Remove unnecessary dtype conversion & register int64 kernels

Author: Yibing Liu

paddle/operators/sequence_erase_op.cc

@@ -86,4 +86,5 @@ REGISTER_OP_WITHOUT_GRADIENT(sequence_erase, ops::SequenceEraseOp,
                              ops::SequenceEraseOpMaker);
 REGISTER_OP_CPU_KERNEL(
     sequence_erase,
-    ops::SequenceEraseKernel<paddle::platform::CPUDeviceContext, int32_t>);
+    ops::SequenceEraseKernel<paddle::platform::CPUDeviceContext, int32_t>,
+    ops::SequenceEraseKernel<paddle::platform::CPUDeviceContext, int64_t>);

paddle/operators/sequence_erase_op.cu

@@ -28,16 +28,12 @@ __global__ void LabelErasedIdx(const T* in_dat, const int64_t in_len,
                                size_t* num_erased) {
   int index = blockIdx.x * blockDim.x + threadIdx.x;
   if (index < in_len) {
-    int erased = 0;
     for (size_t i = 0; i < tokens_len; ++i) {
       if (in_dat[index] == tokens[i]) {
-        erased = 1;
+        num_erased[index + 1] = 1;
+        break;
       }
     }
-    num_erased[index + 1] = erased;
-    if (index == 0) {
-      num_erased[0] = 0;
-    }
   }
 }
 
@@ -60,26 +56,6 @@ __global__ void SetOutput(const T* in_dat, const int64_t in_len,
   }
 }
 
-template <typename T, typename Vector>
-thrust::device_vector<T> set_device_vector(Vector& vector) {
-  thrust::host_vector<T> host_vec(vector.size());
-  for (size_t i = 0; i < vector.size(); ++i) {
-    host_vec[i] = vector[i];
-  }
-  thrust::device_vector<T> dev_vec = host_vec;
-  return dev_vec;
-}
-
-template <typename T>
-std::vector<T> get_std_vector(thrust::device_vector<T>& dev_vec) {
-  thrust::host_vector<T> host_vec = dev_vec;
-  std::vector<T> std_vec(host_vec.size(), 0);
-  for (size_t i = 0; i < host_vec.size(); ++i) {
-    std_vec[i] = host_vec[i];
-  }
-  return std_vec;
-}
-
 template <typename T>
 class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
  public:
@@ -95,12 +71,11 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
     auto in_len = in->numel();
     auto in_dat = in->data<T>();
     // Copy tokens to GPU
-    thrust::device_vector<int> dev_tokens =
-        set_device_vector<int, std::vector<int>>(tokens);
+    thrust::device_vector<int> dev_tokens(tokens.begin(), tokens.end());
     int* dev_tokens_ptr = thrust::raw_pointer_cast(dev_tokens.data());
 
     // Count number of elements to be erased
-    thrust::device_vector<size_t> num_erased(in_len + 1);
+    thrust::device_vector<size_t> num_erased(in_len + 1, 0);
     size_t* num_erased_ptr = thrust::raw_pointer_cast(num_erased.data());
     auto stream = ctx.cuda_device_context().stream();
     LabelErasedIdx<<<(in_len - 1) / PADDLE_CUDA_NUM_THREADS + 1,
@@ -112,8 +87,7 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
     // Copy LoD to GPU
     auto lod0 = lod[0];
     auto lod_len = lod0.size();
-    thrust::device_vector<size_t> dev_in_lod =
-        set_device_vector<size_t, paddle::framework::Vector<size_t>>(lod0);
+    thrust::device_vector<size_t> dev_in_lod = lod0;
     size_t* dev_in_lod_ptr = thrust::raw_pointer_cast(dev_in_lod.data());
 
     // Calc output LoD
@@ -124,7 +98,7 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
         num_erased_ptr, dev_in_lod_ptr, lod_len, dev_out_lod_ptr);
 
     // Set LoD for output
-    std::vector<size_t> out_lod0 = get_std_vector<size_t>(dev_out_lod);
+    thrust::host_vector<size_t> out_lod0 = dev_out_lod;
     framework::LoD out_lod;
     out_lod.push_back(out_lod0);
     out->set_lod(out_lod);
@@ -142,4 +116,5 @@ class SequenceEraseOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 REGISTER_OP_CUDA_KERNEL(sequence_erase,
-                        paddle::operators::SequenceEraseOpCUDAKernel<int32_t>);
+                        paddle::operators::SequenceEraseOpCUDAKernel<int32_t>,
+                        paddle::operators::SequenceEraseOpCUDAKernel<int64_t>);

python/paddle/v2/fluid/tests/test_sequence_erase_op.py

@@ -29,7 +29,7 @@ def sequence_erase(in_seq, lod0, tokens):
     return np.array(out_seq).astype("int32"), new_lod0
 
 
-class TestSequenceEraseOp(OpTest):
+class TestSequenceEraseOpInt32(OpTest):
     def setUp(self):
         self.op_type = "sequence_erase"
         in_seq = np.random.randint(0, 10, (30, 1)).astype("int32")
@@ -44,6 +44,21 @@ def test_check_output(self):
         self.check_output()
 
 
+class TestSequenceEraseOpInt64(OpTest):
+    def setUp(self):
+        self.op_type = "sequence_erase"
+        in_seq = np.random.randint(0, 10, (30, 1)).astype("int64")
+        lod = [[0, 9, 13, 24, 30]]
+        tokens = [2, 3, 5]
+        out_seq, new_lod0 = sequence_erase(in_seq, lod[0], tokens)
+        self.attrs = {'tokens': tokens}
+        self.inputs = {'X': (in_seq, lod)}
+        self.outputs = {'Out': (out_seq, [new_lod0])}
+
+    def test_check_output(self):
+        self.check_output()
+
+
 class TestSequenceEraseOpEmpty(OpTest):
     def setUp(self):
         self.op_type = "sequence_erase"