!7281 use faster sort algorithm in detection_post_process op

Merge pull request !7281 from wangzhe/tmp

mindspore/lite/nnacl/fp32/detection_post_process.c

@@ -19,18 +19,84 @@
 #include "nnacl/errorcode.h"
 #include "nnacl/op_base.h"
 
-int ScoreWithIndexCmp(const void *a, const void *b) {
-  ScoreWithIndex *pa = (ScoreWithIndex *)a;
-  ScoreWithIndex *pb = (ScoreWithIndex *)b;
+bool ScoreWithIndexCmp(ScoreWithIndex *pa, ScoreWithIndex *pb) {
   if (pa->score > pb->score) {
-    return -1;
+    return true;
   } else if (pa->score < pb->score) {
-    return 1;
+    return false;
   } else {
-    return pa->index - pb->index;
+    return pa->index < pb->index;
   }
 }
 
+void PushHeap(ScoreWithIndex *root, int cur, int top_index, ScoreWithIndex value) {
+  int parent = (cur - 1) / 2;
+  while (cur > top_index && ScoreWithIndexCmp(root + parent, &value)) {
+    *(root + cur) = root[parent];
+    cur = parent;
+    parent = (cur - 1) / 2;
+  }
+  *(root + cur) = value;
+}
+
+void AdjustHeap(ScoreWithIndex *root, int cur, int limit, ScoreWithIndex value) {
+  int top_index = cur;
+  int second_child = cur;
+  while (second_child < (limit - 1) / 2) {
+    second_child = 2 * (second_child + 1);
+    if (ScoreWithIndexCmp(root + second_child, root + second_child - 1)) {
+      second_child--;
+    }
+    *(root + cur) = *(root + second_child);
+    cur = second_child;
+  }
+  if ((limit & 1) == 0 && second_child == (limit - 2) / 2) {
+    second_child = 2 * (second_child + 1);
+    *(root + cur) = *(root + second_child - 1);
+    cur = second_child - 1;
+  }
+  PushHeap(root, cur, top_index, value);
+}
+
+void PopHeap(ScoreWithIndex *root, int limit, ScoreWithIndex *result) {
+  ScoreWithIndex value = *result;
+  *result = *root;
+  AdjustHeap(root, 0, limit, value);
+}
+
+void MakeHeap(ScoreWithIndex *values, int limit) {
+  if (limit < 2) return;
+  int parent = (limit - 2) / 2;
+  while (true) {
+    AdjustHeap(values, parent, limit, values[parent]);
+    if (parent == 0) {
+      return;
+    }
+    parent--;
+  }
+}
+
+void SortHeap(ScoreWithIndex *root, int limit) {
+  while (limit > 1) {
+    --limit;
+    PopHeap(root, limit, root + limit);
+  }
+}
+
+void HeapSelect(ScoreWithIndex *root, int cur, int limit) {
+  MakeHeap(root, cur);
+  for (int i = cur; i < limit; ++i) {
+    if (ScoreWithIndexCmp(root + i, root)) {
+      PopHeap(root, cur, root + i);
+    }
+  }
+}
+
+void PartialSort(ScoreWithIndex *values, int num_to_sort, int num_values) {
+  HeapSelect(values, num_to_sort, num_values);
+  SortHeap(values, num_to_sort);
+}
+
 float IntersectionOverUnion(const BboxCorner *a, const BboxCorner *b) {
   const float area_a = (a->ymax - a->ymin) * (a->xmax - a->xmin);
   const float area_b = (b->ymax - b->ymin) * (b->xmax - b->xmin);
@@ -70,7 +136,7 @@ int NmsSingleClass(const int candidate_num, const float *decoded_boxes, const in
   const int output_num = candidate_num < max_detections ? candidate_num : max_detections;
   int possible_candidate_num = candidate_num;
   int selected_num = 0;
-  qsort(score_with_index, candidate_num, sizeof(ScoreWithIndex), ScoreWithIndexCmp);
+  PartialSort(score_with_index, candidate_num, candidate_num);
   for (int i = 0; i < candidate_num; ++i) {
     nms_candidate[i] = 1;
   }
@@ -134,7 +200,7 @@ int NmsMultiClassesRegular(const int num_boxes, const int num_classes_with_bg, c
     }
     all_classes_output_num =
       all_classes_sorted_num < param->max_detections_ ? all_classes_sorted_num : param->max_detections_;
-    qsort(score_with_index_all, all_classes_sorted_num, sizeof(ScoreWithIndex), ScoreWithIndexCmp);
+    PartialSort(score_with_index_all, all_classes_output_num, all_classes_sorted_num);
     for (int i = 0; i < all_classes_output_num; ++i) {
       score_with_index_all[i].index = indexes[score_with_index_all[i].index];
     }
@@ -178,8 +244,7 @@ int NmsMultiClassesFast(const int num_boxes, const int num_classes_with_bg, cons
       // save box and class info to index
       score_with_class_all[i * param->num_classes_ + j - first_class_index].index = i * num_classes_with_bg + j;
     }
-    qsort(score_with_class_all + i * param->num_classes_, param->num_classes_, sizeof(ScoreWithIndex),
-          ScoreWithIndexCmp);
+    PartialSort(score_with_class_all + i * param->num_classes_, max_classes_per_anchor, param->num_classes_);
     const float score_max = (score_with_class_all + i * param->num_classes_)->score;
     if (score_max >= param->nms_score_threshold_) {
       score_with_class[candidate_num].index = i;

mindspore/lite/nnacl/fp32/detection_post_process.h

@@ -43,10 +43,6 @@ typedef struct {
 extern "C" {
 #endif
 
-void nms_multi_classes_regular();
-
-void nms_multi_classes_fase();
-
 int DetectionPostProcess(const int num_boxes, const int num_classes_with_bg, float *input_boxes, float *input_scores,
                          float *input_anchors, float *output_boxes, float *output_classes, float *output_scores,
                          float *output_num, DetectionPostProcessParameter *param);

mindspore/lite/src/runtime/kernel/arm/fp32/detection_post_process.h

@@ -30,17 +30,12 @@ class DetectionPostProcessCPUKernel : public LiteKernel {
   DetectionPostProcessCPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs,
                                 const std::vector<lite::Tensor *> &outputs, const lite::InnerContext *ctx,
                                 const mindspore::lite::PrimitiveC *primitive)
-      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {
-    param_ = reinterpret_cast<DetectionPostProcessCPUKernel *>(parameter);
-  }
+      : LiteKernel(parameter, inputs, outputs, ctx, primitive) {}
   ~DetectionPostProcessCPUKernel() override;
 
   int Init() override;
   int ReSize() override;
   int Run() override;
-
- private:
-  DetectionPostProcessCPUKernel *param_;
 };
 }  // namespace mindspore::kernel
 #endif  // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP32_DETECTION_POST_PROCESS_H_