[GitHub] [tvm] SebastianBoblestETAS commented on a diff in pull request #11341: [TE] Optimized version of concatenation layer

[GitBox <gi...@apache.org>]

SebastianBoblestETAS commented on code in PR #11341:
URL: https://github.com/apache/tvm/pull/11341#discussion_r902474819


##########
python/tvm/topi/x86/concat.py:
##########
@@ -0,0 +1,109 @@
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+"concatenate related operators"
+from typing import Optional
+import tvm
+from tvm import te
+import numpy as np
+from ..utils import get_const_int, const_vector
+
+
+def concatenate(data: tvm.te.Tensor, axis: Optional[int] = 0):
+    """Join a sequence of arrays along an existing axis. Optimized for CPU exeution.
+
+    Parameters
+    ----------
+    data : tuple of tvm.te.Tensor
+        The arrays to concatenate
+
+    axis : int, optional
+        The axis along which the arrays will be joined. Default is 0.
+
+    Returns
+    -------
+    ret : tvm.te.Tensor
+    """
+
+    def gen_ir_1d(data_bufs, in_outers_tensor, in_cumsum_tensor, out_buf):
+        """Custom conactenation execution."""
+        i_b = tvm.tir.ir_builder.create()
+        data_bufs1 = [i_b.buffer_ptr(data_buf) for data_buf in data_bufs]
+        out_buf = i_b.buffer_ptr(out_buf)
+        outers = i_b.buffer_ptr(in_outers_tensor)
+        cumsum = i_b.buffer_ptr(in_cumsum_tensor)
+        for i in range(len(data)):
+            with i_b.for_range(0, outers[i], name="j") as j:
+                out_buf[cumsum[i] + j] = data_bufs1[i][j]
+        return i_b.get()
+
+    def gen_ir(data_bufs, in_outers_tensor, in_cumsum_tensor, out_buf, inner, outer):
+        """Common case of conactenation execution."""
+        i_b = tvm.tir.ir_builder.create()
+        data_bufs1 = [i_b.buffer_ptr(data_buf) for data_buf in data_bufs]
+        out_buf = i_b.buffer_ptr(out_buf)
+        outers = i_b.buffer_ptr(in_outers_tensor)
+        cumsum = i_b.buffer_ptr(in_cumsum_tensor)
+        if inner > 1:
+            with i_b.for_range(0, inner, name="inn", kind="parallel") as inn:
+                pos = inn * outer
+                for i in range(len(data)):
+                    offset = inn * outers[i]
+                    with i_b.for_range(0, outers[i], name="j") as j:
+                        out_buf[pos + cumsum[i] + j] = data_bufs1[i][offset + j]
+        else:
+            for i in range(len(data)):
+                with i_b.for_range(0, outers[i], name="j", kind="parallel") as j:
+                    out_buf[cumsum[i] + j] = data_bufs1[i][j]
+        return i_b.get()
+
+    if axis < 0:
+        axis += len(data[0].shape)
+    concat_axis_sizes = [int(t.shape[axis]) for t in data]
+    join_size = int(np.sum(concat_axis_sizes))
+    in_outers = [int(np.prod(i.shape[axis:])) for i in data]
+    in_outers_cumsum = [0, *np.cumsum(in_outers, dtype="int64")[0:-1]]
+    dtype = data[0].dtype
+    out_shape = data[0].shape[:axis] + [join_size] + data[0].shape[axis + 1 :]
+    in_outers_tensor = const_vector(in_outers)
+    in_cumsum_tensor = const_vector(in_outers_cumsum, name="cumsum")

Review Comment:
   I can confirm this. We are currently working on a PR to change the behavior here.
   Just as a reference the comparison of the resulting C code with plain list of ints
   ```
    for (int32_t j = 0; j < 4; ++j) {
       concatenate_ext[j] = placeholder[j];
     }
     for (int32_t j1 = 0; j1 < 4; ++j1) {
       concatenate_ext[(j1 + 4)] = placeholder1[j1];
     }
     return 0;
   ```
   and with `te.tensor.Tensor`:
   
   ```
     void* const_vector_let = (&(global_workspace_6_var[64]));
     void* cumsum_let = (&(global_workspace_6_var[48]));
     for (int32_t i = 0; i < 2; ++i) {
       ((int64_t*)const_vector_let)[i] = ((i == 1) ? (int64_t)4 : ((i == 0) ? (int64_t)4 : (int64_t)0));
     }
     for (int32_t i1 = 0; i1 < 2; ++i1) {
       ((int64_t*)cumsum_let)[i1] = ((i1 == 1) ? (int64_t)4 : (int64_t)0);
     }
   
     int64_t cumsum_let[2] = {0, 4};
   
     for (int64_t j = 0; j < ((int64_t*)const_vector_let)[0]; ++j) {
       concatenate_ext[(((int64_t*)cumsum_let)[0] + j)] = placeholder[j];
     }
     for (int64_t j1 = 0; j1 < ((int64_t*)const_vector_let)[1]; ++j1) {
       concatenate_ext[(((int64_t*)cumsum_let)[1] + j1)] = placeholder1[j1];
     }
     return 0;
   ```
   
   
   @UlrikHjort-Bosch @vdkhoi @MichaelJKlaiber



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org