undo breaking change in segmented radix sort config (#548)

ROCm · Apr 9, 2024 · 3178ba4 · 3178ba4
1 parent 609ae19
commit 3178ba4
Show file tree

Hide file tree

Showing 8 changed files with 827 additions and 1,056 deletions.
diff --git a/benchmark/benchmark_device_segmented_radix_sort_keys.cpp b/benchmark/benchmark_device_segmented_radix_sort_keys.cpp
@@ -285,6 +285,7 @@ int main(int argc, char* argv[])
  // Add benchmarks
  std::vector<benchmark::internal::Benchmark*> benchmarks;
 #ifdef BENCHMARK_CONFIG_TUNING
+ (void)min_size;
  const int parallel_instance = parser.get<int>("parallel_instance");
  const int parallel_instances = parser.get<int>("parallel_instances");
  config_autotune_register::register_benchmark_subset(benchmarks,

diff --git a/benchmark/benchmark_device_segmented_radix_sort_keys.parallel.hpp b/benchmark/benchmark_device_segmented_radix_sort_keys.parallel.hpp
@@ -291,16 +291,15 @@ struct device_segmented_radix_sort_benchmark_generator
  rocprim::segmented_radix_sort_config<
  LongBits,
  ShortBits,
- BlockSize,
- ItemsPerThread,
- EnableUnpartitionedWarpSort,
+ rocprim::kernel_config<BlockSize, ItemsPerThread>,
  rocprim::WarpSortConfig<LogicalWarpSizeSmall / 2,
  ItemsPerThread / 2,
  BlockSize,
  PartitioningThreshold,
  LogicalWarpSizeSmall,
  ItemsPerThread,
- BlockSize>>>>());
+ BlockSize>,
+ EnableUnpartitionedWarpSort>>>());
  }
  };
 
@@ -321,15 +320,15 @@ struct device_segmented_radix_sort_benchmark_generator
  }
  else
  {
- storage.emplace_back(std::make_unique<device_segmented_radix_sort_benchmark<
-  Key,
-  rocprim::segmented_radix_sort_config<
-  LongBits,
-  ShortBits,
-  BlockSize,
-  ItemsPerThread,
-  EnableUnpartitionedWarpSort,
-  rocprim::DisabledWarpSortConfig>>>());
+ storage.emplace_back(
+ std::make_unique<device_segmented_radix_sort_benchmark<
+ Key,
+ rocprim::segmented_radix_sort_config<
+ LongBits,
+ ShortBits,
+ rocprim::kernel_config<BlockSize, ItemsPerThread>,
+ rocprim::DisabledWarpSortConfig,
+ EnableUnpartitionedWarpSort>>>());
  }
  }
  };

diff --git a/benchmark/benchmark_device_segmented_radix_sort_pairs.cpp b/benchmark/benchmark_device_segmented_radix_sort_pairs.cpp
@@ -309,6 +309,7 @@ int main(int argc, char* argv[])
  // Add benchmarks
  std::vector<benchmark::internal::Benchmark*> benchmarks;
 #ifdef BENCHMARK_CONFIG_TUNING
+ (void)min_size;
  const int parallel_instance = parser.get<int>("parallel_instance");
  const int parallel_instances = parser.get<int>("parallel_instances");
  config_autotune_register::register_benchmark_subset(benchmarks,

diff --git a/benchmark/benchmark_device_segmented_radix_sort_pairs.parallel.hpp b/benchmark/benchmark_device_segmented_radix_sort_pairs.parallel.hpp
@@ -330,16 +330,15 @@ struct device_segmented_radix_sort_benchmark_generator
  rocprim::segmented_radix_sort_config<
  LongBits,
  ShortBits,
- BlockSize,
- ItemsPerThread,
- EnableUnpartitionedWarpSort,
+ rocprim::kernel_config<BlockSize, ItemsPerThread>,
  rocprim::WarpSortConfig<LogicalWarpSizeSmall / 2,
  ItemsPerThread / 2,
  BlockSize,
  PartitioningThreshold,
  LogicalWarpSizeSmall,
  ItemsPerThread,
- BlockSize>>>>());
+ BlockSize>,
+ EnableUnpartitionedWarpSort>>>());
  }
  };
 
@@ -360,16 +359,16 @@ struct device_segmented_radix_sort_benchmark_generator
  }
  else
  {
- storage.emplace_back(std::make_unique<device_segmented_radix_sort_benchmark<
-  Key,
-  Value,
-  rocprim::segmented_radix_sort_config<
-  LongBits,
-  ShortBits,
-  BlockSize,
-  ItemsPerThread,
-  EnableUnpartitionedWarpSort,
-  rocprim::DisabledWarpSortConfig>>>());
+ storage.emplace_back(
+ std::make_unique<device_segmented_radix_sort_benchmark<
+ Key,
+ Value,
+ rocprim::segmented_radix_sort_config<
+ LongBits,
+ ShortBits,
+ rocprim::kernel_config<BlockSize, ItemsPerThread>,
+ rocprim::DisabledWarpSortConfig,
+ EnableUnpartitionedWarpSort>>>());
  }
  }
  };