topic I confirm that there is no in Intel® Integrated Performance Primitives

Incorrect results from ippsSqrt_64f_A53 when ippSetFlushToZero is enabled

MARK_S_8 — Mon, 05 Sep 2016 15:57:23 GMT

C++ test program attached, which gives the following output:

ippCore 9.0.3 (r51269)
ippsSqrt_64f_A53:
x = 2.225074e-308 5.000000e-01 2.000000e+00 1.797693e+308
Sqrt_64f_A53(x) = 1.491668e-154 7.071068e-01 1.414214e+00 3.300384e+154
sqrt(x) = 1.491668e-154 7.071068e-01 1.414214e+00 1.340781e+154

For the input value DBL_MAX the result is wrong. Is this a know issue?

Hi Mark,

Zhen_Z_Intel — Tue, 06 Sep 2016 05:31:25 GMT

Hi Mark,

I tested both in IPP 9.0 & 2017, the problem is fixed in new version. The output value is correct by using IPP 2017, please update to use new release. Thank you.

Best regards,
Fiona

Hi Fiona,

Hristov__Hristo — Tue, 06 Sep 2016 08:52:40 GMT

Hi Fiona,

what is this new version IPP 9.0 & 2017 that we do not know about?

According to https://software.intel.com/en-us/articles/intel-ipp-90-library-release-notes, the latest is IPP 9.0 Update 3

Hi Mark,

Gennady_F_Intel — Tue, 06 Sep 2016 10:29:45 GMT

Hi Mark,

sorry for confusion, I see the problem on my side and this is an unknown issue for IPP v.9.0.update 3 and IPP 2017. IPP 2017 is the newest version which will be released within this week.

C:\...\compilers_and_libraries_2017.0.109\windows\redist\intel64\ipp>_2017.exe

ippIP AVX2 (l9) 2017.0.0 (r52494) 2017.0.0.52494
ippsSqrt_64f_A53:
x = 2.225074e-308 5.000000e-001 2.000000e+000 1.797693e+308
Sqrt_64f_A53(x) = 1.491668e-154 7.071068e-001 1.414214e+000 3.300384e+154
sqrt(x) = 1.491668e-154 7.071068e-001 1.414214e+000 1.340781e+154

the problem is escalated. We will let you know the update asap.

regards, Gennady

Hello Mark, the problem has

Gennady_F_Intel — Thu, 30 Mar 2017 14:53:38 GMT

Hello Mark, the problem has been fixed in IPP 2017 update 2. Could you check the fix and let us know the results.

I confirm that there is no

Aleksey_Y_ — Mon, 03 Apr 2017 06:59:56 GMT

I confirm that there is no error in the latest version.