Fix workspace of BoxNMS (#20212)

src/operator/contrib/bounding_box.cu

@@ -40,23 +40,23 @@ using mshadow::Stream;
 
 template <typename DType>
 struct TempWorkspace {
-  index_t scores_temp_space;
+  size_t scores_temp_space;
   DType* scores;
-  index_t scratch_space;
+  size_t scratch_space;
   uint8_t* scratch;
-  index_t buffer_space;
+  size_t buffer_space;
   DType* buffer;
-  index_t nms_scratch_space;
+  size_t nms_scratch_space;
   uint32_t* nms_scratch;
-  index_t indices_temp_spaces;
+  size_t indices_temp_spaces;
   index_t* indices;
 };
 
-inline index_t ceil_div(index_t x, index_t y) {
+inline size_t ceil_div(size_t x, size_t y) {
   return (x + y - 1) / y;
 }
 
-inline index_t align(index_t x, index_t alignment) {
+inline size_t align(size_t x, size_t alignment) {
   return ceil_div(x, alignment)  * alignment;
 }
 
@@ -150,7 +150,7 @@ void CompactData(const Tensor<gpu, 1, index_t>& indices,
                  const int score_index,
                  Stream<gpu>* s) {
   const int n_threads = 512;
-  const index_t max_blocks = 320;
+  const size_t max_blocks = 320;
   index_t N = source.shape_.Size();
   const auto blocks = std::min(ceil_div(N, n_threads), max_blocks);
   if (topk > 0) {
@@ -175,9 +175,9 @@ void WorkspaceForSort(const index_t num_elem,
                       const index_t topk,
                       const int alignment,
                       TempWorkspace<DType>* workspace) {
-  const index_t sort_scores_temp_space =
+  const size_t sort_scores_temp_space =
     mxnet::op::SortByKeyWorkspaceSize<DType, index_t, gpu>(num_elem, false, false);
-  const index_t sort_topk_scores_temp_space =
+  const size_t sort_topk_scores_temp_space =
     mxnet::op::SortByKeyWorkspaceSize<DType, index_t, gpu>(topk, false, false);
   workspace->scratch_space = align(std::max(sort_scores_temp_space, sort_topk_scores_temp_space),
                                    alignment);
@@ -529,14 +529,15 @@ TempWorkspace<DType> GetWorkspace(const index_t num_batch,
   workspace.nms_scratch_space = align(NMS<DType>::THRESHOLD / (sizeof(uint32_t) * 8) *
                                       num_batch * topk * sizeof(uint32_t), alignment);
 
-  const index_t workspace_size = workspace.scores_temp_space +
-                                 workspace.scratch_space +
-                                 workspace.nms_scratch_space +
-                                 workspace.indices_temp_spaces;
+  const size_t workspace_size = workspace.scores_temp_space +
+                                workspace.scratch_space +
+                                workspace.buffer_space +
+                                workspace.nms_scratch_space +
+                                workspace.indices_temp_spaces;
 
   // Obtain the memory for workspace
-  Tensor<gpu, 1, uint8_t> scratch_memory = ctx.requested[box_nms_enum::kTempSpace]
-    .get_space_typed<gpu, 1, uint8_t>(mshadow::Shape1(workspace_size), s);
+  Tensor<gpu, 1, double> scratch_memory = ctx.requested[box_nms_enum::kTempSpace]
+    .get_space_typed<gpu, 1, double>(mshadow::Shape1(ceil_div(workspace_size, sizeof(double))), s);
 
   // Populate workspace pointers
   workspace.scores = reinterpret_cast<DType*>(scratch_memory.dptr_);