Rules to avoid common extended inline assembly mistakes

GCC and Clang inline assembly is an interface between high and low level programming languages. It is subtle and treacherous. Many are ensnared in its traps, usually unknowingly. As such, the asm keyword is essentially the unsafe keyword of C and C++. Nearly every inline assembly tutorial, including the awful ibilio page at the top of search engines for decades, propagate fundamental, serious mistakes, and most examples are incorrect. The dangerous part is that the examples usually produce the expected results! The situation is dire. This article isn’t a tutorial, but basic rules to avoid the most common mistakes, or to spot them in code review.

• c
• cpp

Everything I've learned so far about running local LLMs

This article was discussed on Hacker News.

Over the past month I’ve been exploring the rapidly evolving world of Large Language Models (LLM). It’s now accessible enough to run a LLM on a Raspberry Pi smarter than the original ChatGPT (November 2022). A modest desktop or laptop supports even smarter AI. It’s also private, offline, unlimited, and registration-free. The technology is improving at breakneck speed, and information is outdated in a matter of months. This article snapshots my practical, hands-on knowledge and experiences — information I wish I had when starting. Keep in mind that I’m a LLM layman, I have no novel insights to share, and it’s likely I’ve misunderstood certain aspects. In a year this article will mostly be a historical footnote, which is simultaneously exciting and scary.

• ai
• rant
• tutorial

Windows dynamic linking depends on the active code page

Windows paths have been WTF-16-encoded for decades, but module names in the import tables of Portable Executable are octets. If a name contains values beyond ASCII — technically out of spec — then the dynamic linker must somehow decode those octets into Unicode in order to construct a lookup path. There are multiple ways this could be done, and the most obvious is the process’s active code page (ACP), which is exactly what happens. As a consequence, the specific DLL loaded by the linker may depend on the system code page. In this article I’ll contrive such a situation.

• c
• win32

Slim Reader/Writer Locks are neato

I’m 18 years late, but Slim Reader/Writer Locks have a fantastic interface: pointer-sized (“slim”), zero-initialized, and non-allocating. Lacking cleanup, they compose naturally with arena allocation. Sounds like a futex? That’s because they’re built on futexes introduced at the same time. They’re also complemented by condition variables with the same desirable properties. My only quibble is that slim locks could easily have been 32-bit objects, but it hardly matters. This article, while treating Win32 as a foreign interface, discusses a paper-thin C++ wrapper interface around lock and condition variables, in my own style.

• win32
• cpp

Giving C++ std::regex a C makeover

Suppose you’re working in C using one of the major toolchains — that is, it’s mainly a C++ implementation — and you need regular expressions. You could integrate a library, but there’s a regex implementation in the C++ standard library included with your compiler, just within reach. As a resourceful engineer, using an asset already in hand seems prudent. But it’s a C++ interface, and you’re using C instead of C++ for a reason, perhaps to avoid dealing with C++. Have no worries. This article is about wrapping std::regex in a tidy C interface which not only hides all the C++ machinery, but utterly tames it. It’s not so much practical as a potpourri of interesting techniques.

• c
• cpp
• win32

Deep list copy: More than meets the eye

I recently came across a take-home C programming test which had more depth and complexity than I suspect the interviewer intended. While considering it, I also came up with a novel, or at least unconventional, solution. The problem is to deep copy a linked list where each node references a random list element in addition to usual linkage — similar to LeetCode problem 138. This reference is one of identity rather than value, which has murky consequences.

• c

Symbol inspection tools for w64devkit: vc++filt and peports

I introduced two new tools to w64devkit, vc++filt and peports (pronounced like purports), which aid manual symbol inspection and complement one another. As of this writing, the latter is not yet in a release, but it’s feature-complete and trivial to build if you wanted to try it out early. This article explains the motivation and purpose for each.

Arenas and the almighty concatenation operator

I continue to streamline an arena-based paradigm, and stumbled upon a concise technique for dynamic growth — an efficient, generic “concatenate anything to anything” within an arena built atop a core of 9-ish lines of code. The key insight originated from a reader suggestion about dynamic arrays. The subject of concatenation can be a string, dynamic array, or even something else. The “system” is extensible, and especially useful for path handling.

• c
• cpp

Guidelines for computing sizes and subscripts

Occasionally we need to compute the size of an object that does not yet exist, or a subscript that may fall out of bounds. It’s easy to miss the edge cases where results overflow, creating a nasty, subtle bug, even in the presence of type safety. Ideally such computations happen in specialized code, such as inside an allocator (calloc, reallocarray) and not outside by the allocatee (i.e. malloc). Mitigations exist with different trade-offs: arbitrary precision, or using a wider fixed integer — i.e. 128-bit integers on 64-bit hosts. In the typical case, working only with fixed size-type integers, I’ve come up with a set of guidelines to avoid overflows in the edge cases.

• c
• cpp
• go
• tutorial

Speculations on arenas and custom strings in C++

My techniques with arena allocation and strings are oriented around C. I’m always looking for a better way, and lately I’ve been experimenting with building them using C++ features. What are the trade-offs? Are the benefits worth the costs? In this article I lay out my goals, review implementation possibilities, and discuss my findings. Following along will require familiarity with those previous two articles.

• c
• cpp