| Re: [eigen] some problems with building with AltiVec enabled |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
- To: eigen@xxxxxxxxxxxxxxxxxxx
- Subject: Re: [eigen] some problems with building with AltiVec enabled
- From: Konstantinos Margaritis <markos@xxxxxxxx>
- Date: Fri, 22 Aug 2008 13:37:49 +0300
- Organization: CODEX
Ok, I have an update of the AltiVec packetmath.h, seems ei_predux* methods
were buggy, so I fixed them. Plus, ei_pdiv, ei_pmul were also problematic, I
had to invent a new algorithm for ei_pmul (integer case) as AltiVec does not
have an instruction for that and most algorithms on that are about unsigned
ints (Eigen2 uses signed 32-bit ints). ei_pdiv was disabled for now, but I
intend to look at it again closer. Right now many tests pass, except these:
* cholesky (some times it fails, 3/10 failures)
* geometry (3/10 failures)
* inverse (1/10 failures)
* sum (2/10 failures)
All the rest tests pass. I'll look into that a bit more closer to see what's
causing these tests fail.
Konstantinos
PS. I also added an overloaded stream operator << to be able to print AltiVec
vectors directly using cout (proved to be very useful for debugging :)
I suggest that SSE should use sth similar, I would provide a patch for that,
but I'm not sure if unions are the way to do sth like that in SSE :)
// This file is part of Eigen, a lightweight C++ template library
// for linear algebra. Eigen itself is part of the KDE project.
//
// Copyright (C) 2008 Konstantinos Margaritis <markos@xxxxxxxx>
// Copyright (C) 2008 Gael Guennebaud <g.gael@xxxxxxx>
//
// Eigen is free software; you can redistribute it and/or
// modify it under the terms of the GNU Lesser General Public
// License as published by the Free Software Foundation; either
// version 3 of the License, or (at your option) any later version.
//
// Alternatively, you can redistribute it and/or
// modify it under the terms of the GNU General Public License as
// published by the Free Software Foundation; either version 2 of
// the License, or (at your option) any later version.
//
// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY
// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public
// License and a copy of the GNU General Public License along with
// Eigen. If not, see <http://www.gnu.org/licenses/>.
#ifndef EIGEN_PACKET_MATH_ALTIVEC_H
#define EIGEN_PACKET_MATH_ALTIVEC_H
#ifndef EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD
#define EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD 4
#endif
typedef vector float v4f;
typedef vector int v4i;
typedef vector unsigned int v4ui;
typedef vector __bool int v4bi;
static const v4i v0i = vec_splat_s32(0);
static const v4i v1i = vec_splat_s32(1);
static const v4i v16i_ = vec_splat_s32(-16);
static const v4f v0f = (v4f) v0i;
static const v4f v1f = (v4f) v1i;
template<> struct ei_packet_traits<float> { typedef v4f type; enum {size=4}; };
template<> struct ei_packet_traits<int> { typedef v4i type; enum {size=4}; };
template<> struct ei_unpacket_traits<v4f> { typedef float type; enum {size=4}; };
template<> struct ei_unpacket_traits<v4i> { typedef int type; enum {size=4}; };
std::ostream & operator <<(std::ostream & s, const v4f & v)
{
union {
v4f v;
float n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3] << std::endl;
return s;
}
std::ostream & operator <<(std::ostream & s, const v4i & v)
{
union {
v4i v;
int n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3] << std::endl;
return s;
}
std::ostream & operator <<(std::ostream & s, const v4ui & v)
{
union {
v4ui v;
unsigned int n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3] << std::endl;
return s;
}
std::ostream & operator <<(std::ostream & s, const v4bi & v)
{
union {
vector __bool int v;
unsigned int n[4];
} vt;
vt.v = v;
s << vt.n[0] << ", " << vt.n[1] << ", " << vt.n[2] << ", " << vt.n[3] << std::endl;
return s;
}
template<> inline v4f ei_padd(const v4f& a, const v4f& b) { return vec_add(a,b); }
template<> inline v4i ei_padd(const v4i& a, const v4i& b) { return vec_add(a,b); }
template<> inline v4f ei_psub(const v4f& a, const v4f& b) { return vec_sub(a,b); }
template<> inline v4i ei_psub(const v4i& a, const v4i& b) { return vec_sub(a,b); }
template<> inline v4f ei_pmul(const v4f& a, const v4f& b) { return vec_madd(a,b, v0f); }
template<> inline v4i ei_pmul(const v4i& a, const v4i& b)
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/algorithms.html#Multiply32
//Set up constants
v4i bswap, low_prod, high_prod, prod, prod_, a_, b_, a1, b1;
v4i v1_, v0_, v1sel;
v1_ = (v4i) vec_splat_u32(-1);
v0_ = (v4i) vec_sl((v4ui) v1_, (v4ui) v1_);
// std::cout << "a: " << a << std::endl;
// std::cout << "b: " << b << std::endl;
a_ = vec_sub(v0i, a);
b_ = vec_sub(v0i, b);
a1 = vec_max(a, a_);
b1 = vec_max(b, b_);
v4bi sgn = (v4bi) vec_cmplt(vec_and(vec_xor(a, b), v0_), v0i);
// std::cout << "a: " << a1 << std::endl;
// std::cout << "b: " << b1 << std::endl;
// std::cout << "a_: " << a_ << std::endl;
// std::cout << "b_: " << b_ << std::endl;
// std::cout << "sgn(a*b): " << std::hex << sgn << std::dec << std::endl;
// Do real work
bswap = (v4i) vec_rl((v4ui) b1, (v4ui) v16i_ );
low_prod = vec_mulo((vector short)a1, (vector short)b1);
high_prod = vec_msum((vector short)a1, (vector short)bswap, v0i);
high_prod = (v4i) vec_sl((v4ui) high_prod, (v4ui) v16i_);
prod = vec_add( low_prod, high_prod );
// std::cout << "prod: " << prod << std::endl;
// std::cout << "v0i: " << v0i << std::endl;
prod_ = vec_nor(prod, prod);
prod_ = vec_sel(v0i, prod_, sgn);
v1sel = vec_sel(v0i, v1i, sgn);
// std::cout << "prod_: " << prod_ << std::endl;
// std::cout << "v1sel: " << v1sel << std::endl;
prod_ = vec_add(prod_, v1sel);
// std::cout << "prod_: " << prod_ << std::endl;
prod = vec_sel(prod, prod_, sgn);
// std::cout << "final prod: " << prod << std::endl;
return prod;
}
template<> inline v4f ei_pdiv(const v4f& a, const v4f& b) {
std::cout << "a: " << a << std::endl;
std::cout << "b: " << b << std::endl;
// Altivec does not offer a divide instruction, we have to do a reciprocal approximation
v4f y = vec_re(b);
std::cout << "1/b: " << y << std::endl;
// Set up some constants for inverse reciprocals
vector unsigned int v1_;
v4f v0_, t;
v1_ = vec_splat_u32(-1);
std::cout << "-1: " << (v4i) v1_ << std::endl;
v0_ = (v4f) vec_sl(v1_, v1_);
std::cout << "-0.0: " << v0_ << std::endl;
// Do a Newton-Raphson iteration to get the needed accuracy
t = vec_nmsub(y, b, v1f);
std::cout << "t: " << t << std::endl;
y = vec_nmsub(t, y, v0_);
v4f res = vec_madd(a, y, v0_);
std::cout << "res: " << res << std::endl;
return res;
}
template<> inline v4f ei_pmadd(const v4f& a, const v4f& b, const v4f& c) { return vec_madd(a, b, c); }
template<> inline v4f ei_pmin(const v4f& a, const v4f& b) { return vec_min(a,b); }
template<> inline v4i ei_pmin(const v4i& a, const v4i& b) { return vec_min(a,b); }
template<> inline v4f ei_pmax(const v4f& a, const v4f& b) { return vec_max(a,b); }
template<> inline v4i ei_pmax(const v4i& a, const v4i& b) { return vec_max(a,b); }
template<> inline v4f ei_pload(const float* from) { return vec_ld(0, from); }
template<> inline v4i ei_pload(const int* from) { return vec_ld(0, from); }
template<> inline v4f ei_ploadu(const float* from)
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
vector unsigned char MSQ, LSQ;
vector unsigned char mask;
MSQ = vec_ld(0, (unsigned char *)from); // most significant quadword
LSQ = vec_ld(15, (unsigned char *)from); // least significant quadword
mask = vec_lvsl(0, from); // create the permute mask
return (v4f) vec_perm(MSQ, LSQ, mask); // align the data
}
template<> inline v4i ei_ploadu(const int* from)
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
vector unsigned char MSQ, LSQ;
vector unsigned char mask;
MSQ = vec_ld(0, (unsigned char *)from); // most significant quadword
LSQ = vec_ld(15, (unsigned char *)from); // least significant quadword
mask = vec_lvsl(0, from); // create the permute mask
return (v4i) vec_perm(MSQ, LSQ, mask); // align the data
}
template<> inline v4f ei_pset1(const float& from)
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
float __attribute__(aligned(16)) af[4];
af[0] = from;
v4f vc = vec_ld(0, af);
vc = vec_splat(vc, 0);
return vc;
}
template<> inline v4i ei_pset1(const int& from)
{
int __attribute__(aligned(16)) ai[4];
ai[0] = from;
v4i vc = vec_ld(0, ai);
vc = vec_splat(vc, 0);
return vc;
}
template<> inline void ei_pstore(float* to, const v4f& from) { vec_st(from, 0, to); }
template<> inline void ei_pstore(int* to, const v4i& from) { vec_st(from, 0, to); }
template<> inline void ei_pstoreu(float* to, const v4f& from)
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
// Warning: not thread safe!
vector unsigned char MSQ, LSQ, edges;
vector unsigned char edgeAlign, align;
MSQ = vec_ld(0, (unsigned char *)to); // most significant quadword
LSQ = vec_ld(15, (unsigned char *)to); // least significant quadword
edgeAlign = vec_lvsl(0, to); // permute map to extract edges
edges=vec_perm(LSQ,MSQ,edgeAlign); // extract the edges
align = vec_lvsr( 0, to ); // permute map to misalign data
MSQ = vec_perm(edges,(vector unsigned char)from,align); // misalign the data (MSQ)
LSQ = vec_perm((vector unsigned char)from,edges,align); // misalign the data (LSQ)
vec_st( LSQ, 15, (unsigned char *)to ); // Store the LSQ part first
vec_st( MSQ, 0, (unsigned char *)to ); // Store the MSQ part
}
template<> inline void ei_pstoreu(int* to , const v4i& from )
{
// Taken from http://developer.apple.com/hardwaredrivers/ve/alignment.html
// Warning: not thread safe!
vector unsigned char MSQ, LSQ, edges;
vector unsigned char edgeAlign, align;
MSQ = vec_ld(0, (unsigned char *)to); // most significant quadword
LSQ = vec_ld(15, (unsigned char *)to); // least significant quadword
edgeAlign = vec_lvsl(0, to); // permute map to extract edges
edges=vec_perm(LSQ,MSQ,edgeAlign); // extract the edges
align = vec_lvsr( 0, to ); // permute map to misalign data
MSQ = vec_perm(edges,(vector unsigned char)from,align); // misalign the data (MSQ)
LSQ = vec_perm((vector unsigned char)from,edges,align); // misalign the data (LSQ)
vec_st( LSQ, 15, (unsigned char *)to ); // Store the LSQ part first
vec_st( MSQ, 0, (unsigned char *)to ); // Store the MSQ part
}
template<> inline float ei_pfirst(const v4f& a)
{
float __attribute__(aligned(16)) af[4];
vec_st(a, 0, af);
return af[0];
}
template<> inline int ei_pfirst(const v4i& a)
{
int __attribute__(aligned(16)) ai[4];
vec_st(a, 0, ai);
return ai[0];
}
inline v4f ei_preduxp(const v4f* vecs)
{
v4f v[4], sum[4];
// It's easier and faster to transpose then add as columns
// Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation
// Do the transpose, first set of moves
v[0] = vec_mergeh(vecs[0], vecs[2]);
v[1] = vec_mergel(vecs[0], vecs[2]);
v[2] = vec_mergeh(vecs[1], vecs[3]);
v[3] = vec_mergel(vecs[1], vecs[3]);
// Get the resulting vectors
sum[0] = vec_mergeh(v[0], v[2]);
sum[1] = vec_mergel(v[0], v[2]);
sum[2] = vec_mergeh(v[1], v[3]);
sum[3] = vec_mergel(v[1], v[3]);
// Now do the summation:
// Lines 0+1
sum[0] = vec_add(sum[0], sum[1]);
// Lines 2+3
sum[1] = vec_add(sum[2], sum[3]);
// Add the results
sum[0] = vec_add(sum[0], sum[1]);
return sum[0];
}
inline float ei_predux(const v4f& a)
{
v4f b, sum;
b = (v4f)vec_sld(a, a, 8);
sum = vec_add(a, b);
b = (v4f)vec_sld(sum, sum, 4);
sum = vec_add(sum, b);
return ei_pfirst(sum);
}
inline v4i ei_preduxp(const v4i* vecs)
{
v4i v[4], sum[4];
// It's easier and faster to transpose then add as columns
// Check: http://www.freevec.org/function/matrix_4x4_transpose_floats for explanation
// Do the transpose, first set of moves
v[0] = vec_mergeh(vecs[0], vecs[2]);
v[1] = vec_mergel(vecs[0], vecs[2]);
v[2] = vec_mergeh(vecs[1], vecs[3]);
v[3] = vec_mergel(vecs[1], vecs[3]);
// Get the resulting vectors
sum[0] = vec_mergeh(v[0], v[2]);
sum[1] = vec_mergel(v[0], v[2]);
sum[2] = vec_mergeh(v[1], v[3]);
sum[3] = vec_mergel(v[1], v[3]);
// Now do the summation:
// Lines 0+1
sum[0] = vec_add(sum[0], sum[1]);
// Lines 2+3
sum[1] = vec_add(sum[2], sum[3]);
// Add the results
sum[0] = vec_add(sum[0], sum[1]);
return sum[0];
}
inline int ei_predux(const v4i& a)
{
v4i sum;
sum = vec_sums(a, v0i);
sum = vec_sld(sum, v0i, 12);
return ei_pfirst(sum);
}
template<int Offset>
struct ei_palign_impl<Offset, v4f>
{
inline static void run(v4f& first, const v4f& second)
{
first = vec_sld(first, second, Offset*4);
}
};
template<int Offset>
struct ei_palign_impl<Offset, v4i>
{
inline static void run(v4i& first, const v4i& second)
{
first = vec_sld(first, second, Offset*4);
}
};
#endif // EIGEN_PACKET_MATH_ALTIVEC_H