Users/hc/gnntest

.gitattributes

@@ -0,0 +1,31 @@
+inputs/cora filter=lfs diff=lfs merge=lfs -text
+inputs/tester filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora_edgelist.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-labels.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.sgr filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.csgr filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-dims.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora_full.npz filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora_labels.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-val_mask.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.el filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.features filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.ft filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-train_mask.txt filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-feat.bin filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-feats.bin filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora.gr filter=lfs diff=lfs merge=lfs -text
+inputs/cora/cora-test_mask.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-test-feats.bin filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-test-labels.bin filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-test-labels-dims.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-test_mask.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-dims.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester.el filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-feats.bin filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-mlabels.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-train_mask.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester.csgr filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-labels.txt filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-test-mapping.bin filter=lfs diff=lfs merge=lfs -text
+inputs/tester/tester-val_mask.txt filter=lfs diff=lfs merge=lfs -text

.github/workflows/build-and-test.yml

@@ -53,11 +53,11 @@ jobs:
         shell: bash -l {0}
     strategy:
       matrix:
-        build-type: ['Release', 'Debug']
-        sanitizer-type: ['nosan', 'san']
-        exclude:
-          - build-type: 'Debug'
-            sanitizer-type: 'san'
+        build-type: ['Release']
+        sanitizer-type: ['nosan']
+        # exclude:
+        #   - build-type: 'Debug'
+        #     sanitizer-type: 'san'
     needs: docker-create-ubuntu-2204
 
     steps:
@@ -105,9 +105,9 @@ jobs:
           make docker
 
     - name: Run Tests
-      timeout-minutes: 5
+      timeout-minutes: 15
       run: |
-        CONTAINER_CMD="make run-tests" \
+        CONTAINER_CMD="bash -lc 'source /opt/intel/oneapi/setvars.sh &&  make run-tests'" \
           IMAGE_NAME="${{ env.IMAGE_NAME }}" \
           VERSION="${{ env.IMAGE_VERSION }}" \
           make docker

Makefile

@@ -117,11 +117,12 @@ run-tests:
 	@ctest --test-dir build -R wmd --verbose
 	@ctest --test-dir build -R large-vec --verbose
 	@ctest --test-dir build -R compile-lscsr --verbose
+	@ctest --test-dir build/libgnn/test --verbose
 
 # this command is slow since hooks are not stored in the container image
 # this is mostly for CI use
 docker-pre-commit:
 	@docker --context ${CONTAINER_CONTEXT} run --rm \
 	-v ${SRC_DIR}/:${CONTAINER_SRC_DIR} --privileged \
 	--workdir=${CONTAINER_WORKDIR} -t \
-	${IMAGE_NAME}:${VERSION} bash -lc "git config --global --add safe.directory /pando-galois && make hooks && make pre-commit"
+	${IMAGE_NAME}:${VERSION} bash -lc "git config --global --add safe.directory /galois && make hooks && make pre-commit"

libcusp/include/galois/graphs/NewGeneric.h

@@ -33,6 +33,7 @@
 #include "shad/ShadGraphConverter.h"
 
 #include <optional>
+#include <filesystem>
 #include <sstream>
 
 #define CUSP_PT_TIMER 0
@@ -236,6 +237,8 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
         "GraphPartitioningTime", GRNAME);
     Tgraph_construct.start();
 
+    std::cout << "CuSP side file path:" << localGraphFileName << "\n" << std::flush;
+
     if (readFromFile) {
       galois::gDebug("[", base_DistGraph::id,
                      "] Reading local graph from file ", localGraphFileName);
@@ -245,6 +248,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
     }
 
     galois::graphs::OfflineGraph* offlineGraph{nullptr};
+    std::cout << "offline graph starts..\n" << std::flush;
 
     std::string host_prefix =
         std::string("[") +
@@ -255,6 +259,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
     galois::graphs::BufferedGraph<EdgeTy> bufGraph;
     bufGraph.resetReadCounters();
 
+    std::cout << "offline graph starts.. 1\n" << std::flush;
     std::vector<unsigned> dummy;
     // not actually getting masters, but getting assigned readers for nodes
     if (masterBlockFile == "") {
@@ -302,7 +307,18 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
             md, base_DistGraph::numGlobalNodes, base_DistGraph::numGlobalEdges,
             shadConverter.getOutIndexBuffer(), dummy, nodeWeight, edgeWeight);
       } else {
+        std::cout << filename << " is given to offline graph\n";
+        if (FILE* fp = fopen(filename.c_str(), "r")) {
+          std::cout << "succeeded to read the input file at NewGeneric.h\n" << std::flush;
+          fclose(fp);
+        } else {
+          std::cout << "failed to read the input file at NewGeneric.h\n" << std::flush;
+        }
+
+        std::filesystem::path p{filename.c_str()};
+        std::cout << "File size:" << std::filesystem::file_size(p) << "\n" << std::flush;
         offlineGraph = new galois::graphs::OfflineGraph(filename);
+        std::cout << "offline graph reading failed\n" << std::flush;
         base_DistGraph::numGlobalNodes = offlineGraph->size();
         base_DistGraph::numGlobalEdges = offlineGraph->sizeEdges();
         base_DistGraph::computeMasters(md, *offlineGraph, dummy, nodeWeight,
@@ -316,14 +332,17 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       base_DistGraph::readersFromFile(*offlineGraph, masterBlockFile);
     }
 
+    std::cout << "offline graph starts.. 2\n" << std::flush;
     graphPartitioner = std::make_unique<Partitioner>(
         host, _numHosts, base_DistGraph::numGlobalNodes,
         base_DistGraph::numGlobalEdges);
     // TODO abstract this away somehow
     graphPartitioner->saveGIDToHost(base_DistGraph::gid2host);
 
+    std::cout << "offline graph starts.. 3\n" << std::flush;
     // get training nodes and split evenly among hosts
     std::vector<uint32_t> trainPoints = this->getGNNBreakpoints(filename);
+    std::cout << "offline graph starts.. 4\n" << std::flush;
     // TODO(hc)
     if (!trainPoints.empty()) {
       std::vector<unsigned> testDistribution =
@@ -359,6 +378,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
         galois::gWarn("partitioning policy used doesn't use trainpoints");
       }
     }
+    std::cout << "offline graph starts.. 5\n" << std::flush;
 
     // signifies how many outgoing edges a particular host should expect from
     // this host
@@ -377,6 +397,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       hasIncomingEdge.resize(base_DistGraph::numHosts);
     }
 
+    std::cout << "offline graph starts.. 6\n" << std::flush;
     // phase 0
 
     galois::gDebug("[", base_DistGraph::id, "] Starting graph reading.");
@@ -401,6 +422,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
                                 host_prefix);
     }
 
+    std::cout << "offline graph starts.. 7\n" << std::flush;
     graphReadTimer.stop();
     galois::gDebug("[", base_DistGraph::id, "] Reading graph complete.");
 
@@ -414,6 +436,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       galois::gDebug("[", base_DistGraph::id, "] Master assignment complete.");
     }
 
+    std::cout << "offline graph starts.. 8\n" << std::flush;
     galois::StatTimer inspectionTimer("EdgeInspection", GRNAME);
     inspectionTimer.start();
     bufGraph.resetReadCounters();
@@ -439,6 +462,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       edgeCutInspection(bufGraph, inspectionTimer, edgeOffset,
                         prefixSumOfEdges);
     }
+    std::cout << "offline graph starts.. 9\n" << std::flush;
     // inspection timer is stopped in edgeInspection function
 
     // flip partitioners that have a master assignment phase to stage 2
@@ -448,6 +472,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       graphPartitioner->enterStage2();
     }
 
+    std::cout << "offline graph starts.. 10\n" << std::flush;
     // get memory back from inspection metadata
     numOutgoingEdges.clear();
     hasIncomingEdge.clear();
@@ -463,6 +488,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
                                        base_DistGraph::numEdges);
     base_DistGraph::graph.constructNodes();
 
+    std::cout << "offline graph starts.. 11\n" << std::flush;
     // edge end fixing
     auto& base_graph = base_DistGraph::graph;
     galois::do_all(
@@ -477,10 +503,12 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
     freeVector(prefixSumOfEdges); // should no longer use this variable
     galois::CondStatTimer<MORE_DIST_STATS> TfillMirrors("FillMirrors", GRNAME);
 
+    std::cout << "offline graph starts.. 12\n" << std::flush;
     TfillMirrors.start();
     fillMirrors();
     TfillMirrors.stop();
 
+    std::cout << "offline graph starts.. 13\n" << std::flush;
     if (_edgeStateRounds > 1) {
       // reset edge load since we need exact same answers again
       resetEdgeLoad();
@@ -495,6 +523,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       bufGraph.resetAndFree();
     }
 
+    std::cout << "offline graph starts.. 14\n" << std::flush;
     // Finalization
 
     // TODO this is a hack; fix it somehow
@@ -517,6 +546,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
     galois::CondStatTimer<MORE_DIST_STATS> Tthread_ranges("ThreadRangesTime",
                                                           GRNAME);
 
+    std::cout << "offline graph starts.. 15\n" << std::flush;
     Tthread_ranges.start();
     base_DistGraph::determineThreadRanges();
     Tthread_ranges.stop();
@@ -528,6 +558,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
     Tgraph_construct.stop();
     galois::gDebug("[", base_DistGraph::id, "] Graph construction complete.");
 
+    std::cout << "offline graph starts.. 16\n" << std::flush;
     if (useWMD) {
       // Different from the gr format file that has been used by Galois
       // and does not contain node data in the file,
@@ -537,6 +568,7 @@ class NewDistGraphGeneric : public DistGraph<NodeTy, EdgeTy> {
       assignNodeDataFromSHADProp(&shadConverter);
     }
 
+    std::cout << "offline graph starts.. 17\n" << std::flush;
     // report state rounds
     if (base_DistGraph::id == 0) {
       galois::runtime::reportStat_Single(GRNAME, "CuSPStateRounds",

libcusp/test/shad-dist-graph.cpp

@@ -53,13 +53,10 @@ int main() {
   sumGlobalNodes += graph->numMasters();
   sumGlobalEdges += graph->sizeEdges();
 
-  uint64_t reducedSumGlobalNodes = sumGlobalNodes.reduce();
-  uint64_t reducedSumGlobalEdges = sumGlobalEdges.reduce();
-
-  assert(reducedSumGlobalNodes == numNodes);
-  assert(reducedSumGlobalNodes == graph->globalSize());
-  assert(reducedSumGlobalEdges == numEdges);
-  assert(reducedSumGlobalEdges == graph->globalSizeEdges());
+  assert(sumGlobalNodes.reduce() == numNodes);
+  assert(sumGlobalNodes.reduce() == graph->globalSize());
+  assert(sumGlobalEdges.reduce() == numEdges);
+  assert(sumGlobalEdges.reduce() == graph->globalSizeEdges());
 
   std::cout << "Num. nodes/edges tests has been passed\n";