set_epi32(
            neg_inf, neg_inf, neg_inf, neg_inf, neg_inf, neg_inf, neg_inf, neg_inf))
  {
  }
#endif
  __forceinline avxi(StepTy) : m256(_mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0))
  {
  }
 
 
  __forceinline const int32_t &operator[](const size_t i) const
  {
    assert(i < 8);
    return v[i];
  }
  __forceinline int32_t &operator[](const size_t i)
  {
    assert(i < 8);
    return v[i];
  }
};
 
 
__forceinline const avxi cast(const __m256 &a)
{
  return _mm256_castps_si256(a);
}
__forceinline const avxi operator+(const avxi &a)
{
  return a;
}
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator-(const avxi &a)
{
  return _mm256_sub_epi32(_mm256_setzero_si256(), a.m256);
}
__forceinline const avxi abs(const avxi &a)
{
  return _mm256_abs_epi32(a.m256);
}
#else
__forceinline const avxi operator-(const avxi &a)
{
  return avxi(_mm_sub_epi32(_mm_setzero_si128(), a.l), _mm_sub_epi32(_mm_setzero_si128(), a.h));
}
__forceinline const avxi abs(const avxi &a)
{
  return avxi(_mm_abs_epi32(a.l), _mm_abs_epi32(a.h));
}
#endif
 
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator+(const avxi &a, const avxi &b)
{
  return _mm256_add_epi32(a.m256, b.m256);
}
#else
__forceinline const avxi operator+(const avxi &a, const avxi &b)
{
  return avxi(_mm_add_epi32(a.l, b.l), _mm_add_epi32(a.h, b.h));
}
#endif
__forceinline const avxi operator+(const avxi &a, const int32_t b)
{
  return a + avxi(b);
}
__forceinline const avxi operator+(const int32_t a, const avxi &b)
{
  return avxi(a) + b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator-(const avxi &a, const avxi &b)
{
  return _mm256_sub_epi32(a.m256, b.m256);
}
#else
__forceinline const avxi operator-(const avxi &a, const avxi &b)
{
  return avxi(_mm_sub_epi32(a.l, b.l), _mm_sub_epi32(a.h, b.h));
}
#endif
__forceinline const avxi operator-(const avxi &a, const int32_t b)
{
  return a - avxi(b);
}
__forceinline const avxi operator-(const int32_t a, const avxi &b)
{
  return avxi(a) - b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator*(const avxi &a, const avxi &b)
{
  return _mm256_mullo_epi32(a.m256, b.m256);
}
#else
__forceinline const avxi operator*(const avxi &a, const avxi &b)
{
  return avxi(_mm_mullo_epi32(a.l, b.l), _mm_mullo_epi32(a.h, b.h));
}
#endif
__forceinline const avxi operator*(const avxi &a, const int32_t b)
{
  return a * avxi(b);
}
__forceinline const avxi operator*(const int32_t a, const avxi &b)
{
  return avxi(a) * b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator&(const avxi &a, const avxi &b)
{
  return _mm256_and_si256(a.m256, b.m256);
}
#else
__forceinline const avxi operator&(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_and_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)));
}
#endif
__forceinline const avxi operator&(const avxi &a, const int32_t b)
{
  return a & avxi(b);
}
__forceinline const avxi operator&(const int32_t a, const avxi &b)
{
  return avxi(a) & b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator|(const avxi &a, const avxi &b)
{
  return _mm256_or_si256(a.m256, b.m256);
}
#else
__forceinline const avxi operator|(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_or_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)));
}
#endif
__forceinline const avxi operator|(const avxi &a, const int32_t b)
{
  return a | avxi(b);
}
__forceinline const avxi operator|(const int32_t a, const avxi &b)
{
  return avxi(a) | b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator^(const avxi &a, const avxi &b)
{
  return _mm256_xor_si256(a.m256, b.m256);
}
#else
__forceinline const avxi operator^(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_xor_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)));
}
#endif
__forceinline const avxi operator^(const avxi &a, const int32_t b)
{
  return a ^ avxi(b);
}
__forceinline const avxi operator^(const int32_t a, const avxi &b)
{
  return avxi(a) ^ b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi operator<<(const avxi &a, const int32_t n)
{
  return _mm256_slli_epi32(a.m256, n);
}
__forceinline const avxi operator>>(const avxi &a, const int32_t n)
{
  return _mm256_srai_epi32(a.m256, n);
}
 
__forceinline const avxi sra(const avxi &a, const int32_t b)
{
  return _mm256_srai_epi32(a.m256, b);
}
__forceinline const avxi srl(const avxi &a, const int32_t b)
{
  return _mm256_srli_epi32(a.m256, b);
}
#else
__forceinline const avxi operator<<(const avxi &a, const int32_t n)
{
  return avxi(_mm_slli_epi32(a.l, n), _mm_slli_epi32(a.h, n));
}
__forceinline const avxi operator>>(const avxi &a, const int32_t n)
{
  return avxi(_mm_srai_epi32(a.l, n), _mm_srai_epi32(a.h, n));
}
 
__forceinline const avxi sra(const avxi &a, const int32_t b)
{
  return avxi(_mm_srai_epi32(a.l, b), _mm_srai_epi32(a.h, b));
}
__forceinline const avxi srl(const avxi &a, const int32_t b)
{
  return avxi(_mm_srli_epi32(a.l, b), _mm_srli_epi32(a.h, b));
}
#endif
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi min(const avxi &a, const avxi &b)
{
  return _mm256_min_epi32(a.m256, b.m256);
}
#else
__forceinline const avxi min(const avxi &a, const avxi &b)
{
  return avxi(_mm_min_epi32(a.l, b.l), _mm_min_epi32(a.h, b.h));
}
#endif
__forceinline const avxi min(const avxi &a, const int32_t b)
{
  return min(a, avxi(b));
}
__forceinline const avxi min(const int32_t a, const avxi &b)
{
  return min(avxi(a), b);
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxi max(const avxi &a, const avxi &b)
{
  return _mm256_max_epi32(a.m256, b.m256);
}
#else
__forceinline const avxi max(const avxi &a, const avxi &b)
{
  return avxi(_mm_max_epi32(a.l, b.l), _mm_max_epi32(a.h, b.h));
}
#endif
__forceinline const avxi max(const avxi &a, const int32_t b)
{
  return max(a, avxi(b));
}
__forceinline const avxi max(const int32_t a, const avxi &b)
{
  return max(avxi(a), b);
}
 
 
__forceinline avxi &operator+=(avxi &a, const avxi &b)
{
  return a = a + b;
}
__forceinline avxi &operator+=(avxi &a, const int32_t b)
{
  return a = a + b;
}
 
__forceinline avxi &operator-=(avxi &a, const avxi &b)
{
  return a = a - b;
}
__forceinline avxi &operator-=(avxi &a, const int32_t b)
{
  return a = a - b;
}
 
__forceinline avxi &operator*=(avxi &a, const avxi &b)
{
  return a = a * b;
}
__forceinline avxi &operator*=(avxi &a, const int32_t b)
{
  return a = a * b;
}
 
__forceinline avxi &operator&=(avxi &a, const avxi &b)
{
  return a = a & b;
}
__forceinline avxi &operator&=(avxi &a, const int32_t b)
{
  return a = a & b;
}
 
__forceinline avxi &operator|=(avxi &a, const avxi &b)
{
  return a = a | b;
}
__forceinline avxi &operator|=(avxi &a, const int32_t b)
{
  return a = a | b;
}
 
__forceinline avxi &operator^=(avxi &a, const avxi &b)
{
  return a = a ^ b;
}
__forceinline avxi &operator^=(avxi &a, const int32_t b)
{
  return a = a ^ b;
}
 
__forceinline avxi &operator<<=(avxi &a, const int32_t b)
{
  return a = a << b;
}
__forceinline avxi &operator>>=(avxi &a, const int32_t b)
{
  return a = a >> b;
}
 
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxb operator==(const avxi &a, const avxi &b)
{
  return _mm256_castsi256_ps(_mm256_cmpeq_epi32(a.m256, b.m256));
}
#else
__forceinline const avxb operator==(const avxi &a, const avxi &b)
{
  return avxb(_mm_castsi128_ps(_mm_cmpeq_epi32(a.l, b.l)),
              _mm_castsi128_ps(_mm_cmpeq_epi32(a.h, b.h)));
}
#endif
__forceinline const avxb operator==(const avxi &a, const int32_t b)
{
  return a == avxi(b);
}
__forceinline const avxb operator==(const int32_t a, const avxi &b)
{
  return avxi(a) == b;
}
 
__forceinline const avxb operator!=(const avxi &a, const avxi &b)
{
  return !(a == b);
}
__forceinline const avxb operator!=(const avxi &a, const int32_t b)
{
  return a != avxi(b);
}
__forceinline const avxb operator!=(const int32_t a, const avxi &b)
{
  return avxi(a) != b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxb operator<(const avxi &a, const avxi &b)
{
  return _mm256_castsi256_ps(_mm256_cmpgt_epi32(b.m256, a.m256));
}
#else
__forceinline const avxb operator<(const avxi &a, const avxi &b)
{
  return avxb(_mm_castsi128_ps(_mm_cmplt_epi32(a.l, b.l)),
              _mm_castsi128_ps(_mm_cmplt_epi32(a.h, b.h)));
}
#endif
__forceinline const avxb operator<(const avxi &a, const int32_t b)
{
  return a < avxi(b);
}
__forceinline const avxb operator<(const int32_t a, const avxi &b)
{
  return avxi(a) < b;
}
 
__forceinline const avxb operator>=(const avxi &a, const avxi &b)
{
  return !(a < b);
}
__forceinline const avxb operator>=(const avxi &a, const int32_t b)
{
  return a >= avxi(b);
}
__forceinline const avxb operator>=(const int32_t a, const avxi &b)
{
  return avxi(a) >= b;
}
 
#if defined(__KERNEL_AVX2__)
__forceinline const avxb operator>(const avxi &a, const avxi &b)
{
  return _mm256_castsi256_ps(_mm256_cmpgt_epi32(a.m256, b.m256));
}
#else
__forceinline const avxb operator>(const avxi &a, const avxi &b)
{
  return avxb(_mm_castsi128_ps(_mm_cmpgt_epi32(a.l, b.l)),
              _mm_castsi128_ps(_mm_cmpgt_epi32(a.h, b.h)));
}
#endif
__forceinline const avxb operator>(const avxi &a, const int32_t b)
{
  return a > avxi(b);
}
__forceinline const avxb operator>(const int32_t a, const avxi &b)
{
  return avxi(a) > b;
}
 
__forceinline const avxb operator<=(const avxi &a, const avxi &b)
{
  return !(a > b);
}
__forceinline const avxb operator<=(const avxi &a, const int32_t b)
{
  return a <= avxi(b);
}
__forceinline const avxb operator<=(const int32_t a, const avxi &b)
{
  return avxi(a) <= b;
}
 
__forceinline const avxi select(const avxb &m, const avxi &t, const avxi &f)
{
  return _mm256_castps_si256(_mm256_blendv_ps(_mm256_castsi256_ps(f), _mm256_castsi256_ps(t), m));
}
 
 
#if defined(__KERNEL_AVX2__)
__forceinline avxi unpacklo(const avxi &a, const avxi &b)
{
  return _mm256_unpacklo_epi32(a.m256, b.m256);
}
__forceinline avxi unpackhi(const avxi &a, const avxi &b)
{
  return _mm256_unpackhi_epi32(a.m256, b.m256);
}
#else
__forceinline avxi unpacklo(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_unpacklo_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)));
}
__forceinline avxi unpackhi(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_unpackhi_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b)));
}
#endif
 
template<size_t i> __forceinline const avxi shuffle(const avxi &a)
{
  return _mm256_castps_si256(_mm256_permute_ps(_mm256_castsi256_ps(a), _MM_SHUFFLE(i, i, i, i)));
}
 
template<size_t i0, size_t i1> __forceinline const avxi shuffle(const avxi &a)
{
  return _mm256_permute2f128_si256(a, a, (i1 << 4) | (i0 << 0));
}
 
template<size_t i0, size_t i1> __forceinline const avxi shuffle(const avxi &a, const avxi &b)
{
  return _mm256_permute2f128_si256(a, b, (i1 << 4) | (i0 << 0));
}
 
template<size_t i0, size_t i1, size_t i2, size_t i3>
__forceinline const avxi shuffle(const avxi &a)
{
  return _mm256_castps_si256(
      _mm256_permute_ps(_mm256_castsi256_ps(a), _MM_SHUFFLE(i3, i2, i1, i0)));
}
 
template<size_t i0, size_t i1, size_t i2, size_t i3>
__forceinline const avxi shuffle(const avxi &a, const avxi &b)
{
  return _mm256_castps_si256(_mm256_shuffle_ps(
      _mm256_castsi256_ps(a), _mm256_castsi256_ps(b), _MM_SHUFFLE(i3, i2, i1, i0)));
}
 
template<> __forceinline const avxi shuffle<0, 0, 2, 2>(const avxi &b)
{
  return _mm256_castps_si256(_mm256_moveldup_ps(_mm256_castsi256_ps(b)));
}
template<> __forceinline const avxi shuffle<1, 1, 3, 3>(const avxi &b)
{
  return _mm256_castps_si256(_mm256_movehdup_ps(_mm256_castsi256_ps(b)));
}
template<> __forceinline const avxi shuffle<0, 1, 0, 1>(const avxi &b)
{
  return _mm256_castps_si256(
      _mm256_castpd_ps(_mm256_movedup_pd(_mm256_castps_pd(_mm256_castsi256_ps(b)))));
}
 
__forceinline const avxi broadcast(const int *ptr)
{
  return _mm256_castps_si256(_mm256_broadcast_ss((const float *)ptr));
}
template<size_t i> __forceinline const avxi insert(const avxi &a, const ssei &b)
{
  return _mm256_insertf128_si256(a, b, i);
}
template<size_t i> __forceinline const ssei extract(const avxi &a)
{
  return _mm256_extractf128_si256(a, i);
}
 
 
__forceinline const avxi vreduce_min2(const avxi &v)
{
  return min(v, shuffle<1, 0, 3, 2>(v));
}
__forceinline const avxi vreduce_min4(const avxi &v)
{
  avxi v1 = vreduce_min2(v);
  return min(v1, shuffle<2, 3, 0, 1>(v1));
}
__forceinline const avxi vreduce_min(const avxi &v)
{
  avxi v1 = vreduce_min4(v);
  return min(v1, shuffle<1, 0>(v1));
}
 
__forceinline const avxi vreduce_max2(const avxi &v)
{
  return max(v, shuffle<1, 0, 3, 2>(v));
}
__forceinline const avxi vreduce_max4(const avxi &v)
{
  avxi v1 = vreduce_max2(v);
  return max(v1, shuffle<2, 3, 0, 1>(v1));
}
__forceinline const avxi vreduce_max(const avxi &v)
{
  avxi v1 = vreduce_max4(v);
  return max(v1, shuffle<1, 0>(v1));
}
 
__forceinline const avxi vreduce_add2(const avxi &v)
{
  return v + shuffle<1, 0, 3, 2>(v);
}
__forceinline const avxi vreduce_add4(const avxi &v)
{
  avxi v1 = vreduce_add2(v);
  return v1 + shuffle<2, 3, 0, 1>(v1);
}
__forceinline const avxi vreduce_add(const avxi &v)
{
  avxi v1 = vreduce_add4(v);
  return v1 + shuffle<1, 0>(v1);
}
 
__forceinline int reduce_min(const avxi &v)
{
  return extract<0>(extract<0>(vreduce_min(v)));
}
__forceinline int reduce_max(const avxi &v)
{
  return extract<0>(extract<0>(vreduce_max(v)));
}
__forceinline int reduce_add(const avxi &v)
{
  return extract<0>(extract<0>(vreduce_add(v)));
}
 
__forceinline uint32_t select_min(const avxi &v)
{
  return __bsf(movemask(v == vreduce_min(v)));
}
__forceinline uint32_t select_max(const avxi &v)
{
  return __bsf(movemask(v == vreduce_max(v)));
}
 
__forceinline uint32_t select_min(const avxb &valid, const avxi &v)
{
  const avxi a = select(valid, v, avxi(pos_inf));
  return __bsf(movemask(valid & (a == vreduce_min(a))));
}
__forceinline uint32_t select_max(const avxb &valid, const avxi &v)
{
  const avxi a = select(valid, v, avxi(neg_inf));
  return __bsf(movemask(valid & (a == vreduce_max(a))));
}
 
 
ccl_device_inline void print_avxi(const char *label, const avxi &a)
{
  printf("%s: %d %d %d %d %d %d %d %d\n", label, a[0], a[1], a[2], a[3], a[4], a[5], a[6], a[7]);
}
 
CCL_NAMESPACE_END
 
#endif