MLBEDSW-6260: Add support for using DMA to copy feature maps

- Reshape ops can be bypassed and there is no need to process them by the NPU.
There are use cases when the IFM must be preserved so a memcpy is needed.
This is implemented by an AvgPool.
- In order to reduce the cost of the AvgPool the IFM can be copied by DMA.
This is faster and also it can be turned into a real NOP in cases where
the IFM and the OFM can use the same memory space.
- Added new memcpy op. Only NHWC format supported since DMA can not change
the format on the fly.
- Allow ofm to reuse ifm for memcpy op
- Make sure the DMA copy size is 16 byte aligned

Change-Id: I3605a48d47646ff60d2bb3644dd3a23f872235a7
Signed-off-by: Johan Alfven <johan.alfven@arm.com>

1 files changed, 17 insertions, 1 deletions

diff --git a/ethosu/vela/high_level_command_stream.py b/ethosu/vela/high_level_command_stream.py
index 609f8556..09c1805d 100644
--- a/ethosu/vela/high_level_command_stream.py
+++ b/ethosu/vela/high_level_command_stream.py
@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright 2020-2022 Arm Limited and/or its affiliates <open-source-office@arm.com>
+# SPDX-FileCopyrightText: Copyright 2020-2023 Arm Limited and/or its affiliates <open-source-office@arm.com>
 #
 # SPDX-License-Identifier: Apache-2.0
 #
@@ -293,3 +293,19 @@ class DMA(Command):
     def get_operation_count(self):
         # returns numpy array of (DPU blocks, dma_ops)
         return np.array((0, 1))
+
+
+class NOP(Command):
+    def __init__(self, ps, in_tensor, out_tensor):
+        self.ps = ps
+        self.in_tensor = in_tensor
+        self.out_tensor = out_tensor
+
+    def __str__(self):
+        return f"<NOP: in={self.in_tensor.name}, out={self.out_tensor.name}>"
+
+    __repr__ = __str__
+
+    def get_operation_count(self):
+        # returns numpy array of (DPU blocks, dma_ops)
+        return np.array((0, 0))