Enable conservative memory allocations for RAFT IVF-Flat benchmarks. (#…

…1634)

In the benchmark we add the whole dataset at once to the index, there is no need to allocate extra space to amortize future extension of the dataset.  We can save memory by enabling the conservative allocation, and this enables running larger benchmarks.

Authors:
  - Tamas Bela Feher (https://github.com/tfeher)

Approvers:
  - Corey J. Nolet (https://github.com/cjnolet)

URL: #1634

cpp/bench/ann/src/raft/raft_ivf_flat_wrapper.h

@@ -96,7 +96,8 @@ RaftIvfFlatGpu<T, IdxT>::RaftIvfFlatGpu(Metric metric, int dim, const BuildParam
     dimension_(dim),
     mr_(rmm::mr::get_current_device_resource(), 1024 * 1024 * 1024ull)
 {
-  index_params_.metric = parse_metric_type(metric);
+  index_params_.metric                         = parse_metric_type(metric);
+  index_params_.conservative_memory_allocation = true;
   rmm::mr::set_current_device_resource(&mr_);
   RAFT_CUDA_TRY(cudaGetDevice(&device_));
 }