devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

bug #86 : use internal:: namespace instead of ei_ prefix

Browse Source
This commit is contained in:
Benoit Jacob
2010-10-25 10:15:22 -04:00
parent ca85a1f6c5
commit 4716040703
330 changed files with 7615 additions and 7032 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
bench/BenchSparseUtil.h
View File

@@ -31,7 +31,7 @@ void fillMatrix(float density, int rows, int cols, EigenSparseMatrix& dst)
{
for(int i = 0; i < rows; i++)
{
Scalar v = (ei_random<float>(0,1) < density) ? ei_random<Scalar>() : 0;
Scalar v = (internal::random<float>(0,1) < density) ? internal::random<Scalar>() : 0;
if (v!=0)
dst.insert(i,j) = v;
}
@@ -48,12 +48,12 @@ void fillMatrix2(int nnzPerCol, int rows, int cols, EigenSparseMatrix& dst)
std::set<int> aux;
for(int i = 0; i < nnzPerCol; i++)
{
int k = ei_random<int>(0,rows-1);
int k = internal::random<int>(0,rows-1);
while (aux.find(k)!=aux.end())
k = ei_random<int>(0,rows-1);
k = internal::random<int>(0,rows-1);
aux.insert(k);
dst.insert(k,j) = ei_random<Scalar>();
dst.insert(k,j) = internal::random<Scalar>();
}
}
dst.finalize();

6
bench/benchBlasGemm.cpp
View File

@@ -208,9 +208,9 @@ void check_product(void)
int M, N, K;
for (uint i=0; i<1000; ++i)
{
M = ei_random<int>(1,64);
N = ei_random<int>(1,768);
K = ei_random<int>(1,768);
M = internal::random<int>(1,64);
N = internal::random<int>(1,768);
K = internal::random<int>(1,768);
M = (0 + M) * 1;
std::cout << M << " x " << N << " x " << K << "\n";
check_product(M, N, K);

4
bench/benchCholesky.cpp
View File

@@ -49,8 +49,8 @@ __attribute__ ((noinline)) void benchLLT(const MatrixType& m)
BenchTimer timerNoSqrt, timerSqrt;
Scalar acc = 0;
int r = ei_random<int>(0,covMat.rows()-1);
int c = ei_random<int>(0,covMat.cols()-1);
int r = internal::random<int>(0,covMat.rows()-1);
int c = internal::random<int>(0,covMat.cols()-1);
for (int t=0; t<TRIES; ++t)
{
timerNoSqrt.start();

4
bench/benchEigenSolver.cpp
View File

@@ -48,8 +48,8 @@ __attribute__ ((noinline)) void benchEigenSolver(const MatrixType& m)
BenchTimer timerSa, timerStd;
Scalar acc = 0;
int r = ei_random<int>(0,covMat.rows()-1);
int c = ei_random<int>(0,covMat.cols()-1);
int r = internal::random<int>(0,covMat.rows()-1);
int c = internal::random<int>(0,covMat.cols()-1);
{
SelfAdjointEigenSolver<SquareMatrixType> ei(covMat);
for (int t=0; t<TRIES; ++t)

70
bench/benchVecAdd.cpp
View File

@@ -22,9 +22,9 @@ int main(int argc, char* argv[])
{
int size = SIZE * 8;
int size2 = size * size;
Scalar* a = ei_aligned_new<Scalar>(size2);
Scalar* b = ei_aligned_new<Scalar>(size2+4)+1;
Scalar* c = ei_aligned_new<Scalar>(size2);
Scalar* a = internal::aligned_new<Scalar>(size2);
Scalar* b = internal::aligned_new<Scalar>(size2+4)+1;
Scalar* c = internal::aligned_new<Scalar>(size2);
for (int i=0; i<size; ++i)
{
@@ -90,46 +90,46 @@ void benchVec(VectorXf& a, VectorXf& b, VectorXf& c)
void benchVec(Scalar* a, Scalar* b, Scalar* c, int size)
{
typedef ei_packet_traits<Scalar>::type PacketScalar;
const int PacketSize = ei_packet_traits<Scalar>::size;
typedef internal::packet_traits<Scalar>::type PacketScalar;
const int PacketSize = internal::packet_traits<Scalar>::size;
PacketScalar a0, a1, a2, a3, b0, b1, b2, b3;
for (int k=0; k<REPEAT; ++k)
for (int i=0; i<size; i+=PacketSize*8)
{
// a0 = ei_pload(&a[i]);
// b0 = ei_pload(&b[i]);
// a1 = ei_pload(&a[i+1*PacketSize]);
// b1 = ei_pload(&b[i+1*PacketSize]);
// a2 = ei_pload(&a[i+2*PacketSize]);
// b2 = ei_pload(&b[i+2*PacketSize]);
// a3 = ei_pload(&a[i+3*PacketSize]);
// b3 = ei_pload(&b[i+3*PacketSize]);
// ei_pstore(&a[i], ei_padd(a0, b0));
// a0 = ei_pload(&a[i+4*PacketSize]);
// b0 = ei_pload(&b[i+4*PacketSize]);
// a0 = internal::pload(&a[i]);
// b0 = internal::pload(&b[i]);
// a1 = internal::pload(&a[i+1*PacketSize]);
// b1 = internal::pload(&b[i+1*PacketSize]);
// a2 = internal::pload(&a[i+2*PacketSize]);
// b2 = internal::pload(&b[i+2*PacketSize]);
// a3 = internal::pload(&a[i+3*PacketSize]);
// b3 = internal::pload(&b[i+3*PacketSize]);
// internal::pstore(&a[i], internal::padd(a0, b0));
// a0 = internal::pload(&a[i+4*PacketSize]);
// b0 = internal::pload(&b[i+4*PacketSize]);
//
// ei_pstore(&a[i+1*PacketSize], ei_padd(a1, b1));
// a1 = ei_pload(&a[i+5*PacketSize]);
// b1 = ei_pload(&b[i+5*PacketSize]);
// internal::pstore(&a[i+1*PacketSize], internal::padd(a1, b1));
// a1 = internal::pload(&a[i+5*PacketSize]);
// b1 = internal::pload(&b[i+5*PacketSize]);
//
// ei_pstore(&a[i+2*PacketSize], ei_padd(a2, b2));
// a2 = ei_pload(&a[i+6*PacketSize]);
// b2 = ei_pload(&b[i+6*PacketSize]);
// internal::pstore(&a[i+2*PacketSize], internal::padd(a2, b2));
// a2 = internal::pload(&a[i+6*PacketSize]);
// b2 = internal::pload(&b[i+6*PacketSize]);
//
// ei_pstore(&a[i+3*PacketSize], ei_padd(a3, b3));
// a3 = ei_pload(&a[i+7*PacketSize]);
// b3 = ei_pload(&b[i+7*PacketSize]);
// internal::pstore(&a[i+3*PacketSize], internal::padd(a3, b3));
// a3 = internal::pload(&a[i+7*PacketSize]);
// b3 = internal::pload(&b[i+7*PacketSize]);
//
// ei_pstore(&a[i+4*PacketSize], ei_padd(a0, b0));
// ei_pstore(&a[i+5*PacketSize], ei_padd(a1, b1));
// ei_pstore(&a[i+6*PacketSize], ei_padd(a2, b2));
// ei_pstore(&a[i+7*PacketSize], ei_padd(a3, b3));
// internal::pstore(&a[i+4*PacketSize], internal::padd(a0, b0));
// internal::pstore(&a[i+5*PacketSize], internal::padd(a1, b1));
// internal::pstore(&a[i+6*PacketSize], internal::padd(a2, b2));
// internal::pstore(&a[i+7*PacketSize], internal::padd(a3, b3));
ei_pstore(&a[i+2*PacketSize], ei_padd(ei_ploadu(&a[i+2*PacketSize]), ei_ploadu(&b[i+2*PacketSize])));
ei_pstore(&a[i+3*PacketSize], ei_padd(ei_ploadu(&a[i+3*PacketSize]), ei_ploadu(&b[i+3*PacketSize])));
ei_pstore(&a[i+4*PacketSize], ei_padd(ei_ploadu(&a[i+4*PacketSize]), ei_ploadu(&b[i+4*PacketSize])));
ei_pstore(&a[i+5*PacketSize], ei_padd(ei_ploadu(&a[i+5*PacketSize]), ei_ploadu(&b[i+5*PacketSize])));
ei_pstore(&a[i+6*PacketSize], ei_padd(ei_ploadu(&a[i+6*PacketSize]), ei_ploadu(&b[i+6*PacketSize])));
ei_pstore(&a[i+7*PacketSize], ei_padd(ei_ploadu(&a[i+7*PacketSize]), ei_ploadu(&b[i+7*PacketSize])));
internal::pstore(&a[i+2*PacketSize], internal::padd(internal::ploadu(&a[i+2*PacketSize]), internal::ploadu(&b[i+2*PacketSize])));
internal::pstore(&a[i+3*PacketSize], internal::padd(internal::ploadu(&a[i+3*PacketSize]), internal::ploadu(&b[i+3*PacketSize])));
internal::pstore(&a[i+4*PacketSize], internal::padd(internal::ploadu(&a[i+4*PacketSize]), internal::ploadu(&b[i+4*PacketSize])));
internal::pstore(&a[i+5*PacketSize], internal::padd(internal::ploadu(&a[i+5*PacketSize]), internal::ploadu(&b[i+5*PacketSize])));
internal::pstore(&a[i+6*PacketSize], internal::padd(internal::ploadu(&a[i+6*PacketSize]), internal::ploadu(&b[i+6*PacketSize])));
internal::pstore(&a[i+7*PacketSize], internal::padd(internal::ploadu(&a[i+7*PacketSize]), internal::ploadu(&b[i+7*PacketSize])));
}
}

6
bench/bench_gemm.cpp
View File

@@ -116,11 +116,11 @@ EIGEN_DONT_INLINE void gemm(const A& a, const B& b, C& c)
int main(int argc, char ** argv)
{
std::ptrdiff_t l1 = ei_queryL1CacheSize();
std::ptrdiff_t l2 = ei_queryTopLevelCacheSize();
std::ptrdiff_t l1 = internal::queryL1CacheSize();
std::ptrdiff_t l2 = internal::queryTopLevelCacheSize();
std::cout << "L1 cache size = " << (l1>0 ? l1/1024 : -1) << " KB\n";
std::cout << "L2/L3 cache size = " << (l2>0 ? l2/1024 : -1) << " KB\n";
typedef ei_gebp_traits<Scalar,Scalar> Traits;
typedef internal::gebp_traits<Scalar,Scalar> Traits;
std::cout << "Register blocking = " << Traits::mr << " x " << Traits::nr << "\n";
int rep = 1; // number of repetitions per try

112
bench/bench_norm.cpp
View File

@@ -32,18 +32,18 @@ EIGEN_DONT_INLINE typename T::Scalar lapackNorm(T& v)
Scalar ssq = 1;
for (int i=0;i<n;++i)
{
Scalar ax = ei_abs(v.coeff(i));
Scalar ax = internal::abs(v.coeff(i));
if (scale >= ax)
{
ssq += ei_abs2(ax/scale);
ssq += internal::abs2(ax/scale);
}
else
{
ssq = Scalar(1) + ssq * ei_abs2(scale/ax);
ssq = Scalar(1) + ssq * internal::abs2(scale/ax);
scale = ax;
}
}
return scale * ei_sqrt(ssq);
return scale * internal::sqrt(ssq);
}
template<typename T>
@@ -73,15 +73,15 @@ EIGEN_DONT_INLINE typename T::Scalar divacNorm(T& v)
v(i) = v(2*i) + v(2*i+1);
n = n/2;
}
return ei_sqrt(v(0));
return internal::sqrt(v(0));
}
#ifdef EIGEN_VECTORIZE
Packet4f ei_plt(const Packet4f& a, Packet4f& b) { return _mm_cmplt_ps(a,b); }
Packet2d ei_plt(const Packet2d& a, Packet2d& b) { return _mm_cmplt_pd(a,b); }
Packet4f internal::plt(const Packet4f& a, Packet4f& b) { return _mm_cmplt_ps(a,b); }
Packet2d internal::plt(const Packet2d& a, Packet2d& b) { return _mm_cmplt_pd(a,b); }
Packet4f ei_pandnot(const Packet4f& a, Packet4f& b) { return _mm_andnot_ps(a,b); }
Packet2d ei_pandnot(const Packet2d& a, Packet2d& b) { return _mm_andnot_pd(a,b); }
Packet4f internal::pandnot(const Packet4f& a, Packet4f& b) { return _mm_andnot_ps(a,b); }
Packet2d internal::pandnot(const Packet2d& a, Packet2d& b) { return _mm_andnot_pd(a,b); }
#endif
template<typename T>
@@ -112,7 +112,7 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v)
if(iemin > 1 - 2*it || 1+it>iemax || (it==2 && ibeta<5)
|| (it<=4 && ibeta <= 3 ) || it<2)
{
ei_assert(false && "the algorithm cannot be guaranteed on this computer");
eigen_assert(false && "the algorithm cannot be guaranteed on this computer");
}
iexp = -((1-iemin)/2);
b1 = std::pow(ibeta, iexp); // lower boundary of midrange
@@ -126,60 +126,60 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v)
overfl = rbig*s2m; // overfow boundary for abig
eps = std::pow(ibeta, 1-it);
relerr = ei_sqrt(eps); // tolerance for neglecting asml
relerr = internal::sqrt(eps); // tolerance for neglecting asml
abig = 1.0/eps - 1.0;
if (Scalar(nbig)>abig) nmax = abig; // largest safe n
else nmax = nbig;
}
typedef typename ei_packet_traits<Scalar>::type Packet;
const int ps = ei_packet_traits<Scalar>::size;
Packet pasml = ei_pset1(Scalar(0));
Packet pamed = ei_pset1(Scalar(0));
Packet pabig = ei_pset1(Scalar(0));
Packet ps2m = ei_pset1(s2m);
Packet ps1m = ei_pset1(s1m);
Packet pb2 = ei_pset1(b2);
Packet pb1 = ei_pset1(b1);
typedef typename internal::packet_traits<Scalar>::type Packet;
const int ps = internal::packet_traits<Scalar>::size;
Packet pasml = internal::pset1(Scalar(0));
Packet pamed = internal::pset1(Scalar(0));
Packet pabig = internal::pset1(Scalar(0));
Packet ps2m = internal::pset1(s2m);
Packet ps1m = internal::pset1(s1m);
Packet pb2 = internal::pset1(b2);
Packet pb1 = internal::pset1(b1);
for(int j=0; j<v.size(); j+=ps)
{
Packet ax = ei_pabs(v.template packet<Aligned>(j));
Packet ax_s2m = ei_pmul(ax,ps2m);
Packet ax_s1m = ei_pmul(ax,ps1m);
Packet maskBig = ei_plt(pb2,ax);
Packet maskSml = ei_plt(ax,pb1);
Packet ax = internal::pabs(v.template packet<Aligned>(j));
Packet ax_s2m = internal::pmul(ax,ps2m);
Packet ax_s1m = internal::pmul(ax,ps1m);
Packet maskBig = internal::plt(pb2,ax);
Packet maskSml = internal::plt(ax,pb1);
// Packet maskMed = ei_pand(maskSml,maskBig);
// Packet scale = ei_pset1(Scalar(0));
// scale = ei_por(scale, ei_pand(maskBig,ps2m));
// scale = ei_por(scale, ei_pand(maskSml,ps1m));
// scale = ei_por(scale, ei_pandnot(ei_pset1(Scalar(1)),maskMed));
// ax = ei_pmul(ax,scale);
// ax = ei_pmul(ax,ax);
// pabig = ei_padd(pabig, ei_pand(maskBig, ax));
// pasml = ei_padd(pasml, ei_pand(maskSml, ax));
// pamed = ei_padd(pamed, ei_pandnot(ax,maskMed));
// Packet maskMed = internal::pand(maskSml,maskBig);
// Packet scale = internal::pset1(Scalar(0));
// scale = internal::por(scale, internal::pand(maskBig,ps2m));
// scale = internal::por(scale, internal::pand(maskSml,ps1m));
// scale = internal::por(scale, internal::pandnot(internal::pset1(Scalar(1)),maskMed));
// ax = internal::pmul(ax,scale);
// ax = internal::pmul(ax,ax);
// pabig = internal::padd(pabig, internal::pand(maskBig, ax));
// pasml = internal::padd(pasml, internal::pand(maskSml, ax));
// pamed = internal::padd(pamed, internal::pandnot(ax,maskMed));
pabig = ei_padd(pabig, ei_pand(maskBig, ei_pmul(ax_s2m,ax_s2m)));
pasml = ei_padd(pasml, ei_pand(maskSml, ei_pmul(ax_s1m,ax_s1m)));
pamed = ei_padd(pamed, ei_pandnot(ei_pmul(ax,ax),ei_pand(maskSml,maskBig)));
pabig = internal::padd(pabig, internal::pand(maskBig, internal::pmul(ax_s2m,ax_s2m)));
pasml = internal::padd(pasml, internal::pand(maskSml, internal::pmul(ax_s1m,ax_s1m)));
pamed = internal::padd(pamed, internal::pandnot(internal::pmul(ax,ax),internal::pand(maskSml,maskBig)));
}
Scalar abig = ei_predux(pabig);
Scalar asml = ei_predux(pasml);
Scalar amed = ei_predux(pamed);
Scalar abig = internal::predux(pabig);
Scalar asml = internal::predux(pasml);
Scalar amed = internal::predux(pamed);
if(abig > Scalar(0))
{
abig = ei_sqrt(abig);
abig = internal::sqrt(abig);
if(abig > overfl)
{
ei_assert(false && "overflow");
eigen_assert(false && "overflow");
return rbig;
}
if(amed > Scalar(0))
{
abig = abig/s2m;
amed = ei_sqrt(amed);
amed = internal::sqrt(amed);
}
else
{
@@ -191,24 +191,24 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v)
{
if (amed > Scalar(0))
{
abig = ei_sqrt(amed);
amed = ei_sqrt(asml) / s1m;
abig = internal::sqrt(amed);
amed = internal::sqrt(asml) / s1m;
}
else
{
return ei_sqrt(asml)/s1m;
return internal::sqrt(asml)/s1m;
}
}
else
{
return ei_sqrt(amed);
return internal::sqrt(amed);
}
asml = std::min(abig, amed);
abig = std::max(abig, amed);
if(asml <= abig*relerr)
return abig;
else
return abig * ei_sqrt(Scalar(1) + ei_abs2(asml/abig));
return abig * internal::sqrt(Scalar(1) + internal::abs2(asml/abig));
#endif
}
@@ -234,12 +234,12 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v)
void check_accuracy(double basef, double based, int s)
{
double yf = basef * ei_abs(ei_random<double>());
double yd = based * ei_abs(ei_random<double>());
double yf = basef * internal::abs(internal::random<double>());
double yd = based * internal::abs(internal::random<double>());
VectorXf vf = VectorXf::Ones(s) * yf;
VectorXd vd = VectorXd::Ones(s) * yd;
std::cout << "reference\t" << ei_sqrt(double(s))*yf << "\t" << ei_sqrt(double(s))*yd << "\n";
std::cout << "reference\t" << internal::sqrt(double(s))*yf << "\t" << internal::sqrt(double(s))*yd << "\n";
std::cout << "sqsumNorm\t" << sqsumNorm(vf) << "\t" << sqsumNorm(vd) << "\n";
std::cout << "hypotNorm\t" << hypotNorm(vf) << "\t" << hypotNorm(vd) << "\n";
std::cout << "blueNorm\t" << blueNorm(vf) << "\t" << blueNorm(vd) << "\n";
@@ -255,11 +255,11 @@ void check_accuracy_var(int ef0, int ef1, int ed0, int ed1, int s)
VectorXd vd(s);
for (int i=0; i<s; ++i)
{
vf[i] = ei_abs(ei_random<double>()) * std::pow(double(10), ei_random<int>(ef0,ef1));
vd[i] = ei_abs(ei_random<double>()) * std::pow(double(10), ei_random<int>(ed0,ed1));
vf[i] = internal::abs(internal::random<double>()) * std::pow(double(10), internal::random<int>(ef0,ef1));
vd[i] = internal::abs(internal::random<double>()) * std::pow(double(10), internal::random<int>(ed0,ed1));
}
//std::cout << "reference\t" << ei_sqrt(double(s))*yf << "\t" << ei_sqrt(double(s))*yd << "\n";
//std::cout << "reference\t" << internal::sqrt(double(s))*yf << "\t" << internal::sqrt(double(s))*yd << "\n";
std::cout << "sqsumNorm\t" << sqsumNorm(vf) << "\t" << sqsumNorm(vd) << "\t" << sqsumNorm(vf.cast<long double>()) << "\t" << sqsumNorm(vd.cast<long double>()) << "\n";
std::cout << "hypotNorm\t" << hypotNorm(vf) << "\t" << hypotNorm(vd) << "\t" << hypotNorm(vf.cast<long double>()) << "\t" << hypotNorm(vd.cast<long double>()) << "\n";
std::cout << "blueNorm\t" << blueNorm(vf) << "\t" << blueNorm(vd) << "\t" << blueNorm(vf.cast<long double>()) << "\t" << blueNorm(vd.cast<long double>()) << "\n";
@@ -273,7 +273,7 @@ int main(int argc, char** argv)
{
int tries = 10;
int iters = 100000;
double y = 1.1345743233455785456788e12 * ei_random<double>();
double y = 1.1345743233455785456788e12 * internal::random<double>();
VectorXf v = VectorXf::Ones(1024) * y;
// return 0;

4
bench/bench_reverse.cpp
View File

@@ -28,8 +28,8 @@ __attribute__ ((noinline)) void bench_reverse(const MatrixType& m)
BenchTimer timerB, timerH, timerV;
Scalar acc = 0;
int r = ei_random<int>(0,rows-1);
int c = ei_random<int>(0,cols-1);
int r = internal::random<int>(0,rows-1);
int c = internal::random<int>(0,cols-1);
for (int t=0; t<TRIES; ++t)
{
timerB.start();

8
bench/benchmarkSlice.cpp
View File

@@ -24,10 +24,10 @@ int main(int argc, char *argv[])
for(int a = 0; a < REPEAT; a++)
{
int r, c, nr, nc;
r = Eigen::ei_random<int>(0,10);
c = Eigen::ei_random<int>(0,10);
nr = Eigen::ei_random<int>(50,80);
nc = Eigen::ei_random<int>(50,80);
r = Eigen::internal::random<int>(0,10);
c = Eigen::internal::random<int>(0,10);
nr = Eigen::internal::random<int>(50,80);
nc = Eigen::internal::random<int>(50,80);
m.block(r,c,nr,nc) += Mat::Ones(nr,nc);
m.block(r,c,nr,nc) *= SCALAR(10);
m.block(r,c,nr,nc) -= Mat::constant(nr,nc,10);

16
bench/btl/libs/eigen2/eigen2_interface.hh
View File

@@ -106,13 +106,13 @@ public :
static inline void symv(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){
X.noalias() = (A.template selfadjointView<Lower>() * B);
// ei_product_selfadjoint_vector<real,0,LowerTriangularBit,false,false>(N,A.data(),N, B.data(), 1, X.data(), 1);
// internal::product_selfadjoint_vector<real,0,LowerTriangularBit,false,false>(N,A.data(),N, B.data(), 1, X.data(), 1);
}
template<typename Dest, typename Src> static void triassign(Dest& dst, const Src& src)
{
typedef typename Dest::Scalar Scalar;
typedef typename ei_packet_traits<Scalar>::type Packet;
typedef typename internal::packet_traits<Scalar>::type Packet;
const int PacketSize = sizeof(Packet)/sizeof(Scalar);
int size = dst.cols();
for(int j=0; j<size; j+=1)
@@ -121,7 +121,7 @@ public :
Scalar* A0 = dst.data() + j*dst.stride();
int starti = j;
int alignedEnd = starti;
int alignedStart = (starti) + ei_first_aligned(&A0[starti], size-starti);
int alignedStart = (starti) + internal::first_aligned(&A0[starti], size-starti);
alignedEnd = alignedStart + ((size-alignedStart)/(2*PacketSize))*(PacketSize*2);
// do the non-vectorizable part of the assignment
@@ -155,7 +155,7 @@ public :
}
static EIGEN_DONT_INLINE void syr2(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){
// ei_product_selfadjoint_rank2_update<real,0,LowerTriangularBit>(N,A.data(),N, X.data(), 1, Y.data(), 1, -1);
// internal::product_selfadjoint_rank2_update<real,0,LowerTriangularBit>(N,A.data(),N, X.data(), 1, Y.data(), 1, -1);
for(int j=0; j<N; ++j)
A.col(j).tail(N-j) += X[j] * Y.tail(N-j) + Y[j] * X.tail(N-j);
}
@@ -166,7 +166,7 @@ public :
}
static EIGEN_DONT_INLINE void rot(gene_vector & A, gene_vector & B, real c, real s, int N){
ei_apply_rotation_in_the_plane(A, B, JacobiRotation<real>(c,s));
internal::apply_rotation_in_the_plane(A, B, JacobiRotation<real>(c,s));
}
static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){
@@ -203,7 +203,7 @@ public :
static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){
C = X;
ei_llt_inplace<Lower>::blocked(C);
internal::llt_inplace<Lower>::blocked(C);
//C = X.llt().matrixL();
// C = X;
// Cholesky<gene_matrix>::computeInPlace(C);
@@ -218,14 +218,14 @@ public :
Matrix<DenseIndex,1,Dynamic> piv(N);
DenseIndex nb;
C = X;
ei_partial_lu_inplace(C,piv,nb);
internal::partial_lu_inplace(C,piv,nb);
// C = X.partialPivLu().matrixLU();
}
static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){
typename Tridiagonalization<gene_matrix>::CoeffVectorType aux(N-1);
C = X;
ei_tridiagonalization_inplace(C, aux);
internal::tridiagonalization_inplace(C, aux);
}
static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){

524
bench/btl/libs/hand_vec/hand_vec_interface.hh
View File

@@ -29,8 +29,8 @@ class hand_vec_interface : public f77_interface_base<real> {
public :
typedef typename ei_packet_traits<real>::type Packet;
static const int PacketSize = ei_packet_traits<real>::size;
typedef typename internal::packet_traits<real>::type Packet;
static const int PacketSize = internal::packet_traits<real>::size;
typedef typename f77_interface_base<real>::stl_matrix stl_matrix;
typedef typename f77_interface_base<real>::stl_vector stl_vector;
@@ -38,16 +38,16 @@ public :
typedef typename f77_interface_base<real>::gene_vector gene_vector;
static void free_matrix(gene_matrix & A, int N){
ei_aligned_free(A);
internal::aligned_free(A);
}
static void free_vector(gene_vector & B){
ei_aligned_free(B);
internal::aligned_free(B);
}
static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){
int N = A_stl.size();
A = (real*)ei_aligned_malloc(N*N*sizeof(real));
A = (real*)internal::aligned_malloc(N*N*sizeof(real));
for (int j=0;j<N;j++)
for (int i=0;i<N;i++)
A[i+N*j] = A_stl[j][i];
@@ -55,7 +55,7 @@ public :
static inline void vector_from_stl(gene_vector & B, stl_vector & B_stl){
int N = B_stl.size();
B = (real*)ei_aligned_malloc(N*sizeof(real));
B = (real*)internal::aligned_malloc(N*sizeof(real));
for (int i=0;i<N;i++)
B[i] = B_stl[i];
}
@@ -84,10 +84,10 @@ public :
register real* __restrict__ A2 = A + (i+2)*N;
register real* __restrict__ A3 = A + (i+3)*N;
Packet ptmp0 = ei_pset1(B[i]);
Packet ptmp1 = ei_pset1(B[i+1]);
Packet ptmp2 = ei_pset1(B[i+2]);
Packet ptmp3 = ei_pset1(B[i+3]);
Packet ptmp0 = internal::pset1(B[i]);
Packet ptmp1 = internal::pset1(B[i+1]);
Packet ptmp2 = internal::pset1(B[i+2]);
Packet ptmp3 = internal::pset1(B[i+3]);
// register Packet ptmp0, ptmp1, ptmp2, ptmp3;
// asm(
//
@@ -162,73 +162,73 @@ public :
register Packet A13;
for (int j = 0;j<ANP;j+=2*PacketSize)
{
// A00 = ei_pload(&A0[j]);
// A01 = ei_ploadu(&A1[j]);
// A02 = ei_ploadu(&A2[j]);
// A03 = ei_ploadu(&A3[j]);
// A10 = ei_pload(&A0[j+PacketSize]);
// A11 = ei_ploadu(&A1[j+PacketSize]);
// A12 = ei_ploadu(&A2[j+PacketSize]);
// A13 = ei_ploadu(&A3[j+PacketSize]);
// A00 = internal::pload(&A0[j]);
// A01 = internal::ploadu(&A1[j]);
// A02 = internal::ploadu(&A2[j]);
// A03 = internal::ploadu(&A3[j]);
// A10 = internal::pload(&A0[j+PacketSize]);
// A11 = internal::ploadu(&A1[j+PacketSize]);
// A12 = internal::ploadu(&A2[j+PacketSize]);
// A13 = internal::ploadu(&A3[j+PacketSize]);
//
// A00 = ei_pmul(ptmp0, A00);
// A01 = ei_pmul(ptmp1, A01);
// A02 = ei_pmul(ptmp2, A02);
// A03 = ei_pmul(ptmp3, A03);
// A10 = ei_pmul(ptmp0, A10);
// A11 = ei_pmul(ptmp1, A11);
// A12 = ei_pmul(ptmp2, A12);
// A13 = ei_pmul(ptmp3, A13);
// A00 = internal::pmul(ptmp0, A00);
// A01 = internal::pmul(ptmp1, A01);
// A02 = internal::pmul(ptmp2, A02);
// A03 = internal::pmul(ptmp3, A03);
// A10 = internal::pmul(ptmp0, A10);
// A11 = internal::pmul(ptmp1, A11);
// A12 = internal::pmul(ptmp2, A12);
// A13 = internal::pmul(ptmp3, A13);
//
// A00 = ei_padd(A00,A01);
// A02 = ei_padd(A02,A03);
// A00 = ei_padd(A00,ei_pload(&X[j]));
// A00 = ei_padd(A00,A02);
// ei_pstore(&X[j],A00);
// A00 = internal::padd(A00,A01);
// A02 = internal::padd(A02,A03);
// A00 = internal::padd(A00,internal::pload(&X[j]));
// A00 = internal::padd(A00,A02);
// internal::pstore(&X[j],A00);
//
// A10 = ei_padd(A10,A11);
// A12 = ei_padd(A12,A13);
// A10 = ei_padd(A10,ei_pload(&X[j+PacketSize]));
// A10 = ei_padd(A10,A12);
// ei_pstore(&X[j+PacketSize],A10);
// A10 = internal::padd(A10,A11);
// A12 = internal::padd(A12,A13);
// A10 = internal::padd(A10,internal::pload(&X[j+PacketSize]));
// A10 = internal::padd(A10,A12);
// internal::pstore(&X[j+PacketSize],A10);
ei_pstore(&X[j],
ei_padd(ei_pload(&X[j]),
ei_padd(
ei_padd(ei_pmul(ptmp0,ei_pload(&A0[j])),ei_pmul(ptmp1,ei_ploadu(&A1[j]))),
ei_padd(ei_pmul(ptmp2,ei_ploadu(&A2[j])),ei_pmul(ptmp3,ei_ploadu(&A3[j]))) )));
internal::pstore(&X[j],
internal::padd(internal::pload(&X[j]),
internal::padd(
internal::padd(internal::pmul(ptmp0,internal::pload(&A0[j])),internal::pmul(ptmp1,internal::ploadu(&A1[j]))),
internal::padd(internal::pmul(ptmp2,internal::ploadu(&A2[j])),internal::pmul(ptmp3,internal::ploadu(&A3[j]))) )));
ei_pstore(&X[j+PacketSize],
ei_padd(ei_pload(&X[j+PacketSize]),
ei_padd(
ei_padd(ei_pmul(ptmp0,ei_pload(&A0[j+PacketSize])),ei_pmul(ptmp1,ei_ploadu(&A1[j+PacketSize]))),
ei_padd(ei_pmul(ptmp2,ei_ploadu(&A2[j+PacketSize])),ei_pmul(ptmp3,ei_ploadu(&A3[j+PacketSize]))) )));
internal::pstore(&X[j+PacketSize],
internal::padd(internal::pload(&X[j+PacketSize]),
internal::padd(
internal::padd(internal::pmul(ptmp0,internal::pload(&A0[j+PacketSize])),internal::pmul(ptmp1,internal::ploadu(&A1[j+PacketSize]))),
internal::padd(internal::pmul(ptmp2,internal::ploadu(&A2[j+PacketSize])),internal::pmul(ptmp3,internal::ploadu(&A3[j+PacketSize]))) )));
}
for (int j = ANP;j<AN;j+=PacketSize)
ei_pstore(&X[j],
ei_padd(ei_pload(&X[j]),
ei_padd(
ei_padd(ei_pmul(ptmp0,ei_pload(&A0[j])),ei_pmul(ptmp1,ei_ploadu(&A1[j]))),
ei_padd(ei_pmul(ptmp2,ei_ploadu(&A2[j])),ei_pmul(ptmp3,ei_ploadu(&A3[j]))) )));
internal::pstore(&X[j],
internal::padd(internal::pload(&X[j]),
internal::padd(
internal::padd(internal::pmul(ptmp0,internal::pload(&A0[j])),internal::pmul(ptmp1,internal::ploadu(&A1[j]))),
internal::padd(internal::pmul(ptmp2,internal::ploadu(&A2[j])),internal::pmul(ptmp3,internal::ploadu(&A3[j]))) )));
}
// process remaining scalars
for (int j=AN;j<N;j++)
X[j] += ei_pfirst(ptmp0) * A0[j] + ei_pfirst(ptmp1) * A1[j] + ei_pfirst(ptmp2) * A2[j] + ei_pfirst(ptmp3) * A3[j];
X[j] += internal::pfirst(ptmp0) * A0[j] + internal::pfirst(ptmp1) * A1[j] + internal::pfirst(ptmp2) * A2[j] + internal::pfirst(ptmp3) * A3[j];
}
for (int i=bound;i<N;i++)
{
real tmp0 = B[i];
Packet ptmp0 = ei_pset1(tmp0);
Packet ptmp0 = internal::pset1(tmp0);
int iN0 = i*N;
if (AN>0)
{
bool aligned0 = (iN0 % PacketSize) == 0;
if (aligned0)
for (int j = 0;j<AN;j+=PacketSize)
ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pload(&X[j])));
internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pload(&X[j])));
else
for (int j = 0;j<AN;j+=PacketSize)
ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),ei_pload(&X[j])));
internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),internal::pload(&X[j])));
}
// process remaining scalars
for (int j=AN;j<N;j++)
@@ -254,18 +254,18 @@ public :
register real* __restrict__ A1 = A + (j+1)*N;
real t0 = B[j];
Packet ptmp0 = ei_pset1(t0);
Packet ptmp0 = internal::pset1(t0);
real t1 = B[j+1];
Packet ptmp1 = ei_pset1(t1);
Packet ptmp1 = internal::pset1(t1);
real t2 = 0;
Packet ptmp2 = ei_pset1(t2);
Packet ptmp2 = internal::pset1(t2);
real t3 = 0;
Packet ptmp3 = ei_pset1(t3);
Packet ptmp3 = internal::pset1(t3);
int starti = j+2;
int alignedEnd = starti;
int alignedStart = (starti) + ei_first_aligned(&X[starti], N-starti);
int alignedStart = (starti) + internal::first_aligned(&X[starti], N-starti);
alignedEnd = alignedStart + ((N-alignedStart)/(PacketSize))*(PacketSize);
X[j] += t0 * A0[j];
@@ -282,21 +282,21 @@ public :
}
asm("#begin symv");
for (size_t i=alignedStart; i<alignedEnd; i+=PacketSize) {
Packet A0i = ei_ploadu(&A0[i]);
Packet A1i = ei_ploadu(&A1[i]);
// Packet A0i1 = ei_ploadu(&A0[i+PacketSize]);
Packet Xi = ei_pload(&X[i]);
Packet Bi = ei_pload/*u*/(&B[i]);
// Packet Xi1 = ei_pload(&X[i+PacketSize]);
// Packet Bi1 = ei_pload/*u*/(&B[i+PacketSize]);
Xi = ei_padd(ei_padd(Xi, ei_pmul(ptmp0, A0i)), ei_pmul(ptmp1, A1i));
ptmp2 = ei_padd(ptmp2, ei_pmul(A0i, Bi));
ptmp3 = ei_padd(ptmp3, ei_pmul(A1i, Bi));
// Xi1 = ei_padd(Xi1, ei_pmul(ptmp1, A0i1));
// ptmp2 = ei_padd(ptmp2, ei_pmul(A0i1, Bi1));
Packet A0i = internal::ploadu(&A0[i]);
Packet A1i = internal::ploadu(&A1[i]);
// Packet A0i1 = internal::ploadu(&A0[i+PacketSize]);
Packet Xi = internal::pload(&X[i]);
Packet Bi = internal::pload/*u*/(&B[i]);
// Packet Xi1 = internal::pload(&X[i+PacketSize]);
// Packet Bi1 = internal::pload/*u*/(&B[i+PacketSize]);
Xi = internal::padd(internal::padd(Xi, internal::pmul(ptmp0, A0i)), internal::pmul(ptmp1, A1i));
ptmp2 = internal::padd(ptmp2, internal::pmul(A0i, Bi));
ptmp3 = internal::padd(ptmp3, internal::pmul(A1i, Bi));
// Xi1 = internal::padd(Xi1, internal::pmul(ptmp1, A0i1));
// ptmp2 = internal::padd(ptmp2, internal::pmul(A0i1, Bi1));
//
ei_pstore(&X[i],Xi);
// ei_pstore(&X[i+PacketSize],Xi1);
internal::pstore(&X[i],Xi);
// internal::pstore(&X[i+PacketSize],Xi1);
// asm(
// "prefetchnta 64(%[A0],%[i],4) \n\t"
// //"movups (%[A0],%[i],4), %%xmm8 \n\t"
@@ -341,8 +341,8 @@ public :
}
X[j] += t2 + ei_predux(ptmp2);
X[j+1] += t3 + ei_predux(ptmp3);
X[j] += t2 + internal::predux(ptmp2);
X[j+1] += t3 + internal::predux(ptmp3);
}
for (int j=bound;j<N;j++)
{
@@ -372,13 +372,13 @@ public :
// for (int i=0;i<bound;i+=4)
// {
// real tmp0 = B[i];
// Packet ptmp0 = ei_pset1(tmp0);
// Packet ptmp0 = internal::pset1(tmp0);
// real tmp1 = B[i+1];
// Packet ptmp1 = ei_pset1(tmp1);
// Packet ptmp1 = internal::pset1(tmp1);
// real tmp2 = B[i+2];
// Packet ptmp2 = ei_pset1(tmp2);
// Packet ptmp2 = internal::pset1(tmp2);
// real tmp3 = B[i+3];
// Packet ptmp3 = ei_pset1(tmp3);
// Packet ptmp3 = internal::pset1(tmp3);
// int iN0 = i*N;
// int iN1 = (i+1)*N;
// int iN2 = (i+2)*N;
@@ -392,59 +392,59 @@ public :
// {
// for (int j = 0;j<AN;j+=PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pload(&X[j]),
// ei_padd(
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pmul(ptmp1,ei_pload(&A[j+iN1]))),
// ei_padd(ei_pmul(ptmp2,ei_pload(&A[j+iN2])),ei_pmul(ptmp3,ei_pload(&A[j+iN3]))) )));
// internal::pstore(&X[j],
// internal::padd(internal::pload(&X[j]),
// internal::padd(
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pmul(ptmp1,internal::pload(&A[j+iN1]))),
// internal::padd(internal::pmul(ptmp2,internal::pload(&A[j+iN2])),internal::pmul(ptmp3,internal::pload(&A[j+iN3]))) )));
// }
// }
// else if (aligned1==2)
// {
// for (int j = 0;j<AN;j+=PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pload(&X[j]),
// ei_padd(
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+iN1]))),
// ei_padd(ei_pmul(ptmp2,ei_pload(&A[j+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+iN3]))) )));
// internal::pstore(&X[j],
// internal::padd(internal::pload(&X[j]),
// internal::padd(
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+iN1]))),
// internal::padd(internal::pmul(ptmp2,internal::pload(&A[j+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+iN3]))) )));
// }
// }
// else
// {
// for (int j = 0;j<ANP;j+=2*PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pload(&X[j]),
// ei_padd(
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+iN1]))),
// ei_padd(ei_pmul(ptmp2,ei_ploadu(&A[j+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+iN3]))) )));
// internal::pstore(&X[j],
// internal::padd(internal::pload(&X[j]),
// internal::padd(
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+iN1]))),
// internal::padd(internal::pmul(ptmp2,internal::ploadu(&A[j+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+iN3]))) )));
//
// ei_pstore(&X[j+PacketSize],
// ei_padd(ei_pload(&X[j+PacketSize]),
// ei_padd(
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+PacketSize+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+PacketSize+iN1]))),
// ei_padd(ei_pmul(ptmp2,ei_ploadu(&A[j+PacketSize+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+PacketSize+iN3]))) )));
// internal::pstore(&X[j+PacketSize],
// internal::padd(internal::pload(&X[j+PacketSize]),
// internal::padd(
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+PacketSize+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+PacketSize+iN1]))),
// internal::padd(internal::pmul(ptmp2,internal::ploadu(&A[j+PacketSize+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+PacketSize+iN3]))) )));
//
// // ei_pstore(&X[j+2*PacketSize],
// // ei_padd(ei_pload(&X[j+2*PacketSize]),
// // ei_padd(
// // ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+2*PacketSize+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+2*PacketSize+iN1]))),
// // ei_padd(ei_pmul(ptmp2,ei_ploadu(&A[j+2*PacketSize+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+2*PacketSize+iN3]))) )));
// // internal::pstore(&X[j+2*PacketSize],
// // internal::padd(internal::pload(&X[j+2*PacketSize]),
// // internal::padd(
// // internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+2*PacketSize+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+2*PacketSize+iN1]))),
// // internal::padd(internal::pmul(ptmp2,internal::ploadu(&A[j+2*PacketSize+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+2*PacketSize+iN3]))) )));
// //
// // ei_pstore(&X[j+3*PacketSize],
// // ei_padd(ei_pload(&X[j+3*PacketSize]),
// // ei_padd(
// // ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+3*PacketSize+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+3*PacketSize+iN1]))),
// // ei_padd(ei_pmul(ptmp2,ei_ploadu(&A[j+3*PacketSize+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+3*PacketSize+iN3]))) )));
// // internal::pstore(&X[j+3*PacketSize],
// // internal::padd(internal::pload(&X[j+3*PacketSize]),
// // internal::padd(
// // internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+3*PacketSize+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+3*PacketSize+iN1]))),
// // internal::padd(internal::pmul(ptmp2,internal::ploadu(&A[j+3*PacketSize+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+3*PacketSize+iN3]))) )));
//
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ei_pstore(&X[j],
// ei_padd(ei_pload(&X[j]),
// ei_padd(
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),ei_pmul(ptmp1,ei_ploadu(&A[j+iN1]))),
// ei_padd(ei_pmul(ptmp2,ei_ploadu(&A[j+iN2])),ei_pmul(ptmp3,ei_ploadu(&A[j+iN3]))) )));
// internal::pstore(&X[j],
// internal::padd(internal::pload(&X[j]),
// internal::padd(
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),internal::pmul(ptmp1,internal::ploadu(&A[j+iN1]))),
// internal::padd(internal::pmul(ptmp2,internal::ploadu(&A[j+iN2])),internal::pmul(ptmp3,internal::ploadu(&A[j+iN3]))) )));
// }
// }
// // process remaining scalars
@@ -454,17 +454,17 @@ public :
// for (int i=bound;i<N;i++)
// {
// real tmp0 = B[i];
// Packet ptmp0 = ei_pset1(tmp0);
// Packet ptmp0 = internal::pset1(tmp0);
// int iN0 = i*N;
// if (AN>0)
// {
// bool aligned0 = (iN0 % PacketSize) == 0;
// if (aligned0)
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pload(&X[j])));
// internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pload(&X[j])));
// else
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),ei_pload(&X[j])));
// internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),internal::pload(&X[j])));
// }
// // process remaining scalars
// for (int j=AN;j<N;j++)
@@ -483,9 +483,9 @@ public :
// for (int i=0;i<N;i+=2)
// {
// real tmp0 = B[i];
// Packet ptmp0 = ei_pset1(tmp0);
// Packet ptmp0 = internal::pset1(tmp0);
// real tmp1 = B[i+1];
// Packet ptmp1 = ei_pset1(tmp1);
// Packet ptmp1 = internal::pset1(tmp1);
// int iN0 = i*N;
// int iN1 = (i+1)*N;
// if (AN>0)
@@ -497,27 +497,27 @@ public :
// {
// for (int j = 0;j<AN;j+=PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_pload(&A[j+iN1])),ei_pload(&X[j]))));
// internal::pstore(&X[j],
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::pload(&A[j+iN1])),internal::pload(&X[j]))));
// }
// }
// else if (aligned0)
// {
// for (int j = 0;j<AN;j+=PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+iN1])),ei_pload(&X[j]))));
// internal::pstore(&X[j],
// internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+iN1])),internal::pload(&X[j]))));
// }
// }
// else if (aligned1)
// {
// for (int j = 0;j<AN;j+=PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_pload(&A[j+iN1])),ei_pload(&X[j]))));
// internal::pstore(&X[j],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::pload(&A[j+iN1])),internal::pload(&X[j]))));
// }
// }
// else
@@ -525,26 +525,26 @@ public :
// int ANP = (AN/(4*PacketSize))*4*PacketSize;
// for (int j = 0;j<ANP;j+=4*PacketSize)
// {
// ei_pstore(&X[j],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+iN1])),ei_pload(&X[j]))));
// internal::pstore(&X[j],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+iN1])),internal::pload(&X[j]))));
//
// ei_pstore(&X[j+PacketSize],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+PacketSize+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+PacketSize+iN1])),ei_pload(&X[j+PacketSize]))));
// internal::pstore(&X[j+PacketSize],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+PacketSize+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+PacketSize+iN1])),internal::pload(&X[j+PacketSize]))));
//
// ei_pstore(&X[j+2*PacketSize],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+2*PacketSize+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+2*PacketSize+iN1])),ei_pload(&X[j+2*PacketSize]))));
// internal::pstore(&X[j+2*PacketSize],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+2*PacketSize+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+2*PacketSize+iN1])),internal::pload(&X[j+2*PacketSize]))));
//
// ei_pstore(&X[j+3*PacketSize],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+3*PacketSize+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+3*PacketSize+iN1])),ei_pload(&X[j+3*PacketSize]))));
// internal::pstore(&X[j+3*PacketSize],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+3*PacketSize+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+3*PacketSize+iN1])),internal::pload(&X[j+3*PacketSize]))));
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ei_pstore(&X[j],
// ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),
// ei_padd(ei_pmul(ptmp1,ei_ploadu(&A[j+iN1])),ei_pload(&X[j]))));
// internal::pstore(&X[j],
// internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),
// internal::padd(internal::pmul(ptmp1,internal::ploadu(&A[j+iN1])),internal::pload(&X[j]))));
// }
// }
// // process remaining scalars
@@ -555,17 +555,17 @@ public :
// for (int i=remaining;i<N;i++)
// {
// real tmp0 = B[i];
// Packet ptmp0 = ei_pset1(tmp0);
// Packet ptmp0 = internal::pset1(tmp0);
// int iN0 = i*N;
// if (AN>0)
// {
// bool aligned0 = (iN0 % PacketSize) == 0;
// if (aligned0)
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_pload(&A[j+iN0])),ei_pload(&X[j])));
// internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::pload(&A[j+iN0])),internal::pload(&X[j])));
// else
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pmul(ptmp0,ei_ploadu(&A[j+iN0])),ei_pload(&X[j])));
// internal::pstore(&X[j], internal::padd(internal::pmul(ptmp0,internal::ploadu(&A[j+iN0])),internal::pload(&X[j])));
// }
// // process remaining scalars
// for (int j=AN;j<N;j++)
@@ -583,7 +583,7 @@ public :
// for (int i=0;i<N;i++)
// {
// real tmp = B[i];
// Packet ptmp = ei_pset1(tmp);
// Packet ptmp = internal::pset1(tmp);
// int iN = i*N;
// if (AN>0)
// {
@@ -595,45 +595,45 @@ public :
// int ANP = (AN/(8*PacketSize))*8*PacketSize;
// for (int j = 0;j<ANP;j+=PacketSize*8)
// {
// A0 = ei_pload(&A[j+iN]);
// X0 = ei_pload(&X[j]);
// A1 = ei_pload(&A[j+PacketSize+iN]);
// X1 = ei_pload(&X[j+PacketSize]);
// A2 = ei_pload(&A[j+2*PacketSize+iN]);
// X2 = ei_pload(&X[j+2*PacketSize]);
// ei_pstore(&X[j], ei_padd(X0, ei_pmul(ptmp,A0)));
// A0 = ei_pload(&A[j+3*PacketSize+iN]);
// X0 = ei_pload(&X[j+3*PacketSize]);
// ei_pstore(&X[j+PacketSize], ei_padd(ei_pload(&X1), ei_pmul(ptmp,A1)));
// A1 = ei_pload(&A[j+4*PacketSize+iN]);
// X1 = ei_pload(&X[j+4*PacketSize]);
// ei_pstore(&X[j+2*PacketSize], ei_padd(ei_pload(&X2), ei_pmul(ptmp,A2)));
// A2 = ei_pload(&A[j+5*PacketSize+iN]);
// X2 = ei_pload(&X[j+5*PacketSize]);
// ei_pstore(&X[j+3*PacketSize], ei_padd(ei_pload(&X0), ei_pmul(ptmp,A0)));
// A0 = ei_pload(&A[j+6*PacketSize+iN]);
// X0 = ei_pload(&X[j+6*PacketSize]);
// ei_pstore(&X[j+4*PacketSize], ei_padd(ei_pload(&X1), ei_pmul(ptmp,A1)));
// A1 = ei_pload(&A[j+7*PacketSize+iN]);
// X1 = ei_pload(&X[j+7*PacketSize]);
// ei_pstore(&X[j+5*PacketSize], ei_padd(ei_pload(&X2), ei_pmul(ptmp,A2)));
// ei_pstore(&X[j+6*PacketSize], ei_padd(ei_pload(&X0), ei_pmul(ptmp,A0)));
// ei_pstore(&X[j+7*PacketSize], ei_padd(ei_pload(&X1), ei_pmul(ptmp,A1)));
// A0 = internal::pload(&A[j+iN]);
// X0 = internal::pload(&X[j]);
// A1 = internal::pload(&A[j+PacketSize+iN]);
// X1 = internal::pload(&X[j+PacketSize]);
// A2 = internal::pload(&A[j+2*PacketSize+iN]);
// X2 = internal::pload(&X[j+2*PacketSize]);
// internal::pstore(&X[j], internal::padd(X0, internal::pmul(ptmp,A0)));
// A0 = internal::pload(&A[j+3*PacketSize+iN]);
// X0 = internal::pload(&X[j+3*PacketSize]);
// internal::pstore(&X[j+PacketSize], internal::padd(internal::pload(&X1), internal::pmul(ptmp,A1)));
// A1 = internal::pload(&A[j+4*PacketSize+iN]);
// X1 = internal::pload(&X[j+4*PacketSize]);
// internal::pstore(&X[j+2*PacketSize], internal::padd(internal::pload(&X2), internal::pmul(ptmp,A2)));
// A2 = internal::pload(&A[j+5*PacketSize+iN]);
// X2 = internal::pload(&X[j+5*PacketSize]);
// internal::pstore(&X[j+3*PacketSize], internal::padd(internal::pload(&X0), internal::pmul(ptmp,A0)));
// A0 = internal::pload(&A[j+6*PacketSize+iN]);
// X0 = internal::pload(&X[j+6*PacketSize]);
// internal::pstore(&X[j+4*PacketSize], internal::padd(internal::pload(&X1), internal::pmul(ptmp,A1)));
// A1 = internal::pload(&A[j+7*PacketSize+iN]);
// X1 = internal::pload(&X[j+7*PacketSize]);
// internal::pstore(&X[j+5*PacketSize], internal::padd(internal::pload(&X2), internal::pmul(ptmp,A2)));
// internal::pstore(&X[j+6*PacketSize], internal::padd(internal::pload(&X0), internal::pmul(ptmp,A0)));
// internal::pstore(&X[j+7*PacketSize], internal::padd(internal::pload(&X1), internal::pmul(ptmp,A1)));
// //
// // ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_pload(&A[j+iN]))));
// // ei_pstore(&X[j+PacketSize], ei_padd(ei_pload(&X[j+PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+PacketSize+iN]))));
// // ei_pstore(&X[j+2*PacketSize], ei_padd(ei_pload(&X[j+2*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+2*PacketSize+iN]))));
// // ei_pstore(&X[j+3*PacketSize], ei_padd(ei_pload(&X[j+3*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+3*PacketSize+iN]))));
// // ei_pstore(&X[j+4*PacketSize], ei_padd(ei_pload(&X[j+4*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+4*PacketSize+iN]))));
// // ei_pstore(&X[j+5*PacketSize], ei_padd(ei_pload(&X[j+5*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+5*PacketSize+iN]))));
// // ei_pstore(&X[j+6*PacketSize], ei_padd(ei_pload(&X[j+6*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+6*PacketSize+iN]))));
// // ei_pstore(&X[j+7*PacketSize], ei_padd(ei_pload(&X[j+7*PacketSize]), ei_pmul(ptmp,ei_pload(&A[j+7*PacketSize+iN]))));
// // internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::pload(&A[j+iN]))));
// // internal::pstore(&X[j+PacketSize], internal::padd(internal::pload(&X[j+PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+PacketSize+iN]))));
// // internal::pstore(&X[j+2*PacketSize], internal::padd(internal::pload(&X[j+2*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+2*PacketSize+iN]))));
// // internal::pstore(&X[j+3*PacketSize], internal::padd(internal::pload(&X[j+3*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+3*PacketSize+iN]))));
// // internal::pstore(&X[j+4*PacketSize], internal::padd(internal::pload(&X[j+4*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+4*PacketSize+iN]))));
// // internal::pstore(&X[j+5*PacketSize], internal::padd(internal::pload(&X[j+5*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+5*PacketSize+iN]))));
// // internal::pstore(&X[j+6*PacketSize], internal::padd(internal::pload(&X[j+6*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+6*PacketSize+iN]))));
// // internal::pstore(&X[j+7*PacketSize], internal::padd(internal::pload(&X[j+7*PacketSize]), internal::pmul(ptmp,internal::pload(&A[j+7*PacketSize+iN]))));
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_pload(&A[j+iN]))));
// internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::pload(&A[j+iN]))));
// #else
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_pload(&A[j+iN]))));
// internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::pload(&A[j+iN]))));
// #endif
// }
// else
@@ -642,20 +642,20 @@ public :
// int ANP = (AN/(8*PacketSize))*8*PacketSize;
// for (int j = 0;j<ANP;j+=PacketSize*8)
// {
// ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_ploadu(&A[j+iN]))));
// ei_pstore(&X[j+PacketSize], ei_padd(ei_pload(&X[j+PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+PacketSize+iN]))));
// ei_pstore(&X[j+2*PacketSize], ei_padd(ei_pload(&X[j+2*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+2*PacketSize+iN]))));
// ei_pstore(&X[j+3*PacketSize], ei_padd(ei_pload(&X[j+3*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+3*PacketSize+iN]))));
// ei_pstore(&X[j+4*PacketSize], ei_padd(ei_pload(&X[j+4*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+4*PacketSize+iN]))));
// ei_pstore(&X[j+5*PacketSize], ei_padd(ei_pload(&X[j+5*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+5*PacketSize+iN]))));
// ei_pstore(&X[j+6*PacketSize], ei_padd(ei_pload(&X[j+6*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+6*PacketSize+iN]))));
// ei_pstore(&X[j+7*PacketSize], ei_padd(ei_pload(&X[j+7*PacketSize]), ei_pmul(ptmp,ei_ploadu(&A[j+7*PacketSize+iN]))));
// internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::ploadu(&A[j+iN]))));
// internal::pstore(&X[j+PacketSize], internal::padd(internal::pload(&X[j+PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+PacketSize+iN]))));
// internal::pstore(&X[j+2*PacketSize], internal::padd(internal::pload(&X[j+2*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+2*PacketSize+iN]))));
// internal::pstore(&X[j+3*PacketSize], internal::padd(internal::pload(&X[j+3*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+3*PacketSize+iN]))));
// internal::pstore(&X[j+4*PacketSize], internal::padd(internal::pload(&X[j+4*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+4*PacketSize+iN]))));
// internal::pstore(&X[j+5*PacketSize], internal::padd(internal::pload(&X[j+5*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+5*PacketSize+iN]))));
// internal::pstore(&X[j+6*PacketSize], internal::padd(internal::pload(&X[j+6*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+6*PacketSize+iN]))));
// internal::pstore(&X[j+7*PacketSize], internal::padd(internal::pload(&X[j+7*PacketSize]), internal::pmul(ptmp,internal::ploadu(&A[j+7*PacketSize+iN]))));
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_ploadu(&A[j+iN]))));
// internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::ploadu(&A[j+iN]))));
// #else
// for (int j = 0;j<AN;j+=PacketSize)
// ei_pstore(&X[j], ei_padd(ei_pload(&X[j]), ei_pmul(ptmp,ei_ploadu(&A[j+iN]))));
// internal::pstore(&X[j], internal::padd(internal::pload(&X[j]), internal::pmul(ptmp,internal::ploadu(&A[j+iN]))));
// #endif
// }
// }
@@ -673,13 +673,13 @@ public :
for (int i=0;i<bound;i+=4)
{
real tmp0 = 0;
Packet ptmp0 = ei_pset1(real(0));
Packet ptmp0 = internal::pset1(real(0));
real tmp1 = 0;
Packet ptmp1 = ei_pset1(real(0));
Packet ptmp1 = internal::pset1(real(0));
real tmp2 = 0;
Packet ptmp2 = ei_pset1(real(0));
Packet ptmp2 = internal::pset1(real(0));
real tmp3 = 0;
Packet ptmp3 = ei_pset1(real(0));
Packet ptmp3 = internal::pset1(real(0));
int iN0 = i*N;
int iN1 = (i+1)*N;
int iN2 = (i+2)*N;
@@ -691,39 +691,39 @@ public :
{
for (int j = 0;j<AN;j+=PacketSize)
{
Packet b = ei_pload(&B[j]);
ptmp0 = ei_padd(ptmp0, ei_pmul(b, ei_pload(&A[j+iN0])));
ptmp1 = ei_padd(ptmp1, ei_pmul(b, ei_pload(&A[j+iN1])));
ptmp2 = ei_padd(ptmp2, ei_pmul(b, ei_pload(&A[j+iN2])));
ptmp3 = ei_padd(ptmp3, ei_pmul(b, ei_pload(&A[j+iN3])));
Packet b = internal::pload(&B[j]);
ptmp0 = internal::padd(ptmp0, internal::pmul(b, internal::pload(&A[j+iN0])));
ptmp1 = internal::padd(ptmp1, internal::pmul(b, internal::pload(&A[j+iN1])));
ptmp2 = internal::padd(ptmp2, internal::pmul(b, internal::pload(&A[j+iN2])));
ptmp3 = internal::padd(ptmp3, internal::pmul(b, internal::pload(&A[j+iN3])));
}
}
else if (align1==2)
{
for (int j = 0;j<AN;j+=PacketSize)
{
Packet b = ei_pload(&B[j]);
ptmp0 = ei_padd(ptmp0, ei_pmul(b, ei_pload(&A[j+iN0])));
ptmp1 = ei_padd(ptmp1, ei_pmul(b, ei_ploadu(&A[j+iN1])));
ptmp2 = ei_padd(ptmp2, ei_pmul(b, ei_pload(&A[j+iN2])));
ptmp3 = ei_padd(ptmp3, ei_pmul(b, ei_ploadu(&A[j+iN3])));
Packet b = internal::pload(&B[j]);
ptmp0 = internal::padd(ptmp0, internal::pmul(b, internal::pload(&A[j+iN0])));
ptmp1 = internal::padd(ptmp1, internal::pmul(b, internal::ploadu(&A[j+iN1])));
ptmp2 = internal::padd(ptmp2, internal::pmul(b, internal::pload(&A[j+iN2])));
ptmp3 = internal::padd(ptmp3, internal::pmul(b, internal::ploadu(&A[j+iN3])));
}
}
else
{
for (int j = 0;j<AN;j+=PacketSize)
{
Packet b = ei_pload(&B[j]);
ptmp0 = ei_padd(ptmp0, ei_pmul(b, ei_pload(&A[j+iN0])));
ptmp1 = ei_padd(ptmp1, ei_pmul(b, ei_ploadu(&A[j+iN1])));
ptmp2 = ei_padd(ptmp2, ei_pmul(b, ei_ploadu(&A[j+iN2])));
ptmp3 = ei_padd(ptmp3, ei_pmul(b, ei_ploadu(&A[j+iN3])));
Packet b = internal::pload(&B[j]);
ptmp0 = internal::padd(ptmp0, internal::pmul(b, internal::pload(&A[j+iN0])));
ptmp1 = internal::padd(ptmp1, internal::pmul(b, internal::ploadu(&A[j+iN1])));
ptmp2 = internal::padd(ptmp2, internal::pmul(b, internal::ploadu(&A[j+iN2])));
ptmp3 = internal::padd(ptmp3, internal::pmul(b, internal::ploadu(&A[j+iN3])));
}
}
tmp0 = ei_predux(ptmp0);
tmp1 = ei_predux(ptmp1);
tmp2 = ei_predux(ptmp2);
tmp3 = ei_predux(ptmp3);
tmp0 = internal::predux(ptmp0);
tmp1 = internal::predux(ptmp1);
tmp2 = internal::predux(ptmp2);
tmp3 = internal::predux(ptmp3);
}
// process remaining scalars
for (int j=AN;j<N;j++)
@@ -742,17 +742,17 @@ public :
for (int i=bound;i<N;i++)
{
real tmp0 = 0;
Packet ptmp0 = ei_pset1(real(0));
Packet ptmp0 = internal::pset1(real(0));
int iN0 = i*N;
if (AN>0)
{
if (iN0 % PacketSize==0)
for (int j = 0;j<AN;j+=PacketSize)
ptmp0 = ei_padd(ptmp0, ei_pmul(ei_pload(&B[j]), ei_pload(&A[j+iN0])));
ptmp0 = internal::padd(ptmp0, internal::pmul(internal::pload(&B[j]), internal::pload(&A[j+iN0])));
else
for (int j = 0;j<AN;j+=PacketSize)
ptmp0 = ei_padd(ptmp0, ei_pmul(ei_pload(&B[j]), ei_ploadu(&A[j+iN0])));
tmp0 = ei_predux(ptmp0);
ptmp0 = internal::padd(ptmp0, internal::pmul(internal::pload(&B[j]), internal::ploadu(&A[j+iN0])));
tmp0 = internal::predux(ptmp0);
}
// process remaining scalars
for (int j=AN;j<N;j++)
@@ -769,7 +769,7 @@ public :
// for (int i=0;i<N;i++)
// {
// real tmp = 0;
// Packet ptmp = ei_pset1(real(0));
// Packet ptmp = internal::pset1(real(0));
// int iN = i*N;
// if (AN>0)
// {
@@ -781,21 +781,21 @@ public :
// for (int j = 0;j<ANP;j+=PacketSize*8)
// {
// ptmp =
// ei_padd(ei_pmul(ei_pload(&B[j]), ei_pload(&A[j+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+PacketSize]), ei_pload(&A[j+PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+2*PacketSize]), ei_pload(&A[j+2*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+3*PacketSize]), ei_pload(&A[j+3*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+4*PacketSize]), ei_pload(&A[j+4*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+5*PacketSize]), ei_pload(&A[j+5*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+6*PacketSize]), ei_pload(&A[j+6*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+7*PacketSize]), ei_pload(&A[j+7*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j]), internal::pload(&A[j+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+PacketSize]), internal::pload(&A[j+PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+2*PacketSize]), internal::pload(&A[j+2*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+3*PacketSize]), internal::pload(&A[j+3*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+4*PacketSize]), internal::pload(&A[j+4*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+5*PacketSize]), internal::pload(&A[j+5*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+6*PacketSize]), internal::pload(&A[j+6*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+7*PacketSize]), internal::pload(&A[j+7*PacketSize+iN])),
// ptmp))))))));
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ptmp = ei_padd(ptmp, ei_pmul(ei_pload(&B[j]), ei_pload(&A[j+iN])));
// ptmp = internal::padd(ptmp, internal::pmul(internal::pload(&B[j]), internal::pload(&A[j+iN])));
// #else
// for (int j = 0;j<AN;j+=PacketSize)
// ptmp = ei_padd(ptmp, ei_pmul(ei_pload(&B[j]), ei_pload(&A[j+iN])));
// ptmp = internal::padd(ptmp, internal::pmul(internal::pload(&B[j]), internal::pload(&A[j+iN])));
// #endif
// }
// else
@@ -805,24 +805,24 @@ public :
// for (int j = 0;j<ANP;j+=PacketSize*8)
// {
// ptmp =
// ei_padd(ei_pmul(ei_pload(&B[j]), ei_ploadu(&A[j+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+PacketSize]), ei_ploadu(&A[j+PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+2*PacketSize]), ei_ploadu(&A[j+2*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+3*PacketSize]), ei_ploadu(&A[j+3*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+4*PacketSize]), ei_ploadu(&A[j+4*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+5*PacketSize]), ei_ploadu(&A[j+5*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+6*PacketSize]), ei_ploadu(&A[j+6*PacketSize+iN])),
// ei_padd(ei_pmul(ei_pload(&B[j+7*PacketSize]), ei_ploadu(&A[j+7*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j]), internal::ploadu(&A[j+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+PacketSize]), internal::ploadu(&A[j+PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+2*PacketSize]), internal::ploadu(&A[j+2*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+3*PacketSize]), internal::ploadu(&A[j+3*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+4*PacketSize]), internal::ploadu(&A[j+4*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+5*PacketSize]), internal::ploadu(&A[j+5*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+6*PacketSize]), internal::ploadu(&A[j+6*PacketSize+iN])),
// internal::padd(internal::pmul(internal::pload(&B[j+7*PacketSize]), internal::ploadu(&A[j+7*PacketSize+iN])),
// ptmp))))))));
// }
// for (int j = ANP;j<AN;j+=PacketSize)
// ptmp = ei_padd(ptmp, ei_pmul(ei_pload(&B[j]), ei_ploadu(&A[j+iN])));
// ptmp = internal::padd(ptmp, internal::pmul(internal::pload(&B[j]), internal::ploadu(&A[j+iN])));
// #else
// for (int j = 0;j<AN;j+=PacketSize)
// ptmp = ei_padd(ptmp, ei_pmul(ei_pload(&B[j]), ei_ploadu(&A[j+iN])));
// ptmp = internal::padd(ptmp, internal::pmul(internal::pload(&B[j]), internal::ploadu(&A[j+iN])));
// #endif
// }
// tmp = ei_predux(ptmp);
// tmp = internal::predux(ptmp);
// }
// // process remaining scalars
// for (int j=AN;j<N;j++)
@@ -835,7 +835,7 @@ public :
int AN = (N/PacketSize)*PacketSize;
if (AN>0)
{
Packet pcoef = ei_pset1(coef);
Packet pcoef = internal::pset1(coef);
#ifdef PEELING
const int peelSize = 3;
int ANP = (AN/(peelSize*PacketSize))*peelSize*PacketSize;
@@ -846,33 +846,33 @@ public :
Packet x0,x1,x2,y0,y1,y2;
for (int j = 0;j<ANP;j+=PacketSize*peelSize)
{
x0 = ei_pload(X+j);
x1 = ei_pload(X1+j);
x2 = ei_pload(X2+j);
x0 = internal::pload(X+j);
x1 = internal::pload(X1+j);
x2 = internal::pload(X2+j);
y0 = ei_pload(Y+j);
y1 = ei_pload(Y1+j);
y2 = ei_pload(Y2+j);
y0 = internal::pload(Y+j);
y1 = internal::pload(Y1+j);
y2 = internal::pload(Y2+j);
y0 = ei_pmadd(pcoef, x0, y0);
y1 = ei_pmadd(pcoef, x1, y1);
y2 = ei_pmadd(pcoef, x2, y2);
y0 = internal::pmadd(pcoef, x0, y0);
y1 = internal::pmadd(pcoef, x1, y1);
y2 = internal::pmadd(pcoef, x2, y2);
ei_pstore(Y+j, y0);
ei_pstore(Y1+j, y1);
ei_pstore(Y2+j, y2);
// ei_pstore(&Y[j+2*PacketSize], ei_padd(ei_pload(&Y[j+2*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+2*PacketSize]))));
// ei_pstore(&Y[j+3*PacketSize], ei_padd(ei_pload(&Y[j+3*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+3*PacketSize]))));
// ei_pstore(&Y[j+4*PacketSize], ei_padd(ei_pload(&Y[j+4*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+4*PacketSize]))));
// ei_pstore(&Y[j+5*PacketSize], ei_padd(ei_pload(&Y[j+5*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+5*PacketSize]))));
// ei_pstore(&Y[j+6*PacketSize], ei_padd(ei_pload(&Y[j+6*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+6*PacketSize]))));
// ei_pstore(&Y[j+7*PacketSize], ei_padd(ei_pload(&Y[j+7*PacketSize]), ei_pmul(pcoef,ei_pload(&X[j+7*PacketSize]))));
internal::pstore(Y+j, y0);
internal::pstore(Y1+j, y1);
internal::pstore(Y2+j, y2);
// internal::pstore(&Y[j+2*PacketSize], internal::padd(internal::pload(&Y[j+2*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+2*PacketSize]))));
// internal::pstore(&Y[j+3*PacketSize], internal::padd(internal::pload(&Y[j+3*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+3*PacketSize]))));
// internal::pstore(&Y[j+4*PacketSize], internal::padd(internal::pload(&Y[j+4*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+4*PacketSize]))));
// internal::pstore(&Y[j+5*PacketSize], internal::padd(internal::pload(&Y[j+5*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+5*PacketSize]))));
// internal::pstore(&Y[j+6*PacketSize], internal::padd(internal::pload(&Y[j+6*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+6*PacketSize]))));
// internal::pstore(&Y[j+7*PacketSize], internal::padd(internal::pload(&Y[j+7*PacketSize]), internal::pmul(pcoef,internal::pload(&X[j+7*PacketSize]))));
}
for (int j = ANP;j<AN;j+=PacketSize)
ei_pstore(&Y[j], ei_padd(ei_pload(&Y[j]), ei_pmul(pcoef,ei_pload(&X[j]))));
internal::pstore(&Y[j], internal::padd(internal::pload(&Y[j]), internal::pmul(pcoef,internal::pload(&X[j]))));
#else
for (int j = 0;j<AN;j+=PacketSize)
ei_pstore(&Y[j], ei_padd(ei_pload(&Y[j]), ei_pmul(pcoef,ei_pload(&X[j]))));
internal::pstore(&Y[j], internal::padd(internal::pload(&Y[j]), internal::pmul(pcoef,internal::pload(&X[j]))));
#endif
}
// process remaining scalars

12
bench/check_cache_queries.cpp
View File

@@ -17,10 +17,10 @@ using namespace std;
int main()
{
cout << "Eigen's L1 = " << ei_queryL1CacheSize() << endl;
cout << "Eigen's L2/L3 = " << ei_queryTopLevelCacheSize() << endl;
cout << "Eigen's L1 = " << internal::queryL1CacheSize() << endl;
cout << "Eigen's L2/L3 = " << internal::queryTopLevelCacheSize() << endl;
int l1, l2, l3;
ei_queryCacheSizes(l1, l2, l3);
internal::queryCacheSizes(l1, l2, l3);
cout << "Eigen's L1, L2, L3 = " << l1 << " " << l2 << " " << l3 << endl;
#ifdef EIGEN_CPUID
@@ -40,14 +40,14 @@ int main()
cout << endl;
int max_funcs = abcd[0];
ei_queryCacheSizes_intel_codes(l1, l2, l3);
internal::queryCacheSizes_intel_codes(l1, l2, l3);
cout << "Eigen's intel codes L1, L2, L3 = " << l1 << " " << l2 << " " << l3 << endl;
if(max_funcs>=4)
{
ei_queryCacheSizes_intel_direct(l1, l2, l3);
internal::queryCacheSizes_intel_direct(l1, l2, l3);
cout << "Eigen's intel direct L1, L2, L3 = " << l1 << " " << l2 << " " << l3 << endl;
}
ei_queryCacheSizes_amd(l1, l2, l3);
internal::queryCacheSizes_amd(l1, l2, l3);
cout << "Eigen's amd L1, L2, L3 = " << l1 << " " << l2 << " " << l3 << endl;
cout << endl;

10
bench/eig33.cpp
View File

@@ -65,7 +65,7 @@ inline void computeRoots(const Matrix& m, Roots& roots)
{
typedef typename Matrix::Scalar Scalar;
const Scalar s_inv3 = 1.0/3.0;
const Scalar s_sqrt3 = ei_sqrt(Scalar(3.0));
const Scalar s_sqrt3 = internal::sqrt(Scalar(3.0));
// The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0. The
// eigenvalues are the roots to this equation, all guaranteed to be
@@ -88,10 +88,10 @@ inline void computeRoots(const Matrix& m, Roots& roots)
q = Scalar(0);
// Compute the eigenvalues by solving for the roots of the polynomial.
Scalar rho = ei_sqrt(-a_over_3);
Scalar theta = std::atan2(ei_sqrt(-q),half_b)*s_inv3;
Scalar cos_theta = ei_cos(theta);
Scalar sin_theta = ei_sin(theta);
Scalar rho = internal::sqrt(-a_over_3);
Scalar theta = std::atan2(internal::sqrt(-q),half_b)*s_inv3;
Scalar cos_theta = internal::cos(theta);
Scalar sin_theta = internal::sin(theta);
roots(0) = c2_over_3 + Scalar(2)*rho*cos_theta;
roots(1) = c2_over_3 - rho*(cos_theta + s_sqrt3*sin_theta);
roots(2) = c2_over_3 - rho*(cos_theta - s_sqrt3*sin_theta);

20
bench/quat_slerp.cpp
View File

@@ -25,16 +25,16 @@ EIGEN_DONT_INLINE Q slerp_legacy(const Q& a, const Q& b, typename Q::Scalar t)
typedef typename Q::Scalar Scalar;
static const Scalar one = Scalar(1) - dummy_precision<Scalar>();
Scalar d = a.dot(b);
Scalar absD = ei_abs(d);
Scalar absD = internal::abs(d);
if (absD>=one)
return a;
// theta is the angle between the 2 quaternions
Scalar theta = std::acos(absD);
Scalar sinTheta = ei_sin(theta);
Scalar sinTheta = internal::sin(theta);
Scalar scale0 = ei_sin( ( Scalar(1) - t ) * theta) / sinTheta;
Scalar scale1 = ei_sin( ( t * theta) ) / sinTheta;
Scalar scale0 = internal::sin( ( Scalar(1) - t ) * theta) / sinTheta;
Scalar scale1 = internal::sin( ( t * theta) ) / sinTheta;
if (d<0)
scale1 = -scale1;
@@ -47,7 +47,7 @@ EIGEN_DONT_INLINE Q slerp_legacy_nlerp(const Q& a, const Q& b, typename Q::Scala
typedef typename Q::Scalar Scalar;
static const Scalar one = Scalar(1) - epsilon<Scalar>();
Scalar d = a.dot(b);
Scalar absD = ei_abs(d);
Scalar absD = internal::abs(d);
Scalar scale0;
Scalar scale1;
@@ -61,10 +61,10 @@ EIGEN_DONT_INLINE Q slerp_legacy_nlerp(const Q& a, const Q& b, typename Q::Scala
{
// theta is the angle between the 2 quaternions
Scalar theta = std::acos(absD);
Scalar sinTheta = ei_sin(theta);
Scalar sinTheta = internal::sin(theta);
scale0 = ei_sin( ( Scalar(1) - t ) * theta) / sinTheta;
scale1 = ei_sin( ( t * theta) ) / sinTheta;
scale0 = internal::sin( ( Scalar(1) - t ) * theta) / sinTheta;
scale1 = internal::sin( ( t * theta) ) / sinTheta;
if (d<0)
scale1 = -scale1;
}
@@ -132,8 +132,8 @@ EIGEN_DONT_INLINE Q slerp_gael(const Q& a, const Q& b, typename Q::Scalar t)
else
{
Scalar sinTheta = std::sin(theta);
scale0 = ei_sin( ( Scalar(1) - t ) * theta) / sinTheta;
scale1 = ei_sin( ( t * theta) ) / sinTheta;
scale0 = internal::sin( ( Scalar(1) - t ) * theta) / sinTheta;
scale1 = internal::sin( ( t * theta) ) / sinTheta;
if (d<0)
scale1 = -scale1;
}

2
bench/quatmul.cpp
View File

@@ -14,7 +14,7 @@ EIGEN_DONT_INLINE void quatmul_default(const Quat& a, const Quat& b, Quat& c)
template<typename Quat>
EIGEN_DONT_INLINE void quatmul_novec(const Quat& a, const Quat& b, Quat& c)
{
c = ei_quat_product<0, Quat, Quat, typename Quat::Scalar, Aligned>::run(a,b);
c = internal::quat_product<0, Quat, Quat, typename Quat::Scalar, Aligned>::run(a,b);
}
template<typename Quat> void bench(const std::string& label)

6
bench/sparse_cholesky.cpp
View File

@@ -46,10 +46,10 @@ void fillSpdMatrix(float density, int rows, int cols, EigenSparseSelfAdjointMat
dst.startFill(rows*cols*density);
for(int j = 0; j < cols; j++)
{
dst.fill(j,j) = ei_random<Scalar>(10,20);
dst.fill(j,j) = internal::random<Scalar>(10,20);
for(int i = j+1; i < rows; i++)
{
Scalar v = (ei_random<float>(0,1) < density) ? ei_random<Scalar>() : 0;
Scalar v = (internal::random<float>(0,1) < density) ? internal::random<Scalar>() : 0;
if (v!=0)
dst.fill(i,j) = v;
}
@@ -116,7 +116,7 @@ int main(int argc, char *argv[])
int count = 0;
for (int j=0; j<cols; ++j)
for (int i=j; i<rows; ++i)
if (!ei_isMuchSmallerThan(ei_abs(chol.matrixL()(i,j)), 0.1))
if (!internal::isMuchSmallerThan(internal::abs(chol.matrixL()(i,j)), 0.1))
count++;
std::cout << "dense: " << "nnz = " << count << "\n";
// std::cout << "dense:\n" << m1 << "\n\n" << chol.matrixL() << endl;

10
bench/sparse_randomsetter.cpp
View File

@@ -51,7 +51,7 @@ void dostuff(const char* name, EigenSparseMatrix& sm1)
SetterType* set1 = new SetterType(sm1);
t.reset(); t.start();
for (int k=0; k<nentries; ++k)
(*set1)(ei_random<int>(0,rows-1),ei_random<int>(0,cols-1)) += 1;
(*set1)(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1)) += 1;
t.stop();
std::cout << "std::map => \t" << t.value()-rtime
<< " nnz=" << set1->nonZeros() << std::flush;
@@ -78,7 +78,7 @@ int main(int argc, char *argv[])
t.reset(); t.start();
for (int k=0; k<nentries; ++k)
dummy = ei_random<int>(0,rows-1) + ei_random<int>(0,cols-1);
dummy = internal::random<int>(0,rows-1) + internal::random<int>(0,cols-1);
t.stop();
rtime = t.value();
std::cout << "rtime = " << rtime << " (" << dummy << ")\n\n";
@@ -94,7 +94,7 @@ int main(int argc, char *argv[])
// RandomSetter<EigenSparseMatrix,GnuHashMapTraits,Bits> set1(sm1);
// t.reset(); t.start();
// for (int k=0; k<n; ++k)
// set1(ei_random<int>(0,rows-1),ei_random<int>(0,cols-1)) += 1;
// set1(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1)) += 1;
// t.stop();
// std::cout << "gnu::hash_map => \t" << t.value()-rtime
// << " nnz=" << set1.nonZeros() << "\n";getchar();
@@ -103,7 +103,7 @@ int main(int argc, char *argv[])
// RandomSetter<EigenSparseMatrix,GoogleDenseHashMapTraits,Bits> set1(sm1);
// t.reset(); t.start();
// for (int k=0; k<n; ++k)
// set1(ei_random<int>(0,rows-1),ei_random<int>(0,cols-1)) += 1;
// set1(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1)) += 1;
// t.stop();
// std::cout << "google::dense => \t" << t.value()-rtime
// << " nnz=" << set1.nonZeros() << "\n";getchar();
@@ -112,7 +112,7 @@ int main(int argc, char *argv[])
// RandomSetter<EigenSparseMatrix,GoogleSparseHashMapTraits,Bits> set1(sm1);
// t.reset(); t.start();
// for (int k=0; k<n; ++k)
// set1(ei_random<int>(0,rows-1),ei_random<int>(0,cols-1)) += 1;
// set1(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1)) += 1;
// t.stop();
// std::cout << "google::sparse => \t" << t.value()-rtime
// << " nnz=" << set1.nonZeros() << "\n";getchar();

10
bench/sparse_setter.cpp
View File

@@ -75,7 +75,7 @@ int main(int argc, char *argv[])
for (int i=0; i<cols*NBPERROW; )
{
// DynamicSparseMatrix<int> stencil(SIZE,SIZE);
Vector2i ij(ei_random<int>(0,rows-1),ei_random<int>(0,cols-1));
Vector2i ij(internal::random<int>(0,rows-1),internal::random<int>(0,cols-1));
// if(stencil.coeffRef(ij.x(), ij.y())==0)
{
// stencil.coeffRef(ij.x(), ij.y()) = 1;
@@ -90,9 +90,9 @@ int main(int argc, char *argv[])
values.reserve(n);
for (int i=0; i<n; ++i)
{
int i = ei_random<int>(0,pool.size());
int i = internal::random<int>(0,pool.size());
coords.push_back(pool[i]);
values.push_back(ei_random<Scalar>());
values.push_back(internal::random<Scalar>());
}
}
else
@@ -100,8 +100,8 @@ int main(int argc, char *argv[])
for (int j=0; j<cols; ++j)
for (int i=0; i<NBPERROW; ++i)
{
coords.push_back(Vector2i(ei_random<int>(0,rows-1),j));
values.push_back(ei_random<Scalar>());
coords.push_back(Vector2i(internal::random<int>(0,rows-1),j));
values.push_back(internal::random<Scalar>());
}
}
std::cout << "nnz = " << coords.size() << "\n";

4
bench/sparse_trisolver.cpp
View File

@@ -44,11 +44,11 @@ void fillMatrix(float density, int rows, int cols, EigenSparseTriMatrix& dst)
{
for(int i = 0; i < j; i++)
{
Scalar v = (ei_random<float>(0,1) < density) ? ei_random<Scalar>() : 0;
Scalar v = (internal::random<float>(0,1) < density) ? internal::random<Scalar>() : 0;
if (v!=0)
dst.fill(i,j) = v;
}
dst.fill(j,j) = ei_random<Scalar>();
dst.fill(j,j) = internal::random<Scalar>();
}
dst.endFill();
}