MLBEDSW-6496 mlperf_deeplabv3_mnv2_ade20k_int8 fails at verify_output for u65

Added check to see if additional stripe data is needed from producer op
when cascading to make sure the stripes are not overwriting data still
being used. Also changed scheduler to make sure ResizeBilinear always
runs with even stripe height.

Signed-off-by: Fredrik Svedberg <fredrik.svedberg@arm.com>
Change-Id: If7d723e6be29575c2b55c400eebbe8275a1aa328
diff --git a/ethosu/vela/scheduler.py b/ethosu/vela/scheduler.py
index 7e989a7..d01942b 100644
--- a/ethosu/vela/scheduler.py
+++ b/ethosu/vela/scheduler.py
@@ -44,6 +44,7 @@
 from .architecture_allocator import ArchitectureBlockConfig
 from .architecture_allocator import find_block_config
 from .architecture_allocator import get_ifm_area_required
+from .architecture_allocator import to_upscale
 from .architecture_features import ArchitectureFeatures
 from .architecture_features import Block
 from .cascade_builder import CascadeBuilder
@@ -906,7 +907,8 @@
             striped_schedule.cost_map[sched_op] = cost
 
             # Calculate the preceeding Op's stripe
-            stripe = sched_op.ifm.shape.with_height(stripe.height * sched_op.kernel.stride.y)
+            height = stripe.height + stripe.height % to_upscale(sched_op.resampling_mode)
+            stripe = sched_op.ifm.shape.with_height(height * sched_op.kernel.stride.y)
 
         return striped_schedule