BeeOnRope

94 ms·

871.
by BeeOnRope 9y ago
To be fair, unless I missed it, that article is mostly just saying "The L1 and L2 are much faster than L3 and beyond". That's correct - but it doesn't imply that the L2 is somehow as fast as the L1 (if that were true, wh
872.
by BeeOnRope 9y ago
Yes, it's a streaming situation - and that's my point: contrary to conventional wisdom, even the streaming situation - for a single core - is generally limited by concurrency and latency, and not by DRAM bandwidth, and there i
873.
by BeeOnRope 9y ago
Like any domain-specific algorithms PNG uses deflate as a final step after using image-specific filters which take advantage of typical 2D features. So in general png will do much better than gzip on image data, but it will generally alwa
874.
by BeeOnRope 9y ago
This test isn't very informative because both .png and .jpg are already compressed formats, with "better than gzip" strength so gzip/deflate isn't going to be able to compress the underlying data. You only see some
875.
by BeeOnRope 9y ago
You can't get close to 50 GB/s of bandwidth to a single core, even if your memory configuration and controller supports that because the per core bandwidth is limited by the latency per cache line and the maximum number of concurr
876.
by BeeOnRope 9y ago
> Intel processors support two loads + one write per core. That can be 2x256-bit reads + 1x256-bit write. Each clock, every clock. At least to L1 and L2... You can approach 2 reads + 1 write in L1, but you won't get close for data i
877.
by BeeOnRope 9y ago
Sure a legal audit is standard and usually much simpler than a full source audit for security, which has a complexity proportional to the project size.
878.
by BeeOnRope 9y ago
Java has taken a couple different shots at this, going back a decade or more, and the newer option is currently enabled in Java 9. Some background: https://stackoverflow.com/q/8833385/149138
879.
by BeeOnRope 9y ago
They are implemented in different ways, depending on the instrinsic. Many are implemented in the in-memory intermediate SSA (?) representation the JVM uses for its JIT, so they essentially get compiled down to assembly at the same time as t
880.
by BeeOnRope 9y ago
That version exists - it's called equals(), and it generates different code that doesn't need to distinguish the less than/greater than cases.
881.
by BeeOnRope 9y ago
There aren't really performance guarantees for CPU and the ones that are there won't help here. When this patch is released big providers will have the same hardware as before and sell it in the same way - but the OS and userspace
882.
by BeeOnRope 9y ago
Yup, I think so. In that case (PCID hardware on a PCID-enabled kernel), the performance effect should be more limited to the syscall itself. That said, why is the hit still so big with PCID? Surely just the CR3-swap by itself shouldn't
883.
by BeeOnRope 9y ago
A major underlying cause is that we're doing things in hardware that ought to be done in software. We really need to stop shipping software as native blobs and start shipping it as pseudocode, allowing the OS to manage native executio
884.
by BeeOnRope 9y ago
In fact, the tight syscall loop isn't even necessarily the worst case: the primary cost of this change isn't the a direct cost in the syscall, but the CR3 switch which invalidates the TLB and incurs an ongoing cost for some time
885.
by BeeOnRope 9y ago
Unfortunately, Skylake doesn't include AVX-512, only the recently released Skylake-X, which is available on server chips and a handful of high end consumer parts.
886.
by BeeOnRope 9y ago
This just shows how much a leg up garbage collected languages have when it comes to standalone lock-free structures. Sure, this is technically a wait-free hashmap implemented in C++, which is pretty cool... but it makes two atomic operati
887.
by BeeOnRope 9y ago
> If it's not vectorizable, LUT result is often used for indirect jump/call (like large switch statement) or memory access (say, a histogram etc.). You've lost me. The parent comments were specifically talking about using
888.
by BeeOnRope 9y ago
This was news to me, but in the other direction: the claim is that 50-60 percent of fish served as sushi has been frozen at some point, implying that 40-50 has never been frozen. Given that most of the non-tuna species need to be frozen a
889.
by BeeOnRope 9y ago
To be fair, replacing a series of ALU ops with a lookup table doesn't usually add a "data dependency" - the data dependency probably already existed, but perhaps flowed through registers rather than memory. What adding a loo
890.
by BeeOnRope 9y ago
Because I need FIFO ordering for tab switching with Ctrl+Tab, and every time I check, Chrome doesn't support it and extensions can't offer it (because they cannot override the shortcut). Firefox supports it via extensions, althoug
891.
by BeeOnRope 9y ago
Take any tests with scaling governor "powersave" using the intel_pstate with a grain of salt. This powersave mode has nothing in common with the acpi_cpufreq's version of powersave and is quite aggressive at clocking down the