From da9a553ad77c32becfb1fb187e44bc19ee04ec01 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Thu, 4 Jun 2020 09:00:03 +0800
Subject: [PATCH 01/19] [skip ci] tmp work save

---
 python/taichi/lang/expr.py   |  3 +++
 python/taichi/lang/matrix.py | 14 ++++++++++----
 python/taichi/misc/gui.py    | 27 +++++++++++++++++++--------
 python/taichi/misc/image.py  |  2 +-
 4 files changed, 33 insertions(+), 13 deletions(-)

diff --git a/python/taichi/lang/expr.py b/python/taichi/lang/expr.py
index 83332b710f3b7..d30083aa1e8eb 100644
--- a/python/taichi/lang/expr.py
+++ b/python/taichi/lang/expr.py
@@ -142,6 +142,9 @@ def shape(self):
             self.materialize_layout_callback()
         return self.snode().shape()
 
+    def shape_ext(self):
+        return ()
+
     def data_type(self):
         return self.snode().data_type()
 
diff --git a/python/taichi/lang/matrix.py b/python/taichi/lang/matrix.py
index 14f45b8448c8c..0f12aff946a03 100644
--- a/python/taichi/lang/matrix.py
+++ b/python/taichi/lang/matrix.py
@@ -497,11 +497,17 @@ def assign_renamed(x, y):
         from .meta import fill_matrix
         fill_matrix(self, val)
 
+    def shape_ext(self, as_vector=None):
+        if as_vector is None:
+            as_vector = self.m == 1
+        shape_ext = (self.n, ) if as_vector else (self.n, self.m)
+        return shape_ext
+
     def to_numpy(self, keep_dims=False):
         # Discussion: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858
         as_vector = self.m == 1 and not keep_dims
-        dim_ext = (self.n, ) if as_vector else (self.n, self.m)
-        ret = np.empty(self.loop_range().shape() + dim_ext,
+        shape_ext = self.shape_ext(as_vector)
+        ret = np.empty(self.loop_range().shape() + shape_ext,
                        dtype=to_numpy_type(
                            self.loop_range().snode().data_type()))
         from .meta import matrix_to_ext_arr
@@ -513,8 +519,8 @@ def to_numpy(self, keep_dims=False):
     def to_torch(self, device=None, keep_dims=False):
         import torch
         as_vector = self.m == 1 and not keep_dims
-        dim_ext = (self.n, ) if as_vector else (self.n, self.m)
-        ret = torch.empty(self.loop_range().shape() + dim_ext,
+        shape_ext = self.shape_ext(as_vector)
+        ret = torch.empty(self.loop_range().shape() + shape_ext,
                           dtype=to_pytorch_type(
                               self.loop_range().snode().data_type()),
                           device=device)
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index 7267d8eb8d65a..071f361aa3ac3 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -49,30 +49,41 @@ def clear(self, color=None):
 
     def set_image(self, img):
         import numpy as np
+        import taichi as ti
         from .image import cook_image
+        from .util import get_os_name
+
+        ti.profiler_start('cook_image')
         img = cook_image(img)
+        ti.profiler_stop()
+
+        ti.profiler_start('astype')
         if img.dtype in [np.uint8, np.uint16, np.uint32, np.uint64]:
             img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
         elif img.dtype in [np.float32, np.float64]:
-            img = np.clip(img.astype(np.float32), 0, 1)
+            if img.dtype == np.float32 and get_os_name() == 'win':
+                img = np.clip(img.astype(np.float32), 0, 1)
         else:
             raise ValueError(
                 f'Data type {img.dtype} not supported in GUI.set_image')
+        ti.profiler_stop()
+
+        ti.profiler_start('concatenate')
         if len(img.shape) == 2:
             img = img[..., None]
         if img.shape[2] == 1:
-            img = img + np.zeros(shape=(1, 1, 4))
+            img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
         if img.shape[2] == 3:
-            img = np.concatenate([
-                img,
+            img = np.concatenate([img,
                 np.zeros(shape=(img.shape[0], img.shape[1], 1),
-                         dtype=np.float32)
-            ],
-                                 axis=2)
-        img = img.astype(np.float32)
+                         dtype=np.float32)], axis=2)
         assert img.shape[:
                          2] == self.res, "Image resolution does not match GUI resolution"
+        ti.profiler_stop()
+
+        ti.profiler_start('set_img')
         self.core.set_img(np.ascontiguousarray(img).ctypes.data)
+        ti.profiler_stop()
 
     def circle(self, pos, color=0xFFFFFF, radius=1):
         self.canvas.circle_single(pos[0], pos[1], color, radius)
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 30469709e4820..fcdbc72d2fb1f 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -4,7 +4,7 @@
 
 def cook_image(img):
     if isinstance(img, ti.Matrix):
-        img = img.to_numpy(as_vector=True)
+        img = img.to_numpy()
     if isinstance(img, ti.Expr):
         img = img.to_numpy()
     assert isinstance(img, np.ndarray)

From 7e57cfbcc0161a6d2af77532e5c1a453504314a4 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Thu, 4 Jun 2020 14:52:01 +0800
Subject: [PATCH 02/19] reduce some redulent numpy copy for f32 image

---
 examples/simple_uv.py           | 29 +++++++++++++
 python/taichi/lang/meta.py      | 22 ++++++++++
 python/taichi/misc/gui.py       | 77 +++++++++++++++++++++------------
 python/taichi/misc/image.py     | 17 ++------
 taichi/python/export_visual.cpp |  5 +++
 5 files changed, 110 insertions(+), 40 deletions(-)
 create mode 100644 examples/simple_uv.py

diff --git a/examples/simple_uv.py b/examples/simple_uv.py
new file mode 100644
index 0000000000000..8f77642db8fd4
--- /dev/null
+++ b/examples/simple_uv.py
@@ -0,0 +1,29 @@
+import taichi as ti
+import numpy as np
+
+ti.init(arch=ti.cpu, print_ir=True)
+
+res = 1280, 720
+pixels = ti.Vector(3, dt=ti.f32, shape=res)
+
+@ti.kernel
+def paint(size_x: ti.template(), size_y: ti.template()):
+  for i, j in pixels:
+    u = i / size_x
+    v = j / size_y
+    pixels[i, j] = [u, v, 0]
+
+gui = ti.GUI('UV', res)
+
+paint(res[0], res[1])
+gui.set_image(pixels)
+
+while not gui.get_event(ti.GUI.ESCAPE):
+    ti.profiler_start('paint')
+    paint(res[0], res[1])
+    ti.sync()
+    ti.profiler_stop()
+    gui.set_image(pixels)
+    gui.show()
+
+ti.profiler_print()
diff --git a/python/taichi/lang/meta.py b/python/taichi/lang/meta.py
index 9fdc8e231461c..8707b21a71d8e 100644
--- a/python/taichi/lang/meta.py
+++ b/python/taichi/lang/meta.py
@@ -15,6 +15,28 @@ def tensor_to_ext_arr(tensor: ti.template(), arr: ti.ext_arr()):
         arr[I] = tensor[I]
 
 
+@ti.func
+def cook_image_type(x):
+    x = ti.cast(x, ti.f32)
+    # Issue 1005, don't know why win GUI doesn't do clamp for us
+    if ti.static(ti.get_os_name() == 'win'):
+        x = ti.min(1, ti.max(0, x))
+    return x
+
+
+@ti.kernel
+def tensor_to_image(tensor: ti.template(), arr: ti.ext_arr()):
+    for I in ti.grouped(tensor):
+        arr[I] = cook_image_type(tensor[I])
+
+
+@ti.kernel
+def vector_to_image(mat: ti.template(), arr: ti.ext_arr()):
+    for I in ti.grouped(mat):
+        for p in ti.static(range(mat.n)):
+            arr[I, p] = cook_image_type(mat[I][p])
+
+
 @ti.kernel
 def tensor_to_tensor(tensor: ti.template(), other: ti.template()):
     for I in ti.grouped(tensor):
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index 071f361aa3ac3..c1ed3086b23b0 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -32,6 +32,7 @@ def __init__(self, name, res=512, background_color=0x0):
         if isinstance(res, numbers.Number):
             res = (res, res)
         self.res = res
+        self.img = np.ascontiguousarray(np.zeros(self.res + (4,), np.float32))
         self.core = ti.core.GUI(name, ti.veci(*res))
         self.canvas = self.core.get_canvas()
         self.background_color = background_color
@@ -50,39 +51,61 @@ def clear(self, color=None):
     def set_image(self, img):
         import numpy as np
         import taichi as ti
-        from .image import cook_image
-        from .util import get_os_name
 
-        ti.profiler_start('cook_image')
-        img = cook_image(img)
-        ti.profiler_stop()
+        def cook_image(img):
+            if img.dtype in [np.uint8, np.uint16, np.uint32, np.uint64]:
+                img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
+            elif img.dtype in [np.float32, np.float64]:
+                img = img.astype(np.float32)
+                from .util import get_os_name
+                if img.dtype == np.float32 and get_os_name() == 'win':
+                    img = np.clip(img, 0, 1)
+            else:
+                raise ValueError(
+                    f'Data type {img.dtype} not supported in GUI.set_image')
+            
+            if len(img.shape) == 2:
+                img = img[..., None]
+            if img.shape[2] == 1:
+                img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
+            if img.shape[2] == 3:
+                img = np.concatenate([img,
+                    np.zeros(shape=(img.shape[0], img.shape[1], 1),
+                             dtype=np.float32)], axis=2)
+            assert img.shape[:2] == self.res, "Image resolution does not match GUI resolution"
+            return np.ascontiguousarray(img)
+
+        ti.profiler_start('copy_image')
+        if isinstance(img, ti.Expr):
+            if ti.core.is_integral(img.data_type()):
+                # image of uint is not optimized by xxx_to_image
+                self.img = cook_image(img.to_numpy())
+            else:
+                assert img.shape() == self.res, "Image resolution does not match GUI resolution"
+                from taichi.lang.meta import tensor_to_image
+                tensor_to_image(img, self.img)
+                ti.sync()
+
+        elif isinstance(img, ti.Matrix):
+            if ti.core.is_integral(img.data_type()):
+                self.img = cook_image(img.to_numpy())
+            else:
+                assert img.shape() == self.res, "Image resolution does not match GUI resolution"
+                assert img.n in [3, 4], "Only greyscale, RGB or RGBA images are supported in GUI.set_image"
+                assert img.m == 1
+                from taichi.lang.meta import vector_to_image
+                vector_to_image(img, self.img)
+                ti.sync()
+
+        elif isinstance(img, np.ndarray):
+            self.img = cook_image(img)
 
-        ti.profiler_start('astype')
-        if img.dtype in [np.uint8, np.uint16, np.uint32, np.uint64]:
-            img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
-        elif img.dtype in [np.float32, np.float64]:
-            if img.dtype == np.float32 and get_os_name() == 'win':
-                img = np.clip(img.astype(np.float32), 0, 1)
         else:
-            raise ValueError(
-                f'Data type {img.dtype} not supported in GUI.set_image')
-        ti.profiler_stop()
-
-        ti.profiler_start('concatenate')
-        if len(img.shape) == 2:
-            img = img[..., None]
-        if img.shape[2] == 1:
-            img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
-        if img.shape[2] == 3:
-            img = np.concatenate([img,
-                np.zeros(shape=(img.shape[0], img.shape[1], 1),
-                         dtype=np.float32)], axis=2)
-        assert img.shape[:
-                         2] == self.res, "Image resolution does not match GUI resolution"
+            raise ValueError(f"GUI.set_image only takes Taichi tensor or NumPy array, not {type(img)}")
         ti.profiler_stop()
 
         ti.profiler_start('set_img')
-        self.core.set_img(np.ascontiguousarray(img).ctypes.data)
+        self.core.set_img(self.img.ctypes.data)
         ti.profiler_stop()
 
     def circle(self, pos, color=0xFFFFFF, radius=1):
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index fcdbc72d2fb1f..881aa856e8fe2 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -1,19 +1,9 @@
 import numpy as np
 import taichi as ti
 
-
-def cook_image(img):
-    if isinstance(img, ti.Matrix):
-        img = img.to_numpy()
-    if isinstance(img, ti.Expr):
-        img = img.to_numpy()
-    assert isinstance(img, np.ndarray)
-    assert len(img.shape) in [2, 3]
-    return img
-
-
 def imwrite(img, filename):
-    img = cook_image(img)
+    img = img.to_numpy()
+    assert len(img.shape) in [2, 3]
     resx, resy = img.shape[:2]
     if len(img.shape) == 3:
         comp = img.shape[2]
@@ -35,7 +25,8 @@ def imread(filename, channels=0):
 
 
 def imshow(img, winname='Taichi'):
-    img = cook_image(img)
+    img = img.to_numpy()
+    assert len(img.shape) in [2, 3]
     gui = ti.GUI(winname, res=img.shape[:2])
     while not gui.get_event(ti.GUI.ESCAPE):
         gui.set_image(img)
diff --git a/taichi/python/export_visual.cpp b/taichi/python/export_visual.cpp
index f28148b417129..6626cb0142792 100644
--- a/taichi/python/export_visual.cpp
+++ b/taichi/python/export_visual.cpp
@@ -22,6 +22,11 @@ void export_visual(py::module &m) {
              std::memcpy((void *)img.get_data().data(), (void *)ptr,
                          img.get_data_size());
            })
+      .def("get_img_ptr",
+           [&](GUI *gui) -> std::size_t {
+             auto &img = gui->canvas->img;
+             return (size_t)img.get_data().data();
+           })
       .def("screenshot", &GUI::screenshot)
       .def("has_key_event", &GUI::has_key_event)
       .def("wait_key_event", &GUI::wait_key_event)

From 3aaece3e1c3e3d20abbfa255c345568f81983feb Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Thu, 4 Jun 2020 15:03:28 +0800
Subject: [PATCH 03/19] [skip ci] no hoc

---
 examples/simple_uv.py     | 11 +----------
 misc/simple_uv.py         | 31 +++++++++++++++++++++++++++++++
 python/taichi/misc/gui.py |  4 ----
 3 files changed, 32 insertions(+), 14 deletions(-)
 create mode 100644 misc/simple_uv.py

diff --git a/examples/simple_uv.py b/examples/simple_uv.py
index 8f77642db8fd4..7c68ae7adc8c1 100644
--- a/examples/simple_uv.py
+++ b/examples/simple_uv.py
@@ -1,7 +1,7 @@
 import taichi as ti
 import numpy as np
 
-ti.init(arch=ti.cpu, print_ir=True)
+ti.init()
 
 res = 1280, 720
 pixels = ti.Vector(3, dt=ti.f32, shape=res)
@@ -14,16 +14,7 @@ def paint(size_x: ti.template(), size_y: ti.template()):
     pixels[i, j] = [u, v, 0]
 
 gui = ti.GUI('UV', res)
-
-paint(res[0], res[1])
-gui.set_image(pixels)
-
 while not gui.get_event(ti.GUI.ESCAPE):
-    ti.profiler_start('paint')
     paint(res[0], res[1])
-    ti.sync()
-    ti.profiler_stop()
     gui.set_image(pixels)
     gui.show()
-
-ti.profiler_print()
diff --git a/misc/simple_uv.py b/misc/simple_uv.py
new file mode 100644
index 0000000000000..92a4573e86535
--- /dev/null
+++ b/misc/simple_uv.py
@@ -0,0 +1,31 @@
+import taichi as ti
+import numpy as np
+
+ti.init(arch=ti.cpu, print_ir=True)
+
+res = 1280, 720
+pixels = ti.Vector(3, dt=ti.f32, shape=res)
+
+@ti.kernel
+def paint(size_x: ti.template(), size_y: ti.template()):
+  for i, j in pixels:
+    u = i / size_x
+    v = j / size_y
+    pixels[i, j] = [u, v, 0]
+
+gui = ti.GUI('UV', res)
+
+paint(res[0], res[1])
+gui.set_image(pixels)
+
+for i in range(200):
+    ti.profiler_start('paint')
+    paint(res[0], res[1])
+    ti.sync()
+    ti.profiler_stop()
+    ti.profiler_start('set_image')
+    gui.set_image(pixels)
+    ti.profiler_stop()
+    gui.show()
+
+ti.profiler_print()
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index c1ed3086b23b0..9f6a5aae0d3fa 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -75,7 +75,6 @@ def cook_image(img):
             assert img.shape[:2] == self.res, "Image resolution does not match GUI resolution"
             return np.ascontiguousarray(img)
 
-        ti.profiler_start('copy_image')
         if isinstance(img, ti.Expr):
             if ti.core.is_integral(img.data_type()):
                 # image of uint is not optimized by xxx_to_image
@@ -102,11 +101,8 @@ def cook_image(img):
 
         else:
             raise ValueError(f"GUI.set_image only takes Taichi tensor or NumPy array, not {type(img)}")
-        ti.profiler_stop()
 
-        ti.profiler_start('set_img')
         self.core.set_img(self.img.ctypes.data)
-        ti.profiler_stop()
 
     def circle(self, pos, color=0xFFFFFF, radius=1):
         self.canvas.circle_single(pos[0], pos[1], color, radius)

From 9a355b8982d01677576a9d0298b8e0453abb0fca Mon Sep 17 00:00:00 2001
From: Taichi Gardener <taichigardener@gmail.com>
Date: Thu, 4 Jun 2020 03:07:55 -0400
Subject: [PATCH 04/19] [skip ci] enforce code format

---
 examples/simple_uv.py       | 10 ++++++----
 misc/simple_uv.py           | 10 ++++++----
 python/taichi/misc/gui.py   | 28 +++++++++++++++++++---------
 python/taichi/misc/image.py |  1 +
 4 files changed, 32 insertions(+), 17 deletions(-)

diff --git a/examples/simple_uv.py b/examples/simple_uv.py
index 7c68ae7adc8c1..262853b95fa34 100644
--- a/examples/simple_uv.py
+++ b/examples/simple_uv.py
@@ -6,12 +6,14 @@
 res = 1280, 720
 pixels = ti.Vector(3, dt=ti.f32, shape=res)
 
+
 @ti.kernel
 def paint(size_x: ti.template(), size_y: ti.template()):
-  for i, j in pixels:
-    u = i / size_x
-    v = j / size_y
-    pixels[i, j] = [u, v, 0]
+    for i, j in pixels:
+        u = i / size_x
+        v = j / size_y
+        pixels[i, j] = [u, v, 0]
+
 
 gui = ti.GUI('UV', res)
 while not gui.get_event(ti.GUI.ESCAPE):
diff --git a/misc/simple_uv.py b/misc/simple_uv.py
index 92a4573e86535..9403acdf93111 100644
--- a/misc/simple_uv.py
+++ b/misc/simple_uv.py
@@ -6,12 +6,14 @@
 res = 1280, 720
 pixels = ti.Vector(3, dt=ti.f32, shape=res)
 
+
 @ti.kernel
 def paint(size_x: ti.template(), size_y: ti.template()):
-  for i, j in pixels:
-    u = i / size_x
-    v = j / size_y
-    pixels[i, j] = [u, v, 0]
+    for i, j in pixels:
+        u = i / size_x
+        v = j / size_y
+        pixels[i, j] = [u, v, 0]
+
 
 gui = ti.GUI('UV', res)
 
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index 9f6a5aae0d3fa..a91c53448d0f0 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -32,7 +32,7 @@ def __init__(self, name, res=512, background_color=0x0):
         if isinstance(res, numbers.Number):
             res = (res, res)
         self.res = res
-        self.img = np.ascontiguousarray(np.zeros(self.res + (4,), np.float32))
+        self.img = np.ascontiguousarray(np.zeros(self.res + (4, ), np.float32))
         self.core = ti.core.GUI(name, ti.veci(*res))
         self.canvas = self.core.get_canvas()
         self.background_color = background_color
@@ -63,16 +63,20 @@ def cook_image(img):
             else:
                 raise ValueError(
                     f'Data type {img.dtype} not supported in GUI.set_image')
-            
+
             if len(img.shape) == 2:
                 img = img[..., None]
             if img.shape[2] == 1:
                 img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
             if img.shape[2] == 3:
-                img = np.concatenate([img,
+                img = np.concatenate([
+                    img,
                     np.zeros(shape=(img.shape[0], img.shape[1], 1),
-                             dtype=np.float32)], axis=2)
-            assert img.shape[:2] == self.res, "Image resolution does not match GUI resolution"
+                             dtype=np.float32)
+                ],
+                                     axis=2)
+            assert img.shape[:
+                             2] == self.res, "Image resolution does not match GUI resolution"
             return np.ascontiguousarray(img)
 
         if isinstance(img, ti.Expr):
@@ -80,7 +84,8 @@ def cook_image(img):
                 # image of uint is not optimized by xxx_to_image
                 self.img = cook_image(img.to_numpy())
             else:
-                assert img.shape() == self.res, "Image resolution does not match GUI resolution"
+                assert img.shape(
+                ) == self.res, "Image resolution does not match GUI resolution"
                 from taichi.lang.meta import tensor_to_image
                 tensor_to_image(img, self.img)
                 ti.sync()
@@ -89,8 +94,11 @@ def cook_image(img):
             if ti.core.is_integral(img.data_type()):
                 self.img = cook_image(img.to_numpy())
             else:
-                assert img.shape() == self.res, "Image resolution does not match GUI resolution"
-                assert img.n in [3, 4], "Only greyscale, RGB or RGBA images are supported in GUI.set_image"
+                assert img.shape(
+                ) == self.res, "Image resolution does not match GUI resolution"
+                assert img.n in [
+                    3, 4
+                ], "Only greyscale, RGB or RGBA images are supported in GUI.set_image"
                 assert img.m == 1
                 from taichi.lang.meta import vector_to_image
                 vector_to_image(img, self.img)
@@ -100,7 +108,9 @@ def cook_image(img):
             self.img = cook_image(img)
 
         else:
-            raise ValueError(f"GUI.set_image only takes Taichi tensor or NumPy array, not {type(img)}")
+            raise ValueError(
+                f"GUI.set_image only takes Taichi tensor or NumPy array, not {type(img)}"
+            )
 
         self.core.set_img(self.img.ctypes.data)
 
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 881aa856e8fe2..d250a7cce438f 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -1,6 +1,7 @@
 import numpy as np
 import taichi as ti
 
+
 def imwrite(img, filename):
     img = img.to_numpy()
     assert len(img.shape) in [2, 3]

From 1780b6a86accc9bff8320bf64970979cc97fc1f6 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 10:32:36 +0800
Subject: [PATCH 05/19] fix image test

---
 python/taichi/misc/image.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index d250a7cce438f..0e35fe7f56adc 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -3,7 +3,8 @@
 
 
 def imwrite(img, filename):
-    img = img.to_numpy()
+    if not isinstance(np.ndarray):
+        img = img.to_numpy()
     assert len(img.shape) in [2, 3]
     resx, resy = img.shape[:2]
     if len(img.shape) == 3:
@@ -26,7 +27,8 @@ def imread(filename, channels=0):
 
 
 def imshow(img, winname='Taichi'):
-    img = img.to_numpy()
+    if not isinstance(np.ndarray):
+        img = img.to_numpy()
     assert len(img.shape) in [2, 3]
     gui = ti.GUI(winname, res=img.shape[:2])
     while not gui.get_event(ti.GUI.ESCAPE):

From 721b1f15490af4964a9174586737e4959d05badc Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 12:47:09 +0800
Subject: [PATCH 06/19] [skip ci] remove misc/simple_uv.py

---
 examples/simple_uv.py      |  8 ++++----
 misc/simple_uv.py          | 33 ---------------------------------
 python/taichi/lang/meta.py |  2 +-
 3 files changed, 5 insertions(+), 38 deletions(-)
 delete mode 100644 misc/simple_uv.py

diff --git a/examples/simple_uv.py b/examples/simple_uv.py
index 262853b95fa34..f9e30568c1ede 100644
--- a/examples/simple_uv.py
+++ b/examples/simple_uv.py
@@ -8,15 +8,15 @@
 
 
 @ti.kernel
-def paint(size_x: ti.template(), size_y: ti.template()):
+def paint():
     for i, j in pixels:
-        u = i / size_x
-        v = j / size_y
+        u = i / res[0]
+        v = j / res[1]
         pixels[i, j] = [u, v, 0]
 
 
 gui = ti.GUI('UV', res)
 while not gui.get_event(ti.GUI.ESCAPE):
-    paint(res[0], res[1])
+    paint()
     gui.set_image(pixels)
     gui.show()
diff --git a/misc/simple_uv.py b/misc/simple_uv.py
deleted file mode 100644
index 9403acdf93111..0000000000000
--- a/misc/simple_uv.py
+++ /dev/null
@@ -1,33 +0,0 @@
-import taichi as ti
-import numpy as np
-
-ti.init(arch=ti.cpu, print_ir=True)
-
-res = 1280, 720
-pixels = ti.Vector(3, dt=ti.f32, shape=res)
-
-
-@ti.kernel
-def paint(size_x: ti.template(), size_y: ti.template()):
-    for i, j in pixels:
-        u = i / size_x
-        v = j / size_y
-        pixels[i, j] = [u, v, 0]
-
-
-gui = ti.GUI('UV', res)
-
-paint(res[0], res[1])
-gui.set_image(pixels)
-
-for i in range(200):
-    ti.profiler_start('paint')
-    paint(res[0], res[1])
-    ti.sync()
-    ti.profiler_stop()
-    ti.profiler_start('set_image')
-    gui.set_image(pixels)
-    ti.profiler_stop()
-    gui.show()
-
-ti.profiler_print()
diff --git a/python/taichi/lang/meta.py b/python/taichi/lang/meta.py
index 8707b21a71d8e..74b411a23c380 100644
--- a/python/taichi/lang/meta.py
+++ b/python/taichi/lang/meta.py
@@ -18,7 +18,7 @@ def tensor_to_ext_arr(tensor: ti.template(), arr: ti.ext_arr()):
 @ti.func
 def cook_image_type(x):
     x = ti.cast(x, ti.f32)
-    # Issue 1005, don't know why win GUI doesn't do clamp for us
+    # Issue #1000, don't know why win GUI doesn't do clamp for us
     if ti.static(ti.get_os_name() == 'win'):
         x = ti.min(1, ti.max(0, x))
     return x

From c0953bdaf5d293d0ef6067658a4b8c3a590299fd Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 12:58:12 +0800
Subject: [PATCH 07/19] [skip ci] remove get_img_ptr

---
 taichi/python/export_visual.cpp | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/taichi/python/export_visual.cpp b/taichi/python/export_visual.cpp
index 6626cb0142792..f28148b417129 100644
--- a/taichi/python/export_visual.cpp
+++ b/taichi/python/export_visual.cpp
@@ -22,11 +22,6 @@ void export_visual(py::module &m) {
              std::memcpy((void *)img.get_data().data(), (void *)ptr,
                          img.get_data_size());
            })
-      .def("get_img_ptr",
-           [&](GUI *gui) -> std::size_t {
-             auto &img = gui->canvas->img;
-             return (size_t)img.get_data().data();
-           })
       .def("screenshot", &GUI::screenshot)
       .def("has_key_event", &GUI::has_key_event)
       .def("wait_key_event", &GUI::wait_key_event)

From 36b11e5e39ac69ea7823ae58ed154a65d798a598 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 12:59:15 +0800
Subject: [PATCH 08/19] [skip ci] comment

---
 python/taichi/misc/gui.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index d16c2083f0a0c..b0ebbc3653d5c 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -32,6 +32,7 @@ def __init__(self, name, res=512, background_color=0x0):
         if isinstance(res, numbers.Number):
             res = (res, res)
         self.res = res
+        # The GUI canvas uses RGBA for storage, therefore we needs NxMx4 for a image
         self.img = np.ascontiguousarray(np.zeros(self.res + (4, ), np.float32))
         self.core = ti.core.GUI(name, ti.veci(*res))
         self.canvas = self.core.get_canvas()

From 39564c8507511dc9fc706d94ef35300ac448a5d3 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 13:11:45 +0800
Subject: [PATCH 09/19] fix image test again

---
 docs/gui.rst                | 14 +++++++-------
 python/taichi/misc/image.py |  4 ++--
 2 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/docs/gui.rst b/docs/gui.rst
index 94a50322d20b2..436db4dce7116 100644
--- a/docs/gui.rst
+++ b/docs/gui.rst
@@ -89,7 +89,7 @@ Paint on a window
     .. note ::
 
         When using ``float32`` or ``float64`` as the data type,
-        ``img`` entries will be clipped into range ``[0, 1]``.
+        ``img`` entries will be clipped into range ``[0, 1]`` for display.
 
 
 .. function:: gui.circle(pos, color = 0xFFFFFF, radius = 1)
@@ -169,14 +169,14 @@ Every event have a key and type.
 
 ::
 
-  ti.GUI.ESCAPE
-  ti.GUI.SHIFT
-  ti.GUI.LEFT
-  'a'
+  ti.GUI.ESCAPE  # Esc
+  ti.GUI.SHIFT   # Shift
+  ti.GUI.LEFT    # Left Arrow
+  'a'            # we use lowercase for alphabet
   'b'
   ...
-  ti.GUI.LMB
-  ti.GUI.RMB
+  ti.GUI.LMB     # Left Mouse Button
+  ti.GUI.RMB     # Right Mouse Button
 
 *Event type* is the type of event, for now, there are just three type of event:
 
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 0e35fe7f56adc..3ae20e210ec91 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -3,7 +3,7 @@
 
 
 def imwrite(img, filename):
-    if not isinstance(np.ndarray):
+    if not isinstance(img, np.ndarray):
         img = img.to_numpy()
     assert len(img.shape) in [2, 3]
     resx, resy = img.shape[:2]
@@ -27,7 +27,7 @@ def imread(filename, channels=0):
 
 
 def imshow(img, winname='Taichi'):
-    if not isinstance(np.ndarray):
+    if not isinstance(img, np.ndarray):
         img = img.to_numpy()
     assert len(img.shape) in [2, 3]
     gui = ti.GUI(winname, res=img.shape[:2])

From 8b6e8fe462bec202b9e0dc0d428e65b6ab39028c Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Fri, 5 Jun 2020 22:26:39 +0800
Subject: [PATCH 10/19] [skip ci] no shape_ext

---
 python/taichi/lang/expr.py   |  3 ---
 python/taichi/lang/matrix.py | 10 ++--------
 2 files changed, 2 insertions(+), 11 deletions(-)

diff --git a/python/taichi/lang/expr.py b/python/taichi/lang/expr.py
index 9f56ac3518e44..3009607755244 100644
--- a/python/taichi/lang/expr.py
+++ b/python/taichi/lang/expr.py
@@ -146,9 +146,6 @@ def shape(self):
             self.materialize_layout_callback()
         return self.snode().shape()
 
-    def shape_ext(self):
-        return ()
-
     def data_type(self):
         return self.snode().data_type()
 
diff --git a/python/taichi/lang/matrix.py b/python/taichi/lang/matrix.py
index 75ccca7f02c9f..1feb7a8a907f3 100644
--- a/python/taichi/lang/matrix.py
+++ b/python/taichi/lang/matrix.py
@@ -604,12 +604,6 @@ def assign_renamed(x, y):
         from .meta import fill_matrix
         fill_matrix(self, val)
 
-    def shape_ext(self, as_vector=None):
-        if as_vector is None:
-            as_vector = self.m == 1
-        shape_ext = (self.n, ) if as_vector else (self.n, self.m)
-        return shape_ext
-
     @python_scope
     def to_numpy(self, keep_dims=False, as_vector=None):
         # Discussion: https://github.com/taichi-dev/taichi/pull/1046#issuecomment-633548858
@@ -621,7 +615,7 @@ def to_numpy(self, keep_dims=False, as_vector=None):
                 DeprecationWarning,
                 stacklevel=3)
         as_vector = self.m == 1 and not keep_dims
-        shape_ext = self.shape_ext(as_vector)
+        shape_ext = (self.n, ) if as_vector else (self.n, self.m)
         ret = np.empty(self.loop_range().shape() + shape_ext,
                        dtype=to_numpy_type(
                            self.loop_range().snode().data_type()))
@@ -635,7 +629,7 @@ def to_numpy(self, keep_dims=False, as_vector=None):
     def to_torch(self, device=None, keep_dims=False):
         import torch
         as_vector = self.m == 1 and not keep_dims
-        shape_ext = self.shape_ext(as_vector)
+        shape_ext = (self.n, ) if as_vector else (self.n, self.m)
         ret = torch.empty(self.loop_range().shape() + shape_ext,
                           dtype=to_pytorch_type(
                               self.loop_range().snode().data_type()),

From 2f9b38df076934457a10243ed895aa9a02b750ab Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 12:37:16 +0800
Subject: [PATCH 11/19] [skip ci]


From 77b6aa0dc881ad6f8e77333b33736060432db976 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E5=BD=AD=E4=BA=8E=E6=96=8C?= <1931127624@qq.com>
Date: Sat, 6 Jun 2020 12:42:10 +0800
Subject: [PATCH 12/19] [skip ci] Apply suggestions from code review

Co-authored-by: Yuanming Hu <yuanming-hu@users.noreply.github.com>
---
 python/taichi/lang/meta.py  | 4 ++--
 python/taichi/misc/gui.py   | 4 ++--
 python/taichi/misc/image.py | 4 ++--
 3 files changed, 6 insertions(+), 6 deletions(-)

diff --git a/python/taichi/lang/meta.py b/python/taichi/lang/meta.py
index 74b411a23c380..ed2350ff75ec3 100644
--- a/python/taichi/lang/meta.py
+++ b/python/taichi/lang/meta.py
@@ -18,9 +18,9 @@ def tensor_to_ext_arr(tensor: ti.template(), arr: ti.ext_arr()):
 @ti.func
 def cook_image_type(x):
     x = ti.cast(x, ti.f32)
-    # Issue #1000, don't know why win GUI doesn't do clamp for us
+    # Issue #1000, we don't know why Windows GUI does not do clamping for us
     if ti.static(ti.get_os_name() == 'win'):
-        x = ti.min(1, ti.max(0, x))
+        x = min(1, max(0, x))
     return x
 
 
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index b0ebbc3653d5c..417ea7c297bee 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -32,7 +32,7 @@ def __init__(self, name, res=512, background_color=0x0):
         if isinstance(res, numbers.Number):
             res = (res, res)
         self.res = res
-        # The GUI canvas uses RGBA for storage, therefore we needs NxMx4 for a image
+        # The GUI canvas uses RGBA for storage, therefore we need NxMx4 for an image.
         self.img = np.ascontiguousarray(np.zeros(self.res + (4, ), np.float32))
         self.core = ti.core.GUI(name, ti.veci(*res))
         self.canvas = self.core.get_canvas()
@@ -110,7 +110,7 @@ def cook_image(img):
 
         else:
             raise ValueError(
-                f"GUI.set_image only takes Taichi tensor or NumPy array, not {type(img)}"
+                f"GUI.set_image only takes a Taichi tensor or NumPy array, not {type(img)}"
             )
 
         self.core.set_img(self.img.ctypes.data)
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 3ae20e210ec91..88e3b7b049750 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -26,11 +26,11 @@ def imread(filename, channels=0):
     return img.swapaxes(0, 1)[:, ::-1, :]
 
 
-def imshow(img, winname='Taichi'):
+def imshow(img, window_name='Taichi'):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
     assert len(img.shape) in [2, 3]
-    gui = ti.GUI(winname, res=img.shape[:2])
+    gui = ti.GUI(window_name, res=img.shape[:2])
     while not gui.get_event(ti.GUI.ESCAPE):
         gui.set_image(img)
         gui.show()

From e7f59ce9beb0b21ffeadfa20c5a936b590a30798 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 12:43:34 +0800
Subject: [PATCH 13/19] [skip travis] self.cook_image

---
 python/taichi/misc/gui.py | 60 +++++++++++++++++++--------------------
 1 file changed, 30 insertions(+), 30 deletions(-)

diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index 417ea7c297bee..bd6fc4755198a 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -49,41 +49,41 @@ def clear(self, color=None):
             color = self.background_color
         self.canvas.clear(color)
 
+    def cook_image(self, img):
+        if img.dtype in [np.uint8, np.uint16, np.uint32, np.uint64]:
+            img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
+        elif img.dtype in [np.float32, np.float64]:
+            img = img.astype(np.float32)
+            from .util import get_os_name
+            if img.dtype == np.float32 and get_os_name() == 'win':
+                img = np.clip(img, 0, 1)
+        else:
+            raise ValueError(
+                f'Data type {img.dtype} not supported in GUI.set_image')
+
+        if len(img.shape) == 2:
+            img = img[..., None]
+        if img.shape[2] == 1:
+            img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
+        if img.shape[2] == 3:
+            img = np.concatenate([
+                img,
+                np.zeros(shape=(img.shape[0], img.shape[1], 1),
+                         dtype=np.float32)
+            ],
+                                 axis=2)
+        assert img.shape[:
+                         2] == self.res, "Image resolution does not match GUI resolution"
+        return np.ascontiguousarray(img)
+
     def set_image(self, img):
         import numpy as np
         import taichi as ti
 
-        def cook_image(img):
-            if img.dtype in [np.uint8, np.uint16, np.uint32, np.uint64]:
-                img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
-            elif img.dtype in [np.float32, np.float64]:
-                img = img.astype(np.float32)
-                from .util import get_os_name
-                if img.dtype == np.float32 and get_os_name() == 'win':
-                    img = np.clip(img, 0, 1)
-            else:
-                raise ValueError(
-                    f'Data type {img.dtype} not supported in GUI.set_image')
-
-            if len(img.shape) == 2:
-                img = img[..., None]
-            if img.shape[2] == 1:
-                img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
-            if img.shape[2] == 3:
-                img = np.concatenate([
-                    img,
-                    np.zeros(shape=(img.shape[0], img.shape[1], 1),
-                             dtype=np.float32)
-                ],
-                                     axis=2)
-            assert img.shape[:
-                             2] == self.res, "Image resolution does not match GUI resolution"
-            return np.ascontiguousarray(img)
-
         if isinstance(img, ti.Expr):
             if ti.core.is_integral(img.data_type()):
                 # image of uint is not optimized by xxx_to_image
-                self.img = cook_image(img.to_numpy())
+                self.img = self.cook_image(img.to_numpy())
             else:
                 assert img.shape(
                 ) == self.res, "Image resolution does not match GUI resolution"
@@ -93,7 +93,7 @@ def cook_image(img):
 
         elif isinstance(img, ti.Matrix):
             if ti.core.is_integral(img.data_type()):
-                self.img = cook_image(img.to_numpy())
+                self.img = self.cook_image(img.to_numpy())
             else:
                 assert img.shape(
                 ) == self.res, "Image resolution does not match GUI resolution"
@@ -106,7 +106,7 @@ def cook_image(img):
                 ti.sync()
 
         elif isinstance(img, np.ndarray):
-            self.img = cook_image(img)
+            self.img = self.cook_image(img)
 
         else:
             raise ValueError(

From 686d8081be452c1e415b0a257811fa265264cea1 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 12:49:05 +0800
Subject: [PATCH 14/19] [skip travis] fix torch test

---
 python/taichi/lang/matrix.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/python/taichi/lang/matrix.py b/python/taichi/lang/matrix.py
index bf34fb8cb7423..45c212b6e70c0 100644
--- a/python/taichi/lang/matrix.py
+++ b/python/taichi/lang/matrix.py
@@ -3,7 +3,7 @@
 import copy
 import numbers
 import numpy as np
-from .util import taichi_scope, python_scope, deprecated, to_numpy_type
+from .util import taichi_scope, python_scope, deprecated, to_numpy_type, to_pytorch_type
 from .common_ops import TaichiOperations
 from collections.abc import Iterable
 

From 0d46229be5cdaf069067cc079a08fd41dea69c2a Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 12:50:36 +0800
Subject: [PATCH 15/19] [skip ci] better message

---
 python/taichi/misc/image.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 88e3b7b049750..73a956981dde2 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -5,7 +5,7 @@
 def imwrite(img, filename):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
-    assert len(img.shape) in [2, 3]
+    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
     resx, resy = img.shape[:2]
     if len(img.shape) == 3:
         comp = img.shape[2]
@@ -29,7 +29,7 @@ def imread(filename, channels=0):
 def imshow(img, window_name='Taichi'):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
-    assert len(img.shape) in [2, 3]
+    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
     gui = ti.GUI(window_name, res=img.shape[:2])
     while not gui.get_event(ti.GUI.ESCAPE):
         gui.set_image(img)

From 3d34e110992e6a4b8c5ed6b08bea30404d356132 Mon Sep 17 00:00:00 2001
From: Taichi Gardener <taichigardener@gmail.com>
Date: Sat, 6 Jun 2020 00:51:03 -0400
Subject: [PATCH 16/19] [skip ci] enforce code format

---
 python/taichi/misc/image.py | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 73a956981dde2..9f2db5cbc4ebd 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -5,7 +5,8 @@
 def imwrite(img, filename):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
-    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
+    assert len(img.shape) in [2,
+                              3], "Image must be either RGB/RGBA or greyscale"
     resx, resy = img.shape[:2]
     if len(img.shape) == 3:
         comp = img.shape[2]
@@ -29,7 +30,8 @@ def imread(filename, channels=0):
 def imshow(img, window_name='Taichi'):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
-    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
+    assert len(img.shape) in [2,
+                              3], "Image must be either RGB/RGBA or greyscale"
     gui = ti.GUI(window_name, res=img.shape[:2])
     while not gui.get_event(ti.GUI.ESCAPE):
         gui.set_image(img)

From cd3e1b43b5a3aa2342e29abda75c5dd392d6ba89 Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 16:32:18 +0800
Subject: [PATCH 17/19] fix win32 not clamping

---
 examples/image_io.py       | 24 ------------------------
 python/taichi/lang/meta.py |  8 ++++----
 python/taichi/misc/gui.py  |  2 --
 taichi/gui/win32.cpp       |  7 ++++---
 taichi/gui/x11.cpp         |  2 +-
 5 files changed, 9 insertions(+), 34 deletions(-)
 delete mode 100644 examples/image_io.py

diff --git a/examples/image_io.py b/examples/image_io.py
deleted file mode 100644
index 11b8aa440f01d..0000000000000
--- a/examples/image_io.py
+++ /dev/null
@@ -1,24 +0,0 @@
-import taichi as ti
-import os
-
-pixel = ti.var(ti.u8, shape=(512, 512, 3))
-
-
-@ti.kernel
-def paint():
-    for I in ti.grouped(pixel):
-        pixel[I] = ti.random() * 255
-
-
-paint()
-pixel = pixel.to_numpy()
-ti.imshow(pixel, 'Random Generated')
-for ext in ['bmp', 'png', 'jpg']:
-    fn = 'taichi-example-random-img.' + ext
-    ti.imwrite(pixel, fn)
-    pixel_r = ti.imread(fn)
-    if ext != 'jpg':
-        assert (pixel_r == pixel).all()
-    else:
-        ti.imshow(pixel_r, 'JPEG Read Result')
-    os.remove(fn)
diff --git a/python/taichi/lang/meta.py b/python/taichi/lang/meta.py
index ed2350ff75ec3..ae38f5f7b5a5f 100644
--- a/python/taichi/lang/meta.py
+++ b/python/taichi/lang/meta.py
@@ -18,16 +18,16 @@ def tensor_to_ext_arr(tensor: ti.template(), arr: ti.ext_arr()):
 @ti.func
 def cook_image_type(x):
     x = ti.cast(x, ti.f32)
-    # Issue #1000, we don't know why Windows GUI does not do clamping for us
-    if ti.static(ti.get_os_name() == 'win'):
-        x = min(1, max(0, x))
     return x
 
 
 @ti.kernel
 def tensor_to_image(tensor: ti.template(), arr: ti.ext_arr()):
     for I in ti.grouped(tensor):
-        arr[I] = cook_image_type(tensor[I])
+        t = cook_image_type(tensor[I])
+        arr[I, 0] = t
+        arr[I, 1] = t
+        arr[I, 2] = t
 
 
 @ti.kernel
diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index bd6fc4755198a..aee8603341fab 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -55,8 +55,6 @@ def cook_image(self, img):
         elif img.dtype in [np.float32, np.float64]:
             img = img.astype(np.float32)
             from .util import get_os_name
-            if img.dtype == np.float32 and get_os_name() == 'win':
-                img = np.clip(img, 0, 1)
         else:
             raise ValueError(
                 f'Data type {img.dtype} not supported in GUI.set_image')
diff --git a/taichi/gui/win32.cpp b/taichi/gui/win32.cpp
index eb3e045c1de0b..0c24ead17c5fc 100644
--- a/taichi/gui/win32.cpp
+++ b/taichi/gui/win32.cpp
@@ -180,9 +180,10 @@ void GUI::redraw() {
   for (int i = 0; i < width; i++) {
     for (int j = 0; j < height; j++) {
       auto c = reinterpret_cast<unsigned char *>(data + (j * width) + i);
-      c[0] = (unsigned char)(canvas->img[i][height - j - 1][2] * 255.0_f);
-      c[1] = (unsigned char)(canvas->img[i][height - j - 1][1] * 255.0_f);
-      c[2] = (unsigned char)(canvas->img[i][height - j - 1][0] * 255.0_f);
+      auto d = canvas->img[i][height - j - 1];
+      c[0] = uint8(clamp(int(d[2] * 255.0_f), 0, 255));
+      c[1] = uint8(clamp(int(d[1] * 255.0_f), 0, 255));
+      c[2] = uint8(clamp(int(d[0] * 255.0_f), 0, 255));
       c[3] = 0;
     }
   }
diff --git a/taichi/gui/x11.cpp b/taichi/gui/x11.cpp
index 84aaf919ef174..b8f2e0a02dc69 100644
--- a/taichi/gui/x11.cpp
+++ b/taichi/gui/x11.cpp
@@ -35,7 +35,7 @@ class CXImage {
         *p++ = uint8(clamp(int(c[2] * 255.0_f), 0, 255));
         *p++ = uint8(clamp(int(c[1] * 255.0_f), 0, 255));
         *p++ = uint8(clamp(int(c[0] * 255.0_f), 0, 255));
-        *p++ = uint8(clamp(int(c[3] * 255.0_f), 0, 255));
+        *p++ = 0;
       }
     }
   }

From 79e4883017bf51d3a08d10206911fe7a7e8df88c Mon Sep 17 00:00:00 2001
From: archibate <17721388340@163.com>
Date: Sat, 6 Jun 2020 17:37:24 +0800
Subject: [PATCH 18/19] [skip ci] fix imread/write

---
 python/taichi/misc/gui.py   | 17 +++++++----------
 python/taichi/misc/image.py | 22 +++++++++++++++++-----
 2 files changed, 24 insertions(+), 15 deletions(-)

diff --git a/python/taichi/misc/gui.py b/python/taichi/misc/gui.py
index aee8603341fab..0205ccd651a26 100644
--- a/python/taichi/misc/gui.py
+++ b/python/taichi/misc/gui.py
@@ -54,24 +54,21 @@ def cook_image(self, img):
             img = img.astype(np.float32) * (1 / np.iinfo(img.dtype).max)
         elif img.dtype in [np.float32, np.float64]:
             img = img.astype(np.float32)
-            from .util import get_os_name
         else:
             raise ValueError(
                 f'Data type {img.dtype} not supported in GUI.set_image')
 
         if len(img.shape) == 2:
             img = img[..., None]
+
         if img.shape[2] == 1:
-            img = img + np.zeros(shape=(1, 1, 4), dtype=np.float32)
+            img = img + np.zeros((1, 1, 4), np.float32)
         if img.shape[2] == 3:
-            img = np.concatenate([
-                img,
-                np.zeros(shape=(img.shape[0], img.shape[1], 1),
-                         dtype=np.float32)
-            ],
-                                 axis=2)
-        assert img.shape[:
-                         2] == self.res, "Image resolution does not match GUI resolution"
+            zeros = np.zeros((img.shape[0], img.shape[1], 1), np.float32)
+            img = np.concatenate([img, zeros], axis=2)
+
+        res = img.shape[:2]
+        assert res == self.res, "Image resolution does not match GUI resolution"
         return np.ascontiguousarray(img)
 
     def set_image(self, img):
diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index 9f2db5cbc4ebd..c1370b7aea9e8 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -5,13 +5,24 @@
 def imwrite(img, filename):
     if not isinstance(img, np.ndarray):
         img = img.to_numpy()
-    assert len(img.shape) in [2,
-                              3], "Image must be either RGB/RGBA or greyscale"
+
+    if img.dtype in [np.uint16, np.uint32, np.uint64]:
+        img = (img // (np.iinfo(img.dtype).max / 256)).astype(np.uint8)
+    elif img.dtype in [np.float32, np.float64]:
+        img = (np.clip(img, 0, 1) * 255.0 + 0.5).astype(np.uint8)
+    elif img.dtype != np.uint8:
+        raise ValueError(
+            f'Data type {img.dtype} not supported in ti.imwrite')
+
+    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
+    assert img.shape[2] in [1, 3, 4], "Image must be either RGB/RGBA or greyscale"
+
     resx, resy = img.shape[:2]
-    if len(img.shape) == 3:
-        comp = img.shape[2]
-    else:
+    if len(img.shape) == 2:
         comp = 1
+    else:
+        comp = img.shape[2]
+
     img = np.ascontiguousarray(img.swapaxes(0, 1)[::-1, :, :])
     ptr = img.ctypes.data
     ti.core.imwrite(filename, ptr, resx, resy, comp)
@@ -33,6 +44,7 @@ def imshow(img, window_name='Taichi'):
     assert len(img.shape) in [2,
                               3], "Image must be either RGB/RGBA or greyscale"
     gui = ti.GUI(window_name, res=img.shape[:2])
+    img = gui.cook_image(img)
     while not gui.get_event(ti.GUI.ESCAPE):
         gui.set_image(img)
         gui.show()

From 5d9a5ae127d342c629477ef2db65d9b20d816df9 Mon Sep 17 00:00:00 2001
From: Taichi Gardener <taichigardener@gmail.com>
Date: Sat, 6 Jun 2020 05:38:15 -0400
Subject: [PATCH 19/19] [skip ci] enforce code format

---
 python/taichi/misc/image.py | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/python/taichi/misc/image.py b/python/taichi/misc/image.py
index c1370b7aea9e8..09cbb1fd98973 100644
--- a/python/taichi/misc/image.py
+++ b/python/taichi/misc/image.py
@@ -11,11 +11,12 @@ def imwrite(img, filename):
     elif img.dtype in [np.float32, np.float64]:
         img = (np.clip(img, 0, 1) * 255.0 + 0.5).astype(np.uint8)
     elif img.dtype != np.uint8:
-        raise ValueError(
-            f'Data type {img.dtype} not supported in ti.imwrite')
+        raise ValueError(f'Data type {img.dtype} not supported in ti.imwrite')
 
-    assert len(img.shape) in [2, 3], "Image must be either RGB/RGBA or greyscale"
-    assert img.shape[2] in [1, 3, 4], "Image must be either RGB/RGBA or greyscale"
+    assert len(img.shape) in [2,
+                              3], "Image must be either RGB/RGBA or greyscale"
+    assert img.shape[2] in [1, 3,
+                            4], "Image must be either RGB/RGBA or greyscale"
 
     resx, resy = img.shape[:2]
     if len(img.shape) == 2: