Add microkernels for fixed-width and fixed-width dictionary in Parquet decode

cpp/CMakeLists.txt

@@ -411,6 +411,7 @@ add_library(
   src/io/parquet/reader_impl_helpers.cpp
   src/io/parquet/reader_impl_preprocess.cu
   src/io/parquet/writer_impl.cu
+  src/io/parquet/decode_fixed.cu
   src/io/statistics/orc_column_statistics.cu
   src/io/statistics/parquet_column_statistics.cu
   src/io/text/byte_range_info.cpp

cpp/src/io/parquet/decode_fixed.hpp

@@ -0,0 +1,42 @@
+/*
+ * Copyright (c) 2024, NVIDIA CORPORATION.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#pragma once
+
+#include "parquet_gpu.hpp"
+
+namespace cudf {
+namespace io {
+namespace parquet {
+namespace detail {
+void DecodePageDataFixed(cudf::detail::hostdevice_vector<PageInfo>& pages,
+                         cudf::detail::hostdevice_vector<ColumnChunkDesc> const& chunks,
+                         std::size_t num_rows,
+                         size_t min_row,
+                         int level_type_size,
+                         rmm::cuda_stream_view stream);
+
+void DecodePageDataFixedDict(cudf::detail::hostdevice_vector<PageInfo>& pages,
+                             cudf::detail::hostdevice_vector<ColumnChunkDesc> const& chunks,
+                             std::size_t num_rows,
+                             size_t min_row,
+                             int level_type_size,
+                             rmm::cuda_stream_view stream);
+
+}  // namespace detail
+}  // namespace parquet
+}  // namespace io
+}  // namespace cudf

cpp/src/io/parquet/decode_preprocess.cu

@@ -342,8 +342,8 @@ CUDF_KERNEL void __launch_bounds__(preprocess_block_size)
   // the level stream decoders
   __shared__ rle_run<level_t> def_runs[rle_run_buffer_size];
   __shared__ rle_run<level_t> rep_runs[rle_run_buffer_size];
-  rle_stream<level_t, preprocess_block_size> decoders[level_type::NUM_LEVEL_TYPES] = {{def_runs},
-                                                                                      {rep_runs}};
+  rle_stream<level_t, preprocess_block_size, rolling_buf_size>
+    decoders[level_type::NUM_LEVEL_TYPES] = {{def_runs}, {rep_runs}};
 
   // setup page info
   if (!setupLocalPageInfo(
@@ -353,20 +353,17 @@ CUDF_KERNEL void __launch_bounds__(preprocess_block_size)
 
   // initialize the stream decoders (requires values computed in setupLocalPageInfo)
   // the size of the rolling batch buffer
-  int const max_batch_size = rolling_buf_size;
-  level_t* rep             = reinterpret_cast<level_t*>(pp->lvl_decode_buf[level_type::REPETITION]);
-  level_t* def             = reinterpret_cast<level_t*>(pp->lvl_decode_buf[level_type::DEFINITION]);
+  level_t* rep = reinterpret_cast<level_t*>(pp->lvl_decode_buf[level_type::REPETITION]);
+  level_t* def = reinterpret_cast<level_t*>(pp->lvl_decode_buf[level_type::DEFINITION]);
   decoders[level_type::DEFINITION].init(s->col.level_bits[level_type::DEFINITION],
                                         s->abs_lvl_start[level_type::DEFINITION],
                                         s->abs_lvl_end[level_type::DEFINITION],
-                                        max_batch_size,
                                         def,
                                         s->page.num_input_values);
   if (has_repetition) {
     decoders[level_type::REPETITION].init(s->col.level_bits[level_type::REPETITION],
                                           s->abs_lvl_start[level_type::REPETITION],
                                           s->abs_lvl_end[level_type::REPETITION],
-                                          max_batch_size,
                                           rep,
                                           s->page.num_input_values);
   }