Open MPI logo

Hardware Locality Users' Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Users mailing list

Subject: [hwloc-users] Istopo question
From: Norman Lo (normanlo_at_[hidden])
Date: 2010-01-28 12:56:56


Hi,

I checked out the current version of hwloc from the svn. I ran lstopo
and got the follow result (snapshot of it):

Towards the beginning:

  Socket#0 cpuset 0x000000ff
    L3Cache(8192KB) cpuset 0x000000ff
      L2Cache(256KB) cpuset 0x00000011
        L1Cache(32KB) cpuset 0x00000011
          Core#0 cpuset 0x00000011
            P#0 cpuset 0x00000001
            P#4 cpuset 0x00000010
      L2Cache(256KB) cpuset 0x00000022
        L1Cache(32KB) cpuset 0x00000022
          Core#1 cpuset 0x00000022
            P#1 cpuset 0x00000002
            P#5 cpuset 0x00000020
      L2Cache(256KB) cpuset 0x00000044
        L1Cache(32KB) cpuset 0x00000044
          Core#2 cpuset 0x00000044
            P#2 cpuset 0x00000004
            P#6 cpuset 0x00000040
      L2Cache(256KB) cpuset 0x00000088
        L1Cache(32KB) cpuset 0x00000088
          Core#3 cpuset 0x00000088
            P#3 cpuset 0x00000008
            P#7 cpuset 0x00000080

Towards the end:Machine(3016MB) + Socket#0 + L3#0(8192KB)
  L2#0(256KB) + L1#0(32KB) + Core#0
    P#0
    P#1
  L2#1(256KB) + L1#1(32KB) + Core#1
    P#2
    P#3
  L2#2(256KB) + L1#2(32KB) + Core#2
    P#4
    P#5
  L2#3(256KB) + L1#3(32KB) + Core#3
    P#6
    P#7

It seems to me that they are reporting different things. If I understand
correctly, the lstopo reports that P#1 and P#4 are in core 1 (and so
forth) while at the end, it reports that core#0 has P#0 and P#1. I'm
wondering why there is such a descrepancy or if I have misinterpret
something here. The machine I am running on is a i7-920 (4 cores with
hyperthreading).

Any answer would be greatly appreciated. Thank you very much in advance.

Norman