Copy seq_lengths before creating descriptor

PiperOrigin-RevId: 519771897

jax/experimental/rnn.py

@@ -261,6 +261,8 @@ def lstm_ref(x: Array, h_0: Array, c_0: Array, W_ih: Dict[int, Array],
   See https://pytorch.org/docs/stable/generated/torch.nn.LSTM.html#lstm
   https://docs.nvidia.com/deeplearning/cudnn/api/index.html#cudnnRNNMode_t
   """
+  if seq_lengths.dtype != jnp.dtype("int32"):
+    raise NotImplementedError("`seq_lengths` can only be int32.")
   if dropout != 0.0:
     raise NotImplementedError(
         'Dropout not supported in LSTM reference because we cannot determine CUDNN dropout mask.'
@@ -326,6 +328,8 @@ def lstm_cell(carry, x, *, W_ih, W_hh, b_ih, b_hh):
 def lstm_fwd(x: Array, h_0: Array, c_0: Array, w: Array, seq_lengths: Array,
              input_size: int, hidden_size: int, num_layers: int, dropout: float,
              bidirectional: bool):
+  if seq_lengths.dtype != jnp.dtype("int32"):
+    raise NotImplementedError("`seq_lengths` can only be int32.")
   y, h_n, c_n, workspace, reserve_space = rnn_fwd_p.bind(
       x,
       h_0,

jaxlib/gpu/rnn_kernels.cc

@@ -316,8 +316,12 @@ static absl::Status DnnRNNBackward_(gpuStream_t stream, void** buffers,
   cudnnRNNDataLayout_t layout = CUDNN_RNN_DATA_LAYOUT_BATCH_MAJOR_UNPACKED;
   float padding = 0.0f;
 
+  auto seq_lengths_buf = buffers[11];
   std::vector<int32_t> seq_length_vector(d.batch_size, d.max_seq_length);
   int32_t* seq_length_array = &seq_length_vector[0];
+  JAX_RETURN_IF_ERROR(JAX_AS_STATUS(gpuMemcpy(
+      seq_length_array, seq_lengths_buf,
+      seq_length_vector.size() * sizeof(int32_t), gpuMemcpyDeviceToHost)));
 
   cudnnRNNDataDescriptor_t input_data_desc;
   JAX_RETURN_IF_ERROR(
@@ -367,7 +371,6 @@ static absl::Status DnnRNNBackward_(gpuStream_t stream, void** buffers,
   auto workspace_buf = buffers[8];
   auto reserve_space_buf = buffers[9];
   auto zeroed_dw_buf = buffers[10];
-  auto seq_lengths_buf = buffers[11];
   auto dx_buf = buffers[12];
   auto dh_0_buf = buffers[13];
   auto dc_0_buf = buffers[14];

tests/experimental_rnn_test.py

@@ -96,6 +96,70 @@ def f(x, h_0, c_0, weights):
     np.testing.assert_allclose(h_n_ref, h_n, rtol=1e-05, atol=1e-5)
     np.testing.assert_allclose(c_n_ref, c_n, rtol=1e-05, atol=1e-5)
 
+  @jtu.skip_on_devices("cpu", "tpu", "rocm")
+  def test_lstm_with_varying_seq_lens(self):
+    batch_size = 6
+    seq_len = 7
+    input_size = 8
+    hidden_size = 12
+    num_layers = 5
+    bidirectional = False
+    num_directions = 2 if bidirectional else 1
+
+    seq_lengths = jnp.array([4, 5, 1, 1, 1, 1], dtype=jnp.dtype("int32"))
+
+    root_key = jax.random.PRNGKey(1)
+    k1, k2, k3, k4 = jax.random.split(root_key, 4)
+    x = jax.random.normal(
+        k1, (batch_size, seq_len, input_size), dtype=jnp.float32)
+    h_0 = jax.random.normal(
+        k2, (num_directions * num_layers, batch_size, hidden_size),
+        dtype=jnp.float32)
+    c_0 = jax.random.normal(
+        k3, (num_directions * num_layers, batch_size, hidden_size),
+        dtype=jnp.float32)
+    weights = rnn.init_lstm_weight(k4, input_size, hidden_size, num_layers,
+                                   bidirectional)
+
+    @jax.jit
+    def f(x, h_0, c_0, weights):
+      return rnn.lstm(
+          x,
+          h_0,
+          c_0,
+          weights,
+          seq_lengths=seq_lengths,
+          input_size=input_size,
+          hidden_size=hidden_size,
+          num_layers=num_layers,
+          dropout=False,
+          bidirectional=bidirectional)
+
+    jtu.check_grads(f, (x, h_0, c_0, weights), modes=['rev'], order=1)
+
+    # TODO(sharadmv): enable when lstm_ref works with seq_lengths
+    # W_ih, W_hh, b_ih, b_hh = rnn.unpack_lstm_weights(weights, input_size,
+    #                                                  hidden_size, num_layers,
+    #                                                  bidirectional)
+    # y_ref, h_n_ref, c_n_ref = rnn.lstm_ref(
+    #     x,
+    #     h_0,
+    #     c_0,
+    #     W_ih,
+    #     W_hh,
+    #     b_ih,
+    #     b_hh,
+    #     seq_lengths=seq_lengths,
+    #     input_size=input_size,
+    #     hidden_size=hidden_size,
+    #     num_layers=num_layers,
+    #     dropout=False,
+    #     bidirectional=bidirectional)
+
+    # np.testing.assert_allclose(y_ref, y, rtol=1e-05, atol=1e-5)
+    # np.testing.assert_allclose(h_n_ref, h_n, rtol=1e-05, atol=1e-5)
+    # np.testing.assert_allclose(c_n_ref, c_n, rtol=1e-05, atol=1e-5)
+
 
 if __name__ == '__main__':
   absltest.main(testLoader=jtu.JaxTestLoader())