gallo/atrip
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Implement deleting Vppph

Browse Source
This commit is contained in:
Alejandro Gallo 2022-04-05 17:16:11 +02:00
parent 7d37bd9f8b
commit fed19ff52c
7 changed files with 196 additions and 181 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
atrip.org
View File

@ -2935,13 +2935,13 @@ namespace atrip {
template <typename F=double> template <typename F=double>
struct Input { struct Input {
CTF::Tensor<F> *ei = nullptr CTF::Tensor<F> *ei = nullptr
, *ea = nullptr , *ea = nullptr
, *Tph = nullptr , *Tph = nullptr
, *Tpphh = nullptr , *Tpphh = nullptr
, *Vpphh = nullptr , *Vpphh = nullptr
, *Vhhhp = nullptr , *Vhhhp = nullptr
, *Vppph = nullptr , *Vppph = nullptr
; ;
Input& with_epsilon_i(CTF::Tensor<F> * t) { ei = t; return *this; } Input& with_epsilon_i(CTF::Tensor<F> * t) { ei = t; return *this; }
Input& with_epsilon_a(CTF::Tensor<F> * t) { ea = t; return *this; } Input& with_epsilon_a(CTF::Tensor<F> * t) { ea = t; return *this; }
Input& with_Tai(CTF::Tensor<F> * t) { Tph = t; return *this; } Input& with_Tai(CTF::Tensor<F> * t) { Tph = t; return *this; }
@ -2955,6 +2955,7 @@ namespace atrip {
GROUP_AND_SORT, GROUP_AND_SORT,
}; };
ADD_ATTRIBUTE(bool, deleteVppph, false)
ADD_ATTRIBUTE(bool, rankRoundRobin, false) ADD_ATTRIBUTE(bool, rankRoundRobin, false)
ADD_ATTRIBUTE(bool, chrono, false) ADD_ATTRIBUTE(bool, chrono, false)
ADD_ATTRIBUTE(bool, barrier, false) ADD_ATTRIBUTE(bool, barrier, false)
@ -3063,14 +3064,6 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
MPI_Comm_size(child_comm, &child_size); MPI_Comm_size(child_comm, &child_size);
} }
// BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
WITH_CHRONO("nv-slices",
LOG(0,"Atrip") << "BUILD NV-SLICES\n";
TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
HHHA<F> hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
)
// BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1 // BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1
WITH_CHRONO("nv-nv-slices", WITH_CHRONO("nv-nv-slices",
LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n"; LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n";
@ -3079,6 +3072,18 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe); TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
) )
// delete the Vppph so that we don't have a HWM situation for the NV slices
if (in.deleteVppph) {
delete in.Vppph;
}
// BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
WITH_CHRONO("nv-slices",
LOG(0,"Atrip") << "BUILD NV-SLICES\n";
TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
HHHA<F> hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
)
// all tensors // all tensors
std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh}; std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh};
@ -3251,12 +3256,12 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
const double doublesFlops const double doublesFlops
= double(No) = double(No)
* double(No) ,* double(No)
* double(No) ,* double(No)
* (double(No) + double(Nv)) ,* (double(No) + double(Nv))
* 2.0 ,* 2.0
* (traits::isComplex<F>() ? 2.0 : 1.0) ,* (traits::isComplex<F>() ? 2.0 : 1.0)
* 6.0 ,* 6.0
/ 1e9 / 1e9
; ;

53
bench/test_main.cxx
View File

@ -17,13 +17,15 @@ int main(int argc, char** argv) {
MPI_Init(&argc, &argv); MPI_Init(&argc, &argv);
int no(10), nv(10), itMod(-1), percentageMod(10); int no(10), nv(10), itMod(-1), percentageMod(10);
bool nochrono(false), barrier(false), rankRoundRobin(false); bool nochrono(false), barrier(false), rankRoundRobin(false),
keepVppph(false);
std::string tuplesDistributionString = "naive"; std::string tuplesDistributionString = "naive";
CLI::App app{"Main bench for atrip"}; CLI::App app{"Main bench for atrip"};
app.add_option("--no", no, "Occupied orbitals"); app.add_option("--no", no, "Occupied orbitals");
app.add_option("--nv", nv, "Virtual orbitals"); app.add_option("--nv", nv, "Virtual orbitals");
app.add_option("--mod", itMod, "Iteration modifier"); app.add_option("--mod", itMod, "Iteration modifier");
app.add_flag("--keep-vppph", keepVppph, "Do not delete Vppph");
app.add_flag("--nochrono", nochrono, "Do not print chrono"); app.add_flag("--nochrono", nochrono, "Do not print chrono");
app.add_flag("--rank-round-robin", rankRoundRobin, "Do rank round robin"); app.add_flag("--rank-round-robin", rankRoundRobin, "Do rank round robin");
app.add_flag("--barrier", barrier, "Use the first barrier"); app.add_flag("--barrier", barrier, "Use the first barrier");
@ -38,12 +40,13 @@ int main(int argc, char** argv) {
constexpr double elem_to_gb = 8.0 / 1024.0 / 1024.0 / 1024.0; constexpr double elem_to_gb = 8.0 / 1024.0 / 1024.0 / 1024.0;
// USER PRINTING TEST BEGIN // USER PRINTING TEST BEGIN
const double doublesFlops = no * no * no const double doublesFlops
* (no + nv) = no * no * no
* 2.0 * (no + nv)
* 6.0 * 2.0
/ 1.0e9 * 6.0
; / 1.0e9
;
double lastElapsedTime = 0; double lastElapsedTime = 0;
bool firstHeaderPrinted = false; bool firstHeaderPrinted = false;
atrip::registerIterationDescriptor atrip::registerIterationDescriptor
@ -110,23 +113,25 @@ int main(int argc, char** argv) {
Vppph.fill_random(0, 1); Vppph.fill_random(0, 1);
atrip::Atrip::init(); atrip::Atrip::init();
const auto in = atrip::Atrip::Input<double>() const auto in
// Tensors = atrip::Atrip::Input<double>()
.with_epsilon_i(&ei) // Tensors
.with_epsilon_a(&ea) .with_epsilon_i(&ei)
.with_Tai(&Tph) .with_epsilon_a(&ea)
.with_Tabij(&Tpphh) .with_Tai(&Tph)
.with_Vabij(&Vpphh) .with_Tabij(&Tpphh)
.with_Vijka(&Vhhhp) .with_Vabij(&Vpphh)
.with_Vabci(&Vppph) .with_Vijka(&Vhhhp)
// some options .with_Vabci(&Vppph)
.with_barrier(barrier) // some options
.with_chrono(!nochrono) .with_deleteVppph(!keepVppph)
.with_rankRoundRobin(rankRoundRobin) .with_barrier(barrier)
.with_iterationMod(itMod) .with_chrono(!nochrono)
.with_percentageMod(percentageMod) .with_rankRoundRobin(rankRoundRobin)
.with_tuplesDistribution(tuplesDistribution) .with_iterationMod(itMod)
; .with_percentageMod(percentageMod)
.with_tuplesDistribution(tuplesDistribution)
;
auto out = atrip::Atrip::run(in); auto out = atrip::Atrip::run(in);

15
include/atrip/Atrip.hpp
View File

@ -40,13 +40,13 @@ namespace atrip {
template <typename F=double> template <typename F=double>
struct Input { struct Input {
CTF::Tensor<F> *ei = nullptr CTF::Tensor<F> *ei = nullptr
, *ea = nullptr , *ea = nullptr
, *Tph = nullptr , *Tph = nullptr
, *Tpphh = nullptr , *Tpphh = nullptr
, *Vpphh = nullptr , *Vpphh = nullptr
, *Vhhhp = nullptr , *Vhhhp = nullptr
, *Vppph = nullptr , *Vppph = nullptr
; ;
Input& with_epsilon_i(CTF::Tensor<F> * t) { ei = t; return *this; } Input& with_epsilon_i(CTF::Tensor<F> * t) { ei = t; return *this; }
Input& with_epsilon_a(CTF::Tensor<F> * t) { ea = t; return *this; } Input& with_epsilon_a(CTF::Tensor<F> * t) { ea = t; return *this; }
Input& with_Tai(CTF::Tensor<F> * t) { Tph = t; return *this; } Input& with_Tai(CTF::Tensor<F> * t) { Tph = t; return *this; }
@ -60,6 +60,7 @@ namespace atrip {
GROUP_AND_SORT, GROUP_AND_SORT,
}; };
ADD_ATTRIBUTE(bool, deleteVppph, false)
ADD_ATTRIBUTE(bool, rankRoundRobin, false) ADD_ATTRIBUTE(bool, rankRoundRobin, false)
ADD_ATTRIBUTE(bool, chrono, false) ADD_ATTRIBUTE(bool, chrono, false)
ADD_ATTRIBUTE(bool, barrier, false) ADD_ATTRIBUTE(bool, barrier, false)

196
include/atrip/Slice.hpp
View File

@ -44,36 +44,36 @@ struct Slice {
// Prolog:1 ends here // Prolog:1 ends here
// [[file:../../atrip.org::*Location][Location:1]] // [[file:../../atrip.org::*Location][Location:1]]
struct Location { size_t rank; size_t source; }; struct Location { size_t rank; size_t source; };
// Location:1 ends here // Location:1 ends here
// [[file:../../atrip.org::*Type][Type:1]] // [[file:../../atrip.org::*Type][Type:1]]
enum Type enum Type
{ A = 10 { A = 10
, B , B
, C , C
// Two-parameter slices // Two-parameter slices
, AB = 20 , AB = 20
, BC , BC
, AC , AC
// for abci and the doubles // for abci and the doubles
, CB , CB
, BA , BA
, CA , CA
// The non-typed slice // The non-typed slice
, Blank = 404 , Blank = 404
}; };
// Type:1 ends here // Type:1 ends here
// [[file:../../atrip.org::*State][State:1]] // [[file:../../atrip.org::*State][State:1]]
enum State { enum State {
Fetch = 0, Fetch = 0,
Dispatched = 2, Dispatched = 2,
Ready = 1, Ready = 1,
SelfSufficient = 911, SelfSufficient = 911,
Recycled = 123, Recycled = 123,
Acceptor = 405 Acceptor = 405
}; };
// State:1 ends here // State:1 ends here
// [[file:../../atrip.org::*The Info structure][The Info structure:1]] // [[file:../../atrip.org::*The Info structure][The Info structure:1]]
@ -101,25 +101,25 @@ using Ty_x_Tu = std::pair< Type, PartialTuple >;
// The Info structure:1 ends here // The Info structure:1 ends here
// [[file:../../atrip.org::*Name][Name:1]] // [[file:../../atrip.org::*Name][Name:1]]
enum Name enum Name
{ TA = 100 { TA = 100
, VIJKA = 101 , VIJKA = 101
, VABCI = 200 , VABCI = 200
, TABIJ = 201 , TABIJ = 201
, VABIJ = 202 , VABIJ = 202
}; };
// Name:1 ends here // Name:1 ends here
// [[file:../../atrip.org::*Database][Database:1]] // [[file:../../atrip.org::*Database][Database:1]]
struct LocalDatabaseElement { struct LocalDatabaseElement {
Slice<F>::Name name; Slice<F>::Name name;
Slice<F>::Info info; Slice<F>::Info info;
}; };
// Database:1 ends here // Database:1 ends here
// [[file:../../atrip.org::*Database][Database:2]] // [[file:../../atrip.org::*Database][Database:2]]
using LocalDatabase = std::vector<LocalDatabaseElement>; using LocalDatabase = std::vector<LocalDatabaseElement>;
using Database = LocalDatabase; using Database = LocalDatabase;
// Database:2 ends here // Database:2 ends here
// [[file:../../atrip.org::*MPI Types][MPI Types:1]] // [[file:../../atrip.org::*MPI Types][MPI Types:1]]
@ -359,91 +359,91 @@ static Slice<F>& findByInfo(std::vector<Slice<F>> &slices,
// Static utilities:6 ends here // Static utilities:6 ends here
// [[file:../../atrip.org::*Attributes][Attributes:1]] // [[file:../../atrip.org::*Attributes][Attributes:1]]
Info info; Info info;
// Attributes:1 ends here // Attributes:1 ends here
// [[file:../../atrip.org::*Attributes][Attributes:2]] // [[file:../../atrip.org::*Attributes][Attributes:2]]
F *data; F *data;
// Attributes:2 ends here // Attributes:2 ends here
// [[file:../../atrip.org::*Attributes][Attributes:3]] // [[file:../../atrip.org::*Attributes][Attributes:3]]
MPI_Request request; MPI_Request request;
// Attributes:3 ends here // Attributes:3 ends here
// [[file:../../atrip.org::*Attributes][Attributes:4]] // [[file:../../atrip.org::*Attributes][Attributes:4]]
const size_t size; const size_t size;
// Attributes:4 ends here // Attributes:4 ends here
// [[file:../../atrip.org::*Member functions][Member functions:1]] // [[file:../../atrip.org::*Member functions][Member functions:1]]
void markReady() noexcept { void markReady() noexcept {
info.state = Ready; info.state = Ready;
info.recycling = Blank; info.recycling = Blank;
} }
// Member functions:1 ends here // Member functions:1 ends here
// [[file:../../atrip.org::*Member functions][Member functions:2]] // [[file:../../atrip.org::*Member functions][Member functions:2]]
bool isUnwrapped() const noexcept { bool isUnwrapped() const noexcept {
return info.state == Ready return info.state == Ready
|| info.state == SelfSufficient || info.state == SelfSufficient
; ;
} }
// Member functions:2 ends here // Member functions:2 ends here
// [[file:../../atrip.org::*Member functions][Member functions:3]] // [[file:../../atrip.org::*Member functions][Member functions:3]]
bool isUnwrappable() const noexcept { bool isUnwrappable() const noexcept {
return isUnwrapped() return isUnwrapped()
|| info.state == Recycled || info.state == Recycled
|| info.state == Dispatched || info.state == Dispatched
; ;
} }
inline bool isDirectlyFetchable() const noexcept { inline bool isDirectlyFetchable() const noexcept {
return info.state == Ready || info.state == Dispatched; return info.state == Ready || info.state == Dispatched;
} }
void free() noexcept { void free() noexcept {
info.tuple = {0, 0}; info.tuple = {0, 0};
info.type = Blank; info.type = Blank;
info.state = Acceptor; info.state = Acceptor;
info.from = {0, 0}; info.from = {0, 0};
info.recycling = Blank; info.recycling = Blank;
data = nullptr; data = nullptr;
} }
inline bool isFree() const noexcept { inline bool isFree() const noexcept {
return info.tuple == PartialTuple{0, 0} return info.tuple == PartialTuple{0, 0}
&& info.type == Blank && info.type == Blank
&& info.state == Acceptor && info.state == Acceptor
&& info.from.rank == 0 && info.from.rank == 0
&& info.from.source == 0 && info.from.source == 0
&& info.recycling == Blank && info.recycling == Blank
&& data == nullptr && data == nullptr
; ;
} }
// Member functions:3 ends here // Member functions:3 ends here
// [[file:../../atrip.org::*Member functions][Member functions:4]] // [[file:../../atrip.org::*Member functions][Member functions:4]]
inline bool isRecyclable() const noexcept { inline bool isRecyclable() const noexcept {
return ( info.state == Dispatched return ( info.state == Dispatched
|| info.state == Ready || info.state == Ready
|| info.state == Fetch || info.state == Fetch
) )
&& hasValidDataPointer() && hasValidDataPointer()
; ;
} }
// Member functions:4 ends here // Member functions:4 ends here
// [[file:../../atrip.org::*Member functions][Member functions:5]] // [[file:../../atrip.org::*Member functions][Member functions:5]]
inline bool hasValidDataPointer() const noexcept { inline bool hasValidDataPointer() const noexcept {
return data != nullptr return data != nullptr
&& info.state != Acceptor && info.state != Acceptor
&& info.type != Blank && info.type != Blank
; ;
} }
// Member functions:5 ends here // Member functions:5 ends here
// [[file:../../atrip.org::*Member functions][Member functions:6]] // [[file:../../atrip.org::*Member functions][Member functions:6]]
void unwrapAndMarkReady() { void unwrapAndMarkReady() {
if (info.state == Ready) return; if (info.state == Ready) return;
if (info.state != Dispatched) if (info.state != Dispatched)
throw throw
@ -475,14 +475,14 @@ void unwrapAndMarkReady() {
// Member functions:6 ends here // Member functions:6 ends here
// [[file:../../atrip.org::*Epilog][Epilog:1]] // [[file:../../atrip.org::*Epilog][Epilog:1]]
Slice(size_t size_) Slice(size_t size_)
: info({}) : info({})
, data(nullptr) , data(nullptr)
, size(size_) , size(size_)
{} {}
}; // struct Slice }; // struct Slice
// Epilog:1 ends here // Epilog:1 ends here
// [[file:../../atrip.org::*Debug][Debug:1]] // [[file:../../atrip.org::*Debug][Debug:1]]

44
include/atrip/Tuples.hpp
View File

@ -467,31 +467,31 @@ std::vector<ABCTuple> main(MPI_Comm universe, size_t Nv) {
// Main:1 ends here // Main:1 ends here
// [[file:../../atrip.org::*Main][Main:2]] // [[file:../../atrip.org::*Main][Main:2]]
size_t const size_t const
tuplesPerRankLocal tuplesPerRankLocal
= nodeTuples.size() / nodeInfos[rank].ranksPerNode = nodeTuples.size() / nodeInfos[rank].ranksPerNode
+ size_t(nodeTuples.size() % nodeInfos[rank].ranksPerNode != 0) + size_t(nodeTuples.size() % nodeInfos[rank].ranksPerNode != 0)
; ;
size_t tuplesPerRankGlobal; size_t tuplesPerRankGlobal;
MPI_Reduce(&tuplesPerRankLocal, MPI_Reduce(&tuplesPerRankLocal,
&tuplesPerRankGlobal, &tuplesPerRankGlobal,
1, 1,
MPI_UINT64_T, MPI_UINT64_T,
MPI_MAX, MPI_MAX,
0, 0,
universe); universe);
MPI_Bcast(&tuplesPerRankGlobal, MPI_Bcast(&tuplesPerRankGlobal,
1, 1,
MPI_UINT64_T, MPI_UINT64_T,
0, 0,
universe); universe);
LOG(1,"Atrip") << "Tuples per rank: " << tuplesPerRankGlobal << "\n"; LOG(1,"Atrip") << "Tuples per rank: " << tuplesPerRankGlobal << "\n";
LOG(1,"Atrip") << "ranks per node " << nodeInfos[rank].ranksPerNode << "\n"; LOG(1,"Atrip") << "ranks per node " << nodeInfos[rank].ranksPerNode << "\n";
LOG(1,"Atrip") << "#nodes " << nNodes << "\n"; LOG(1,"Atrip") << "#nodes " << nNodes << "\n";
// Main:2 ends here // Main:2 ends here
// [[file:../../atrip.org::*Main][Main:3]] // [[file:../../atrip.org::*Main][Main:3]]
@ -531,7 +531,7 @@ if (computeDistribution) {
// Main:4 ends here // Main:4 ends here
// [[file:../../atrip.org::*Main][Main:5]] // [[file:../../atrip.org::*Main][Main:5]]
return result; return result;
} }
// Main:5 ends here // Main:5 ends here

2
include/atrip/Utils.hpp
View File

@ -33,7 +33,7 @@ namespace atrip {
// Prolog:1 ends here // Prolog:1 ends here
// [[file:../../atrip.org::*Pretty printing][Pretty printing:1]] // [[file:../../atrip.org::*Pretty printing][Pretty printing:1]]
template <typename T> template <typename T>
std::string pretty_print(T&& value) { std::string pretty_print(T&& value) {
std::stringstream stream; std::stringstream stream;
#if ATRIP_DEBUG > 2 #if ATRIP_DEBUG > 2

20
src/atrip/Atrip.cxx
View File

@ -96,14 +96,6 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
MPI_Comm_size(child_comm, &child_size); MPI_Comm_size(child_comm, &child_size);
} }
// BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
WITH_CHRONO("nv-slices",
LOG(0,"Atrip") << "BUILD NV-SLICES\n";
TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
HHHA<F> hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
)
// BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1 // BUILD SLICES PARAMETRIZED BY NV x NV =============================={{{1
WITH_CHRONO("nv-nv-slices", WITH_CHRONO("nv-nv-slices",
LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n"; LOG(0,"Atrip") << "BUILD NV x NV-SLICES\n";
@ -112,6 +104,18 @@ Atrip::Output Atrip::run(Atrip::Input<F> const& in) {
TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe); TABHH<F> tabhh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
) )
// delete the Vppph so that we don't have a HWM situation for the NV slices
if (in.deleteVppph) {
delete in.Vppph;
}
// BUILD SLICES PARAMETRIZED BY NV ==================================={{{1
WITH_CHRONO("nv-slices",
LOG(0,"Atrip") << "BUILD NV-SLICES\n";
TAPHH<F> taphh(*in.Tpphh, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
HHHA<F> hhha(*in.Vhhhp, (size_t)No, (size_t)Nv, (size_t)np, child_comm, universe);
)
// all tensors // all tensors
std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh}; std::vector< SliceUnion<F>* > unions = {&taphh, &hhha, &abph, &abhh, &tabhh};