Eigen/src/Core/arch/AVX/TypeCasting.h

// This file is part of Eigen, a lightweight C++ template library
// for linear algebra.
//
// Copyright (C) 2015 Benoit Steiner <benoit.steiner.goog@gmail.com>
//
// This Source Code Form is subject to the terms of the Mozilla
// Public License v. 2.0. If a copy of the MPL was not distributed
// with this file, You can obtain one at http://mozilla.org/MPL/2.0/.

#ifndef EIGEN_TYPE_CASTING_AVX_H
#define EIGEN_TYPE_CASTING_AVX_H

// IWYU pragma: private
#include "../../InternalHeaderCheck.h"

namespace Eigen {

namespace internal {

#ifndef EIGEN_VECTORIZE_AVX512
template <>
struct type_casting_traits<float, bool> : vectorized_type_casting_traits<float, bool> {};
template <>
struct type_casting_traits<bool, float> : vectorized_type_casting_traits<bool, float> {};

template <>
struct type_casting_traits<float, int> : vectorized_type_casting_traits<float, int> {};
template <>
struct type_casting_traits<int, float> : vectorized_type_casting_traits<int, float> {};

template <>
struct type_casting_traits<float, double> : vectorized_type_casting_traits<float, double> {};
template <>
struct type_casting_traits<double, float> : vectorized_type_casting_traits<double, float> {};

template <>
struct type_casting_traits<double, int> : vectorized_type_casting_traits<double, int> {};
template <>
struct type_casting_traits<int, double> : vectorized_type_casting_traits<int, double> {};

template <>
struct type_casting_traits<half, float> : vectorized_type_casting_traits<half, float> {};
template <>
struct type_casting_traits<float, half> : vectorized_type_casting_traits<float, half> {};

template <>
struct type_casting_traits<bfloat16, float> : vectorized_type_casting_traits<bfloat16, float> {};
template <>
struct type_casting_traits<float, bfloat16> : vectorized_type_casting_traits<float, bfloat16> {};

#ifdef EIGEN_VECTORIZE_AVX2
template <>
struct type_casting_traits<double, int64_t> : vectorized_type_casting_traits<double, int64_t> {};
template <>
struct type_casting_traits<int64_t, double> : vectorized_type_casting_traits<int64_t, double> {};
#endif
#endif

template <>
EIGEN_STRONG_INLINE Packet16b pcast<Packet8f, Packet16b>(const Packet8f& a, const Packet8f& b) {
  __m256 nonzero_a = _mm256_cmp_ps(a, pzero(a), _CMP_NEQ_UQ);
  __m256 nonzero_b = _mm256_cmp_ps(b, pzero(b), _CMP_NEQ_UQ);
  constexpr char kFF = '\255';
#ifndef EIGEN_VECTORIZE_AVX2
  __m128i shuffle_mask128_a_lo = _mm_set_epi8(kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, 12, 8, 4, 0);
  __m128i shuffle_mask128_a_hi = _mm_set_epi8(kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, 12, 8, 4, 0, kFF, kFF, kFF, kFF);
  __m128i shuffle_mask128_b_lo = _mm_set_epi8(kFF, kFF, kFF, kFF, 12, 8, 4, 0, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF);
  __m128i shuffle_mask128_b_hi = _mm_set_epi8(12, 8, 4, 0, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF);
  __m128i a_hi = _mm_shuffle_epi8(_mm256_extractf128_si256(_mm256_castps_si256(nonzero_a), 1), shuffle_mask128_a_hi);
  __m128i a_lo = _mm_shuffle_epi8(_mm256_extractf128_si256(_mm256_castps_si256(nonzero_a), 0), shuffle_mask128_a_lo);
  __m128i b_hi = _mm_shuffle_epi8(_mm256_extractf128_si256(_mm256_castps_si256(nonzero_b), 1), shuffle_mask128_b_hi);
  __m128i b_lo = _mm_shuffle_epi8(_mm256_extractf128_si256(_mm256_castps_si256(nonzero_b), 0), shuffle_mask128_b_lo);
  __m128i merged = _mm_or_si128(_mm_or_si128(b_lo, b_hi), _mm_or_si128(a_lo, a_hi));
  return _mm_and_si128(merged, _mm_set1_epi8(1));
#else
  __m256i a_shuffle_mask = _mm256_set_epi8(kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, 12, 8, 4, 0, kFF, kFF, kFF, kFF, kFF,
                                           kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, 12, 8, 4, 0);
  __m256i b_shuffle_mask = _mm256_set_epi8(12, 8, 4, 0, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF,
                                           kFF, kFF, kFF, 12, 8, 4, 0, kFF, kFF, kFF, kFF, kFF, kFF, kFF, kFF);
  __m256i a_shuff = _mm256_shuffle_epi8(_mm256_castps_si256(nonzero_a), a_shuffle_mask);
  __m256i b_shuff = _mm256_shuffle_epi8(_mm256_castps_si256(nonzero_b), b_shuffle_mask);
  __m256i a_or_b = _mm256_or_si256(a_shuff, b_shuff);
  __m256i merged = _mm256_or_si256(a_or_b, _mm256_castsi128_si256(_mm256_extractf128_si256(a_or_b, 1)));
  return _mm256_castsi256_si128(_mm256_and_si256(merged, _mm256_set1_epi8(1)));
#endif
}

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet16b, Packet8f>(const Packet16b& a) {
  const __m256 cst_one = _mm256_set1_ps(1.0f);
#ifdef EIGEN_VECTORIZE_AVX2
  __m256i a_extended = _mm256_cvtepi8_epi32(a);
  __m256i abcd_efgh = _mm256_cmpeq_epi32(a_extended, _mm256_setzero_si256());
#else
  __m128i abcd_efhg_ijkl_mnop = _mm_cmpeq_epi8(a, _mm_setzero_si128());
  __m128i aabb_ccdd_eeff_gghh = _mm_unpacklo_epi8(abcd_efhg_ijkl_mnop, abcd_efhg_ijkl_mnop);
  __m128i aaaa_bbbb_cccc_dddd = _mm_unpacklo_epi8(aabb_ccdd_eeff_gghh, aabb_ccdd_eeff_gghh);
  __m128i eeee_ffff_gggg_hhhh = _mm_unpackhi_epi8(aabb_ccdd_eeff_gghh, aabb_ccdd_eeff_gghh);
  __m256i abcd_efgh = _mm256_setr_m128i(aaaa_bbbb_cccc_dddd, eeee_ffff_gggg_hhhh);
#endif
  __m256 result = _mm256_andnot_ps(_mm256_castsi256_ps(abcd_efgh), cst_one);
  return result;
}

template <>
EIGEN_STRONG_INLINE Packet8i pcast<Packet8f, Packet8i>(const Packet8f& a) {
  return _mm256_cvttps_epi32(a);
}

template <>
EIGEN_STRONG_INLINE Packet8i pcast<Packet4d, Packet8i>(const Packet4d& a, const Packet4d& b) {
  return _mm256_set_m128i(_mm256_cvttpd_epi32(b), _mm256_cvttpd_epi32(a));
}

template <>
EIGEN_STRONG_INLINE Packet4i pcast<Packet4d, Packet4i>(const Packet4d& a) {
  return _mm256_cvttpd_epi32(a);
}

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet8i, Packet8f>(const Packet8i& a) {
  return _mm256_cvtepi32_ps(a);
}

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet4d, Packet8f>(const Packet4d& a, const Packet4d& b) {
  return _mm256_set_m128(_mm256_cvtpd_ps(b), _mm256_cvtpd_ps(a));
}

template <>
EIGEN_STRONG_INLINE Packet4f pcast<Packet4d, Packet4f>(const Packet4d& a) {
  return _mm256_cvtpd_ps(a);
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet8i, Packet4d>(const Packet8i& a) {
  return _mm256_cvtepi32_pd(_mm256_castsi256_si128(a));
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet4i, Packet4d>(const Packet4i& a) {
  return _mm256_cvtepi32_pd(a);
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet8f, Packet4d>(const Packet8f& a) {
  return _mm256_cvtps_pd(_mm256_castps256_ps128(a));
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet4f, Packet4d>(const Packet4f& a) {
  return _mm256_cvtps_pd(a);
}

template <>
EIGEN_STRONG_INLINE Packet8i preinterpret<Packet8i, Packet8f>(const Packet8f& a) {
  return _mm256_castps_si256(a);
}

template <>
EIGEN_STRONG_INLINE Packet8f preinterpret<Packet8f, Packet8i>(const Packet8i& a) {
  return _mm256_castsi256_ps(a);
}

template <>
EIGEN_STRONG_INLINE Packet8ui preinterpret<Packet8ui, Packet8i>(const Packet8i& a) {
  return Packet8ui(a);
}

template <>
EIGEN_STRONG_INLINE Packet8i preinterpret<Packet8i, Packet8ui>(const Packet8ui& a) {
  return Packet8i(a);
}

// truncation operations

template <>
EIGEN_STRONG_INLINE Packet4f preinterpret<Packet4f, Packet8f>(const Packet8f& a) {
  return _mm256_castps256_ps128(a);
}

template <>
EIGEN_STRONG_INLINE Packet2d preinterpret<Packet2d, Packet4d>(const Packet4d& a) {
  return _mm256_castpd256_pd128(a);
}

template <>
EIGEN_STRONG_INLINE Packet4i preinterpret<Packet4i, Packet8i>(const Packet8i& a) {
  return _mm256_castsi256_si128(a);
}

template <>
EIGEN_STRONG_INLINE Packet4ui preinterpret<Packet4ui, Packet8ui>(const Packet8ui& a) {
  return _mm256_castsi256_si128(a);
}

#ifdef EIGEN_VECTORIZE_AVX2
template <>
EIGEN_STRONG_INLINE Packet4l pcast<Packet4d, Packet4l>(const Packet4d& a) {
#if defined(EIGEN_VECTORIZE_AVX512DQ) && defined(EIGEN_VECTORIZE_AVS512VL)
  return _mm256_cvttpd_epi64(a);
#else

  // if 'a' exceeds the numerical limits of int64_t, the behavior is undefined

  // e <= 0 corresponds to |a| < 1, which should result in zero. incidentally, intel intrinsics with shift arguments
  // greater than or equal to 64 produce zero. furthermore, negative shifts appear to be interpreted as large positive
  // shifts (two's complement), which also result in zero. therefore, e does not need to be clamped to [0, 64)

  constexpr int kTotalBits = sizeof(double) * CHAR_BIT, kMantissaBits = std::numeric_limits<double>::digits - 1,
                kExponentBits = kTotalBits - kMantissaBits - 1, kBias = (1 << (kExponentBits - 1)) - 1;

  const __m256i cst_one = _mm256_set1_epi64x(1);
  const __m256i cst_total_bits = _mm256_set1_epi64x(kTotalBits);
  const __m256i cst_bias = _mm256_set1_epi64x(kBias);

  __m256i a_bits = _mm256_castpd_si256(a);
  // shift left by 1 to clear the sign bit, and shift right by kMantissaBits + 1 to recover biased exponent
  __m256i biased_e = _mm256_srli_epi64(_mm256_slli_epi64(a_bits, 1), kMantissaBits + 1);
  __m256i e = _mm256_sub_epi64(biased_e, cst_bias);

  // shift to the left by kExponentBits + 1 to clear the sign and exponent bits
  __m256i shifted_mantissa = _mm256_slli_epi64(a_bits, kExponentBits + 1);
  // shift to the right by kTotalBits - e to convert the significand to an integer
  __m256i result_significand = _mm256_srlv_epi64(shifted_mantissa, _mm256_sub_epi64(cst_total_bits, e));

  // add the implied bit
  __m256i result_exponent = _mm256_sllv_epi64(cst_one, e);
  // e <= 0 is interpreted as a large positive shift (2's complement), which also conveniently results in zero
  __m256i result = _mm256_add_epi64(result_significand, result_exponent);
  // handle negative arguments
  __m256i sign_mask = _mm256_cmpgt_epi64(_mm256_setzero_si256(), a_bits);
  result = _mm256_sub_epi64(_mm256_xor_si256(result, sign_mask), sign_mask);
  return result;
#endif
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet4l, Packet4d>(const Packet4l& a) {
#if defined(EIGEN_VECTORIZE_AVX512DQ) && defined(EIGEN_VECTORIZE_AVS512VL)
  return _mm256_cvtepi64_pd(a);
#else
  EIGEN_ALIGN16 int64_t aux[4];
  pstore(aux, a);
  return _mm256_set_pd(static_cast<double>(aux[3]), static_cast<double>(aux[2]), static_cast<double>(aux[1]),
                       static_cast<double>(aux[0]));
#endif
}

template <>
EIGEN_STRONG_INLINE Packet4d pcast<Packet2l, Packet4d>(const Packet2l& a, const Packet2l& b) {
  return _mm256_set_m128d((pcast<Packet2l, Packet2d>(b)), (pcast<Packet2l, Packet2d>(a)));
}

template <>
EIGEN_STRONG_INLINE Packet4ul preinterpret<Packet4ul, Packet4l>(const Packet4l& a) {
  return Packet4ul(a);
}

template <>
EIGEN_STRONG_INLINE Packet4l preinterpret<Packet4l, Packet4ul>(const Packet4ul& a) {
  return Packet4l(a);
}

template <>
EIGEN_STRONG_INLINE Packet4l preinterpret<Packet4l, Packet4d>(const Packet4d& a) {
  return _mm256_castpd_si256(a);
}

template <>
EIGEN_STRONG_INLINE Packet4d preinterpret<Packet4d, Packet4l>(const Packet4l& a) {
  return _mm256_castsi256_pd(a);
}

// truncation operations
template <>
EIGEN_STRONG_INLINE Packet2l preinterpret<Packet2l, Packet4l>(const Packet4l& a) {
  return _mm256_castsi256_si128(a);
}
#endif

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet8h, Packet8f>(const Packet8h& a) {
  return half2float(a);
}

template <>
EIGEN_STRONG_INLINE Packet8f pcast<Packet8bf, Packet8f>(const Packet8bf& a) {
  return Bf16ToF32(a);
}

template <>
EIGEN_STRONG_INLINE Packet8h pcast<Packet8f, Packet8h>(const Packet8f& a) {
  return float2half(a);
}

template <>
EIGEN_STRONG_INLINE Packet8bf pcast<Packet8f, Packet8bf>(const Packet8f& a) {
  return F32ToBf16(a);
}

}  // end namespace internal

}  // end namespace Eigen

#endif  // EIGEN_TYPE_CASTING_AVX_H