Haswell CPU article online

Article: Intel's Haswell CPU Microarchitecture
By: Ricardo B (ricardo.b.delete@this.xxxxx.xx), November 13, 2012 6:09 pm
Room: Moderated Discussions
hobold (hobold.delete@this.vectorizer.org) on November 13, 2012 5:13 pm wrote:
>
> I would guess that in the context of a fully cache coherent manycore machine, when you
> try to optimize the scatter operation to store more than a single vector element at a
> time, you run into problems. Memory transactions probably don't make that any easier.
>
> For example, if a scatter operation has to be aborted and restarted for some reason, does it (semantically)
> execute all or nothing? Or can it be in a partly completed state? Larrabee made partly completed
> state information architecturally visible (in a mandatory boolean mask register), but did not support
> transactional memory. As far as I know, AVX* does not expose such internal state.
>
> Or when two "simultaneous" scatter operations from two different cores fight over overlapping
> memory addresses, and then one or both operations has to be undone and later rerun ... is it even
> possible to decide on a consistent specification of what the memory contents ought to be?
>
> Does the coherency protocol support groups of in-flight memory accesses that are semantically
> related to one another? With respect to one or more memory transactions?
>
>
> I could be blowing this issue out of proportions due to personal
> cluelessness. But it does seem rather complicated to me.

AVX2 does expose internal state.
«This instruction can be suspended by an exception if at least one element is already
gathered (i.e., if the exception is triggered by an element other than the rightmost
one with its mask bit set). When this happens, the destination register and the mask
operand are partially updated; those elements that have been gathered are placed
into the destination register and have their mask bits set to zero. If any traps or
interrupts are pending from already gathered elements, they will be delivered in lieu
of the exception; in this case, EFLAG.RF is set to one so an instruction breakpoint is
not re-triggered when the instruction is continued.»

This, of course, makes it simple to implement gather: it's actually a µOP sequence, with individual µOP for each load. All it's required is to keep the proper order.

Using a similar method for stores should not be complicated either.

I'd say Intel simply decided scatter wasn't worth the effort.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Haswell CPU article onlineDavid Kanter11/13/12 03:43 PM
  Haswell CPU article onlineEric11/13/12 04:10 PM
    Haswell CPU article onlinehobold11/13/12 05:13 PM
      Haswell CPU article onlineRicardo B11/13/12 06:09 PM
    Haswell CPU article onlineanonymou511/13/12 05:44 PM
      Haswell CPU article onlinenone11/14/12 03:40 AM
  Haswell CPU article onlinetarlinian11/13/12 04:56 PM
    Fixed (NT)David Kanter11/13/12 06:06 PM
      Haswell CPU article onlineJacob Marley11/14/12 02:18 AM
  Haswell CPU article onlinerandomshinichi11/14/12 02:53 AM
    LLC == Last Level Cache (usually L3) (NT)Paul A. Clayton11/14/12 05:50 AM
    Haswell CPU article onlineJoe11/14/12 10:38 AM
      LLC vs. L3 vs. L4David Kanter11/14/12 11:09 AM
        LLC vs. L3 vs. L4; LLC = Link Layer ControllerRay11/14/12 10:08 PM
          A pit there are only 17000 TLAs... (NT)EduardoS11/15/12 03:14 AM
  Haswell CPU article onlineanon11/14/12 05:10 AM
    Move elimination can be a µop fusionPaul A. Clayton11/14/12 06:41 AM
      That should be "mov R10 <- R9"! (NT)Paul A. Clayton11/14/12 06:43 AM
      Move elimination can be a µop fusionanon11/14/12 07:25 AM
        It does avoid the scheduler (NT)Paul A. Clayton11/14/12 08:47 AM
      Move elimination can be a µop fusionStubabe11/14/12 01:43 PM
        Move elimination can be a µop fusionanon11/14/12 09:33 PM
          Move elimination can be a µop fusionFelid11/15/12 12:49 AM
            Move elimination can be a µop fusionanon11/15/12 01:23 AM
              Move elimination can be a µop fusionStuart11/15/12 05:04 AM
                Move elimination can be a µop fusionStubabe11/15/12 05:14 AM
                  Move elimination can be a µop fusionanon11/15/12 05:48 AM
                    Move elimination can be a µop fusionEduardoS11/15/12 06:00 AM
                      Move elimination can be a µop fusionanon11/15/12 06:14 AM
                        Move elimination can be a µop fusionEduardoS11/15/12 06:21 AM
                          Move elimination can be a µop fusionanon11/15/12 06:31 AM
                    Move elimination can be a µop fusionStubabe11/15/12 11:38 AM
                      There can be only one dependencePaul A. Clayton11/15/12 12:50 PM
                    Move elimination can be a µop fusionFelid11/15/12 03:19 PM
                      Move elimination can be a µop fusionanon11/16/12 04:07 AM
                        Move elimination can be a µop fusionFelid11/16/12 07:43 PM
                  Move elimination can be a µop fusionFelid11/15/12 02:50 PM
                    Move elimination can be a µop fusionFelid11/15/12 03:03 PM
                      Correction!Felid11/19/12 01:23 AM
                    Thanks, I wasn't aware of the change in SB. Good to know... (NT)Stubabe11/15/12 03:43 PM
            Move fusion assumes adjacencyPaul A. Clayton11/15/12 07:15 AM
              Move fusion assumes adjacencyFelid11/15/12 02:40 PM
        Move elimination can be a µop fusionPatrick Chase11/21/12 11:52 AM
          Move elimination can be a µop fusionPatrick Chase11/21/12 12:12 PM
    Haswell CPU article onlineRicardo B11/14/12 09:12 AM
  Haswell CPU article onlinegmb11/14/12 08:28 AM
  Haswell CPU article onlineFelid11/14/12 11:58 PM
    Haswell CPU article onlineDavid Kanter11/15/12 09:59 AM
      Haswell CPU article onlineFelid11/15/12 02:15 PM
        Instruction queueDavid Kanter11/16/12 12:23 PM
          Instruction queueFelid11/16/12 01:05 PM
  128-bit division unit?Eric Bron11/16/12 04:57 AM
    128-bit division unit?David Kanter11/16/12 08:59 AM
      128-bit division unit?Eric Bron11/16/12 09:47 AM
        128-bit division unit?Felid11/16/12 12:46 PM
          128-bit division unit?Eric Bron11/16/12 01:24 PM
            128-bit division unit?Felid11/16/12 07:19 PM
              128-bit division unit?Eric Bron11/18/12 08:41 AM
            128-bit division unit?Michael S11/17/12 12:50 PM
              128-bit division unit?Felid11/17/12 01:44 PM
                128-bit division unit?Michael S11/17/12 02:45 PM
                  128-bit division unit?Felid11/17/12 05:49 PM
                    128-bit division unit?Michael S11/17/12 06:56 PM
              128-bit division unit?Eric Bron11/18/12 08:35 AM
  Haswell CPU article onlineJim F11/18/12 09:45 AM
    Haswell CPU article onlineGabriele Svelto11/18/12 12:52 PM
  Probable bottleneckLaurent Birtz11/23/12 01:45 PM
    Probable bottleneckEduardoS11/23/12 01:58 PM
      Probable bottleneckLaurent Birtz11/24/12 10:10 AM
    Probable bottleneckStubabe11/25/12 03:08 AM
      Probable bottleneckEduardoS11/25/12 08:15 AM
        Probable bottleneckStubabe11/28/12 04:36 PM
          Urgh. Post got mangled by LESS THAN signStubabe11/28/12 04:41 PM
          Probable bottleneckLaurent Birtz11/29/12 08:34 AM
  Haswell CPU article onlineMr. Camel11/28/12 03:47 PM
    Haswell CPU article onlineEduardoS11/28/12 04:06 PM
      Haswell CPU article onlineMr. Camel11/28/12 07:23 PM
        Haswell CPU article onlineEduardoS11/28/12 07:27 PM
          Haswell CPU article onlineMr. Camel12/12/12 01:39 PM
            Much faster iGPU clock ...Mark Roulo12/12/12 03:53 PM
              Much faster iGPU clock ...Exophase12/12/12 11:46 PM
                Much faster iGPU clock ... or not :-)Mark Roulo12/13/12 09:11 AM
                  Much faster iGPU clock ... or not :-)EduardoS12/13/12 10:38 PM
                    Much faster iGPU clock ... or not :-)Michael S12/14/12 05:33 AM
                      Much faster iGPU clock ... or not :-)EduardoS12/14/12 07:06 AM
                        Much faster iGPU clock ... or not :-)Doug S12/14/12 12:13 PM
                          Much faster iGPU clock ... or not :-)EduardoS12/14/12 12:43 PM
                  Much faster iGPU clock ... or not :-)Mr. Camel12/14/12 10:50 AM
              Much faster iGPU clock ...Michael S12/13/12 02:44 AM
                Much faster iGPU clock ...Mark Roulo12/13/12 09:09 AM
  Haswell CPU article onlineYang12/09/12 08:28 PM
    possible spam bot? (NT)I.S.T.12/10/12 03:40 PM
  CPU Crystal Well behavior w/ eGPU?Robert Williams04/17/13 02:16 PM
    CPU Crystal Well behavior w/ eGPU?Nicolas Capens04/17/13 03:30 PM
      CPU Crystal Well behavior w/ eGPU?RecessionCone04/17/13 04:20 PM
        CPU Crystal Well behavior w/ eGPU?Robert Williams04/17/13 07:37 PM
    CPU Crystal Well behavior w/ eGPU?Eric Bron04/17/13 09:10 PM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell blue?