7 ms·

Also noteworthy that on Intel at least, using AVX/AVX2 reduces the frequency of the CPU for a while. It can even go below base clock.

scottlamb 8y ago
iirc, it's complicated. Some instructions don't reduce the frequency; some reduce it a little; some reduce it a lot. I'm not sure AVX2 is as ubiquitous as the README says: "We assume AVX2 support which is available in all recent mainstream x86 processors produced by AMD and Intel." I guess "mainstream" is somewhat subjective, but some recent Chromebooks have Celeron processors with no AVX2: https://us-store.acer.com/chromebook-14-cb3-431-c5fm https://us-store.acer.com/chromebook-14-cb3-431-c5fm https://ark.intel.com/products/91831/Intel-Celeron-Processor-N3160-2M-Cache-up-to-2_24-GHz https://ark.intel.com/products/91831/Intel-Celeron-Processor...
- Ultimatt 8y ago
  Because someone wanting 2.2GB/s JSON parsing is deploying to a chromebook...
  - scottlamb 8y ago
    It doesn't seem that laughable to me to want faster JSON parsing on a Chromebook, given how heavily JSON is used to communicate between webservers and client-side Javascript. "Faster" meaning faster than Chromebooks do now; 2.2 GB/s may simply be unachievable hardware-wise with these cheap processors. They're kinda slow, so any speed increase would be welcome.
- sitkack 8y ago
  AVX2 also incurs some pretty large penalties for switching between SSE and AVX2. Depending on the amount of time taken in the library between calls, it could be problematic. This looks mostly applicable to server scenarios where the runtime environment is highly controlled.
  - BeeOnRope 8y ago
    There is no real penalty for switching between SSE and AVX2, unless you do it wrong. What are you referring to specifically? Are you talking about state transition penalties that can occur if you forget a vzeroupper? That's the only thing I'm aware of which kind of matches that.

scottlamb 8y ago

iirc, it's complicated. Some instructions don't reduce the frequency; some reduce it a little; some reduce it a lot. I'm not sure AVX2 is as ubiquitous as the README says: "We assume AVX2 support which is available in all recent mainstream x86 processors produced by AMD and Intel." I guess "mainstream" is somewhat subjective, but some recent Chromebooks have Celeron processors with no AVX2: https://us-store.acer.com/chromebook-14-cb3-431-c5fm https://us-store.acer.com/chromebook-14-cb3-431-c5fm https://ark.intel.com/products/91831/Intel-Celeron-Processor-N3160-2M-Cache-up-to-2_24-GHz https://ark.intel.com/products/91831/Intel-Celeron-Processor...

Ultimatt 8y ago
Because someone wanting 2.2GB/s JSON parsing is deploying to a chromebook...
- scottlamb 8y ago
  It doesn't seem that laughable to me to want faster JSON parsing on a Chromebook, given how heavily JSON is used to communicate between webservers and client-side Javascript. "Faster" meaning faster than Chromebooks do now; 2.2 GB/s may simply be unachievable hardware-wise with these cheap processors. They're kinda slow, so any speed increase would be welcome.
sitkack 8y ago
AVX2 also incurs some pretty large penalties for switching between SSE and AVX2. Depending on the amount of time taken in the library between calls, it could be problematic. This looks mostly applicable to server scenarios where the runtime environment is highly controlled.
- BeeOnRope 8y ago
  There is no real penalty for switching between SSE and AVX2, unless you do it wrong. What are you referring to specifically? Are you talking about state transition penalties that can occur if you forget a vzeroupper? That's the only thing I'm aware of which kind of matches that.