Wave Metrics

- Make multithreaded memcpy for staging transfers for GPU table scan.
- Make variants of bit unpacking in GpuDecoder-inl.cuh. Make selective decoding templatized as opposed to runtime switching.
- Add pieces to GpuDecoderTest, like comparing calling via launchDecode or (multi-function blocks) or decodeGlobal (single function thread blocks).
- Add a metric for driver thread waiting for first continuable stream.
- Check approx correctness of Wave runtimeStats.
- Refactor QueryBenchmarkBase.* from TpchBenchmark. Logic to do sweeps across parameter combinations.
- Add persistent file format to Wave mock format.
- Add benchmark for scan, filter, filter expr, projection. aggregation combinations with Wave and Dwrf.

velox/benchmarks/CMakeLists.txt

@@ -39,3 +39,27 @@ if(${VELOX_ENABLE_BENCHMARKS})
   add_subdirectory(tpch)
   add_subdirectory(filesystem)
 endif()
+
+add_library(velox_query_benchmark QueryBenchmarkBase.cpp)
+target_link_libraries(
+  velox_query_benchmark
+  velox_aggregates
+  velox_exec
+  velox_exec_test_lib
+  velox_dwio_common
+  velox_dwio_common_exception
+  velox_dwio_parquet_reader
+  velox_dwio_common_test_utils
+  velox_hive_connector
+  velox_exception
+  velox_memory
+  velox_process
+  velox_serialization
+  velox_encode
+  velox_type
+  velox_type_fbhive
+  velox_caching
+  velox_vector_test_lib
+  ${FOLLY_BENCHMARK}
+  Folly::folly
+  fmt::fmt)