topic Empty Concurreny column in Software Archive

Empty Concurreny column

Pindor — Fri, 26 Mar 2010 12:36:26 GMT

I recently purchased the Intel Parallel Studio for use on my WindowsXP, Core2Quad machine.

I've created a small example code to exercise the following snippet of code. When run, I expected the concurrency column as detailed in the code. However, the column is all blank. Why? And how to I configure it to always show the actual time used on every statement?

.. snippet ..

const size_t tripCnt(5000);

for (size_t i(0); i

{

a[0] = static_cast(i);

// ippmCopy_ma_32f_SS is not threaded

// (so Parallel Amplifier should show 1 CPU/poor concurrency BUT doesn't!)

status = ippmCopy_ma_32f_SS(

a, srcStride0, srcStride1, srcStride2,

b, dstStride0, dstStride1, dstStride2,

height, width, 1);

// ippsConv_32f is threaded (according to ThreadedFunctionsList.txt)

// (so Parallel Amplifier should show 4 CPU/ideal concurrency BUT doesn't!)

// (however, the Summary pane does show 4 CPUs in use)

status = ippsConv_32f(a, height*width, b, height, c);

}

.. end snippet ..

Empty Concurreny column

Peter_W_Intel — Mon, 29 Mar 2010 06:59:26 GMT

I tried your code on my side, actually there is no data to show - because it ran shortly! Usually CPU consumed time is little than 10ms, there is no result to show.

I changed tripCnt from 5000 to 500000, there are two functions "ippsZero_8u" and "ippConv_32f" to be displayed. However IPP doesn't provide .pdb files and source file, you only canselectone of them then click on "Top-down Tree" to view Call Stack, but snippet.exe/snippet.cpp consumes "0ms" - no performance data to show.

Regards, Peter

Empty Concurreny column

Gennady_F_Intel — Wed, 31 Mar 2010 08:34:17 GMT

it should be noted that all ippm*.* ( small matrix) functions are highly optimized for very small inputs like 3x3...6x6. They will perform the correct output, but for that input sizes

will works non-optimized (reference) code.

--Gennady