qnn.py 9.93 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
# Licensed to the Apache Software Foundation (ASF) under one
# or more contributor license agreements.  See the NOTICE file
# distributed with this work for additional information
# regarding copyright ownership.  The ASF licenses this file
# to you 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
#
#   http://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.
#pylint: disable=invalid-name
"""QNN dialect operators."""

from __future__ import absolute_import as _abs
21 22
from tvm.expr import FloatImm, IntImm
from tvm.relay.expr import Tuple
23 24 25 26 27 28 29
from . import _make

def requantize(data,
               input_scale,
               input_zero_point,
               output_scale,
               output_zero_point,
30
               rounding="UPWARD",
31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76
               out_dtype="int8"):
    r"""Requantized operator.

    The requantize operator converts one quantized tensor representation to
    another quantized tensor representation. For the output tensor, we are
    provided with output scale and zero point. The computation is as follows

    Q_output = zp_output +  (scale_input)/(scale_output) * (Q_input - zp_input)

    Parameters
    ----------
    data : tvm.relay.Expr
        The input data to the operator.

    input_scale: float
        The quantization scale for the input tensor.

    input_zero_point: int
        The zero point of the input tensor.

    output_scale: float
        The quantization scale for the output tensor.

    output_zero_point: int
        The zero point of the output tensor.

    rounding : string, optional
        Defines the rounding direction when the value is midway between two
        representable values.

    out_dtype : str, optional
        Specifies the output data type.

    Returns
    -------
    result : tvm.relay.Expr
        The computed result.
    """

    return _make.requantize(data,
                            input_scale,
                            input_zero_point,
                            output_scale,
                            output_zero_point,
                            rounding,
                            out_dtype)
77

78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98

def quantize(data,
             output_scale,
             output_zero_point,
             out_dtype='int8'):
    r""" Quantize op
    This operator takes float32 as input and produces quantized int8 or unit8 as output.
    The input tensor can be of any shape. The output shape is the same as input shape.

    Q_output = clamp((round(input_tensor/output_scale) + output_zero_point),
                     out_dtype::min,
                     out_dtype::max)

    Parameters
    ----------
    data : tvm.relay.Expr
        The input tensor to be quantized. Can be of type float32.
    output_zero_point : int
        The output zero_point.
    output_scale : float
        The output scale.
shoubhik committed
99
    out_dtype : str, optional
100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
        The data type of the input tensor. Can be [int8, uint8]
    Returns
    -------
    result : tvm.relay.Expr
        The computed result.
    """

    return _make.quantize(data,
                          output_scale,
                          output_zero_point,
                          out_dtype)


def dequantize(data,
               input_scale,
               input_zero_point):
    r""" Dequantize op
    This operator takes quantized int8 and unit8 as input and produces
    dequantized float32 as output. The output shape is the same as input shape. The input
    tensor can be of any shape.

    Parameters
    ----------
    data : tvm.relay.Expr
        The input tensor to be dequantized. Can be of type [int8, uint8].
    input_zero_point : int
        The output zero_point.
    input_scale : float
        The output scale.
    Returns
    -------
    result : tvm.relay.Expr
        The computed result.
    """

    return _make.dequantize(data,
                            input_scale,
                            input_zero_point)
138 139


140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174
def concatenate(data,
                input_scales,
                input_zero_points,
                output_scale,
                output_zero_point,
                axis):
    """Concatenate the quantized input tensors along the given axis.

    Parameters
    ----------
    data : Union(List[relay.Expr], Tuple[relay.Expr])
        The list of quantized tensors.

    input_scales : List[float32]
        The list of scales of input quantized tensors.

    input_zero_points : List[int32]
        The list of zero points of input quantized tensors.

    output_scale : float32
        The scale of the output quantized tensor.

    output_zero_point : int32
        The zero point of the output quantized tensor.

    axis : int
        The axis along which the tensors are concatenated.

    Returns
    -------
    result: relay.Expr
        The concatenated quantized tensor.
    """

    data = list(data)
175 176 177 178 179 180 181 182 183 184 185
    if not data:
        raise ValueError("relay.concatenate requires data to be non-empty.")
    if not isinstance(axis, int):
        raise ValueError("For now, we only support integer axis")

    return _make.concatenate(Tuple(data),
                             [FloatImm("float64", x) for x in input_scales],
                             [IntImm("int32", x) for x in input_zero_points],
                             output_scale,
                             output_zero_point,
                             axis)
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265


def conv2d(data,
           kernel,
           input_zero_point,
           kernel_zero_point,
           strides=(1, 1),
           padding=(0, 0),
           dilation=(1, 1),
           groups=1,
           channels=None,
           kernel_size=None,
           data_layout="NCHW",
           kernel_layout="OIHW",
           out_layout="",
           out_dtype="int32"):
    r"""Quantized 2D convolution.

    This operator convolves quantized data with quantized kernel. The scale of
    the output quantized tensor is the product of the kernel_scale and
    input_scale of the input quantized tensors. The zero point of the output
    quantized tensor is 0. By default, the dtype of output is int32. Please also
    refer to Requantize operator to understand how to scale back the int32
    output to (u)int8.

    Parameters
    ----------
    data : tvm.relay.Expr
        The input data to the operator.

    kernel : tvm.relay.Expr
        The kernel expressions.

    input_zero_point: int
           The zero point of the data distribution.

    kernel_zero_point: int
           The zero point of the quantized_kernel distribution.

    strides : tuple of int, optional
        The strides of convolution.

    padding : tuple of int, optional
        The padding of convolution on both sides of inputs before convolution.

    dilation : tuple of int, optional
        Specifies the dilation rate to be used for dilated convolution.

    groups : int, optional
        Number of groups for grouped convolution.

    channels : int, optional
        Number of output channels of this convolution.

    kernel_size : tuple of int, optional
        The spatial of the convolution kernel.

    data_layout : str, optional
        Layout of the input.

    kernel_layout : str, optional
        Layout of the kernel.

    out_layout : str, optional
        Layout of the output, by default, out_layout is the same as data_layout

    out_dtype : str, optional
        Specifies the output data type for mixed precision conv2d.

    Returns
    -------
    result : tvm.relay.Expr
        The computed result.
    """

    return _make.conv2d(data, kernel,
                        input_zero_point, kernel_zero_point,
                        strides, padding, dilation,
                        groups, channels, kernel_size,
                        data_layout, kernel_layout, out_layout, out_dtype)
266 267


shoubhik committed
268 269 270 271 272 273 274
def add(lhs,
        rhs,
        lhs_scale,
        lhs_zero_point,
        rhs_scale,
        rhs_zero_point,
        output_scale,
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313
        output_zero_point):
    """Quantized addition with numpy-style broadcasting.

    Parameters
    ----------
    lhs : relay.Expr
        The left hand side quantized input data.

    rhs : relay.Expr
        The right hand side quantized input data.

    lhs_scale: float
        The scale of the lhs quantized expr.

    lhs_zero_point: int
       The zero point of lhs quantized expr.

    rhs_scale: float
        The scale of the rhs quantized expr.

    rhs_zero_point: int
       The zero point of rhs quantized expr.

    output_scale: float
        The scale of the output quantized expr.

    output_zero_point: int
       The zero point of output quantized expr.

    Returns
    -------
    result : relay.Expr
        The computed result.

    """
    return _make.add(lhs, rhs,
                     lhs_scale, lhs_zero_point,
                     rhs_scale, rhs_zero_point,
                     output_scale, output_zero_point)
shoubhik committed
314 315


316 317 318 319 320 321
def dense(data,
          weight,
          input_zero_point,
          kernel_zero_point,
          units=None,
          out_dtype="int32"):
shoubhik committed
322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351
    """Qnn Dense operator.
    Applies a quantized linear transformation

     .. math::

     `Y = X * W`

    Parameters
    ----------
    data : tvm.relay.Expr
        The quantized input data to the operator.
    weight : tvm.relay.Expr
        The quantized weight expressions.
    units : int, optional
        Number of hidden units of the dense transformation.
    out_dtype : str, optional
        Specifies the output data type for mixed precision dense can be int32 or int16.

    Returns
    -------
    result : tvm.relay.Expr
        The computed result.
    """

    return _make.dense(data,
                       weight,
                       units,
                       input_zero_point,
                       kernel_zero_point,
                       out_dtype)