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

Add missing packet ops for bool, and make it pass the same packet op unit tests as other arithmetic types.

Browse Source 
This change also contains a few minor cleanups:
  1. Remove packet op pnot, which is not needed for anything other than pcmp_le_or_nan,
     which can be done in other ways.
  2. Remove the "HasInsert" enum, which is no longer needed since we removed the
     corresponding packet ops.
  3. Add faster pselect op for Packet4i when SSE4.1 is supported.

Among other things, this makes the fast transposeInPlace() method available for Matrix<bool>.

Run on ************** (72 X 2994 MHz CPUs); 2020-05-09T10:51:02.372347913-07:00
CPU: Intel Skylake Xeon with HyperThreading (36 cores) dL1:32KB dL2:1024KB dL3:24MB
Benchmark                        Time(ns)        CPU(ns)     Iterations
-----------------------------------------------------------------------
BM_TransposeInPlace<float>/4            9.77           9.77    71670320
BM_TransposeInPlace<float>/8           21.9           21.9     31929525
BM_TransposeInPlace<float>/16          66.6           66.6     10000000
BM_TransposeInPlace<float>/32         243            243        2879561
BM_TransposeInPlace<float>/59         844            844         829767
BM_TransposeInPlace<float>/64         933            933         750567
BM_TransposeInPlace<float>/128       3944           3945         177405
BM_TransposeInPlace<float>/256      16853          16853          41457
BM_TransposeInPlace<float>/512     204952         204968           3448
BM_TransposeInPlace<float>/1k     1053889        1053861            664
BM_TransposeInPlace<bool>/4            14.4           14.4     48637301
BM_TransposeInPlace<bool>/8            36.0           36.0     19370222
BM_TransposeInPlace<bool>/16           31.5           31.5     22178902
BM_TransposeInPlace<bool>/32          111            111        6272048
BM_TransposeInPlace<bool>/59          626            626        1000000
BM_TransposeInPlace<bool>/64          428            428        1632689
BM_TransposeInPlace<bool>/128        1677           1677         417377
BM_TransposeInPlace<bool>/256        7126           7126          96264
BM_TransposeInPlace<bool>/512       29021          29024          24165
BM_TransposeInPlace<bool>/1k       116321         116330           6068
This commit is contained in:
Rasmus Munk Larsen
2020-05-11 13:23:31 -07:00
parent d640276d31
commit 9b411757ab
9 changed files with 247 additions and 156 deletions
Download Patch File Download Diff File Expand all files Collapse all files

75
Eigen/src/Core/GenericPacketMath.h
View File

@@ -58,7 +58,6 @@ struct default_packet_traits
HasConj = 1,
HasSetLinear = 1,
HasBlend = 0,
HasInsert = 0,
HasDiv = 0,
HasSqrt = 0,
@@ -191,8 +190,10 @@ psub(const Packet& a, const Packet& b) { return a-b; }
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pnegate(const Packet& a) { return -a; }
/** \internal \returns conj(a) (coeff-wise) */
template<> EIGEN_DEVICE_FUNC inline bool
pnegate(const bool& a) { return !a; }
/** \internal \returns conj(a) (coeff-wise) */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pconj(const Packet& a) { return numext::conj(a); }
@@ -269,14 +270,30 @@ pldexp(const Packet &a, const Packet &exponent) {
return ldexp(a, static_cast<int>(exponent));
}
// Notice: The following ops accept and operator on bitwise masks.
// The value of each field in a masks is Scalar(0) or ~Scalar(0).
// For boolean packet like Packet16b, this is different from the
// representation of true and false, which are 1 and 0.
// As an example
// ptrue<Packet16b>() = 0xffffffffffffffffffffffffffffffff
// while
// pset1<Packet16b>(true) = 0x01010101010101010101010101010101
/** \internal \returns zero bits */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pzero(const Packet& /*a*/) { Packet b; memset((void*)&b, 0, sizeof(b)); return b;}
template<> EIGEN_DEVICE_FUNC inline float pzero<float>(const float& a) {
EIGEN_UNUSED_VARIABLE(a);
return 0.f;
}
template<> EIGEN_DEVICE_FUNC inline double pzero<double>(const double& a) {
EIGEN_UNUSED_VARIABLE(a);
return 0.;
}
/** \internal \returns one bits */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
ptrue(const Packet& /*a*/) { Packet b; memset((void*)&b, 0xff, sizeof(b)); return b;}
template <typename RealScalar>
EIGEN_DEVICE_FUNC inline std::complex<RealScalar> ptrue(const std::complex<RealScalar>& /*a*/) {
RealScalar b;
b = ptrue(b);
return std::complex<RealScalar>(b, b);
}
/** \internal \returns the bitwise and of \a a and \a b */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -292,36 +309,7 @@ pxor(const Packet& a, const Packet& b) { return a ^ b; }
/** \internal \returns the bitwise and of \a a and not \a b */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pandnot(const Packet& a, const Packet& b) { return a & (~b); }
/** \internal \returns ones */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
ptrue(const Packet& /*a*/) { Packet b; memset((void*)&b, 0xff, sizeof(b)); return b;}
/** \internal \returns zeros */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pzero(const Packet& a) { return pxor(a,a); }
template<> EIGEN_DEVICE_FUNC inline float pzero<float>(const float& a) {
EIGEN_UNUSED_VARIABLE(a);
return 0.f;
}
template<> EIGEN_DEVICE_FUNC inline double pzero<double>(const double& a) {
EIGEN_UNUSED_VARIABLE(a);
return 0.;
}
template <typename RealScalar>
EIGEN_DEVICE_FUNC inline std::complex<RealScalar> ptrue(const std::complex<RealScalar>& /*a*/) {
RealScalar b;
b = ptrue(b);
return std::complex<RealScalar>(b, b);
}
/** \internal \returns the bitwise not of \a a */
template <typename Packet> EIGEN_DEVICE_FUNC inline Packet
pnot(const Packet& a) { return pxor(ptrue(a), a);}
pandnot(const Packet& a, const Packet& b) { return pand(a, pxor(ptrue(b), b)); }
/** \internal \returns a <= b as a bit mask */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -337,7 +325,7 @@ pcmp_eq(const Packet& a, const Packet& b) { return a==b ? ptrue(a) : pzero(a); }
/** \internal \returns a < b or a==NaN or b==NaN as a bit mask */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pcmp_lt_or_nan(const Packet& a, const Packet& b) { return pnot(pcmp_le(b,a)); }
pcmp_lt_or_nan(const Packet& a, const Packet& b) { return a>=b ? pzero(a) : ptrue(a); }
/** \internal \returns \a or \b for each field in packet according to \mask */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
@@ -355,7 +343,10 @@ template<> EIGEN_DEVICE_FUNC inline double pselect<double>(
return numext::equal_strict(cond,0.) ? b : a;
}
template<> EIGEN_DEVICE_FUNC inline bool pselect<bool>(
const bool& cond, const bool& a, const bool& b) {
return cond ? a : b;
}
/** \internal \returns the min of \a a and \a b (coeff-wise) */
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet