ethosu/vela/supported_operators.py

# Copyright (C) 2020 Arm Limited or its affiliates. All rights reserved.
#
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the License); you may
# not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an AS IS BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# Description:
# The SupportedOperators class which is a collection of all supported operators and parameter checks.
import numpy as np

from .data_type import BaseType
from .data_type import DataType


class SupportedOperators:
    def __init__(self, softmax_support):
        self.softmax_support = softmax_support
        # Categorised lists of supported operators
        self.npu_pre_ops = set(("QuantizedResizeBilinear", "SplitSliceRead",))
        self.convolution_ops = set(("Conv2DBiasAct", "Conv2D", "QuantizedConv2D",))
        self.depthwise_convolution_ops = set(
            ("DepthwiseConv2dBiasAct", "DepthwiseConv2dNative", "QuantizedDepthwiseConv2D,")
        )
        self.transpose_convolution_ops = set(("Conv2DBackpropInput",))
        self.max_pooling_ops = set(("QuantizedMaxPool", "MaxPool", "MaxPoolAct",))
        self.avg_pooling_ops = set(("QuantizedAvgPool", "AvgPool", "AvgPoolAct",))
        self.pooling_ops = set(("ReduceSum",)) | self.max_pooling_ops | self.avg_pooling_ops
        self.resizing_ops = set(("ResizeBilinear",))
        self.fc_vector_products = set(("QuantizedMatMul", "MatMul", "FullyConnectedAct",))
        self.mac_main_ops = (
            # convolutions
            self.convolution_ops
            # depth-wise convolutions
            | self.depthwise_convolution_ops
            # transpose convolutions
            | self.transpose_convolution_ops
            # pooling
            | self.pooling_ops
            # resizing/upscaling
            | self.resizing_ops
            # FC layers
            | self.fc_vector_products
            # RNN/LSTM/GRU
            | set(("BlockLSTM",))
        )
        self.unary_elem_wise_main_ops = set(("LeakyRelu", "Abs", "CLZ",))
        self.binary_elem_wise_min_max_ops = set(("Minimum", "Maximum",))
        self.binary_elem_wise_shift_ops = set(("SHL", "SHR",))
        self.binary_elem_wise_add_mul_sub = set(
            ("AddAct", "MulAct", "SubAct", "QuantizedAdd", "QuantizedSub", "QuantizedMul", "Mul", "Add", "Sub",)
        )
        self.binary_elem_wise_main_ops = (
            self.binary_elem_wise_min_max_ops | self.binary_elem_wise_add_mul_sub | self.binary_elem_wise_shift_ops
        )
        self.elem_wise_main_ops = self.binary_elem_wise_main_ops | self.unary_elem_wise_main_ops
        self.activation_ops = set(
            (
                "QuantizedRelu",
                "QuantizedRelu1",
                "QuantizedRelu6",
                "Relu",
                "Relu6",
                "ReluN1To1",
                "Sigmoid",
                "Tanh",
                "Softmax",
            )
        )
        self.npu_post_ops = (
            # activation functions
            self.activation_ops
            # concatenation write direction
            | set(("ConcatSliceWrite",))
            # bias add and batch norm
            | set(("QuantizedBiasAdd", "Requantize", "QuantizedBatchNorm", "BiasAdd", "FusedBatchNorm",))
            # Quantization
            | set(("Quantize",))
        )
        self.split_ops = set(("Split", "SplitV", "StridedSlice", "Slice", "UnpackReshaped", "Unpack",))
        self.concat_ops = set(("Concat", "ConcatV2", "QuantizedConcat", "ConcatTFLite", "PackReshaped", "Pack",))
        self.memory_only_ops = (
            set(("Squeeze", "Reshape", "QuantizedReshape", "ExpandDims",)) | self.concat_ops | self.split_ops
        )
        self.shapeless_input_ops = self.binary_elem_wise_main_ops | set(("Split", "SplitV",))
        self.supported_fused_activations = set(("Relu", "Relu6", "ReluN1To1", "Tanh", "Sigmoid", "LUT",))
        self.supported_operators = (
            self.npu_pre_ops | self.mac_main_ops | self.elem_wise_main_ops | self.npu_post_ops | self.memory_only_ops
        )
        # Setup supported operator restriction checkers
        self.supported_operator_restrictions = {}
        self.supported_operator_restrictions.update(
            {op: self.check_convolution_restrictions for op in self.convolution_ops}
        )
        self.supported_operator_restrictions.update(
            {op: self.check_depthwise_convolution_restrictions for op in self.depthwise_convolution_ops}
        )
        self.supported_operator_restrictions.update(
            {op: self.check_transpose_convolution_restrictions for op in self.transpose_convolution_ops}
        )
        self.supported_operator_restrictions.update({op: self.check_pooling_restrictions for op in self.pooling_ops})
        self.supported_operator_restrictions.update({op: self.check_resize_restrictions for op in self.resizing_ops})
        self.supported_operator_restrictions.update(
            {op: self.check_vector_product_restrictions for op in self.fc_vector_products}
        )
        self.supported_operator_restrictions.update(
            {op: self.check_element_wise_restrictions for op in self.elem_wise_main_ops}
        )
        self.supported_operator_restrictions.update(
            {op: self.check_memory_only_restrictions for op in self.memory_only_ops}
        )
        self.supported_operator_restrictions.update(
            {op: self.check_quantization_restrictions_binary_elem_wise for op in self.binary_elem_wise_min_max_ops}
        )
        self.supported_operator_restrictions.update({op: self.check_activation_ops for op in self.activation_ops})

    def is_operator_supported(self, op):
        if op.type not in self.supported_operators:
            return False
        if not self.check_generic_restrictions(op):
            return False
        if op.type in self.supported_operator_restrictions:
            return self.supported_operator_restrictions[op.type](op)
        return True

    def check_generic_restrictions(self, op):
        # check fully defined shapes
        for t in op.inputs:
            if not t:
                continue
            if not t.has_fully_defined_shape():
                print("Warning:", op.type, "has input(s) of undefined shape, placing on CPU")
                return False
            if t.shape == [] and op.type not in self.shapeless_input_ops:
                print(
                    "Warning:",
                    op.type,
                    "has input(s) of shape [].",
                    "Scalar input or broadcasting is not supported for this operator,",
                    "placing on CPU",
                )
                return False
        for t in op.outputs:
            if not t.has_fully_defined_shape():
                print("Warning:", op.type, "has output(s) of undefined shape, placing on CPU")
                return False
            if t.shape == []:
                print(
                    "Warning:",
                    op.type,
                    "has output(s) of shape [].",
                    "Scalar input or broadcasting is not supported for this operator,",
                    "placing on CPU",
                )
                return False

        # check data type
        tensors = [t for t in op.get_ifm_ifm2_weights_ofm() if t is not None]
        if not tensors:
            tensors = op.inputs
        for t in tensors:
            if not (t.dtype.type & BaseType.Int):
                return False
            if (
                t.element_size() > 2
                and op.type
                not in set(("Requantize", "ReduceSum", "CLZ",))
                | self.binary_elem_wise_add_mul_sub
                | self.binary_elem_wise_shift_ops
            ):
                return False
            # check size
            if any(dim > 65536 for dim in t.shape):
                return False

        # check fused activations
        if (
            "fused_activation_function" in op.attrs
            and op.attrs["fused_activation_function"] is not None
            and op.attrs["fused_activation_function"] not in self.supported_fused_activations
        ):
            return False
        return True

    def check_convolution_restrictions(self, op):
        # check stride
        if op.attrs["stride_w"] > 3 or op.attrs["stride_h"] > 3:
            return False

        # check dilation
        dilation_w_factor = op.attrs.get("dilation_w_factor", 1)
        dilation_h_factor = op.attrs.get("dilation_h_factor", 1)
        if dilation_w_factor > 2 or dilation_h_factor > 2:
            return False

        # check data type
        ifm_tensor, _, weight_tensor, _ = op.get_ifm_ifm2_weights_ofm()
        if weight_tensor.element_size() > 1:
            return False

        # check kernel size [HWIO]
        dilated_weight_w = weight_tensor.shape[1] + (weight_tensor.shape[1] - 1) * (dilation_w_factor - 1)
        dilated_weight_h = weight_tensor.shape[0] + (weight_tensor.shape[0] - 1) * (dilation_h_factor - 1)

        if dilated_weight_w > 64 or dilated_weight_h > 64:
            return False

        # check weight sums over [HWI]
        zero_point = weight_tensor.quantization.zero_point
        quant_weights = weight_tensor.quant_values.astype(np.int64)
        weights = quant_weights - zero_point
        totals = np.sum(np.absolute(weights), axis=(0, 1, 2))

        if np.amax(totals) > 127 * 65536:
            return False

        # check batch size
        if ifm_tensor.shape[0] != 1:
            return False
        return True

    def check_depthwise_convolution_restrictions(self, op):
        # check depth
        ifm_tensor, _, _, ofm_tensor = op.get_ifm_ifm2_weights_ofm()
        if op.attrs["depth_multiplier"] > 1 and not (
            (ifm_tensor.shape[3] == 1) and (ofm_tensor.shape[3] == op.attrs["depth_multiplier"])
        ):
            return False
        return self.check_convolution_restrictions(op)

    def check_transpose_convolution_restrictions(self, op):
        # check stride
        stride_h, stride_w = op.attrs["stride_h"], op.attrs["stride_w"]
        if stride_h != stride_w != 2:
            return False

        # check output dimensions
        ifm_tensor, weight_tensor, _, ofm_tensor = op.get_ifm_weights_biases_ofm()
        ifm_h, ifm_w = ifm_tensor.shape[1], ifm_tensor.shape[2]
        ofm_h, ofm_w = ofm_tensor.shape[1], ofm_tensor.shape[2]
        if op.attrs["padding"] == b"SAME":
            if (ofm_h != ifm_h * stride_h) or (ofm_w != ifm_w * stride_w):
                return False
        elif op.attrs["padding"] == b"VALID":
            kernel_h, kernel_w = weight_tensor.shape[0], weight_tensor.shape[1]
            if (ofm_h != (ifm_h) * stride_h + max(kernel_h - stride_h, 0)) or (
                ofm_w != (ifm_w) * stride_w + max(kernel_w - stride_w, 0)
            ):
                return False

        return self.check_convolution_restrictions(op)

    def check_pooling_restrictions(self, op):
        # check stride
        if op.attrs["stride_w"] > 3 or op.attrs["stride_h"] > 3:
            return False

        # check data type
        ifm_tensor, _, _, ofm_tensor = op.get_ifm_ifm2_weights_ofm()
        if ifm_tensor.dtype != ofm_tensor.dtype:
            if op.type != "ReduceSum":
                return False
            # TODO: else check ReduceSum restrictions.

        # check batch size
        if ifm_tensor.shape[0] != 1:
            return False

        if op.type in self.avg_pooling_ops:
            # check kernel size
            if op.attrs["padding"] == b"SAME" and (op.attrs["filter_width"] > 8 or op.attrs["filter_height"] > 8):
                return False
            if op.attrs["padding"] == b"VALID" and (
                op.attrs["filter_width"] * op.attrs["filter_height"] > 256 * 256 or op.attrs["filter_height"] > 256
            ):
                return False

        if op.type in self.max_pooling_ops:
            # check kernel size (any padding)
            if op.attrs["filter_width"] * op.attrs["filter_height"] > 256 * 256 or op.attrs["filter_height"] > 256:
                return False
        return True

    def check_resize_restrictions(self, op):
        # check unsupported upscaling factor
        if op.type == "ResizeBilinear":
            if op.inputs[0].shape[1] == 1 and op.inputs[0].shape[2] == 1:
                return True
            if op.inputs[0].shape == op.outputs[0].shape:
                return True
            upscaled_shape = np.array(op.inputs[0].shape[1:3])
            out_shape = np.array(op.outputs[0].shape[1:3])
            while (upscaled_shape < out_shape).all():
                upscaled_shape *= 2
                if op.attrs["align_corners"]:
                    upscaled_shape -= 1
                if np.array_equal(out_shape, upscaled_shape):
                    return True
        return False

    def check_vector_product_restrictions(self, op):
        # check data type
        ifm_tensor, _, weight_tensor, _ = op.get_ifm_ifm2_weights_ofm()
        if weight_tensor.element_size() > 1:
            return False

        return True

    def check_element_wise_restrictions(self, op):
        # check data type
        ifm_tensor, ifm2_tensor, _, ofm_tensor = op.get_ifm_ifm2_weights_ofm()
        # input and output datatype must match for these operators
        if (
            op.type in self.binary_elem_wise_min_max_ops | self.unary_elem_wise_main_ops
            and ifm_tensor.dtype != ofm_tensor.dtype
        ):
            return False
        if op.type in self.binary_elem_wise_add_mul_sub:
            # both inputs must have same type
            if ifm_tensor.dtype != ifm2_tensor.dtype:
                return False
            # signed input check
            if ifm_tensor.dtype.type & BaseType.Signed:
                # output must be signed
                if ofm_tensor.dtype.type & BaseType.Unsigned:
                    return False
                # and 8, 16 or 32-bit
                if ofm_tensor.element_size() not in (1, 2, 4):
                    return False
            # unsigned input check, output must be same type or int32
            if ifm_tensor.dtype.type & BaseType.Unsigned and not (
                ifm_tensor.dtype == ofm_tensor.dtype or ofm_tensor.dtype == DataType.int32
            ):
                return False
        elif op.type in self.binary_elem_wise_shift_ops | set(("CLZ")):
            if ifm_tensor.dtype != DataType.int32 or ifm2_tensor.dtype != DataType.int32:
                return False
            if op.type in ("CLZ", "SHL") and ofm_tensor.dtype != DataType.int32:
                return False

        # check batch size
        if len(ifm_tensor.shape) > 2 and ifm_tensor.shape[0] != 1:
            return False
        if op.type in self.binary_elem_wise_main_ops:  # if op type is unary, ifm2_tensor is None
            if len(ifm2_tensor.shape) > 2 and ifm2_tensor.shape[0] != 1:
                return False

        # negative alpha values are not supported
        if op.type == "LeakyRelu" and op.attrs["alpha"] < 0:
            return False

        return True

    def check_memory_only_restrictions(self, op):
        if op.type == "StridedSlice":
            # check stride size
            if len(op.inputs) > 3 and any(stride != 1 for stride in op.inputs[3].values):
                return False
            # check "end - begin" doesnt result in any zero or negative elements
            if any((end - begin) <= 0 for begin, end in zip(op.inputs[1].values, op.inputs[2].values)):
                return False
            # check ellipsis_mask
            if op.attrs["ellipsis_mask"] != 0:
                return False
            # check if both new_axis_mask and shrink_axis_mask have bit set
            if op.attrs["new_axis_mask"] != 0 and op.attrs["shrink_axis_mask"] != 0:
                return False
        return True

    def check_quantization_restrictions_binary_elem_wise(self, op):
        # makes sure IFM1, IFM2 and OFM quantization are equal for binary ops
        assert len(op.inputs) >= 2 and len(op.outputs) == 1

        if (
            op.inputs[0].quantization is None
            or not op.inputs[0].quantization.is_scaling_equal(op.inputs[1].quantization)
            or not op.inputs[0].quantization.is_scaling_equal(op.outputs[0].quantization)
        ):
            print(
                "Warning: Input/output tensors with different quantization is unsupported for the", op.type, "operator"
            )
            return False

        return True

    def check_activation_ops(self, op):
        if op.type == "Softmax":
            if not self.softmax_support:
                return False

            ifm_tensor = op.inputs[0]
            ofm_tensor = op.outputs[0]

            # check data type
            if ifm_tensor.dtype != ofm_tensor.dtype:
                return False

            if ifm_tensor.dtype not in (DataType.uint8, DataType.int8, DataType.int16):
                return False

            # check batch size
            if len(ifm_tensor.shape) in (2, 4) and ifm_tensor.shape[0] != 1:
                return False

        return True