def verify_squeeze(src_shape, axis):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.squeeze(A, axis=axis)
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="squeeze")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.squeeze(data_npy, axis=axis)
data_nd = tvm.nd.array(data_npy, ctx)
if out_npy.shape == ():
out_nd_shape = (1, )
else:
out_nd_shape = out_npy.shape
out_nd = tvm.nd.empty(out_nd_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")
def verify_squeeze(src_shape, axis):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.squeeze(A, axis=axis)
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="squeeze")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.squeeze(data_npy, axis=axis)
data_nd = tvm.nd.array(data_npy, ctx)
if out_npy.shape == ():
out_nd_shape = (1,)
else:
out_nd_shape = out_npy.shape
out_nd = tvm.nd.empty(out_nd_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"]:
check_device(device)
def Mean(device="llvm",
lib_path="./",
ndim=None,
dtype=None,
axis=None,
keep_dims=None):
'''
mean
Args:
device:
lib_path:
ndim:
dtype:
axis:
keepDims:
Returns:
'''
if axis[-1] >= ndim:
return
shape = [tvm.var("n" + str(i)) for i in range(ndim)]
axis_str = ""
for dim in axis:
axis_str += str(dim)
opname = "Mean_ndim%d_%s_axis%s_%s" % (ndim, dtype, axis_str, "keepDims"
if keep_dims else "notkeepDims")
print(opname)
# define compute
in_tensor = tvm.placeholder(shape, dtype=dtype, name='in_tensor')
c_shape = shape[:]
reduced_num = 1
for dim in axis:
c_shape[dim] = 1
reduced_num *= shape[dim]
def _ComputeSum(*b_idx):
reduce_axis = [tvm.reduce_axis((0, shape[dim])) for dim in axis]
a_idx = list(b_idx)
for i, dim in enumerate(axis):
a_idx[dim] = reduce_axis[i]
a_idx = tuple(a_idx)
return tvm.sum(in_tensor[a_idx], axis=reduce_axis)
out_tensor = tvm.compute(c_shape, _ComputeSum)
out_tensor = tvm.compute(c_shape, lambda *i: out_tensor(*i) / reduced_num)
if not keep_dims:
out_tensor = topi.squeeze(out_tensor, axis)
# define schedule & generate lib
tensor_list = [in_tensor, out_tensor]
s = tvm.create_schedule(out_tensor.op)
Genlib(s, tensor_list, device, opname, lib_path)
def test_squeeze():
verify_squeeze((1, 2, 3, 4), 0)
verify_squeeze((1, 2, 1, 4), None)
verify_squeeze((1, 1, 1, 4), (1, 2))
verify_squeeze((1, 1, 1, 1), None)
# a special case to trigger inline let expression
A = tvm.placeholder((2, ), 'float32', 'A')
E = topi.squeeze(A)
C = tvm.compute((1, ), lambda i: E[(2 * A[0] - 1).astype('int32')])
for device in ['cuda', 'opencl']:
ctx = tvm.context(device, 0)
if ctx.exist:
with tvm.target.create(device):
s = topi.generic.schedule_injective(C)
func = tvm.build(s, [A, C])
a = tvm.nd.array(np.array((1, 2)).astype('float32'), ctx=ctx)
c = tvm.nd.empty((1, ), dtype='float32', ctx=ctx)
func(a, c)
assert c.asnumpy()[0] == 2
def test_squeeze():
verify_squeeze((1, 2, 3, 4), 0)
verify_squeeze((1, 2, 1, 4), None)
verify_squeeze((1, 1, 1, 4), (1, 2))
verify_squeeze((1, 1, 1, 1), None)
# a special case to trigger inline let expression
A = tvm.placeholder((2,), 'float32', 'A')
E = topi.squeeze(A)
C = tvm.compute((1,), lambda i: E[(2 * A[0] - 1).astype('int32')])
for device in ['cuda', 'opencl']:
ctx = tvm.context(device, 0)
if ctx.exist:
with tvm.target.create(device):
s = topi.generic.schedule_injective(C)
func = tvm.build(s, [A, C])
a = tvm.nd.array(np.array((1, 2)).astype('float32'), ctx=ctx)
c = tvm.nd.empty((1,), dtype='float32', ctx=ctx)
func(a, c)
assert c.asnumpy()[0] == 2
def verify_squeeze(src_shape, axis):
A = te.placeholder(shape=src_shape, name="A")
B = topi.squeeze(A, axis=axis)
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="squeeze")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.squeeze(data_npy, axis=axis)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd_shape = out_npy.shape
out_nd = tvm.nd.empty(out_nd_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_squeeze(src_shape, axis):
A = tvm.placeholder(shape=src_shape, name="A")
B = topi.squeeze(A, axis=axis)
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="squeeze")
data_npy = np.random.normal(size=src_shape).astype(A.dtype)
out_npy = np.squeeze(data_npy, axis=axis)
data_nd = tvm.nd.array(data_npy, ctx)
out_nd_shape = out_npy.shape
out_nd = tvm.nd.empty(out_nd_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 compute_squeeze(attrs, inputs, out_info):
"""Compute definition of reshape"""
axis = attrs.get_int_tuple("axis")
axis = tuple(axis) if axis else None
return topi.squeeze(inputs[0], axis)