atpoints - drop old hack to avoid bug

Author: jeremylt

backends/cuda-gen/ceed-cuda-gen-operator.c

@@ -209,7 +209,7 @@ static int CeedOperatorApplyAddCore_Cuda_gen(CeedOperator op, CUstream stream, c
   int block[3] = {data->thread_1d, ((!is_tensor || data->dim == 1) ? 1 : data->thread_1d), -1};
 
   if (is_tensor) {
-    CeedCallBackend(BlockGridCalculate(num_elem, min_grid_size / cuda_data->device_prop.multiProcessorCount, is_at_points ? 1 : max_threads_per_block,
+    CeedCallBackend(BlockGridCalculate(num_elem, min_grid_size / cuda_data->device_prop.multiProcessorCount, max_threads_per_block,
                                        cuda_data->device_prop.maxThreadsDim[2], cuda_data->device_prop.warpSize, block, &grid));
   } else {
     CeedInt elems_per_block = CeedIntMin(cuda_data->device_prop.maxThreadsDim[2], CeedIntMax(512 / data->thread_1d, 1));

backends/hip-gen/ceed-hip-gen-operator-build.cpp

@@ -1589,7 +1589,6 @@ extern "C" int CeedOperatorBuildKernel_Hip_gen(CeedOperator op, bool *is_good_bu
   // Compile
   CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem));
   CeedCallBackend(BlockGridCalculate_Hip_gen(is_all_tensor ? max_dim : 1, num_elem, data->max_P_1d, is_all_tensor ? Q_1d : Q, block_sizes));
-  if (is_at_points) block_sizes[2] = 1;
   {
     bool is_compile_good = false;
 
@@ -2625,7 +2624,6 @@ extern "C" int CeedOperatorBuildKernelLinearAssembleQFunction_Hip_gen(CeedOperat
   // Compile
   CeedCallBackend(CeedOperatorGetNumElements(op, &num_elem));
   CeedCallBackend(BlockGridCalculate_Hip_gen(max_dim, num_elem, data->max_P_1d, Q_1d, block_sizes));
-  block_sizes[2] = 1;
   {
     bool is_compile_good = false;
 

backends/hip-gen/ceed-hip-gen-operator.c

@@ -153,7 +153,6 @@ static int CeedOperatorApplyAddCore_Hip_gen(CeedOperator op, hipStream_t stream,
 
   if (is_tensor) {
     CeedCallBackend(BlockGridCalculate_Hip_gen(data->dim, num_elem, data->max_P_1d, data->Q_1d, block_sizes));
-    if (is_at_points) block_sizes[2] = 1;
   } else {
     CeedInt elems_per_block = 64 * data->thread_1d > 256 ? 256 / data->thread_1d : 64;