diff options
-rw-r--r-- | ethosu/vela/graph_optimiser.py | 4 | ||||
-rw-r--r-- | ethosu/vela/high_level_command_stream.py | 2 | ||||
-rw-r--r-- | ethosu/vela/high_level_command_to_npu_op.py | 9 |
3 files changed, 10 insertions, 5 deletions
diff --git a/ethosu/vela/graph_optimiser.py b/ethosu/vela/graph_optimiser.py index 7c60368d..573b7ebb 100644 --- a/ethosu/vela/graph_optimiser.py +++ b/ethosu/vela/graph_optimiser.py @@ -196,6 +196,10 @@ def remove_SplitSliceRead(op, arch): cons_op.set_input_tensor(op.ifm, cons_op.type.info.indices.ifms[1]) cons_op.ifm_shapes[1] = op.ifm_shapes[0] + if "skirt" in cons_op.attrs: + assert cons_op.attrs["explicit_padding"] == cons_op.attrs["skirt"] + cons_op.attrs["skirt"] = None + cons_op.attrs["force_padding"] = True op.ofm.consumer_list.remove(cons_op) op.ofm.ops = [] op.ifm.consumer_list.remove(op) diff --git a/ethosu/vela/high_level_command_stream.py b/ethosu/vela/high_level_command_stream.py index 075574e4..53c5e01f 100644 --- a/ethosu/vela/high_level_command_stream.py +++ b/ethosu/vela/high_level_command_stream.py @@ -62,7 +62,7 @@ class Box: new_start_coord[-1] = 0 new_end_coord[-1] = ifm_shape.depth - if npu_block_type == NpuBlockType.ElementWise and len(new_end_coord) >= 1: + if npu_block_type in (NpuBlockType.ElementWise, NpuBlockType.ConvolutionMxN) and len(new_end_coord) >= 1: new_end_coord[-1] = min(new_end_coord[-1], ifm_shape.depth) if len(new_end_coord) >= 2: new_end_coord[-2] = min(new_end_coord[-2], ifm_shape.width * upscaling_factor) diff --git a/ethosu/vela/high_level_command_to_npu_op.py b/ethosu/vela/high_level_command_to_npu_op.py index c56eb04d..ad9e2664 100644 --- a/ethosu/vela/high_level_command_to_npu_op.py +++ b/ethosu/vela/high_level_command_to_npu_op.py @@ -151,10 +151,11 @@ def create_padding(cmd: NpuStripe, primary_op: Operation) -> NpuPadding: # Indexing from end since a 1x1 Avgpool might have been added with non 4-dimensional input/output, # because of activation function needed to be fused. - if len(cmd.ifm_box.start_coord) >= 2 and cmd.ifm_box.start_coord[-2] > 0: - left = 0 - if len(cmd.ifm_box.end_coord) >= 2 and cmd.ifm_box.end_coord[-2] < cmd.ps.ifm_shapes[0].width: - right = 0 + if not primary_op.attrs.get("force_padding"): + if len(cmd.ifm_box.start_coord) >= 2 and cmd.ifm_box.start_coord[-2] > 0: + left = 0 + if len(cmd.ifm_box.end_coord) >= 2 and cmd.ifm_box.end_coord[-2] < cmd.ps.ifm_shapes[0].width: + right = 0 return NpuPadding(top=top, left=left, bottom=bottom, right=right) |