aboutsummaryrefslogtreecommitdiff
path: root/ethosu/vela/high_level_command_stream_generator.py
diff options
context:
space:
mode:
authorAndreas Nevalainen <andreas.nevalainen@arm.com>2020-10-28 15:42:08 +0100
committerpatrik.gustavsson <patrik.gustavsson@arm.com>2020-11-11 08:34:16 +0000
commit897cc14968e017b1f48f376f7f7cefc515c5fe88 (patch)
tree27d17a59f1529c1ed0b1cc59e60438ee7f5d29d2 /ethosu/vela/high_level_command_stream_generator.py
parent73320a48dfa711f5938b0e3d8e03b9858558b899 (diff)
downloadethos-u-vela-897cc14968e017b1f48f376f7f7cefc515c5fe88.tar.gz
MLBEDSW-3222: Bias tensors in fast storage
For IFM streamed cascades bias tensors are read several times. Moves these tensors to fast storage and add DMA commands. Change-Id: I630f6275986c1b5e3f126c925b11e22500fb1128 Signed-off-by: Andreas Nevalainen <andreas.nevalainen@arm.com>
Diffstat (limited to 'ethosu/vela/high_level_command_stream_generator.py')
-rw-r--r--ethosu/vela/high_level_command_stream_generator.py5
1 files changed, 5 insertions, 0 deletions
diff --git a/ethosu/vela/high_level_command_stream_generator.py b/ethosu/vela/high_level_command_stream_generator.py
index 01fab0e8..871a0486 100644
--- a/ethosu/vela/high_level_command_stream_generator.py
+++ b/ethosu/vela/high_level_command_stream_generator.py
@@ -238,6 +238,7 @@ def generate_high_level_command_stream_for_pass(strat, passes, block_configs, id
y_step = y_dim
weight_box = None
+ scale_box = None
for start in range(y_start, y_dim, y_step):
end = min(start + y_step, y_dim)
@@ -299,6 +300,10 @@ def generate_high_level_command_stream_for_pass(strat, passes, block_configs, id
if ifm_y_present >= ifm_y_needed:
break
+ if scale_tensor is not None and scale_tensor.purpose == TensorPurpose.FSBias and scale_box is None:
+ scale_box = Box([0] * len(scale_tensor.shape), list(scale_tensor.shape))
+ yield from dma_if_necessary(ps, scale_box, scale_tensor)
+
if weight_tensor is not None and weight_box is None:
weight_box = Box.make_weight_box(
weight_tensor.shape, npu_block_type, weights_transposed=weight_tensor.weight_transpose_depthwise