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 numext::fma and missing pmadd implementations.

Browse Source
This commit is contained in:
Antonio Sánchez
2025-03-23 01:05:53 +00:00
parent 754bd24f5e
commit d935916ac6
11 changed files with 319 additions and 105 deletions
Download Patch File Download Diff File Expand all files Collapse all files

87
Eigen/src/Core/arch/AltiVec/PacketMath.h
View File

@@ -424,55 +424,6 @@ struct unpacket_traits<Packet8bf> {
masked_store_available = false
};
};
inline std::ostream& operator<<(std::ostream& s, const Packet16c& v) {
union {
Packet16c v;
signed char n[16];
} vt;
vt.v = v;
for (int i = 0; i < 16; i++) s << vt.n[i] << ", ";
return s;
}
inline std::ostream& operator<<(std::ostream& s, const Packet16uc& v) {
union {
Packet16uc v;
unsigned char n[16];
} vt;
vt.v = v;
for (int i = 0; i < 16; i++) s << vt.n[i] << ", ";
return s;
}
inline std::ostream& operator<<(std::ostream& s, const Packet4f& v) {
union {
Packet4f v;
float n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
return s;
}
inline std::ostream& operator<<(std::ostream& s, const Packet4i& v) {
union {
Packet4i v;
int n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
return s;
}
inline std::ostream& operator<<(std::ostream& s, const Packet4ui& v) {
union {
Packet4ui v;
unsigned int n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3];
return s;
}
template <typename Packet>
EIGEN_STRONG_INLINE Packet pload_common(const __UNPACK_TYPE__(Packet) * from) {
@@ -2384,6 +2335,44 @@ EIGEN_STRONG_INLINE Packet8bf pmadd(const Packet8bf& a, const Packet8bf& b, cons
return F32ToBf16(pmadd_even, pmadd_odd);
}
template <>
EIGEN_STRONG_INLINE Packet8bf pmsub(const Packet8bf& a, const Packet8bf& b, const Packet8bf& c) {
Packet4f a_even = Bf16ToF32Even(a);
Packet4f a_odd = Bf16ToF32Odd(a);
Packet4f b_even = Bf16ToF32Even(b);
Packet4f b_odd = Bf16ToF32Odd(b);
Packet4f c_even = Bf16ToF32Even(c);
Packet4f c_odd = Bf16ToF32Odd(c);
Packet4f pmadd_even = pmsub<Packet4f>(a_even, b_even, c_even);
Packet4f pmadd_odd = pmsub<Packet4f>(a_odd, b_odd, c_odd);
return F32ToBf16(pmadd_even, pmadd_odd);
}
template <>
EIGEN_STRONG_INLINE Packet8bf pnmadd(const Packet8bf& a, const Packet8bf& b, const Packet8bf& c) {
Packet4f a_even = Bf16ToF32Even(a);
Packet4f a_odd = Bf16ToF32Odd(a);
Packet4f b_even = Bf16ToF32Even(b);
Packet4f b_odd = Bf16ToF32Odd(b);
Packet4f c_even = Bf16ToF32Even(c);
Packet4f c_odd = Bf16ToF32Odd(c);
Packet4f pmadd_even = pnmadd<Packet4f>(a_even, b_even, c_even);
Packet4f pmadd_odd = pnmadd<Packet4f>(a_odd, b_odd, c_odd);
return F32ToBf16(pmadd_even, pmadd_odd);
}
template <>
EIGEN_STRONG_INLINE Packet8bf pnmsub(const Packet8bf& a, const Packet8bf& b, const Packet8bf& c) {
Packet4f a_even = Bf16ToF32Even(a);
Packet4f a_odd = Bf16ToF32Odd(a);
Packet4f b_even = Bf16ToF32Even(b);
Packet4f b_odd = Bf16ToF32Odd(b);
Packet4f c_even = Bf16ToF32Even(c);
Packet4f c_odd = Bf16ToF32Odd(c);
Packet4f pmadd_even = pnmsub<Packet4f>(a_even, b_even, c_even);
Packet4f pmadd_odd = pnmsub<Packet4f>(a_odd, b_odd, c_odd);
return F32ToBf16(pmadd_even, pmadd_odd);
}
template <>
EIGEN_STRONG_INLINE Packet8bf pmin<Packet8bf>(const Packet8bf& a, const Packet8bf& b) {
BF16_TO_F32_BINARY_OP_WRAPPER(pmin<Packet4f>, a, b);