From candel at itp.phys.ethz.ch Sun Apr 6 17:00:33 2003 From: candel at itp.phys.ethz.ch (Arno Candel) Date: Sun, 06 Apr 2003 16:00:33 -0100 Subject: Remote access of distributed multipatched arrays Message-ID: <3E905D31.8010104@itp.phys.ethz.ch> Hi everybody, I encountered a problem doing local if-statements using multiple values of several distributed multipatched brick arrays to calculate the value of one such array. Below I have included a simple test program which shows the relevant problem. Three 3D distributed multipatched brick arrays are created and allocated on different domains and thus on different patches. Two arrays are now used to calculate the values of the other array, using local if-statements. Both a "stupid serial" and a "parallel" version of the calculation part are implemented. The use of data-parallel expression or stencils was not possible, in my view. The "stupid serial" version runs flawlessly, but takes far too long on parallel execution, as every node does all the work. Unfortunately, the "parallel" version crashes when executed in parallel (mpirun -np 2 Test -mpi) as remote access to values of distributed arrays seems forbidden. Due to the different domains of the multiple arrays, they are differently stored on the contexts and simultaneous local access is not possible everywhere. Now, I would expect that there is a way to allow remote access via communication in a loop over local patches. This would still scale much better than the "stupid serial" version. I would greatly appreciate your suggestions and comments. Arno Candel simple test program: ********************************************************************** #include "Pooma/Pooma.h" #include "Pooma/Particles.h" #include "Pooma/BrickArrays.h" #include "Pooma/Arrays.h" #include "Utilities/Clock.h" #include "Layout/GridLayout.h" struct FillB { double operator()(int i, int j, int k) const { return i*sin(i*j/2.0)/(k+1); } }; struct FillC { double operator()(int i, int j, int k) const { return k*cos(3.0*j*k)/(i+1); } }; int main(int argc, char* argv[]) { Pooma::initialize(argc,argv); Pooma::Clock clock; double time; Inform pout("Test"); Loc<3> ex=Loc<3>(1,0,0); Loc<3> ez=Loc<3>(0,0,1); int MX=3,MY=4,MZ=7; // dimensions of calculational domain Loc<3> blocks(1,1,Pooma::contexts()); // parallelization along z-axis GuardLayers<3> intguards(1), extguards(0); // one internal guard layer, no external guard layer GridPartition<3> Part=GridPartition<3>(blocks,intguards,extguards); // Create three distributed MultiPatch brick arrays // with differing domains and thus different distributions among contexts Array<3, double, MultiPatch > > A, B, C; Interval<3> ADom=Interval<3>(Interval<1>(1,MX), Interval<1>(1,MY), Interval<1>(1,MZ)); Interval<3> BDom=Interval<3>(Interval<1>(0,MX), Interval<1>(1,MY), Interval<1>(0,MZ+1)); Interval<3> CDom=Interval<3>(Interval<1>(0,MX), Interval<1>(1,MY), Interval<1>(1,2*MZ)); GridLayout<3> Alayout = GridLayout<3>(ADom,Part,DistributedTag()); GridLayout<3> Blayout = GridLayout<3>(BDom,Part,DistributedTag()); GridLayout<3> Clayout = GridLayout<3>(CDom,Part,DistributedTag()); A.initialize(Alayout); B.initialize(Blayout); C.initialize(Clayout); A=1; B=Array<3, double, IndexFunction >(); C=Array<3, double, IndexFunction >(); // "stupid serial" version Pooma::blockAndEvaluate(); time=clock.value(); for (int i=ADom[0].first();i<=ADom[0].last();i++) for (int j=ADom[1].first();j<=ADom[1].last();j++) for (int k=ADom[2].first();k<=ADom[2].last();k++) { Loc<3> x(i,j,k); if ( (B(x+ez)<0.5) && (C(x-ex)>0.3) ) { A(x)=B(x-ez-ex)+C(x+ez); } } Pooma::blockAndEvaluate(); pout << A << "\nstupid serial version took " << clock.value()-time << " secs" << std::endl; // "parallel" version, iterate only over local patches of A Pooma::blockAndEvaluate(); time=clock.value(); for (GridLayout<3>::const_iterator it = Alayout.beginLocal(); it != Alayout.endLocal(); it++) { for (int i=it->domain()[0].first();i<=it->domain()[0].last();i++) for (int j=it->domain()[1].first();j<=it->domain()[1].last();j++) for (int k=it->domain()[2].first();k<=it->domain()[2].last();k++) { Loc<3> x(i,j,k); if ( (B(x+ez)<0.5) && (C(x-ex)>0.3) ) { A(x)=B(x-ez-ex)+C(x+ez); // problem: local patches of A might not contain needed values of B and C and remote access is forbidden! // Will crash in parallel execution! } } } Pooma::blockAndEvaluate(); pout << A << "\nparallel version took " << clock.value()-time << " secs" << std::endl; Pooma::finalize(); return 0; } From rguenth at tat.physik.uni-tuebingen.de Sun Apr 6 18:31:58 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Sun, 6 Apr 2003 20:31:58 +0200 (CEST) Subject: [pooma-dev] Remote access of distributed multipatched arrays In-Reply-To: <3E905D31.8010104@itp.phys.ethz.ch> Message-ID: On Sun, 6 Apr 2003, Arno Candel wrote: > Hi everybody, > > I encountered a problem doing local if-statements using multiple values > of several distributed multipatched brick arrays to calculate the value > of one such array. > > Below I have included a simple test program which shows the relevant > problem. Three 3D distributed multipatched brick arrays are created and > allocated on different domains and thus on different patches. Two arrays > are now used to calculate the values of the other array, using local > if-statements. Both a "stupid serial" and a "parallel" version of the > calculation part are implemented. The use of data-parallel expression or > stencils was not possible, in my view. This is not the case. > // "stupid serial" version > Pooma::blockAndEvaluate(); > time=clock.value(); > for (int i=ADom[0].first();i<=ADom[0].last();i++) > for (int j=ADom[1].first();j<=ADom[1].last();j++) > for (int k=ADom[2].first();k<=ADom[2].last();k++) { > Loc<3> x(i,j,k); > if ( (B(x+ez)<0.5) && (C(x-ex)>0.3) ) { > A(x)=B(x-ez-ex)+C(x+ez); > } > } > Pooma::blockAndEvaluate(); > pout << A << "\nstupid serial version took " << clock.value()-time > << " secs" << std::endl; Just use A(Adom) = where(B(Adom+ez)<0.5 && C(Adom-ex)>0.3, B(Adom-ez-ex) + C(Adom+ez)); here - also notice that setting external guards to zero will not work, as you need at least as much external guards as internal ones (usually). > > // "parallel" version, iterate only over local patches of A > Pooma::blockAndEvaluate(); > time=clock.value(); > for (GridLayout<3>::const_iterator it = Alayout.beginLocal(); it != > Alayout.endLocal(); it++) { > for (int i=it->domain()[0].first();i<=it->domain()[0].last();i++) > for (int j=it->domain()[1].first();j<=it->domain()[1].last();j++) > for (int k=it->domain()[2].first();k<=it->domain()[2].last();k++) { > Loc<3> x(i,j,k); > if ( (B(x+ez)<0.5) && (C(x-ex)>0.3) ) { > A(x)=B(x-ez-ex)+C(x+ez); > // problem: local patches of A might not contain needed > values of B and C and remote access is forbidden! > // Will crash in parallel execution! remote access is not forbidden. But due to the guard layer setup (see above) you may experience crashes. Otherwise this should work fine. Hope this helps. Richard. From rguenth at tat.physik.uni-tuebingen.de Tue Apr 22 15:57:58 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Tue, 22 Apr 2003 17:57:58 +0200 (CEST) Subject: [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnCopy() Message-ID: Hi! Currently for a multipatch engine makeOwnCopy() does no good (it doesnt copy), it seems because just calling makeOwnCopy() on the RefCountedBlockPtr does not copy the underlying engines. I didnt figure out why and so came up with the following fix which works for me. I have also an updated Field/BasicTest3.cpp which now checks every engine combination for makeOwnCopy and it now succeeds for both MultiPatch and MultiPatch >. Ok to apply? Richard. 2003 Apr 22 Richard Guenther * Engine/MultiPatchEngine.cpp: do makeOwnCopy() manually. ===== MultiPatchEngine.cpp 1.3 vs edited ===== --- 1.3/r2/src/Engine/MultiPatchEngine.cpp Fri Jan 24 10:35:52 2003 +++ edited/MultiPatchEngine.cpp Tue Apr 22 17:42:04 2003 @@ -244,8 +244,17 @@ Engine >:: makeOwnCopy() { - data_m.makeOwnCopy(); - + PAssert(data_m.isValid()); + if (data_m.isShared()) { + PatchContainer_t model = data_m; + data_m = PatchContainer_t(model.size()); + for (int i=0; i >::makeOwnC opy() Message-ID: Hmmm. I don't have time to play with this, but something looks wrong. The sequence data_m.makeOwnCopy(); if (data_m.isShared()) the isShared() test should always return false after a call to makeOwnCopy - that should be an invariant property of any ref-counted thing. That was certainly my intention. Jim ------------------------------------------------------------------------ James A. Crotinger email: jimc at numerix.com NumeriX, LLC phone: (505) 424-4477 x104 2960 Rodeo Park Dr. W. Santa Fe, NM 87505 -----Original Message----- From: Richard Guenther [mailto:rguenth at tat.physik.uni-tuebingen.de] Sent: Tuesday, April 22, 2003 9:58 AM To: pooma-dev at pooma.codesourcery.com Cc: Jeffrey D. Oldham Subject: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnCopy() Hi! Currently for a multipatch engine makeOwnCopy() does no good (it doesnt copy), it seems because just calling makeOwnCopy() on the RefCountedBlockPtr does not copy the underlying engines. I didnt figure out why and so came up with the following fix which works for me. I have also an updated Field/BasicTest3.cpp which now checks every engine combination for makeOwnCopy and it now succeeds for both MultiPatch and MultiPatch >. Ok to apply? Richard. 2003 Apr 22 Richard Guenther * Engine/MultiPatchEngine.cpp: do makeOwnCopy() manually. ===== MultiPatchEngine.cpp 1.3 vs edited ===== --- 1.3/r2/src/Engine/MultiPatchEngine.cpp Fri Jan 24 10:35:52 2003 +++ edited/MultiPatchEngine.cpp Tue Apr 22 17:42:04 2003 @@ -244,8 +244,17 @@ Engine >:: makeOwnCopy() { - data_m.makeOwnCopy(); - + PAssert(data_m.isValid()); + if (data_m.isShared()) { + PatchContainer_t model = data_m; + data_m = PatchContainer_t(model.size()); + for (int i=0; i From rguenth at tat.physik.uni-tuebingen.de Tue Apr 22 17:51:22 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Tue, 22 Apr 2003 19:51:22 +0200 (CEST) Subject: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnC opy() In-Reply-To: Message-ID: On Tue, 22 Apr 2003, James Crotinger wrote: > Hmmm. I don't have time to play with this, but something looks wrong. The > sequence > > data_m.makeOwnCopy(); > if (data_m.isShared()) Well ;) I attached a unified patch, the data_m.makeOwnCopy() is removed, the rest added. The problem is, the data_m.makeOwnCopy() does not work. Richard. > 2003 Apr 22 Richard Guenther > > * Engine/MultiPatchEngine.cpp: do makeOwnCopy() manually. > > ===== MultiPatchEngine.cpp 1.3 vs edited ===== > --- 1.3/r2/src/Engine/MultiPatchEngine.cpp Fri Jan 24 10:35:52 2003 > +++ edited/MultiPatchEngine.cpp Tue Apr 22 17:42:04 2003 > @@ -244,8 +244,17 @@ > Engine >:: > makeOwnCopy() > { > - data_m.makeOwnCopy(); > - > + PAssert(data_m.isValid()); > + if (data_m.isShared()) { > + PatchContainer_t model = data_m; > + data_m = PatchContainer_t(model.size()); > + for (int i=0; i + data_m[i] = model[i]; > + data_m[i].makeOwnCopy(); > + } > + pDirty_m = new bool(*pDirty_m); > + } > + > return *this; > } > > From rguenth at tat.physik.uni-tuebingen.de Tue Apr 22 17:56:15 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Tue, 22 Apr 2003 19:56:15 +0200 (CEST) Subject: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnC opy() In-Reply-To: Message-ID: On Tue, 22 Apr 2003, James Crotinger wrote: > Hmmm. I don't have time to play with this, but something looks wrong. Btw., the failing testcase is the following (which is fixed with my patch), failing tester.check()s marked with XXXX: #include "Pooma/Fields.h" #include "Pooma/Tiny.h" #include "Utilities/Tester.h" #include #include template void check(Pooma::Tester& tester, Field& f) { typedef Field Field_t; f.all() = 2; // make a deep copy Field_t h(f); h.makeOwnCopy(); // write to copy and check if it did not clobber f h.all() = 1; tester.check("f != h", all(f.all() == 2) && all(h.all() == 1)); // XXXX } //----------------------------------------------------------------------------- // Main program: //----------------------------------------------------------------------------- int main(int argc, char *argv[]) { Pooma::initialize(argc, argv); Pooma::Tester tester(argc, argv); // Create the physical domain. const int NX = 5, NY = 5; Interval<1> I(NX), J(NY); Vector<2,double> origin; Vector<2,double> spacings; for (int d = 0; d < 2; d++) { origin(d) = d; spacings(d) = d + 1; } Centering<2> vert = canonicalCentering<2>(VertexType, Continuous); // MultiPatch tester.out() << "MultiPatch...\n"; { GridLayout<2> layout1(Interval<2>(I, J), Loc<2>(1, 1), GuardLayers<2>(1), DistributedTag()); Field, int, MultiPatch > f(vert, layout1, origin, spacings); check(tester, f); } // MultiPatch > tester.out() << "MultiPatch >...\n"; { GridLayout<2> layout1(Interval<2>(I, J), Loc<2>(1, 1), GuardLayers<2>(1), DistributedTag()); Field, int, MultiPatch > > f(vert, layout1, origin, spacings); check(tester, f); } int ret = tester.results("BasicTest3"); Pooma::finalize(); return ret; } From jcrotinger at proximation.com Tue Apr 22 18:23:27 2003 From: jcrotinger at proximation.com (James Crotinger) Date: Tue, 22 Apr 2003 12:23:27 -0600 Subject: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnC opy() Message-ID: Ah - gotcha. I missed the '-'. The right thing is to fix makeOwnCopy, of course. This looks like a traits problem - the MultiPatchEngine::PatchContainer_t is a RCBPtr of engines, but the underlying engines have shallow copy semantics. There is a traits class that marks shallow-copy objects appropriately and that can be used to implement copy correctly. In ElementProperties.h there is an ElementProperties base class called MakeOwnCopyProperties that is set up to handle this - what there should be, somewhere, is a specialization: template template <> struct ElementProperties > : public MakeOwnCopyProperties > { }; What puzzles me is that I see this specialization (a partial one) for Brick and several others. So this ought to work for MultiPatch and any of the other engines that have done this specialization. (It will not work on views - we went around on this and ultimately decided that doing makeOwnCopy on a view should not be allowed) Hmmm. I don't see a specialization for Remote. (Just saw your example.) Is this failing for both Brick and Remote? The lack of a ElementProperties specialization for Remote is a bug. Any class that has shallow copy semantics and that implements the makeOwnCopy() function should specialize ElementProperties as above. At any rate, it would probably be good to step through the old code and see where it is ending up in the wrong specialization of ElementProperties. That's a bug that could bite in other places. Looks like spotted another bug wrt the dirty flag - that deep copy is needed regardless of how the engines are copied. Jim ------------------------------------------------------------------------ James A. Crotinger email: jimc at numerix.com NumeriX, LLC phone: (505) 424-4477 x104 2960 Rodeo Park Dr. W. Santa Fe, NM 87505 -----Original Message----- From: Richard Guenther [mailto:rguenth at tat.physik.uni-tuebingen.de] Sent: Tuesday, April 22, 2003 11:51 AM To: James Crotinger Cc: 'pooma-dev at pooma.codesourcery.com'; 'Jeffrey D. Oldham' Subject: RE: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnC opy() On Tue, 22 Apr 2003, James Crotinger wrote: > Hmmm. I don't have time to play with this, but something looks wrong. The > sequence > > data_m.makeOwnCopy(); > if (data_m.isShared()) Well ;) I attached a unified patch, the data_m.makeOwnCopy() is removed, the rest added. The problem is, the data_m.makeOwnCopy() does not work. Richard. > 2003 Apr 22 Richard Guenther > > * Engine/MultiPatchEngine.cpp: do makeOwnCopy() manually. > > ===== MultiPatchEngine.cpp 1.3 vs edited ===== > --- 1.3/r2/src/Engine/MultiPatchEngine.cpp Fri Jan 24 10:35:52 2003 > +++ edited/MultiPatchEngine.cpp Tue Apr 22 17:42:04 2003 > @@ -244,8 +244,17 @@ > Engine >:: > makeOwnCopy() > { > - data_m.makeOwnCopy(); > - > + PAssert(data_m.isValid()); > + if (data_m.isShared()) { > + PatchContainer_t model = data_m; > + data_m = PatchContainer_t(model.size()); > + for (int i=0; i + data_m[i] = model[i]; > + data_m[i].makeOwnCopy(); > + } > + pDirty_m = new bool(*pDirty_m); > + } > + > return *this; > } > > -------------- next part -------------- An HTML attachment was scrubbed... URL: From rguenth at tat.physik.uni-tuebingen.de Tue Apr 22 18:52:14 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Tue, 22 Apr 2003 20:52:14 +0200 (CEST) Subject: [pooma-dev] [PATCH] Fix Engine<.., MultiPatch<..> >::makeOwnC opy() In-Reply-To: Message-ID: On Tue, 22 Apr 2003, James Crotinger wrote: > Ah - gotcha. I missed the '-'. > > The right thing is to fix makeOwnCopy, of course. This looks like a traits > problem - the MultiPatchEngine::PatchContainer_t is a RCBPtr of engines, but > the underlying engines have shallow copy semantics. There is a traits class > that marks shallow-copy objects appropriately and that can be used to > implement copy correctly. In ElementProperties.h there is an > ElementProperties base class called MakeOwnCopyProperties that is set up to > handle this - what there should be, somewhere, is a specialization: > > template > template <> > struct ElementProperties > > : public MakeOwnCopyProperties > > { }; > > What puzzles me is that I see this specialization (a partial one) for Brick > and several others. So this ought to work for MultiPatch and any of > the other engines that have done this specialization. (It will not work on > views - we went around on this and ultimately decided that doing makeOwnCopy > on a view should not be allowed) > > Hmmm. I don't see a specialization for Remote. (Just saw your > example.) Is this failing for both Brick and Remote? The lack of a It fails for multipatch engines, both MultiPatch and MultiPatch >, it works for Remote. > ElementProperties specialization for Remote is a bug. Any class that > has shallow copy semantics and that implements the makeOwnCopy() function > should specialize ElementProperties as above. The problem is, the specialization for ElementProperties for the RCB is somehow not working correctly. I dont have time to track this down, so I went the route used in FieldEngine::makeOwnCopy(), do it all manually, and this solved my problem. > At any rate, it would probably be good to step through the old code and see > where it is ending up in the wrong specialization of ElementProperties. > That's a bug that could bite in other places. > > Looks like spotted another bug wrt the dirty flag - that deep copy is needed > regardless of how the engines are copied. Yes, indeed. I'll leave it to Jeffrey, wether to apply this patch or leave the bug in. Thanks, Richard. From gtalbot at locuspharma.com Wed Apr 23 20:10:35 2003 From: gtalbot at locuspharma.com (George Talbot) Date: 23 Apr 2003 16:10:35 -0400 Subject: GCC 3.2 on Redhat 8.0--a fair number of compiler warnings? Message-ID: <24752b2dfd3ec609e6c9582a2ede3a5b3ea6f32a@locuspharma.com> Hi, New POOMA II user. Got everything to compile and link today. The performance of this stuff is shocking. I ran a hand coded version of your 2D diffusion program on my box for 250 steps, 800x800 array. 1m34s, uniprocessor. Using a stencil, same box, uniprocessor. 25s. WOW. Using -shmem -np 2 to distribute the shared array version on both of my processors on my box. 12s. Truly amazing. I think this may be a dramatically useful tool. The only thing that irks me a bit is that there are a fair number of compiler warnings, which I will append, along with the source I used, to this e-mail. Is this normal? Swimming in a sea of warnings from the POOMA headers, I might miss some on my own code... Take care. -- George T. Talbot Here's the program: #include using namespace std; #define STENCIL #define POOMA #define DISTRIBUTED #ifdef POOMA #include "Pooma/Arrays.h" #endif #ifdef STENCIL class NinePtDiffusion { public: template inline typename C::Element_t operator()(const C& c, int I, int J) const { return (1.0/9.0) * (c.read(I+1, J+1) + c.read(I+1, J ) + c.read(I+1, J-1) + c.read(I , J+1) + c.read(I , J ) + c.read(I , J-1) + c.read(I-1, J+1) + c.read(I-1, J ) + c.read(I-1, J-1)); } inline int lowerExtent(int) const { return 1; } inline int upperExtent(int) const { return 1; } }; #endif int main(int argc, char* argv[]) { #ifdef POOMA Pooma::initialize(argc, argv); #endif long numAveragings = 250, numSteps; numSteps = (numAveragings+1)/2; long n = 800; long numProcessors=2; #ifdef POOMA Inform output; output << "processors=" << numProcessors << " avgs=" << numAveragings << " n=" << n << endl; #else cout << "processors=" << numProcessors << " avgs=" << numAveragings << " n=" << n << endl; #endif #ifdef POOMA Interval<1> N(0, n-1); Interval<2> problemDomain(N, N); Interval<1> I(1, n-2); Interval<1> J(1, n-2); Interval<2> interiorDomain(I, I); #ifdef DISTRIBUTED UniformGridPartition<2> partition(Loc<2>(numProcessors,numProcessors), GuardLayers<2>(1), GuardLayers<2>(0)); UniformGridLayout<2> layout(problemDomain, partition, DistributedTag()); Array<2, double, MultiPatch > > a(layout); Array<2, double, MultiPatch > > b(layout); #else Array<2, double, Brick> a(problemDomain); Array<2, double, Brick> b(problemDomain); #endif a = b = 0.0; Pooma::blockAndEvaluate(); b(n/2, n/2) = 1000.0; #ifdef STENCIL Stencil diffusion; #endif #else double a[n][n], b[n][n]; for (long x=1; x::Grid(const Grid<1>&)': /usr/local/cots/pooma/src/Domain/Grid.h:360: warning: base class `class Domain<1, DomainTraits > >' should be explicitly initialized in the copy constructor In file included from /usr/local/cots/pooma/src/Pooma/Arrays.h:46, from diffusion_test.cpp:9: /usr/local/cots/pooma/src/Engine/RemoteEngine.h: In copy constructor `GatherContexts::GatherContextsData::GatherContextsData(const GatherContexts::GatherContextsData&)': /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1697: warning: base class ` class RefCounted' should be explicitly initialized in the copy constructor diffusion_test.cpp: In function `int main(int, char**)': diffusion_test.cpp:100: warning: unused variable `const double weight' In file included from /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1499, from /usr/local/cots/pooma/src/Array/Array.h:108, from diffusion_test.cpp:77: /usr/local/cots/pooma/src/Layout/UniformGridLayout.h: In copy constructor `UniformGridLayout::UniformGridLayout(const UniformGridLayout&) [with int Dim = 2]': /usr/local/cots/pooma/src/Layout/LayoutBase.h:806: instantiated from `LayoutBaseViewData::LayoutBaseViewData(const L&, const Domain&) [with DT = DomainTraits >, int Dim = 2, int Dim2 = 2, L = UniformGridLayout<2>]' /usr/local/cots/pooma/src/Layout/LayoutBase.h:1368: instantiated from `UniformGridLayoutViewData::UniformGridLayoutViewData(const UniformGridLayout&, const Domain&) [with DT = DomainTraits >, int Dim = 2, int Dim2 = 2]' /usr/local/cots/pooma/src/Layout/LayoutBase.h:1109: instantiated from `LayoutBaseView::LayoutBaseView(lvd*) [with int Dim = 2, int Dim2 = 2, lvd = UniformGridLayoutViewData<2, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1499: instantiated from `UniformGridLayoutView::UniformGridLayoutView(UniformGridLayoutViewData::Layout_t&, const Domain&) [with DT = DomainTraits >, int Dim = 2, int Dim2 = 2]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:992: instantiated from `Engine >::Engine(const Engine >&, const Domain&) [with DT = DomainTraits >, int Dim = 2, T = double, LayoutTag = UniformTag, PatchTag = Remote, int Dim2 = 2]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1457: instantiated from `Array::Array(const Engine&, const Initializer&) [with int Dim2 = 2, T2 = double, EngineTag2 = MultiPatch >, Initializer = Interval<2>, int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:676: instantiated from `static Array >::newDim, T, NewEngine::Engine_t, Array::Domain_t>::Type_t::Tag_t> View0 >::make(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1910: instantiated from `AltView0 >::Type_t Array::operator()() const [with int Dim = 2, T = double, EngineTag = MultiPatch >]'/usr/local/cots/pooma/src/Engine/RemoteEngine.h:1909: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:108: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatch > >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2504: instantiated from `const Array& assign(const Array&, const T1&, const Op&) [with int Dim = 2, T = double, EngineTag = MultiPatch >, T1 = double, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2117: instantiated from `const Array& Array::operator=(const T1&) const [with T1 = double, int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Array/Array.h:2105: instantiated from `Array& Array::operator=(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatch >]' diffusion_test.cpp:77: instantiated from here /usr/local/cots/pooma/src/Layout/UniformGridLayout.h:1831: warning: base class `class Observer >' should be explicitly initialized in the copy constructor In file included from /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1519, from diffusion_test.cpp:108: /usr/local/cots/pooma/src/Layout/LayoutBase.h: In constructor `LayoutBaseViewData::LayoutBaseViewData(const L&, const LV&, const ViewIndexer&, const Domain&, GuardLayers, GuardLayers) [with DT = DomainTraits >, LV = UniformGridLayoutView<2, 2>, int Dim = 2, int Dim2 = 2, L = UniformGridLayout<2>]': /usr/local/cots/pooma/src/Layout/LayoutBase.h:1388: instantiated from `UniformGridLayoutViewData::UniformGridLayoutViewData(const UniformGridLayoutView&, const Domain&) [with DT = DomainTraits >, int Dim = 2, int Dim2 = 2]' /usr/local/cots/pooma/src/Layout/LayoutBase.h:1109: instantiated from `LayoutBaseView::LayoutBaseView(lvd*) [with int Dim = 2, int Dim2 = 2, lvd = UniformGridLayoutViewData<2, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1519: instantiated from `UniformGridLayoutView::UniformGridLayoutView(const UniformGridLayoutView&, const Domain&) [with DT = DomainTraits >, int Dim = 2, int Dim2 = 2]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1012: instantiated from `Engine >::Engine(const Engine >&, const Domain&) [with DT = DomainTraits >, int Dim = 2, T = double, LayoutTag = UniformTag, PatchTag = Remote, int Dim2 = 2]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1457: instantiated from `Array::Array(const Engine&, const Initializer&) [with int Dim2 = 2, T2 = double, EngineTag2 = MultiPatchView, 2>, Initializer = Interval<2>, int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:676: instantiated from `static Array >::newDim, T, NewEngine::Engine_t, Array::Domain_t>::Type_t::Tag_t> View0 >::make(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1910: instantiated from `AltView0 >::Type_t Array::operator()() const [with int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1909: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, StencilEngine, 2> > > >, Op = OpAssign]' diffusion_test.cpp:108: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, StencilEngine, 2> > > >, Op = OpAssign]' diffusion_test.cpp:2486: instantiated from `const Array& assign(const Array&, const Array&, const Op&) [with int Dim = 2, T = double, EngineTag = MultiPatchView, 2>, int OtherDim = 2, OtherT = double, OtherEngineTag = StencilEngine, 2> > >, Op = OpAssign]' diffusion_test.cpp:2117: instantiated from `const Array& Array::operator=(const T1&) const [with T1 = Array<2, double, StencilEngine, 2> > > >, int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' diffusion_test.cpp:105: instantiated from here /usr/local/cots/pooma/src/Layout/LayoutBase.h:1027: warning: member initializers for `ViewIndexer<2, 2> LayoutBaseViewData<2, 2, UniformGridLayout<2> >::indexer_m' /usr/local/cots/pooma/src/Layout/LayoutBase.h:1010: warning: and `long int LayoutBaseViewData<2, 2, UniformGridLayout<2> >::id_m' /usr/local/cots/pooma/src/Layout/LayoutBase.h:871: warning: will be re-ordered to match declaration order /usr/local/cots/pooma/src/Layout/LayoutBase.h:1040: warning: member initializers for `bool LayoutBaseViewData<2, 2, UniformGridLayout<2> >::subdomainsComputed_m' /usr/local/cots/pooma/src/Layout/LayoutBase.h:1021: warning: and ` GuardLayers<2> LayoutBaseViewData<2, 2, UniformGridLayout<2> >::internalGuards_m' /usr/local/cots/pooma/src/Layout/LayoutBase.h:871: warning: will be re-ordered to match declaration order /usr/local/cots/pooma/src/Engine/Stencil.h: In constructor `StencilIntersector::StencilIntersector(const Interval&, const Intersect&) [with int Dim = 2, Intersect = Intersector<2>]': diffusion_test.cpp:863: instantiated from `static int LeafFunctor >, ExpressionApply > >::apply(const Engine >&, const ExpressionApply >&) [with int D = 2, T = double, S = NinePtDiffusion, E = Array<2, double, MultiPatchView, 2> >, Intersect = Intersector<2>]' diffusion_test.cpp:2307: instantiated from `static int LeafFunctor, ExpressionApply >::apply(const Array&, const ExpressionApply&) [with int Dim = 2, T = double, E = StencilEngine, 2> > >, Tag = IntersectorTag >]' diffusion_test.cpp:80: instantiated from `static LeafFunctor::Type_t ForEach::apply(const Expr&, const FTag&, const CTag&) [with Expr = Array<2, double, StencilEngine, 2> > > >, FTag = ExpressionApply > >, CTag = NullCombine]' diffusion_test.cpp:88: instantiated from `ForEach::Type_t forEach(const Expr&, const FTag&, const CTag&) [with Expr = Array<2, double, StencilEngine, 2> > > >, FTag = ExpressionApply > >, CTag = NullCombine]' diffusion_test.cpp:276: instantiated from `void expressionApply(const A&, const Tag&) [with A = Array<2, double, StencilEngine, 2> > > >, Tag = IntersectorTag >]' diffusion_test.cpp:1914: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, StencilEngine, 2> > > >, Op = OpAssign]' diffusion_test.cpp:108: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, StencilEngine, 2> > > >, Op = OpAssign]' diffusion_test.cpp:2486: instantiated from `const Array& assign(const Array&, const Array&, const Op&) [with int Dim = 2, T = double, EngineTag = MultiPatchView, 2>, int OtherDim = 2, OtherT = double, OtherEngineTag = StencilEngine, 2> > >, Op = OpAssign]' diffusion_test.cpp:2117: instantiated from `const Array& Array::operator=(const T1&) const [with T1 = Array<2, double, StencilEngine, 2> > > >, int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' diffusion_test.cpp:105: instantiated from here /usr/local/cots/pooma/src/Engine/Stencil.h:844: warning: member initializers for `Intersector<2> StencilIntersector<2, Intersector<2> >::intersector_m' /usr/local/cots/pooma/src/Engine/Stencil.h:843: warning: and `Interval<2> StencilIntersector<2, Intersector<2> >::domain_m' /usr/local/cots/pooma/src/Engine/Stencil.h:788: warning: will be re-ordered to match declaration order /usr/local/cots/pooma/src/Domain/DomainTraits.Interval.h: In static member function `static void DomainTraits >::setDomain(int (&)[2], const T1&, const T2&) [with T1 = int, T2 = unsigned int]': /usr/local/cots/pooma/src/Domain/Interval.h:369: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Interval.h:264: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Domain/DomainTraits.Interval.h: In static member function `static void DomainTraits >::setDomain(int (&)[2], const T1&, const T2&) [with T1 = int, T2 = long unsigned int]': /usr/local/cots/pooma/src/Domain/Interval.h:377: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Interval.h:264: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Domain/DomainTraits.Range.h: In static member function `static void DomainTraits >::setDomain(int (&)[3], const T1&, const T2&) [with T1 = int, T2 = unsigned int]': /usr/local/cots/pooma/src/Domain/Range.h:396: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Range.h:239: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Domain/DomainTraits.Range.h: In static member function `static void DomainTraits >::setDomain(int (&)[3], const T1&, const T2&) [with T1 = int, T2 = long unsigned int]': /usr/local/cots/pooma/src/Domain/Range.h:407: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Range.h:239: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Domain/DomainTraits.Grid.h: In static member function `static void DomainTraits >::setDomain(IndirectionList&, const T1&, const T2&) [with T1 = int, T2 = unsigned int]': /usr/local/cots/pooma/src/Domain/Grid.h:395: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Grid.h:245: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Domain/DomainTraits.Grid.h: In static member function `static void DomainTraits >::setDomain(IndirectionList&, const T1&, const T2&) [with T1 = int, T2 = long unsigned int]': /usr/local/cots/pooma/src/Domain/Grid.h:404: instantiated from here /usr/local/cots/pooma/src/Domain/DomainTraits.Grid.h:245: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Engine/MultiPatchEngine.cpp: In copy constructor `Engine >::Engine(const Engine >&) [with int Dim = 2, T = double, LayoutTag = UniformTag, PatchTag = Remote]': /usr/local/cots/pooma/src/Engine/MultiPatchEngine.h:992: instantiated from `Engine >::Engine(const Engine >&, const Domain&) [with DT = DomainTraits >, int Dim = 2, T = double, LayoutTag = UniformTag, PatchTag = Remote, int Dim2 = 2]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1457: instantiated from `Array::Array(const Engine&, const Initializer&) [with int Dim2 = 2, T2 = double, EngineTag2 = MultiPatch >, Initializer = Interval<2>, int Dim = 2, T = double, EngineTag = MultiPatchView, 2>]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:676: instantiated from `static Array >::newDim, T, NewEngine::Engine_t, Array::Domain_t>::Type_t::Tag_t> View0 >::make(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1910: instantiated from `AltView0 >::Type_t Array::operator()() const [with int Dim = 2, T = double, EngineTag = MultiPatch >]'/usr/local/cots/pooma/src/Engine/RemoteEngine.h:1909: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:108: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatch > >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2504: instantiated from `const Array& assign(const Array&, const T1&, const Op&) [with int Dim = 2, T = double, EngineTag = MultiPatch >, T1 = double, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2117: instantiated from `const Array& Array::operator=(const T1&) const [with T1 = double, int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Array/Array.h:2105: instantiated from `Array& Array::operator=(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatch >]' diffusion_test.cpp:77: instantiated from here /usr/local/cots/pooma/src/Engine/MultiPatchEngine.cpp:151: warning: base class `class Observer >' should be explicitly initialized in the copy constructor /usr/local/cots/pooma/src/Layout/UniformGridLayout.h: In member function `int UniformGridLayoutData::globalID(const Loc&) const [with int Dim = 2]': /usr/local/cots/pooma/src/Layout/LayoutBase.h:472: instantiated from `int LayoutBase::globalID(const Loc&) const [with int Dim = 2, LBD = UniformGridLayoutData<2>]' /usr/local/cots/pooma/src/Utilities/RefCountedBlockPtr.h:758: instantiated from `T& RefCountedBlockPtr::operator[](int) const [with T = Engine<2, double, Remote >, bool BoundsChecked = false, Controller = RefBlockController > >]' /usr/local/cots/pooma/src/Engine/RemoteEngine.h:458: instantiated from `RemoteProxy Engine >::operator()(const Loc&) const [with int Dim = 2, T = double, Tag = Brick]' /usr/local/cots/pooma/src/Tulip/RemoteProxy.h:1367: instantiated from `Engine::ElementRef_t Engine >::operator()(const Loc&) const [with int Dim = 2, T = double, LayoutTag = UniformTag, PatchTag = Remote]' /usr/local/cots/pooma/src/Tulip/RemoteProxy.h:166: instantiated from `static Array::ElementRef_t View1Implementation, Domain, true>::make(const Array&, const S1&, const S2&, const Combine&) [with S1 = long int, S2 = long int, Combine = CombineDomainOpt, true>, int Dim = 2, T = double, EngineTag = MultiPatch >, Domain = Loc<2>]' /usr/local/cots/pooma/src/Tulip/RemoteProxy.h:823: instantiated from `static View1Implementation, NewDomain2::SliceType_t, View2, Sub1, Sub2>::sv>::Type_t View2, Sub1, Sub2>::make(const Array&, const Sub1&, const Sub2&) [with int Dim = 2, T = double, EngineTag = MultiPatch >, Sub1 = long int, Sub2 = long int]' /usr/local/cots/pooma/src/Tulip/RemoteProxy.h:1926: instantiated from `View2, Sub1, Sub2>::Type_t Array::operator()(const Sub1&, const Sub2&) const [with Sub1 = long int, Sub2 = long int, int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Tulip/RemoteProxy.h:176: instantiated from `RemoteProxy& RemoteProxy::operator=(const S&) [with S = double, T = double]' diffusion_test.cpp:80: instantiated from here /usr/local/cots/pooma/src/Layout/UniformGridLayout.h:2127: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Layout/UniformGridLayout.h: In member function `int UniformGridLayoutData::touches(const OtherDomain&, OutIter, const ConstructTag&) const [with OtherDomain = Range<2>, OutIter = std::back_insert_iterator, Interval<2> >, std::allocator, Interval<2> > > > >, ConstructTag = TouchesConstructNodeObj, int Dim = 2]': /usr/local/cots/pooma/src/Layout/LayoutBase.h:641: instantiated from `int LayoutBase::touches(const OtherDomain&, OutIter, const ConstructTag&) const [with OtherDomain = Range<2>, OutIter = std::back_insert_iterator, Interval<2> >, std::allocator, Interval<2> > > > >, ConstructTag = TouchesConstructNodeObj, int Dim = 2, LBD = UniformGridLayoutData<2>]' /usr/local/cots/pooma/src/Engine/Intersector.h:703: instantiated from `int LayoutBase::touches(const OtherDomain&, OutIter) const [with OtherDomain = Range<2>, OutIter = std::back_insert_iterator, Interval<2> >, std::allocator, Interval<2> > > > >, int Dim = 2, LBD = UniformGridLayoutData<2>]' /usr/local/cots/pooma/src/Engine/Intersector.h:942: instantiated from `int LayoutBaseViewData::touches(const OtherDomain&, OutIter, const ConstructTag&) const [with OtherDomain = Interval<2>, OutIter = std::back_insert_iterator, std::allocator > > >, ConstructTag = TouchesConstructINode<2>, int Dim = 2, int Dim2 = 2, L = UniformGridLayout<2>]' /usr/local/cots/pooma/src/Engine/Intersector.h:1285: instantiated from `int LayoutBaseView::touches(const OtherDomain&, OutIter, const ConstructTag&) const [with OtherDomain = Interval<2>, OutIter = std::back_insert_iterator, std::allocator > > >, ConstructTag = TouchesConstructINode<2>, int Dim = 2, int Dim2 = 2, lvd = UniformGridLayoutViewData<2, 2>]' /usr/local/cots/pooma/src/Engine/Intersector.h:333: instantiated from `void IntersectorData::touches(const Layout&) [with Layout = UniformGridLayoutView<2, 2>, int Dim = 2]' /usr/local/cots/pooma/src/Engine/Intersector.h:171: instantiated from `bool IntersectorData::intersect(const Engine&, const GuardLayers&) [with Engine = Engine<2, double, MultiPatchView, 2> >, int Dim2 = 2, int Dim = 2]' /usr/local/cots/pooma/src/Engine/Intersector.h:440: instantiated from `bool Intersector::intersect(const Engine&, const GuardLayers&) [with Engine = Engine<2, double, MultiPatchView, 2> >, int Dim2 = 2, int Dim = 2]' /usr/local/cots/pooma/src/Array/Array.h:1729: instantiated from `static int LeafFunctor >, ExpressionApply > >::applyHandler(const Engine >&, const ExpressionApply >&, const WrappedInt<1>&) [with int Dim = 2, T = double, LT = UniformTag, PatchTag = Remote, int BD = 2, Intersect = Intersector<2>]' /usr/local/cots/pooma/src/Array/Array.h:1716: instantiated from `static int LeafFunctor >, ExpressionApply > >::apply(const Engine >&, const ExpressionApply >&) [with int Dim = 2, T = double, LT = UniformTag, PatchTag = Remote, int BD = 2, Intersect = Intersector<2>]' /usr/local/cots/pooma/src/Array/Array.h:2307: instantiated from `static int LeafFunctor, ExpressionApply >::apply(const Array&, const ExpressionApply&) [with int Dim = 2, T = double, E = MultiPatchView, 2>, Tag = IntersectorTag >]' /usr/local/cots/pooma/src/Array/Array.h:80: instantiated from `static LeafFunctor::Type_t ForEach::apply(const Expr&, const FTag&, const CTag&) [with Expr = Array<2, double, MultiPatchView, 2> >, FTag = ExpressionApply > >, CTag = NullCombine]' /usr/local/cots/pooma/src/Array/Array.h:88: instantiated from `ForEach::Type_t forEach(const Expr&, const FTag&, const CTag&) [with Expr = Array<2, double, MultiPatchView, 2> >, FTag = ExpressionApply > >, CTag = NullCombine]' /usr/local/cots/pooma/src/Array/Array.h:276: instantiated from `void expressionApply(const A&, const Tag&) [with A = Array<2, double, MultiPatchView, 2> >, Tag = IntersectorTag >]' /usr/local/cots/pooma/src/Array/Array.h:1913: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatchView, 2> >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:108: instantiated from `void Evaluator::evaluate(const LHS&, const Op&, const RHS&) const [with LHS = Array<2, double, MultiPatch > >, RHS = Array<2, double, ConstantFunction>, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2504: instantiated from `const Array& assign(const Array&, const T1&, const Op&) [with int Dim = 2, T = double, EngineTag = MultiPatch >, T1 = double, Op = OpAssign]' /usr/local/cots/pooma/src/Array/Array.h:2117: instantiated from `const Array& Array::operator=(const T1&) const [with T1 = double, int Dim = 2, T = double, EngineTag = MultiPatch >]' /usr/local/cots/pooma/src/Array/Array.h:2105: instantiated from `Array& Array::operator=(const Array&) [with int Dim = 2, T = double, EngineTag = MultiPatch >]' diffusion_test.cpp:77: instantiated from here /usr/local/cots/pooma/src/Layout/UniformGridLayout.h:297: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Partition/DistributedMapper.h: In member function `void DistributedMapper::map(const std::vector, Interval >*, std::allocator, Interval >*> >&) const [with int Dim = 2]': /usr/local/cots/pooma/src/Utilities/RefCountedBlockPtr.h:411: instantiated from here /usr/local/cots/pooma/src/Partition/DistributedMapper.h:68: warning: comparison between signed and unsigned integer expressions /usr/local/cots/pooma/src/Partition/BisectionMapper.h: In member function `void BisectionMapper::map(const std::vector, Interval >*, std::allocator, Interval >*> >&) const [with int Dim = 2]': /usr/local/cots/pooma/src/Partition/DistributedMapper.h:86: instantiated from `void DistributedMapper::map(const std::vector, Interval >*, std::allocator, Interval >*> >&) const [with int Dim = 2]' /usr/local/cots/pooma/src/Utilities/RefCountedBlockPtr.h:411: instantiated from here /usr/local/cots/pooma/src/Partition/BisectionMapper.h:86: warning: comparison between signed and unsigned integer expressions From rguenth at tat.physik.uni-tuebingen.de Wed Apr 23 20:50:16 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Wed, 23 Apr 2003 22:50:16 +0200 (CEST) Subject: [pooma-dev] GCC 3.2 on Redhat 8.0--a fair number of compiler warnings? In-Reply-To: <24752b2dfd3ec609e6c9582a2ede3a5b3ea6f32a@locuspharma.com> Message-ID: On 23 Apr 2003, George Talbot wrote: > Hi, > > New POOMA II user. Got everything to compile and link today. The > performance of this stuff is shocking. > > I ran a hand coded version of your 2D diffusion program on my box for > 250 steps, 800x800 array. 1m34s, uniprocessor. > > Using a stencil, same box, uniprocessor. 25s. WOW. > > Using -shmem -np 2 to distribute the shared array version on both of my > processors on my box. 12s. Truly amazing. I think this may be a > dramatically useful tool. You will find gcc-3.3 even more useful, combined with careful inline parameter tuning! > The only thing that irks me a bit is that there are a fair number of > compiler warnings, which I will append, along with the source I used, to > this e-mail. Is this normal? Swimming in a sea of warnings from the > POOMA headers, I might miss some on my own code... Without -Wall I dont see warnings from the POOMA headers with g++/icpc, but you are right, theres something to clean up. But as time is always short, fixing warnings that dont annoy me is not very high on my priority list. But I'm sure we're accepting patches to correct them (maybe...). Richard. From gtalbot at locuspharma.com Wed Apr 23 20:57:03 2003 From: gtalbot at locuspharma.com (George Talbot) Date: 23 Apr 2003 16:57:03 -0400 Subject: [pooma-dev] GCC 3.2 on Redhat 8.0--a fair number of compiler warnings? In-Reply-To: References: Message-ID: On Wed, 2003-04-23 at 16:50, Richard Guenther wrote: > On 23 Apr 2003, George Talbot wrote: > > > Hi, > > > > New POOMA II user. Got everything to compile and link today. The > > performance of this stuff is shocking. > > > > I ran a hand coded version of your 2D diffusion program on my box for > > 250 steps, 800x800 array. 1m34s, uniprocessor. > > > > Using a stencil, same box, uniprocessor. 25s. WOW. > > > > Using -shmem -np 2 to distribute the shared array version on both of my > > processors on my box. 12s. Truly amazing. I think this may be a > > dramatically useful tool. > > You will find gcc-3.3 even more useful, combined with careful inline > parameter tuning! Really? What enhancements make POOMA speedier? The DFA scheduler? Type-based alias analysis for aggregates? All of the above? ;^) > > The only thing that irks me a bit is that there are a fair number of > > compiler warnings, which I will append, along with the source I used, to > > this e-mail. Is this normal? Swimming in a sea of warnings from the > > POOMA headers, I might miss some on my own code... > > Without -Wall I dont see warnings from the POOMA headers with g++/icpc, > but you are right, theres something to clean up. But as time is always > short, fixing warnings that dont annoy me is not very high on my priority > list. But I'm sure we're accepting patches to correct them (maybe...). So I'm on my own, eh? ;^) If I come up with any good patches for cleaning up warnings I encounter, shall I post them to this list? Are there any preferred formats? --George P.S. Thanks for such an amazing tool. From rguenth at tat.physik.uni-tuebingen.de Wed Apr 23 21:06:40 2003 From: rguenth at tat.physik.uni-tuebingen.de (Richard Guenther) Date: Wed, 23 Apr 2003 23:06:40 +0200 (CEST) Subject: [pooma-dev] GCC 3.2 on Redhat 8.0--a fair number of compiler warnings? In-Reply-To: Message-ID: On 23 Apr 2003, George Talbot wrote: > On Wed, 2003-04-23 at 16:50, Richard Guenther wrote: > > On 23 Apr 2003, George Talbot wrote: > > > > > Hi, > > > > > > New POOMA II user. Got everything to compile and link today. The > > > performance of this stuff is shocking. > > > > > > I ran a hand coded version of your 2D diffusion program on my box for > > > 250 steps, 800x800 array. 1m34s, uniprocessor. > > > > > > Using a stencil, same box, uniprocessor. 25s. WOW. > > > > > > Using -shmem -np 2 to distribute the shared array version on both of my > > > processors on my box. 12s. Truly amazing. I think this may be a > > > dramatically useful tool. > > > > You will find gcc-3.3 even more useful, combined with careful inline > > parameter tuning! > > Really? What enhancements make POOMA speedier? The DFA scheduler? > Type-based alias analysis for aggregates? All of the above? ;^) Mostly the DFA scheduler and the possibility to specify --param min-inline-insns=300 ;) The rest surely helps, too. > > > The only thing that irks me a bit is that there are a fair number of > > > compiler warnings, which I will append, along with the source I used, to > > > this e-mail. Is this normal? Swimming in a sea of warnings from the > > > POOMA headers, I might miss some on my own code... > > > > Without -Wall I dont see warnings from the POOMA headers with g++/icpc, > > but you are right, theres something to clean up. But as time is always > > short, fixing warnings that dont annoy me is not very high on my priority > > list. But I'm sure we're accepting patches to correct them (maybe...). > > So I'm on my own, eh? ;^) If I come up with any good patches for > cleaning up warnings I encounter, shall I post them to this list? Are > there any preferred formats? I'd use unified diffs. But be warned that there may be license/copyright issues with accepting your work - Jeffrey may comment on this. But I think its about identifying the most annoying ones that come up again and again. Richard. From gtalbot at locuspharma.com Fri Apr 25 14:21:39 2003 From: gtalbot at locuspharma.com (George Talbot) Date: 25 Apr 2003 10:21:39 -0400 Subject: [pooma-dev] GCC 3.2 on Redhat 8.0--a fair number of compiler warnings? In-Reply-To: <24752b2dfd3ec609e6c9582a2ede3a5b3ea6f32a@locuspharma.com> References: <24752b2dfd3ec609e6c9582a2ede3a5b3ea6f32a@locuspharma.com> Message-ID: Not so good news. So I went back to the hand-coded program and reversed the indices, indexing x in the inner loop and y in the outer to reflect the memory layout of the array. Now the hand-coded variant is twice as fast as the equivalent POOMA program. This is with GCC 3.2. Distributing it among the two processors on my workstation doesn't bring the performance up to the single-processor hand-coded version. I will try today to beat the hand-coded program with MPI over more than two processors. However, I would like to know something. Is the diffusion example from the book a realistic example to show the performance of POOMA vs. hand-coded stuff? I need to be able to describe the break-even point where a POOMA app will outperform a uniprocessor app. Of course, I probably need to try to apply it to one of my in-house programs to be realistic, but I was hoping that someone on the list could suggest a particular demonstration program that would show POOMA outperforming a uniprocessor hand-coded implementation. --George On Wed, 2003-04-23 at 16:10, George Talbot wrote: > Hi, > > New POOMA II user. Got everything to compile and link today. The > performance of this stuff is shocking. > > I ran a hand coded version of your 2D diffusion program on my box for > 250 steps, 800x800 array. 1m34s, uniprocessor. > > Using a stencil, same box, uniprocessor. 25s. WOW. > > Using -shmem -np 2 to distribute the shared array version on both of my > processors on my box. 12s. Truly amazing. I think this may be a > dramatically useful tool. > > The only thing that irks me a bit is that there are a fair number of > compiler warnings, which I will append, along with the source I used, to > this e-mail. Is this normal? Swimming in a sea of warnings from the > POOMA headers, I might miss some on my own code... > > Take care. > > -- > George T. Talbot > > > Here's the program: > > #include > using namespace std; > > #define STENCIL > #define POOMA > #define DISTRIBUTED > > #ifdef POOMA > #include "Pooma/Arrays.h" > #endif > > #ifdef STENCIL > class NinePtDiffusion > { > public: > template > inline typename C::Element_t operator()(const C& c, int I, int J) > const > { > return (1.0/9.0) * > (c.read(I+1, J+1) + c.read(I+1, J ) + c.read(I+1, > J-1) + > c.read(I , J+1) + c.read(I , J ) + c.read(I , > J-1) + > c.read(I-1, J+1) + c.read(I-1, J ) + c.read(I-1, > J-1)); > } > > inline int lowerExtent(int) const { return 1; } > inline int upperExtent(int) const { return 1; } > }; > #endif > > int main(int argc, char* argv[]) > { > #ifdef POOMA > Pooma::initialize(argc, argv); > #endif > > long numAveragings = 250, numSteps; > > numSteps = (numAveragings+1)/2; > > long n = 800; > > long numProcessors=2; > > #ifdef POOMA > Inform output; > > output << "processors=" << numProcessors > << " avgs=" << numAveragings > << " n=" << n > << endl; > #else > cout << "processors=" << numProcessors > << " avgs=" << numAveragings > << " n=" << n > << endl; > #endif > > #ifdef POOMA > Interval<1> N(0, n-1); > Interval<2> problemDomain(N, N); > > Interval<1> I(1, n-2); > Interval<1> J(1, n-2); > Interval<2> interiorDomain(I, I); > > #ifdef DISTRIBUTED > UniformGridPartition<2> > partition(Loc<2>(numProcessors,numProcessors), GuardLayers<2>(1), > GuardLayers<2>(0)); > UniformGridLayout<2> layout(problemDomain, partition, > DistributedTag()); > > Array<2, double, MultiPatch > > > a(layout); > Array<2, double, MultiPatch > > > b(layout); > #else > Array<2, double, Brick> a(problemDomain); > Array<2, double, Brick> b(problemDomain); > #endif > > a = b = 0.0; > > Pooma::blockAndEvaluate(); > b(n/2, n/2) = 1000.0; > > #ifdef STENCIL > Stencil diffusion; > #endif > > #else > double a[n][n], b[n][n]; > > for (long x=1; x { > for (long y=1; y { > a[x][y] = b[x][y] = 0.0; > } > } > > b[n/2][n/2] = 1000.0; > #endif > > const double weight = 1.0/9.0; > > for (long i=0; i { > #ifdef STENCIL > a(interiorDomain) = diffusion(b, interiorDomain); > b(interiorDomain) = diffusion(a, interiorDomain); > #elif defined (POOMA) > a(I,J) = weight * > (b(I+1, J+1) + b(I+1, J ) + b(I+1, J-1) + > b(I , J+1) + b(I , J ) + b(I , J-1) + > b(I-1, J+1) + b(I-1, J ) + b(I-1, J-1)); > > b(I,J) = weight * > (a(I+1, J+1) + a(I+1, J ) + a(I+1, J-1) + > a(I , J+1) + a(I , J ) + a(I , J-1) + > a(I-1, J+1) + a(I-1, J ) + a(I-1, J-1)); > #else > for (long x=1; x { > for (long y=1; y { > a[x][y] = weight * > (b[x+1][y+1] + b[x+1][y] + b[x+1][y-1] + > b[x] [y+1] + b[x] [y] + b[x] [y-1] + > b[x-1][y+1] + b[x-1][y] + b[x-1][y-1]); > } > } > > for (long x=1; x { > for (long y=1; y { > b[x][y] = weight * > (a[x+1][y+1] + a[x+1][y] + a[x+1][y-1] + > a[x] [y+1] + a[x] [y] + a[x] [y-1] + > a[x-1][y+1] + a[x-1][y] + a[x-1][y-1]); > } > } > #endif > } > > #ifdef POOMA > Pooma::blockAndEvaluate(); > > output << "Center position: " << a(n/2, n/2) << endl; > > Pooma::finalize(); > #else > cout << "Center position: " << a[n/2][n/2] << endl; > #endif > > return 0; > } > > > Here's the compiler warnings, plus the compiler command line: > > g++ -c -pipe -Wall -W -O2 -D_REENTRANT -DQT_NO_DEBUG > -DQT_THREAD_SUPPORT -I/usr/local/qt/mkspecs/default -I. > -I/usr/local/cots/cheetah/linux/src > -I/usr/local/cots/cheetah/linux/lib/g++-ex -I/usr/local/cots/pooma/src > -I/usr/local/cots/pooma/linux/lib/PoomaConfiguration > -I/usr/local/qt/include -o diffusion_test.o diffusion_test.cpp > In file included from /usr/local/cots/pooma/src/Pooma/Domains.h:42, > from > /usr/local/cots/pooma/src/Domain/DomainRemoveOverlap.h:32, from /usr/local/cots/pooma/src/Layout/SparseTileLayout.cpp:36, > from > /usr/local/cots/pooma/src/Layout/SparseTileLayout.h:1201, > from > /usr/local/cots/pooma/src/Engine/IsValidLocation.h:55, > from /usr/local/cots/pooma/src/Array/PrintArray.h:57, > from /usr/local/cots/pooma/src/Array/Array.h:58, > from /usr/local/cots/pooma/src/Pooma/BrickArrays.h:42, > from /usr/local/cots/pooma/src/Pooma/UMPArrays.h:40, > from /usr/local/cots/pooma/src/Pooma/Arrays.h:39, > from diffusion_test.cpp:9: > /usr/local/cots/pooma/src/Domain/Grid.h: In copy constructor > `Grid<1>::Grid(const Grid<1>&)': > /usr/local/cots/pooma/src/Domain/Grid.h:360: warning: base class `class > Domain<1, DomainTraits > >' should be explicitly initialized > in the > copy constructor > In file included from /usr/local/cots/pooma/src/Pooma/Arrays.h:46, > from diffusion_test.cpp:9: > /usr/local/cots/pooma/src/Engine/RemoteEngine.h: In copy constructor > `GatherContexts::GatherContextsData::GatherContextsData(const > GatherContexts::GatherContextsData&)': > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1697: warning: base > class ` > class RefCounted' should be explicitly initialized in the copy > constructor > diffusion_test.cpp: In function `int main(int, char**)': > diffusion_test.cpp:100: warning: unused variable `const double weight' > In file included from > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1499, > from /usr/local/cots/pooma/src/Array/Array.h:108, > from diffusion_test.cpp:77: > /usr/local/cots/pooma/src/Layout/UniformGridLayout.h: In copy > constructor > `UniformGridLayout::UniformGridLayout(const > UniformGridLayout&) > [with int Dim = 2]': > /usr/local/cots/pooma/src/Layout/LayoutBase.h:806: instantiated from > `LayoutBaseViewData::LayoutBaseViewData(const L&, const > Domain&) [with DT = DomainTraits >, int Dim = 2, > int Dim2 = 2, L = UniformGridLayout<2>]' > /usr/local/cots/pooma/src/Layout/LayoutBase.h:1368: instantiated from > `UniformGridLayoutViewData::UniformGridLayoutViewData(const > UniformGridLayout&, const Domain&) [with DT = > DomainTraits >, int Dim > = 2, int Dim2 = 2]' > /usr/local/cots/pooma/src/Layout/LayoutBase.h:1109: instantiated from > `LayoutBaseView::LayoutBaseView(lvd*) [with int Dim = 2, > int Dim2 = 2, lvd = UniformGridLayoutViewData<2, 2>]' > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1499: instantiated > from `UniformGridLayoutView Dim2>::UniformGridLayoutView(UniformGridLayoutViewData Dim2>::Layout_t&, const Domain&) [with DT = > DomainTraits >, int Dim = 2, int Dim2 = 2]' > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:992: instantiated from > `Engine > >::Engine(const Engine T, MultiPatch >&, const Domain&) [with DT > = DomainTraits >, int Dim = 2, T = double, LayoutTag = > UniformTag, PatchTag = Remote, int Dim2 = 2]' > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:1457: instantiated > from `Array::Array(const Engine EngineTag2>&, const Initializer&) [with int Dim2 = 2, T2 = double, > EngineTag2 = MultiPatch >, Initializer = > Interval<2>, int Dim = 2, T = double, EngineTag = > MultiPatchView, 2>]' > /usr/local/cots/pooma/src/Engine/RemoteEngine.h:676: instantiated from > `static Array >::newDim, T, NewEngine T1, E1>::Engine_t, Array::Domain_t>::Type_t::Tag_t> > View0 >::make(const Array&) [with int Dim > = 2, T = double, EngineTag = MultiPatch