[NPU] 更新3个 box_iou 相关 NPU 算子适配层 (#3276)

* Update boxes_overlap_bev_npu.cpp

更换 boxes_overlap_bev_npu 的底层 aclnn 调用算子

* Update box_iou_rotated.py

更新 box_iou_rotated python 的 npu 相关适配层

* Update box_iou_rotated_npu.cpp

Update box_iou_rotated_npu.cpp

* Create diff_iou_rotated_npu.cpp

diff_iou_rotated 算子添加 npu 适配

* Update box_iou_rotated_npu.cpp

* Update diff_iou_rotated_npu.cpp

* Update boxes_overlap_bev_npu.cpp

* Update boxes_overlap_bev_npu.cpp

* Update diff_iou_rotated_npu.cpp

* Update boxes_overlap_bev_npu.cpp

Author: JYYCaN

mmcv/ops/box_iou_rotated.py

@@ -142,11 +142,6 @@ def box_iou_rotated(bboxes1: torch.Tensor,
         flip_mat[-1] = -1
         bboxes1 = bboxes1 * flip_mat
         bboxes2 = bboxes2 * flip_mat
-    if bboxes1.device.type == 'npu':
-        scale_mat = bboxes1.new_ones(bboxes1.shape[-1])
-        scale_mat[-1] = 1.0 / 0.01745329252
-        bboxes1 = bboxes1 * scale_mat
-        bboxes2 = bboxes2 * scale_mat
     bboxes1 = bboxes1.contiguous()
     bboxes2 = bboxes2.contiguous()
     ext_module.box_iou_rotated(

mmcv/ops/csrc/pytorch/npu/box_iou_rotated_npu.cpp

@@ -8,40 +8,10 @@ void box_iou_rotated_impl(const Tensor boxes1, const Tensor boxes2, Tensor ious,
 
 void box_iou_rotated_npu(const Tensor boxes1, const Tensor boxes2, Tensor ious,
                          const int mode_flag, const bool aligned) {
-  at::Tensor boxes = at::ones_like(boxes1);
-  at::Tensor query_boxes = at::ones_like(boxes2);
-  boxes = boxes1.transpose(0, 1).unsqueeze(0);
-  query_boxes = boxes2.transpose(0, 1).unsqueeze(0);
-
-  bool is_trans = false;
-  string modeStr = "iou";
-  if (mode_flag == 1) {
-    modeStr = "iof";
-  }
-  bool is_cross = true;
-  if (aligned) {
-    is_cross = false;
-  }
-  float v_threshold = 0;
-  float e_threshold = 0;
-
-  OpCommand cmd;
-  cmd.Name("RotatedIou")
-      .Input(boxes)
-      .Input(query_boxes)
-      .Output(ious)
-      .Attr("trans", is_trans)
-      .Attr("mode", modeStr)
-      .Attr("is_cross", is_cross)
-      .Attr("v_threshold", v_threshold)
-      .Attr("e_threshold", e_threshold)
-      .Run();
-
-  if (is_cross) {
-    ious = ious.view({boxes1.size(0), boxes2.size(0)});
-  } else {
-    ious = ious.view({boxes1.size(0), 1});
-  }
+  TORCH_CHECK(boxes1.size(1) == 5, "boxes1 must be 2D tensor (N, 5)");
+  TORCH_CHECK(boxes1.size(1) == 5, "boxes1 must be 2D tensor (N, 5)");
+  EXEC_NPU_CMD(aclnnBoxIou, boxes1, boxes2, mode_flag, aligned, ious);
+  return;
 }
 
 REGISTER_NPU_IMPL(box_iou_rotated_impl, box_iou_rotated_npu);

mmcv/ops/csrc/pytorch/npu/boxes_overlap_bev_npu.cpp

@@ -3,6 +3,11 @@
 using namespace NPU_NAME_SPACE;
 using namespace std;
 
+namespace {
+constexpr int32_t MODE_FLAG_OVERLAP = 0;
+constexpr int32_t FORMAT_FLAG_XYZWHDR = 3;
+};  // namespace
+
 void iou3d_boxes_overlap_bev_forward_impl(const int num_a, const Tensor boxes_a,
                                           const int num_b, const Tensor boxes_b,
                                           Tensor ans_overlap);
@@ -13,13 +18,13 @@ void iou3d_boxes_overlap_bev_forward_npu(const int num_a, const Tensor boxes_a,
   TORCH_CHECK(boxes_a.size(1) == 7, "boxes_a must be 2D tensor (N, 7)");
   TORCH_CHECK(boxes_b.size(1) == 7, "boxes_b must be 2D tensor (N, 7)");
 
-  auto format_flag = 3;
   auto clockwise = true;
-  auto mode_flag = 0;
-  auto aligned = false;
-  auto margin = 1e-2;
+  bool aligned = false;
+  double margin = 1e-5;
+  int32_t mode_flag = MODE_FLAG_OVERLAP;
+  int32_t format_flag = FORMAT_FLAG_XYZWHDR;
 
-  EXEC_NPU_CMD(aclnnBoxesOverlapBev, boxes_a, boxes_b, format_flag, clockwise,
+  EXEC_NPU_CMD(aclnnBoxesOverlapBevV1, boxes_a, boxes_b, format_flag, clockwise,
                mode_flag, aligned, margin, ans_overlap);
   return;
 }

mmcv/ops/csrc/pytorch/npu/diff_iou_rotated_npu.cpp

@@ -0,0 +1,30 @@
+#include "pytorch_npu_helper.hpp"
+using namespace NPU_NAME_SPACE;
+using namespace std;
+
+Tensor diff_iou_rotated_sort_vertices_npu(Tensor vertices, Tensor mask,
+                                          Tensor num_valid) {
+  TORCH_CHECK(vertices.dim() == 4,
+              "vertices must be a 4D Tensor, but got: ", vertices.dim());
+  TORCH_CHECK(mask.dim() == 3,
+              "mask must be a 3D Tensor, but got: ", mask.dim());
+  TORCH_CHECK(num_valid.dim() == 2,
+              "num_valid must be a 2D Tensor, but got: ", num_valid.dim());
+
+  uint32_t B = vertices.size(0);
+  uint32_t N = vertices.size(1);
+
+  at::Tensor sortedIdx = at::empty({B, N, 9}, num_valid.options());
+  at::Tensor mask_fp = mask.to(at::kFloat);
+
+  EXEC_NPU_CMD(aclnnDiffIouRotatedSortVertices, vertices, mask_fp, num_valid,
+               sortedIdx);
+
+  return sortedIdx;
+}
+
+Tensor diff_iou_rotated_sort_vertices_forward_impl(Tensor vertices, Tensor mask,
+                                                   Tensor num_valid);
+
+REGISTER_NPU_IMPL(diff_iou_rotated_sort_vertices_forward_impl,
+                  diff_iou_rotated_sort_vertices_npu);