commit 458a208c44f70a9848f1e8e2e91f28ce3641c48f
author Patrik Gustavsson <patrik.gustavsson@arm.com> Thu Aug 13 13:41:05 2020 +0200
committer Fredrik Knutsson <fredrik.knutsson.hunnebo@gmail.com> Fri Aug 14 10:49:15 2020 +0000
tree 37f23561f75d61746383dafc987b411646baaed8
parent be733cf04bb262d4eee791d76f01cecd64ff9255

MLBEDSW-2570 Avoid usage of NHCWB16 for some cases

Avoid usage of NHCWB16 when Stack/Pack/Concat is performed in axis 3,
and the "concat start" of each slice to be combined is not a multiple
of 16.

Signed-off-by: Patrik Gustavsson <patrik.gustavsson@arm.com>
Change-Id: If3f7b4a3424be3c86fc2dc48e8649ce4c4f49485

ethosu/vela/tensor.py

diff --git a/ethosu/vela/tensor.py b/ethosu/vela/tensor.py
index 3574970..ecca0e0 100644
--- a/ethosu/vela/tensor.py
+++ b/ethosu/vela/tensor.py
@@ -300,6 +300,7 @@
         "npu_tensor",
         "equivalence_id",
         "resampling_mode",
+        "avoid_NHCWB16",
     )
     AllocationQuantum = 16
 
@@ -346,6 +347,8 @@
         self.block_traversal = TensorBlockTraversal.Default
         self.resampling_mode = resampling_mode.NONE
 
+        self.avoid_NHCWB16 = False
+
     def element_size(self):
         if self.element_size_bytes == 0:
             return self.dtype.size_in_bits() / 8
@@ -380,6 +383,7 @@
         res.resampling_mode = self.resampling_mode
 
         res.copy_compressed_weight_info(self)
+        res.avoid_NHCWB16 = self.avoid_NHCWB16
         return res
 
     def clone_into_fast_storage(self, arch):