def pop_stmt(self):
"""Create a statement from the statements within current stage."""
stmts = self.stmt_stack.pop()
if not stmts or callable(stmts[-1]):
stmts.append(_stmt.Evaluate(0))
stmtwrap = lambda x: x[0] if len(x) == 1 else _stmt.SeqStmt(
list(reversed(x)))
ret_stmt = [stmts[-1]]
for s in reversed(stmts[:-1]):
if callable(s):
ret_stmt = [s(stmtwrap(ret_stmt))]
else:
assert isinstance(s, _stmt.Stmt)
ret_stmt.append(s)
return stmtwrap(ret_stmt)
def make_nop():
"""Returns a 'no operation' node in HalideIR."""
return _stmt.Evaluate(tvm.runtime.const(0, dtype='int32'))
def extern(shape,
inputs,
fcompute,
name="extern",
dtype=None,
in_buffers=None,
out_buffers=None,
tag="",
attrs=None):
"""Compute several tensor via extern function.
Parameters
----------
shape: tuple or list of tuples.
The shape of the outputs.
inputs: list of Tensor
The inputs
fcompute: lambda function of inputs, outputs-> stmt
Specifies the IR statement to do the computation.
See the following note for function signature of fcompute
.. note::
**Parameters**
- **ins** (list of :any:`Buffer`) - Placeholder for each inputs
- **outs** (list of :any:`Buffer`) - Placeholder for each outputs
**Returns**
- **stmt** (:any:`Stmt`) - The statement that carries out array computation.
name: str, optional
The name hint of the tensor
dtype: str or list of str, optional
The data types of outputs,
by default dtype will be same as inputs.
in_buffers: Buffer or list of Buffer, optional
Input buffers.
out_buffers: Buffer or list of Buffers, optional
Output buffers.
tag: str, optional
Additonal tag information about the compute.
attrs: dict, optional
The additional auxiliary attributes about the compute.
Returns
-------
tensor: Tensor or list of Tensors
The created tensor or tuple of tensors it it contains multiple outputs.
Example
-------
In the code below, C is generated by calling external PackedFunc
`tvm.contrib.cblas.matmul`
.. code-block:: python
A = tvm.placeholder((n, l), name='A')
B = tvm.placeholder((l, m), name='B')
C = tvm.extern((n, m), [A, B],
lambda ins, outs: tvm.call_packed(
"tvm.contrib.cblas.matmul",
ins[0], ins[1], outs[0], 0, 0), name="C")
"""
if _tag.TagScope.get_current() is not None:
if tag != "":
raise ValueError("nested tag is not allowed for now")
tag = _tag.TagScope.get_current().tag
shape = (shape, ) if isinstance(shape,
(_expr.PrimExpr, _Integral)) else shape
if shape == () or isinstance(shape[0], (_expr.PrimExpr, _Integral)):
shape = [shape]
if in_buffers is not None:
in_buffers = [in_buffers
] if not isinstance(in_buffers, list) else in_buffers
if len(inputs) != len(in_buffers):
raise RuntimeError(
"Number of inputs and in_buffers mismatch: %d vs %d." %
(len(inputs), len(in_buffers)))
if out_buffers is not None:
out_buffers = [out_buffers
] if not isinstance(out_buffers, list) else out_buffers
if len(shape) != len(out_buffers):
raise RuntimeError(
"Number of outputs and out_buffers mismatch: %d vs %d." %
(len(shape), len(out_buffers)))
input_placeholders = in_buffers or []
output_placeholders = out_buffers or []
types = set()
for t in inputs:
if not isinstance(t, _tensor.Tensor):
raise ValueError("expect inputs to be tensor")
if in_buffers is None:
input_placeholders.append(decl_buffer(t.shape, t.dtype, t.op.name))
types.add(t.dtype)
if dtype is None:
if len(types) != 1:
raise ValueError(
"Cannot infer output type, please provide dtype argument")
infered_type = types.pop()
dtype = [infered_type for _ in shape]
if isinstance(dtype, str):
dtype = [dtype]
if out_buffers is None:
for shp, dt in zip(shape, dtype):
output_placeholders.append(decl_buffer(shp, dt, name))
body = fcompute(input_placeholders, output_placeholders)
if isinstance(body, _expr.PrimExpr):
body = _stmt.Evaluate(body)
op = _api_internal._ExternOp(name, tag, attrs, inputs, input_placeholders,
output_placeholders, body)
res = [op.output(i) for i in range(len(output_placeholders))]
return res[0] if len(res) == 1 else res
def decl_tensor_intrin(op,
fcompute,
name="tensor_intrin",
binds=None,
scalar_params=None):
"""Declare a tensor intrinsic function.
Parameters
----------
op: Operation
The symbolic description of the intrinsic operation
fcompute: lambda function of inputs, outputs-> stmt
Specifies the IR statement to do the computation.
See the following note for function signature of fcompute
.. note::
**Parameters**
- **ins** (list of :any:`Buffer`) - Placeholder for each inputs
- **outs** (list of :any:`Buffer`) - Placeholder for each outputs
**Returns**
- **stmt** (:any:`Stmt`, or tuple of three stmts)
- If a single stmt is returned, it represents the body
- If tuple of three stmts are returned they corresponds to body,
reduce_init, reduce_update
name: str, optional
The name of the intrinsic.
binds: dict of :any:`Tensor` to :any:`Buffer`, optional
Dictionary that maps the Tensor to Buffer which specified the data layout
requirement of the function. By default, a new compact buffer is created
for each tensor in the argument.
scalar_params: a list of variables used by op, whose values will be passed
as scalar_inputs when the tensor intrinsic is called.
Returns
-------
intrin: TensorIntrin
A TensorIntrin that can be used in tensorize schedule.
"""
if not isinstance(op, _tensor.Operation):
raise TypeError("expect Operation")
inputs = op.input_tensors
binds = binds if binds else {}
tensors = list(inputs)
for i in range(op.num_outputs):
tensors.append(op.output(i))
binds_list = []
for t in inputs:
if not isinstance(t.op, _tensor.PlaceholderOp):
raise ValueError("Do not yet support composition op")
cfg = current_build_config()
for t in tensors:
buf = (binds[t] if t in binds else _api.decl_buffer(
t.shape,
t.dtype,
t.op.name,
data_alignment=cfg.data_alignment,
offset_factor=cfg.offset_factor))
binds_list.append(buf)
if scalar_params:
body = fcompute(binds_list[:len(inputs)], binds_list[len(inputs):],
scalar_params)
else:
body = fcompute(binds_list[:len(inputs)], binds_list[len(inputs):])
scalar_params = []
if isinstance(body, (_expr.PrimExpr, _stmt.Stmt)):
body = [body]
body = [
_stmt.Evaluate(x) if isinstance(x, _expr.PrimExpr) else x for x in body
]
if len(body) < 3:
body += [None] * (3 - len(body))
return _api_internal._TensorIntrin(name, op, inputs, binds_list,
scalar_params, *body)