fix merge conflict

build.sh

@@ -148,10 +148,8 @@ function buildLibCudfJniInDocker {
                 -DCUDF_CPP_BUILD_DIR=$workspaceRepoDir/java/target/libcudf-cmake-build \
                 -DCUDA_STATIC_RUNTIME=ON \
                 -DPER_THREAD_DEFAULT_STREAM=ON \
-                -DRMM_LOGGING_LEVEL=OFF \
                 -DUSE_GDS=ON \
                 -DGPU_ARCHS=${CUDF_CMAKE_CUDA_ARCHITECTURES} \
-                -DCUDF_JNI_ARROW_STATIC=ON \
                 -DCUDF_JNI_LIBCUDF_STATIC=ON \
                 -Dtest=*,!CuFileTest"
 }

cpp/CMakeLists.txt

@@ -155,6 +155,8 @@ include(cmake/thirdparty/get_gtest.cmake)
 include(cmake/Modules/JitifyPreprocessKernels.cmake)
 # find cuFile
 include(cmake/Modules/FindcuFile.cmake)
+# find KvikIO
+include(cmake/thirdparty/get_kvikio.cmake)
 
 # Workaround until https://github.com/rapidsai/rapids-cmake/issues/176 is resolved
 if(NOT BUILD_SHARED_LIBS)
@@ -344,10 +346,12 @@ add_library(
   src/join/join.cu
   src/join/join_utils.cu
   src/join/mixed_join.cu
-  src/join/mixed_join_kernels.cu
+  src/join/mixed_join_kernel.cu
+  src/join/mixed_join_kernel_nulls.cu
   src/join/mixed_join_kernels_semi.cu
   src/join/mixed_join_semi.cu
-  src/join/mixed_join_size_kernels.cu
+  src/join/mixed_join_size_kernel.cu
+  src/join/mixed_join_size_kernel_nulls.cu
   src/join/mixed_join_size_kernels_semi.cu
   src/join/semi_join.cu
   src/lists/contains.cu
@@ -585,7 +589,7 @@ add_dependencies(cudf jitify_preprocess_run)
 target_link_libraries(
   cudf
   PUBLIC ${ARROW_LIBRARIES} libcudacxx::libcudacxx cudf::Thrust rmm::rmm
-  PRIVATE cuco::cuco ZLIB::ZLIB nvcomp::nvcomp
+  PRIVATE cuco::cuco ZLIB::ZLIB nvcomp::nvcomp kvikio::kvikio
 )
 
 # Add Conda library, and include paths if specified

cpp/benchmarks/copying/contiguous_split.cu

@@ -17,6 +17,7 @@
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>
+
 #include <cudf_test/column_wrapper.hpp>
 
 #include <cudf/column/column.hpp>

cpp/benchmarks/copying/copy_if_else.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>

cpp/benchmarks/hashing/hash.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>

cpp/benchmarks/io/csv/csv_reader.cpp

@@ -14,8 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>
@@ -52,6 +50,7 @@ void BM_csv_read_varying_input(benchmark::State& state)
 
   auto mem_stats_logger = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);  // flush_l2_cache = true, stream = 0
     cudf_io::read_csv(read_options);
   }
@@ -98,6 +97,7 @@ void BM_csv_read_varying_options(benchmark::State& state)
   cudf::size_type const chunk_row_cnt = view.num_rows() / num_chunks;
   auto mem_stats_logger               = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);  // flush_l2_cache = true, stream = 0
     for (int32_t chunk = 0; chunk < num_chunks; ++chunk) {
       // only read the header in the first chunk

cpp/benchmarks/io/csv/csv_writer.cpp

@@ -14,8 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>

cpp/benchmarks/io/cuio_common.cpp

@@ -141,3 +141,31 @@ std::vector<cudf::size_type> segments_in_chunk(int num_segments, int num_chunks,
 
   return selected_segments;
 }
+
+// Executes the command and returns stderr output
+std::string exec_cmd(std::string_view cmd)
+{
+  // Switch stderr and stdout to only capture stderr
+  auto const redirected_cmd = std::string{"( "}.append(cmd).append(" 3>&2 2>&1 1>&3) 2>/dev/null");
+  std::unique_ptr<FILE, decltype(&pclose)> pipe(popen(redirected_cmd.c_str(), "r"), pclose);
+  CUDF_EXPECTS(pipe != nullptr, "popen() failed");
+
+  std::array<char, 128> buffer;
+  std::string error_out;
+  while (fgets(buffer.data(), buffer.size(), pipe.get()) != nullptr) {
+    error_out += buffer.data();
+  }
+  return error_out;
+}
+
+void try_drop_l3_cache()
+{
+  static bool is_drop_cache_enabled = std::getenv("CUDF_BENCHMARK_DROP_CACHE") != nullptr;
+  if (not is_drop_cache_enabled) { return; }
+
+  std::array drop_cache_cmds{"/sbin/sysctl vm.drop_caches=3", "sudo /sbin/sysctl vm.drop_caches=3"};
+  CUDF_EXPECTS(std::any_of(drop_cache_cmds.cbegin(),
+                           drop_cache_cmds.cend(),
+                           [](auto& cmd) { return exec_cmd(cmd).empty(); }),
+               "Failed to execute the drop cache command");
+}

cpp/benchmarks/io/cuio_common.hpp

@@ -16,12 +16,12 @@
 
 #pragma once
 
+#include <cudf_test/file_utilities.hpp>
+
 #include <cudf/io/data_sink.hpp>
 #include <cudf/io/datasource.hpp>
 #include <cudf/io/types.hpp>
 
-#include <cudf_test/file_utilities.hpp>
-
 using cudf::io::io_type;
 
 #define RD_BENCHMARK_DEFINE_ALL_SOURCES(benchmark, name, type_or_group)                  \
@@ -132,3 +132,13 @@ std::vector<std::string> select_column_names(std::vector<std::string> const& col
  * The segments could be Parquet row groups or ORC stripes.
  */
 std::vector<cudf::size_type> segments_in_chunk(int num_segments, int num_chunks, int chunk);
+
+/**
+ * @brief Drops L3 cache if `CUDF_BENCHMARK_DROP_CACHE` environment variable is set.
+ *
+ * Has no effect if the environment variable is not set.
+ * May require sudo access ro run successfully.
+ *
+ * @throw cudf::logic_error if the environment variable is set and the command fails
+ */
+void try_drop_l3_cache();

cpp/benchmarks/io/orc/orc_reader.cpp

@@ -14,8 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>
@@ -60,6 +58,7 @@ void BM_orc_read_varying_input(benchmark::State& state)
 
   auto mem_stats_logger = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);  // flush_l2_cache = true, stream = 0
     cudf_io::read_orc(read_opts);
   }
@@ -117,6 +116,7 @@ void BM_orc_read_varying_options(benchmark::State& state)
   cudf::size_type const chunk_row_cnt = view.num_rows() / num_chunks;
   auto mem_stats_logger               = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);  // flush_l2_cache = true, stream = 0
 
     cudf::size_type rows_read = 0;

cpp/benchmarks/io/orc/orc_writer.cpp

@@ -14,15 +14,13 @@
  * limitations under the License.
  */
 
-#include "cudf/io/types.hpp"
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>
 
 #include <cudf/io/orc.hpp>
+#include <cudf/io/types.hpp>
 
 // to enable, run cmake with -DBUILD_BENCHMARKS=ON
 

cpp/benchmarks/io/parquet/parquet_reader.cpp

@@ -14,8 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>
@@ -60,6 +58,7 @@ void BM_parq_read_varying_input(benchmark::State& state)
 
   auto mem_stats_logger = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer const raii(state, true);  // flush_l2_cache = true, stream = 0
     cudf_io::read_parquet(read_opts);
   }
@@ -117,6 +116,7 @@ void BM_parq_read_varying_options(benchmark::State& state)
   cudf::size_type const chunk_row_cnt = view.num_rows() / num_chunks;
   auto mem_stats_logger               = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);  // flush_l2_cache = true, stream = 0
 
     cudf::size_type rows_read = 0;

cpp/benchmarks/io/parquet/parquet_writer.cpp

@@ -14,8 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>

cpp/benchmarks/io/parquet/parquet_writer_chunks.cpp

@@ -14,21 +14,14 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
-
-#include <cudf/column/column.hpp>
-#include <cudf/table/table.hpp>
-
-#include <cudf_test/base_fixture.hpp>
-#include <cudf_test/column_utilities.hpp>
-#include <cudf_test/column_wrapper.hpp>
-
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/io/cuio_common.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>
 
+#include <cudf/column/column.hpp>
 #include <cudf/io/parquet.hpp>
+#include <cudf/table/table.hpp>
 
 // to enable, run cmake with -DBUILD_BENCHMARKS=ON
 

cpp/benchmarks/io/text/multibyte_split.cpp

@@ -19,10 +19,9 @@
 #include <benchmarks/io/cuio_common.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>
 
-#include <cudf_test/column_wrapper.hpp>
-
 #include <cudf_test/file_utilities.hpp>
 
+#include <cudf/column/column_factories.hpp>
 #include <cudf/io/text/data_chunk_source_factories.hpp>
 #include <cudf/io/text/multibyte_split.hpp>
 #include <cudf/scalar/scalar_factories.hpp>
@@ -38,8 +37,6 @@
 #include <fstream>
 #include <memory>
 
-using cudf::test::fixed_width_column_wrapper;
-
 temp_directory const temp_dir("cudf_gbench");
 
 enum data_chunk_source_type {
@@ -137,6 +134,7 @@ static void BM_multibyte_split(benchmark::State& state)
 
   auto mem_stats_logger = cudf::memory_stats_logger();
   for (auto _ : state) {
+    try_drop_l3_cache();
     cuda_event_timer raii(state, true);
     auto output = cudf::io::text::multibyte_split(*source, delim);
   }

cpp/benchmarks/iterator/iterator.cu

@@ -14,13 +14,14 @@
  * limitations under the License.
  */
 
-#include "../fixture/benchmark_fixture.hpp"
-#include "../synchronization/synchronization.hpp"
+#include <benchmarks/fixture/benchmark_fixture.hpp>
+#include <benchmarks/synchronization/synchronization.hpp>
+
+#include <cudf_test/column_wrapper.hpp>
 
 #include <cudf/detail/iterator.cuh>
 #include <cudf/detail/utilities/device_operators.cuh>
 #include <cudf/detail/utilities/vector_factories.hpp>
-#include <cudf_test/column_wrapper.hpp>
 
 #include <rmm/device_uvector.hpp>
 
@@ -31,8 +32,6 @@
 #include <thrust/iterator/transform_iterator.h>
 #include <thrust/pair.h>
 
-#include <benchmark/benchmark.h>
-
 #include <random>
 
 template <typename T>

cpp/benchmarks/join/join_common.hpp

@@ -21,10 +21,8 @@
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>
 
-#include <cudf_test/base_fixture.hpp>
-#include <cudf_test/column_wrapper.hpp>
-
 #include <cudf/ast/expressions.hpp>
+#include <cudf/column/column_factories.hpp>
 #include <cudf/detail/valid_if.cuh>
 #include <cudf/filling.hpp>
 #include <cudf/join.hpp>

cpp/benchmarks/merge/merge.cpp

@@ -14,18 +14,15 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
+#include <benchmarks/fixture/benchmark_fixture.hpp>
+#include <benchmarks/synchronization/synchronization.hpp>
+#include <cudf_test/column_wrapper.hpp>
 
 #include <cudf/column/column.hpp>
+#include <cudf/merge.hpp>
 #include <cudf/table/table.hpp>
 #include <cudf/table/table_view.hpp>
 
-#include <cudf/merge.hpp>
-
-#include <benchmarks/fixture/benchmark_fixture.hpp>
-#include <benchmarks/synchronization/synchronization.hpp>
-#include <cudf_test/column_wrapper.hpp>
-
 #include <thrust/iterator/constant_iterator.h>
 
 #include <random>

cpp/benchmarks/null_mask/set_null_mask.cpp

@@ -19,8 +19,6 @@
 
 #include <cudf/null_mask.hpp>
 
-#include <benchmark/benchmark.h>
-
 class SetNullmask : public cudf::benchmark {
 };
 

cpp/benchmarks/reduction/scan.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>

cpp/benchmarks/replace/clamp.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>

cpp/benchmarks/replace/nans.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
-#include <benchmark/benchmark.h>
 #include <benchmarks/common/generate_input.hpp>
 #include <benchmarks/fixture/benchmark_fixture.hpp>
 #include <benchmarks/synchronization/synchronization.hpp>