def _meshgrid(height, width):
y_zero = tvm.compute((height, ),
lambda i: -1 + i * 2.0 / (height - 1),
name='y0')
x_zero = tvm.compute((width, ),
lambda i: -1 + i * 2.0 / (width - 1),
name='x0')
y_temp = tvm.compute((height * width, ),
lambda i: y_zero[i // width],
name='y')
x_temp = tvm.compute((height * width, ),
lambda i: x_zero[i % width],
name='x')
y_temp = topi.reshape(y_temp, (1, height * width))
x_temp = topi.reshape(x_temp, (1, height * width))
ones = tvm.compute((1, height * width), lambda i, j: 1, name='ones')
grid = tvm.compute(
(3, height * width),
lambda i, j: 0.5 * (i - 1) * (i - 2) * x_temp[0, j] + i *
(2 - i) * y_temp[0, j] + 0.5 * i * (i - 1) * ones[0, j],
name='grid')
# grid = topi.concatenate((x,y,ones),0) #can not use topi.concatenate
return grid
def _transform(theta, input_dim, out_size, input_shape, dtype):
num_batch = input_shape[0]
height = input_shape[1]
width = input_shape[2]
num_channels = input_shape[3]
theta = topi.reshape(theta, (num_batch, 2, 3))
theta = topi.cast(theta, dtype)
out_height = out_size[0]
out_width = out_size[1]
grid = _meshgrid(out_height, out_width)
grid = topi.reshape(grid, (num_batch, 3, out_height*out_width))
grid = topi.cast(grid, dtype=dtype)
k = tvm.reduce_axis((0, 3), 'k')
T_g = tvm.compute((num_batch, 2, out_height*out_width),lambda b, y, x: tvm.sum(theta[b, y, k] * grid[b, k, x], axis = k), name = 'T_g')
x_s = tvm.compute((num_batch, 1, out_height*out_width), lambda i,j,k:T_g[i,0,k], name = 'x_s')
y_s = tvm.compute((num_batch, 1, out_height*out_width), lambda i,j,k:T_g[i,1,k], name = 'y_s')
x_s_flat = topi.reshape(x_s, (num_batch*out_height*out_width,))
y_s_flat = topi.reshape(y_s, (num_batch*out_height*out_width,))
input_transformed = _interpolate(input_dim, input_shape, x_s_flat, y_s_flat, out_size, dtype)
output = topi.reshape(input_transformed, [num_batch, out_height, out_width, num_channels])
return output
def forward_(x_op, name):
# this is 2d only specialized implementation of topi.nn.softmax
if x_op.ndim == 1:
x = topi.reshape(x_op.tvm_tensor, (1, x_op.size))
m = 1
n = x_op.shape[0]
elif x_op.ndim == 2:
x = x_op.tvm_tensor
m, n = x_op.shape
else:
raise ValueError(f'Given ndim {x_op.ndim} is not supported')
k = tvm.reduce_axis((0, n), name='k')
max_elem = tvm.compute((m, ),
lambda i: tvm.max(x[i, k], axis=k),
name=f'{name}:max_elem')
k = tvm.reduce_axis((0, n), name='k')
expsum = tvm.compute(
(m, ),
lambda i: tvm.sum(tvm.exp(x[i, k] - max_elem[i]), axis=k),
name=f'{name}:expsum')
softmax = tvm.compute(
x.shape,
lambda i, j: tvm.exp(x[i, j] - max_elem[i]) / expsum[i],
name=f'{name}:softmax')
if x_op.ndim == 1:
softmax = topi.reshape(softmax, x_op.shape)
return softmax, max_elem, expsum
def reshape(tensor, shape, sph=None, dst_scope='buffer0'):
res = topi.reshape(tensor, shape)
MarkScope(res, dst_scope)
PragmaCopy(res)
return res
def verify_reshape(src_shape, dst_shape):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.reshape(A, dst_shape)
def check_device(device):
ctx = tvm.context(device, 0)
if not ctx.exist:
print("Skip because %s is not enabled" % device)
return
print("Running on target: %s" % device)
with tvm.target.create(device):
s = topi.generic.schedule_injective(B)
foo = tvm.build(s, [A, B], device, name="reshape")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.reshape(data_npy, newshape=dst_shape)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd = tvm.nd.empty(dst_shape, ctx=ctx, dtype=B.dtype)
foo(data_nd, out_nd)
np.testing.assert_allclose(out_nd.asnumpy(), out_npy)
for device in [
"llvm", "nvptx", "cuda", "opencl", "metal", "rocm", "vulkan",
"sdaccel"
]:
check_device(device)
def FullConnection(device="llvm",
lib_path="./",
ndim_a=None,
dtype=None,
has_bias=None):
'''
full connection
Args:
device:
lib_path:
ndim_a:
dtype:
hasBias:
Returns:
'''
n_dim, ci, h_dim, kernel_tensor = (tvm.var("n_dim"), tvm.var("out_tensor"), tvm.var("h_dim"), \
tvm.var("kernel_tensor"))
co = tvm.var("co")
if ndim_a == 4:
shape_a = (n_dim, ci, h_dim, kernel_tensor)
chw = ci * h_dim * kernel_tensor
else:
shape_a = (n_dim, ci)
chw = ci
shape_w = (co, chw)
opname = "FullConnection_ndimA%d_%s_%s" % (ndim_a, dtype, "hasBias"
if has_bias else "notHasBias")
is_var = True
vh, vw, vc = 1, 1, 1
print(opname)
in_tensor = tvm.placeholder(shape_a, dtype=dtype, name='in_tensor')
kernel_tensor = tvm.placeholder(shape_w, dtype=dtype, name='kernel_tensor')
input_tensor = topi.reshape(in_tensor,
(n_dim,
chw)) if len(shape_a) == 4 else in_tensor
out_tensor = _matmul_spatial_pack_asm((is_var, 0, ci, vh, vw, vc), input_tensor, kernel_tensor, \
layout='NC', out_dtype=dtype)
if has_bias:
bias = tvm.placeholder((co, ), dtype=dtype, name='bias')
out_tensor = tvm.compute((n_dim, co),
lambda n, co: out_tensor[n, co] + bias[co],
tag='injective')
tensor_list = [in_tensor, kernel_tensor, bias, out_tensor
] if has_bias else [in_tensor, kernel_tensor, out_tensor]
cfg = {
'is_var': is_var,
'is_transpose': 0,
'core_id': 0,
'CI': ci,
'VH': vh,
'VW': vw,
'VC': vc
}
s = _matmul_schedule_asm(cfg, [out_tensor])
Genlib(s, tensor_list, device, opname, lib_path)
def __call__(self, inputs):
outputs = topi.nn.conv2d(inputs, self.weight, self.stride,
self.padding, self.dilation)
if self.bias: # TODO: check bias shape
reshaped_bias = topi.reshape(
self.bias, (self.in_channels, self.out_channels, 1, 1))
outputs += reshaped_bias
return outputs
def algorithm_forward(self):
assert self.x.size == self.t.size, \
"only supports one-hot vector"
self.softmax, self.max_elem, self.expsum = Softmax.forward_(
self.x, name=f'{self.label}')
if self.x.ndim == 1:
y = topi.reshape(self.softmax, (1, self.x.size))
t = topi.reshape(self.t.tvm_tensor, (1, self.t.size))
m = 1
n = self.x.shape[0]
elif self.x.ndim == 2:
y = self.softmax
t = self.t.tvm_tensor
m, n = self.x.shape
else:
raise NotImplementedError
self.ty = tvm.compute((m, n),
lambda i, j: tvm.log(y[i, j]) * t[i, j],
name=f'{self.label}:ty')
k = tvm.reduce_axis((0, n), name='k')
self.sum_ty = tvm.compute((m, ),
lambda i: tvm.sum(self.ty[i, k], axis=k),
name=f'{self.label}:sum_ty')
# TODO: need to validate the shape and keepdims
# self.shape would be like (1,1,1), which size is 1
expected_size = 1
assert self.size == expected_size, \
f'size of SoftmaxWithCrossEntropyLoss must be {expected_size}, not {self.size}'
k = tvm.reduce_axis((0, m), name='k')
self.total = tvm.compute(self.shape,
lambda *idxs: tvm.sum(self.sum_ty[k], axis=k),
name=f'{self.label}:total')
self.tvm_tensor = tvm.compute(self.shape,
lambda *idxs: -self.total[idxs] / m,
name=f'{self.label}:tensor')
def _declaration_conv_NCHWc_int8(wkl, sch, data, kernel):
""" Declaration for int8 conv"""
out_dtype = wkl.out_dtype
HPAD, WPAD = wkl.hpad, wkl.wpad
HSTR, WSTR = wkl.hstride, wkl.wstride
batch_size = data.shape[0]
out_height = (wkl.height + 2 * HPAD - wkl.hkernel) // HSTR + 1
out_width = (wkl.width + 2 * WPAD - wkl.wkernel) // WSTR + 1
DOPAD = (HPAD != 0 or WPAD != 0)
if DOPAD:
data_pad = pad(data, (0, 0, HPAD, WPAD, 0), name="data_pad")
else:
data_pad = data
oshape = (batch_size, wkl.out_filter // sch.oc_bn, out_height, out_width,
sch.oc_bn)
# Intel performs dot product of 2 "4" Int8 values
n_elems = 4
assert sch.ic_bn % n_elems == 0
ic_outer = tvm.reduce_axis((0, wkl.in_filter // (sch.ic_bn)),
name='ic_outer')
ic_f_inner = tvm.reduce_axis((0, sch.ic_bn // n_elems), name='ic_f_inner')
ic_s_inner = tvm.reduce_axis((0, n_elems), name='ic_s_inner')
# Reshaping kernel as the last 2 dimensions are 1x1 (k_h x k_w)
k_shape = kernel.shape
kernel = topi.reshape(kernel,
(k_shape[0], k_shape[1], k_shape[2], k_shape[3],
k_shape[4] * k_shape[5] * k_shape[6]))
conv = tvm.compute(
oshape,
lambda n, oc_chunk, oh, ow, oc_block: tvm.sum(
data_pad[n, ic_outer, oh * HSTR, ow * WSTR, ic_f_inner * n_elems +
ic_s_inner].astype(out_dtype) * kernel[
oc_chunk, ic_outer, ic_f_inner, oc_block, ic_s_inner].
astype(out_dtype),
axis=[ic_outer, ic_f_inner, ic_s_inner]),
name='conv2d_NCHWc_int8',
tag="conv2d_NCHWc_int8")
return conv
def verify_reshape(src_shape, dst_shape):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.reshape(A, dst_shape)
def check_device(device):
ctx = tvm.context(device, 0)
if not ctx.exist:
print("Skip because %s is not enabled" % device)
return
print("Running on target: %s" % device)
with tvm.target.create(device):
s = topi.generic.schedule_injective(B)
foo = tvm.build(s, [A, B], device, name="reshape")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.reshape(data_npy, newshape=dst_shape)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd = tvm.nd.empty(dst_shape, ctx=ctx, dtype=B.dtype)
foo(data_nd, out_nd)
tvm.testing.assert_allclose(out_nd.asnumpy(), out_npy)
for device in get_all_backend():
check_device(device)
def verify_reshape(src_shape, dst_shape):
A = te.placeholder(shape=src_shape, name="A")
B = topi.reshape(A, dst_shape)
def check_device(device):
ctx = tvm.context(device, 0)
if not ctx.exist:
print("Skip because %s is not enabled" % device)
return
print("Running on target: %s" % device)
with tvm.target.create(device):
s = topi.testing.get_injective_schedule(device)(B)
foo = tvm.build(s, [A, B], device, name="reshape")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.reshape(data_npy, newshape=dst_shape)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd = tvm.nd.empty(dst_shape, ctx=ctx, dtype=B.dtype)
foo(data_nd, out_nd)
tvm.testing.assert_allclose(out_nd.asnumpy(), out_npy)
for device in get_all_backend():
check_device(device)
def verify_reshape(src_shape, dst_shape):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.reshape(A, dst_shape)
s = topi.cuda.schedule_injective(B)
def check_device(device):
if not tvm.module.enabled(device):
print("Skip because %s is not enabled" % device)
return
ctx = tvm.gpu(0) if device == "cuda" else tvm.cl(0)
foo = tvm.build(s, [A, B], device, name="reshape")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.reshape(data_npy, newshape=dst_shape)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd = tvm.nd.empty(dst_shape, ctx=ctx, dtype=B.dtype)
foo(data_nd, out_nd)
np.testing.assert_allclose(out_nd.asnumpy(), out_npy)
check_device("cuda")
check_device("opencl")
check_device("metal")
inits = [(np.zeros, 'shape'), (np.zeros, 'shape'), (np.zeros, 'shape'),
(np.zeros, 'shape'), (np.zeros, 'shape'), (np.ones, 'shape'),
(np.zeros, 'shape'), (np.zeros, 'shape'), (np.random.normal, 'size'),
(np.random.normal, 'size')]
# Graph input
x = tvm.placeholder((batch_size, num_timesteps * num_input), 'float32')
y = tvm.placeholder((batch_size, num_classes), 'float32')
s = tvm.placeholder((batch_size, num_hidden), 'float32')
h = tvm.placeholder((batch_size, num_hidden), 'float32')
# Tensors and vars for training graph
weights = [tvm.placeholder(x, 'float32') for x in sizes]
#Construct model
xs = topi.split(topi.reshape(x, (batch_size, num_timesteps, num_input)),
num_timesteps,
axis=1)
xs = [topi.reshape(x, (batch_size, num_input)) for x in xs]
new_s = s
new_h = h
for i in range(num_timesteps):
inp = topi.concatenate([xs[i], new_h], 1)
g = topi.tanh(topi.matmul(inp, weights[0]) + weights[1])
j = topi.sigmoid(topi.matmul(inp, weights[2]) + weights[3])
f = topi.sigmoid(topi.matmul(inp, weights[4]) + weights[5])
o = topi.sigmoid(topi.matmul(inp, weights[6]) + weights[7])
new_s = new_s * f + g * j
new_h = topi.tanh(new_s) * o
def compute_reshape_like(attrs, inputs, out_info):
"""Compute definition of reshape_like"""
return topi.reshape(inputs[0], inputs[1].shape)
def ConvVar(device="llvm", lib_path="./", optype=None,\
ndim=None, layout=None, dtype=None, kernels=None,\
strides=None, pad=None, dilations=None,\
hasbias=None, activation_type=None,\
config_entity=None, impl_dtype=None, channel_multiplier=None,\
use_arm32=False, cfg=None):
'''
convolution
Args:
device:
lib_path:
optype:
ndim:
layout:
dtype:
kernels:
strides:
pad:
dilations:
hasbias:
activationType:
configEntity:
impl_dtype:
channel_multiplier:
use_arm32:
cfg:
Returns:
'''
use_depthwise = optype == 'ConvolutionDepthwise'
use_deconv = optype == 'Deconvolution'
use_deconv_depthwise = optype == 'DeConvolutionDepthwise'
has_bias = hasbias
ow = 1 if cfg is None else cfg['VW']
oh = 1 if cfg is None else cfg['VH']
oc = 1 if cfg is None else cfg['VC']
kh, kw = kernels
op_name = "%s_ndim%d_%s_k%d_s%d_p%d%d%d%d_d%d_act%d_vc%d_vh%d_vw%d_hasbias%d" % ( \
map_conv[optype], ndim, dtype, \
kh, strides[0], pad[0], pad[1], pad[2], pad[3], dilations[0], \
activation_enum_map[activation_type], oc, oh, ow, hasbias)
batch = tvm.var("batch")
in_channel = tvm.var("in_channel")
in_height, in_width = tvm.var("in_height"), tvm.var("in_width")
pad_up, pad_down, pad_left, pad_right = pad
opname = op_name
print("Conv", opname, config_entity)
if impl_dtype is None:
impl_dtype = dtype
if use_depthwise:
multiplier = channel_multiplier
out_channel = in_channel * multiplier
elif use_deconv_depthwise:
multiplier = channel_multiplier
out_channel = in_channel * multiplier
else:
out_channel = tvm.var("out_channel")
# define placeholder
input_tensor = in_tensor = tvm.placeholder(
(batch, in_channel, in_height, in_width),
dtype=dtype,
name='in_tensor')
if use_depthwise:
temp_tensor = kernel_tensor = tvm.placeholder((in_channel, multiplier, kh, kw), dtype=dtype,\
name='kernel_tensor')
elif use_deconv:
temp_tensor = kernel_tensor = tvm.placeholder((in_channel, out_channel, kh, kw), dtype=dtype,\
name='kernel_tensor')
elif use_deconv_depthwise:
temp_tensor = kernel_tensor = tvm.placeholder((in_channel, multiplier, kh, kw), dtype=dtype,\
name='kernel_tensor')
else:
temp_tensor = kernel_tensor = tvm.placeholder((out_channel, in_channel, kh, kw), dtype=dtype,\
name='kernel_tensor')
if has_bias:
bias = tvm.placeholder((out_channel, ), dtype=dtype, name='bias')
bias1 = topi.reshape(bias, (out_channel, 1, 1))
if impl_dtype != dtype:
input_tensor = AsType(input_tensor, impl_dtype)
temp_tensor = AsType(temp_tensor, impl_dtype)
if has_bias:
bias1 = AsType(bias1, impl_dtype)
# define compute & schedule
if pad_up != pad_down or pad_left != pad_right:
input_tensor = topi.nn.pad(input_tensor, [0, 0, pad_up, pad_left],
[0, 0, pad_down, pad_right],
name='data_pad')
padding = 0, 0
else:
padding = pad_up, pad_left
if use_depthwise:
cfg1 = (True, 1, 1,
1) if cfg is None else (True, cfg["tile_oh"], cfg["tile_ow"],
cfg["tile_co"])
out_tensor = _depthwise_spatial_pack(cfg1, input_tensor, temp_tensor, strides, padding, dilations,\
out_dtype=impl_dtype)
elif use_deconv:
def GetInput(input_tensor, temp_tensor, padding):
_, out_c, filter_h, filter_w = temp_tensor.shape
if out_c is None:
print("temp_tensor.shape err")
stride_h, stride_w = strides
# dilate stage
dilated_input = topi.nn.dilate(input_tensor,
[1, 1, stride_h, stride_w],
name='DilatedInput')
# padding stage
fpad_top, fpad_left, fpad_bottom, fpad_right = topi.nn.get_pad_tuple(
padding, (filter_h, filter_w))
bpad_top = filter_h - 1 - fpad_top
bpad_bottom = filter_h - 1 - fpad_bottom
bpad_left = filter_w - 1 - fpad_left
bpad_right = filter_w - 1 - fpad_right
padded_input = topi.nn.pad(dilated_input, \
[0, 0, bpad_top, bpad_left], \
[0, 0, bpad_bottom, bpad_right], \
name='PaddedInput')
return padded_input
special_deconv = kh == 2 and kw == 2 and strides[0] == 2 and strides[
1] == 2
# special_deconv = False
if special_deconv:
out_tensor = OptimalOut(input_tensor, temp_tensor, in_channel)
else:
out_tensor = BaseImplementation(input_tensor, temp_tensor,
GetInput, layout, padding)
elif use_deconv_depthwise:
def GetInput(input_tensor, temp_tensor, padding):
_, out_c, filter_h, filter_w = temp_tensor.shape
if out_c is None:
print("temp_tensor.shape err")
stride_h, stride_w = strides
# dilate stage
dilated_input = topi.nn.dilate(input_tensor,
[1, 1, stride_h, stride_w],
name='DilatedInput')
# padding stage
fpad_top, fpad_left, fpad_bottom, fpad_right = topi.nn.get_pad_tuple(
padding, (filter_h, filter_w))
bpad_top = filter_h - 1 - fpad_top
bpad_bottom = filter_h - 1 - fpad_bottom
bpad_left = filter_w - 1 - fpad_left
bpad_right = filter_w - 1 - fpad_right
padded_input = topi.nn.pad(dilated_input, \
[0, 0, bpad_top, bpad_left], \
[0, 0, bpad_bottom, bpad_right], \
name='PaddedInput')
return padded_input
temp_tensor = topi.flip(temp_tensor, axis=-1)
temp_tensor = topi.flip(temp_tensor, axis=-2)
out_tensor = topi.nn.depthwise_conv2d_nchw(GetInput(input_tensor, temp_tensor, padding), temp_tensor, (1, 1), \
padding, (1, 1), out_dtype=input_tensor.dtype)
else:
cfg1 = (True, 1, 1,
1) if cfg is None else (True, cfg["tile_oh"], cfg["tile_ow"],
cfg["tile_co"])
out_tensor = _conv_spatial_pack_asm(cfg1, input_tensor, temp_tensor, strides, padding, dilations,\
out_dtype=impl_dtype)
if has_bias:
out_tensor = tvm.compute(out_tensor.shape, lambda n, co, h, w: out_tensor[n, co, h, w] + bias1[co][0][0],\
tag="injective")
out_tensor = TopiActivation(out_tensor, activation_type)
if impl_dtype != dtype:
out_tensor = AsType(out_tensor, dtype)
# create schedule
if use_arm32:
s = tvm.create_schedule(out_tensor.op)
elif use_depthwise:
s = schedule_depthwise_conv2d_nchw_arm(cfg, [out_tensor])
elif use_deconv:
if special_deconv:
s = tvm.create_schedule([out_tensor.op])
else:
s = topi.generic.schedule_conv2d_nchw([out_tensor])
elif use_deconv_depthwise:
s = tvm.create_schedule([out_tensor.op])
else:
s = schedule_conv2d_nchw_arm_cpu([out_tensor])
# generate lib
attr = [
batch, in_channel, in_height, in_width, out_channel, in_tensor,
kernel_tensor
]
tensor_list = [*attr, bias, out_tensor
] if has_bias else [*attr, out_tensor]
Genlib(s, tensor_list, device, opname, lib_path)
def flatten_topi(inputs):
N, C, H, W = inputs.shape
return topi.reshape(inputs, [N, C * H * W])
packed_output_shape = (N, G, K // G, P, Q)
output_shape = (N, K, P, Q)
I = te.placeholder(input_shape, name="I")
W = te.placeholder(weight_shape, name="W")
### reductions
rc = te.reduce_axis((0, C // G), name='rc')
ry = te.reduce_axis((0, R), name='ry')
rx = te.reduce_axis((0, S), name='rx')
ig = C // G
og = K // G
### (K,C//G,R,S) to (G,K//G,C//G,R,S)
W_pack = topi.reshape(W, (packed_weight_shape))
O = te.compute(
packed_output_shape, lambda n, g, co, x, y: te.sum(I[n, rc + (
g * ig), x + rx, y + ry] * W_pack[g, co, rc, rx, ry],
axis=[rc, ry, rx]))
s = te.create_schedule(O.op)
s[W_pack].compute_inline()
ir = tvm.lower(s, [I, W, O])
print(ir)
### COMPILE AND RUN
tgt_host = "llvm"
tgt = "llvm"
def Deconv(device="llvm",
lib_path="./",
optype=None,
ndim=None,
dtype=None,
kernels=None,
strides=None,
pad=None,
dilations=None,
hasbias=None,
activation_type=None,
config_entity=None,
impl_dtype=None,
use_arm32=False,
cfg=None):
'''
Deconvolution
Args:
device:
lib_path:
optype:
ndim:
dtype:
kernels:
strides:
pad:
dilations:
hasbias:
activationType:
configEntity:
impl_dtype:
use_arm32:
cfg:
Returns:
'''
if cfg is None:
cfg = {
'CI': tvm.var('ci'),
'VH': 2,
'VW': 2,
'VC': 4,
'VI': 4,
'tile_oh': 2,
'tile_ow': 2,
'tile_co': 4,
'ann_reduce': ['none', 'none'],
"ann_spatial": ['none', 'none', 'none']
}
has_bias = hasbias
batch = tvm.var("batch")
in_channel = tvm.var("in_channel")
in_height, in_width = tvm.var("in_height"), tvm.var("in_width")
kh, kw = kernels
ow = cfg['VW']
oh = cfg['VH']
oc = cfg['VC']
op_name = "%s_ndim%d_%s_k%d_s%d_p%d%d%d%d_d%d_act%d_vc%d_vh%d_vw%d_hasbias%d" % (\
map_conv[optype], ndim, dtype,\
kh, strides[0], pad[0], pad[1], pad[2], pad[3], dilations[0],\
activation_enum_map[activation_type], oc, oh, ow, hasbias)
opname = op_name
print("DEconv", opname, config_entity)
if impl_dtype is None:
impl_dtype = dtype
out_channel = tvm.var("out_channel")
# define placeholder
input_tensor = in_tensor = tvm.placeholder((batch, in_channel, in_height, in_width, 4), \
dtype=dtype, name='in_tensor')
temp_tensor = kernel_tensor = tvm.placeholder((in_channel*4, out_channel, kh, kw), dtype=dtype, \
name='kernel_tensor')
if has_bias:
bias = tvm.placeholder((out_channel, ), dtype=dtype, name='bias')
bias1 = topi.reshape(bias, (out_channel, 1, 1))
if impl_dtype != dtype:
input_tensor = AsType(input_tensor, impl_dtype)
temp_tensor = AsType(temp_tensor, impl_dtype)
if has_bias:
bias1 = AsType(bias1, impl_dtype)
# define compute & schedule
cfg1 = (True, 1, 1, 1) if cfg is None else (True, cfg["tile_oh"],
cfg["tile_ow"], cfg["tile_co"])
out_tensor = _conv_spatial_pack_deconv(cfg1,
input_tensor,
temp_tensor,
out_dtype=impl_dtype)
if has_bias:
out_tensor = tvm.compute(out_tensor.shape, lambda n, co, h, w, c4: \
out_tensor[n, co, h, w, c4] + bias1[co*4 + c4][0][0], tag="injective")
out_tensor = TopiActivation(out_tensor, activation_type)
if impl_dtype != dtype:
out_tensor = AsType(out_tensor, dtype)
# create schedule
if use_arm32:
s = tvm.create_schedule(out_tensor.op)
else:
s = schedule_conv2d_nchw_arm_cpu_deconv(cfg, [out_tensor])
attr = [
batch, in_channel, in_height, in_width, out_channel, in_tensor,
kernel_tensor
]
if has_bias: attr.append(bias)
attr.append(out_tensor)
tensor_list = attr
Genlib(s, tensor_list, device, opname, lib_path)
def _interpolate(im, im_shape, x, y, out_size, dtype):
num_batch = im_shape[0]
height = im_shape[1]
width = im_shape[2]
channels = im_shape[3]
out_height = out_size[0]
out_width = out_size[1]
max_y = int(im_shape[1] - 1)
max_x = int(im_shape[2] - 1)
#[-1,1] -> [0, width-1]
x = topi.multiply(topi.add(x, tvm.const(1, dtype=dtype)), width / tvm.const(2, dtype=dtype))
y = topi.multiply(topi.add(y, tvm.const(1, dtype=dtype)), height / tvm.const(2, dtype=dtype))
# do sampling
dim3 = out_height * out_width * num_batch
x0 = topi.cast(topi.floor(x), 'int32')
y0 = topi.cast(topi.floor(y), 'int32')
x1 = topi.add(x0,tvm.const(1, dtype="int32"))
y1 = topi.add(y0,tvm.const(1, dtype="int32"))
x0 = topi.clip(x0, 0, max_x)
x1 = topi.clip(x1, 0, max_x)
y0 = topi.clip(y0, 0, max_y)
y1 = topi.clip(y1, 0, max_y)
dim2 = width
dim1 = width * height
base = tvm.compute((dim3,),lambda i:(i // (out_height * out_width)) * width * height, name = 'base')
base_y0 = topi.add(base, topi.multiply(y0, dim2))
base_y1 = topi.add(base, topi.multiply(y1, dim2))
idx_a = topi.add(base_y0, x0)
idx_b = topi.add(base_y1, x0)
idx_c = topi.add(base_y0, x1)
idx_d = topi.add(base_y1, x1)
im_flat = topi.reshape(im, (num_batch * height * width, channels))
im_flat = topi.cast(im_flat, dtype)
Ia = tvm.compute((dim3, channels),lambda i,j: im_flat[idx_a[i], j], name = 'Ia')
Ib = tvm.compute((dim3, channels),lambda i,j: im_flat[idx_b[i], j], name = 'Ib')
Ic = tvm.compute((dim3, channels),lambda i,j: im_flat[idx_c[i], j], name = 'Ic')
Id = tvm.compute((dim3, channels),lambda i,j: im_flat[idx_d[i], j], name = 'Id')
x0_f = topi.cast(x0, dtype)
x1_f = topi.cast(x1, dtype)
y0_f = topi.cast(y0, dtype)
y1_f = topi.cast(y1, dtype)
wa = topi.expand_dims(topi.multiply(topi.subtract(x1_f, x), topi.subtract(y1_f, y)), 1)
wb = topi.expand_dims(topi.multiply(topi.subtract(x1_f, x), topi.subtract(y, y0_f)), 1)
wc = topi.expand_dims(topi.multiply(topi.subtract(x, x0_f), topi.subtract(y1_f, y)), 1)
wd = topi.expand_dims(topi.multiply(topi.subtract(x, x0_f), topi.subtract(y, y0_f)), 1)
output = topi.add(topi.add(topi.add(topi.multiply(wa, Ia), topi.multiply(wb, Ib)),topi.multiply(wc, Ic)), topi.multiply(wd, Id))
return output
def _interpolate(im, im_shape, x, y, out_size, dtype):
num_batch = im_shape[0]
height = im_shape[1]
width = im_shape[2]
channels = im_shape[3]
out_height = out_size[0]
out_width = out_size[1]
max_y = int(im_shape[1] - 1)
max_x = int(im_shape[2] - 1)
# [-1,1] -> [0, width-1]
x_temp = topi.multiply(topi.add(x, tvm.const(1, dtype=dtype)),
width / tvm.const(2, dtype=dtype))
y_temp = topi.multiply(topi.add(y, tvm.const(1, dtype=dtype)),
height / tvm.const(2, dtype=dtype))
# do sampling
dim3 = out_height * out_width * num_batch
x_zero = topi.cast(topi.floor(x_temp), 'int32')
y_zero = topi.cast(topi.floor(y_temp), 'int32')
x_one = topi.add(x_zero, tvm.const(1, dtype="int32"))
y_one = topi.add(y_zero, tvm.const(1, dtype="int32"))
x_zero = topi.clip(x_zero, 0, max_x)
x_one = topi.clip(x_one, 0, max_x)
y_zero = topi.clip(y_zero, 0, max_y)
y_one = topi.clip(y_one, 0, max_y)
dim2 = width
base = tvm.compute((dim3, ),
lambda i:
(i // (out_height * out_width)) * width * height,
name='base')
base_y0 = topi.add(base, topi.multiply(y_zero, dim2))
base_y1 = topi.add(base, topi.multiply(y_one, dim2))
idx_a = topi.add(base_y0, x_zero)
idx_b = topi.add(base_y1, x_zero)
idx_c = topi.add(base_y0, x_one)
idx_d = topi.add(base_y1, x_one)
im_flat = topi.reshape(im, (num_batch * height * width, channels))
im_flat = topi.cast(im_flat, dtype)
i_a = tvm.compute((dim3, channels),
lambda i, j: im_flat[idx_a[i], j],
name='Ia')
i_b = tvm.compute((dim3, channels),
lambda i, j: im_flat[idx_b[i], j],
name='Ib')
i_c = tvm.compute((dim3, channels),
lambda i, j: im_flat[idx_c[i], j],
name='Ic')
i_d = tvm.compute((dim3, channels),
lambda i, j: im_flat[idx_d[i], j],
name='Id')
x0_f = topi.cast(x_zero, dtype)
x1_f = topi.cast(x_one, dtype)
y0_f = topi.cast(y_zero, dtype)
y1_f = topi.cast(y_zero, dtype)
w_a = topi.expand_dims(
topi.multiply(topi.subtract(x1_f, x_temp),
topi.subtract(y1_f, y_temp)), 1)
w_b = topi.expand_dims(
topi.multiply(topi.subtract(x1_f, x_temp),
topi.subtract(y_temp, y0_f)), 1)
w_c = topi.expand_dims(
topi.multiply(topi.subtract(x_temp, x0_f),
topi.subtract(y1_f, y_temp)), 1)
w_d = topi.expand_dims(
topi.multiply(topi.subtract(x_temp, x0_f),
topi.subtract(y_temp, y0_f)), 1)
output = topi.add(
topi.add(
topi.add(topi.multiply(w_a, i_a), topi.multiply(w_b, i_b)),
topi.multiply(w_c, i_c)), topi.multiply(w_d, i_d))
return output
def compute_reshape_like(attrs, inputs, out_info):
"""Compute definition of reshape_like"""
return topi.reshape(inputs[0], inputs[1].shape)
def compute_reshape(attrs, inputs, out_info):
"""Compute definition of reshape"""
oshape = out_info[0].shape
return topi.reshape(inputs[0], oshape)
(np.zeros, 'shape'),
(np.ones, 'shape'),
(np.zeros, 'shape'),
(np.zeros, 'shape'),
(np.random.normal, 'size'),
(np.random.normal, 'size')
]
x = tvm.placeholder((batch_size, num_timesteps * num_input), 'float32')
y = tvm.placeholder((batch_size, num_classes), 'float32')
s = tvm.placeholder((batch_size, num_hidden), 'float32')
h = tvm.placeholder((batch_size, num_hidden), 'float32')
weights = [tvm.placeholder(x, 'float32', name="weights") for x in sizes]
xs = topi.split(topi.reshape(x, (batch_size, num_timesteps, num_input)), num_timesteps, axis=1)
xs = [topi.reshape(x, (batch_size, num_input)) for x in xs]
new_s = s
new_h = h
for i in range(num_timesteps):
inp = topi.concatenate([xs[i], new_h], 1)
g = topi.tanh(topi.matmul(inp, weights[0]) + weights[1])
j = topi.sigmoid(topi.matmul(inp, weights[2]) + weights[3])
f = topi.sigmoid(topi.matmul(inp, weights[4]) + weights[5])
o = topi.sigmoid(topi.matmul(inp, weights[6]) + weights[7])
new_s = new_s * f + g * j
new_h = topi.tanh(new_s) * o
logits = topi.matmul(new_h, weights[8]) + weights[9]