Clock cycles per instruction?

[christian_convey]

If this is the wrong forum, I apologize - it's the closest match I could find for my question.

I'm trying to find out how many clock cycles are required for various double-precision operations, both in their simple forms, and in their SSE and (if applicable) AVX forms. For example, I'm trying to understand the relative costs of doube-precision comparisons, multiplications, and divisions for Intel's recent processors (Core 2 Duo up through i7's.)

Can anyone point me in the right direction?

Thanks very much,
Christian

[Matthias_Kretz]

Look for the Optimization Reference Manual on http://intel.com/product/processors/manuals/. It contains much more good information and in appendix C the complete list of instruction latencies and throughputs.

BTW, nowadays the preferred method of doing scalar floating point operations is by using the SSE scalar instructions which have the same performance as their vector counterparts. Therefore vectorization normally yields a speedup of the number of entries in the vector (at least for the arithmetics).

PS: Can somebody please replace this forum software with something that works? I had to copy that URL manually because pasting only works 5% of the time...

[Gaiger_Chen]

You means "Intel 64 and IA-32 Architectures Optimization Reference Manual"??

The link is changed to this.

[Matthias_Kretz]

That's the one I meant. But I just wanted to link to the page which contains all those very useful manuals. Since the manuals get updated from time to time. So the overview page is really the best pointer. Anyway, of course I go the URL wrong when typing it. Pasting still doesn't work (as does toggling to HTML Editor). :(

http://intel.com/products/processor/manuals/

[stephanie00]

This is not exactly what you are looking for, but it's pretty amazing. I checked wikipedia.org for computer benchmark info, and got a big page with lots of info and outside references:

http://en.wikipedia.org/wiki/Benchmark_%28computing%29

I know you want to compare different algorithms on the same processor, and that is different than benchmarking different processors against the same algorithm, but maybe poke around some of the links to see if you can find what you want, or see other terms to search on. Pretty interesting links that I followed....

[Taronyu]

Profilers may be useful too. Have a look at Intel's development software, they offer some performance analysis tools (non-free). Or you look for a free one, maybe give Google a shot

[Nicolae_P_Intel]

I would suggest looking atIntel Architecture Code Analyzer ( http://software.intel.com/en-us/articles/intel-architecture-code-analyzer/) tool which you can get for free. Some other free useful tools can be found at whatif.intel.com.

[Taronyu]

Quoting Nicolae Popovici (Intel)

I would suggest looking atIntel Architecture Code Analyzer ( http://software.intel.com/en-us/articles/intel-architecture-code-analyzer/) tool which you can get for free. Some other free useful tools can be found at whatif.intel.com.

Thanks for sharing that link! I wasn't looking for a program like this but it might be useful for me in the future :)

[0xr]

I think here's everything you need:http://www.freeweb.hu/instlatx64/