This is a short post that explains how to write a high-performance matrix multiplication program on modern processors. In this tutorial I will use a single core of the Skylake-client CPU with AVX2, but the principles in this post also apply to other processors with different instruction sets (such as AVX512).
Matrix multiplication is a mathematical operation that defines the product of
| // The two main components are the interfaces | |
| // Generic<T, Context> and GenericArg<"identifier"> | |
| // Generic basically structurally replaces types in T that are GenericArg<S> | |
| // for some `S extends keyof Context` with `Context[S]` | |
| // See the test cases for specific uses. | |
| // ====== TESTING | |
| // Pass through for trivial types | |
| type Test00 = Generic<number>; |
Here's a list of mildly interesting things about the C language that I learned mostly by consuming Clang's ASTs. Although surprises are getting sparser, I might continue to update this document over time.
There are many more mildly interesting features of C++, but the language is literally known for being weird, whereas C is usually considered smaller and simpler, so this is (almost) only about C.
struct foo {
struct bar {
int x;
| from dataclasses import dataclass, fields as datafields | |
| from ujson import dumps, loads | |
| # Note: ujson seamlessly serializes dataclasses, unlike stdlib's json | |
| @dataclass | |
| class Point: | |
| x: float | |
| y: float | |
| # Shallow dataclass can be rebuilt from dict/json: |