MEM : ALU ratio

Article: Knights Landing Details
By: Nicolas Capens (nicolas.capens.delete@this.gmail.com), January 20, 2014 3:48 pm
Room: Moderated Discussions
Stubabe (Stubabe.delete@this.nospam.com) on January 17, 2014 12:58 pm wrote:
> Nicolas Capens (nicolas.capens.delete@this.gmail.com) on January 16, 2014 9:23 am wrote:
> >Haswell has four arithmetic execution ports instead of two, despite 'only' two load ports, >because
> they expect to have a purpose for them! Those two extra ports can also require a load >operand.
> If it was really necessary to have two load ports just for the two floating-point >execution ports,
> you couldn't execute many other instructions, let alone two! But Intel has >explicitly mentioned
> that a fourth execution unit was added to offload the vector ports. As a >consequence they also
> added a third AGU to sustain two loads per cycle, but nothing exceeding >that 1:2 ratio.
>
> You have heard of specfp_rate? Haswell's FMA units ARE constrained by its load bandwidth even now.
> With a latency of 5 you need 10 software visible registers just for the accumulators to fully exploit 2
> FMA units and you still have to source 2 other operands for each of those 10 instruction for registers/ram.
> Haswell with 2 load ports and 16 registers is already starved. That's ignoring the 2 other "spare" execution
> units. But those will typically be used to blend/shuffle or to do flow control/loop overhead/pointer respectively
> so no they don't go to waste just because there is insufficient load bandwidth.
>
> Knights [whatever] may have 32 registers but that not really enough to prevent high load bandwidth either
> as you still need at least one load per FMA for matrix stuff, and often have a high ratio of load:store so
> reusing the store port for loads is probably not a bad idea (I doubt they added a 3rd memory issue pipe).

One load per FMA is not necessary. NVIDIA's GK110 architecture can sustain its theoretical DPFLOPS rating in DGEMM, with two DPFMA units and just one load unit. It even achieves 45% of its peak performance in SGEMM with six SPFMA units and one load unit. So Haswell's two load units suffice to supply all four arithmetic ports with memory operands. You need Hyper-Threading to help hide the latency of Haswell's 5-cycle FMA, but Xeon Phi has a 4-cycle latency, twice the registers, and 4-way SMT. So one load port would suffice for KNL to sustain a peak IPC of 2 instructions.

> > > I'm confident there is no L0 cache. I never said that the L1D
> > > is dual ported, I said that it supports 2x64B reads/clock.
> >
> > Then it needs two read ports. Even if it's implemented with multiple banks with a single read
> > port each, it adds complexity to receive two addresses, select banks, and deal with conflicts,
> > and you're also consuming twice the SRAM lookup power on every dual access in the same cycle.
> > With a tiny L0 cache you could avoid all that and even reduce register file pressure.
>
> Yes

Great.

> but how it that implemented?
> Either in the Load units themselves -> Still need two Issue ports for Load + 2 writeback paths.
>
> OR
>
> Its an extra stage in the pipeline for floating point operations (and just uses the bypass network)
> I.e. Horrific on an OOO design as the scheduler needs to check the L0 indexes on each issue (as L0 hit/miss
> will effect instruction latency and coissue)or if we assume the FPU is in order (as in Silvermont) then that's
> an extra (stallable) pipeline stage that will introduce bubbles on a L0 miss - also power inefficent.

I assume the vector pipelines are in-order with a 4-cycle latency and round-robin scheduling, just like KNC. So the L0 cache does not affect scheduling. One L0 miss becomes an L1 access, so that's fine. Two L0 misses would create a bubble in one of the pipes, but that should be rare enough and cheap enough compared to the energy savings from L0 hits, and the savings from simplifying the register file.

> The real problem with a tiny L0 is its piss poor hit rate. A two entry structure would only work
> for A,A,A,A or A,B,A,C,A,D type load patterns, in any other case it will miss. So even if it does
> save power in a few corner cases its just an expensive (power) waste of die area in every other.

That's exactly the load patterns we're observing in the examples Eric has posted. Remember, the L0 hit ratio does not have to be very high. It just allows to reuse memory operands without requiring a second L1 load port and to allow reducing the register file complexity.

Here are some papers which clarify why a very high hit rate for an L0 like structure isn't necessary:
"The Filter Cache: An Energy Efficient Memory Structure" (Johnson Kin et al., 1997)
"Reducing Data Cache Energy Consumption via Cached Load/Store Queue" (Dan Nicolaescu et al., 2003)

> You appear only to consider the case where you L0 helps with out factoring the continued cost of it when
> it provides no benefit (during a miss).

First of all there are also continued savings from simplifying the register file. Secondly the L0 cache is utterly tiny so any cost is easily won back tenfold by hits. Look at the 'filter cache' and 'cached load/store queue' for similar very cheap additions with big savings.

> The problem with placing it in the pipeline or linking to instruction
> issue is that appears to be very hard to clock gate, at least a second load unit can be reasonably easily
> clock gated when not needed (or just get the store issue pipe to double duty for load as well).

Please elaborate. I don't see the above two papers mentioning any problems with clock gating.

> it would hurt FLOP/s and FLOP/watt. The P4s trace cache comes to mind.
> > > When the facts change, I might change my conclusion. Until then, I'm quite confident
> > > of where I stand. And I'm still open to that wager about the existence of the L0...
> >
> > I'm not trying to convince you that I'm correct about the L0 cache. I'm trying to convince you that
> > a dual-ported L1 is not a given due to x86 being a load-op ISA. Other x86 architectures, as well
> > as GPUs, have a lower MEM:ALU ratio and they're doing fine. But a single L1 load port wouldn't explain
> > the code Eric has posted. So I proposed the L0 cache as something which would explain all the given
> > "facts" at once. Whether that's the solution Intel ended up implementing, I don't know. It's just
> > one of the viable possibilities besides assuming it "must have" two L1 read ports.
>
> A far better explanation (if you really want to avoid two load ports) is Load elimination (especially as Intel
> actually have a patent filed for that) i.e. where identical loads are eliminated by PRF redirection during rename.
> But the only reason a complier would explicitly take advantage of such a technique is to reduce total instruction
> counts or to deal with register pressure, it would not offer power savings v normal register use.

Why would it not save power? You're saving accessing the L1 cache for the same operand.

> Given the facts (or more absence of them) and that the complier appears to issue
> similar code for both Knights landing and knights corner I think the most likely
> explanation is the complier's back end is bugged for these targets.

It amazes me that you find a bug an acceptable explanation for this horrendous code (assuming no hardware structure to efficiently support it), while the insignificant power consumption of an L0 cache is somehow unacceptable. The software is just as important as the hardware. I can assure you that no compiler developer would make even an early prototype available that would produce code like this, unless it was intentional and the result of knowledge about the architecture. It can't be due to an unimplemented optimization, because the Haswell code looks as expected and it would be trivial to use it for the AVX-512. Compiler developers look at assembly output all day long, so this isn't something that just slipped their attention.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Knights Landing details (new article)David Kanter2014/01/03 12:58 AM
  eDRAM as cacheiz2014/01/03 04:39 AM
    eDRAM optionsEric Bron2014/01/09 03:45 AM
  Knights Landing details (new article)Emil Briggs2014/01/03 06:06 AM
  Knights Landing details (new article)Michael S2014/01/03 07:05 AM
    PCI-E and QPIDavid Kanter2014/01/03 12:11 PM
  eDRAM still seems too expensive ...Mark Roulo2014/01/03 10:48 AM
    Nevermind ... I see that you addressed this :-)Mark Roulo2014/01/03 10:51 AM
    eDRAM still seems too expensive ...Eric Bron2014/01/03 01:42 PM
  eDRAM or stacked DRAM?Patrick Chase2014/01/03 11:21 AM
    eDRAM or stacked DRAM?Wes Felter2014/01/03 03:00 PM
      eDRAM or stacked DRAM?Patrick Chase2014/01/03 07:26 PM
        eDRAM or stacked DRAM?tarlinian2014/06/23 09:59 PM
          eDRAM or stacked DRAM?Maynard Handley2014/06/24 01:47 AM
            eDRAM or stacked DRAM?Michael S2014/06/24 03:13 AM
            eDRAM or stacked DRAM?David Kanter2014/06/24 12:09 PM
              eDRAM or stacked DRAM?anon2014/06/24 07:50 PM
                eDRAM or stacked DRAM?Eric Bron2014/06/24 10:02 PM
                  eDRAM or stacked DRAM?anon2014/06/24 10:39 PM
                eDRAM or stacked DRAM?Michael S2014/06/25 01:46 AM
              eDRAM or stacked DRAM?Michael S2014/06/25 01:29 AM
          eDRAM or stacked DRAM?Eric Bron2014/06/24 05:37 AM
            eDRAM or stacked DRAM?tarlinian2014/06/24 08:53 AM
              eDRAM or stacked DRAM?Eric Bron2014/06/24 09:09 AM
                eDRAM or stacked DRAM?tarlinian2014/06/24 09:40 AM
                  eDRAM or stacked DRAM?Eric Bron2014/06/24 10:10 AM
                    eDRAM or stacked DRAM?Eric Bron2014/06/24 10:12 AM
          eDRAM or stacked DRAM?Wes Felter2014/06/24 10:09 PM
            eDRAM or stacked DRAM?Michael S2014/06/25 02:02 AM
  Why not tag-inclusive L3?Paul A. Clayton2014/01/03 04:28 PM
    Why not tag-inclusive L3?Eric Bron2014/01/04 03:22 AM
  Knights Landing L/S bandwidthNicolas Capens2014/01/04 05:43 AM
    Knights Landing L/S bandwidthEric Bron2014/01/04 06:20 AM
      Knights Landing L/S bandwidthNicolas Capens2014/01/04 02:55 PM
        Knights Landing L/S bandwidthEric Bron2014/01/04 03:27 PM
          Knights Landing L/S bandwidthhobold2014/01/04 04:23 PM
            Knights Landing L/S bandwidthEric Bron2014/01/04 05:20 PM
              Knights Landing L/S bandwidthMichael S2014/01/05 03:42 AM
                Knights Landing L/S bandwidthEric Bron2014/01/05 03:49 AM
                  Knights Landing L/S bandwidthPatrick Chase2014/01/11 08:13 PM
                    Knights Landing L/S bandwidthNicolas Capens2014/01/13 08:39 PM
                Knights Landing L/S bandwidthNicolas Capens2014/01/05 03:18 PM
                  Knights Landing L/S bandwidthMichael S2014/01/06 04:09 AM
                    Knights Landing L/S bandwidthEric Bron2014/01/06 05:11 AM
                      Knights Landing L/S bandwidthMichael S2014/01/06 05:40 AM
                        Knights Landing L/S bandwidthEric Bron2014/01/06 05:54 AM
                        Knights Landing L/S bandwidthEric Bron2014/01/08 09:00 AM
                    Knights Landing L/S bandwidthNicolas Capens2014/01/07 03:31 PM
                      Knights Landing L/S bandwidthMichael S2014/01/07 04:17 PM
                        Knights Landing L/S bandwidthNicolas Capens2014/01/07 09:55 PM
                          Knights Landing L/S bandwidthMichael S2014/01/08 01:42 AM
                            Knights Landing L/S bandwidthGabriele Svelto2014/01/08 08:30 AM
                              Occam's razorNicolas Capens2014/01/08 02:33 PM
                                Occam's razorGabriele Svelto2014/01/08 02:51 PM
                                  Occam's razorEric Bron2014/01/08 03:28 PM
                                    Occam's razorbakaneko2014/01/09 04:45 AM
                                      Occam's razoranon2014/01/09 05:02 AM
                                        Occam's razorbakaneko2014/01/09 06:24 AM
                                          Occam's razorbakaneko2014/01/09 06:51 AM
                                            Occam's razoranon2014/01/09 07:18 AM
                                          Occam's razoranon2014/01/09 07:16 AM
                                            Occam's razorbakaneko2014/01/09 08:43 AM
                                              Occam's razoranon2014/01/09 09:17 AM
                                                Occam's razorbakaneko2014/01/09 11:12 AM
                                                  Occam's razorEric Bron2014/01/09 11:18 AM
                                                    Occam's razorbakaneko2014/01/09 11:58 AM
                                                  Occam's razoranon2014/01/09 12:35 PM
                                                    Occam's razorbakaneko2014/01/12 10:48 AM
                                                  99.9% not a new extensionNicolas Capens2014/01/10 11:39 AM
                                                    Compiler complexityGabriele Svelto2014/01/11 03:58 AM
                                                      Compiler complexityNicolas Capens2014/01/11 01:20 PM
                                                        Compiler complexityGabriele Svelto2014/01/11 03:17 PM
                                                          Patent pendingNicolas Capens2014/01/14 07:21 PM
                                                    99.9% not a new extensionbakaneko2014/01/12 11:08 AM
                                  L0 data cacheEric Bron2014/01/08 04:52 PM
                                  Occam's razorDavid Kanter2014/01/08 04:53 PM
                                    Occam's razorNicolas Capens2014/01/09 03:07 AM
                                      Occam's razorRicardo B2014/01/09 05:21 AM
                                        Virtually indexed, untaggedNicolas Capens2014/01/10 11:27 AM
                                          Virtually indexed, untaggedGabriele Svelto2014/01/11 04:08 AM
                                            Virtually indexed, untaggedNicolas Capens2014/01/11 09:45 PM
                                              Virtually indexed, untaggedDavid Kanter2014/01/12 02:13 AM
                                                Virtually indexed, untaggedanon2014/01/12 04:02 AM
                                                Virtually indexed, untaggedNicolas Capens2014/01/16 09:55 AM
                                              Virtually indexed, untaggedMichael S2014/01/12 04:09 AM
                                                Virtually indexed, untaggedNicolas Capens2014/01/16 10:47 AM
                                      Occam's razorDavid Kanter2014/01/09 06:42 PM
                                        Occam's razorNicolas Capens2014/01/10 02:22 PM
                                          Occam's razorDavid Kanter2014/01/10 04:06 PM
                                            MEM : ALU ratioNicolas Capens2014/01/11 12:24 AM
                                              MEM : ALU ratioGabriele Svelto2014/01/11 03:47 AM
                                                MEM : ALU ratioEric Bron2014/01/11 04:41 AM
                                                  MEM : ALU ratioEric Bron2014/01/11 05:06 AM
                                                    MEM : ALU ratioDavid Kanter2014/01/11 08:28 PM
                                                      MEM : ALU ratioEric Bron nli2014/01/12 02:54 AM
                                                  MEM : ALU ratioGabriele Svelto2014/01/11 10:15 AM
                                                MEM : ALU ratioNicolas Capens2014/01/14 06:56 PM
                                                  Etiquette in linking to papersPaul A. Clayton2014/01/14 07:44 PM
                                                  MEM : ALU ratioanon2014/01/14 08:32 PM
                                                    L0 power costNicolas Capens2014/01/16 02:05 PM
                                                      L0 power costanon2014/01/16 10:01 PM
                                                        L0 power costNicolas Capens2014/01/19 12:30 AM
                                                          Links revealedPaul A. Clayton2014/01/19 04:47 PM
                                                          L0 power costanon2014/01/20 01:19 AM
                                                            L0 power costNicolas Capens2014/01/20 02:49 PM
                                                              L0 power costanon2014/01/21 01:18 AM
                                                                Q.E.D.Nicolas Capens2014/01/21 08:44 PM
                                                                  Q.E.D.anon2014/01/21 09:24 PM
                                                                    Straw manNicolas Capens2014/01/23 11:56 PM
                                                                      Straw mananon2014/01/25 06:46 AM
                                                                        Still waiting for an explanationNicolas Capens2014/01/26 12:19 AM
                                                                          Still waiting for an explanationExophase2014/01/26 01:13 PM
                                                                            Still waiting for an explanationbakaneko2014/01/26 11:52 PM
                                                                  Q.E.D.Ricardo B2014/01/22 06:58 PM
                                                                    Q.E.D.Michael S2014/01/23 04:59 AM
                                                                      L0 entry countNicolas Capens2014/01/24 01:11 AM
                                                                        L0 entry countEric Bron2014/01/24 02:08 AM
                                                                          L0 entry countMichael S2014/01/24 06:18 AM
                                                                            L0 entry countEric Bron2014/01/24 07:15 AM
                                                                              L0 entry countMichael S2014/01/24 08:10 AM
                                                                                L0 entry countEric Bron2014/01/24 08:20 AM
                                                                          L0 entry countNicolas Capens2014/01/24 02:33 PM
                                                                            L0 entry countEric Bron2014/01/24 03:20 PM
                                                                              L0 entry count and L1 read port orthogonalityNicolas Capens2014/01/26 01:14 AM
                                                                                L0 entry count and L1 read port orthogonalityEric Bron2014/01/26 03:49 AM
                                                                    L0 hit rateNicolas Capens2014/01/24 12:49 AM
                                                                      L0 hit rateRicardo B2014/01/24 06:42 AM
                                                                        L0 hit rateExophase2014/01/24 01:37 PM
                                                                          L0 hit rateEric Bron2014/01/24 02:12 PM
                                                                        L0 vs RF powerNicolas Capens2014/01/24 02:43 PM
                                              MEM : ALU ratioDavid Kanter2014/01/11 01:47 PM
                                                MEM : ALU ratioNicolas Capens2014/01/16 09:23 AM
                                                  MEM : ALU ratioStubabe2014/01/17 12:58 PM
                                                    MEM : ALU ratioStubabe2014/01/17 01:42 PM
                                                      MEM : ALU ratioMichael S2014/01/18 04:57 PM
                                                        MEM : ALU ratiobakaneko2014/01/19 12:47 AM
                                                    MEM : ALU ratioNicolas Capens2014/01/20 03:48 PM
                                                      It's called "tunnel vision" (NT)iz2014/01/20 04:36 PM
                                                      MEM : ALU ratioMichael S2014/01/20 04:37 PM
                                                        MEM : ALU ratioStubabe2014/01/21 04:54 PM
                                                        MEM : ALU ratioNicolas Capens2014/01/21 10:07 PM
                                                          MEM : ALU ratioMichael S2014/01/22 08:17 AM
                                                            MEM : ALU ratioNicolas Capens2014/01/24 03:33 PM
                                                      MEM : ALU ratioStubabe2014/01/21 04:32 PM
                                                        MEM : ALU ratioMichael S2014/01/22 08:56 AM
                                                          MEM : ALU ratioStubabe2014/01/23 09:06 AM
                                                            MEM : ALU ratioEric Bron2014/01/23 09:45 AM
                                                              editEric Bron2014/01/23 09:49 AM
                                                            MEM : ALU ratioMichael S2014/01/23 09:58 AM
                                                              MEM : ALU ratioEric Bron2014/01/23 10:29 AM
                                                                MEM : ALU ratioMichael S2014/01/23 10:33 AM
                                                              MEM : ALU ratioStubabe2014/01/24 04:50 AM
                                                MEM : ALU ratiobakaneko2014/01/23 10:36 AM
                                              MEM : ALU ratioNoSpammer2014/01/11 03:39 PM
                                                L1 vs L0 access costNicolas Capens2014/01/16 03:17 PM
                                                  L1 vs L0 access costNoSpammer2014/01/19 01:48 PM
                                                    L1 vs L0 access costdmcq2014/01/22 05:45 AM
                                                      L1 vs L0 access costGabriele Svelto2014/01/22 07:29 AM
                                                        L1 vs L0 access costdmcq2014/01/22 01:33 PM
                                                          L1 vs L0 access costGabriele Svelto2014/01/22 04:33 PM
                                                            L1 vs L0 access costdmcq2014/01/24 04:19 AM
                                                    L1 vs L0 access costNicolas Capens2014/01/24 02:16 AM
                                      Occam's razorPatrick Chase2014/01/13 11:19 AM
                                  Occam's razorNicolas Capens2014/01/09 12:40 AM
                                    Occam's razorGabriele Svelto2014/01/09 02:41 AM
                                      Occam's razorEric Bron2014/01/09 02:54 AM
                                        Occam's razorGabriele Svelto2014/01/09 06:35 AM
                                          Occam's razorEric Bron2014/01/09 07:14 AM
                                            avoiding redundant loadsEric Bron2014/01/09 07:18 AM
                                            AVX2 versionEric Bron2014/01/09 07:32 AM
                                      Occam's razorAmiba Gelos2014/01/09 03:01 AM
                                        Occam's razorEric Bron2014/01/09 03:06 AM
                                          Occam's razorAmiba Gelos2014/01/09 03:43 AM
                                            Occam's razorEric Bron2014/01/09 04:02 AM
                                        L0 access latencyNicolas Capens2014/01/09 04:27 AM
                                          L0 access latencyAmiba Gelos2014/01/09 05:16 AM
                                            compared to L0$ i would say banking is far more likely (NT)Amiba Gelos2014/01/09 05:20 AM
                                            L0 access latencyNicolas Capens2014/01/10 03:20 PM
                                      Occam's razorNicolas Capens2014/01/09 04:19 AM
                                    Occam's razorNoSpammer2014/01/09 12:55 PM
                                      Occam's razorNicolas Capens2014/01/10 03:40 PM
                                        Occam's razorMichael S2014/01/11 10:21 AM
                                        Occam's razorMichael S2014/01/12 03:21 PM
                                          KNC compiler outputNicolas Capens2014/01/16 06:39 PM
                                            KNC compiler outputMichael S2014/01/18 05:13 PM
                                    L0 cache coherencyDavid Kanter2014/01/11 08:39 PM
                                Occam's razoranon2014/01/09 05:12 AM
                            Knights Landing L/S bandwidthEric Bron2014/01/08 10:46 AM
                              Knights Landing L/S bandwidthMichael S2014/01/08 11:23 AM
                            Knights Landing L/S bandwidthNicolas Capens2014/01/08 02:02 PM
                              Knights Landing L/S bandwidthMichael S2014/01/08 02:29 PM
                                Knights Landing L/S bandwidthEric Bron2014/01/08 02:54 PM
                                  Knights Landing L/S bandwidthMichael S2014/01/08 03:00 PM
                                    Knights Landing L/S bandwidthEric Bron2014/01/08 03:13 PM
                                      Knights Landing L/S bandwidthMichael S2014/01/08 03:28 PM
                                        Knights Landing L/S bandwidthEric Bron2014/01/08 03:32 PM
                                          Knights Landing L/S bandwidthMichael S2014/01/08 03:40 PM
                                            Knights Landing L/S bandwidthEric Bron2014/01/08 03:51 PM
                                              Knights Landing L/S bandwidthMichael S2014/01/09 12:18 PM
                          Knights Landing L/S bandwidthPatrick Chase2014/01/12 10:03 PM
                            Also page/line splits?David Kanter2014/01/12 10:50 PM
                              Also page/line splits?anon2014/01/13 01:44 AM
                                Also page/line splits?none2014/01/13 03:09 AM
                                  Also page/line splits?anon2014/01/13 04:19 AM
                            Knights Landing L/S bandwidthExophase2014/01/13 12:15 AM
                            Knights Landing L/S bandwidthanon2014/01/13 01:41 AM
                              Knights Landing L/S bandwidthPatrick Chase2014/01/13 11:14 AM
                            Aliased writesNicolas Capens2014/01/14 09:46 PM
                      Knights Landing L/S bandwidthRicardo B2014/01/07 04:27 PM
                        Knights Landing L/S bandwidthNicolas Capens2014/01/07 10:28 PM
                          Knights Landing L/S bandwidthRicardo B2014/01/08 02:13 AM
                            Knights Landing L/S bandwidthEric Bron2014/01/08 11:10 AM
                            Knights Landing L/S bandwidthNicolas Capens2014/01/08 03:31 PM
                              Knights Landing L/S bandwidthRicardo B2014/01/08 03:58 PM
                                Knights Landing L/S bandwidthG. Gouvine2014/01/09 09:10 AM
                                  Knights Landing L/S bandwidthRicardo B2014/01/09 11:19 AM
                                    Efficient load queue vs. efficient L0 cacheNicolas Capens2014/01/11 12:28 PM
                                      Efficient load queue vs. efficient L0 cacheG. Gouvine2014/01/13 02:11 AM
                                        Efficient load queue vs. efficient L0 cacheMichael S2014/01/13 03:43 AM
                                Register file read port requirementsNicolas Capens2014/01/11 12:55 AM
                                  Register file read port requirementsRicardo B2014/01/11 05:24 AM
                                    Register file read port requirementsEric Bron2014/01/11 05:32 AM
                                      Register file read port requirementsMichael S2014/01/11 09:57 AM
                                        Register file read port requirementsEric Bron2014/01/11 11:16 AM
                                          Register file read port requirementsMichael S2014/01/11 11:46 AM
                                            Register file read port requirementsEric Bron2014/01/11 12:12 PM
                                              Register file read port requirementsMichael S2014/01/11 12:36 PM
                                                Register file read port requirementsEric Bron2014/01/11 12:51 PM
                                              Register file read port requirementsPatrick Chase2014/01/13 02:27 PM
                                                Register file read port requirementsEric Bron2014/01/13 04:24 PM
                                                  Register file read port requirementsPatrick Chase2014/01/13 06:02 PM
                                                    Register file read port requirementsEric Bron2014/01/14 04:50 AM
                                                      Register file read port requirementsMichael S2014/01/14 11:36 AM
                                                        Register file read port requirementsEric Bron nli2014/01/14 01:04 PM
                                            Register file read port requirementsPatrick Chase2014/01/13 02:17 PM
                                              Register file read port requirementsMichael S2014/01/15 04:27 AM
                                        Register file read port requirementsEric Bron2014/01/11 11:28 AM
                                          Register file read port requirementsMichael S2014/01/11 12:07 PM
                                            Register file read port requirementsPatrick Chase2014/01/13 02:40 PM
                                          Register file read port requirementsPatrick Chase2014/01/13 02:34 PM
                                      Register file read port requirementsRicardo B2014/01/11 12:55 PM
                                        Register file read port requirementsEric Bron2014/01/11 01:17 PM
                                          Register file read port requirementsRicardo B2014/01/11 02:36 PM
                                            Register file read port requirementsEric Bron2014/01/11 02:42 PM
                                              Register file read port requirementsRicardo B2014/01/11 03:20 PM
                                                Register file read port requirementsEric Bron2014/01/11 03:26 PM
                                                  Register file read port requirementsMichael S2014/01/11 04:07 PM
                                                    Register file read port requirementsRicardo B2014/01/11 04:38 PM
                                                      Register file read port requirementsMichael S2014/01/11 04:49 PM
                                                Register file read port requirementsEric Bron2014/01/11 03:39 PM
                                                  Register file read port requirementsEric Bron2014/01/11 03:41 PM
                                                  Register file read port requirementsRicardo B2014/01/11 04:30 PM
                                    Register file read port requirementsNicolas Capens2014/01/11 12:09 PM
              Knights Landing L/S bandwidthanon2014/01/05 06:55 AM
                Knights Landing L/S bandwidthEric Bron2014/01/05 07:30 AM
                  Knights Landing L/S bandwidthanon2014/01/06 01:07 AM
                    Knights Landing L/S bandwidthEric Bron2014/01/06 02:38 AM
                      Knights Landing L/S bandwidthanon2014/01/06 04:01 AM
                        Knights Landing L/S bandwidthEric Bron2014/01/06 04:44 AM
                          Knights Landing L/S bandwidthanon2014/01/06 05:39 AM
                            Knights Landing L/S bandwidthEric Bron2014/01/06 06:00 AM
                              Knights Landing L/S bandwidthanon2014/01/06 06:44 AM
                                Knights Landing L/S bandwidthMichael S2014/01/06 08:54 AM
                                  Knights Landing L/S bandwidthEric Bron2014/01/06 10:11 AM
                                    Knights Landing L/S bandwidthMichael S2014/01/06 10:14 AM
                                      Knights Landing L/S bandwidthEric Bron2014/01/06 11:37 AM
                                        Knights Landing L/S bandwidthRicardo B2014/01/08 06:25 AM
                                          Knights Landing L/S bandwidthEric Bron2014/01/08 08:36 AM
                                            Knights Landing L/S bandwidthEric Bron2014/01/08 08:41 AM
                                            KNC code generator with EVEX back-end?Michael S2014/01/08 09:43 AM
                                              KNC code generator with EVEX back-end?Exophase2014/01/08 10:00 AM
                                                KNC code generator with EVEX back-end?Ricardo B2014/01/08 11:39 AM
                                                  KNC code generator with EVEX back-end?Eric Bron2014/01/08 12:15 PM
                                                    KNC code generator with EVEX back-end?Exophase2014/01/08 01:17 PM
                                                      KNC code generator with EVEX back-end?Ricardo B2014/01/08 02:06 PM
                                                        KNC code generator with EVEX back-end?Exophase2014/01/08 02:24 PM
                                                        KNC code generator with EVEX back-end?Eric Bron2014/01/08 02:38 PM
                                                    KNC code generator with EVEX back-end?Michael S2014/01/08 01:54 PM
                                              KNC code generator with EVEX back-end?Eric Bron2014/01/08 10:25 AM
                                              KNC code generator with EVEX back-end?Eric Bron2014/01/08 10:35 AM
                                                KNC code generator with EVEX back-end?Michael S2014/01/08 11:07 AM
                                                  KNC code generator with EVEX back-end?Eric Bron2014/01/08 11:24 AM
                                                    KNC code generator with EVEX back-end?Michael S2014/01/08 11:43 AM
                                                      KNC code generator with EVEX back-end?Eric Bron2014/01/08 01:23 PM
                                              KNC code generator with EVEX back-end?Eric Bron2014/01/08 10:43 AM
                                          AVX2 code much different than AVX-512Eric Bron2014/01/08 08:52 AM
                                            evil questionhobold2014/01/08 10:22 AM
                                              evil questionEric Bron2014/01/08 10:27 AM
                                                evil questionhobold2014/01/08 02:33 PM
                                                  evil questionMichael S2014/01/08 02:37 PM
                                                    stupid question (was: evil question)hobold2014/01/09 05:41 AM
                                                      stupid question (was: evil question)Eric Bron2014/01/09 05:52 AM
                                                        stupid question (was: evil question)Michael S2014/01/09 08:00 AM
                                                          stupid question (was: evil question)Michael S2014/01/09 08:12 AM
                                                            stupid question (was: evil question)Eric Bron2014/01/09 10:47 AM
                                                              stupid question (was: evil question)Michael S2014/01/09 11:48 AM
                                                                more decisive (hopefully) test caseMichael S2014/01/09 12:01 PM
                                                                  more decisive (hopefully) test caseEric Bron2014/01/09 12:08 PM
                                                                    more decisive (hopefully) test caseMichael S2014/01/09 12:24 PM
                                                                      more decisive (hopefully) test caseEric Bron2014/01/09 12:27 PM
                                                                        more decisive (hopefully) test caseMichael S2014/01/09 12:33 PM
                                                                  AVX2Eric Bron2014/01/09 12:14 PM
                                                                    AVX2Michael S2014/01/09 12:30 PM
                                                                      AVX2Eric Bron2014/01/09 12:40 PM
                                                                  another tryMichael S2014/01/09 03:02 PM
                                                                    another tryEric Bron2014/01/09 03:33 PM
                                                                      another tryMichael S2014/01/09 04:20 PM
                                                                      another try - ignore misformated mess aboveMichael S2014/01/09 04:24 PM
                                                                        another try - ignore misformated mess aboveGabriele Svelto2014/01/10 01:01 AM
                                                                          another try - ignore misformated mess aboveEric Bron2014/01/10 03:05 AM
                                                                            another try - ignore misformated mess aboveMichael S2014/01/11 10:23 AM
                                                                              another try - ignore misformated mess aboveEric Bron2014/01/11 11:08 AM
                                                                                another try - ignore misformated mess aboveMichael S2014/01/11 12:09 PM
                                                                                  another try - ignore misformated mess aboveMichael S2014/01/11 12:12 PM
                                                                                    another try - ignore misformated mess aboveEric Bron2014/01/11 12:24 PM
                                                                                      another try - ignore misformated mess aboveMichael S2014/01/11 01:24 PM
                                                                                        another try - ignore misformated mess aboveEric Bron2014/01/11 02:11 PM
                                                                                          another try - ignore misformated mess aboveMichael S2014/01/11 02:18 PM
                                                                                            another try - ignore misformated mess aboveEric Bron2014/01/11 02:27 PM
                                                                                              another try - ignore misformated mess aboveMichael S2014/01/11 02:29 PM
                                                                                                another try - ignore misformated mess aboveEric Bron2014/01/11 02:46 PM
                                                                                                  another try - ignore misformated mess aboveEric Bron2014/01/11 02:46 PM
                                                                                                  another try - ignore misformated mess aboveMichael S2014/01/11 03:28 PM
                                                                                        another try - ignore misformated mess aboveEric Bron2014/01/11 02:17 PM
                                                                                          another try - ignore misformated mess aboveMichael S2014/01/11 02:24 PM
                                                                    KNC versionMichael S2014/01/11 05:19 PM
                                                                      KNC versionEric Bron nli2014/01/12 02:59 AM
                                                                        KNC versionGabriele Svelto2014/01/12 09:06 AM
                                                  evil questionEric Bron2014/01/08 02:41 PM
              Knights Landing L/S bandwidthPatrick Chase2014/01/05 11:20 PM
                Knights Landing L/S bandwidthEric Bron2014/01/06 02:45 AM
                  Knights Landing L/S bandwidthanon2014/01/06 04:12 AM
                    Knights Landing L/S bandwidthMichael S2014/01/06 04:17 AM
                      Knights Landing L/S bandwidthanon2014/01/06 05:20 AM
          Knights Landing L/S bandwidthNicolas Capens2014/01/04 05:34 PM
            Knights Landing L/S bandwidthEric Bron2014/01/04 05:44 PM
              Knights Landing L/S bandwidthNicolas Capens2014/01/05 12:25 PM
                Knights Landing L/S bandwidthEric Bron2014/01/05 01:50 PM
                  Knights Landing L/S bandwidthNicolas Capens2014/01/05 03:34 PM
                    Might even help with gatherNicolas Capens2014/01/05 03:40 PM
                      What is an L0 cache?David Kanter2014/01/05 10:44 PM
                        What is an L0 cache?anon2014/01/06 05:57 AM
                          What is an L0 cache?Nicolas Capens2014/01/06 12:57 PM
                            What is an L0 cache?anon2014/01/06 02:18 PM
    Knights Landing L/S bandwidthDavid Kanter2014/01/04 10:58 AM
      Knights Landing L/S bandwidthNicolas Capens2014/01/04 04:24 PM
        Knights Landing L/S bandwidthEric Bron2014/01/04 04:46 PM
          Knights Landing L/S bandwidthKonrad Schwarz2014/01/08 12:48 AM
            Knights Landing L/S bandwidthMichael S2014/01/08 02:45 AM
        Knights Landing L/S bandwidthDavid Kanter2014/01/05 01:44 AM
          Knights Landing L/S bandwidthEric Bron2014/01/05 03:55 AM
          Knights Landing L/S bandwidthNicolas Capens2014/01/05 12:18 PM
            Knights Landing L/S bandwidthMaynard Handley2014/01/05 11:33 PM
              Knights Landing L/S bandwidthEric Bron2014/01/06 04:02 AM
                Knights Landing L/S bandwidthMichael S2014/01/06 04:23 AM
                  Knights Landing L/S bandwidthEric Bron2014/01/06 04:35 AM
                    Knights Landing L/S bandwidthMichael S2014/01/06 05:20 AM
                      Knights Landing L/S bandwidthMichael S2014/01/06 05:32 AM
                      Knights Landing L/S bandwidthEric Bron2014/01/06 05:36 AM
                        Knights Landing L/S bandwidthMichael S2014/01/06 06:00 AM
                          Knights Landing L/S bandwidthEric Bron2014/01/06 06:07 AM
                          Knights Landing L/S bandwidthEric Bron2014/01/06 06:14 AM
                            editsEric Bron2014/01/06 06:22 AM
                              optimized versionEric Bron2014/01/06 06:35 AM
                                yet more optimized versionEric Bron2014/01/06 06:42 AM
                                  latest version for todayEric Bron2014/01/06 06:51 AM
                                    Probably just L2 bandwith limitedNicolas Capens2014/01/06 11:48 AM
                                  yet more optimized versionMaynard Handley2014/01/06 06:54 PM
                                optimized versionMaynard Handley2014/01/06 06:52 PM
                                  optimized versionMichael S2014/01/07 10:42 AM
                                    optimized versionNicolas Capens2014/01/07 12:36 PM
                                      optimized versionMichael S2014/01/07 03:41 PM
                                        optimized versionNicolas Capens2014/01/07 10:52 PM
                                          optimized versionMichael S2014/01/08 02:10 AM
                                    optimized versionEric Bron2014/01/07 02:34 PM
                                      optimized versionMichael S2014/01/07 03:18 PM
                                        optimized versionEric Bron2014/01/07 03:30 PM
                                          optimized versionEric Bron2014/01/07 03:33 PM
                                            optimized versionMichael S2014/01/07 03:57 PM
                                    optimized versionMaynard Handley2014/01/07 06:50 PM
                                      optimized versionMichael S2014/01/08 02:39 AM
                Knights Landing L/S bandwidthMaynard Handley2014/01/06 06:47 PM
              Knights Landing L/S bandwidthNicolas Capens2014/01/06 09:18 AM
                Knights Landing L/S bandwidthMaynard Handley2014/01/06 06:56 PM
                  Knights Landing L/S bandwidthNicolas Capens2014/01/07 12:18 PM
        Knights Landing L/S bandwidthNoSpammer2014/01/05 01:15 PM
          Knights Landing L/S bandwidthNicolas Capens2014/01/05 03:06 PM
            Knights Landing L/S bandwidthNoSpammer2014/01/06 04:20 AM
              Knights Landing L/S bandwidthNicolas Capens2014/01/06 11:54 AM
                Knights Landing L/S bandwidthNoSpammer2014/01/06 01:24 PM
                  Knights Landing L/S bandwidthNicolas Capens2014/01/06 09:15 PM
                    Knights Landing L/S bandwidthNoSpammer2014/01/07 03:58 AM
                      Knights Landing L/S bandwidthNicolas Capens2014/01/07 03:18 PM
                        Knights Landing L/S bandwidthNoSpammer2014/01/08 01:38 PM
                          Knights Landing L/S bandwidthNicolas Capens2014/01/08 11:14 PM
  AVX512F questionMichael S2014/01/06 10:18 AM
    AVX512F questionNicolas Capens2014/01/06 12:01 PM
  Knights Landing - time for obituary?Michael S2018/07/31 03:00 PM
    Knights Landing - time for obituary?Adrian2018/07/31 09:24 PM
      Knights Landing - time for obituary?SoftwareEngineer2018/08/01 02:15 AM
        auto-vectorization is a dead endMichael S2018/08/01 03:48 AM
          Auto-vectorization of random C is a dead endMark Roulo2018/08/01 11:07 AM
            Auto-vectorization of random C is a dead endPassing Through2018/08/01 01:35 PM
              Auto-vectorization of random C is a dead endDavid Kanter2018/08/01 10:44 PM
                Auto-vectorization of random C is a dead endPassing Through2018/08/02 01:51 AM
            Auto-vectorization of random C is a dead endSoftwareEngineer2018/08/02 01:19 AM
              Auto-vectorization of random C is a dead endMark Roulo2018/08/02 09:50 AM
                Auto-vectorization of random C is a dead endMichael S2018/08/02 12:11 PM
                  Auto-vectorization of random C is a dead endj2018/08/02 11:37 PM
                    Auto-vectorization of random C is a dead endMichael S2018/08/03 03:50 AM
                      Auto-vectorization of random C is a dead endrwessel2018/08/03 11:06 PM
                  Auto-vectorization of random C is a dead endRicardo B2018/08/03 04:20 AM
                    Auto-vectorization of random C is a dead endMichael S2018/08/03 05:37 AM
                      Auto-vectorization of random C is a dead endRicardo B2018/08/03 11:22 AM
            Auto-vectorization of random C is a dead endTravis2018/08/03 07:58 PM
              Potential way to autovectorization in the future. Jouni Osmala2018/08/03 10:22 PM
                Potential way to autovectorization in the future. Jukka Larja2018/08/04 04:03 AM
                  Potential way to autovectorization in the future. Passing Through2018/08/04 06:47 AM
                Potential way to autovectorization in the future. Travis2018/08/04 01:50 PM
                  Potential way to autovectorization in the future. Michael S2018/08/04 02:33 PM
                    Potential way to autovectorization in the future. Travis2018/08/04 02:48 PM
                    Potential way to autovectorization in the future. Passing Through2018/08/04 02:58 PM
                  Skylake server/client AVX PRF speculationJeff S.2018/08/04 05:42 PM
                    Skylake server/client AVX PRF speculationanonymou52018/08/04 06:21 PM
                      Skylake server/client AVX PRF speculationJeff S.2018/08/04 06:38 PM
                        Skylake server/client AVX PRF speculationanonymou52018/08/04 07:45 PM
                          Skylake server/client AVX PRF speculationJeff S.2018/08/04 08:08 PM
                            Skylake server/client AVX PRF speculationanonymou52018/08/04 08:18 PM
                              Skylake server/client AVX PRF speculationNomad2018/08/05 11:10 PM
                                Skylake server/client AVX PRF speculationanonymou52018/08/06 12:14 PM
                                Skylake server/client AVX PRF speculationTravis2018/08/06 08:43 PM
                            Skylake server/client AVX PRF speculationTravis2018/08/06 08:39 PM
              Auto-vectorization of random C is a dead endBrett2018/08/04 01:55 PM
                Auto-vectorization of random C is a dead endTravis2018/08/04 02:38 PM
                  Auto-vectorization of random C is a dead endPassing Through2018/08/04 03:00 PM
                    New record for shortest post by Ireland - AI crashed? (NT)Travis2018/08/04 03:34 PM
                      New record for shortest post by Ireland - AI crashed?Passing Through2018/08/04 04:12 PM
                        New record for shortest post by Ireland - AI crashed?anonymou52018/08/04 06:00 PM
                          New record for shortest post by Ireland - AI crashed?Brett2018/08/04 06:40 PM
                            New record for shortest post by Ireland - AI crashed?anonymou52018/08/04 07:38 PM
                  Auto-vectorization of random C is a dead endnoko2018/08/04 09:46 PM
        The story of ispc (a 12 entry blog series)Simon Farnsworth2018/08/01 03:50 AM
          the 1st link is empty (NT)Michael S2018/08/01 04:05 AM
            the 1st link is emptySimon Farnsworth2018/08/01 06:42 AM
          Interesting read, thanks! (NT)SoftwareEngineer2018/08/01 06:57 AM
          Amazing readLaurent2018/08/01 09:00 AM
            Amazing readPassing Through2018/08/01 01:13 PM
              Amazing readDoug S2018/08/01 02:30 PM
                Amazing readPassing Through2018/08/01 02:49 PM
          ISPC vs OpenCL?j2018/08/02 11:41 PM
            ISPC vs OpenCL?coppcie2018/08/03 03:55 AM
            ISPC vs OpenCL?Passing Through2018/08/03 04:07 AM
              Go awayForum Reader2018/08/03 08:11 AM
              ISPC vs OpenCL?Gian-Carlo Pascutto2018/09/11 06:50 AM
            ISPC vs OpenCL?SoftwareEngineer2018/08/03 04:18 AM
        Knights Landing - time for obituary?Kevin G2018/08/01 07:14 AM
          Knights Landing - time for obituary?SoftwareEngineer2018/08/01 07:29 AM
        Knights Landing - time for obituary?Passing Through2018/08/01 07:38 AM
          Knights Landing - time for obituary?Eric Bron2018/08/02 06:57 AM
            Knights Landing - time for obituary?Passing Through2018/08/02 12:29 PM
              Knights Landing - time for obituary?Eric Bron2018/08/02 01:49 PM
                Knights Landing - time for obituary?Passing Through2018/08/02 02:17 PM
          chess algorithms vs, low level optimizationsEric Bron2018/08/02 07:15 AM
            AlphaZero vs StockfishMichael S2018/08/02 07:55 AM
              AlphaZero vs StockfishEric Bron2018/08/02 08:24 AM
                AlphaZero vs StockfishMichael S2018/08/02 09:01 AM
                  AlphaZero vs StockfishEric Bron2018/08/02 09:11 AM
                  Leela 4th vs all othersEric Bron nli2018/09/11 03:40 AM
              AlphaZero vs StockfishGian-Carlo Pascutto2018/09/11 06:31 AM
                AlphaZero vs StockfishEric Bron2018/09/11 09:26 AM
                  AlphaZero vs StockfishEric Bron2018/09/11 09:58 AM
                    AlphaZero vs StockfishPer Hesselgren2018/12/31 10:04 AM
                Leela Chess ZeroPer Hesselgren2018/12/31 12:00 PM
              AlphaZero vs Stockfish (on Xeon)Per Hesselgren2018/12/31 09:59 AM
        C/C++ and vector/parallel/distributedRichardC2018/08/02 05:50 AM
      Knights Landing - time for obituary?Passing Through2018/08/01 07:52 AM
        Knights Landing - time for obituary?Kevin G2018/08/01 02:03 PM
          Knights Landing - time for obituary?Passing Through2018/08/01 02:33 PM
      Knights Landing - time for obituary?Kevin G2018/08/01 08:26 AM
        Knights Landing - time for obituary?Kevin G2018/08/01 08:26 AM
        Knights Landing - time for obituary?juanrga2018/08/01 02:26 PM
          Knights Landing - time for obituary?hobel2018/08/02 05:46 AM
    Knights Landing - time for obituary?juanrga2018/07/31 11:25 PM
      Right, time for obituary for whole LRB lineageAM2018/08/02 11:46 AM
        Right, time for obituary for whole LRB lineageAdrian2018/08/02 11:46 PM
          LRBNI, AVX512, etc...Michael S2018/08/03 05:23 AM
        Right, time for obituary for whole LRB lineagejuanrga2018/08/03 04:11 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell tangerine? 🍊