Reduce the duplication of layout transfrom matrices

* Change the nhcwb16_to_nhwc matrix for binary and unary elementwise such that it matches the other NPU ops * Reduce the number of places where the same layout transform matrices are defined
apache · Mar 28, 2022 · b0d1ae6 · b0d1ae6
1 parent aab1d7c
commit b0d1ae6
Show file tree

Hide file tree

Showing 11 changed files with 95 additions and 174 deletions.
diff --git a/python/tvm/relay/backend/contrib/ethosu/te/binary_elementwise.py b/python/tvm/relay/backend/contrib/ethosu/te/binary_elementwise.py
@@ -22,6 +22,7 @@
 from tvm.contrib.ethosu.cascader import TESubgraph, EthosuPart, Propagator, register_matcher
 
 from .dma import dma_ofm_compute, dma_ifm_compute
+from .common import get_layout_transform_matrices
 
 
 def binary_elementwise_compute(
@@ -196,21 +197,8 @@ def binary_elementwise_compute(
             attrs=binary_elementwise_attrs,
         )
 
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 16, 0, 1, -16],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(int(ifm_channels))
+
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, 1, 0, 0, 0],

diff --git a/python/tvm/relay/backend/contrib/ethosu/te/common.py b/python/tvm/relay/backend/contrib/ethosu/te/common.py
@@ -0,0 +1,53 @@
+# 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.
+"""Common methods for the NPU tensor expressions"""
+
+from typing import Tuple, List
+
+
+def get_layout_transform_matrices(ofm_channels: int) -> Tuple[List[List[float]], List[List[float]]]:
+    """Get the NHWC->NHCWB16 and NHCWB16->NHWC layout transform matrices.
+
+    Parameters
+    ----------
+    ofm_channels : int
+        The number of output channels in a NHWC layout
+
+    Returns
+    -------
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc : Tuple[List[List[float]], List[List[float]]]
+        The layout transformation matrices
+    """
+
+    nhwc_to_nhcwb16 = [
+        [1, 0, 0, 0, 0],
+        [0, 1, 0, 0, 0],
+        [0, 0, 0, 1 / 16, 0],
+        [0, 0, 1, 0, 0],
+        [0, 0, 0, 0, 16],
+        [0, 0, 0, 0, 1],
+    ]
+
+    nhcwb16_to_nhwc = [
+        [1, 0, 0, 0, 0, 0],
+        [0, 1, 0, 0, 0, 0],
+        [0, 0, 0, 1, 0, 0],
+        [0, 0, 0, 0, 0, ofm_channels],
+        [0, 0, 0, 0, 0, 1],
+    ]
+
+    return nhwc_to_nhcwb16, nhcwb16_to_nhwc
diff --git a/python/tvm/relay/backend/contrib/ethosu/te/convolution.py b/python/tvm/relay/backend/contrib/ethosu/te/convolution.py
@@ -23,6 +23,7 @@
 from tvm.contrib.ethosu.cascader import TESubgraph, EthosuPart, Propagator, register_matcher
 
 from .dma import dma_ofm_compute, dma_ifm_compute
+from .common import get_layout_transform_matrices
 
 
 def conv2d_compute(
@@ -175,21 +176,8 @@ def conv2d_compute(
         attrs=conv2d_attrs,
     )
 
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 0, ofm_channels],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(ofm_channels)
+
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, stride_h, 0, 0, (dilated_kernel_h - stride_h)],

diff --git a/python/tvm/relay/backend/contrib/ethosu/te/depthwise.py b/python/tvm/relay/backend/contrib/ethosu/te/depthwise.py
@@ -23,6 +23,7 @@
 from tvm.contrib.ethosu.cascader import TESubgraph, EthosuPart, Propagator, register_matcher
 
 from .dma import dma_ofm_compute, dma_ifm_compute
+from .common import get_layout_transform_matrices
 
 
 def depthwise_conv2d_compute(
@@ -169,21 +170,8 @@ def depthwise_conv2d_compute(
         attrs=depthwise_conv2d_attrs,
     )
 
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 0, channels],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(channels)
+
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, stride_h, 0, 0, (dilated_kernel_h - stride_h)],

diff --git a/python/tvm/relay/backend/contrib/ethosu/te/pooling.py b/python/tvm/relay/backend/contrib/ethosu/te/pooling.py
@@ -23,6 +23,7 @@
 from tvm.contrib.ethosu.cascader import TESubgraph, EthosuPart, Propagator, register_matcher
 
 from .dma import dma_ofm_compute, dma_ifm_compute
+from .common import get_layout_transform_matrices
 
 
 def pooling_compute(
@@ -157,21 +158,8 @@ def pooling_compute(
         attrs=pooling_attrs,
     )
 
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 0, int(ofm_channels)],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(int(ofm_channels))
+
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, stride_h, 0, 0, (pool_shape_h - stride_h)],

diff --git a/python/tvm/relay/backend/contrib/ethosu/te/unary_elementwise.py b/python/tvm/relay/backend/contrib/ethosu/te/unary_elementwise.py
@@ -21,6 +21,7 @@
 from tvm import te
 from tvm.contrib.ethosu.cascader import TESubgraph, EthosuPart, Propagator, register_matcher
 from .dma import dma_ofm_compute, dma_ifm_compute
+from .common import get_layout_transform_matrices
 
 
 def unary_elementwise_compute(
@@ -129,21 +130,8 @@ def clz_imp(inp):
         attrs=unary_elementwise_attrs,
     )
 
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 16, 0, 1, -16],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(int(ofm_channels))
+
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, 1, 0, 0, 0],

diff --git a/src/contrib/ethosu/cascader/block_config.cc b/src/contrib/ethosu/cascader/block_config.cc
@@ -46,8 +46,8 @@ BlockConfig::BlockConfig(const std::vector<int>& input_shape, const std::vector<
   auto n = make_object<BlockConfigNode>();
   n->input_shape_ = std::move(input_shape);
   n->output_shape_ = std::move(output_shape);
-  n->compute_cycles_ = std::move(compute_cycles);
-  n->output_cycles_ = std::move(output_cycles);
+  n->compute_cycles_ = compute_cycles;
+  n->output_cycles_ = output_cycles;
   data_ = std::move(n);
 }
 

diff --git a/tests/python/contrib/test_ethosu/cascader/infra.py b/tests/python/contrib/test_ethosu/cascader/infra.py
@@ -55,6 +55,7 @@ def make_simple_home_map(graph, var_region, const_region):
 
 if ethosu_enabled:
     from tvm.relay.backend.contrib.ethosu.tir.compiler import extract_constants, lower_to_te
+    from tvm.relay.backend.contrib.ethosu.te.common import get_layout_transform_matrices
 
     def create_te_graph(func):
         func, consts = extract_constants(func)
@@ -79,21 +80,9 @@ def make_matrices(
         dilation_h, dilation_w = dilation
         dilated_kernel_h = (kernel_h - 1) * dilation_h + 1
         dilated_kernel_w = (kernel_w - 1) * dilation_w + 1
-        nhwc_to_nhcwb16 = [
-            [1, 0, 0, 0, 0],
-            [0, 1, 0, 0, 0],
-            [0, 0, 0, 1 / 16, 0],
-            [0, 0, 1, 0, 0],
-            [0, 0, 0, 0, 16],
-            [0, 0, 0, 0, 1],
-        ]
-        nhcwb16_to_nhwc = [
-            [1, 0, 0, 0, 0, 0],
-            [0, 1, 0, 0, 0, 0],
-            [0, 0, 0, 1, 0, 0],
-            [0, 0, 0, 0, 0, ofm_channels],
-            [0, 0, 0, 0, 0, 1],
-        ]
+
+        nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(ofm_channels)
+
         if op_type == "ethosu_conv2d":
             ifm_matrix = [
                 [1, 0, 0, 0, 0],

diff --git a/tests/python/contrib/test_ethosu/cascader/test_ethosu_binary_elementwise_matcher.py b/tests/python/contrib/test_ethosu/cascader/test_ethosu_binary_elementwise_matcher.py
@@ -27,25 +27,12 @@
     match_ethosu_binary_elementwise,
     binary_elementwise_compute,
 )
+from tvm.relay.backend.contrib.ethosu.te.common import get_layout_transform_matrices
 
 
-def _make_matrices(broadcast, ifm_layout, ifm2_layout, ofm_layout):
+def _make_matrices(broadcast, ifm_layout, ifm2_layout, ofm_layout, ofm_channels):
     broadcast_h, broadcast_w, broadcast_c = broadcast
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-    nhcwb16_to_nhwc = [
-        [1, 0, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0, 0],
-        [0, 0, 0, 1, 0, 0],
-        [0, 0, 16, 0, 1, -16],
-        [0, 0, 0, 0, 0, 1],
-    ]
+    nhwc_to_nhcwb16, nhcwb16_to_nhwc = get_layout_transform_matrices(ofm_channels)
     ifm_matrix = [
         [1, 0, 0, 0, 0],
         [0, 1, 0, 0, 0],
@@ -93,14 +80,8 @@ def test_ethosu_binary_elementwise_matcher(
     ifm2_shape = [1] + [1 if (b == 1) else a for a, b in zip(ofm_shape[1:], ifm2_broadcast)]
     ifm_channels = ifm_shape[3]
     ifm2_channels = ifm2_shape[3]
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
+    ofm_channels = ofm_shape[3]
+    nhwc_to_nhcwb16, _ = get_layout_transform_matrices(ofm_channels)
     broadcast = [1 if a == 1 else 0 for a in ifm2_shape[1:]]
     if ifm_layout == "NHCWB16":
         ifm_shape = [
@@ -173,10 +154,7 @@ def test_ethosu_binary_elementwise_matcher(
     output_stripe_config = cs.StripeConfig(ofm_shape, ofm_shape, ofm_shape, order, stripes, offset)
 
     (ifm_transform, ifm2_transform) = _make_matrices(
-        broadcast,
-        ifm_layout,
-        ifm2_layout,
-        ofm_layout,
+        broadcast, ifm_layout, ifm2_layout, ofm_layout, ofm_channels
     )
 
     device_config = cs.EthosuDeviceConfig("ethos-u55-256")
@@ -190,19 +168,10 @@ def test_ethosu_binary_elementwise_matcher(
     propagated_ifm = ifm_propagator.propagate(output_stripe_config).shape
     propagated_ifm2 = ifm2_propagator.propagate(output_stripe_config).shape
 
-    # Layout conversions will align the propagated IFMs to the brick, i.e. 16
-    # so the expected ifm(2)_shape needs to be rounded up to 16
-    if ifm_layout != ofm_layout:
-        assert ifm_shape[:-1] == propagated_ifm[:-1]
-        assert ((ifm_shape[-1] + 16 - 1) // 16) * 16 == propagated_ifm[-1]
-    else:
-        assert ifm_shape == propagated_ifm
-
-    if ifm2_layout != ofm_layout:
-        assert ifm2_shape[:-1] == propagated_ifm2[:-1]
-        assert ((ifm2_shape[-1] + 16 - 1) // 16) * 16 == propagated_ifm2[-1]
-    else:
-        assert ifm2_shape == propagated_ifm2
+    # The layout transforms that have the exact number of output channels in them
+    # will lose no information about the number of channels
+    assert ifm_shape == propagated_ifm
+    assert ifm2_shape == propagated_ifm2
 
 
 if __name__ == "__main__":

diff --git a/tests/python/contrib/test_ethosu/cascader/test_ethosu_block_config.py b/tests/python/contrib/test_ethosu/cascader/test_ethosu_block_config.py
@@ -22,6 +22,7 @@
 import math
 
 import tvm.contrib.ethosu.cascader as cs
+from tvm.relay.backend.contrib.ethosu.te.common import get_layout_transform_matrices
 
 from .infra import make_matrices
 
@@ -244,18 +245,11 @@ def test_best_block_config(
     acc_config,
     expected_block_configs,
 ):
-    nhwc_to_nhcwb16 = [
-        [1, 0, 0, 0, 0],
-        [0, 1, 0, 0, 0],
-        [0, 0, 0, 1 / 16, 0],
-        [0, 0, 1, 0, 0],
-        [0, 0, 0, 0, 16],
-        [0, 0, 0, 0, 1],
-    ]
-
     ofm_channels = out_shape[3]
     ifm_channels = in_shape[3]
 
+    nhwc_to_nhcwb16, _ = get_layout_transform_matrices(ofm_channels)
+
     ifm_matrix, ifm_offset, weight_matrix, weight_offset, _, _ = make_matrices(
         op_type,
         kernel,