Skip to content

Commit

Permalink
Merge pull request #22 from JohanMabille/basic
Browse files Browse the repository at this point in the history 
basic operations
  • Loading branch information
@JohanMabille
JohanMabille committed May 24, 2017
2 parents 0153203 + 4517d43 commit bfd5b04
Show file tree
Hide file tree
Showing 7 changed files with 134 additions and 4 deletions.
21 changes: 21 additions & 0 deletions include/xsimd/types/xsimd_avx_double.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,8 @@ namespace xsimd
void store_aligned(double* dst) const;
void store_unaligned(double* dst) const;

double operator[](std::size_t index) const;

private:

__m256d m_value;
Expand All @@ -100,6 +102,8 @@ namespace xsimd

batch<double, 4> min(const batch<double, 4>& lhs, const batch<double, 4>& rhs);
batch<double, 4> max(const batch<double, 4>& lhs, const batch<double, 4>& rhs);
batch<double, 4> fmin(const batch<double, 4>& lhs, const batch<double, 4>& rhs);
batch<double, 4> fmax(const batch<double, 4>& lhs, const batch<double, 4>& rhs);

batch<double, 4> abs(const batch<double, 4>& rhs);
batch<double, 4> sqrt(const batch<double, 4>& rhs);
Expand Down Expand Up @@ -236,6 +240,13 @@ namespace xsimd
_mm256_storeu_pd(dst, m_value);
}

inline double batch<double, 4>::operator[](std::size_t index) const
{
alignas(32) double x[4];
store_aligned(x);
return x[index & 3];
}

inline batch<double, 4> operator-(const batch<double, 4>& rhs)
{
return _mm256_xor_pd(rhs, _mm256_castsi256_pd(_mm256_set1_epi64x(0x8000000000000000)));
Expand Down Expand Up @@ -311,6 +322,16 @@ namespace xsimd
return _mm256_max_pd(lhs, rhs);
}

inline batch<double, 4> fmin(const batch<double, 4>& lhs, const batch<double, 4>& rhs)
{
return min(lhs, rhs);
}

inline batch<double, 4> fmax(const batch<double, 4>& lhs, const batch<double, 4>& rhs)
{
return max(lhs, rhs);
}

inline batch<double, 4> abs(const batch<double, 4>& rhs)
{
__m256d sign_mask = _mm256_set1_pd(-0.); // -0. = 1 << 63
Expand Down
21 changes: 21 additions & 0 deletions include/xsimd/types/xsimd_avx_float.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -79,6 +79,8 @@ namespace xsimd
void store_aligned(float* dst) const;
void store_unaligned(float* dst) const;

float operator[](std::size_t index) const;

private:

__m256 m_value;
Expand All @@ -102,6 +104,8 @@ namespace xsimd

batch<float, 8> min(const batch<float, 8>& lhs, const batch<float, 8>& rhs);
batch<float, 8> max(const batch<float, 8>& lhs, const batch<float, 8>& rhs);
batch<float, 8> fmin(const batch<float, 8>& lhs, const batch<float, 8>& rhs);
batch<float, 8> fmax(const batch<float, 8>& lhs, const batch<float, 8>& rhs);

batch<float, 8> abs(const batch<float, 8>& rhs);
batch<float, 8> sqrt(const batch<float, 8>& rhs);
Expand Down Expand Up @@ -240,6 +244,13 @@ namespace xsimd
_mm256_store_ps(dst, m_value);
}

inline float batch<float, 8>::operator[](std::size_t index) const
{
alignas(32) float x[8];
store_aligned(x);
return x[index & 7];
}

inline batch<float, 8> operator-(const batch<float, 8>& rhs)
{
return _mm256_xor_ps(rhs, _mm256_castsi256_ps(_mm256_set1_epi32(0x80000000)));
Expand Down Expand Up @@ -315,6 +326,16 @@ namespace xsimd
return _mm256_max_ps(lhs, rhs);
}

inline batch<float, 8> fmin(const batch<float, 8>& lhs, const batch<float, 8>& rhs)
{
return min(lhs, rhs);
}

inline batch<float, 8> fmax(const batch<float, 8>& lhs, const batch<float, 8>& rhs)
{
return max(lhs, rhs);
}

inline batch<float, 8> abs(const batch<float, 8>& rhs)
{
__m256 sign_mask = _mm256_set1_ps(-0.f); // -0.f = 1 << 31
Expand Down
9 changes: 9 additions & 0 deletions include/xsimd/types/xsimd_avx_int.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,8 @@ namespace xsimd
void store_aligned(int* dst) const;
void store_unaligned(int* dst) const;

int operator[](std::size_t index) const;

private:

__m256i m_value;
Expand Down Expand Up @@ -273,6 +275,13 @@ namespace xsimd
_mm256_storeu_si256((__m256i*)dst, m_value);
}

inline int batch<int, 8>::operator[](std::size_t index) const
{
alignas(32) int x[8];
store_aligned(x);
return x[index & 7];
}

inline batch<int, 8> operator-(const batch<int, 8>& rhs)
{
#if XSIMD_X86_INSTR_SET >= XSIMD_X86_AVX2_VERSION
Expand Down
21 changes: 21 additions & 0 deletions include/xsimd/types/xsimd_sse_double.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,8 @@ namespace xsimd
void store_aligned(double* dst) const;
void store_unaligned(double* dst) const;

double operator[](std::size_t index) const;

private:

__m128d m_value;
Expand All @@ -100,6 +102,8 @@ namespace xsimd

batch<double, 2> min(const batch<double, 2>& lhs, const batch<double, 2>& rhs);
batch<double, 2> max(const batch<double, 2>& lhs, const batch<double, 2>& rhs);
batch<double, 2> fmin(const batch<double, 2>& lhs, const batch<double, 2>& rhs);
batch<double, 2> fmax(const batch<double, 2>& lhs, const batch<double, 2>& rhs);

batch<double, 2> abs(const batch<double, 2>& rhs);
batch<double, 2> sqrt(const batch<double, 2>& rhs);
Expand Down Expand Up @@ -235,6 +239,13 @@ namespace xsimd
_mm_storeu_pd(dst, m_value);
}

inline double batch<double, 2>::operator[](std::size_t index) const
{
alignas(16) double x[2];
store_aligned(x);
return x[index & 1];
}

inline batch<double, 2> operator-(const batch<double, 2>& rhs)
{
return _mm_xor_pd(rhs, _mm_castsi128_pd(_mm_setr_epi32(0, 0x80000000,
Expand Down Expand Up @@ -311,6 +322,16 @@ namespace xsimd
return _mm_max_pd(lhs, rhs);
}

inline batch<double, 2> fmin(const batch<double, 2>& lhs, const batch<double, 2>& rhs)
{
return min(lhs, rhs);
}

inline batch<double, 2> fmax(const batch<double, 2>& lhs, const batch<double, 2>& rhs)
{
return max(lhs, rhs);
}

inline batch<double, 2> abs(const batch<double, 2>& rhs)
{
__m128d sign_mask = _mm_set1_pd(-0.); // -0. = 1 << 63
Expand Down
21 changes: 21 additions & 0 deletions include/xsimd/types/xsimd_sse_float.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -77,6 +77,8 @@ namespace xsimd
void store_aligned(float* dst) const;
void store_unaligned(float* dst) const;

float operator[](std::size_t index) const;

private:

__m128 m_value;
Expand All @@ -100,6 +102,8 @@ namespace xsimd

batch<float, 4> min(const batch<float, 4>& lhs, const batch<float, 4>& rhs);
batch<float, 4> max(const batch<float, 4>& lhs, const batch<float, 4>& rhs);
batch<float, 4> fmin(const batch<float, 4>& lhs, const batch<float, 4>& rhs);
batch<float, 4> fmax(const batch<float, 4>& lhs, const batch<float, 4>& rhs);

batch<float, 4> abs(const batch<float, 4>& rhs);
batch<float, 4> sqrt(const batch<float, 4>& rhs);
Expand Down Expand Up @@ -234,6 +238,13 @@ namespace xsimd
_mm_storeu_ps(dst, m_value);
}

inline float batch<float, 4>::operator[](std::size_t index) const
{
alignas(16) float x[4];
store_aligned(x);
return x[index & 3];
}

inline batch<float, 4> operator-(const batch<float, 4>& rhs)
{
return _mm_xor_ps(rhs, _mm_castsi128_ps(_mm_set1_epi32(0x80000000)));
Expand Down Expand Up @@ -309,6 +320,16 @@ namespace xsimd
return _mm_max_ps(lhs, rhs);
}

inline batch<float, 4> fmin(const batch<float, 4>& lhs, const batch<float, 4>& rhs)
{
return min(lhs, rhs);
}

inline batch<float, 4> fmax(const batch<float, 4>& lhs, const batch<float, 4>& rhs)
{
return max(lhs, rhs);
}

inline batch<float, 4> abs(const batch<float, 4>& rhs)
{
__m128 sign_mask = _mm_set1_ps(-0.f); // -0.f = 1 << 31
Expand Down
9 changes: 9 additions & 0 deletions include/xsimd/types/xsimd_sse_int.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,8 @@ namespace xsimd
void store_aligned(int* dst) const;
void store_unaligned(int* dst) const;

int operator[](std::size_t index) const;

private:

__m128i m_value;
Expand Down Expand Up @@ -232,6 +234,13 @@ namespace xsimd
_mm_storeu_si128((__m128i*)dst, m_value);
}

inline int batch<int, 4>::operator[](std::size_t index) const
{
alignas(16) int x[4];
store_aligned(x);
return x[index & 3];
}

inline batch<int, 4> operator-(const batch<int, 4>& rhs)
{
return _mm_sub_epi32(_mm_setzero_si128(), rhs);
Expand Down
36 changes: 32 additions & 4 deletions test/xsimd_basic_test.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,7 @@ namespace xsimd
res_type lhs;
res_type rhs;

value_type extract_res;
res_type minus_res;
res_type add_vv_res;
res_type add_vs_res;
Expand Down Expand Up @@ -106,6 +107,7 @@ namespace xsimd
{
lhs[i] = value_type(i) / 4 + value_type(1.2) * std::sqrt(value_type(i + 0.25));
rhs[i] = value_type(10.2) / (i+2) + value_type(0.25);
extract_res = lhs[1];
minus_res[i] = -lhs[i];
add_vv_res[i] = lhs[i] + rhs[i];
add_vs_res[i] = lhs[i] + s;
Expand Down Expand Up @@ -154,6 +156,7 @@ namespace xsimd
res_type lhs;
res_type rhs;

value_type extract_res;
res_type minus_res;
res_type add_vv_res;
res_type add_vs_res;
Expand Down Expand Up @@ -221,6 +224,7 @@ namespace xsimd
{
lhs[i] = value_type(i) * 10;
rhs[i] = value_type(4) + value_type(i);
extract_res = lhs[1];
minus_res[i] = -lhs[i];
add_vv_res[i] = lhs[i] + rhs[i];
add_vs_res[i] = lhs[i] + s;
Expand Down Expand Up @@ -274,6 +278,12 @@ namespace xsimd
out << space << name << " " << val_type << std::endl;
out << dash << name_shift << '-' << shift << dash << std::endl << std::endl;

out << "operator[] : ";
detail::load_vec(lhs, tester.lhs);
value_type es = lhs[1];
tmp_success = check_almost_equal(es, tester.extract_res, out);
success = success && tmp_success;

out << "load/store aligned : ";
detail::load_vec(lhs, tester.lhs);
detail::store_vec(lhs, res);
Expand Down Expand Up @@ -435,6 +445,18 @@ namespace xsimd
tmp_success = check_almost_equal(res, tester.max_res, out);
success = success && tmp_success;

out << "fmin(simd, simd) : ";
vres = fmin(lhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.min_res, out);
success = success && tmp_success;

out << "fmax(simd, simd) : ";
vres = fmax(lhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.max_res, out);
success = success && tmp_success;

out << "abs(simd) : ";
vres = abs(lhs);
detail::store_vec(vres, res);
Expand All @@ -459,13 +481,13 @@ namespace xsimd
tmp_success = check_almost_equal(res, tester.fms_res, out);
success = success && tmp_success;

out << "fnma(simd, simd, simd) : ";
out << "fnma(simd, simd, simd) : ";
vres = fnma(lhs, rhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.fnma_res, out);
success = success && tmp_success;

out << "fnms(simd, simd, simd) : ";
out << "fnms(simd, simd, simd) : ";
vres = fnms(lhs, rhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.fnms_res, out);
Expand Down Expand Up @@ -506,6 +528,12 @@ namespace xsimd
out << space << name << " " << val_type << std::endl;
out << dash << name_shift << '-' << shift << dash << std::endl << std::endl;

out << "operator[] : ";
detail::load_vec(lhs, tester.lhs);
value_type es = lhs[1];
tmp_success = check_almost_equal(es, tester.extract_res, out);
success = success && tmp_success;

out << "load/store aligned : ";
detail::load_vec(lhs, tester.lhs);
detail::store_vec(lhs, res);
Expand Down Expand Up @@ -658,13 +686,13 @@ namespace xsimd
tmp_success = check_almost_equal(res, tester.fms_res, out);
success = success && tmp_success;

out << "fnma(simd, simd, simd) : ";
out << "fnma(simd, simd, simd) : ";
vres = fnma(lhs, rhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.fnma_res, out);
success = success && tmp_success;

out << "fnms(simd, simd, simd) : ";
out << "fnms(simd, simd, simd) : ";
vres = fnms(lhs, rhs, rhs);
detail::store_vec(vres, res);
tmp_success = check_almost_equal(res, tester.fnms_res, out);
Expand Down

0 comments on commit bfd5b04

Please sign in to comment.