GitHub-Proxy/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2025-09-24 07:13:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

Used AVX instructions to vectorize the predux_min<Packet8f>, predux_min<Packet4d>, predux_max<Packet8f>, and predux_max<Packet4d> packet primitives.

Browse Source
This commit is contained in:
Benoit Steiner 2014-03-24 13:33:40 -07:00
parent 08f7b3221d
commit 7ae9b0805d
1 changed files with 10 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
Eigen/src/Core/arch/AVX/PacketMath.h
View File

@ -304,37 +304,27 @@ template<> EIGEN_STRONG_INLINE double predux_mul<Packet4d>(const Packet4d& a)
template<> EIGEN_STRONG_INLINE float predux_min<Packet8f>(const Packet8f& a) template<> EIGEN_STRONG_INLINE float predux_min<Packet8f>(const Packet8f& a)
{ {
float result = a[0]; Packet8f tmp = _mm256_min_ps(a, _mm256_permute2f128_ps(a,a,1));
for (int i = 1; i < 8; ++i) { tmp = _mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
if (a[i] < result) result = a[i]; return pfirst(_mm256_min_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
}
return result;
} }
template<> EIGEN_STRONG_INLINE double predux_min<Packet4d>(const Packet4d& a) template<> EIGEN_STRONG_INLINE double predux_min<Packet4d>(const Packet4d& a)
{ {
double result = a[0]; Packet4d tmp = _mm256_min_pd(a, _mm256_permute2f128_pd(a,a,1));
for (int i = 1; i < 4; ++i) { return pfirst(_mm256_min_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
if (a[i] < result) result = a[i];
}
return result;
} }
template<> EIGEN_STRONG_INLINE float predux_max<Packet8f>(const Packet8f& a) template<> EIGEN_STRONG_INLINE float predux_max<Packet8f>(const Packet8f& a)
{ {
float result = a[0]; Packet8f tmp = _mm256_max_ps(a, _mm256_permute2f128_ps(a,a,1));
for (int i = 1; i < 8; ++i) { tmp = _mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,_MM_SHUFFLE(1,0,3,2)));
if (a[i] > result) result = a[i]; return pfirst(_mm256_max_ps(tmp, _mm256_shuffle_ps(tmp,tmp,1)));
}
return result;
} }
template<> EIGEN_STRONG_INLINE double predux_max<Packet4d>(const Packet4d& a) template<> EIGEN_STRONG_INLINE double predux_max<Packet4d>(const Packet4d& a)
{ {
double result = a[0]; Packet4d tmp = _mm256_max_pd(a, _mm256_permute2f128_pd(a,a,1));
for (int i = 1; i < 4; ++i) { return pfirst(_mm256_max_pd(tmp, _mm256_shuffle_pd(tmp, tmp, 1)));
if (a[i] > result) result = a[i];
}
return result;
} }