MLBEDSW-5437: MLCE: Vela ignores back2back Activations in model
- Back-to-back 16-bit activation ops were packed into the same pass
because there was no check to disallow it
- The solution is to set the appropriate incompatible-flags
Signed-off-by: Tim Hall <tim.hall@arm.com>
Change-Id: Idb3c741a7b52e0d81c1f687f6ecf78352b7872dd
diff --git a/ethosu/vela/pass_packing.py b/ethosu/vela/pass_packing.py
index b28f4eb..530ad78 100644
--- a/ethosu/vela/pass_packing.py
+++ b/ethosu/vela/pass_packing.py
@@ -105,7 +105,7 @@
# ops_set
npu_post_fuse_limited_ops,
# incompatible_pack_flags
- PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Main,
+ PassFlags.Cpu | PassFlags.MemoryOnly | PassFlags.Main | PassFlags.PostFusingLimited,
# flags_to_set
PassFlags.Npu | PassFlags.PostFusingLimited,
# flags_to_clear