Merge branch 'branch-24.12' into nvtext-pylibcudf-tokenize

rapidsai · Oct 30, 2024 · b720d08 · b720d08
2 parents e958b24 + 5ee7d7c
commit b720d08
Show file tree

Hide file tree

Showing 12 changed files with 744 additions and 166 deletions.
diff --git a/.github/workflows/auto-assign.yml b/.github/workflows/auto-assign.yml
@@ -13,5 +13,5 @@ jobs:
     steps:
       - uses: actions-ecosystem/action-add-assignees@v1
         with:
-          repo_token: "${{ secrets.GITHUB_TOKEN }}"
+          github_token: "${{ secrets.GITHUB_TOKEN }}"
           assignees: ${{ github.actor }}
diff --git a/cpp/benchmarks/ast/transform.cpp b/cpp/benchmarks/ast/transform.cpp
@@ -52,14 +52,14 @@ enum class TreeType {
 template <typename key_type, TreeType tree_type, bool reuse_columns, bool Nullable>
 static void BM_ast_transform(nvbench::state& state)
 {
-  auto const table_size  = static_cast<cudf::size_type>(state.get_int64("table_size"));
+  auto const num_rows    = static_cast<cudf::size_type>(state.get_int64("num_rows"));
   auto const tree_levels = static_cast<cudf::size_type>(state.get_int64("tree_levels"));
 
   // Create table data
   auto const n_cols = reuse_columns ? 1 : tree_levels + 1;
   auto const source_table =
     create_sequence_table(cycle_dtypes({cudf::type_to_id<key_type>()}, n_cols),
-                          row_count{table_size},
+                          row_count{num_rows},
                           Nullable ? std::optional<double>{0.5} : std::nullopt);
   auto table = source_table->view();
 
@@ -99,8 +99,8 @@ static void BM_ast_transform(nvbench::state& state)
   auto const& expression_tree_root = expressions.back();
 
   // Use the number of bytes read from global memory
-  state.add_global_memory_reads<key_type>(static_cast<size_t>(table_size) * (tree_levels + 1));
-  state.add_global_memory_writes<key_type>(table_size);
+  state.add_global_memory_reads<key_type>(static_cast<size_t>(num_rows) * (tree_levels + 1));
+  state.add_global_memory_writes<key_type>(num_rows);
 
   state.exec(nvbench::exec_tag::sync,
              [&](nvbench::launch&) { cudf::compute_column(table, expression_tree_root); });
@@ -109,15 +109,15 @@ static void BM_ast_transform(nvbench::state& state)
 template <cudf::ast::ast_operator cmp_op, cudf::ast::ast_operator reduce_op>
 static void BM_string_compare_ast_transform(nvbench::state& state)
 {
-  auto const string_width    = static_cast<cudf::size_type>(state.get_int64("string_width"));
-  auto const num_rows        = static_cast<cudf::size_type>(state.get_int64("num_rows"));
-  auto const num_comparisons = static_cast<cudf::size_type>(state.get_int64("num_comparisons"));
-  auto const hit_rate        = static_cast<cudf::size_type>(state.get_int64("hit_rate"));
+  auto const string_width = static_cast<cudf::size_type>(state.get_int64("string_width"));
+  auto const num_rows     = static_cast<cudf::size_type>(state.get_int64("num_rows"));
+  auto const tree_levels  = static_cast<cudf::size_type>(state.get_int64("tree_levels"));
+  auto const hit_rate     = static_cast<cudf::size_type>(state.get_int64("hit_rate"));
 
-  CUDF_EXPECTS(num_comparisons > 0, "benchmarks require 1 or more comparisons");
+  CUDF_EXPECTS(tree_levels > 0, "benchmarks require 1 or more comparisons");
 
   // Create table data
-  auto const num_cols = num_comparisons * 2;
+  auto const num_cols = tree_levels * 2;
   std::vector<std::unique_ptr<cudf::column>> columns;
   std::for_each(
     thrust::make_counting_iterator(0), thrust::make_counting_iterator(num_cols), [&](size_t) {
@@ -150,7 +150,7 @@ static void BM_string_compare_ast_transform(nvbench::state& state)
   expressions.emplace_back(cudf::ast::operation(cmp_op, column_refs[0], column_refs[1]));
 
   std::for_each(thrust::make_counting_iterator(1),
-                thrust::make_counting_iterator(num_comparisons),
+                thrust::make_counting_iterator(tree_levels),
                 [&](size_t idx) {
                   auto const& lhs = expressions.back();
                   auto const& rhs = expressions.emplace_back(
@@ -177,7 +177,7 @@ static void BM_string_compare_ast_transform(nvbench::state& state)
   NVBENCH_BENCH(name)                                                                      \
     .set_name(#name)                                                                       \
     .add_int64_axis("tree_levels", {1, 5, 10})                                             \
-    .add_int64_axis("table_size", {100'000, 1'000'000, 10'000'000, 100'000'000})
+    .add_int64_axis("num_rows", {100'000, 1'000'000, 10'000'000, 100'000'000})
 
 AST_TRANSFORM_BENCHMARK_DEFINE(
   ast_int32_imbalanced_unique, int32_t, TreeType::IMBALANCED_LEFT, false, false);
@@ -202,7 +202,7 @@ AST_TRANSFORM_BENCHMARK_DEFINE(
     .set_name(#name)                                                           \
     .add_int64_axis("string_width", {32, 64, 128, 256})                        \
     .add_int64_axis("num_rows", {32768, 262144, 2097152})                      \
-    .add_int64_axis("num_comparisons", {1, 2, 3, 4})                           \
+    .add_int64_axis("tree_levels", {1, 2, 3, 4})                               \
     .add_int64_axis("hit_rate", {50, 100})
 
 AST_STRING_COMPARE_TRANSFORM_BENCHMARK_DEFINE(ast_string_equal_logical_and,

diff --git a/cpp/benchmarks/binaryop/binaryop.cpp b/cpp/benchmarks/binaryop/binaryop.cpp
@@ -40,18 +40,18 @@ enum class TreeType {
 template <typename key_type, TreeType tree_type, bool reuse_columns>
 static void BM_binaryop_transform(nvbench::state& state)
 {
-  auto const table_size{static_cast<cudf::size_type>(state.get_int64("table_size"))};
+  auto const num_rows{static_cast<cudf::size_type>(state.get_int64("num_rows"))};
   auto const tree_levels{static_cast<cudf::size_type>(state.get_int64("tree_levels"))};
 
   // Create table data
   auto const n_cols       = reuse_columns ? 1 : tree_levels + 1;
   auto const source_table = create_sequence_table(
-    cycle_dtypes({cudf::type_to_id<key_type>()}, n_cols), row_count{table_size});
+    cycle_dtypes({cudf::type_to_id<key_type>()}, n_cols), row_count{num_rows});
   cudf::table_view table{*source_table};
 
   // Use the number of bytes read from global memory
-  state.add_global_memory_reads<key_type>(static_cast<size_t>(table_size) * (tree_levels + 1));
-  state.add_global_memory_writes<key_type>(table_size);
+  state.add_global_memory_reads<key_type>(static_cast<size_t>(num_rows) * (tree_levels + 1));
+  state.add_global_memory_writes<key_type>(num_rows);
 
   state.exec(nvbench::exec_tag::sync, [&](nvbench::launch&) {
     // Execute tree that chains additions like (((a + b) + c) + d)
@@ -74,15 +74,15 @@ static void BM_binaryop_transform(nvbench::state& state)
 template <cudf::binary_operator cmp_op, cudf::binary_operator reduce_op>
 static void BM_string_compare_binaryop_transform(nvbench::state& state)
 {
-  auto const string_width    = static_cast<cudf::size_type>(state.get_int64("string_width"));
-  auto const num_rows        = static_cast<cudf::size_type>(state.get_int64("num_rows"));
-  auto const num_comparisons = static_cast<cudf::size_type>(state.get_int64("num_comparisons"));
-  auto const hit_rate        = static_cast<cudf::size_type>(state.get_int64("hit_rate"));
+  auto const string_width = static_cast<cudf::size_type>(state.get_int64("string_width"));
+  auto const num_rows     = static_cast<cudf::size_type>(state.get_int64("num_rows"));
+  auto const tree_levels  = static_cast<cudf::size_type>(state.get_int64("tree_levels"));
+  auto const hit_rate     = static_cast<cudf::size_type>(state.get_int64("hit_rate"));
 
-  CUDF_EXPECTS(num_comparisons > 0, "benchmarks require 1 or more comparisons");
+  CUDF_EXPECTS(tree_levels > 0, "benchmarks require 1 or more comparisons");
 
   // Create table data
-  auto const num_cols = num_comparisons * 2;
+  auto const num_cols = tree_levels * 2;
   std::vector<std::unique_ptr<cudf::column>> columns;
   std::for_each(
     thrust::make_counting_iterator(0), thrust::make_counting_iterator(num_cols), [&](size_t) {
@@ -113,7 +113,7 @@ static void BM_string_compare_binaryop_transform(nvbench::state& state)
       cudf::binary_operation(table.get_column(0), table.get_column(1), cmp_op, bool_type, stream);
     std::for_each(
       thrust::make_counting_iterator(1),
-      thrust::make_counting_iterator(num_comparisons),
+      thrust::make_counting_iterator(tree_levels),
       [&](size_t idx) {
         std::unique_ptr<cudf::column> comparison = cudf::binary_operation(
           table.get_column(idx * 2), table.get_column(idx * 2 + 1), cmp_op, bool_type, stream);
@@ -133,7 +133,7 @@ static void BM_string_compare_binaryop_transform(nvbench::state& state)
   }                                                                                   \
   NVBENCH_BENCH(name)                                                                 \
     .add_int64_axis("tree_levels", {1, 2, 5, 10})                                     \
-    .add_int64_axis("table_size", {100'000, 1'000'000, 10'000'000, 100'000'000})
+    .add_int64_axis("num_rows", {100'000, 1'000'000, 10'000'000, 100'000'000})
 
 BINARYOP_TRANSFORM_BENCHMARK_DEFINE(binaryop_int32_imbalanced_unique,
                                     int32_t,
@@ -158,7 +158,7 @@ BINARYOP_TRANSFORM_BENCHMARK_DEFINE(binaryop_double_imbalanced_unique,
     .set_name(#name)                                                                \
     .add_int64_axis("string_width", {32, 64, 128, 256})                             \
     .add_int64_axis("num_rows", {32768, 262144, 2097152})                           \
-    .add_int64_axis("num_comparisons", {1, 2, 3, 4})                                \
+    .add_int64_axis("tree_levels", {1, 2, 3, 4})                                    \
     .add_int64_axis("hit_rate", {50, 100})
 
 STRING_COMPARE_BINARYOP_TRANSFORM_BENCHMARK_DEFINE(string_compare_binaryop_transform,

diff --git a/cpp/benchmarks/binaryop/compiled_binaryop.cpp b/cpp/benchmarks/binaryop/compiled_binaryop.cpp
@@ -23,10 +23,10 @@
 template <typename TypeLhs, typename TypeRhs, typename TypeOut>
 void BM_compiled_binaryop(nvbench::state& state, cudf::binary_operator binop)
 {
-  auto const table_size = static_cast<cudf::size_type>(state.get_int64("table_size"));
+  auto const num_rows = static_cast<cudf::size_type>(state.get_int64("num_rows"));
 
   auto const source_table = create_random_table(
-    {cudf::type_to_id<TypeLhs>(), cudf::type_to_id<TypeRhs>()}, row_count{table_size});
+    {cudf::type_to_id<TypeLhs>(), cudf::type_to_id<TypeRhs>()}, row_count{num_rows});
 
   auto lhs = cudf::column_view(source_table->get_column(0));
   auto rhs = cudf::column_view(source_table->get_column(1));
@@ -37,9 +37,9 @@ void BM_compiled_binaryop(nvbench::state& state, cudf::binary_operator binop)
   cudf::binary_operation(lhs, rhs, binop, output_dtype);
 
   // use number of bytes read and written to global memory
-  state.add_global_memory_reads<TypeLhs>(table_size);
-  state.add_global_memory_reads<TypeRhs>(table_size);
-  state.add_global_memory_writes<TypeOut>(table_size);
+  state.add_global_memory_reads<TypeLhs>(num_rows);
+  state.add_global_memory_reads<TypeRhs>(num_rows);
+  state.add_global_memory_writes<TypeOut>(num_rows);
 
   state.exec(nvbench::exec_tag::sync,
              [&](nvbench::launch&) { cudf::binary_operation(lhs, rhs, binop, output_dtype); });
@@ -55,7 +55,7 @@ void BM_compiled_binaryop(nvbench::state& state, cudf::binary_operator binop)
   }                                                                           \
   NVBENCH_BENCH(name)                                                         \
     .set_name("compiled_binary_op_" BM_STRINGIFY(name))                       \
-    .add_int64_axis("table_size", {10'000, 100'000, 1'000'000, 10'000'000, 100'000'000})
+    .add_int64_axis("num_rows", {10'000, 100'000, 1'000'000, 10'000'000, 100'000'000})
 
 #define build_name(a, b, c, d) a##_##b##_##c##_##d
 

diff --git a/cpp/doxygen/developer_guide/DEVELOPER_GUIDE.md b/cpp/doxygen/developer_guide/DEVELOPER_GUIDE.md
@@ -1483,6 +1483,17 @@ struct, and therefore `cudf::struct_view` is the data type of a `cudf::column` o
 
 `cudf::type_dispatcher` dispatches to the `struct_view` data type when invoked on a `STRUCT` column.
 
+# Empty Columns
+
+The libcudf columns support empty, typed content. These columns have no data and no validity mask.
+Empty strings or lists columns may or may not contain a child offsets column.
+It is undefined behavior (UB) to access the offsets child of an empty strings or lists column.
+Nested columns like lists and structs may require other children columns to provide the
+nested structure of the empty types.
+
+Use `cudf::make_empty_column()` to create fixed-width and strings columns.
+Use `cudf::empty_like()` to create an empty column from an existing `cudf::column_view`.
+
 # cuIO: file reading and writing
 
 cuIO is a component of libcudf that provides GPU-accelerated reading and writing of data file