Merge branch 'branch-25.02' into imprv-numba_cuda_bump

rapidsai · Dec 23, 2024 · f8325f4 · f8325f4
2 parents 4b50a89 + f7f7e33
commit f8325f4
Show file tree

Hide file tree

Showing 140 changed files with 3,484 additions and 1,473 deletions.
diff --git a/.github/workflows/pr.yaml b/.github/workflows/pr.yaml
@@ -13,6 +13,7 @@ jobs:
   # Please keep pr-builder as the top job here
   pr-builder:
     needs:
+      - check-nightly-ci
       - changed-files
       - checks
       - conda-cpp-build
@@ -54,6 +55,18 @@ jobs:
       - name: Telemetry setup
         if: ${{ vars.TELEMETRY_ENABLED == 'true' }}
         uses: rapidsai/shared-actions/telemetry-dispatch-stash-base-env-vars@main
+  check-nightly-ci:
+    # Switch to ubuntu-latest once it defaults to a version of Ubuntu that
+    # provides at least Python 3.11 (see
+    # https://docs.python.org/3/library/datetime.html#datetime.date.fromisoformat)
+    runs-on: ubuntu-24.04
+    env:
+      RAPIDS_GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+    steps:
+      - name: Check if nightly CI is passing
+        uses: rapidsai/shared-actions/check_nightly_success/dispatch@main
+        with:
+          repo: cudf
   changed-files:
     secrets: inherit
     needs: telemetry-setup

diff --git a/conda/environments/all_cuda-118_arch-x86_64.yaml b/conda/environments/all_cuda-118_arch-x86_64.yaml
@@ -71,7 +71,7 @@ dependencies:
 - ptxcompiler
 - pyarrow>=14.0.0,<19.0.0a0
 - pydata-sphinx-theme!=0.14.2
-- pynvml>=11.4.1,<12.0.0a0
+- pynvml>=12.0.0,<13.0.0a0
 - pytest-benchmark
 - pytest-cases>=3.8.2
 - pytest-cov

diff --git a/conda/environments/all_cuda-125_arch-x86_64.yaml b/conda/environments/all_cuda-125_arch-x86_64.yaml
@@ -69,7 +69,7 @@ dependencies:
 - pyarrow>=14.0.0,<19.0.0a0
 - pydata-sphinx-theme!=0.14.2
 - pynvjitlink>=0.0.0a0
-- pynvml>=11.4.1,<12.0.0a0
+- pynvml>=12.0.0,<13.0.0a0
 - pytest-benchmark
 - pytest-cases>=3.8.2
 - pytest-cov

diff --git a/conda/recipes/dask-cudf/meta.yaml b/conda/recipes/dask-cudf/meta.yaml
@@ -43,7 +43,7 @@ requirements:
   run:
     - python
     - cudf ={{ version }}
-    - pynvml >=11.4.1,<12.0.0a0
+    - pynvml >=12.0.0,<13.0.0a0
     - rapids-dask-dependency ={{ minor_version }}
     - {{ pin_compatible('cuda-version', max_pin='x', min_pin='x') }}
 

diff --git a/cpp/CMakeLists.txt b/cpp/CMakeLists.txt
@@ -446,14 +446,15 @@ add_library(
   src/groupby/sort/group_quantiles.cu
   src/groupby/sort/group_std.cu
   src/groupby/sort/group_sum.cu
-  src/groupby/sort/scan.cpp
   src/groupby/sort/group_count_scan.cu
   src/groupby/sort/group_max_scan.cu
   src/groupby/sort/group_min_scan.cu
   src/groupby/sort/group_product_scan.cu
   src/groupby/sort/group_rank_scan.cu
   src/groupby/sort/group_replace_nulls.cu
   src/groupby/sort/group_sum_scan.cu
+  src/groupby/sort/host_udf_aggregation.cpp
+  src/groupby/sort/scan.cpp
   src/groupby/sort/sort_helper.cu
   src/hash/md5_hash.cu
   src/hash/murmurhash3_x86_32.cu

diff --git a/cpp/include/cudf/aggregation.hpp b/cpp/include/cudf/aggregation.hpp
@@ -110,8 +110,9 @@ class aggregation {
     COLLECT_SET,     ///< collect values into a list without duplicate entries
     LEAD,            ///< window function, accesses row at specified offset following current row
     LAG,             ///< window function, accesses row at specified offset preceding current row
-    PTX,             ///< PTX  UDF based reduction
-    CUDA,            ///< CUDA UDF based reduction
+    PTX,             ///< PTX  based UDF aggregation
+    CUDA,            ///< CUDA based UDF aggregation
+    HOST_UDF,        ///< host based UDF aggregation
     MERGE_LISTS,     ///< merge multiple lists values into one list
     MERGE_SETS,      ///< merge multiple lists values into one list then drop duplicate entries
     MERGE_M2,        ///< merge partial values of M2 aggregation,
@@ -120,7 +121,7 @@ class aggregation {
     TDIGEST,         ///< create a tdigest from a set of input values
     MERGE_TDIGEST,   ///< create a tdigest by merging multiple tdigests together
     HISTOGRAM,       ///< compute frequency of each element
-    MERGE_HISTOGRAM  ///< merge partial values of HISTOGRAM aggregation,
+    MERGE_HISTOGRAM  ///< merge partial values of HISTOGRAM aggregation
   };
 
   aggregation() = delete;
@@ -599,6 +600,18 @@ std::unique_ptr<Base> make_udf_aggregation(udf_type type,
                                            std::string const& user_defined_aggregator,
                                            data_type output_type);
 
+// Forward declaration of `host_udf_base` for the factory function of `HOST_UDF` aggregation.
+struct host_udf_base;
+
+/**
+ * @brief Factory to create a HOST_UDF aggregation.
+ *
+ * @param host_udf An instance of a class derived from `host_udf_base` to perform aggregation
+ * @return A HOST_UDF aggregation object
+ */
+template <typename Base = aggregation>
+std::unique_ptr<Base> make_host_udf_aggregation(std::unique_ptr<host_udf_base> host_udf);
+
 /**
  * @brief Factory to create a MERGE_LISTS aggregation.
  *