Update forked branch-0.18 from release

.github/labeler.yml

@@ -0,0 +1,30 @@
+# https://github.com/actions/labeler#common-examples
+# Adapted from https://github.com/rapidsai/cugraph/blob/main/.github/CODEOWNERS
+# Labels culled from https://github.com/rapidsai/cugraph/labels
+
+python:
+  - 'python/**'
+  - 'notebooks/**'
+  - 'benchmarks/**'
+
+doc:
+  - 'docs/**'
+  - '**/*.md'
+
+datasets:
+  - 'datasets/**'
+
+cuGraph:
+  - 'cpp/**'
+
+CMake:
+  - '**/CMakeLists.txt'
+  - '**/cmake/**'
+
+Ops:
+  - '.github/**'
+  - 'ci/**'
+  - 'conda/**'
+  - '**/Dockerfile'
+  - '**/.dockerignore'
+  - 'docker/**'

CHANGELOG.md

@@ -6,6 +6,7 @@
 ## Improvements
 
 ## Bug Fixes
+- PR #1321 Fix benchmark script trap setup to come after the PATH variable update
 
 # cuGraph 0.17.0 (10 Dec 2020)
 ## New Features

ci/benchmark/build.sh

@@ -20,18 +20,18 @@ function cleanup {
   rm -f testoutput.txt
 }
 
-# Set cleanup trap for Jenkins
-if [ ! -z "$JENKINS_HOME" ] ; then
-  gpuci_logger "Jenkins environment detected, setting cleanup trap"
-  trap cleanup EXIT
-fi
-
 # Set path, build parallel level, and CUDA version
 cd $WORKSPACE
 export PATH=/opt/conda/bin:/usr/local/cuda/bin:$PATH
 export PARALLEL_LEVEL=${PARALLEL_LEVEL:-4}
 export CUDA_REL=${CUDA_VERSION%.*}
 
+# Set cleanup trap for Jenkins
+if [ ! -z "$JENKINS_HOME" ] ; then
+  gpuci_logger "Jenkins environment detected, setting cleanup trap"
+  trap cleanup EXIT
+fi
+
 # Set home
 export HOME=$WORKSPACE
 

ci/checks/style.sh

@@ -1,5 +1,5 @@
 #!/bin/bash
-# Copyright (c) 2018-2020, NVIDIA CORPORATION.
+# Copyright (c) 2018-2021, NVIDIA CORPORATION.
 ########################
 # cuGraph Style Tester #
 ########################
@@ -52,13 +52,14 @@ COPYRIGHT=`env PYTHONPATH=ci/utils python ci/checks/copyright.py --git-modified-
 CR_RETVAL=$?
 ERRORCODE=$((ERRORCODE | ${CR_RETVAL}))
 
-# Output results if failure otherwise show pass
 if [ "$CR_RETVAL" != "0" ]; then
   echo -e "\n\n>>>> FAILED: copyright check; begin output\n\n"
   echo -e "$COPYRIGHT"
   echo -e "\n\n>>>> FAILED: copyright check; end output\n\n"
 else
-  echo -e "\n\n>>>> PASSED: copyright check\n\n"
+  echo -e "\n\n>>>> PASSED: copyright check; begin debug output\n\n"
+  echo -e "$COPYRIGHT"
+  echo -e "\n\n>>>> PASSED: copyright check; end debug output\n\n"
 fi
 
 exit ${ERRORCODE}

ci/gpu/build.sh

@@ -91,7 +91,7 @@ conda list --show-channel-urls
 
 if [[ -z "$PROJECT_FLASH" || "$PROJECT_FLASH" == "0" ]]; then
   gpuci_logger "Build from source"
-  $WORKSPACE/build.sh -v clean libcugraph cugraph --allgpuarch
+  $WORKSPACE/build.sh -v clean libcugraph cugraph
 fi
 
 ################################################################################

ci/utils/git_helpers.py

@@ -1,4 +1,4 @@
-# Copyright (c) 2019-2020, NVIDIA CORPORATION.
+# Copyright (c) 2019-2021, NVIDIA CORPORATION.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -59,14 +59,24 @@ def uncommittedFiles():
     return ret
 
 
-def changedFilesBetween(b1, b2):
-    """Returns a list of files changed between branches b1 and b2"""
+def changedFilesBetween(baseName, branchName, commitHash):
+    """
+    Returns a list of files changed between branches baseName and latest commit
+    of branchName.
+    """
     current = branch()
-    __git("checkout", "--quiet", b1)
-    __git("checkout", "--quiet", b2)
-    files = __gitdiff("--name-only", "--ignore-submodules", "%s...%s" %
-                      (b1, b2))
-    __git("checkout", "--quiet", current)
+    # checkout "base" branch
+    __git("checkout", "--force", baseName)
+    # checkout branch for comparing
+    __git("checkout", "--force", branchName)
+    # checkout latest commit from branch
+    __git("checkout", "-fq", commitHash)
+
+    files = __gitdiff("--name-only", "--ignore-submodules",
+                      f"{baseName}..{branchName}")
+
+    # restore the original branch
+    __git("checkout", "--force", current)
     return files.splitlines()
 
 
@@ -87,26 +97,37 @@ def changesInFileBetween(file, b1, b2, pathFilter=None):
 
 def modifiedFiles(pathFilter=None):
     """
-    If inside a CI-env (ie. currentBranch=current-pr-branch and the env-var
-    PR_TARGET_BRANCH is defined), then lists out all files modified between
-    these 2 branches. Else, lists out all the uncommitted files in the current
-    branch.
+    If inside a CI-env (ie. TARGET_BRANCH and COMMIT_HASH are defined, and
+    current branch is "current-pr-branch"), then lists out all files modified
+    between these 2 branches. Else, lists out all the uncommitted files in the
+    current branch.
 
     Such utility function is helpful while putting checker scripts as part of
     cmake, as well as CI process. This way, during development, only the files
     touched (but not yet committed) by devs can be checked. But, during the CI
     process ALL files modified by the dev, as submiited in the PR, will be
     checked. This happens, all the while using the same script.
     """
-    if "PR_TARGET_BRANCH" in os.environ and branch() == "current-pr-branch":
-        allFiles = changedFilesBetween(os.environ["PR_TARGET_BRANCH"],
-                                       branch())
+    targetBranch = os.environ.get("TARGET_BRANCH")
+    commitHash = os.environ.get("COMMIT_HASH")
+    currentBranch = branch()
+    print(f"   [DEBUG] TARGET_BRANCH={targetBranch}, COMMIT_HASH={commitHash}, "
+          f"currentBranch={currentBranch}")
+
+    if targetBranch and commitHash and (currentBranch == "current-pr-branch"):
+        print("   [DEBUG] Assuming a CI environment.")
+        allFiles = changedFilesBetween(targetBranch, currentBranch, commitHash)
     else:
+        print("   [DEBUG] Did not detect CI environment.")
         allFiles = uncommittedFiles()
+
     files = []
     for f in allFiles:
         if pathFilter is None or pathFilter(f):
             files.append(f)
+
+    filesToCheckString = "\n\t".join(files) if files else "<None>"
+    print(f"   [DEBUG] Found files to check:\n\t{filesToCheckString}\n")
     return files
 
 

cpp/src/community/louvain.cu

@@ -15,19 +15,8 @@
  */
 
 #include <community/louvain.cuh>
-
-// "FIXME": remove the guards after support for Pascal will be dropped;
-//
-// Disable louvain(experimenta::graph_view_t,...)
-// versions for GPU architectures < 700
-//(this is because cuco/static_map.cuh would not
-// compile on those)
-//
-#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 700
 #include <experimental/graph.hpp>
-#else
 #include <experimental/louvain.cuh>
-#endif
 
 namespace cugraph {
 
@@ -58,13 +47,22 @@ std::pair<size_t, weight_t> louvain(
 {
   CUGRAPH_EXPECTS(clustering != nullptr, "Invalid input argument: clustering is null");
 
-#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 700
-  CUGRAPH_FAIL("Louvain not supported on Pascal and older architectures");
-#else
-  experimental::Louvain<experimental::graph_view_t<vertex_t, edge_t, weight_t, false, multi_gpu>>
-    runner(handle, graph_view);
-  return runner(clustering, max_level, resolution);
-#endif
+  // "FIXME": remove this check and the guards below
+  //
+  // Disable louvain(experimental::graph_view_t,...)
+  // versions for GPU architectures < 700
+  // (cuco/static_map.cuh depends on features not supported on or before Pascal)
+  //
+  cudaDeviceProp device_prop;
+  CUDA_CHECK(cudaGetDeviceProperties(&device_prop, 0));
+
+  if (device_prop.major < 7) {
+    CUGRAPH_FAIL("Louvain not supported on Pascal and older architectures");
+  } else {
+    experimental::Louvain<experimental::graph_view_t<vertex_t, edge_t, weight_t, false, multi_gpu>>
+      runner(handle, graph_view);
+    return runner(clustering, max_level, resolution);
+  }
 }
 
 }  // namespace detail
@@ -78,11 +76,7 @@ std::pair<size_t, typename graph_t::weight_type> louvain(raft::handle_t const &h
 {
   CUGRAPH_EXPECTS(clustering != nullptr, "Invalid input argument: clustering is null");
 
-#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 700
-  CUGRAPH_FAIL("Louvain not supported on Pascal and older architectures");
-#else
   return detail::louvain(handle, graph, clustering, max_level, resolution);
-#endif
 }
 
 // Explicit template instantations

cpp/src/experimental/louvain.cuh

@@ -21,7 +21,6 @@
 
 #include <rmm/thrust_rmm_allocator.h>
 #include <compute_partition.cuh>
-#include <cuco/static_map.cuh>
 #include <experimental/shuffle.cuh>
 #include <utilities/comm_utils.cuh>
 #include <utilities/graph_utils.cuh>
@@ -33,6 +32,23 @@
 #include <patterns/transform_reduce_e.cuh>
 #include <patterns/transform_reduce_v.cuh>
 
+// "FIXME": remove the guards below and references to CUCO_STATIC_MAP_DEFINED
+//
+// cuco/static_map.cuh depends on features not supported on or before Pascal.
+//
+// If we build for sm_60 or before, the inclusion of cuco/static_map.cuh wil
+// result in compilation errors.
+//
+// If we're Pascal or before we do nothing here and will suppress including
+// some code below.  If we are later than Pascal we define CUCO_STATIC_MAP_DEFINED
+// which will result in the full implementation being pulled in.
+//
+#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 700
+#else
+#define CUCO_STATIC_MAP_DEFINED
+#include <cuco/static_map.cuh>
+#endif
+
 //#define TIMING
 
 #ifdef TIMING
@@ -44,6 +60,7 @@ namespace experimental {
 
 namespace detail {
 
+#ifdef CUCO_STATIC_MAP_DEFINED
 template <typename data_t>
 struct create_cuco_pair_t {
   cuco::pair_type<data_t, data_t> __device__ operator()(data_t data)
@@ -54,6 +71,7 @@ struct create_cuco_pair_t {
     return tmp;
   }
 };
+#endif
 
 //
 // These classes should allow cuco::static_map to generate hash tables of
@@ -443,7 +461,9 @@ class Louvain {
                                                  weight_t resolution)
   {
     size_t num_level{0};
+    weight_t best_modularity = weight_t{-1};
 
+#ifdef CUCO_STATIC_MAP_DEFINED
     weight_t total_edge_weight;
     total_edge_weight = experimental::transform_reduce_e(
       handle_,
@@ -453,8 +473,6 @@ class Louvain {
       [] __device__(auto, auto, weight_t wt, auto, auto) { return wt; },
       weight_t{0});
 
-    weight_t best_modularity = weight_t{-1};
-
     //
     //  Initialize every cluster to reference each vertex to itself
     //
@@ -480,6 +498,7 @@ class Louvain {
     }
 
     timer_display(std::cout);
+#endif
 
     return std::make_pair(num_level, best_modularity);
   }
@@ -593,6 +612,7 @@ class Louvain {
     }
   }
 
+#ifdef CUCO_STATIC_MAP_DEFINED
   virtual weight_t update_clustering(weight_t total_edge_weight, weight_t resolution)
   {
     timer_start("update_clustering");
@@ -1616,6 +1636,7 @@ class Louvain {
     cugraph::detail::offsets_to_indices(
       current_graph_view_.offsets(), local_num_rows_, src_indices_v_.data().get());
   }
+#endif
 
   std::
     tuple<rmm::device_vector<vertex_t>, rmm::device_vector<vertex_t>, rmm::device_vector<weight_t>>

cpp/tests/CMakeLists.txt

@@ -302,12 +302,11 @@ ConfigureTest(EXPERIMENTAL_PAGERANK_TEST "${EXPERIMENTAL_PAGERANK_TEST_SRCS}" ""
 ###################################################################################################
 # - Experimental LOUVAIN tests -------------------------------------------------------------------
 
-# FIXME: Re-enable once failures are fixed
-#set(EXPERIMENTAL_LOUVAIN_TEST_SRCS
-#    "${CMAKE_SOURCE_DIR}/../thirdparty/mmio/mmio.c"
-#    "${CMAKE_CURRENT_SOURCE_DIR}/experimental/louvain_test.cu")
-#
-#ConfigureTest(EXPERIMENTAL_LOUVAIN_TEST "${EXPERIMENTAL_LOUVAIN_TEST_SRCS}" "")
+set(EXPERIMENTAL_LOUVAIN_TEST_SRCS
+    "${CMAKE_SOURCE_DIR}/../thirdparty/mmio/mmio.c"
+    "${CMAKE_CURRENT_SOURCE_DIR}/experimental/louvain_test.cu")
+
+ConfigureTest(EXPERIMENTAL_LOUVAIN_TEST "${EXPERIMENTAL_LOUVAIN_TEST_SRCS}" "")
 
 ###################################################################################################
 # - Experimental KATZ_CENTRALITY tests ------------------------------------------------------------

cpp/tests/experimental/bfs_test.cpp

@@ -33,8 +33,8 @@
 #include <vector>
 
 template <typename vertex_t, typename edge_t>
-void bfs_reference(edge_t* offsets,
-                   vertex_t* indices,
+void bfs_reference(edge_t const* offsets,
+                   vertex_t const* indices,
                    vertex_t* distances,
                    vertex_t* predecessors,
                    vertex_t num_vertices,

cpp/tests/experimental/katz_centrality_test.cpp

@@ -35,10 +35,10 @@
 #include <vector>
 
 template <typename vertex_t, typename edge_t, typename weight_t, typename result_t>
-void katz_centrality_reference(edge_t* offsets,
-                               vertex_t* indices,
-                               weight_t* weights,
-                               result_t* betas,
+void katz_centrality_reference(edge_t const* offsets,
+                               vertex_t const* indices,
+                               weight_t const* weights,
+                               result_t const* betas,
                                result_t* katz_centralities,
                                vertex_t num_vertices,
                                result_t alpha,
@@ -195,7 +195,8 @@ class Tests_KatzCentrality : public ::testing::TestWithParam<KatzCentrality_Usec
 
     auto threshold_ratio = 1e-3;
     auto threshold_magnitude =
-      (epsilon / static_cast<result_t>(graph_view.get_number_of_vertices())) * threshold_ratio;
+      (1.0 / static_cast<result_t>(graph_view.get_number_of_vertices())) *
+      threshold_ratio;  // skip comparison for low Katz Centrality verties (lowly ranked vertices)
     auto nearly_equal = [threshold_ratio, threshold_magnitude](auto lhs, auto rhs) {
       auto diff = std::abs(lhs - rhs);
       return (diff < std::max(lhs, rhs) * threshold_ratio) || (diff < threshold_magnitude);