MLBEDSW-4688: Fix performance estimates

Putting back the estimates related to unbuffered
weight transfer.

Signed-off-by: Patrik Gustavsson <patrik.gustavsson@arm.com>
Change-Id: I2072066bc1e01814fe3b0b87a912f69646da861c
diff --git a/ethosu/vela/npu_performance.py b/ethosu/vela/npu_performance.py
index b1dae4e..5c61c7d 100644
--- a/ethosu/vela/npu_performance.py
+++ b/ethosu/vela/npu_performance.py
@@ -708,11 +708,17 @@
         bw = access.const_read[0] * bandwidth_compression_scale_approx
         bws[query.const_memory_area][TensorPurpose.Weights][BandwidthDirection.Read] += bw
 
+        if not cost.buffered_weight_tensor:
+            scaled_bws[query.const_memory_area][TensorPurpose.Weights][BandwidthDirection.Read] += bw
+
     if access.const_read[1] > 0:
         # Scales & biases
         bw = access.const_read[1] * op.parent_op.bias.element_size()
         bws[query.const_memory_area][TensorPurpose.FSBias][BandwidthDirection.Read] += bw
 
+        if not cost.buffered_weight_tensor:
+            scaled_bws[query.const_memory_area][TensorPurpose.FSBias][BandwidthDirection.Read] += bw
+
     update_summary_cycles(arch, scaled_bws, cycles_a)
 
     return bws, macs, cycles_a