Revert "Tuning block sizes for aggregation setup routines."

This reverts commit 4968914.

core/src/aggregation/coarseAgenerators/low_deg_coarse_A_generator.cu

@@ -868,7 +868,7 @@ void fill_A_kernel_NxN_large( const int  R_num_rows, // same as num_aggregates.
 
 ///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 
-enum { WARP_SIZE = 32, SMEM_SIZE = 128 };
+enum { WARP_SIZE = 32, GRID_SIZE = 128, SMEM_SIZE = 128 };
 
 template< int CTA_SIZE, bool HAS_DIAG, bool COUNT_ONLY, typename Workspace >
 static
@@ -883,9 +883,6 @@ void compute_sparsity_dispatch( Workspace &hash_wk,
                                 int *Ac_cols,
                                 int *Ac_pos )
 {
-    cudaDeviceProp props = getDeviceProperties();
-    int GRID_SIZE = (props.major >= 7) ? 256 : 128;
-
     const int NUM_WARPS = CTA_SIZE / WARP_SIZE;
     int *h_status;
     thrust::global_thread_handle::cudaMallocHost((void **) &h_status, sizeof(int));
@@ -967,9 +964,6 @@ void fill_A_dispatch( Workspace &hash_wk,
                       Value_type *Ac_vals,
                       bool force_determinism )
 {
-    cudaDeviceProp props = getDeviceProperties();
-    int GRID_SIZE = (props.major >= 7) ? 256 : 128;
-
     const int NUM_WARPS = CTA_SIZE / WARP_SIZE;
     int work_offset = GRID_SIZE * NUM_WARPS;
     cudaMemcpyAsync( hash_wk.get_work_queue(), &work_offset, sizeof(int), cudaMemcpyHostToDevice, thrust::global_thread_handle::get_stream() );