Taichi is a statically typed programming language, meaning that the type of a variable in the Taichi scope is determined at compile time. This means that once a variable has been declared, it cannot be assigned a value of a different type.
x = 1 # x is the integer 1
x = 3.14 # x is an integer, so the value 3.14 is cast to 3 and x takes the value 3
x = ti.Vector([1, 1]) # Error!
- Line 3:
xis an integer since it is assigned an integer value when it is declared for the first time.
- Line 4:
xis reassigned a floating-point number 3.14. However, the value of
xis 3 instead of 3.14. This is because 3.14 is automatically cast to an integer 3 to match the type of
- Line 5: The system throws an error, because
ti.Vectorcannot be cast into an integer.
ti.types module in Taichi defines all of the supported data types. These data types are categorized into two groups: primitive and compound.
- Primitive types encompass commonly utilized numerical data types, such as
- Compound types, on the other hand, encompass array-like or struct-like data types, including
ti.types.struct. These types are composed of multiple members, which can be primitive or other compound types.
The primitive data types in Taichi are scalars, which are the smallest units that make up compound data types. These types are denoted with a letter indicating their category, followed by a number indicating their precision in bits. The category letter can be
i for signed integers,
u for unsigned integers, or
f for floating-point numbers. The precision bits can be 8, 16, 32, or 64. The two most commonly used primitive types are:
i32: 32 位符号整型
f32: 32 位浮点型
The support of Taichi's primitive types by various backends may vary. Consult the following table for detailed information, and note that some backends may require extensions for complete support of a specific primitive type.
初始化 Taichi 运行时，Taichi 会自动使用以下数据类型作为默认基本数据类型：
ti.init(default_ip=ti.i64) # Sets the default integer type to ti.i64
ti.init(default_fp=ti.f64) # Sets the default floating-point type to ti.f64
The numeric literals in Taichi's scope have default integer or floating-point types. For example, if the default floating-point type is
ti.f32, a numeric literal 3.14159265358979 will be cast to a 32-bit floating-point number with a precision of approximately seven decimal digits. To ensure high precision in applications such as engineering simulations, it is recommended to set
In Taichi scope, the two names
float serve as aliases for the default integer and floating-point types, respectively. These default types can be changed using the configuration option
default_fp. For instance, setting the
f64 would allow you to use
int as an alias for
float as an alias for
f64 in your code.
def example_cast() -> int: # the returned type is ti.i64
x = 3.14 # x is of ti.f64 type
y = int(x) # equivalent to ti.i64(x)
Furthermore, in the Python scope, when declaring Taichi's data containers using
ti.ndarray, these two names also serve as aliases for the default integer and floating-point types. 例如：
x = ti.field(float, 5)
# Is equivalent to:
x = ti.field(ti.f64, 5)
However, when using
float outside of Taichi's data containers in regular Python code, they refer to their standard meaning as built-in functions and not aliases for Taichi's
default_fp. Therefore, in Python scope and outside of Taichi's data containers,
float have their standard meaning as built-in functions.
x = numpy.array([1, 2, 3, 4], dtype=int) # NumPy's int64 type
y = int(3.14) # Python's built-in int type
As mentioned at the beginning of this document, the type of a variable in the Taichi scope is determined at compile time, meaning that it is statically typed. The Taichi compiler performs type checking at compile time and therefore, once a variable is declared, you cannot assign to it a value of a different type. However, in certain situations, you may need to switch to a different data type due to the unavailability of the original type for an assignment or calculation. In these cases, you must perform an explicit type casting.
ti.cast()function allows you to convert a given value to a specific target type. For instance, you can use
ti.cast(x, float)to transform a variable
xinto a floating-point type.
a = 3.14
b = ti.cast(a, ti.i32) # 3
c = ti.cast(b, ti.f32) # 3.0
As of Taichi v1.1.0, the capability to perform type casting on scalar variables has been introduced using primitive types such as
ti.i64. This allows you to convert scalar variables to different scalar types with ease.
a = 3.14
x = int(a) # 3
y = float(a) # 3.14
z = ti.i32(a) # 3
w = ti.f64(a) # 3.14
Implicit type casting occurs when a value is placed or assigned where a different data type is expected.
As a general principle, implicit type casting can be a significant source of bugs. As such, Taichi strongly discourages the use of this mechanism and recommends that you explicitly specify the desired data types for all variables and operations.
In Taichi, implicit type casting can occur during binary operations or assignments. The casting rules are implemented specifically for Taichi and are slightly different from those for the C programming language. These rules are prioritized as follows:
i32 + f32 -> f32
i16 + f16 -> f16
低精度位 + 高精度位 -> 高精度位
i16 + i32 -> i32
f16 + f32 -> f32
u8 + u16 -> u16
带符号整数 + 无符号整数 -> 无符号整数
u32 + i32 -> u32
u8 + i8 -> u8
u8 + i16 -> i16(当规则#2 与规则#3 冲突时，规则#2 适用。)
f16 + i32 -> f16(当规则#1 与规则#2 冲突时，规则#1 适用。)
- 位移运算返回 lhs (左侧) 数据类型：
u8 << i32 -> u8
i16 << i8 -> i16
In Taichi, implicit type casting is performed when assigning a value to a variable with a different data type. In cases where the value has a higher precision than the target variable, a warning indicating potential precision loss will be displayed.
Example 1: The variable
ais initialized with the data type
float, and then immediately reassigned the value 1. This reassignment implicitly converts the data type of 1 from
floatwithout generating a warning.
a = 3.14
a = 1
print(a) # 1.0
Example 2: The variable
ais initialized with the data type
int, and is immediately reassigned the value 3.14. This reassignment implicitly converts the data type of 3.14 from
int, which results in a loss of precision due to
inthaving a lower precision than
float. As a result, a warning is generated.
a = 1
a = 3.14
print(a) # 3
复合类型是用户自定义的数据类型，由多个元素组成。 支持的复合类型包括向量、矩阵、ndarray 和结构体：
ndarray 类型在另一篇文档 与外部数组交互 中讨论。
vec4d = ti.types.vector(4, ti.f64) # a 64-bit floating-point 4D vector type
mat4x3i = ti.types.matrix(4, 3, int) # a 4x3 integer matrix type
You can utilize the customized compound types to instantiate vectors and matrices, as well as annotate the data types of function arguments and struct members. For instance:
v = vec4d(1, 2, 3, 4) # Create a vector instance, here v = [1.0 2.0 3.0 4.0]
def length(w: vec4d): # vec4d as type hint
You can use the function
ti.types.struct() to create a struct type, which can be utilized to represent a sphere in 3D space, abstracted by its center and radius. To achieve this, you can call
ti.types.struct() to create two higher-level compound types:
sphere_type, respectively. These types can then be used as templates to initialize two local variables,
sphere2, to represent two instances of spheres.
# Define a compound type vec3 to represent a sphere's center
vec3 = ti.types.vector(3, float)
# Define a compound type sphere_type to represent a sphere
sphere_type = ti.types.struct(center=vec3, radius=float)
# Initialize sphere1, whose center is at [0,0,0] and whose radius is 1.0
sphere1 = sphere_type(center=vec3([0, 0, 0]), radius=1.0)
# Initialize sphere2, whose center is at [1,1,1] and whose radius is 1.0
sphere2 = sphere_type(center=vec3([1, 1, 1]), radius=1.0)
When defining a struct with numerous members, the use of
ti.types.struct can lead to cluttered and unorganized code. Taichi provides a more elegant solution with the
@ti.dataclass decorator, which acts as a lightweight wrapper around the struct type.
The code above accomplishes the same task as the following line, however it offers improved comprehensibility:
Sphere = ti.types.struct(center=vec3, radius=float)
Another benefit of utilizing the
@ti.dataclass over the
ti.types.struct is the ability to define member functions within a dataclass, enabling object-oriented programming (OOP) capabilities. For more information on the topic of objective data-oriented programming, refer to the objective data-oriented programming documentation.
In Taichi, creating instances of vector, matrix, or struct compound types can be achieved by directly calling the type, similar to how it is done with any other data type.
As of Taichi v1.1.0, multiple options are available for initializing instances of structs or dataclasses. The conventional method of calling a compound type directly still holds true. In addition, the following alternatives are also supported:
- Pass positional arguments to the struct, in the order in which the members are defined.
- Utilize keyword arguments to set the specific struct members.
- Members that are not specified will be automatically set to zero.
# The definition above is equivalent to
#Ray = ti.types.struct(ro=vec3, rd=vec3, t=float)
# Use positional arguments to set struct members in order
ray = Ray(vec3(0), vec3(1, 0, 0), 1.0)
# ro is set to vec3(0) and t will be set to 0
ray = Ray(vec3(0), rd=vec3(1, 0, 0))
# both ro and rd are set to vec3(0)
ray = Ray(t=1.0)
# ro is set to vec3(1), rd=vec3(0) and t=0.0
ray = Ray(1)
# All members are set to 0
ray = Ray()
You can create vectors, matrices, and structs using GLSL-like broadcast syntax because their shapes are already known.
For now, the only compound data types that support type casting in Taichi are vectors and matrices. When casting the type of a vector or matrix, it is performed element-wise, resulting in the creation of new vectors and matrices.
u = ti.Vector([2.3, 4.7])
v = int(u) # ti.Vector([2, 4])
# If you are using ti.i32 as default_ip, this is equivalent to:
v = ti.cast(u, ti.i32) # ti.Vector([2, 4])