Fix pfrexp/pldexp for half.

The recent addition of vectorized pow (!330) relies on `pfrexp` and
`pldexp`.  This was missing for `Eigen::half` and `Eigen::bfloat16`.
Adding tests for these packet ops also exposed an issue with handling
negative values in `pfrexp`, returning an incorrect exponent.

Added the missing implementations, corrected the exponent in `pfrexp1`,
and added `packetmath` tests.
This commit is contained in:
Antonio Sanchez 2021-01-20 19:00:09 -08:00 committed by Rasmus Munk Larsen
parent 25d8498f8b
commit b2126fd6b5
7 changed files with 112 additions and 4 deletions

View File

@ -442,7 +442,7 @@ template <typename Packet>
EIGEN_DEVICE_FUNC inline Packet pfrexp(const Packet& a, Packet& exponent) {
int exp;
EIGEN_USING_STD(frexp);
Packet result = frexp(a, &exp);
Packet result = static_cast<Packet>(frexp(a, &exp));
exponent = static_cast<Packet>(exp);
return result;
}
@ -453,7 +453,7 @@ EIGEN_DEVICE_FUNC inline Packet pfrexp(const Packet& a, Packet& exponent) {
template<typename Packet> EIGEN_DEVICE_FUNC inline Packet
pldexp(const Packet &a, const Packet &exponent) {
EIGEN_USING_STD(ldexp)
return ldexp(a, static_cast<int>(exponent));
return static_cast<Packet>(ldexp(a, static_cast<int>(exponent)));
}
/** \internal \returns the min of \a a and \a b (coeff-wise) */

View File

@ -184,6 +184,19 @@ F16_PACKET_FUNCTION(Packet8f, Packet8h, ptanh)
F16_PACKET_FUNCTION(Packet8f, Packet8h, psqrt)
F16_PACKET_FUNCTION(Packet8f, Packet8h, prsqrt)
template <>
EIGEN_STRONG_INLINE Packet8h pfrexp(const Packet8h& a, Packet8h& exponent) {
Packet8f fexponent;
const Packet8h out = float2half(pfrexp<Packet8f>(half2float(a), fexponent));
exponent = float2half(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet8h pldexp(const Packet8h& a, const Packet8h& exponent) {
return float2half(pldexp<Packet8f>(half2float(a), half2float(exponent)));
}
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psin)
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, pcos)
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, plog)
@ -195,6 +208,19 @@ BF16_PACKET_FUNCTION(Packet8f, Packet8bf, ptanh)
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, psqrt)
BF16_PACKET_FUNCTION(Packet8f, Packet8bf, prsqrt)
template <>
EIGEN_STRONG_INLINE Packet8bf pfrexp(const Packet8bf& a, Packet8bf& exponent) {
Packet8f fexponent;
const Packet8bf out = F32ToBf16(pfrexp<Packet8f>(Bf16ToF32(a), fexponent));
exponent = F32ToBf16(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet8bf pldexp(const Packet8bf& a, const Packet8bf& exponent) {
return F32ToBf16(pldexp<Packet8f>(Bf16ToF32(a), Bf16ToF32(exponent)));
}
} // end namespace internal
} // end namespace Eigen

View File

@ -191,6 +191,32 @@ pexp<Packet8d>(const Packet8d& _x) {
F16_PACKET_FUNCTION(Packet16f, Packet16h, pexp)
BF16_PACKET_FUNCTION(Packet16f, Packet16bf, pexp)
template <>
EIGEN_STRONG_INLINE Packet16h pfrexp(const Packet16h& a, Packet16h& exponent) {
Packet16f fexponent;
const Packet16h out = float2half(pfrexp<Packet16f>(half2float(a), fexponent));
exponent = float2half(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet16h pldexp(const Packet16h& a, const Packet16h& exponent) {
return float2half(pldexp<Packet16f>(half2float(a), half2float(exponent)));
}
template <>
EIGEN_STRONG_INLINE Packet16bf pfrexp(const Packet16bf& a, Packet16bf& exponent) {
Packet16f fexponent;
const Packet16bf out = F32ToBf16(pfrexp<Packet16f>(Bf16ToF32(a), fexponent));
exponent = F32ToBf16(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet16bf pldexp(const Packet16bf& a, const Packet16bf& exponent) {
return F32ToBf16(pldexp<Packet16f>(Bf16ToF32(a), Bf16ToF32(exponent)));
}
// Functions for sqrt.
// The EIGEN_FAST_MATH version uses the _mm_rsqrt_ps approximation and one step
// of Newton's method, at a cost of 1-2 bits of precision as opposed to the

View File

@ -31,7 +31,7 @@ pfrexp_float(const Packet& a, Packet& exponent) {
const Packet cst_126f = pset1<Packet>(126.0f);
const Packet cst_half = pset1<Packet>(0.5f);
const Packet cst_inv_mant_mask = pset1frombits<Packet>(~0x7f800000u);
exponent = psub(pcast<PacketI,Packet>(plogical_shift_right<23>(preinterpret<PacketI>(a))), cst_126f);
exponent = psub(pcast<PacketI,Packet>(plogical_shift_right<23>(preinterpret<PacketI>(pabs<Packet>(a)))), cst_126f);
return por(pand(a, cst_inv_mant_mask), cst_half);
}
@ -41,7 +41,7 @@ pfrexp_double(const Packet& a, Packet& exponent) {
const Packet cst_1022d = pset1<Packet>(1022.0);
const Packet cst_half = pset1<Packet>(0.5);
const Packet cst_inv_mant_mask = pset1frombits<Packet>(static_cast<uint64_t>(~0x7ff0000000000000ull));
exponent = psub(pcast<PacketI,Packet>(plogical_shift_right<52>(preinterpret<PacketI>(a))), cst_1022d);
exponent = psub(pcast<PacketI,Packet>(plogical_shift_right<52>(preinterpret<PacketI>(pabs<Packet>(a)))), cst_1022d);
return por(pand(a, cst_inv_mant_mask), cst_half);
}

View File

@ -44,6 +44,18 @@ BF16_PACKET_FUNCTION(Packet4f, Packet4bf, plog)
BF16_PACKET_FUNCTION(Packet4f, Packet4bf, pexp)
BF16_PACKET_FUNCTION(Packet4f, Packet4bf, ptanh)
template <>
EIGEN_STRONG_INLINE Packet4bf pfrexp(const Packet4bf& a, Packet4bf& exponent) {
Packet4f fexponent;
const Packet4bf out = F32ToBf16(pfrexp<Packet4f>(Bf16ToF32(a), fexponent));
exponent = F32ToBf16(fexponent);
return out;
}
template <>
EIGEN_STRONG_INLINE Packet4bf pldexp(const Packet4bf& a, const Packet4bf& exponent) {
return F32ToBf16(pldexp<Packet4f>(Bf16ToF32(a), Bf16ToF32(exponent)));
}
//---------- double ----------

View File

@ -46,6 +46,21 @@ inline bool REF_MUL(const bool& a, const bool& b) {
return a && b;
}
template <typename T>
inline T REF_FREXP(const T& x, T& exp) {
int iexp;
EIGEN_USING_STD(frexp)
const T out = static_cast<T>(frexp(x, &iexp));
exp = static_cast<T>(iexp);
return out;
}
template <typename T>
inline T REF_LDEXP(const T& x, const T& exp) {
EIGEN_USING_STD(ldexp)
return static_cast<T>(ldexp(x, static_cast<int>(exp)));
}
// Uses pcast to cast from one array to another.
template <typename SrcPacket, typename TgtPacket, int SrcCoeffRatio, int TgtCoeffRatio>
struct pcast_array;
@ -552,6 +567,17 @@ void packetmath_real() {
data2[i] = Scalar(internal::random<double>(-87, 88));
}
CHECK_CWISE1_IF(PacketTraits::HasExp, std::exp, internal::pexp);
CHECK_CWISE1_BYREF1_IF(PacketTraits::HasExp, REF_FREXP, internal::pfrexp);
for (int i = 0; i < PacketSize; ++i) {
data1[i] = Scalar(internal::random<double>(-1, 1));
data2[i] = Scalar(internal::random<double>(-1, 1));
}
for (int i = 0; i < PacketSize; ++i) {
data1[i+PacketSize] = Scalar(internal::random<int>(0, 4));
data2[i+PacketSize] = Scalar(internal::random<double>(0, 4));
}
CHECK_CWISE2_IF(PacketTraits::HasExp, REF_LDEXP, internal::pldexp);
for (int i = 0; i < size; ++i) {
data1[i] = Scalar(internal::random<double>(-1, 1) * std::pow(10., internal::random<double>(-6, 6)));
data2[i] = Scalar(internal::random<double>(-1, 1) * std::pow(10., internal::random<double>(-6, 6)));

View File

@ -143,6 +143,9 @@ struct packet_helper
template<typename T>
inline void store(T* to, const Packet& x, unsigned long long umask) const { internal::pstoreu(to, x, umask); }
template<typename T>
inline Packet& forward_reference(Packet& packet, T& /*scalar*/) const { return packet; }
};
template<typename Packet>
@ -162,6 +165,9 @@ struct packet_helper<false,Packet>
template<typename T>
inline void store(T* to, const T& x, unsigned long long) const { *to = x; }
template<typename T>
inline T& forward_reference(Packet& /*packet*/, T& scalar) const { return scalar; }
};
#define CHECK_CWISE1_IF(COND, REFOP, POP) if(COND) { \
@ -180,6 +186,18 @@ struct packet_helper<false,Packet>
VERIFY(test::areApprox(ref, data2, PacketSize) && #POP); \
}
// One input, one output by reference.
#define CHECK_CWISE1_BYREF1_IF(COND, REFOP, POP) if(COND) { \
test::packet_helper<COND,Packet> h; \
for (int i=0; i<PacketSize; ++i) \
ref[i] = Scalar(REFOP(data1[i], ref[i+PacketSize])); \
Packet pout; \
Scalar sout; \
h.store(data2, POP(h.load(data1), h.forward_reference(pout, sout))); \
h.store(data2+PacketSize, h.forward_reference(pout, sout)); \
VERIFY(test::areApprox(ref, data2, 2 * PacketSize) && #POP); \
}
#define CHECK_CWISE3_IF(COND, REFOP, POP) if (COND) { \
test::packet_helper<COND, Packet> h; \
for (int i = 0; i < PacketSize; ++i) \