We can help not only in primitives, but in scalar processing. There was
a big effort in j8.05 to speed up scalar processing. This had 4 main parts:
* each arithmetic primitive has a special case for handling singletons
* processing of sentences in explicit definitions was made faster
* the parser and memory allocator were recoded to be faster on modern CPUs
* names in explicit definitions were prehashed to speed up the lookup of
the value
But we can always do more. Some algorithms are simply scalar in nature,
and for them we need to reduce interpretive overhead. If you have cases
like that, send them to me.
Henry Rich
On 4/26/2017 10:47 PM, Igor Zhuravlov wrote:
Speed-up of matrix product made in JE core is substantial. This operation is
basic for mt addon. Further speed-up of mt code I see in mt algorithms some of
which are scalar. If I encounter bottleneck in primitive, I'll follow your
advise, thank you.
---
This email has been checked for viruses by AVG.
http://www.avg.com
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
