Marc B. Reynolds

FWIW: this comments on: 0x80.pl/notesen/2014...

Oodle 2.9.14 and Intel 13th/14th gen CPUs There’s a hardware problem affecting Intel 13th/14th gen CPUs, mostly desktop ones. This made the rounds through the press last year and has been on forums etc. for much longer than that. For…

fgiesen.wordpress.com/2025/05/21/o...

I'll just note that assuming voters are rational is a bad assumption.

Stefan Gustavson has put out an open access book on procgen textures.

liu.diva-portal.org/smash/record...

github.com/stegu/noisei...

test driver that demos "xoroshiro128+" failing PractRand test driver that demos "xoroshiro128+" failing PractRand - test.c

A test driver that demos "xoroshiro128+" failing PractRand in under 10 seconds.

Nothing special here so this is a bit surprising to me.

gist.github.com/Marc-B-Reyno...

As a member of the commonwealth it seems like it isn't even farfetched.

YouTube video by The Current Gary Numan - Metal (Live at The Current)

The original version of Numan's Metal is good...but the post fame version is extra. Just an example if you haven't hit it before.

www.youtube.com/watch?v=tJaY...

Just sayin'

Plus it's in the same vein as "broken clock". RKF is right about healthy diets and exercise. He's just batshit crazy about everything else. Likewise Musk's zoomers might accidently kill a few termites while they burn down the neighborhood.

In these cases sollya would fail to converge without the seed. Other examples are using some other tool (say multiprecision or CAS) to get a first pass.

Using sollya to "repeat" Cecil Hastings' 1955 approximations for atan. Works as an example of using a 'seed' approximation & using sollya to refine. Using sollya to "repeat" Cecil Hastings' 1955 approximations for atan. Works as an example of using a 'seed' approximation & using sollya to refine. - hastings_atan.sollya

Previously mentioned Hastings 1955 #approximation and remembered I have #sollya files for some. This is the first few atan approximations.

I'm sharing because it's an example of using a "seed" approximation to refine the coefficients using sollya.

gist.github.com/Marc-B-Reyno...

On the robustness of double-word addition algorithms We demonstrate that, even when there are moderate overlaps in the inputs of sloppy or accurate double-word addition algorithms in the QD library, these algorithms still guarantee error bounds of $O(u^...

Another #FloatingPoint refinement of pair arithmetic. This time on the so-called "sloppy add"
arxiv.org/abs/2404.05948

AFIK: latest refinement of the accurate building blocks are in this: hal.science/hal-02972245

& weakened constraint versions by Rump & Lang www.tuhh.de/ti3/paper/ru...

together the so-call Lie space and group (to be nerdy) or angle & axis vs. smooth space representations (to be rotation specific).

Euler's is nice because it connects dots. In 2D we have a single basis 'i' and exactly 2 unit bivectors (0,1) & (0,-1). Extending complex to 3D to get quaternions we replace 'i' with the set of all unit bivectors which is a 3D unit sphere & call a specific one 'u' and off we go. log and exp tie

A couple of #FloatingPoint papers that I caught my eye for future reading:

1) A tightening of FastTwoSum knowledge
2) Attempting to generate approximations which account for runtime evaluation errors (w source 4 external proc for Sollya)

1) hal.science/hal-04875749v1
2) hal.science/hal-04709615v1

What is a succinct rank data structure? How does it work? I've heard of a class of data structures called succinct rank data structures. What do these data structures do? What does "succinct" mean here? And how do they work?

I was reminded of this because I ran across a SO answer of his that builds up to using "Four Russians" for the "binary ranking problem"

stackoverflow.com/a/72580831/3...

An old blog post by Keith Schwarz that builds (in an easy IMHO to follow way) sampling of a discrete distribution using "alias tables" and building via Vose's method.

www.keithschwarz.com/darts-dice-c...

It's a big sadface that we didn't get AVX-512 the 256 bit addition at rollout time for consumer targeted CPUs.

but anyway the version of isolate lowest & highest bit of each run and merge reduces to the same as the above two in that cases & is more "clear" to my mind.

FWIW: I'm failing at giving credit here. Warren also has a snippet for the exact match. The first 'e' expression where 'x' is the result of the sieve. The second is a reduction to make the not-and op more obvious & he notes it's 4 op

e = x & ~(x >> 1) & ~(x << 1)
e = x & ~((x >> 1) | (x << 1))

I mostly quit twitter a few years ago but pop in once or twice a month. So anyway toward the end of last year I logged in and saw a post of hers where she was basically normalizing the KKK. The scooby-doo meme popped in my head.

Ah well. Too late. Anyway if a former White Nationalist, Stormfront moderator and member of the Oath Keepers told me I sounded far-right then I'm question myself about what I was saying and maybe ask for clarification.

FWIW: "What would you call an actually Nazi?" is a very funny question to ask to this person.

Compiler Explorer - C (x86-64 clang 19.1.0) //************************************************************************** // mini PRNG (a unparameterized PCG variant) typedef struct { uint64_t state; } prng_t; static inline uint64_t prng_mix_6...

Repeating a respond on twitter. Using the method from hacker's delight: find all matching bit positions + find any exact fits:

gcc.godbolt.org/z/5j81on5Kz

A mini "Happy New Year" blog post on a scheme for "drop rates":

marc-b-reynolds.github.io/math/2025/01...

completed the thought: marc-b-reynolds.github.io/math/2025/01...

Can't credit but (paraphrased): "...while actually the NYT is written by and for middle managers from New Jersey"

Arseny Kapoulkine reminded me that on Zen2 PDEP is microcoded so to be avoided there. My bad.

Compiler Explorer - C (x86-64 clang (trunk)) // intel specific static inline uint64_t bit_scatter_64(uint64_t x, uint64_t m) { return _pdep_u64(x, m); } static inline uint32_t ctz_64(uint64_t x) { return (uint32_t)_tzcnt_u64(x); } // return x...

Two thoughts:

1) Interesting solution but n^th bit operations on hardware supporting bit-scatter can be performed in a couple of cycles.

gcc.godbolt.org/z/njde5bob6

2) If we can do (1) then is the sideline information worth the data size increase?

I need to digest their no bit-scatter solution.

Rope Optimizations, Part 1 - Zed Blog From the Zed Blog: In this episode of Zed Decoded, Thorsten and Antonio walk through the performance optimizations Antonio when working on Zed's Rope data structure before they then pair on adding mor...

A little over a month ago the Zed (editor) team put up a blog post on their text representation

zed.dev/blog/zed-dec...

(A B-tree of 128 byte payloads...interesting) and an optimization related to that which is "find n^th bit set" for precomputed return positions in payload side-line info.