docs/lang/articles/glossary/glossary.md
https://en.wikipedia.org/wiki/Abstract_syntax_tree
https://en.wikipedia.org/wiki/Ahead-of-time_compilation
https://en.wikipedia.org/wiki/Aliasing_(computing)
https://docs.python.org/3/glossary.html
https://en.wikipedia.org/wiki/AoS_and_SoA
See also structure of arrays.
https://docs.python.org/3/reference/simple_stmts.html#the-assert-statement
https://wiki.osdev.org/Atomic_operation
https://en.wikipedia.org/wiki/Augmented_assignment
https://en.wikipedia.org/wiki/Automatic_differentiation
https://en.wikipedia.org/wiki/Mask_(computing)
https://en.wikipedia.org/wiki/Row-_and_column-major_order
See also row-major order.
A compound type is a user-defined array-like or struct-like data type which comprises multiple members of primitive types or other compound types. Supported compound types in Taichi include vectors, metrics, and structs.
https://www.khronos.org/opengl/wiki/Compute_Shader
A coordinate offset refers to a value added to another base value, which is an element in a Taichi field.
You can use offsets when defining a field to move the field boundaries.
https://en.wikipedia.org/wiki/Data-oriented_design
https://en.wikipedia.org/wiki/Race_condition#Data_race
https://en.wikipedia.org/wiki/Differentiable_programming
https://en.wikipedia.org/wiki/Domain-specific_language
External arrays refer to data containers available in the Python scope.
Taichi supports interaction with the following external arrays - Numpy arrays, PyTorch tensors, and Paddle tensors.
A field is a multi-dimensional array of elements. The elements it accepts can be a scalar, a vector, a matrix, or a struct. It is a global data container provided by Taichi and can be accessed from both the Python scope and the Taichi scope.
The shape of a field is the number of elements in each dimension.
https://en.wikipedia.org/wiki/Shader#Pixel_shaders
https://en.wikipedia.org/wiki/Global_variable
https://developer.nvidia.com/blog/cuda-pro-tip-write-flexible-kernels-grid-stride-loops/
https://en.wikipedia.org/wiki/Imperative_programming
https://en.wikipedia.org/wiki/Instance_(computer_science)
https://en.wikipedia.org/wiki/Intermediate_representation
https://en.wikipedia.org/wiki/Just-in-time_compilation
A kernel is a function decorated with @ti.kernel. A kernel serves as the entry point where Taichi begins to take over the tasks, and it must be called directly by Python code.
https://en.wikipedia.org/wiki/Scope_(computer_science)#Lexical_scope
https://en.wikipedia.org/wiki/Local_variable
https://en.wikipedia.org/wiki/Loop_unrolling
A megakernel is a Taichi kernel that can deal with a large amount of computation to achieve high arithmetic intensity.
Metadata refers to the two fundamental attributes of a Taichi field, i.e., data type and shape.
Use field.dtype and field.shape to retrieve the metadata.
https://en.wikipedia.org/wiki/Metaprogramming
https://en.wikipedia.org/wiki/Object-oriented_programming
https://en.wikipedia.org/wiki/Passive_data_structure
Primitive data types are commonly-used numerical data types from which all other data types are constructed. Supported primitive data types in Taichi include ti.i32 (int32), ti.u8 (uint8), and ti.f64 (float64)
Code outside of the Taichi scope is in the Python scope. The code in the Python scope is native Python and executed by Python's virtual machine, not by Taichi's runtime.
The Python scope corresponds to the host side in CUDA.
https://en.wikipedia.org/wiki/Row-_and_column-major_order
See also coloum-major order.
https://www.khronos.org/opengl/wiki/Shader_Storage_Buffer_Object
A matrix is a two-dimensional data object made of m rows and n columns. If a matrix is comprised of mostly zero values, then it is a sparse matrix.
Taichi provides APIs for sparse matrices.
A static scope is a scope of the argument of ti.static, which is a hint for the compiler to evaluate the argument at compile time.
https://en.wikipedia.org/wiki/Static_single-assignment_form
https://en.wikipedia.org/wiki/AoS_and_SoA
See also array of structures.
A Taichi function is a function decorated with @ti.func.
A Taichi function must be called from inside a kernel or from inside another Taichi function.
The code inside a kernel or a Taichi function is in the Taichi scope. The code in the Taichi scope is compiled by Taichi's runtime and executed in parallel on CPU or GPU devices for high-performance computation.
The Taichi scope corresponds to the device side in CUDA.
Template signatures are what distinguish different instantiations of a kernel template.
For example, The signature of add(x, 42) is (x, ti.i32), which is the same as that of add(x, 1). Therefore, the latter can reuse the previously compiled binary. The signature of add(y, 42) is (y, ti.i32), a different value from the previous signature, hence a new kernel will be instantiated and compiled.
https://en.wikipedia.org/wiki/Thread-local_storage
https://en.wikipedia.org/wiki/Stack_trace