Merge branch 'main' into cupy-pin

jameslamb · web-flow · commit c60693c1aa01 · 2026-06-29T21:16:13.000-05:00
diff --git a/.github/CODEOWNERS b/.github/CODEOWNERS
@@ -1,30 +1,30 @@
 #cpp code owners
-cpp/               @rapidsai/raft-cpp-codeowners
+cpp/               @NVIDIA/raft-cpp-codeowners
 
 #python code owners
-python/            @rapidsai/raft-python-codeowners
+python/            @NVIDIA/raft-python-codeowners
 
 #cmake code owners
-CMakeLists.txt     @rapidsai/raft-cmake-codeowners
-**/cmake/          @rapidsai/raft-cmake-codeowners
-*.cmake            @rapidsai/raft-cmake-codeowners
-python/setup.py    @rapidsai/raft-cmake-codeowners
-build.sh           @rapidsai/raft-cmake-codeowners
-**/build.sh        @rapidsai/raft-cmake-codeowners
+CMakeLists.txt     @NVIDIA/raft-cmake-codeowners
+**/cmake/          @NVIDIA/raft-cmake-codeowners
+*.cmake            @NVIDIA/raft-cmake-codeowners
+python/setup.py    @NVIDIA/raft-cmake-codeowners
+build.sh           @NVIDIA/raft-cmake-codeowners
+**/build.sh        @NVIDIA/raft-cmake-codeowners
 
 #CI code owners
-/.github/                @rapidsai/ci-codeowners
-/ci/                     @rapidsai/ci-codeowners
-/.shellcheckrc           @rapidsai/ci-codeowners
-/.coderabbit.yaml        @rapidsai/ci-codeowners
+/.github/                @NVIDIA/adi-ci-codeowners
+/ci/                     @NVIDIA/adi-ci-codeowners
+/.shellcheckrc           @NVIDIA/adi-ci-codeowners
+/.coderabbit.yaml        @NVIDIA/adi-ci-codeowners
 
 #packaging code owners
-/.pre-commit-config.yaml @rapidsai/packaging-codeowners
-/.devcontainer/          @rapidsai/packaging-codeowners
-/conda/                  @rapidsai/packaging-codeowners
-dependencies.yaml        @rapidsai/packaging-codeowners
-/build.sh                @rapidsai/packaging-codeowners
-pyproject.toml           @rapidsai/packaging-codeowners
+/.pre-commit-config.yaml @NVIDIA/adi-packaging-codeowners
+/.devcontainer/          @NVIDIA/adi-packaging-codeowners
+/conda/                  @NVIDIA/adi-packaging-codeowners
+dependencies.yaml        @NVIDIA/adi-packaging-codeowners
+/build.sh                @NVIDIA/adi-packaging-codeowners
+pyproject.toml           @NVIDIA/adi-packaging-codeowners
 
 # Ops code owners
-/SECURITY.md             @rapidsai/ops-codeowners
+/SECURITY.md             @NVIDIA/adi-ops-codeowners
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
@@ -68,8 +68,6 @@ jobs:
         with:
           repo: ${{ github.repository }}
           target-branch: ${{ fromJSON(steps.get-pr-info.outputs.pr-info).base.ref }}
-          # Tracking Issue: https://github.com/rapidsai/raft/issues/3037
-          max-days-without-success: 30
   changed-files:
     needs: telemetry-setup
     permissions:
@@ -277,6 +275,7 @@ jobs:
     if: fromJSON(needs.changed-files.outputs.changed_file_groups).test_python_conda
     with:
       build_type: pull-request
+      run_codecov: false
       script: ci/test_python.sh
   docs-build:
     needs: [conda-python-build, changed-files]
diff --git a/.github/workflows/test.yaml b/.github/workflows/test.yaml
@@ -68,6 +68,7 @@ jobs:
       build_type: ${{ inputs.build_type }}
       branch: ${{ inputs.branch }}
       date: ${{ inputs.date }}
+      run_codecov: false
       script: ci/test_python.sh
       sha: ${{ inputs.sha }}
   wheel-tests-pylibraft:
diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml
@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION.
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
 repos:
@@ -85,7 +85,7 @@ repos:
                   (?x)
                     ^CHANGELOG[.]md$|
       - repo: https://github.com/rapidsai/pre-commit-hooks
-        rev: v1.4.2
+        rev: v1.6.0
         hooks:
           - id: verify-copyright
             name: verify-copyright-raft
@@ -145,7 +145,7 @@ repos:
               )
           - id: verify-alpha-spec
           - id: verify-codeowners
-            args: [--fix, --project-prefix=raft]
+            args: [--fix, --org=NVIDIA, --project-prefix=raft]
           - id: verify-hardcoded-version
             exclude: |
               (?x)
@@ -171,7 +171,7 @@ repos:
                 ^conda/environments/|
                 (^|/)UCXX_BRANCH$|
                 (^|/)RAPIDS_BRANCH$|
-                [.](md|rst)$
+                [.](md|rst|svg)$
           - id: verify-pyproject-license
             # ignore the top-level pyproject.toml, which doesn't
             # have or need a [project] table
diff --git a/build.sh b/build.sh
@@ -48,6 +48,8 @@ HELP="$0 [<target> ...] [<flag> ...] [--cmake-args=\"<args>\"] [--cache-tool=<to
    --cmake-args=\\\"<args>\\\" - pass arbitrary list of CMake configuration options (escape all quotes in argument)
    --cache-tool=<tool>         - pass the build cache tool (eg: ccache, sccache, distcc) that will be used
                                  to speedup the build process.
+   --logging-level=<level>     - set macro \"RAFT_LOG_ACTIVE_LEVEL\" to RAPIDS_LOGGER_LOG_LEVEL_{given_value}.
+                                 Valid values are (TRACE, DEBUG, INFO, WARN, ERROR, CRITICAL, OFF)
    --time                      - Enable nvcc compilation time logging into cpp/build/nvcc_compile_log.csv.
                                  Results can be interpreted with cpp/scripts/analyze_nvcc_log.py
    -h                          - print this text
@@ -72,6 +74,7 @@ COMPILE_LIBRARY=OFF
 INSTALL_TARGET=install
 BUILD_REPORT_METRICS=""
 BUILD_REPORT_INCL_CACHE_STATS=OFF
+LIBRAFT_LOGGING_LEVEL=INFO
 
 TEST_TARGETS="CORE_TEST;\
 CORE_TEST_NOCUDA;\
@@ -193,6 +196,21 @@ function limitBench {
     fi
 }
 
+function loggingLevel {
+    if [[ -n $(echo "$ARGS" | { grep -E "\-\-logging\-level" || true; } ) ]]; then
+        LIBRAFT_LOGGING_LEVEL=$(echo "$ARGS" | sed -e 's/.*--logging-level[= ]//' -e 's/ .*//')
+        if [[ -n ${LIBRAFT_LOGGING_LEVEL} ]]; then
+            ARGS="${ARGS//--logging-level[= ]${LIBRAFT_LOGGING_LEVEL}/}"
+            VALID_LEVELS="TRACE DEBUG INFO WARN ERROR CRITICAL OFF"
+            if [[ ! " ${VALID_LEVELS} " == *" ${LIBRAFT_LOGGING_LEVEL} "* ]]; then
+                echo "Invalid --logging-level value: '${LIBRAFT_LOGGING_LEVEL}'. Valid values are: ${VALID_LEVELS}"
+                exit 1
+            fi
+            echo "Set LIBRAFT_LOGGING_LEVEL to ${LIBRAFT_LOGGING_LEVEL}"
+        fi
+    fi
+}
+
 function buildMetrics {
     # Check for multiple build-metrics options
     if [[ $(echo "$ARGS" | { grep -Eo "\-\-build\-metrics" || true; } | wc -l ) -gt 1 ]]; then
@@ -223,6 +241,7 @@ if (( NUMARGS != 0 )); then
     cacheTool
     limitTests
     limitBench
+    loggingLevel
     buildMetrics
     for a in ${ARGS}; do
         if ! (echo " ${VALIDARGS} " | grep -q " ${a} "); then
@@ -406,6 +425,7 @@ if (( NUMARGS == 0 )) || hasArg libraft || hasArg docs || hasArg tests || hasArg
           -DBUILD_TESTS=${BUILD_TESTS} \
           -DBUILD_PRIMS_BENCH=${BUILD_PRIMS_BENCH} \
           -DCMAKE_MESSAGE_LOG_LEVEL=${CMAKE_LOG_LEVEL} \
+          -DLIBRAFT_LOGGING_LEVEL="${LIBRAFT_LOGGING_LEVEL}" \
           "${CACHE_ARGS[@]}" \
           "${EXTRA_CMAKE_ARGS[@]}"
 
diff --git a/cpp/include/raft/core/interruptible.hpp b/cpp/include/raft/core/interruptible.hpp
@@ -14,6 +14,7 @@
 
 #include <rmm/cuda_stream_view.hpp>
 
+#include <atomic>
 #include <memory>
 #include <mutex>
 #include <optional>
diff --git a/cpp/include/raft/matrix/detail/gather_inplace.cuh b/cpp/include/raft/matrix/detail/gather_inplace.cuh
@@ -54,7 +54,7 @@ void gatherInplaceImpl(raft::resources const& handle,
                       transform_op,
                       batch_offset,
                       map_length,
-                      cols_per_batch = raft::util::FastIntDiv(cols_per_batch),
+                      cols_per_batch = raft::util::FastIntDiv<IndexT>(cols_per_batch),
                       n,
                       ld] __device__(auto idx) {
       IndexT row   = idx / cols_per_batch;
@@ -74,7 +74,7 @@ void gatherInplaceImpl(raft::resources const& handle,
                     scratch_space = scratch_space.data_handle(),
                     batch_offset,
                     map_length,
-                    cols_per_batch = raft::util::FastIntDiv(cols_per_batch),
+                    cols_per_batch = raft::util::FastIntDiv<IndexT>(cols_per_batch),
                     n,
                     ld] __device__(auto idx) {
       IndexT row                           = idx / cols_per_batch;
diff --git a/cpp/include/raft/matrix/detail/scatter_inplace.cuh b/cpp/include/raft/matrix/detail/scatter_inplace.cuh
@@ -77,7 +77,7 @@ void scatterInplaceImpl(
     auto copy_op = [inout = inout.data_handle(),
                     map   = map.data_handle(),
                     batch_offset,
-                    cols_per_batch = raft::util::FastIntDiv(cols_per_batch),
+                    cols_per_batch = raft::util::FastIntDiv<IndexT>(cols_per_batch),
                     n] __device__(auto idx) {
       IndexT row = idx / cols_per_batch;
       IndexT col = idx % cols_per_batch;
@@ -92,7 +92,7 @@ void scatterInplaceImpl(
                        map           = map.data_handle(),
                        scratch_space = scratch_space.data_handle(),
                        batch_offset,
-                       cols_per_batch = raft::util::FastIntDiv(cols_per_batch),
+                       cols_per_batch = raft::util::FastIntDiv<IndexT>(cols_per_batch),
                        n] __device__(auto idx) {
       IndexT row     = idx / cols_per_batch;
       IndexT col     = idx % cols_per_batch;
diff --git a/cpp/include/raft/util/fast_int_div.cuh b/cpp/include/raft/util/fast_int_div.cuh
@@ -10,25 +10,34 @@
 
 #include <stdint.h>
 
+#include <type_traits>
+
 namespace raft {
 namespace util {
 
 /**
  * @brief Perform fast integer division and modulo using a known divisor
  * From Hacker's Delight, Second Edition, Chapter 10
  *
- * @note This currently only supports 32b signed integers
+ * @note 32b signed integer is supported.
+ * @note 64b signed integers is supported for an input data up to 2^31
+ * because gpu-non-native int128 is avoided for performance.
  * @todo Extend support for signed divisors
  */
+template <typename IntT>
 struct FastIntDiv {
+  static_assert(std::is_same_v<IntT, int32_t> || std::is_same_v<IntT, int64_t>,
+                "FastIntDiv: IntT must be int32_t or int64_t");
+  using UIntT = std::make_unsigned_t<IntT>;
+
   /**
    * @defgroup HostMethods Ctor's that are accessible only from host
    * @{
    * @brief Host-only ctor's
    * @param _d the divisor
    */
-  FastIntDiv(int _d) : d(_d) { computeScalars(); }
-  FastIntDiv& operator=(int _d)
+  FastIntDiv(IntT _d) : d(_d) { computeScalars(); }
+  FastIntDiv& operator=(IntT _d)
   {
     d = _d;
     computeScalars();
@@ -53,9 +62,9 @@ struct FastIntDiv {
   /** @} */
 
   /** divisor */
-  int d;
+  IntT d;
   /** the term 'm' as found in the reference chapter */
-  unsigned m;
+  UIntT m;
   /** the term 'p' as found in the reference chapter */
   int p;
 
@@ -90,10 +99,11 @@ struct FastIntDiv {
  * @param divisor the denominator
  * @return the quotient
  */
-HDI int operator/(int n, const FastIntDiv& divisor)
+template <typename IntT>
+HDI IntT operator/(IntT n, const FastIntDiv<IntT>& divisor)
 {
   if (divisor.d == 1) return n;
-  int ret = (int64_t(divisor.m) * int64_t(n)) >> divisor.p;
+  IntT ret = (int64_t(divisor.m) * int64_t(n)) >> divisor.p;
   if (n < 0) ++ret;
   return ret;
 }
@@ -105,10 +115,11 @@ HDI int operator/(int n, const FastIntDiv& divisor)
  * @param divisor the denominator
  * @return the remainder
  */
-HDI int operator%(int n, const FastIntDiv& divisor)
+template <typename IntT>
+HDI IntT operator%(IntT n, const FastIntDiv<IntT>& divisor)
 {
-  int quotient  = n / divisor;
-  int remainder = n - quotient * divisor.d;
+  IntT quotient  = n / divisor;
+  IntT remainder = n - quotient * divisor.d;
   return remainder;
 }
 
diff --git a/cpp/tests/CMakeLists.txt b/cpp/tests/CMakeLists.txt
@@ -317,6 +317,7 @@ if(BUILD_TESTS)
     util/bitonic_sort.cu
     util/cudart_utils.cpp
     util/device_atomics.cu
+    util/fast_int_div.cu
     util/integer_utils.cpp
     util/integer_utils.cu
     util/memory_type_dispatcher.cu
diff --git a/cpp/tests/util/fast_int_div.cu b/cpp/tests/util/fast_int_div.cu
@@ -0,0 +1,54 @@
+/*
+ * SPDX-FileCopyrightText: Copyright (c) 2024-2026, NVIDIA CORPORATION.
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include <raft/util/fast_int_div.cuh>
+
+#include <gtest/gtest.h>
+
+#include <cstdint>
+#include <limits>
+#include <vector>
+
+namespace raft::util {
+
+constexpr int64_t kInt32Max = std::numeric_limits<int32_t>::max();
+
+template <typename IntT>
+class FastIntDivTest : public ::testing::Test {
+ protected:
+  void CompareWithNativeDivision()
+  {
+    std::vector<IntT> magnitudes{0, 1, 2, 3, 7, 13, 255, 12345, (1 << 20), kInt32Max};
+    std::vector<IntT> divisors{1, 2, 4, 7, 16, 31, 63, 128, 1000, (1 << 15), kInt32Max};
+
+    for (IntT d : divisors) {
+      FastIntDiv fid(d);
+      for (IntT mag : magnitudes) {
+        for (IntT n : {mag, -mag}) {
+          ASSERT_EQ(n / fid, n / d) << "operator/ mismatch for numerator=" << n << " divisor=" << d;
+          ASSERT_EQ(n % fid, n % d) << "operator% mismatch for numerator=" << n << " divisor=" << d;
+        }
+      }
+    }
+  }
+};
+
+using FastIntDivTypes = ::testing::Types<int32_t, int64_t>;
+TYPED_TEST_CASE(FastIntDivTest, FastIntDivTypes);
+
+TYPED_TEST(FastIntDivTest, CompareWithNativeDivision) { this->CompareWithNativeDivision(); }
+
+TEST(FastIntDiv, Int64NumeratorPastInt32Boundary)
+{
+  for (int64_t d : {129, 772, 1000}) {
+    FastIntDiv fid(d);
+    for (int64_t n : {1LL << 31, 2147704000LL, 3LL << 30}) {  // in [2^31, 2^32)
+      ASSERT_EQ(n / fid, n / d) << "operator/ mismatch for numerator=" << n << " divisor=" << d;
+      ASSERT_EQ(n % fid, n % d) << "operator% mismatch for numerator=" << n << " divisor=" << d;
+    }
+  }
+}
+
+}  // namespace raft::util
