3 ms·

GCC alread has this for x64, I thought. https://gcc.gnu.org/onlinedocs/gcc/_005f_005fint128.html https://gcc.gnu.org/onlinedocs/gcc/_005f_005fint128.html RISC

GCC alread has this for x64, I thought. https://gcc.gnu.org/onlinedocs/gcc/_005f_005fint128.html https://gcc.gnu.org/onlinedocs/gcc/_005f_005fint128.html RISC-V has no carry bit and this whole thing becomes awkward. I am under the impression that boost::multiprecision has specialized templates for 128 and 256 bit math, but maybe I'm wrong. In practice when I've wanted extended precision, I've just used GMP or a language with bignums. I would expect the best x86 machine code for many 128 bit operations would use XMM instructions, no? But I haven't investigated.

stackghost 9mo ago
>GMP When I was a teenager I downloaded some pirated games and reverse-engineered the installer/unpacker and discovered it used UHARC, which seemed to be almost magic in how well it could compress data compared to winzip. Knowing absolutely nothing about compression algorithms or information theory, I decided I'd write my own compression algorithm, as one does. I reasoned that since any computer value can be represented as a binary number, you could store the exponents to which you'd raise 2 and then just write those exponents to a file. Voila, lossless compression, can't believe nobody thought of this before. I used GMP so that I could "compress" arbitrarily large files (which would require handling arbitrarily large exponents). Except of course binary bits are already optimal for dense data, so my "compression" algorithm made things orders of magnitude larger, not smaller. But it was fast! All that to say GMP is a great library. Performant, and even an absolute moron like me could figure out how to use it in the days before ChatGPT.
- direwolf20 9mo ago
  Fool! You're supposed to subtract 1 from the number, and drop leading zeroes. Then you can compress the file recursively until it's 0 bytes!
- mnahkies 9mo ago
  I had a similar idea as a teenager - calculate md5 hash and store that plus a hint/offset to then brute force the original content. I had dial up and wanted a more practical way to get large files. Anyway I emailed the Winrar developers about my idea and they politely explained why they didn't think it was feasible (appreciate they even took the time to respond!)
sifar 9mo ago
RVV defines add/sub with carry.
- throwaway81523 9mo ago
  https://rvv-isadoc.readthedocs.io/en/latest/arith_integer.html#vadc https://rvv-isadoc.readthedocs.io/en/latest/arith_integer.ht... Interesting, I didn't know that. I wonder how much hassle the setup for that is. Maybe too much overhead to use for general purpose int128. But, there is also a 128-bit extension that might or might not have implementations.

stackghost 9mo ago

>GMP When I was a teenager I downloaded some pirated games and reverse-engineered the installer/unpacker and discovered it used UHARC, which seemed to be almost magic in how well it could compress data compared to winzip. Knowing absolutely nothing about compression algorithms or information theory, I decided I'd write my own compression algorithm, as one does. I reasoned that since any computer value can be represented as a binary number, you could store the exponents to which you'd raise 2 and then just write those exponents to a file. Voila, lossless compression, can't believe nobody thought of this before. I used GMP so that I could "compress" arbitrarily large files (which would require handling arbitrarily large exponents). Except of course binary bits are already optimal for dense data, so my "compression" algorithm made things orders of magnitude larger, not smaller. But it was fast! All that to say GMP is a great library. Performant, and even an absolute moron like me could figure out how to use it in the days before ChatGPT.

direwolf20 9mo ago
Fool! You're supposed to subtract 1 from the number, and drop leading zeroes. Then you can compress the file recursively until it's 0 bytes!
mnahkies 9mo ago
I had a similar idea as a teenager - calculate md5 hash and store that plus a hint/offset to then brute force the original content. I had dial up and wanted a more practical way to get large files. Anyway I emailed the Winrar developers about my idea and they politely explained why they didn't think it was feasible (appreciate they even took the time to respond!)

sifar 9mo ago

RVV defines add/sub with carry.

throwaway81523 9mo ago
https://rvv-isadoc.readthedocs.io/en/latest/arith_integer.html#vadc https://rvv-isadoc.readthedocs.io/en/latest/arith_integer.ht... Interesting, I didn't know that. I wonder how much hassle the setup for that is. Maybe too much overhead to use for general purpose int128. But, there is also a 128-bit extension that might or might not have implementations.