Skip to content

M
mindspore

magicwindyyd / mindspore
与 Fork 源项目一致

Fork自 MindSpore / mindspore

通知 1
Star 1
Fork 0
M
mindspore
提交 e0fa277a 编写于 作者: C chenzomi
浏览文件
操作

fix bug in fake quant grad

上级 eac1f93e
展开全部 隐藏空白更改
内联 并排
Showing with 897 addition and 357 deletion
mindspore/ccsrc/kernel/gpu/quant/fake_quant_gpu_kernel.cc
浏览文件 @ e0fa277a
......@@ -171,6 +171,6 @@ bool FakeQuantGpuKernel::Launch(const std::vector<AddressPtr> &inputs, const std
return true;
}
MS_REG_GPU_KERNEL(FakeQuantWithMinMax, FakeQuantGpuKernel)
MS_REG_GPU_KERNEL(FakeQuantPerLayer, FakeQuantGpuKernel)
} // namespace kernel
} // namespace mindspore
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.cc
浏览文件 @ e0fa277a
......@@ -30,7 +30,9 @@ FakeQuantGradGpuKernel::FakeQuantGradGpuKernel()
quant_max_(0),
quant_size_(0),
quant_delay_(0),
global_step_(0) {}
global_step_(0),
narrow_range_(false),
symmetric_(false) {}
const std::vector<size_t> &FakeQuantGradGpuKernel::GetInputSizeList() const { return input_size_list_; }
......@@ -59,8 +61,19 @@ bool FakeQuantGradGpuKernel::Init(const CNodePtr &kernel_node) {
MS_LOG(EXCEPTION) << "Attr \'quant_delay_\' " << quant_delay_ << " is less then 0, require larger than 0.";
}
quant_min_ = 0;
quant_max_ = (1 << num_bits_) - 1;
symmetric_ = GetValue<bool>(AnfAlgo::GetCNodePrimitive(kernel_node)->GetAttr("symmetric"));
if (symmetric_) {
quant_min_ = 0 - (1 << (num_bits_ - 1));
quant_max_ = (1 << (num_bits_ - 1)) - 1;
} else {
quant_min_ = 0;
quant_max_ = (1 << num_bits_) - 1;
}
narrow_range_ = GetValue<bool>(AnfAlgo::GetCNodePrimitive(kernel_node)->GetAttr("narrow_range"));
if (narrow_range_) {
quant_min_++;
}
if (quant_size_ == 0) {
quant_size_ = 1;
......@@ -140,6 +153,6 @@ bool FakeQuantGradGpuKernel::Launch(const std::vector<AddressPtr> &inputs, const
return true;
}
MS_REG_GPU_KERNEL(FakeQuantWithMinMaxGrad, FakeQuantGradGpuKernel)
MS_REG_GPU_KERNEL(FakeQuantPerLayerGrad, FakeQuantGradGpuKernel)
} // namespace kernel
} // namespace mindspore
mindspore/ccsrc/kernel/gpu/quant/fake_quant_grad_gpu_kernel.h
浏览文件 @ e0fa277a
......@@ -54,6 +54,8 @@ class FakeQuantGradGpuKernel : public GpuKernel {
int quant_size_;
int quant_delay_;
int global_step_;
bool narrow_range_;
bool symmetric_;
};
} // namespace kernel
} // namespace mindspore
......
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_gpu_kernel.cc
浏览文件 @ e0fa277a
......@@ -175,6 +175,6 @@ bool FakeQuantPerChannelGpuKernel::Launch(const std::vector<AddressPtr> &inputs,
return true;
}
MS_REG_GPU_KERNEL(FakeQuantWithMinMaxPerChannel, FakeQuantPerChannelGpuKernel)
MS_REG_GPU_KERNEL(FakeQuantPerChannel, FakeQuantPerChannelGpuKernel)
} // namespace kernel
} // namespace mindspore
mindspore/ccsrc/kernel/gpu/quant/fake_quant_per_channel_grad_gpu_kernel.cc
浏览文件 @ e0fa277a
......@@ -143,6 +143,6 @@ bool FakeQuantPerChannelGradGpuKernel::Launch(const std::vector<AddressPtr> &inp
return true;
}
MS_REG_GPU_KERNEL(FakeQuantWithMinMaxPerChannelGrad, FakeQuantPerChannelGradGpuKernel)
MS_REG_GPU_KERNEL(FakeQuantPerChannelGrad, FakeQuantPerChannelGradGpuKernel)
} // namespace kernel
} // namespace mindspore
mindspore/nn/layer/quant.py
浏览文件 @ e0fa277a
......@@ -14,6 +14,7 @@
# ============================================================================
"""Aware quantization."""
from functools import partial
import numpy as np
import mindspore.common.dtype as mstype
from mindspore.ops import operations as P
......@@ -101,125 +102,23 @@ class BatchNormFoldCell(Cell):
return batch_mean, batch_std, running_mean, running_std
class FakeQuantWithMinMaxD(Cell):
r"""
Aware Quantization training op of ascend. This OP provide Fake quantization observer
function on data with min and max.
Args:
min_init (int, list): The dimension of channel or 1(layer). Default: -6.
max_init (int, list): The dimension of channel or 1(layer). Default: 6.
num_bits (int): Quantization number bit, support 4 and 8bit. Default: 8.
ema (bool): Exponential Moving Average algorithm update min and max. Default: False.
ema_decay (float): Exponential Moving Average algorithm parameter. Default: 0.999.
per_channel (bool): Quantization by layer or channel. Default: False.
out_channels (int): declarate the min and max channel size, Default: 1.
quant_delay (int): Quantization delay parameters according by global step. Default: 0.
symmetric (bool): Quantization algorithm use symmetric or not. Default: False.
narrow_range (bool): Quantization algorithm use narrow range or not. Default: False.
Inputs:
- **x** (Tensor) - The input of FakeQuantWithMinMax.
Outputs:
Tensor, with the same type and shape as the `x`.
Examples:
>>> fake_quant = nn.FakeQuantWithMinMaxD()
>>> input_x = Tensor(np.array([[1, 2, 1], [-2, 0, -1]]), mindspore.float32)
>>> result = fake_quant(input_x)
"""
def __init__(self,
min_init=-6,
max_init=6,
num_bits=8,
ema=False,
ema_decay=0.999,
per_channel=False,
channel_size=1,
quant_delay=0,
symmetric=False,
narrow_range=False,
training=True):
"""init FakeQuantWithMinMax ascend layer"""
super(FakeQuantWithMinMaxD, self).__init__()
self.min_init = min_init
self.num_bits = num_bits
self.max_init = max_init
self.ema = ema
self.ema_decay = ema_decay
self.per_channel = per_channel
self.channel_size = channel_size
self.quant_delay = quant_delay
self.symmetric = symmetric
self.narrow_range = narrow_range
self.training = training
if not per_channel:
self.fake_quant = P.FakeQuantWithMinMax(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=training)
self.ema_update = P.FakeQuantWithMinMaxUpdate(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=training)
else:
raise RuntimeError("not support per channel")
if isinstance(min_init, Parameter):
self.minq = min_init
self.maxq = max_init
else:
self.minq = Parameter(Tensor(np.array([min_init]).astype(np.float32)),
name='quant_min',
requires_grad=False)
self.maxq = Parameter(Tensor(np.array([max_init]).astype(np.float32)),
name='quant_max',
requires_grad=False)
self.reduce_min = P.ReduceMin()
self.reduce_max = P.ReduceMax()
def extend_repr(self):
s = 'min_init={}, max_init={}, ema={}, ema_decay={}, per_channel={}, channel_size={}, quant_delay={}'.format(
self.min_init, self.max_init, self.ema, self.ema_decay, self.per_channel, self.channel_size,
self.quant_delay)
return s
def construct(self, x, minq, maxq):
if self.training:
min_up, max_up = self.ema_update(x, minq, maxq)
out = self.fake_quant(x, min_up, max_up)
P.Assign()(self.minq, min_up)
P.Assign()(self.maxq, max_up)
else:
out = self.fake_quant(x, minq, maxq)
return out
class FakeQuantWithMinMax(Cell):
r"""
Aware Quantization op. This OP provide Fake quantization observer function on data with min and max.
Args:
min_init (int, list): The dimension of channel or 1(layer). Default: -6.
max_init (int, list): The dimension of channel or 1(layer). Default: 6.
min_init (int, float): The dimension of channel or 1(layer). Default: -6.
max_init (int, float): The dimension of channel or 1(layer). Default: 6.
num_bits (int): Quantization number bit, support 4 and 8bit. Default: 8.
ema (bool): Exponential Moving Average algorithm update min and max. Default: False.
ema_decay (float): Exponential Moving Average algorithm parameter. Default: 0.999.
per_channel (bool): Quantization by layer or channel. Default: False.
channel_axis (int): Quantization by channel axis. Default: 1.
out_channels (int): declarate the min and max channel size, Default: 1.
quant_delay (int): Quantization delay parameters according by global step. Default: 0.
symmetric (bool): Quantization algorithm use symmetric or not. Default: False.
narrow_range (bool): Quantization algorithm use narrow range or not. Default: False.
training (bool): Quantization algorithm training or not. Default: True.
Inputs:
- **x** (Tensor) - The input of FakeQuantWithMinMax.
......@@ -240,95 +139,82 @@ class FakeQuantWithMinMax(Cell):
ema=False,
ema_decay=0.999,
per_channel=False,
channel_axis=1,
out_channels=1,
quant_delay=0,
symmetric=False,
narrow_range=False):
narrow_range=False,
training=True):
"""init FakeQuantWithMinMax layer"""
super(FakeQuantWithMinMax, self).__init__()
self.min_init = min_init
self.num_bits = num_bits
self.max_init = max_init
self.num_bits = num_bits
self.ema = ema
self.ema_decay = ema_decay
self.per_channel = per_channel
self.out_channels = out_channels
self.channel_axis = channel_axis
self.quant_delay = quant_delay
self.symmetric = symmetric
self.narrow_range = narrow_range
self.training = training
self.is_ascend = context.get_context('device_target') == "Ascend"
if per_channel:
# init tensor min and max for fake quant op
if self.per_channel:
min_array = np.array([self.min_init for i in range(0, self.out_channels)]).astype(np.float32)
max_array = np.array([self.max_init for i in range(0, self.channel_size)]).astype(np.float32)
self.minq = Parameter(Tensor(min_array), name='quant_min', requires_grad=False)
self.maxq = Parameter(Tensor(max_array), name='quant_max', requires_grad=False)
self.fake_quant_train = P.FakeQuantWithMinMaxPerChannel(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=True)
self.fake_quant_infer = P.FakeQuantWithMinMaxPerChannel(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=False)
max_array = np.array([self.max_init for i in range(0, self.out_channels)]).astype(np.float32)
else:
min_array = np.array([min_init]).reshape(1).astype(np.float32)
max_array = np.array([max_init]).reshape(1).astype(np.float32)
self.minq = Parameter(Tensor(min_array), name='quant_min', requires_grad=False)
self.maxq = Parameter(Tensor(max_array), name='quant_max', requires_grad=False)
if context.get_context('device_target') == "Ascend":
self.fake_quant_train = FakeQuantWithMinMaxD(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=True,
min_init=self.minq,
max_init=self.maxq)
self.fake_quant_infer = FakeQuantWithMinMaxD(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=False,
min_init=self.minq,
max_init=self.maxq)
elif context.get_context('device_target') == "GPU":
self.fake_quant_train = P.FakeQuantWithMinMax(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=True)
self.fake_quant_infer = P.FakeQuantWithMinMax(num_bits=self.num_bits,
ema=self.ema,
ema_decay=ema_decay,
quant_delay=quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=False)
else:
raise ValueError("Not support platform.")
min_array = np.array([self.min_init]).reshape(1).astype(np.float32)
max_array = np.array([self.max_init]).reshape(1).astype(np.float32)
self.minq = Parameter(Tensor(min_array), name='quant_min', requires_grad=False)
self.maxq = Parameter(Tensor(max_array), name='quant_max', requires_grad=False)
# init fake quant relative op
if per_channel:
quant_fun = partial(P.FakeQuantPerChannel, channel_axis=self.channel_axis)
ema_fun = partial(P.FakeQuantMinMaxPerChannelUpdate, channel_axis=self.channel_axis)
else:
quant_fun = P.FakeQuantPerLayer
ema_fun = P.FakeQuantMinMaxPerLayerUpdate
if self.is_ascend:
self.fake_quant = quant_fun(num_bits=self.num_bits,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=self.training)
else:
self.fake_quant = quant_fun(num_bits=self.num_bits,
ema=self.ema,
ema_decay=ema_decay,
quant_delay=quant_delay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=self.training)
if self.ema:
self.ema_update = ema_fun(num_bits=self.num_bits,
ema=self.ema,
ema_decay=self.ema_decay,
symmetric=self.symmetric,
narrow_range=self.narrow_range,
training=self.training)
def extend_repr(self):
s = 'min={}, max={}, ema={}, ema_decay={}, per_channel={}, quant_delay={}'.format(
self.min_init, self.max_init, self.ema, self.ema_decay, self.per_channel, self.quant_delay)
s = 'num_bits={}, symmetric={}, narrow_range={}, ema={}({}), per_channel={}({}, {}), ' \
'quant_delay={}, min_init={}, max_init={}'.format(
self.num_bits, self.symmetric, self.narrow_range, self.ema, self.ema_decay, self.per_channel,
self.channel_axis, self.out_channels, self.quant_delay, self.min_init, self.max_init)
return s
def construct(self, x):
if self.training:
out = self.fake_quant_train(x, self.minq, self.maxq)
if self.ema and self.is_ascend:
min_up, max_up = self.ema_update(x, self.minq, self.maxq)
out = self.fake_quant(x, min_up, max_up)
P.Assign()(self.minq, min_up)
P.Assign()(self.maxq, max_up)
else:
out = self.fake_quant_infer(x, self.minq, self.maxq)
out = self.fake_quant(x, self.minq, self.maxq)
return out
......@@ -420,7 +306,6 @@ class Conv2dBatchNormQuant(Cell):
self.per_channel = per_channel
self.symmetric = symmetric
self.narrow_range = narrow_range
self.channel_axis = int(group > 1)
self.is_gpu = context.get_context('device_target') == "GPU"
# initialize convolution op and Parameter
......@@ -435,6 +320,7 @@ class Conv2dBatchNormQuant(Cell):
dilation=self.dilation)
if weight_init is None:
weight_init = initializer('normal', [1, in_channels, *self.kernel_size])
channel_axis = 1
else:
self.conv = P.Conv2D(out_channel=out_channels,
kernel_size=self.kernel_size,
......@@ -445,6 +331,7 @@ class Conv2dBatchNormQuant(Cell):
group=group)
if weight_init is None:
weight_init = initializer('normal', [out_channels, in_channels // group, *self.kernel_size])
channel_axis = 0
self.weight = Parameter(weight_init, name='weight')
# initialize batchnorm Parameter
......@@ -472,7 +359,7 @@ class Conv2dBatchNormQuant(Cell):
symmetric=symmetric,
narrow_range=narrow_range)
self.batchnorm_fold = BatchNormFoldCell(epsilon=eps, momentum=momentum, freeze_bn=freeze_bn)
self.correct_mul = P.CorrectionMul(self.channel_axis)
self.correct_mul = P.CorrectionMul(channel_axis)
if context.get_context('device_target') == "Ascend":
self.batchnorm_fold2_train = P.BatchNormFold2_D(freeze_bn=freeze_bn)
self.batchnorm_fold2_infer = P.BatchNormFold2_D(freeze_bn=0)
......@@ -522,7 +409,7 @@ class Conv2dBatchNormQuant(Cell):
out = self.batchnorm_fold2_train(out, self.beta, self.gamma, batch_std, batch_mean, running_std)
F.control_depend(out, self.assignadd(self.step, self.one))
else:
out = self.batchnorm_fold2_infer(out, self.beta, self.gamma, batch_std, batch_mean, running_std)
out = self.batchnorm_fold2_infer(out, self.beta, self.gamma, running_std, running_mean, running_std)
return out
......
mindspore/ops/_grad/grad_quant_ops.py
浏览文件 @ e0fa277a
......@@ -20,10 +20,11 @@ from .grad_base import bprop_getters
from ..composite.multitype_ops.zeros_like_impl import zeros_like
@bprop_getters.register(P.FakeQuantWithMinMax)
@bprop_getters.register(P.FakeQuantPerLayer)
def get_bprop_fakequant_with_minmax(self):
"""Generate bprop for FakeQuantWithMinMax for GPU and Ascend"""
op = P.FakeQuantWithMinMaxGrad(num_bits=self.num_bits, quant_delay=self.quant_delay)
"""Generate bprop for FakeQuantPerLayer for GPU and Ascend"""
op = P.FakeQuantPerLayerGrad(
num_bits=self.num_bits, quant_delay=self.quant_delay)
def bprop(x, x_min, x_max, out, dout):
dx = op(dout, x, x_min, x_max)
......@@ -32,10 +33,14 @@ def get_bprop_fakequant_with_minmax(self):
return bprop
@bprop_getters.register(P.FakeQuantWithMinMaxPerChannel)
@bprop_getters.register(P.FakeQuantPerChannel)
def get_bprop_fakequant_with_minmax_perchannel(self):
"""Generate bprop for FakeQuantWithMinMaxPerChannel for GPU"""
op = P.FakeQuantWithMinMaxPerChannelGrad(num_bits=self.num_bits, quant_delay=self.quant_delay)
"""Generate bprop for FakeQuantPerChannel"""
op = P.FakeQuantPerChannelGrad(num_bits=self.num_bits,
quant_delay=self.quant_delay,
symmetric=self.symmetric,
narrow_range=self.symmetric,
channel_axis=self.channel_axis)
def bprop(x, x_min, x_max, out, dout):
dx = op(dout, x, x_min, x_max)
......@@ -77,7 +82,7 @@ def get_bprop_batchnorm_fold2(self):
d_batch_std, d_batch_mean, d_beta, d_gamma, d_x = op_f(dout, x, gamma, batch_std, batch_mean, running_std,
running_mean, global_step)
return d_x, d_beta, d_gamma, d_batch_std, d_batch_mean, zeros_like(running_std), zeros_like(running_mean), \
zeros_like(global_step)
zeros_like(global_step)
return bprop
......@@ -117,9 +122,19 @@ def get_bprop_batchnorm_fold2_(self):
return bprop
@bprop_getters.register(P.FakeQuantWithMinMaxUpdate)
def get_bprop_fakequant_with_minmax_update(self):
"""Generate bprop for FakeQuantWithMinMaxUpdate for Ascend"""
@bprop_getters.register(P.FakeQuantMinMaxPerLayerUpdate)
def get_bprop_fakequant_with_minmax_per_layer_update(self):
"""Generate bprop for FakeQuantMinMaxPerLayerUpdate for Ascend"""
def bprop(x, x_min, x_max, out, dout):
return zeros_like(x), zeros_like(x_min), zeros_like(x_max)
return bprop
@bprop_getters.register(P.FakeQuantMinMaxPerChannelUpdate)
def get_bprop_fakequant_with_minmax_per_channel_update(self):
"""Generate bprop for FakeQuantMinMaxPerChannelUpdate for Ascend"""
def bprop(x, x_min, x_max, out, dout):
return zeros_like(x), zeros_like(x_min), zeros_like(x_max)
......
mindspore/ops/_op_impl/_custom_op/fake_quant_minmax_perchannel_update.py 0 → 100644
浏览文件 @ e0fa277a
# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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
#
# 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.
# ============================================================================
"""FakeQuantMinMaxPerChannelUpdate op"""
import te.lang.cce
from te import tvm
from te.platform.fusion_manager import fusion_manager
from topi import generic
from topi.cce import util
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
fake_quant_min_max_per_channel_update_op_info = TBERegOp("FakeQuantMinMaxPerChannelUpdate") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("fake_quant_min_max_per_channel_update.so") \
.compute_cost(10) \
.kernel_name("fake_quant_min_max_per_channel_update") \
.partial_flag(True) \
.attr("ema", "optional", "bool", "all") \
.attr("ema_decay", "optional", "float", "all") \
.attr("symmetric", "optional", "bool", "all") \
.attr("narrow_range", "optional", "bool", "all") \
.attr("training", "optional", "bool", "all") \
.attr("num_bits", "optional", "int", "all") \
.attr("channel_axis", "optional", "int", "all") \
.input(0, "x", None, "required", None) \
.input(1, "min", None, "required", None) \
.input(2, "max", None, "required", None) \
.output(0, "min_up", True, "required", "all") \
.output(1, "max_up", True, "required", "all") \
.dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD,
DataType.F32_5HD) \
.get_op_info()
@op_info_register(fake_quant_min_max_per_channel_update_op_info)
def _fake_quant_min_max_per_channel_update_tbe():
"""FakeQuantPerChannelUpdate TBE register"""
return
@fusion_manager.register("fake_quant_min_max_per_channel_update")
def fake_quant_min_max_per_channel_update_compute(x, min_val, max_val,
ema, ema_decay, quant_min, quant_max, training, channel_axis,
kernel_name="fake_quant_min_max_per_channel_update"):
"""FakeQuantPerChannelUpdate compute"""
shape_min = te.lang.cce.util.shape_to_list(min_val.shape)
if not ema:
ema_decay = 0.0
if training:
# CalMinMax
axis = [0, 2, 3]
x_min = te.lang.cce.reduce_min(x, axis=axis)
x_max = te.lang.cce.reduce_max(x, axis=axis)
x_min = te.lang.cce.broadcast(x_min, shape_min)
x_max = te.lang.cce.broadcast(x_max, shape_min)
min_val = te.lang.cce.vadd(te.lang.cce.vmuls(
min_val, ema_decay), te.lang.cce.vmuls(x_min, (1 - ema_decay)))
max_val = te.lang.cce.vadd(te.lang.cce.vmuls(
max_val, ema_decay), te.lang.cce.vmuls(x_max, (1 - ema_decay)))
min_val = te.lang.cce.vmins(min_val, 0)
max_val = te.lang.cce.vmaxs(max_val, 0)
return [min_val, max_val]
@util.check_input_type(dict, dict, dict, dict, dict, bool, float, bool, bool, bool, int, int, str)
def fake_quant_min_max_per_channel_update(x, min_val, max_val, min_up, max_up,
ema, ema_decay, symmetric, narrow_range, training, num_bits, channel_axis,
kernel_name="fake_quant_min_max_per_channel_update"):
"""FakeQuantPerLayer op"""
x_shape = x.get("ori_shape")
x_format = x.get("format")
x_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
min_dtype = min_val.get("dtype")
max_shape = max_val.get("ori_shape")
max_dtype = max_val.get("dtype")
util.check_kernel_name(kernel_name)
util.check_shape_rule(x_shape)
util.check_shape_rule(min_shape, 1, 1, x_shape[channel_axis])
util.check_shape_rule(max_shape, 1, 1, x_shape[channel_axis])
util.check_tensor_shape_size(x_shape)
util.check_tensor_shape_size(min_shape)
util.check_tensor_shape_size(max_shape)
check_list = ["float32", "float16"]
x_dtype = x_dtype.lower()
min_dtype = min_dtype.lower()
max_dtype = max_dtype.lower()
util.check_dtype_rule(x_dtype, check_list)
util.check_dtype_rule(min_dtype, check_list)
util.check_dtype_rule(max_dtype, check_list)
if symmetric:
quant_min = 0 - 2 ** (num_bits - 1)
quant_max = 2 ** (num_bits - 1) - 1
else:
quant_min = 0
quant_max = 2 ** num_bits - 1
if narrow_range:
quant_min = quant_min + 1
shape_c = [min_val.get("shape")[1], min_val.get("shape")[-1]]
input_data = tvm.placeholder(x.get("shape"), name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_c, name="min_val", dtype=x_dtype)
max_data = tvm.placeholder(shape_c, name="max_val", dtype=x_dtype)
res_list = fake_quant_min_max_per_channel_update_compute(input_data, min_data, max_data,
ema, ema_decay, quant_min, quant_max, training, channel_axis, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res_list)
tensor_list = [input_data, min_data, max_data] + list(res_list)
config = {"print_ir": False,
"name": kernel_name,
"tensor_list": tensor_list}
te.lang.cce.cce_build_code(sch, config)
mindspore/ops/_op_impl/_custom_op/fake_quant_with_min_max_update.py → mindspore/ops/_op_impl/_custom_op/fake_quant_minmax_perlayer_update.py
浏览文件 @ e0fa277a
......@@ -13,7 +13,7 @@
# limitations under the License.
# ============================================================================
"""FakeQuantWithMinMaxUpdate op"""
"""FakeQuantMinMaxPerLayerUpdate op"""
from functools import reduce as functools_reduce
import te.lang.cce
from te import tvm
......@@ -23,12 +23,12 @@ from topi.cce import util
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
fake_quant_update_op_info = TBERegOp("FakeQuantWithMinMaxUpdate") \
fake_quant_minmax_update_op_info = TBERegOp("FakeQuantMinMaxPerLayerUpdate") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("fake_quant_with_min_max_update.so") \
.binfile_name("fake_quant_minmax_update.so") \
.compute_cost(10) \
.kernel_name("fake_quant_with_min_max_update") \
.kernel_name("fake_quant_minmax_update") \
.partial_flag(True) \
.attr("ema", "optional", "bool", "all") \
.attr("ema_decay", "optional", "float", "all") \
......@@ -36,7 +36,6 @@ fake_quant_update_op_info = TBERegOp("FakeQuantWithMinMaxUpdate") \
.attr("narrow_range", "optional", "bool", "all") \
.attr("training", "optional", "bool", "all") \
.attr("num_bits", "optional", "int", "all") \
.attr("quant_delay", "optional", "int", "all") \
.input(0, "x", None, "required", None) \
.input(1, "min", None, "required", None) \
.input(2, "max", None, "required", None) \
......@@ -47,16 +46,16 @@ fake_quant_update_op_info = TBERegOp("FakeQuantWithMinMaxUpdate") \
.get_op_info()
@op_info_register(fake_quant_update_op_info)
def _fake_quant_update_tbe():
"""_FakeQuantWithMinMaxUpdate TBE register"""
@op_info_register(fake_quant_minmax_update_op_info)
def _fake_quant_minmax_update_tbe():
"""FakeQuantMinMaxPerLayerUpdate TBE register"""
return
@fusion_manager.register("fake_quant_with_min_max_update")
def fake_quant_with_min_max_update_compute(x, min_val, max_val, ema, ema_decay, quant_min, quant_max, training,
kernel_name="fake_quant_update"):
"""FakeQuantWithMinMaxUpdate compute"""
@fusion_manager.register("fake_quant_minmax_update")
def fake_quant_minmax_update_compute(x, min_val, max_val, ema, ema_decay, quant_min, quant_max, training,
kernel_name="fake_quant_minmax_update"):
"""FakeQuantMinMaxPerLayerUpdate compute"""
shape = te.lang.cce.util.shape_to_list(x.shape)
shape_min = te.lang.cce.util.shape_to_list(min_val.shape)
min_val = te.lang.cce.broadcast(min_val, shape_min, x.dtype)
......@@ -70,19 +69,21 @@ def fake_quant_with_min_max_update_compute(x, min_val, max_val, ema, ema_decay,
x_max = te.lang.cce.reduce_max(x, axis=axis)
x_min = te.lang.cce.broadcast(x_min, shape_min)
x_max = te.lang.cce.broadcast(x_max, shape_min)
min_val = te.lang.cce.vadd(te.lang.cce.vmuls(min_val, ema_decay), te.lang.cce.vmuls(x_min, (1 - ema_decay)))
max_val = te.lang.cce.vadd(te.lang.cce.vmuls(max_val, ema_decay), te.lang.cce.vmuls(x_max, (1 - ema_decay)))
min_val = te.lang.cce.vadd(te.lang.cce.vmuls(
min_val, ema_decay), te.lang.cce.vmuls(x_min, (1 - ema_decay)))
max_val = te.lang.cce.vadd(te.lang.cce.vmuls(
max_val, ema_decay), te.lang.cce.vmuls(x_max, (1 - ema_decay)))
min_val = te.lang.cce.vmins(min_val, 0)
max_val = te.lang.cce.vmaxs(max_val, 0)
return [min_val, max_val]
@util.check_input_type(dict, dict, dict, dict, dict, bool, float, bool, bool, bool, int, int, str)
def fake_quant_with_min_max_update(x, min_val, max_val, min_up, max_up,
ema, ema_decay, symmetric, narrow_range, training, num_bits, quant_delay,
kernel_name="fake_quant_update"):
"""FakeQuantWithMinMax op"""
@util.check_input_type(dict, dict, dict, dict, dict, bool, float, bool, bool, bool, int, str)
def fake_quant_minmax_update(x, min_val, max_val, min_up, max_up,
ema, ema_decay, symmetric, narrow_range, training, num_bits,
kernel_name="fake_quant_minmax_update"):
"""FakeQuantPerLayer op"""
input_shape = x.get("shape")
input_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
......@@ -123,8 +124,8 @@ def fake_quant_with_min_max_update(x, min_val, max_val, min_up, max_up,
input_data = tvm.placeholder(input_shape, name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_min, name="min_data", dtype=min_dtype)
max_data = tvm.placeholder(shape_min, name="max_data", dtype=max_dtype)
res_list = fake_quant_with_min_max_update_compute(input_data, min_data, max_data,
ema, ema_decay, quant_min, quant_max, training, kernel_name)
res_list = fake_quant_minmax_update_compute(input_data, min_data, max_data,
ema, ema_decay, quant_min, quant_max, training, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res_list)
......
mindspore/ops/_op_impl/_custom_op/fake_quant_perchannel.py 0 → 100644
浏览文件 @ e0fa277a
# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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
#
# 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.
# ============================================================================
"""FakeQuantPerChannel op"""
import te.lang.cce
from te import tvm
from te.platform.fusion_manager import fusion_manager
from topi import generic
from topi.cce import util
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
fake_quant_perchannel_op_info = TBERegOp("FakeQuantPerChannel") \
.fusion_type("ELEMWISE") \
.async_flag(False) \
.binfile_name("fake_quant_perchannel.so") \
.compute_cost(10) \
.kernel_name("fake_quant_perchannel") \
.partial_flag(True) \
.attr("symmetric", "optional", "bool", "all") \
.attr("narrow_range", "optional", "bool", "all") \
.attr("num_bits", "optional", "int", "all") \
.attr("channel_axis", "optional", "int", "all") \
.input(0, "x", None, "required", None) \
.input(1, "min", None, "required", None) \
.input(2, "max", None, "required", None) \
.output(0, "y", True, "required", "all") \
.dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default, DataType.F16_Default) \
.dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default, DataType.F32_Default) \
.dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD) \
.get_op_info()
@op_info_register(fake_quant_perchannel_op_info)
def _fake_quant_perchannel_tbe():
"""FakeQuantPerChannel TBE register"""
return
@fusion_manager.register("fake_quant_perchannel")
def fake_quant_perchannel_compute(x, min_val, max_val, y, quant_min, quant_max,
kernel_name="fake_quant_perchannel"):
"""FakeQuantPerChannel"""
x_shape = te.lang.cce.util.shape_to_list(x.shape)
minmax_shape = te.lang.cce.util.shape_to_list(min_val.shape)
quant_min = tvm.const(quant_min, x.dtype)
quant_max = tvm.const(quant_max, x.dtype)
quant_min = te.lang.cce.broadcast(quant_min, minmax_shape, x.dtype)
quant_max = te.lang.cce.broadcast(quant_max, minmax_shape, x.dtype)
# CalNudge(NudgeMinMax)
scale = te.lang.cce.vdiv(te.lang.cce.vsub(
max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
zp_from_min = te.lang.cce.vsub(quant_min, te.lang.cce.vdiv(min_val, scale))
# Nudge zero point
nudge_zp_ = te.lang.cce.vmin(
quant_max, te.lang.cce.vmax(quant_min, zp_from_min))
nudge_zp = te.lang.cce.floor(te.lang.cce.vadds(nudge_zp_, 0.5))
nudge_min = te.lang.cce.vmul(te.lang.cce.vsub(quant_min, nudge_zp), scale)
nudge_max = te.lang.cce.vmul(te.lang.cce.vsub(quant_max, nudge_zp), scale)
# FakeQuant
nudge_min_b = te.lang.cce.broadcast(nudge_min, x_shape)
nudge_max_b = te.lang.cce.broadcast(nudge_max, x_shape)
scale_b = te.lang.cce.broadcast(scale, x_shape)
input_x = te.lang.cce.vmin(nudge_max_b, te.lang.cce.vmax(nudge_min_b, x))
nudge_input_ = te.lang.cce.vdiv(
te.lang.cce.vsub(input_x, nudge_min_b), scale_b)
nudge_input = te.lang.cce.floor(te.lang.cce.vadds(nudge_input_, 0.5))
res = te.lang.cce.vadd(te.lang.cce.vmul(nudge_input, scale_b), nudge_min_b)
return res
@util.check_input_type(dict, dict, dict, dict, bool, bool, int, int, str)
def fake_quant_perchannel(x, min_val, max_val, y,
symmetric, narrow_range, num_bits, channel_axis,
kernel_name="fake_quant_perchannel"):
"""FakeQuantPerChannel"""
x_shape = x.get("shape")
x_shape_ = x.get("ori_shape")
x_format = x.get("format")
x_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
min_dtype = min_val.get("dtype")
max_shape = max_val.get("ori_shape")
max_dtype = max_val.get("dtype")
util.check_kernel_name(kernel_name)
util.check_shape_rule(x_shape)
util.check_shape_rule(min_shape, 1, 1, x_shape_[channel_axis])
util.check_shape_rule(max_shape, 1, 1, x_shape_[channel_axis])
util.check_tensor_shape_size(x_shape)
util.check_tensor_shape_size(min_shape)
util.check_tensor_shape_size(max_shape)
check_list = ["float32", "float16"]
x_dtype = x_dtype.lower()
min_dtype = min_dtype.lower()
max_dtype = max_dtype.lower()
util.check_dtype_rule(x_dtype, check_list)
util.check_dtype_rule(min_dtype, check_list)
util.check_dtype_rule(max_dtype, check_list)
if symmetric:
quant_min = 0 - 2 ** (num_bits - 1)
quant_max = 2 ** (num_bits - 1) - 1
else:
quant_min = 0
quant_max = 2 ** num_bits - 1
if narrow_range:
quant_min = quant_min + 1
shape_c = [1] * len(x_shape)
shape_c[channel_axis] = min_val.get("ori_shape")[0]
if x_format == "NC1HWC0" and channel_axis == 1:
shape_c = min_val.get("shape")
input_data = tvm.placeholder(x_shape, name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_c, name="min_val", dtype=x_dtype)
max_data = tvm.placeholder(shape_c, name="max_val", dtype=x_dtype)
res = fake_quant_perchannel_compute(input_data, min_data, max_data, y,
quant_min, quant_max, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res)
tensor_list = [input_data, min_data, max_data, res]
config = {"print_ir": False,
"name": kernel_name,
"tensor_list": tensor_list}
te.lang.cce.cce_build_code(sch, config)
mindspore/ops/_op_impl/_custom_op/fake_quant_perchannel_grad.py 0 → 100644
浏览文件 @ e0fa277a
# Copyright 2020 Huawei Technologies Co., Ltd
#
# 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
#
# 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.
# ============================================================================
"""FakeQuantPerChannelGrad op"""
import te.lang.cce
from te import tvm
from te.platform.fusion_manager import fusion_manager
from topi import generic
from topi.cce import util
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
SHAPE_SIZE_LIMIT = 2147483648
D_TYPE = 'float32'
fake_quant_perchannel_grad_op_info = TBERegOp("FakeQuantPerChannelGrad") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("fake_quant_perchannel_grad.so") \
.compute_cost(10) \
.kernel_name("fake_quant_perchannel_grad") \
.partial_flag(True) \
.attr("symmetric", "optional", "bool", "all") \
.attr("narrow_range", "optional", "bool", "all") \
.attr("num_bits", "optional", "int", "all") \
.attr("channel_axis", "optional", "int", "all") \
.input(0, "dout", None, "required", None) \
.input(1, "x", None, "required", None) \
.input(2, "min", None, "required", None) \
.input(3, "max", None, "required", None) \
.output(0, "dx", True, "required", "all") \
.dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default, DataType.F16_Default,
DataType.F16_Default) \
.dtype_format(DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD, DataType.F16_5HD) \
.dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default, DataType.F32_Default,
DataType.F32_Default) \
.dtype_format(DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD, DataType.F32_5HD) \
.get_op_info()
def _less_compare_float32(data_x, data_y):
"""_less_compare_float32 compute"""
input_shape = te.lang.cce.util.shape_to_list(data_x.shape)
min_value = tvm.const(2 ** (-126), dtype=D_TYPE)
max_value = tvm.const(2 ** 62, dtype=D_TYPE)
factor_value = tvm.const(2 ** 2, dtype=D_TYPE)
data_zero = te.lang.cce.broadcast(
tvm.const(0, dtype=D_TYPE), input_shape, D_TYPE)
min_value_tensor = te.lang.cce.vadds(data_zero, min_value)
res_sub = te.lang.cce.vsub(data_y, data_x)
res_min = te.lang.cce.vmin(res_sub, min_value_tensor)
res_max = te.lang.cce.vmax(res_min, data_zero)
res_max_mul = te.lang.cce.vmuls(res_max, max_value)
res_max_mul_max = te.lang.cce.vmuls(res_max_mul, max_value)
res = te.lang.cce.vmuls(res_max_mul_max, factor_value)
return res
@op_info_register(fake_quant_perchannel_grad_op_info)
def _fake_quant_perchannel_grad_tbe():
"""FakeQuantPerChannelGrad TBE register"""
return
@fusion_manager.register("fake_quant_perchannel_grad")
def fake_quant_perchannel_grad_compute(dout, x, min_val, max_val, quant_min, quant_max,
kernel_name="fake_quant_perchannel_grad"):
"""FakeQuantPerChannelGrad"""
x_shape = te.lang.cce.util.shape_to_list(x.shape)
minmax_shape = te.lang.cce.util.shape_to_list(min_val.shape)
quant_min = tvm.const(quant_min, x.dtype)
quant_max = tvm.const(quant_max, x.dtype)
quant_min = te.lang.cce.broadcast(quant_min, minmax_shape, x.dtype)
quant_max = te.lang.cce.broadcast(quant_max, minmax_shape, x.dtype)
# CalNudge(NudgeMinMax)
scale = te.lang.cce.vdiv(te.lang.cce.vsub(
max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
zp_from_min = te.lang.cce.vsub(quant_min, te.lang.cce.vdiv(min_val, scale))
# Nudge zero point
nudge_zp_ = te.lang.cce.vmin(
quant_max, te.lang.cce.vmax(quant_min, zp_from_min))
nudge_zp = te.lang.cce.floor(te.lang.cce.vadds(nudge_zp_, 0.5))
nudge_min = te.lang.cce.vmul(te.lang.cce.vsub(quant_min, nudge_zp), scale)
nudge_max = te.lang.cce.vmul(te.lang.cce.vsub(quant_max, nudge_zp), scale)
# FakeQuant Grad
nudge_min_b = te.lang.cce.broadcast(nudge_min, x_shape)
nudge_max_b = te.lang.cce.broadcast(nudge_max, x_shape)
bool_over_min = _less_compare_float32(nudge_min_b, x)
bool_less_max = _less_compare_float32(x, nudge_max_b)
bool_between = te.lang.cce.vmul(bool_over_min, bool_less_max)
res = te.lang.cce.vmul(dout, bool_between)
return res
@util.check_input_type(dict, dict, dict, dict, dict, bool, bool, int, int, str)
def fake_quant_perchannel_grad(dout, x, min_val, max_val, dx,
symmetric, narrow_range, num_bits, channel_axis,
kernel_name="fake_quant_perchannel_grad"):
"""FakeQuantPerChannelGrad"""
x_shape = x.get("shape")
x_shape_ = x.get("ori_shape")
x_format = x.get("format")
x_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
min_dtype = min_val.get("dtype")
max_shape = max_val.get("ori_shape")
max_dtype = max_val.get("dtype")
util.check_kernel_name(kernel_name)
util.check_shape_rule(x_shape)
util.check_shape_rule(min_shape, 1, 1, x_shape_[channel_axis])
util.check_shape_rule(max_shape, 1, 1, x_shape_[channel_axis])
util.check_tensor_shape_size(x_shape)
util.check_tensor_shape_size(min_shape)
util.check_tensor_shape_size(max_shape)
check_list = ["float32", "float16"]
x_dtype = x_dtype.lower()
min_dtype = min_dtype.lower()
max_dtype = max_dtype.lower()
util.check_dtype_rule(x_dtype, check_list)
util.check_dtype_rule(min_dtype, check_list)
util.check_dtype_rule(max_dtype, check_list)
if symmetric:
quant_min = 0 - 2 ** (num_bits - 1)
quant_max = 2 ** (num_bits - 1) - 1
else:
quant_min = 0
quant_max = 2 ** num_bits - 1
if narrow_range:
quant_min = quant_min + 1
shape_c = [1] * len(x_shape)
shape_c[channel_axis] = min_val.get("ori_shape")[0]
if x_format == "NC1HWC0" and channel_axis == 1:
shape_c = min_val.get("shape")
dout_data = tvm.placeholder(x_shape, name="dout", dtype=x_dtype)
input_data = tvm.placeholder(x_shape, name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_c, name="min_val", dtype=x_dtype)
max_data = tvm.placeholder(shape_c, name="max_val", dtype=x_dtype)
res = fake_quant_perchannel_grad_compute(dout_data, input_data, min_data, max_data,
quant_min, quant_max, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res)
tensor_list = [dout_data, input_data, min_data, max_data, res]
config = {"print_ir": False,
"name": kernel_name,
"tensor_list": tensor_list}
te.lang.cce.cce_build_code(sch, config)
mindspore/ops/_op_impl/_custom_op/fake_quant_with_min_max.py → mindspore/ops/_op_impl/_custom_op/fake_quant_perlayer.py
浏览文件 @ e0fa277a
......@@ -13,8 +13,7 @@
# limitations under the License.
# ============================================================================
"""FakeQuantWithMinMax op"""
"""FakeQuantPerLayer op"""
from functools import reduce as functools_reduce
import te.lang.cce
from te import tvm
......@@ -23,20 +22,16 @@ from topi import generic
from topi.cce import util
from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
fake_quant_op_info = TBERegOp("FakeQuantWithMinMax") \
fake_quant_per_layer_op_info = TBERegOp("FakeQuantPerLayer") \
.fusion_type("ELEMWISE") \
.async_flag(False) \
.binfile_name("fake_quant_with_min_max_vars_ema.so") \
.binfile_name("fake_quant_per_layer.so") \
.compute_cost(10) \
.kernel_name("fake_quant_with_min_max_vars_ema") \
.kernel_name("fake_quant_per_layer") \
.partial_flag(True) \
.attr("ema", "optional", "bool", "all") \
.attr("ema_decay", "optional", "float", "all") \
.attr("symmetric", "optional", "bool", "all") \
.attr("narrow_range", "optional", "bool", "all") \
.attr("training", "optional", "bool", "all") \
.attr("num_bits", "optional", "int", "all") \
.attr("quant_delay", "optional", "int", "all") \
.input(0, "x", None, "required", None) \
.input(1, "min", None, "required", None) \
.input(2, "max", None, "required", None) \
......@@ -48,16 +43,16 @@ fake_quant_op_info = TBERegOp("FakeQuantWithMinMax") \
.get_op_info()
@op_info_register(fake_quant_op_info)
def _fake_quant_tbe():
"""FakeQuantWithMinMax TBE register"""
@op_info_register(fake_quant_per_layer_op_info)
def _fake_quant_per_layer_tbe():
"""FakeQuantPerLayer TBE register"""
return
@fusion_manager.register("fake_quant_with_min_max_vars_ema")
def fake_quant_with_min_max_vars_ema_compute(x, min_val, max_val, y, quant_min, quant_max,
kernel_name="correction_mul"):
"""FakeQuantWithMinMax"""
@fusion_manager.register("fake_quant_per_layer")
def fake_quant_per_layer_compute(x, min_val, max_val, y, quant_min, quant_max,
kernel_name="fake_quant_per_layer"):
"""FakeQuantPerLayer"""
shape = te.lang.cce.util.shape_to_list(x.shape)
shape_min = te.lang.cce.util.shape_to_list(min_val.shape)
quant_min = te.lang.cce.broadcast(quant_min, shape_min, x.dtype)
......@@ -66,10 +61,13 @@ def fake_quant_with_min_max_vars_ema_compute(x, min_val, max_val, y, quant_min,
max_val = te.lang.cce.broadcast(max_val, shape_min, x.dtype)
# CalNudge(NudgeMinMax)
scale = te.lang.cce.vdiv(te.lang.cce.vsub(max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
scale = te.lang.cce.vdiv(te.lang.cce.vsub(
max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
zp_from_min = te.lang.cce.vsub(quant_min, te.lang.cce.vdiv(min_val, scale))
# Nudge zero point
nudge_zp = te.lang.cce.round(te.lang.cce.vmin(quant_max, te.lang.cce.vmax(quant_min, zp_from_min)))
nudge_zp_ = te.lang.cce.vmin(
quant_max, te.lang.cce.vmax(quant_min, zp_from_min))
nudge_zp = te.lang.cce.floor(te.lang.cce.vadds(nudge_zp_, 0.5))
nudge_min = te.lang.cce.vmul(te.lang.cce.vsub(quant_min, nudge_zp), scale)
nudge_max = te.lang.cce.vmul(te.lang.cce.vsub(quant_max, nudge_zp), scale)
......@@ -80,17 +78,19 @@ def fake_quant_with_min_max_vars_ema_compute(x, min_val, max_val, y, quant_min,
# FakeQuant
input_x = te.lang.cce.vmin(nudge_max, te.lang.cce.vmax(nudge_min, x))
nudge_input = te.lang.cce.round(te.lang.cce.vdiv(te.lang.cce.vsub(input_x, nudge_min), scale))
nudge_input_ = te.lang.cce.vdiv(
te.lang.cce.vsub(input_x, nudge_min), scale)
nudge_input = te.lang.cce.floor(te.lang.cce.vadds(nudge_input_, 0.5))
res = te.lang.cce.vadd(te.lang.cce.vmul(nudge_input, scale), nudge_min)
return res
@util.check_input_type(dict, dict, dict, dict, bool, float, bool, bool, bool, int, int, str)
def fake_quant_with_min_max_vars_ema(x, min_val, max_val, y,
ema, ema_decay, symmetric, narrow_range, training, num_bits, quant_delay,
kernel_name="fake_quant"):
"""FakeQuantWithMinMax"""
@util.check_input_type(dict, dict, dict, dict, bool, bool, int, str)
def fake_quant_per_layer(x, min_val, max_val, y,
symmetric, narrow_range, num_bits,
kernel_name="fake_quant_per_layer"):
"""FakeQuantPerLayer"""
input_shape = x.get("shape")
input_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
......@@ -131,8 +131,8 @@ def fake_quant_with_min_max_vars_ema(x, min_val, max_val, y,
input_data = tvm.placeholder(input_shape, name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_min, name="min_data", dtype=min_dtype)
max_data = tvm.placeholder(shape_min, name="max_data", dtype=max_dtype)
res = fake_quant_with_min_max_vars_ema_compute(input_data, min_data, max_data, y,
quant_min, quant_max, kernel_name)
res = fake_quant_per_layer_compute(input_data, min_data, max_data, y,
quant_min, quant_max, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res)
......
mindspore/ops/_op_impl/_custom_op/fake_quant_with_min_max_grad.py → mindspore/ops/_op_impl/_custom_op/fake_quant_perlayer_grad.py
浏览文件 @ e0fa277a
......@@ -13,7 +13,7 @@
# limitations under the License.
# ============================================================================
"""FakeQuantWithMinMaxGrad op"""
"""FakeQuantPerLayerGrad op"""
from functools import reduce as functools_reduce
import te.lang.cce
......@@ -26,15 +26,14 @@ from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
SHAPE_SIZE_LIMIT = 2147483648
D_TYPE = 'float32'
fake_quant_grad_op_info = TBERegOp("FakeQuantWithMinMaxGrad") \
fake_quant_per_layer_grad_op_info = TBERegOp("FakeQuantPerLayerGrad") \
.fusion_type("OPAQUE") \
.async_flag(False) \
.binfile_name("fake_quant_with_min_max_grad.so") \
.binfile_name("fake_quant_per_layer_grad.so") \
.compute_cost(10) \
.kernel_name("fake_quant_with_min_max_grad") \
.kernel_name("fake_quant_per_layer_grad") \
.partial_flag(True) \
.attr("num_bits", "optional", "int", "all") \
.attr("quant_delay", "optional", "int", "all") \
.attr("symmetric", "optional", "bool", "all") \
.attr("narrow_range", "optional", "bool", "all") \
.input(0, "dout", None, "required", None) \
......@@ -57,7 +56,8 @@ def _less_compare_float32(data_x, data_y):
min_value = tvm.const(2 ** (-126), dtype=D_TYPE)
max_value = tvm.const(2 ** 62, dtype=D_TYPE)
factor_value = tvm.const(2 ** 2, dtype=D_TYPE)
data_zero = te.lang.cce.broadcast(tvm.const(0, dtype=D_TYPE), shape_inputs, D_TYPE)
data_zero = te.lang.cce.broadcast(
tvm.const(0, dtype=D_TYPE), shape_inputs, D_TYPE)
min_value_tensor = te.lang.cce.vadds(data_zero, min_value)
res_sub = te.lang.cce.vsub(data_y, data_x)
......@@ -71,16 +71,16 @@ def _less_compare_float32(data_x, data_y):
return res
@op_info_register(fake_quant_grad_op_info)
def _fake_quant_grad_tbe():
"""FakeQuantWithMinMaxGrad TBE register"""
@op_info_register(fake_quant_per_layer_grad_op_info)
def _fake_quant_per_layer_grad_tbe():
"""FakeQuantPerLayerGrad TBE register"""
return
@fusion_manager.register("fake_quant_with_min_max_grad")
def fake_quant_with_min_max_grad_compute(dout, x, min_val, max_val, quant_min, quant_max,
kernel_name="fake_quant_with_min_max_grad"):
"""FakeQuantWithMinMaxGrad"""
@fusion_manager.register("fake_quant_per_layer_grad")
def fake_quant_per_layer_grad_compute(dout, x, min_val, max_val, quant_min, quant_max,
kernel_name="fake_quant_per_layer_grad"):
"""FakeQuantPerLayerGrad"""
shape = te.lang.cce.util.shape_to_list(x.shape)
shape_min = te.lang.cce.util.shape_to_list(min_val.shape)
quant_min = tvm.const(quant_min, x.dtype)
......@@ -89,10 +89,13 @@ def fake_quant_with_min_max_grad_compute(dout, x, min_val, max_val, quant_min, q
quant_max = te.lang.cce.broadcast(quant_max, shape_min)
# CalNudge(NudgeMinMax)
scale = te.lang.cce.vdiv(te.lang.cce.vsub(max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
scale = te.lang.cce.vdiv(te.lang.cce.vsub(
max_val, min_val), te.lang.cce.vsub(quant_max, quant_min))
zp_from_min = te.lang.cce.vsub(quant_min, te.lang.cce.vdiv(min_val, scale))
# Nudge zero point
nudge_zp = te.lang.cce.round(te.lang.cce.vmin(quant_max, te.lang.cce.vmax(quant_min, zp_from_min)))
nudge_zp_ = te.lang.cce.vmin(
quant_max, te.lang.cce.vmax(quant_min, zp_from_min))
nudge_zp = te.lang.cce.floor(te.lang.cce.vadds(nudge_zp_, 0.5))
nudge_min = te.lang.cce.vmul(te.lang.cce.vsub(quant_min, nudge_zp), scale)
nudge_max = te.lang.cce.vmul(te.lang.cce.vsub(quant_max, nudge_zp), scale)
nudge_min = te.lang.cce.broadcast(nudge_min, shape)
......@@ -106,11 +109,11 @@ def fake_quant_with_min_max_grad_compute(dout, x, min_val, max_val, quant_min, q
return res
@util.check_input_type(dict, dict, dict, dict, dict, int, int, bool, bool, str)
def fake_quant_with_min_max_grad(dout, x, min_val, max_val, dx,
num_bits, quant_delay, symmetric, narrow_range,
kernel_name="fake_quant_with_min_max_grad"):
"""FakeQuantWithMinMaxGrad"""
@util.check_input_type(dict, dict, dict, dict, dict, int, bool, bool, str)
def fake_quant_per_layer_grad(dout, x, min_val, max_val, dx,
num_bits, symmetric, narrow_range,
kernel_name="fake_quant_per_layer_grad"):
"""FakeQuantPerLayerGrad"""
input_shape = x.get("shape")
input_dtype = x.get("dtype")
min_shape = min_val.get("ori_shape")
......@@ -152,8 +155,8 @@ def fake_quant_with_min_max_grad(dout, x, min_val, max_val, dx,
input_data = tvm.placeholder(input_shape, name="x", dtype=x_dtype)
min_data = tvm.placeholder(shape_min, name="min_data", dtype=min_dtype)
max_data = tvm.placeholder(shape_min, name="max_data", dtype=max_dtype)
res = fake_quant_with_min_max_grad_compute(dout_data, input_data, min_data, max_data, quant_min,
quant_max, kernel_name)
res = fake_quant_per_layer_grad_compute(dout_data, input_data, min_data, max_data, quant_min,
quant_max, kernel_name)
with tvm.target.cce():
sch = generic.auto_schedule(res)
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或