3 ms·
It's strange that they intend for the decompressor to be fast on embedded devices and handhelds, yet give no benchmarks for ARM processors. Instead, they only t
by joosters 12y ago
It's strange that they intend for the decompressor to be fast on embedded devices and handhelds, yet give no benchmarks for ARM processors. Instead, they only talk about i7 performance, a chip that you won't find in the target environments.
Does the algorithm run well on ARM?
- nkurz 12y agoFor that matter, does it run well on a modern x64 processor? I'm confused by people who do optimization at this level, but don't distinguish between the different generations of processors. Core i7 is a marketing term corresponding to Intel's positioning of the CPU within the market, but says little about the performance characteristics of the processor. He mentions later that he's using a "Core i7 970", which is from the Westmere generation that came out in 2010. The decompression speed on the more modern Sandy Bridge, Ivy Bridge, Haswell, or Broadwell chips could be anywhere from the same to 8x better depending on how amenable the algorithm is SIMD improvements. Without testing it's hard to know, but my guess would be that since it was not consciously designed to take advantage of the newer processor features, it probably will be at the low end of this, and could be beaten handily by an approach that targets only new instruction sets.