dzaima

31 ms·

241.
by dzaima 1y ago
That's only if they're unpredictable; sure, perhaps on some workload it'll be unpredictable whether the input to sin/cos is grater than 2*pi, but I'm pretty sure on most it'll be nearly-always a "no".
242.
by dzaima 1y ago
There are a bunch of real situations where you can assume the input will be in a small range. And while reducing from [-pi;pi] or [-2*pi;2*pi] or whatever is gonna slow it down somewhat, I'm pretty sure it wouldn't be too signific
243.
by dzaima 1y ago
At pi/2 that approximation gets you 1.0045, i.e. half a percent off, so it's not particularly good at that. (still could be sufficient for some uses though; but not the best achievable even with that performance)
244.
by dzaima 2y ago
Here that can go both ways - SIMD parsing can allow handling arbitrary changes in reasonable time for files below like maybe 100MB (i.e. everything non-insane), whereas incremental parsing can allow handling small changes in truly-arbitrary
245.
by dzaima 2y ago
It may theoretically be false, and probably false in some cases, but (at least temporarily) there are cases (not all! but some) where some of the current C compilers currently will never result in bad behavior even if UB actually happens, f
246.
by dzaima 2y ago
On getting supported extension status - there's a C API spec that could potentially become an option for an OS-agnostic way: https://github.com/riscv-non-isa/riscv-c-api-doc/blob/main/s... . libc alr
247.
by dzaima 2y ago
Fair point on CSRs, though I'd count that as a subset of unsupported/not-yet-specified instructions; pretty sure all of the "reserved"s in the spec are effectively not-yet-defined instructions too, which'll have equ
248.
by dzaima 2y ago
What's all that undefined behavior? Closest I can think of is executing unsupported instructions, but you have to mess up pretty hard for that to happen, and you're not gonna get predictable behavior here anyway (and sane hardware
249.
by dzaima 2y ago
gcc requires that contributors assign copyright to FSF (or at least "certify the Developer Certificate of Origin")[0], so a fork isn't gonna get upstreamed without the approval of the fork's authors. So that limits the b
250.
by dzaima 2y ago
If anything, an emoji is easier to search for, as less things would be using it than a general "warning" (or whatever would make sense, as the emoji'd thing isn't the warning itself). Do have to get a copy from somewhere
251.
by dzaima 2y ago
Directly running wasm/Java bytecode in hardware is just a bad idea unless you somehow have tons of loopless code that'll only be run once; both are stack-based with an unbounded stack size, so each op would have multiple memory lo
252.
by dzaima 2y ago
This isn't actually timing the sorting, but just the (dumb) function f.
253.
by dzaima 2y ago
There's still no time travel in any of those posts; some may require an explanation of "the compiler replaced the UB division with a call to 'bar'" or something, but "calling 'bar'" is still a su
254.
by dzaima 2y ago
Oops, yeah, need to remove the `do_something();` for the loop to be UB. > Well, assume that we are eg looking at the 'sudo' program. In any case, almost any code can become harmful, if memory gets corrupted, and if attackers co
255.
by dzaima 2y ago
`={}` in place of `={0}` is the new option in C23.
256.
by dzaima 2y ago
> There are all kinds of bugs in these compilers all the time. I don’t think you can assert that. That's a separate discussion; generally you need specific conditions in the source code to have compiler bugs affect your code (and fo
257.
by dzaima 2y ago
But UB extends that to "a bug anywhere can potentially cause anything to go wrong", whereas traditional logic bugs require the bug to be functionally related to the bad-if-misused code (for which you can do things like carefully v
258.
by dzaima 2y ago
The parent post has a mistake - it should use an "int" type instead of "signed char", otherwise the implicit promotion done by a+b breaks the desired showcase. That fixed, it wouldn't be an infinite loop if taking t
259.
by dzaima 2y ago
But an implementation wouldn't choose to specify its behavior for an operation to be something that it knows it cannot do efficiently, if the goal is to be efficient, for extremely obvious reasons. If your hardware does twos complement
260.
by dzaima 2y ago
UB doesn't allow arbitrary time travel; any time travel you see should be explainable by happening to run code that undoes what was done previously, and where it's not possible to explain it as such it's a compiler bug (ref:
261.
by dzaima 2y ago
Don't think any language has standardized SIMD that's particularly nice; Highway is probably a quite nice library on C++, though I haven't used it enough to get comfortable. The thing I use for my projects is Singeli[1], a DS
262.
by dzaima 2y ago
For what it's worth, the RISC-V unprivileged spec says: > The behavior upon decoding a reserved instruction is UNSPECIFIED. > Some platforms may require that opcodes reserved for standard use raise an illegal-instruction exceptio
263.
by dzaima 2y ago
There is some work on an OS-agnostic feature detection C API: https://github.com/riscv-non-isa/riscv-c-api-doc/blob/main/s... . Still quite new though, and potentially might change (as it did a month ago)
264.
by dzaima 2y ago
With RISC-V being an open ISA a vendor can freely add some non-RVV thing on the instruction space used by RVV if it doesn't desire to support RVV. And that already exists with Xtheadvector (aka RVV 0.7.1) where thead made hardware with
265.
by dzaima 2y ago
You'd just produce the ≤AVX2-style result; clang will switch to using the mask-returning comparison instrs if the usage ends up being a mask (gcc doesn't do such fancy things, so I guess getting the last bits of speedup from this
266.
by dzaima 2y ago
A simple enough abstraction is to have the mask type have inputs of both element size and count; so for a ≤256-bit product of those you do the homogeneous-bit elements, and for 512-bit you do the packed-bit values. Conversion methods on the
267.
by dzaima 2y ago
It's to an extent even acceptable to crash via segmentation fault (more specifically, doing whatever unsafe exploitable things may come of the source of the issue) if it takes the same amount of privileges to cause the crash as the thi
268.
by dzaima 2y ago
Ah yeah, gather/scatter are indeed a rather problematic thing for autovectorization. That said, with no-alias info (which Rust has a lot of) it's possible: https://rust.godbolt.org/z/zTfo9nxhd . Unfortunately
269.
by dzaima 2y ago
I don't think Rust is particularly problematic here. As long as you don't want to do funky things like use immutable argument memory as temporary scratch space (with you restoring the values afterwards of course), all it means is
270.
by dzaima 2y ago
Seems rustc nightly does successfully vectorize the first sigmoid example: https://rust.godbolt.org/z/e1WYexqWY Also there's progress on making safe intrinsics safe: https://github.com/rust-lang&#x

More ›