Fix naive implementation, NH3 working for both
This commit is contained in:
parent
a5619146f0
commit
7b617930a6
183
atrip.org
183
atrip.org
@ -1449,94 +1449,7 @@ struct TuplesDistribution {
|
||||
|
||||
|
||||
|
||||
*** Naive list
|
||||
|
||||
The naive implementation of the global tuples list is simple
|
||||
three for loops creating tuples of the sort
|
||||
\( (a,b,c) \) where the following conditions are met at the same time:
|
||||
- \( a \leq b \leq c \)
|
||||
- \(
|
||||
a \neq b \land b \neq c
|
||||
\)
|
||||
|
||||
This means,
|
||||
\( (1, 2, 3)
|
||||
, (1, 1, 3)
|
||||
, (1, 2, 2)
|
||||
\) are acceptable tuples wherease \( (2, 1, 1) \) and \( (1, 1, 1) \) are not.
|
||||
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
ABCTuples getTuplesList(size_t Nv, size_t rank, size_t np) {
|
||||
|
||||
const size_t
|
||||
// total number of tuples for the problem
|
||||
n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv
|
||||
|
||||
// all ranks should have the same number of tuples_per_rank
|
||||
, tuples_per_rank = n / np + size_t(n % np != 0)
|
||||
|
||||
// start index for the global tuples list
|
||||
, start = tuples_per_rank * rank
|
||||
|
||||
// end index for the global tuples list
|
||||
, end = tuples_per_rank * (rank + 1)
|
||||
;
|
||||
|
||||
ABCTuples result(tuples_per_rank, FAKE_TUPLE);
|
||||
|
||||
for (size_t a(0), r(0), g(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++, g++){
|
||||
if ( a == b && b == c ) continue;
|
||||
if ( g > start && g <= end) result[r++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
|
||||
}
|
||||
#+end_src
|
||||
|
||||
and all tuples would simply be
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
ABCTuples getAllTuplesList(const size_t Nv) {
|
||||
const size_t n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv;
|
||||
ABCTuples result(n);
|
||||
|
||||
for (size_t a(0), u(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
result[u++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
#+end_src
|
||||
|
||||
|
||||
With =getTupleList= we can easily define a tuple distribution like
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
struct NaiveDistribution : public TuplesDistribution {
|
||||
ABCTuples getTuples(size_t Nv, MPI_Comm universe) override {
|
||||
int rank, np;
|
||||
MPI_Comm_rank(universe, &rank);
|
||||
MPI_Comm_size(universe, &np);
|
||||
return getTuplesList(Nv, (size_t)rank, (size_t)np);
|
||||
}
|
||||
};
|
||||
#+end_src
|
||||
|
||||
|
||||
*** Group and sort list
|
||||
**** Prolog :noexport:
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
namespace group_and_sort {
|
||||
#+end_src
|
||||
|
||||
**** Node information
|
||||
*** Node information
|
||||
|
||||
- nodeList ::
|
||||
List of hostnames of size \( N_n \)
|
||||
@ -1630,6 +1543,96 @@ getNodeInfos(std::vector<string> const& nodeNames) {
|
||||
}
|
||||
#+end_src
|
||||
|
||||
*** Naive list
|
||||
|
||||
The naive implementation of the global tuples list is simple
|
||||
three for loops creating tuples of the sort
|
||||
\( (a,b,c) \) where the following conditions are met at the same time:
|
||||
- \( a \leq b \leq c \)
|
||||
- \(
|
||||
a \neq b \land b \neq c
|
||||
\)
|
||||
|
||||
This means,
|
||||
\( (1, 2, 3)
|
||||
, (1, 1, 3)
|
||||
, (1, 2, 2)
|
||||
\) are acceptable tuples wherease \( (2, 1, 1) \) and \( (1, 1, 1) \) are not.
|
||||
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
ABCTuples getTuplesList(size_t Nv, size_t rank, size_t np) {
|
||||
|
||||
const size_t
|
||||
// total number of tuples for the problem
|
||||
n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv
|
||||
|
||||
// all ranks should have the same number of tuples_per_rank
|
||||
, tuples_per_rank = n / np + size_t(n % np != 0)
|
||||
|
||||
// start index for the global tuples list
|
||||
, start = tuples_per_rank * rank
|
||||
|
||||
// end index for the global tuples list
|
||||
, end = tuples_per_rank * (rank + 1)
|
||||
;
|
||||
|
||||
LOG(1,"Atrip") << "tuples_per_rank = " << tuples_per_rank << "\n";
|
||||
WITH_RANK << "start, end = " << start << ", " << end << "\n";
|
||||
ABCTuples result(tuples_per_rank, FAKE_TUPLE);
|
||||
|
||||
for (size_t a(0), r(0), g(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
if ( start <= g && g < end) result[r++] = {a, b, c};
|
||||
g++;
|
||||
}
|
||||
|
||||
return result;
|
||||
|
||||
}
|
||||
#+end_src
|
||||
|
||||
and all tuples would simply be
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
ABCTuples getAllTuplesList(const size_t Nv) {
|
||||
const size_t n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv;
|
||||
ABCTuples result(n);
|
||||
|
||||
for (size_t a(0), u(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
result[u++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
#+end_src
|
||||
|
||||
|
||||
With =getTupleList= we can easily define a tuple distribution like
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
struct NaiveDistribution : public TuplesDistribution {
|
||||
ABCTuples getTuples(size_t Nv, MPI_Comm universe) override {
|
||||
int rank, np;
|
||||
MPI_Comm_rank(universe, &rank);
|
||||
MPI_Comm_size(universe, &np);
|
||||
return getTuplesList(Nv, (size_t)rank, (size_t)np);
|
||||
}
|
||||
};
|
||||
#+end_src
|
||||
|
||||
|
||||
*** Group and sort list
|
||||
**** Prolog :noexport:
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
namespace group_and_sort {
|
||||
#+end_src
|
||||
|
||||
**** Utils
|
||||
|
||||
#+begin_src c++ :tangle (atrip-tuples-h)
|
||||
@ -2863,7 +2866,7 @@ Atrip::Output Atrip::run(Atrip::Input const& in) {
|
||||
LOG(0,"Atrip") << "#iterations: "
|
||||
<< nIterations
|
||||
<< "/"
|
||||
<< _all_tuples
|
||||
<< nIterations * np
|
||||
<< "\n";
|
||||
}
|
||||
|
||||
@ -3052,8 +3055,6 @@ Atrip::Output Atrip::run(Atrip::Input const& in) {
|
||||
: tuplesList[i]
|
||||
, *abcNext = i == (tuplesList.size() - 1)
|
||||
? nullptr
|
||||
: isFakeTuple(i + 1)
|
||||
? &tuplesList[tuplesList.size() - 1]
|
||||
: &tuplesList[i + 1]
|
||||
;
|
||||
|
||||
|
||||
@ -36,68 +36,6 @@ struct TuplesDistribution {
|
||||
};
|
||||
// Distributing the tuples:1 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:1]]
|
||||
ABCTuples getTuplesList(size_t Nv, size_t rank, size_t np) {
|
||||
|
||||
const size_t
|
||||
// total number of tuples for the problem
|
||||
n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv
|
||||
|
||||
// all ranks should have the same number of tuples_per_rank
|
||||
, tuples_per_rank = n / np + size_t(n % np != 0)
|
||||
|
||||
// start index for the global tuples list
|
||||
, start = tuples_per_rank * rank
|
||||
|
||||
// end index for the global tuples list
|
||||
, end = tuples_per_rank * (rank + 1)
|
||||
;
|
||||
|
||||
ABCTuples result(tuples_per_rank, FAKE_TUPLE);
|
||||
|
||||
for (size_t a(0), r(0), g(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++, g++){
|
||||
if ( a == b && b == c ) continue;
|
||||
if ( g > start && g <= end) result[r++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
|
||||
}
|
||||
// Naive list:1 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:2]]
|
||||
ABCTuples getAllTuplesList(const size_t Nv) {
|
||||
const size_t n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv;
|
||||
ABCTuples result(n);
|
||||
|
||||
for (size_t a(0), u(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
result[u++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
// Naive list:2 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:3]]
|
||||
struct NaiveDistribution : public TuplesDistribution {
|
||||
ABCTuples getTuples(size_t Nv, MPI_Comm universe) override {
|
||||
int rank, np;
|
||||
MPI_Comm_rank(universe, &rank);
|
||||
MPI_Comm_size(universe, &np);
|
||||
return getTuplesList(Nv, (size_t)rank, (size_t)np);
|
||||
}
|
||||
};
|
||||
// Naive list:3 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Prolog][Prolog:1]]
|
||||
namespace group_and_sort {
|
||||
// Prolog:1 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Node%20information][Node information:1]]
|
||||
std::vector<std::string> getNodeNames(MPI_Comm comm){
|
||||
int rank, np;
|
||||
@ -177,6 +115,71 @@ getNodeInfos(std::vector<string> const& nodeNames) {
|
||||
}
|
||||
// Node information:2 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:1]]
|
||||
ABCTuples getTuplesList(size_t Nv, size_t rank, size_t np) {
|
||||
|
||||
const size_t
|
||||
// total number of tuples for the problem
|
||||
n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv
|
||||
|
||||
// all ranks should have the same number of tuples_per_rank
|
||||
, tuples_per_rank = n / np + size_t(n % np != 0)
|
||||
|
||||
// start index for the global tuples list
|
||||
, start = tuples_per_rank * rank
|
||||
|
||||
// end index for the global tuples list
|
||||
, end = tuples_per_rank * (rank + 1)
|
||||
;
|
||||
|
||||
LOG(1,"Atrip") << "tuples_per_rank = " << tuples_per_rank << "\n";
|
||||
WITH_RANK << "start, end = " << start << ", " << end << "\n";
|
||||
ABCTuples result(tuples_per_rank, FAKE_TUPLE);
|
||||
|
||||
for (size_t a(0), r(0), g(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
if ( start <= g && g < end) result[r++] = {a, b, c};
|
||||
g++;
|
||||
}
|
||||
|
||||
return result;
|
||||
|
||||
}
|
||||
// Naive list:1 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:2]]
|
||||
ABCTuples getAllTuplesList(const size_t Nv) {
|
||||
const size_t n = Nv * (Nv + 1) * (Nv + 2) / 6 - Nv;
|
||||
ABCTuples result(n);
|
||||
|
||||
for (size_t a(0), u(0); a < Nv; a++)
|
||||
for (size_t b(a); b < Nv; b++)
|
||||
for (size_t c(b); c < Nv; c++){
|
||||
if ( a == b && b == c ) continue;
|
||||
result[u++] = {a, b, c};
|
||||
}
|
||||
|
||||
return result;
|
||||
}
|
||||
// Naive list:2 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Naive%20list][Naive list:3]]
|
||||
struct NaiveDistribution : public TuplesDistribution {
|
||||
ABCTuples getTuples(size_t Nv, MPI_Comm universe) override {
|
||||
int rank, np;
|
||||
MPI_Comm_rank(universe, &rank);
|
||||
MPI_Comm_size(universe, &np);
|
||||
return getTuplesList(Nv, (size_t)rank, (size_t)np);
|
||||
}
|
||||
};
|
||||
// Naive list:3 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Prolog][Prolog:1]]
|
||||
namespace group_and_sort {
|
||||
// Prolog:1 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Utils][Utils:1]]
|
||||
// Provides the node on which the slice-element is found
|
||||
// Right now we distribute the slices in a round robin fashion
|
||||
@ -460,13 +463,7 @@ if (makeDistribution) {
|
||||
// Main:4 ends here
|
||||
|
||||
// [[file:~/atrip/atrip.org::*Main][Main:6]]
|
||||
/*
|
||||
result.insert(result.end(),
|
||||
tuplesPerRankGlobal - result.size(),
|
||||
FAKE_TUPLE);
|
||||
*/
|
||||
|
||||
LOG(1,"Atrip") << "scattering tuples \n";
|
||||
LOG(1,"Atrip") << "scattering tuples \n";
|
||||
|
||||
return result;
|
||||
|
||||
|
||||
@ -102,7 +102,7 @@ Atrip::Output Atrip::run(Atrip::Input const& in) {
|
||||
LOG(0,"Atrip") << "#iterations: "
|
||||
<< nIterations
|
||||
<< "/"
|
||||
<< _all_tuples
|
||||
<< nIterations * np
|
||||
<< "\n";
|
||||
}
|
||||
|
||||
@ -291,8 +291,6 @@ Atrip::Output Atrip::run(Atrip::Input const& in) {
|
||||
: tuplesList[i]
|
||||
, *abcNext = i == (tuplesList.size() - 1)
|
||||
? nullptr
|
||||
: isFakeTuple(i + 1)
|
||||
? &tuplesList[tuplesList.size() - 1]
|
||||
: &tuplesList[i + 1]
|
||||
;
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user