ckennelly

4 ms·

1.
by ckennelly 7d ago
With restartable sequences in Linux, there is no need for mutexes or RMW atomics. The mutual exclusion is guaranteed by the kernel.
2.
by ckennelly 7d ago
TCMalloc has different abort handlers for each function: If preempted, we need to know where to restart. If `rseq_cs` is no longer describing a relevant address, that is, the program counter has moved past it, the kernel just ignores it.
3.
by ckennelly 7d ago
TCMalloc TL here :) The code is executed by a single thread. Everything retires in program order. There is no need for a memory barrier between starting the critical section and its body.
4.
Illuminating the processor core with LLVM-mca(abseil.io)
68 points by ckennelly 10mo ago|5 comments
5.
Optimizing Optimization(abseil.io)
2 points by ckennelly 2y ago|0 comments
6.
by ckennelly 4y ago
It depends on your STL implementation's representation of string: https://godbolt.org/z/nMYGYoWbq * libstdc++ has an internal reference to its own address for the SSO. If the moved-from string was referencing its
7.
Beware Microbenchmarks Bearing Gifts(abseil.io)
7 points by ckennelly 4y ago|0 comments
8.
by ckennelly 4y ago
The default setting of max_ptes_none is also problematic. On a stock kernel, it's 511. TCMalloc's docs recommend using max_ptes_none set to 0 for this reason: https://github.com/google/tcmalloc/blob&#x2F
9.
by ckennelly 5y ago
As mentioned in that Stack Overflow post, though, things change again with FSRM (Fast Short Rep Mov). While there are still startup costs, the overhead of calling a function (especially via a PLT) and incurring instruction cache misses is h
10.
by ckennelly 5y ago
https://research.google/pubs/pub50338.pdf goes into more depth on the mem* libc functions and principles for the implementations in llvm libc.
11.
by ckennelly 6y ago
llvm-libc's memcpy is heavily sized optimized for this reason. A dedicated instruction that is fast for all cases, is ideal, though.
12.
by ckennelly 6y ago
You can still get a benefit from sized delete, even without inlining.
13.
by ckennelly 6y ago
Dynamic linking requires calling through the PLT to get to the implementation, so there's a data dependency on determining where the code for it is. Independent of inlining (with LTO, since the C++ language rules requires inhibit optim
14.
by ckennelly 6y ago
The cost in applications really does add up. "Profiling a warehouse-scale computer" (by S. Kanev, et. al.) showed several % of CPU usage allocating and deallocating memory. While a malloc and free does have some data dependencies
15.
by ckennelly 7y ago
This is the modern version--based on Abseil--that we use in production for practically every C++ binary. It includes a number of performance optimizations (per-CPU caches, an improved fast/slow path, and a hugepage-aware backed), alon
16.
Announcing TCMalloc(abseil.io)
9 points by ckennelly 7y ago|3 comments
17.
by ckennelly 8y ago
They're not confidence intervals, but weather.gov's forecasts include a link to the NWS office's forecast discussion (updated every few hours). This can give you a bit of insight into forecaster uncertainty and the variation
18.
by ckennelly 12y ago
Thanks for taking a look. To go through your points: * If the "wrong" GPG binary is loaded, the game is likely already up: If an attacker has that degree of control over your machine, they're already on the other side of thi
19.
Show HN: Asymmetricfs – An Asymmetric, Encrypting Filesystem for Userspace(github.com)
20 points by ckennelly 12y ago|3 comments
20.
Panoptes: A Binary Translation Framework for CUDA(github.com)
2 points by ckennelly 14y ago|0 comments