diff options
| -rw-r--r-- | ethosu/vela/graph_optimiser_util.py | 4 | ||||
| -rw-r--r-- | ethosu/vela/high_level_command_stream.py | 10 | ||||
| -rw-r--r-- | ethosu/vela/high_level_command_to_npu_op.py | 19 |
3 files changed, 22 insertions, 11 deletions
diff --git a/ethosu/vela/graph_optimiser_util.py b/ethosu/vela/graph_optimiser_util.py index 3e15f126..57fd7dbf 100644 --- a/ethosu/vela/graph_optimiser_util.py +++ b/ethosu/vela/graph_optimiser_util.py @@ -235,10 +235,6 @@ def move_splitsliceread_to_consumer(op, cons_op): 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 ddb24824..abf6d837 100644 --- a/ethosu/vela/high_level_command_stream.py +++ b/ethosu/vela/high_level_command_stream.py @@ -78,8 +78,14 @@ class Box: if strides is not None and skirt is not None: if len(new_start_coord) >= 2: stride = strides[2] - new_start_coord[-2] = max(new_start_coord[-2] * stride - skirt[1], 0) - new_end_coord[-2] = min(new_end_coord[-2] * stride + skirt[3], ifm_shape.width) + # if the current op was combined with a split slice read then the valid ifm range is given by the output + # of the split op + if split_offset is None: + new_start_coord[-2] = max(new_start_coord[-2] * stride - skirt[1], 0) + new_end_coord[-2] = min(new_end_coord[-2] * stride + skirt[3], ifm_shape.width) + else: + new_start_coord[-2] = max(new_start_coord[-2] * stride - skirt[1], split_offset[-2]) + new_end_coord[-2] = min(new_end_coord[-2] * stride + skirt[3], split_shape[-2]) if len(new_start_coord) >= 3: stride = strides[1] diff --git a/ethosu/vela/high_level_command_to_npu_op.py b/ethosu/vela/high_level_command_to_npu_op.py index 318960ec..9abfbd40 100644 --- a/ethosu/vela/high_level_command_to_npu_op.py +++ b/ethosu/vela/high_level_command_to_npu_op.py @@ -147,13 +147,22 @@ def create_padding(cmd: NpuStripe, primary_op: Operation) -> NpuPadding: top = cmd.pad_top bottom = cmd.pad_bottom + # the ifm box coordinate range depends upon whether the primary op was combined with a split slice read + ifm_read_offset = primary_op.read_offsets[0] + ifm_read_shape = primary_op.read_shapes[0] + if ifm_read_offset is None or len(ifm_read_offset) < 2: + box_start_coord_min = 0 + box_end_coord_max = cmd.ps.ifm_shapes[0].width + else: + box_start_coord_min = ifm_read_offset[-2] + box_end_coord_max = ifm_read_shape[-2] + # 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 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 + if len(cmd.ifm_box.start_coord) >= 2 and cmd.ifm_box.start_coord[-2] > box_start_coord_min: + left = 0 + if len(cmd.ifm_box.end_coord) >= 2 and cmd.ifm_box.end_coord[-2] < box_end_coord_max: + right = 0 return NpuPadding(top=top, left=left, bottom=bottom, right=right) |