Fix flaky - test_operator_gpu.test_np_insert (#17620)

* fix flaky

* retrigger CI

src/operator/numpy/np_insert_op_slice-inl.h

@@ -171,7 +171,7 @@ void NumpyInsertSliceCompute(const nnvm::NodeAttrs& attrs,
     Tensor<xpu, 1, int64_t> indices(indices_ptr, Shape1(indices_len), s);
     Tensor<xpu, 1, int64_t> sorted_indices(sorted_indices_ptr, Shape1(indices_len), s);
     Tensor<xpu, 1, int> order(order_ptr, Shape1(indices_len), s);
-    int num_bits = common::ilog2ui(static_cast<unsigned int>(indices_len) - 1);
+    int num_bits = 8 * sizeof(int64_t);
     Kernel<SliceToIndices, xpu>::Launch(s, indices_len, indices_ptr, start, step);
     Kernel<range_fwd, xpu>::Launch(s, indices_len, 1, 0, 1, kWriteTo, order_ptr);
     mxnet::op::SortByKey(indices, order, true, &temp_storage, 0, num_bits, &sorted_indices);

src/operator/numpy/np_insert_op_tensor-inl.h

@@ -199,7 +199,7 @@ void NumpyInsertTensorCompute(const nnvm::NodeAttrs& attrs,
     Tensor<xpu, 1, int64_t> indices(indices_ptr, Shape1(indices_len), s);
     Tensor<xpu, 1, int64_t> sorted_indices(sorted_indices_ptr, Shape1(indices_len), s);
     Tensor<xpu, 1, int> order(order_ptr, Shape1(indices_len), s);
-    int num_bits = common::ilog2ui(static_cast<unsigned int>(indices_len) - 1);
+    int num_bits = 8 * sizeof(int64_t);
     Kernel<ObjToIndices, xpu>::Launch(s, indices_len, indices_ptr, N,
                                       inputs[obj_pos].dptr<int64_t>());
     Kernel<range_fwd, xpu>::Launch(s, indices_len, 1, 0, 1, kWriteTo, order_ptr);

tests/python/unittest/test_numpy_op.py

@@ -2661,7 +2661,7 @@ def GetNdim(tp):
     E = (1)
     F = (3, 1)
     G = (3, 2)
-    H = (5, 2, 3, 4)
+    H = (2, 2, 3, 8)
     config = []
     # test scale index
     for idx in range(-1 * GetSize(A), GetSize(A) + 1):
@@ -2705,7 +2705,7 @@ def GetNdim(tp):
     config.append(tuple([H, 0, 1, 3]))
     config.append(tuple([H, [1], E, 2]))
     config.append(tuple([H, [1], 1, 2]))
-    idx = _np.random.randint(-1 * H[3], H[3] + 1, size = (3)).tolist()
+    idx = _np.random.randint(-1 * H[3], H[3] + 1, size = (5)).tolist()
     config.append(tuple([H, idx, E, 3]))
     config.append(tuple([H, idx, 1, 3]))
     # test slice
@@ -2728,9 +2728,9 @@ def GetNdim(tp):
                 obj_onp = obj
             test_insert = TestInsert(obj=obj_mxnp, axis=axis)
 
-            a = mx.nd.random.uniform(-1.0, 1.0, shape=arr_shape).as_np_ndarray().astype(atype)
+            a = mx.nd.random.uniform(-10.0, 10.0, shape=arr_shape).as_np_ndarray().astype(atype)
             a.attach_grad()
-            b = mx.nd.random.uniform(-1.0, 1.0, shape=val_shape).as_np_ndarray().astype(btype)
+            b = mx.nd.random.uniform(-10.0, 10.0, shape=val_shape).as_np_ndarray().astype(btype)
             b.attach_grad()
             expected_ret = _np.insert(a.asnumpy(), obj_onp, b.asnumpy(), axis=axis)
             with mx.autograd.record():