template<typename Scalar> void packetmath()
{
typedef typename internal::packet_traits<Scalar>::type Packet;
const int PacketSize = internal::packet_traits<Scalar>::size;
typedef typename NumTraits<Scalar>::Real RealScalar;
const int size = PacketSize*4;
EIGEN_ALIGN16 Scalar data1[internal::packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Scalar data2[internal::packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Packet packets[PacketSize*2];
EIGEN_ALIGN16 Scalar ref[internal::packet_traits<Scalar>::size*4];
RealScalar refvalue = 0;
for (int i=0; i<size; ++i)
{
data1[i] = internal::random<Scalar>()/RealScalar(PacketSize);
data2[i] = internal::random<Scalar>()/RealScalar(PacketSize);
refvalue = (std::max)(refvalue,internal::abs(data1[i]));
}
internal::pstore(data2, internal::pload<Packet>(data1));
VERIFY(areApprox(data1, data2, PacketSize) && "aligned load/store");
for (int offset=0; offset<PacketSize; ++offset)
{
internal::pstore(data2, internal::ploadu<Packet>(data1+offset));
VERIFY(areApprox(data1+offset, data2, PacketSize) && "internal::ploadu");
}
for (int offset=0; offset<PacketSize; ++offset)
{
internal::pstoreu(data2+offset, internal::pload<Packet>(data1));
VERIFY(areApprox(data1, data2+offset, PacketSize) && "internal::pstoreu");
}
for (int offset=0; offset<PacketSize; ++offset)
{
packets[0] = internal::pload<Packet>(data1);
packets[1] = internal::pload<Packet>(data1+PacketSize);
if (offset==0) internal::palign<0>(packets[0], packets[1]);
else if (offset==1) internal::palign<1>(packets[0], packets[1]);
else if (offset==2) internal::palign<2>(packets[0], packets[1]);
else if (offset==3) internal::palign<3>(packets[0], packets[1]);
internal::pstore(data2, packets[0]);
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[i+offset];
typedef Matrix<Scalar, PacketSize, 1> Vector;
VERIFY(areApprox(ref, data2, PacketSize) && "internal::palign");
}
CHECK_CWISE2(REF_ADD, internal::padd);
CHECK_CWISE2(REF_SUB, internal::psub);
CHECK_CWISE2(REF_MUL, internal::pmul);
#ifndef EIGEN_VECTORIZE_ALTIVEC
if (!internal::is_same<Scalar,int>::value)
CHECK_CWISE2(REF_DIV, internal::pdiv);
#endif
CHECK_CWISE1(internal::negate, internal::pnegate);
CHECK_CWISE1(internal::conj, internal::pconj);
for(int offset=0;offset<3;++offset)
{
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[offset];
internal::pstore(data2, internal::pset1<Packet>(data1[offset]));
VERIFY(areApprox(ref, data2, PacketSize) && "internal::pset1");
}
VERIFY(internal::isApprox(data1[0], internal::pfirst(internal::pload<Packet>(data1))) && "internal::pfirst");
if(PacketSize>1)
{
for(int offset=0;offset<4;++offset)
{
for(int i=0;i<PacketSize/2;++i)
ref[2*i+0] = ref[2*i+1] = data1[offset+i];
internal::pstore(data2,internal::ploaddup<Packet>(data1+offset));
VERIFY(areApprox(ref, data2, PacketSize) && "ploaddup");
}
}
ref[0] = 0;
for (int i=0; i<PacketSize; ++i)
ref[0] += data1[i];
VERIFY(isApproxAbs(ref[0], internal::predux(internal::pload<Packet>(data1)), refvalue) && "internal::predux");
ref[0] = 1;
for (int i=0; i<PacketSize; ++i)
ref[0] *= data1[i];
VERIFY(internal::isApprox(ref[0], internal::predux_mul(internal::pload<Packet>(data1))) && "internal::predux_mul");
for (int j=0; j<PacketSize; ++j)
{
ref[j] = 0;
for (int i=0; i<PacketSize; ++i)
ref[j] += data1[i+j*PacketSize];
packets[j] = internal::pload<Packet>(data1+j*PacketSize);
}
internal::pstore(data2, internal::preduxp(packets));
VERIFY(areApproxAbs(ref, data2, PacketSize, refvalue) && "internal::preduxp");
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[PacketSize-i-1];
internal::pstore(data2, internal::preverse(internal::pload<Packet>(data1)));
VERIFY(areApprox(ref, data2, PacketSize) && "internal::preverse");
}
template<typename Scalar> void packetmath()
{
typedef typename ei_packet_traits<Scalar>::type Packet;
const int PacketSize = ei_packet_traits<Scalar>::size;
typedef typename NumTraits<Scalar>::Real RealScalar;
const int size = PacketSize*4;
EIGEN_ALIGN16 Scalar data1[ei_packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Scalar data2[ei_packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Packet packets[PacketSize*2];
EIGEN_ALIGN16 Scalar ref[ei_packet_traits<Scalar>::size*4];
RealScalar refvalue = 0;
for (int i=0; i<size; ++i)
{
data1[i] = ei_random<Scalar>();
data2[i] = ei_random<Scalar>();
refvalue = std::max(refvalue,ei_abs(data1[i]));
}
ei_pstore(data2, ei_pload(data1));
VERIFY(areApprox(data1, data2, PacketSize) && "aligned load/store");
for (int offset=0; offset<PacketSize; ++offset)
{
ei_pstore(data2, ei_ploadu(data1+offset));
VERIFY(areApprox(data1+offset, data2, PacketSize) && "ei_ploadu");
}
for (int offset=0; offset<PacketSize; ++offset)
{
ei_pstoreu(data2+offset, ei_pload(data1));
VERIFY(areApprox(data1, data2+offset, PacketSize) && "ei_pstoreu");
}
for (int offset=0; offset<PacketSize; ++offset)
{
packets[0] = ei_pload(data1);
packets[1] = ei_pload(data1+PacketSize);
if (offset==0) ei_palign<0>(packets[0], packets[1]);
else if (offset==1) ei_palign<1>(packets[0], packets[1]);
else if (offset==2) ei_palign<2>(packets[0], packets[1]);
else if (offset==3) ei_palign<3>(packets[0], packets[1]);
ei_pstore(data2, packets[0]);
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[i+offset];
typedef Matrix<Scalar, PacketSize, 1> Vector;
VERIFY(areApprox(ref, data2, PacketSize) && "ei_palign");
}
CHECK_CWISE2(REF_ADD, ei_padd);
CHECK_CWISE2(REF_SUB, ei_psub);
CHECK_CWISE2(REF_MUL, ei_pmul);
#ifndef EIGEN_VECTORIZE_ALTIVEC
if (!ei_is_same_type<Scalar,int>::ret)
CHECK_CWISE2(REF_DIV, ei_pdiv);
#endif
CHECK_CWISE2(std::min, ei_pmin);
CHECK_CWISE2(std::max, ei_pmax);
CHECK_CWISE1(ei_abs, ei_pabs);
CHECK_CWISE1(ei_negate, ei_pnegate);
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[0];
ei_pstore(data2, ei_pset1(data1[0]));
VERIFY(areApprox(ref, data2, PacketSize) && "ei_pset1");
VERIFY(ei_isApprox(data1[0], ei_pfirst(ei_pload(data1))) && "ei_pfirst");
ref[0] = 0;
for (int i=0; i<PacketSize; ++i)
ref[0] += data1[i];
VERIFY(isApproxAbs(ref[0], ei_predux(ei_pload(data1)), refvalue) && "ei_predux");
ref[0] = 1;
for (int i=0; i<PacketSize; ++i)
ref[0] *= data1[i];
VERIFY(ei_isApprox(ref[0], ei_predux_mul(ei_pload(data1))) && "ei_predux_mul");
ref[0] = data1[0];
for (int i=0; i<PacketSize; ++i)
ref[0] = std::min(ref[0],data1[i]);
VERIFY(ei_isApprox(ref[0], ei_predux_min(ei_pload(data1))) && "ei_predux_min");
ref[0] = data1[0];
for (int i=0; i<PacketSize; ++i)
ref[0] = std::max(ref[0],data1[i]);
VERIFY(ei_isApprox(ref[0], ei_predux_max(ei_pload(data1))) && "ei_predux_max");
for (int j=0; j<PacketSize; ++j)
{
ref[j] = 0;
for (int i=0; i<PacketSize; ++i)
ref[j] += data1[i+j*PacketSize];
packets[j] = ei_pload(data1+j*PacketSize);
}
ei_pstore(data2, ei_preduxp(packets));
VERIFY(areApproxAbs(ref, data2, PacketSize, refvalue) && "ei_preduxp");
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[PacketSize-i-1];
ei_pstore(data2, ei_preverse(ei_pload(data1)));
VERIFY(areApprox(ref, data2, PacketSize) && "ei_preverse");
}
template<typename Scalar> void packetmath_real()
{
typedef typename internal::packet_traits<Scalar>::type Packet;
const int PacketSize = internal::packet_traits<Scalar>::size;
const int size = PacketSize*4;
EIGEN_ALIGN16 Scalar data1[internal::packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Scalar data2[internal::packet_traits<Scalar>::size*4];
EIGEN_ALIGN16 Scalar ref[internal::packet_traits<Scalar>::size*4];
for (int i=0; i<size; ++i)
{
data1[i] = internal::random<Scalar>(-1e3,1e3);
data2[i] = internal::random<Scalar>(-1e3,1e3);
}
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasSin, internal::sin, internal::psin);
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasCos, internal::cos, internal::pcos);
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasTan, internal::tan, internal::ptan);
for (int i=0; i<size; ++i)
{
data1[i] = internal::random<Scalar>(-1,1);
data2[i] = internal::random<Scalar>(-1,1);
}
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasASin, internal::asin, internal::pasin);
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasACos, internal::acos, internal::pacos);
for (int i=0; i<size; ++i)
{
data1[i] = internal::random<Scalar>(-87,88);
data2[i] = internal::random<Scalar>(-87,88);
}
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasExp, internal::exp, internal::pexp);
for (int i=0; i<size; ++i)
{
data1[i] = internal::random<Scalar>(0,1e6);
data2[i] = internal::random<Scalar>(0,1e6);
}
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasLog, internal::log, internal::plog);
CHECK_CWISE1_IF(internal::packet_traits<Scalar>::HasSqrt, internal::sqrt, internal::psqrt);
ref[0] = data1[0];
for (int i=0; i<PacketSize; ++i)
ref[0] = (std::min)(ref[0],data1[i]);
VERIFY(internal::isApprox(ref[0], internal::predux_min(internal::pload<Packet>(data1))) && "internal::predux_min");
CHECK_CWISE2((std::min), internal::pmin);
CHECK_CWISE2((std::max), internal::pmax);
CHECK_CWISE1(internal::abs, internal::pabs);
ref[0] = data1[0];
for (int i=0; i<PacketSize; ++i)
ref[0] = (std::max)(ref[0],data1[i]);
VERIFY(internal::isApprox(ref[0], internal::predux_max(internal::pload<Packet>(data1))) && "internal::predux_max");
for (int i=0; i<PacketSize; ++i)
ref[i] = data1[0]+Scalar(i);
internal::pstore(data2, internal::plset(data1[0]));
VERIFY(areApprox(ref, data2, PacketSize) && "internal::plset");
}