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

AVX512 Optimizations for Triangular Solve

Browse Source
This commit is contained in:
b-shi
2022-03-16 18:04:50 +00:00
committed by Rasmus Munk Larsen
parent 01b5bc48cc
commit 518fc321cb
7 changed files with 2664 additions and 128 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
Eigen/src/Core/GenericPacketMath.h
View File

@@ -220,6 +220,15 @@ padd(const Packet& a, const Packet& b) { return a+b; }
template<> EIGEN_DEVICE_FUNC inline bool
padd(const bool& a, const bool& b) { return a || b; }
/** \internal \returns a packet version of \a *from, (un-aligned masked add)
* There is no generic implementation. We only have implementations for specialized
* cases. Generic case should not be called.
*/
template<typename Packet> EIGEN_DEVICE_FUNC inline
std::enable_if_t<unpacket_traits<Packet>::masked_fpops_available, Packet>
padd(const Packet& a, const Packet& b, typename unpacket_traits<Packet>::mask_t umask);
/** \internal \returns a - b (coeff-wise) */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
psub(const Packet& a, const Packet& b) { return a-b; }
@@ -359,16 +368,16 @@ struct bytewise_bitwise_helper {
EIGEN_DEVICE_FUNC static inline T bitwise_and(const T& a, const T& b) {
return binary(a, b, bit_and<unsigned char>());
}
EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) {
EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) {
return binary(a, b, bit_or<unsigned char>());
}
EIGEN_DEVICE_FUNC static inline T bitwise_xor(const T& a, const T& b) {
return binary(a, b, bit_xor<unsigned char>());
}
EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) {
EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) {
return unary(a,bit_not<unsigned char>());
}
private:
template<typename Op>
EIGEN_DEVICE_FUNC static inline T unary(const T& a, Op op) {
@@ -810,7 +819,7 @@ Packet plog10(const Packet& a) { EIGEN_USING_STD(log10); return log10(a); }
template<typename Packet> EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS
Packet plog2(const Packet& a) {
typedef typename internal::unpacket_traits<Packet>::type Scalar;
return pmul(pset1<Packet>(Scalar(EIGEN_LOG2E)), plog(a));
return pmul(pset1<Packet>(Scalar(EIGEN_LOG2E)), plog(a));
}
/** \internal \returns the square-root of \a a (coeff-wise) */
@@ -877,7 +886,7 @@ predux(const Packet& a)
template <typename Packet>
EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(
const Packet& a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename unpacket_traits<Packet>::type Scalar;
return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmul<Scalar>)));
}
@@ -885,14 +894,14 @@ EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_mul(
template <typename Packet>
EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(
const Packet &a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename unpacket_traits<Packet>::type Scalar;
return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin<PropagateFast, Scalar>)));
}
template <int NaNPropagation, typename Packet>
EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(
const Packet& a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename unpacket_traits<Packet>::type Scalar;
return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin<NaNPropagation, Scalar>)));
}
@@ -900,14 +909,14 @@ EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_min(
template <typename Packet>
EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(
const Packet &a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename unpacket_traits<Packet>::type Scalar;
return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax<PropagateFast, Scalar>)));
}
template <int NaNPropagation, typename Packet>
EIGEN_DEVICE_FUNC inline typename unpacket_traits<Packet>::type predux_max(
const Packet& a) {
typedef typename unpacket_traits<Packet>::type Scalar;
typedef typename unpacket_traits<Packet>::type Scalar;
return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax<NaNPropagation, Scalar>)));
}