Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[doc] Update documentaion on meta programming for #1374 #1389

Merged
merged 14 commits into from
Jul 17, 2020
Merged

[doc] Update documentaion on meta programming for #1374 #1389

Show file tree
Hide file tree
Changes from 13 commits
Commits
Show all changes
14 commits
Select commit Hold shift + click to select a range
4e9f1d5
[Refactor] [lang] x.data_type() is deprecated, use x.dtype instead
archibate Jun 29, 2020
356999e
improve doc
archibate Jun 29, 2020
987b738
did deprecate
archibate Jun 29, 2020
f63b83c
fix
archibate Jul 2, 2020
0fd86bb
merge master
archibate Jul 2, 2020
0f34e4f
[skip ci] enforce code format
taichi-gardener Jul 2, 2020
30d8f1c
[skip ci] Update docs/vector.rst
archibate Jul 3, 2020
741c153
[skip ci] Merge branch 'master' into deprecate-data_type-use-dtype-in…
archibate Jul 3, 2020
0d3efb6
revert doc
archibate Jul 3, 2020
38cf03e
Revert "revert doc"
archibate Jul 3, 2020
9f434eb
Merge branch 'master' into doc-deprecate-data_type-use-dtype-instead
archibate Jul 14, 2020
6c1d48f
[skip ci] Apply suggestions from code review
archibate Jul 16, 2020
de62ab5
[skip ci] Update docs/meta.rst
archibate Jul 16, 2020
47cf398
Update docs/meta.rst
archibate Jul 17, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
131 changes: 94 additions & 37 deletions docs/meta.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,55 +17,124 @@ Taichi kernels are *lazily instantiated* and a lot of computation can happen at
Template metaprogramming
------------------------

You may use ``ti.template()``
as a type hint to pass a tensor as an argument. For example:

.. code-block:: python

@ti.kernel
def copy(x: ti.template(), y: ti.template()):
for i in x:
y[i] = x[i]

a = ti.var(ti.f32, 4)
b = ti.var(ti.f32, 4)
c = ti.var(ti.f32, 12)
d = ti.var(ti.f32, 12)
copy(a, b)
copy(c, d)


As shown in the example above, template programming may enable us to reuse our
code and provide more flexibility.


Dimensionality-independent programming using grouped indices
------------------------------------------------------------

However, the ``copy`` template shown above is not perfect. For example, it can only be
used to copy 1D tensors. What if we want to copy 2D tensors? Do we have to write
another kernel?

.. code-block:: python

@ti.kernel
def copy2d(x: ti.template(), y: ti.template()):
for i, j in x:
y[i, j] = x[i, j]

Not necessary! Taichi provides ``ti.grouped`` syntax which enables you to pack
loop indices into a grouped vector to unify kernels of different dimensionalities.
For example:

.. code-block:: python

@ti.kernel
def copy(x: ti.template(), y: ti.template()):
for I in ti.grouped(y):
# I is a vector with same dimensionality with x and data type i32
# If y is 0D, then I = None
archibate marked this conversation as resolved.
Show resolved Hide resolved
# If y is 1D, then I = ti.Vector([i])
# If y is 2D, then I = ti.Vector([i, j])
# If y is 3D, then I = ti.Vector([i, j, k])
# ...
x[I] = y[I]

@ti.kernel
def array_op(x: ti.template(), y: ti.template()):
# If tensor x is 2D
for I in ti.grouped(x): # I is a vector of size x.dim() and data type i32
# if tensor x is 2D:
for I in ti.grouped(x): # I is simply a 2D vector with data type i32
y[I + ti.Vector([0, 1])] = I[0] + I[1]
# is equivalent to

# then it is equivalent to:
for i, j in x:
y[i, j + 1] = i + j

Tensor size reflection
----------------------

Sometimes it will be useful to get the dimensionality (``tensor.dim()``) and shape (``tensor.shape()``) of tensors.
These functions can be used in both Taichi kernels and python scripts.
Tensor metadata
---------------

Sometimes it is useful to get the data type (``tensor.dtype``) and shape (``tensor.shape``) of tensors.
These attributes can be accessed in both Taichi- and Python-scopes.

.. code-block:: python

@ti.func
def print_tensor_size(x: ti.template()):
print(x.dim())
for i in ti.static(range(x.dim())):
print(x.shape()[i])
def print_tensor_info(x: ti.template()):
print('Tensor dimensionality is', len(x.shape))
for i in ti.static(range(len(x.shape))):
print('Size alone dimension', i, 'is', x.shape[i])
ti.static_print('Tensor data type is', x.dtype)

See :ref:`scalar_tensor` for more details.

.. note::

For sparse tensors, the full domain shape will be returned.


Matrix & vector metadata
------------------------

Getting the number of matrix columns and rows will allow
you to write dimensionality-independent code. For example, this can be used to unify
2D and 3D physical simulators.

``matrix.m`` equals to the number of columns of a matrix, while ``matrix.n`` equals to
the number of rows of a matrix.
Since vectors are considered as matrices with one column, ``vector.n`` is simply
the dimensionality of the vector.

.. code-block:: python

@ti.kernel
def foo():
matrix = ti.Matrix([[1, 2], [3, 4], [5, 6]])
print(matrix.n) # 2
print(matrix.m) # 3
vector = ti.Vector([7, 8, 9])
print(vector.n) # 3
print(vector.m) # 1


For sparse tensors, the full domain shape will be returned.

Compile-time evaluations
------------------------

Using compile-time evaluation will allow certain computations to happen when kernels are being instantiated.
This saves the overhead of those computations at runtime.

* Use ``ti.static`` for compile-time branching (for those who come from C++17, this is `if constexpr <https://en.cppreference.com/w/cpp/language/if>`_.)
* Use ``ti.static`` for compile-time branching (for those who come from C++17, this is `if constexpr <https://en.cppreference.com/w/cpp/language/if>`_.):

.. code-block:: python

Expand All @@ -77,32 +146,20 @@ This saves the overhead of those computations at runtime.
x[0] = 1


* Use ``ti.static`` for forced loop unrolling
* Use ``ti.static`` for forced loop unrolling:

.. code-block:: python

@ti.kernel
def g2p(f: ti.i32):
for p in range(0, n_particles):
base = ti.cast(x[f, p] * inv_dx - 0.5, ti.i32)
fx = x[f, p] * inv_dx - ti.cast(base, real)
w = [0.5 * (1.5 - fx) ** 2, 0.75 - (fx - 1.0) ** 2,
0.5 * (fx - 0.5) ** 2]
new_v = ti.Vector([0.0, 0.0])
new_C = ti.Matrix([[0.0, 0.0], [0.0, 0.0]])

# Unrolled 9 iterations for higher performance
for i in ti.static(range(3)):
for j in ti.static(range(3)):
dpos = ti.cast(ti.Vector([i, j]), real) - fx
g_v = grid_v_out[base(0) + i, base(1) + j]
weight = w[i](0) * w[j](1)
new_v += weight * g_v
new_C += 4 * weight * ti.outer_product(g_v, dpos) * inv_dx

v[f + 1, p] = new_v
x[f + 1, p] = x[f, p] + dt * v[f + 1, p]
C[f + 1, p] = new_C
@ti.kernel
def func():
for i in ti.static(range(4)):
print(i)

# is equivalent to:
print(0)
print(1)
print(2)
print(3)
Comment on lines +155 to +162
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It seems that we are mixing 2-space and 4-space identation... Maybe we should somehow enforce PEP8 for all the Python code in the doc? I don't have a clear idea how that can be done automatically :-)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't know, e.g.:

.. code-block::

    4tabs

vs

.. code-block::

  2tabs

Is that possible to let ti format take care of that too?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is that possible to let ti format take care of that too?

We probably can, but I don't think that would be easy. We will need to write an rst parster that extract the code, format using yapf, and the put it back...

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No, my idea is: not only parse the code-block, but parse the whole rst, e.g.:

- This  is a  bad line
+ This is a bad line
- This is a bad line that is tooooooooooo long for yapf
+ This is a bad line that is tooooooooooo
+ long for yapf

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You probably need a different tool. I don't think yapf can correctly handle rst...



When to use for loops with ``ti.static``
Expand All @@ -120,7 +177,7 @@ For example, code for resetting this tensor of vectors should be
@ti.kernel
def reset():
for i in x:
for j in ti.static(range(3)):
for j in ti.static(range(x.n)):
# The inner loop must be unrolled since j is a vector index instead
# of a global tensor index.
x[i][j] = 0
28 changes: 6 additions & 22 deletions docs/scalar_tensor.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,49 +107,33 @@ You can access an element of the Taichi tensor by an index or indices.
Meta data
---------

.. function:: a.dim()

:parameter a: (Tensor) the tensor
:return: (scalar) the length of ``a``

::

x = ti.var(ti.i32, (6, 5))
x.dim() # 2

y = ti.var(ti.i32, 6)
y.dim() # 1

z = ti.var(ti.i32, ())
z.dim() # 0


.. function:: a.shape()
.. attribute:: a.shape

:parameter a: (Tensor) the tensor
:return: (tuple) the shape of tensor ``a``

::

x = ti.var(ti.i32, (6, 5))
x.shape() # (6, 5)
x.shape # (6, 5)

y = ti.var(ti.i32, 6)
y.shape() # (6,)
y.shape # (6,)

z = ti.var(ti.i32, ())
z.shape() # ()
z.shape # ()


.. function:: a.data_type()
.. function:: a.dtype

:parameter a: (Tensor) the tensor
:return: (DataType) the data type of ``a``

::

x = ti.var(ti.i32, (2, 3))
x.data_type() # ti.i32
x.dtype # ti.i32


.. function:: a.parent(n = 1)
Expand Down
50 changes: 14 additions & 36 deletions docs/snode.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,32 +33,19 @@ See :ref:`layout` for more details. ``ti.root`` is the root node of the data str
assert x.snode() == y.snode()


.. function:: tensor.shape()
.. function:: tensor.shape

:parameter tensor: (Tensor)
:return: (tuple of integers) the shape of tensor

Equivalent to ``tensor.snode().shape()``.
Equivalent to ``tensor.snode().shape``.

For example,

::

ti.root.dense(ti.ijk, (3, 5, 4)).place(x)
x.shape() # returns (3, 5, 4)


.. function:: tensor.dim()

:parameter tensor: (Tensor)
:return: (scalar) the dimensionality of the tensor

Equivalent to ``len(tensor.shape())``.

::

ti.root.dense(ti.ijk, (8, 9, 10)).place(x)
x.dim() # 3
x.shape # returns (3, 5, 4)


.. function:: tensor.snode()
Expand All @@ -74,7 +61,7 @@ See :ref:`layout` for more details. ``ti.root`` is the root node of the data str
x.snode()


.. function:: snode.shape()
.. function:: snode.shape

:parameter snode: (SNode)
:return: (tuple) the size of node along that axis
Expand All @@ -85,29 +72,16 @@ See :ref:`layout` for more details. ``ti.root`` is the root node of the data str
blk2 = blk1.dense(ti.i, 3)
blk3 = blk2.dense(ti.jk, (5, 2))
blk4 = blk3.dense(ti.k, 2)
blk1.shape() # ()
blk2.shape() # (3, )
blk3.shape() # (3, 5, 2)
blk4.shape() # (3, 5, 4)


.. function:: snode.dim()

:parameter snode: (SNode)
:return: (scalar) the dimensionality of ``snode``

Equivalent to ``len(snode.shape())``.

::

blk1 = ti.root.dense(ti.ijk, (8, 9, 10))
ti.root.dim() # 0
blk1.dim() # 3
blk1.shape # ()
blk2.shape # (3, )
blk3.shape # (3, 5, 2)
blk4.shape # (3, 5, 4)


.. function:: snode.parent()
.. function:: snode.parent(n = 1)

:parameter snode: (SNode)
:parameter n: (optional, scalar) the number of steps, i.e. ``n=1`` for parent, ``n=2`` grandparent, etc.
:return: (SNode) the parent node of ``snode``

::
Expand All @@ -118,6 +92,10 @@ See :ref:`layout` for more details. ``ti.root`` is the root node of the data str
blk1.parent() # ti.root
blk2.parent() # blk1
blk3.parent() # blk2
blk3.parent(1) # blk2
blk3.parent(2) # blk1
blk3.parent(3) # ti.root
blk3.parent(4) # None


Node types
Expand Down
3 changes: 3 additions & 0 deletions docs/vector.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,6 +191,9 @@ Methods
Vectors are special matrices with only 1 column. In fact, ``ti.Vector`` is just an alias of ``ti.Matrix``.


Metadata
--------

.. attribute:: a.n

:parameter a: (Vector or tensor of Vector)
Expand Down