STOKE – a stochastic optimizer for x86_64 assembly

Re:STOKE – a stochastic optimizer for x86_64 assembly by frozenport on Tuesday January 19, 2016 @12:06AM (#10928976)

This is great and reminds me of the ATLAS BLAS[1]

I would really like the authors to discuss the effects of power-saving features, for example dynamic clocking of the CPU.

Indeed ATLAS will often fail to converge or get the wrong tuning when automatic power-tuning is not disabled. In a modern system is becoming increasingly difficult to manage concurrent power saving features, for example things like Intel's Turbo Boost or the clock on your mobile device.

[1]http://math-atlas.sourceforge.net/

[ Parent | on HN ]

Re:STOKE – a stochastic optimizer for x86_64 assembly by jepler on Tuesday January 19, 2016 @08:27AM (#10930539)

Do you have a reference for this problem? I would love to hear more. Is it thought to be a hardware bug, like this recent snafu with arithmetic (but for which power tuning was not mentioned as contributing) http://arstechnica.com/gadgets/2016/01/intel-skylake-bug-cau... ?

[ Parent | on HN ]

Re:STOKE – a stochastic optimizer for x86_64 assembly by frozenport on Tuesday January 19, 2016 @04:25PM (#10934169)

So, modern ATLAS has a CPU throttling check, for example: http://stackoverflow.com/questions/14592401/atlas-install-re...

[ Parent | on HN ]