Running _mm512_popcnt_epi32 on colfax using icc?

[YAkha]

How can i run _mm512_popcnt_epi32 on the colfax KNL7210? Does vpopcntd instruction need to be enabled? How can i do this?

I also need to include <iostream> in the C file, however I am using icc. How can I do this?

#include <stdio.h>
#include <mkl.h>
#include <immintrin.h>
#include <zmmintrin.h>

int main(){
        __m512i k, b, c;
        c = _mm512_and_epi32(k, b);
        printf("%d", c);
        int len;
        len = _mm512_popcnt_epi32(c);
        printf("%d\n", len);
}

I have another doubt. Does the function _mm512_popcnt_epi32 return a pointer to an array of 16 integers denoting the population count of each of the 16 integers packed into the _m512i data type? Is assigning this to an integer incorrect? I thought that the function above returns the total popcnt of all the 512 elements in the _m512i data type.

I am running the code above using the following submission script:

cd ~/benchmarking/

icc matmul.c -o mat.out -xMIC-AVX512

./mat.out

Is there something I am missing here?

Thank you!

[James_C_Intel2 · Accepted solution]

According to the table on the AVX512 Wikipedia page (scroll right to the bottom) (which I believe is accurate) , the vponcnt instructions are not implemented in the Xeon Phi 72xxx series. So, unless you have a very small soldering iron and an electron microscope to re-engineer the chip :-), you can't do what you want to. Indeed, there seem to be no, currently shipping, cores which have those instructions.

[YAkha]

So I will have to implement an unoptimized version of pop count or so, isn't it? And there seem to be a new Xeon Scalable lineup (8xxx), does it have this instruction? Thank you for your help!

[YAkha]

Cownie, James H (Intel) wrote:

According to the table on the AVX512 Wikipedia page (scroll right to the bottom) (which I believe is accurate) , the vponcnt instructions are not implemented in the Xeon Phi 72xxx series. So, unless you have a very small soldering iron and an electron microscope to re-engineer the chip :-), you can't do what you want to. Indeed, there seem to be no, currently shipping, cores which have those instructions.

I was looking at the wiki page, and it says that the instruction: VPOPCNTD is in the extension set VPOPCNTDQ. The Skylake-SP, Skylake-X processors (2017) support: AVX-512 F, CD, BW, DQ, VL.

The Xeon platinum 8180 has 

AVX512DQ	AVX-512 Double and Quad

as a supported extension.

Does that mean that the latest lineup of Xeon based on the Skylake architecture can run the intrinsic (_mm512_popcnt_epi32) ? Doesn't the DevCloud's Xeon Scalable processor family include Xeon Platinum 8180/Skylake arch CPUs which support AVX512DQ?

I am a sophomore in my undergraduate degree and have not done any courses in computer architecture, so i am struggling assimilating these concepts definitively. I would really appreciate your help.

[James_C_Intel2]

 was looking at the wiki page, and it says that the instruction: VPOPCNTD is in the extension set VPOPCNTDQ. The Skylake-SP, Skylake-X processors (2017) support: AVX-512 F, CD, BW, DQ, VL.

The Xeon platinum 8180 has 

AVX512DQ
AVX-512 Double and Quad

as a supported extension.

Does that mean that the latest lineup of Xeon based on the Skylake architecture can run the intrinsic (_mm512_popcnt_epi32) ? Doesn't the DevCloud's Xeon Scalable processor family include Xeon Platinum 8180/Skylake arch CPUs which support AVX512DQ?

You are confusing the DQ extensions with the VPOPCNTDQ extensions. If you look at that table again, you'll see that it has separate columns for DQ and VPOPCNTDQ.

[YAkha]

Cownie, James H (Intel) wrote:

 was looking at the wiki page, and it says that the instruction: VPOPCNTD is in the extension set VPOPCNTDQ. The Skylake-SP, Skylake-X processors (2017) support: AVX-512 F, CD, BW, DQ, VL.

The Xeon platinum 8180 has 

AVX512DQ
AVX-512 Double and Quad

as a supported extension.

Does that mean that the latest lineup of Xeon based on the Skylake architecture can run the intrinsic (_mm512_popcnt_epi32) ? Doesn't the DevCloud's Xeon Scalable processor family include Xeon Platinum 8180/Skylake arch CPUs which support AVX512DQ?

You are confusing the DQ extensions with the VPOPCNTDQ extensions. If you look at that table again, you'll see that it has separate columns for DQ and VPOPCNTDQ.

I apologize, seemed to have missed that in a hurry. I observe that VPOPCNTDQ will be included in Knights Mill and Ice Lake. I suppose I will have to wait till Knights Mill arrives later this year.

Thank you for the assistance !