Comments by "Lawrence D’Oliveiro" (@lawrencedoliveiro9104) on "Low Level" channel.

  1. So has Python. And Python isn’t limited to 18 decimal digits.
    22
  2. Actually, CPUs have been reproducing themselves for decades. Consider, you can no longer do modern CPU design without the help of a computer. And yes, there are legitimate use cases for recursion. I wrote a tool for doing version control on Blender documents. It’s not enough to store the base document in the repo, it might also have external dependencies that also need to be included. Some of these dependencies need to be found by scanning node graphs, which can contain node groups which in turn recursively reference further node graphs. Do you really want to code that with an explicit stack? Recursion is just so much easier to code, and easier to understand.
    9
  3. According to the top500 list, the current fastest supercomputer in the world, RIKEN’s Fugaku, has 7,630,848 cores. Of course, they’re not x86 cores, they’re ARM. And it’s not running Windows, it’s Linux. That might help.
    7
  4. Are these short forms really worth using, though? Often, when you benchmark things, you find that longer, simpler instruction sequences are faster. Which is why RISC architectures have the performance edge that they do.
    6
  5. I suspect that benchmarks will show that simpler instructions are faster.
    4
  6.  @marcossidoruk8033  Microsoft Windows user?
    3
  7. Linux is the most versatile development environment. Even if the target system will not be running Linux, you can still use it for your cross-development platform.
    3
  8.  @ABaumstumpf  RISC absolutely has a performance advantage. Look at the current list of the world’s most powerful supercomputers: you don’t see anything x86 until 5th place.
    3
  9.  @ABaumstumpf  ARM and POWER and the Chinese machine are all RISC. Like I said, the highest that x86 could get was 5th place.
    3
  10. Note that stdint.h defines types which are explicitly known to be 1, 2, 4, 8 bytes long.
    3
  11.  @mikapeltokorpi7671  Not sure why that’s relevant.
    3
  12. It was a performance issue. Those complex instructions were too hard to make fast.
    3
  13.  @brycemw  In other words, these clever CISC-isms are a complete waste of time. This is why we have RISC.
    3
  14.  @marcossidoruk8033  Sounds like a typical Microsoft Windows problem.
    2
  15. Just had a check of a more recent COBOL spec. In the section on numeric literals, for fixed-point ones it says “the implementor shall allow for fixed-point numeric literals of 1 through 31 digits in length”. For floating-point, it says “the significand shall be from 1 to 36 digits in length”, while the exponent has a maximum of 4 digits. So there are your official language limits for literals in COBOL.
    2
  16. Let’s see your tail-recursive version of Ackermann’s Function.
    2
  17.  @brycemw  I doubt Linus Torvalds said any such thing. x86 backward compatibility was basically an albatross around the neck of Intel. If it had “focused on what mattered the most”, it would never have gone down the Itanium dead end, and left the door open for AMD to take the lead with AMD64. I just checked some (early) AMD64 docs, and the REPxx prefixes are listed in the “legacy” category -- not something you would use in new code.
    2
  18. long double will work. Also, this: #include <stdio.h> typedef _Decimal32 /* also available: _Decimal64, _Decimal128 */ real; typedef long double /* no printf support for decimal types (yet), so convert to binary for printing */ ioreal; int main(void) { const real a = 0.1df; const real b = 0.2df; const real c = a + b; fprintf(stdout, "%.17llf + %.17llf = %.17llf\n", (ioreal)a, (ioreal)b, (ioreal)c); return 0; } /*main*/
    1
  19. Oh well ... roll on decimal floats!
    1
  20. Actually, even if you use double instead of long double in that program, the output error is still just one unit in the last place, instead of something like 4 units without the use of decimal floats.
    1
  21.  @groszak1  That’s in binary. It works in decimal.
    1
  22.  @groszak1  It’s been in IEEE754 since 2008.
    1
  23.  @groszak1  It will also come to the standards for major languages like C. Software implementations are available for them. Hardware will follow.
    1
  24. Note also that if you are communicating over AF_UNIX sockets, then the client and server are by definition the same machine.
    1
  25.  @Conenion  The problems scale, up to a point. That’s why a supercomputer needs a high-performance interconnect which makes up such a big part of its cost. If it wasn't for that, a supercomputer would not be much different from, say, a server farm.
    1