Subnormal speed

By: Sylvain Collange (full.name.delete.delete@this.this.gmail.com), December 25, 2013 9:29 am
Room: Moderated Discussions
Michael S (already5chosen.delete@this.yahoo.com) on December 21, 2013 8:59 am wrote:
> Sylvain Collange (full.name.delete.delete@this.this.gmail.com) on December 20, 2013 5:03 pm wrote:
> > Is there any information available about the speed of subnormals on Haswell ?
>
> Haswell FMA implementation is different from others mentioned above. Haswell provides both "slow" FMA/FMUL
> path and "fast" FADD path through [one of the] execution units. So, the chance that that the speed of denormals
> for FADD is different from SandyB/IvyB sounds very slim. The chance that denormal handling for FMA/FMUL
> is clock or two faster than SandyB/IvyB do exist, but full speed denormals are unlikely.

Yes, but subnormals are not as much a problem in FP adders as they are in FP multipliers, so I was thinking a full-speed subnormal implementation would have been feasible.

However I have just found my answer in Agner's microarchitecture guide:
The Haswell has a penalty of approximately 124 clock cycles in all cases where an operation on normal numbers gives a denormal result. There is a similar penalty for a multiplication between a normal and a denormal number, regardless of whether the result is normal or denormal. There is no penalty for adding a normal and a denormal number, regardless of the result.


It looks like the fast normal+subnormal case was introduced (silently?) in Sandy Bridge. It seems indeed that the FP adder in Haswell is the same as in Sandy Brigde, and the FMA does not support subnormals natively in multiplications.
< Previous Post in ThreadNext Post in Thread >
TopicPosted ByDate
Intel 14nmAnonym2013/11/22 01:43 AM
  Intel 14nmDoug S2013/11/22 01:56 AM
    Intel 14nmBrett2013/11/22 07:13 PM
      Intel 14nmDoug S2013/11/22 11:58 PM
  Intel's promises about 14nmhobold2013/11/22 03:23 AM
    Intel's promises about 14nmDavid Kanter2013/11/22 03:28 PM
      Intel's promises about 14nmhobold2013/11/22 03:44 PM
        Intel's promises about 14nmAnonym2013/11/22 04:17 PM
        Intel's promises about 14nmanon2013/11/22 04:29 PM
          Intel's promises about 14nmDoug S2013/11/23 12:03 AM
            Intel's promises about 14nmDavid Kanter2013/11/23 01:43 AM
              Intel's promises about 14nmnone2013/11/23 02:01 AM
              Intel's promises about 14nmGabriele Svelto2013/11/23 02:25 AM
                Intel's promises about 14nmbakaneko2013/11/23 02:55 PM
                  Intel's promises about 14nmAlberto2013/11/23 04:38 PM
                    Intel's promises about 14nmbakaneko2013/11/23 05:13 PM
                  Intel's promises about 14nmDavid Ball2013/11/24 02:23 PM
  Intel 14nmsomeone2013/11/22 09:08 AM
  Intel 14nmtmp9902013/11/23 12:34 AM
    Intel 14nmDavid Kanter2013/11/23 01:46 AM
      Intel 14nmAlberto2013/11/23 02:00 PM
        Intel 14nmDoug S2013/11/23 02:27 PM
          Intel 14nmAlberto2013/11/23 04:26 PM
        Intel 14nmanon2013/11/23 03:22 PM
      Intel 14nmDoug S2013/11/23 02:26 PM
        Intel 14nmDavid Kanter2013/11/23 06:22 PM
    Intel 14nmGabriele Svelto2013/11/23 02:28 AM
  Intel 14nmEric Bron2013/11/23 02:32 AM
    Intel 14nmAnonym2013/11/23 10:05 AM
      Intel 14nmhobold2013/11/23 11:37 AM
        Intel 14nmAnonym2013/11/23 01:53 PM
      Intel 14nmEric Bron2013/11/24 01:31 AM
        Intel 14nmAnonym2013/11/24 04:25 PM
          Intel 14nmSeni2013/11/25 03:42 AM
            Intel 14nmMr. Camel2013/11/25 07:32 AM
              Intel 14nmMr. Dromedary2013/11/25 07:06 PM
                That's enoughDavid Kanter2013/11/25 10:18 PM
          Intel 14nmEric Bron2013/11/25 04:36 AM
            Intel 14nmAnonym2013/11/25 09:21 AM
            Intel's chart is mislabeled Anonym2013/11/27 12:35 PM
              Intel's chart is mislabeled Eric Bron2013/11/27 01:31 PM
                Intel's chart is mislabeled Anonym2013/11/27 05:22 PM
                  Intel's chart is mislabeled Alberto2013/11/29 01:55 AM
                Intel's chart is mislabeled Alberto2013/11/29 02:09 AM
            Intel 14nmDavid Kanter2013/11/27 09:55 PM
              Intel 14nmEric Bron2013/11/28 01:56 AM
  The need for more performanceAnon2013/11/26 03:09 PM
    The need for more performancenone2013/11/26 05:50 PM
      The need for more performanceAnon2013/11/27 12:10 PM
        The need for more performancenone2013/11/28 08:24 AM
          No signs of ARM preparing for wide SIMDAnon2013/12/01 08:38 PM
            No signs of ARM preparing for wide SIMDnone2013/12/02 04:54 AM
              No signs of ARM preparing for wide SIMDAnon2013/12/04 11:39 PM
                ARM is one of the founders of HSA (NT)iz2013/12/05 03:41 AM
                  ARM is one of the founders of HSA Michael S2013/12/05 08:05 AM
                    ARM is one of the founders of HSA anonymous422013/12/05 10:24 AM
                    ARM is one of the founders of HSA iz2013/12/05 11:01 AM
                      ARM and AMD not sitting in a treeanon2013/12/05 12:59 PM
                        ARM and AMD not sitting in a treeTREZA2013/12/05 02:23 PM
                          Out-of-order + multi-core + SIMDanon2013/12/05 03:34 PM
                            Welcome back, Nick! (NT)Poindexter2013/12/06 05:13 AM
                  ARM is one of the founders of Hot Steaming Airanon2013/12/05 12:09 PM
                    ARM is one of the founders of Hot Steaming AirAnon2013/12/05 03:06 PM
                      ARM is one of the founders of Hot Steaming Airanon2013/12/05 04:14 PM
            No signs of ARM preparing for wide SIMDtmp9902013/12/02 05:18 AM
              No signs of ARM preparing for wide SIMDhobold2013/12/02 07:08 AM
                SIMD then, now, and beyondanon2013/12/05 02:48 PM
                  SIMD then, now, and beyondhobold2013/12/06 10:25 AM
                    SIMD then, now, and beyondSylvain Collange2013/12/06 02:30 PM
                      SIMD then, now, and beyondhobold2013/12/06 06:56 PM
                        SIMD then, now, and beyondanonymou52013/12/06 07:29 PM
                          SIMD then, now, and beyondhobold2013/12/07 05:28 AM
                        SIMD then, now, and beyondSylvain Collange2013/12/06 08:01 PM
                          SIMD then, now, and beyondhobold2013/12/07 05:09 AM
                            SIMD then, now, and beyondJouni Osmala2013/12/08 01:36 AM
                              SIMD then, now, and beyondRicardo B2013/12/08 04:56 PM
                                SIMD then, now, and beyondMichael S2013/12/09 01:43 AM
                          SIMD then, now, and beyondhobold2013/12/07 10:41 AM
                            SIMD then, now, and beyondSylvain Collange2013/12/07 12:36 PM
                              SIMD then, now, and beyondEric Bron2013/12/07 12:57 PM
                                SIMD then, now, and beyondSylvain Collange2013/12/07 02:31 PM
                                  SIMD then, now, and beyondEric Bron2013/12/09 12:20 PM
                        SIMD then, now, and beyondEric Bron2013/12/07 01:27 AM
                          SIMD then, now, and beyondhobold2013/12/07 05:17 AM
                            SIMD then, now, and beyondEric Bron2013/12/07 06:00 AM
                              SIMD then, now, and beyondhobold2013/12/07 09:26 AM
                                SIMD then, now, and beyondEric Bron2013/12/07 10:05 AM
                                  SIMD then, now, and beyondEric Bron2013/12/07 10:46 AM
                                  SIMD then, now, and beyondhobold2013/12/07 10:50 AM
                    SIMD then, now, and beyondAnon2013/12/07 04:59 PM
                      SIMD then, now, and beyondDoug S2013/12/07 06:19 PM
                        SIMD then, now, and beyondMichael S2013/12/08 03:09 AM
                          Does REX prefix's opcode stealing count? (NT)Paul A. Clayton2013/12/08 12:27 PM
                            Does REX prefix's opcode stealing count?Michael S2013/12/08 02:35 PM
                          SIMD then, now, and beyondanon2013/12/08 01:26 PM
                      SIMD then, now, and beyondhobold2013/12/08 11:59 AM
                      SIMD then, now, and beyondrwessel2013/12/09 09:55 PM
                        SIMD then, now, and beyondAnon2013/12/11 01:58 PM
                          SIMD then, now, and beyondrwessel2013/12/12 12:57 AM
                            SIMD then, now, and beyondrwessel2013/12/12 01:00 AM
                              SIMD then, now, and beyondMichael S2013/12/12 01:42 AM
                                SIMD then, now, and beyondrwessel2013/12/12 04:58 AM
                                  SIMD then, now, and beyondEric Bron2013/12/12 10:08 PM
                                    SIMD then, now, and beyondanonymou52013/12/13 12:11 AM
                                      SIMD then, now, and beyondEric Bron2013/12/13 01:48 AM
                                        SIMD then, now, and beyondanonymou52013/12/13 09:38 AM
                                          SIMD then, now, and beyondEric Bron2013/12/13 11:58 AM
                                            SIMD then, now, and beyondanonymou52013/12/14 10:45 AM
                                              SIMD then, now, and beyondEric Bron2013/12/14 11:10 AM
                                      SIMD then, now, and beyondEric Bron2013/12/13 02:08 AM
                                      SIMD then, now, and beyondEric Bron2013/12/13 12:04 PM
                                        SIMD then, now, and beyondanonymou52013/12/14 10:12 AM
                                          SIMD then, now, and beyondEric Bron2013/12/14 10:41 AM
                                    SIMD then, now, and beyondSylvain Collange2013/12/13 06:54 AM
                              Itanium and DLPAnon2013/12/16 10:30 PM
                                Itanium and DLPrwessel2013/12/16 11:02 PM
                                  Itanium and DLPDavid Kanter2013/12/17 06:29 PM
                                    Practicality of compiler task interleaving?Paul A. Clayton2013/12/17 09:53 PM
                                      Practicality of compiler task interleaving?rwessel2013/12/18 01:58 PM
                                        Any Law for code quality trends? :-)Paul A. Clayton2013/12/18 05:06 PM
                                    Itanium and DLPnone2013/12/18 05:09 AM
                                      Intel's case for AVX-512hobold2013/12/18 06:28 AM
                                        Intel's case for AVX-512Eric Bron2013/12/19 07:29 AM
                                      Itanium and DLPRicardo B2013/12/18 07:13 AM
                                        Itanium and DLPrwessel2013/12/18 02:17 PM
                                          Itanium and DLPMark Roulo2013/12/18 03:34 PM
                                        Itanium and DLPMaynard Handley2013/12/19 12:26 AM
                                          Itanium and DLPMichael S2013/12/19 02:17 AM
                                            Itanium and DLPnone2013/12/19 03:32 AM
                                            Itanium and DLPSylvain Collange2013/12/19 10:14 AM
                                              Itanium and DLPEduardoS2013/12/19 05:49 PM
                                                Itanium and DLPrwessel2013/12/20 12:27 AM
                                                  Itanium and DLPhobold2013/12/20 03:53 AM
                                                    6-clock latencyMichael S2013/12/20 05:18 AM
                                                      6-clock latencyhobold2013/12/20 07:04 AM
                                                      6-clock latencyMegol2013/12/22 03:47 AM
                                                    Itanium and DLPSylvain Collange2013/12/20 05:03 PM
                                                      Itanium and DLPMichael S2013/12/21 08:59 AM
                                                        Subnormal speedSylvain Collange2013/12/25 09:29 AM
                                                          Subnormal speedMichael S2013/12/25 12:51 PM
                                                            Subnormal speedTREZA2013/12/25 02:47 PM
                                                              Subnormal speedMichael S2013/12/25 03:28 PM
                                                              Subnormal speedEduardoS2013/12/25 04:44 PM
                                                                Subnormal speedMichael S2013/12/26 12:40 PM
                                      Itanium and DLPDavid Kanter2013/12/18 09:19 AM
                                        Itanium and DLPExophase2013/12/18 09:38 AM
                                          Itanium and DLPrwessel2013/12/18 02:06 PM
                                            Itanium and DLPExophase2013/12/20 09:27 AM
                                          Itanium and DLPDavid Kanter2013/12/20 10:58 AM
                                            Itanium and DLPEric Bron2013/12/21 02:18 AM
                                      The same reason why AMD insists on HSAAnon2013/12/27 12:24 PM
                                        Serial code isn't the alternativeDavid Kanter2013/12/27 03:43 PM
                                          Serial code is still the normAnon2013/12/27 09:24 PM
                                            Serial code is still the norm Eric Bron nli2013/12/28 02:18 AM
                                              Serial code is still the normAnon2013/12/29 01:01 AM
                                                Serial code is still the normEric Bron nli2013/12/29 02:55 AM
                                                  Serial code is still the normAnon2013/12/30 01:24 PM
                                            Serial code is still the normEduardoS2013/12/28 07:55 AM
                                              Serial code is still the normEric Bron nli2013/12/28 09:45 AM
                                                Serial code is still the normEduardoS2013/12/28 10:08 AM
                                                  Serial code is still the normEric Bron nli2013/12/28 10:34 AM
                                                    Serial code is still the normEduardoS2013/12/28 10:47 AM
                                                      Serial code is still the normEric Bron nli2013/12/28 11:13 AM
                                                        Serial code is still the normEduardoS2013/12/28 01:19 PM
                                                          Serial code is still the normEric Bron nli2013/12/28 01:43 PM
                                                          Serial code is still the normEric Bron nli2013/12/28 01:55 PM
                                              Gather will be faster. By how much is the real question.Anon2013/12/29 01:45 AM
                                                Gather will be faster. By how much is the real question.rwessel2013/12/29 02:49 AM
                                                Gather will be faster. By how much is the real question.Michael S2013/12/29 06:26 AM
                                                  Fast gather is vital for 512-bitAnon2013/12/30 01:49 PM
                                                    Fast gather is vital for 512-bitMichael S2013/12/31 02:29 AM
                                                      Fast gather is vital for 512-bitEric Bron nli2013/12/31 04:08 AM
                                                        Fast gather is vital for 512-bitMichael S2013/12/31 06:23 AM
                                                          Fast gather is vital for 512-bitEric Bron nli2013/12/31 06:53 AM
                                                            Fast gather is vital for 512-bitMichael S2013/12/31 09:39 AM
                                                              Fast gather is vital for 512-bitDavid Kanter2013/12/31 11:06 AM
                                                                Fast gather is vital for 512-bitMichael S2013/12/31 01:25 PM
                                                                  Fast gather is vital for 512-bitMichael S2013/12/31 01:45 PM
                                                      Fast gather is vital for 512-bitEduardoS2013/12/31 08:03 AM
                                                Gather will be faster. By how much is the real question.EduardoS2013/12/29 08:39 AM
                                                  Cache lines per cycle for gatherAnon2014/01/03 12:58 PM
                                                    Cache lines per cycle for gatherEduardoS2014/01/03 02:34 PM
                                                    Cache lines per cycle for gatherEric Bron2014/01/03 02:41 PM
                                        Homogeneous ISA is not a free lunch eitherPaul A. Clayton2013/12/27 05:42 PM
                                          Homogeneous ISA is a free lunch in absolute termsAnon2013/12/29 12:15 AM
                                      Itanium and DLPEric2013/12/27 04:12 PM
                                        Itanium and DLPDavid Kanter2013/12/27 09:09 PM
                                          Sky Lake and LRB coresEric2013/12/29 03:05 PM
                                            Sky Lake and LRB coresGroo2013/12/29 04:21 PM
                                              Stop posting gated linksDavid Kanter2013/12/29 04:44 PM
                                                Stop posting gated linksGroo2013/12/29 07:40 PM
                                    Itanium and DLPAnon2013/12/27 01:25 PM
                                  Itanium is not a GPUAnon2013/12/27 12:58 PM
                                    Itanium is not a GPUDavid Kanter2013/12/27 04:00 PM
                                      Itanium is not a GPUAnon2013/12/31 01:29 PM
                                        Itanium is not a GPUrwessel2014/01/01 05:13 AM
                                    Itanium is not a GPUrwessel2013/12/29 12:54 AM
                                Itanium and DLPEduardoS2013/12/17 05:31 PM
                                  Itanium and DLPNathan2013/12/18 09:47 AM
                                    Itanium and DLPKlimax2013/12/18 01:11 PM
                                      Itanium and DLPNathan2013/12/18 03:59 PM
                                        Itanium and DLPKlimax2013/12/21 03:28 PM
                                          Itanium and DLPNathan2013/12/23 08:39 AM
                                            Itanium and DLPKlimax2013/12/24 03:01 AM
                                              Itanium and DLPMaynard Handley2013/12/24 10:28 AM
                                                Itanium and DLPKlimax2013/12/26 04:17 PM
                                                  Itanium and DLPMichael S2013/12/27 03:41 AM
                                                    Itanium and DLPKlimax2013/12/27 04:51 AM
                                                    Itanium and DLPEric Bron nli2013/12/27 08:55 AM
                                                      editsEric Bron nli2013/12/27 09:01 AM
              SIMD and big.LITTLEDavid Kanter2013/12/02 06:51 PM
                SIMD and big.LITTLEJouni Osmala2013/12/02 11:21 PM
                  SIMD and big.LITTLEPatrick Chase2013/12/02 11:43 PM
                    SIMD and big.LITTLEKonrad Schwarz2013/12/03 01:47 AM
                      SIMD and big.LITTLESimon Farnsworth2013/12/03 04:22 AM
                        SIMD and big.LITTLEMichael S2013/12/03 04:48 AM
                        SIMD and big.LITTLEKonrad Schwarz2013/12/04 04:14 AM
                          SIMD and big.LITTLESimon Farnsworth2013/12/04 06:42 AM
                            SIMD and big.LITTLEKonrad Schwarz2013/12/04 08:54 AM
                              SIMD and big.LITTLESimon Farnsworth2013/12/04 11:32 AM
                          SIMD and big.LITTLEEduardoS2013/12/04 03:04 PM
                            SIMD and big.LITTLEDoug S2013/12/04 06:34 PM
                              SIMD and big.LITTLEEduardoS2013/12/04 07:42 PM
                                SIMD and big.LITTLEdmcq2013/12/05 05:59 AM
                                  SIMD and big.LITTLEExophase2013/12/05 08:01 AM
                    SIMD and big.LITTLEMichael S2013/12/03 05:06 AM
                      SIMD and big.LITTLEanonymou52013/12/03 08:12 AM
                      SIMD and big.LITTLEPatrick Chase2013/12/03 09:12 AM
                        SIMD and big.LITTLEMichael S2013/12/03 09:56 AM
                          SIMD and big.LITTLEExophase2013/12/03 10:03 AM
                    SIMD and big.LITTLEanon2013/12/03 05:43 AM
                      SIMD and big.LITTLEanon2013/12/03 05:49 AM
                    SIMD and big.LITTLEtmp9902013/12/03 06:01 AM
                SIMD and big.LITTLEtmp9902013/12/03 05:45 AM
                  SIMD and big.LITTLEPatrick Chase2013/12/03 09:32 AM
                    SIMD and big.LITTLEPatrick Chase2013/12/03 09:52 AM
                      SIMD and big.LITTLEtarlinian2013/12/03 10:24 AM
                        SIMD and big.LITTLEPatrick Chase2013/12/03 10:42 AM
                    SIMD and big.LITTLEtmp9902013/12/03 07:11 PM
                  SIMD and big.LITTLEExophase2013/12/03 09:55 AM
                    SIMD and big.LITTLEPatrick Chase2013/12/03 10:12 AM
                      SIMD and big.LITTLEExophase2013/12/03 10:31 AM
                        SIMD and big.LITTLEPatrick Chase2013/12/03 02:56 PM
                          SIMD and big.LITTLEPatrick Chase2013/12/03 03:08 PM
                        SIMD and big.LITTLEIvan2013/12/03 06:56 PM
                          SIMD and big.LITTLEWilco2013/12/03 08:05 PM
                          SIMD and big.LITTLEPatrick Chase2013/12/04 06:06 AM
                    SIMD and big.LITTLEtmp9902013/12/03 07:21 PM
                      SIMD and big.LITTLEWilco2013/12/03 08:22 PM
                        SIMD and big.LITTLEPurana Archer2013/12/04 01:05 AM
                          SIMD and big.LITTLEbakaneko2013/12/04 02:51 AM
                            SIMD and big.LITTLEdmcq2013/12/04 03:28 AM
                              SIMD and big.LITTLEbakaneko2013/12/04 10:20 PM
                          SIMD and big.LITTLEExophase2013/12/04 07:51 AM
                            SIMD and big.LITTLEtmp9902013/12/04 08:39 AM
                              Turbo vs rated frequencyDoug S2013/12/04 02:55 PM
                                Turbo vs rated frequencyUngo2013/12/05 03:06 PM
                                  Turbo vs rated frequencyWilco2013/12/05 04:15 PM
                            SIMD and big.LITTLEWilco2013/12/04 08:46 AM
                              SIMD and big.LITTLEUngo2013/12/05 04:18 PM
                                SIMD and big.LITTLEExophase2013/12/05 04:56 PM
                                  SIMD and big.LITTLEPatrick Chase2013/12/06 11:10 AM
                                SIMD and big.LITTLEWilco2013/12/05 06:05 PM
                                  SIMD and big.LITTLEPatrick Chase2013/12/06 11:53 AM
                                    NTV and DVFSPatrick Chase2013/12/06 01:12 PM
                                      NTV and DVFSPatrick Chase2013/12/07 12:22 AM
                                        NTV and DVFSWilco2013/12/07 02:47 PM
                                    SIMD and big.LITTLEWilco2013/12/07 05:33 PM
                                      SIMD and big.LITTLEDoug S2013/12/07 06:22 PM
                                  SIMD and big.LITTLEDavid Kanter2013/12/08 06:40 PM
                                    SIMD and big.LITTLEtarlinian2013/12/08 11:02 PM
                                SIMD and big.LITTLEPatrick Chase2013/12/06 11:27 AM
                                  SIMD and big.LITTLEExophase2013/12/06 11:52 AM
                                    SIMD and big.LITTLEExophase2013/12/06 12:00 PM
                                    SIMD and big.LITTLEPatrick Chase2013/12/06 12:12 PM
                                  SIMD and big.LITTLEUngo2013/12/11 03:08 PM
                          SIMD and big.LITTLEWilco2013/12/04 09:05 AM
                            SIMD and big.LITTLEDavid Kanter2013/12/06 10:50 PM
                              SIMD and big.LITTLEbakaneko2013/12/06 11:54 PM
                                SIMD and big.LITTLEPatrick Chase2013/12/07 10:42 AM
                                  SIMD and big.LITTLEExophase2013/12/07 11:23 AM
                              SIMD and big.LITTLEWilco2013/12/07 06:44 AM
                                SIMD and big.LITTLEDavid Kanter2013/12/07 11:23 PM
                          SIMD and big.LITTLEPatrick Chase2013/12/04 12:35 PM
                        SIMD and big.LITTLEtmp9902013/12/04 04:19 AM
                          SIMD and big.LITTLEspike2013/12/04 06:11 AM
                            SIMD and big.LITTLEtmp9902013/12/04 06:31 AM
                              I think POWER used confidence for dynamic predication not low-power stall (NT)Paul A. Clayton2013/12/04 06:51 AM
                          SIMD and big.LITTLEtmp9902013/12/04 06:27 AM
                            SIMD and big.LITTLEDoug S2013/12/04 03:07 PM
                              SIMD and big.LITTLEtmp9902013/12/04 08:04 PM
                                SIMD and big.LITTLErwessel2013/12/05 01:28 AM
                              SIMD and big.LITTLEPatrick Chase2013/12/06 12:33 PM
                          SIMD and big.LITTLEWilco2013/12/04 09:32 AM
                      SIMD and big.LITTLEExophase2013/12/04 07:42 AM
                        SIMD and big.LITTLEEduardoS2013/12/04 03:21 PM
                          SIMD and big.LITTLEExophase2013/12/04 10:29 PM
              The time to design a wide SIMD architectureAnon2013/12/05 12:30 AM
                The time to design a wide SIMD architectureDoug S2013/12/05 01:28 AM
                  The time to design a wide SIMD architectureExophase2013/12/05 03:19 PM
                    The time to design a wide SIMD architectureExophase2013/12/05 03:22 PM
                  Homogeneous wide SIMD necessary for mobileAnon2013/12/06 10:18 AM
                    Homogeneous wide SIMD necessary for mobilebakaneko2013/12/06 10:48 AM
                      Wider SIMD can lead to lower consumptionAnon2013/12/06 11:58 PM
                        Wider SIMD can lead to lower consumptionbakaneko2013/12/07 12:15 AM
                          Wider SIMD can lead to lower consumptionAnon2013/12/09 06:39 PM
                            Wider SIMD can lead to lower consumptionrwessel2013/12/09 09:47 PM
                              Wider SIMD can lead to lower consumptionAnon2013/12/10 08:15 PM
                            Wider SIMD can lead to lower consumptionEric Bron2013/12/09 11:31 PM
                              sidenote: hiding conditionals in arithmetichobold2013/12/10 10:04 AM
                                sidenote: hiding conditionals in arithmeticEric Bron2013/12/10 10:33 AM
                                  sidenote: hiding conditionals in arithmetichobold2013/12/10 04:39 PM
                                That's branching using just predicationAnon2013/12/10 08:45 PM
                                  That's branching using just predicationhobold2013/12/11 01:44 AM
                                    That's branching using just predicationAnon2013/12/11 02:10 PM
                                  That's branching using just predicationPatrick Chase2013/12/11 08:38 AM
                                    SIMD and memory latencyAnon2013/12/11 03:40 PM
                                      SIMD and memory latencyPatrick Chase2013/12/12 12:40 AM
                                    That's branching using just predicationanon2013/12/12 01:43 AM
                                      That's branching using just predicationPatrick Chase2013/12/12 11:41 PM
                                    That's branching using just predicationdmcq2013/12/12 04:17 PM
                              Efficient SIMD implementation of virtual callsAnon2013/12/10 08:30 PM
                                Efficient SIMD implementation of virtual callsEric Bron2013/12/11 12:24 AM
                                Efficient SIMD implementation of virtual callsExophase2013/12/11 10:37 AM
                                  Efficient SIMD implementation of virtual callsAnotherAnon2013/12/11 02:48 PM
                                    Efficient SIMD implementation of virtual callsExophase2013/12/11 03:17 PM
                                      Efficient SIMD implementation of virtual callsEric Bron2013/12/11 11:50 PM
                            Wider SIMD can lead to lower consumptionbakaneko2013/12/10 12:41 AM
                The time to design a wide SIMD architecturenone2013/12/05 02:02 AM
                  The time to design a wide SIMD architecturex2013/12/05 07:57 PM
    The need for more performance, or race to the bottomBrett2013/11/27 12:18 AM
      Slate 21 review: dismalAnon2013/11/28 09:39 AM
        Slate 21 review: dismalDoug S2013/11/28 01:07 PM
          Tablets not displacing PCs yetAnon2013/12/01 09:38 PM
            Tablets not displacing PCs yetDoug S2013/12/01 11:23 PM
        Slate 21 review: dismalBrett2013/11/29 12:47 AM
          Slate 21 review: dismalExophase2013/11/29 01:56 AM
    The need for more performanceDoug S2013/11/27 01:22 AM
      PC - death by profesionalization?mpx2013/12/03 06:41 AM
        PC - death by profesionalization?Michael S2013/12/03 09:39 AM
        PC - death by profesionalization?Doug S2013/12/03 01:50 PM
          PC - death by profesionalization?Gabriele Svelto2013/12/04 02:26 AM
            PC - death by profesionalization?Kevin G2013/12/04 08:23 AM
              PC - death by profesionalization?Doug S2013/12/04 03:25 PM
    The need for more performancetmp9902013/11/28 01:57 AM
      The need for more performanceDavid Kanter2013/11/28 12:23 PM
        The need for more performanceMichael S2013/11/28 12:46 PM
          ARMv8 not extensibleDavid Kanter2013/11/29 01:21 AM
            ARMv8 not extensibletmp9902013/11/29 02:53 AM
              ARMv8 not extensibleDavid Kanter2013/11/29 11:20 AM
                ARMv8 not extensibletmp9902013/11/29 09:13 PM
                  ARMv8 not extensibleDoug S2013/11/29 09:55 PM
                    ARMv8 not extensibleBrett2013/11/30 01:01 AM
                      ARMv8 not extensibleDoug S2013/11/30 02:09 PM
                        ARMv8 not extensibleExophase2013/11/30 02:40 PM
                      ARMv8 not extensibleEduardoS2013/11/30 03:22 PM
                        ARMv8 not extensibleBrett2013/12/01 03:25 AM
                          ARMv8 not extensibleEduardoS2013/12/01 12:49 PM
                    ARMv8 not extensibleNiels Jørgen Kruse2013/11/30 04:25 AM
                      ARMv8 not extensibleDoug S2013/11/30 02:02 PM
                ARMv8 not extensibleKevin G2013/12/02 01:07 PM
                  ARMv8 not extensibleDavid Kanter2013/12/03 10:36 AM
                    ARMv8 not extensibleExophase2013/12/03 10:52 AM
                      ARMv8 not extensibleKevin G2013/12/03 11:13 AM
                    ARMv8 not extensibleKevin G2013/12/03 11:08 AM
          The need for more performanceAnon2013/12/01 04:51 PM
        The need for more performancenone2013/11/28 12:54 PM
          The need for more performanceWilco2013/11/28 06:44 PM
            SIMD in the mobile worldAnon2013/12/01 10:31 PM
              SIMD in the mobile worldtmp9902013/12/02 05:11 AM
                SIMD is about way more than FMAsAnon2013/12/04 09:57 PM
                  SIMD is about way more than FMAsrwessel2013/12/06 04:47 AM
                    SIMD is about way more than FMAstmp9902013/12/06 07:48 AM
                      SIMD is about way more than FMAsExophase2013/12/06 10:03 AM
              SIMD in the mobile worldWilco2013/12/02 02:35 PM
                SIMD in the mobile worldEric Bron2013/12/03 11:31 PM
                  SIMD in the mobile worldnone2013/12/04 03:54 AM
                    SIMD in the mobile worldEric Bron2013/12/04 04:12 AM
                  SIMD in the mobile worldtmp9902013/12/04 04:23 AM
                  SIMD in the mobile worldWilco2013/12/04 10:49 AM
                    SIMD in the mobile worldMichael S2013/12/04 11:35 AM
                      SIMD in the mobile worldWilco2013/12/07 06:03 AM
                Where's ARM's wide SIMD?Anon2013/12/04 11:10 PM
                  Where's ARM's wide SIMD?tmp9902013/12/05 06:39 AM
                  Where's ARM's wide SIMD?Dan Fay2013/12/06 04:35 PM
                    How to implement gather/scatter efficientlyWilco2013/12/07 05:29 AM
                      How to implement gather/scatter efficientlyPatrick Chase2013/12/07 11:06 AM
                        How to implement gather/scatter efficientlyDan Fay2013/12/07 11:39 AM
                          How to implement gather/scatter efficientlyPatrick Chase2013/12/07 02:01 PM
                        How to implement gather/scatter efficientlyWilco2013/12/07 11:50 AM
                      Why implement gather/scatter efficiently?Michael S2013/12/08 12:13 PM
                        Why implement gather/scatter efficiently?Ricardo B2013/12/09 07:08 AM
                      Gather doesn't require 16 full addressesAnon2013/12/09 09:18 PM
                        Gather doesn't require 16 full addressesGabriele Svelto2013/12/10 03:38 AM
                          Logic wires scale just fineAnon2013/12/11 09:12 AM
                            Logic wires scale just fineDavid Kanter2013/12/11 10:05 AM
                              Logic wires scale just fineGabriele Svelto2013/12/11 10:44 AM
                                Logic wires scale just fineAnotherAnon2013/12/11 03:10 PM
                            Logic wires scale just fineslacker2013/12/12 11:23 PM
                              Logic wires scale just fineGabriele Svelto2013/12/13 02:41 AM
                                Logic wires scale just fineslacker2013/12/14 06:52 PM
                                  Logic wires scale just fineEric Bron2013/12/15 12:13 AM
                              Logic wires scale just finetarlinian2013/12/13 12:46 PM
                                Logic wires scale just fineslacker2013/12/14 06:45 PM
                        Gather doesn't require 16 full addressesWilco2013/12/10 04:42 AM
                      How to implement gather/scatter efficientlyEric Bron2013/12/10 10:09 AM
                    Cheap and fast gatherAnon2013/12/09 09:42 PM
                      Cheap and fast gatherPatrick Chase2013/12/10 06:55 PM
              SIMD in the mobile worldPatrick Chase2013/12/02 11:25 PM
                SIMD in the mobile worlddmcq2013/12/04 02:49 AM
                  SIMD in the mobile worldWilco2013/12/04 11:43 AM
        Was providing LPAE that stupid?Paul A. Clayton2013/11/28 06:16 PM
          Was providing LPAE that stupid?Linus Torvalds2013/11/29 01:34 PM
            Was providing LPAE that stupid?Paul A. Clayton2013/11/29 05:05 PM
              Oops! forgot the close 'em' for "I" (NT)Paul A. Clayton2013/11/29 05:06 PM
              Was providing LPAE that stupid?none2013/11/30 03:51 AM
            REAL use of PAE.Anon2013/11/29 10:01 PM
              32-bit code runs on 64-bit OS/CPU just fineanon2013/11/29 11:12 PM
                32-bit code runs on 64-bit OS/CPU just fineanon2013/11/30 04:18 PM
                  32-bit code runs on 64-bit OS/CPU just finerwessel2013/12/04 11:38 AM
              REAL use of PAE.Linus Torvalds2013/11/30 10:55 AM
                REAL use of PAE.Doug S2013/11/30 02:21 PM
                  REAL use of PAE.Linus Torvalds2013/11/30 04:30 PM
                    REAL use of PAE.Doug S2013/12/01 02:02 PM
                    REAL use of PAE.anon2013/12/01 09:58 PM
                      REAL use of PAE.Klimax2013/12/02 03:10 AM
                        REAL use of PAE.Michael S2013/12/02 06:29 AM
                          REAL use of PAE.Klimax2013/12/02 06:59 AM
                            REAL use of PAE.Michael S2013/12/02 07:43 AM
                              REAL use of PAE.Patrick Chase2013/12/02 09:06 PM
                                REAL use of PAE.Klimax2013/12/03 12:25 AM
                                REAL use of PAE.rwessel2013/12/04 10:37 AM
                                REAL use of PAE.etienne2013/12/05 03:42 AM
                                  REAL use of PAE.Patrick Chase2013/12/06 10:00 PM
                        REAL use of PAE.nksingh2013/12/02 03:37 PM
                          REAL use of PAE.EduardoS2013/12/02 04:24 PM
                            REAL use of PAE.nksingh2013/12/03 09:27 AM
                              REAL use of PAE.EduardoS2013/12/03 03:12 PM
                                REAL use of PAE.nksingh2013/12/03 06:05 PM
                                  REAL use of PAE.EduardoS2013/12/04 03:37 PM
                                    REAL use of PAE.Patrick Chase2013/12/07 11:18 AM
                              REAL use of PAE.rwessel2013/12/04 10:45 AM
                                REAL use of PAE.nksingh2013/12/04 12:30 PM
                                  REAL use of PAE.rwessel2013/12/04 02:55 PM
                                    REAL use of PAE.nksingh2013/12/05 12:37 AM
                      REAL use of PAE.tmp9902013/12/02 05:07 AM
                        compatibility, obsolescence, change, dependabilityhobold2013/12/02 07:36 AM
                        REAL use of PAE.Gabriele Svelto2013/12/02 07:47 AM
                          REAL use of PAE.Michael S2013/12/02 08:08 AM
                            REAL use of PAE.rwessel2013/12/04 11:04 AM
                    REAL use of PAE.Anon2013/12/02 07:12 PM
                      REAL use of PAE.Linus Torvalds2013/12/02 09:00 PM
                        REAL use of PAE.Yuhong Bao2013/12/02 09:15 PM
                          REAL use of PAE.Patrick Chase2013/12/02 10:24 PM
                            REAL use of PAE.Yuhong Bao2013/12/03 01:08 PM
                              REAL use of PAE.Patrick Chase2013/12/03 03:23 PM
                                REAL use of PAE.Ricardo B2013/12/03 03:52 PM
                                  REAL use of PAE.Patrick Chase2013/12/03 04:53 PM
                                    REAL use of PAE.Ricardo B2013/12/03 05:53 PM
                                  REAL use of PAE.tmp9902013/12/03 07:30 PM
                REAL use of PAE.Anon2013/11/30 04:24 PM
                  REAL use of PAE.Linus Torvalds2013/11/30 04:38 PM
                    REAL use of PAE.rwessel2013/12/04 12:05 PM
                      REAL use of PAE.Linus Torvalds2013/12/05 11:42 AM
                        REAL use of PAE.Doug S2013/12/06 02:39 AM
                          REAL use of PAE.Linus Torvalds2013/12/06 10:30 AM
                            REAL use of PAE.Yuhong Bao2013/12/06 04:04 PM
                        REAL use of PAE.Patrick Chase2013/12/06 12:43 PM
                          REAL use of PAE.Patrick Chase2013/12/06 12:50 PM
                            REAL use of PAE.anon2013/12/07 04:10 AM
                  REAL use of PAE.Klimax2013/12/01 01:05 AM
                    REAL use of PAE.nksingh2013/12/02 03:48 PM
                      REAL use of PAE.Konrad Schwarz2013/12/03 01:59 AM
                        REAL use of PAE.Gabriele Svelto2013/12/03 04:25 AM
                          REAL use of PAE.nksingh2013/12/03 09:31 AM
                            REAL use of PAE.Anon2013/12/04 07:24 PM
                              AWE is not PAE.Stubabe2013/12/05 09:13 AM
                                AWE is not PAE.Anon2013/12/07 02:26 AM
                                  AWE is not PAE.Patrick Chase2013/12/07 10:55 AM
                                    AWE is not PAE.Patrick Chase2013/12/07 01:01 PM
                                      AWE is not PAE.Wilco2013/12/07 01:46 PM
                                        AWE is not PAE.Stubabe2013/12/08 03:13 AM
                                  AWE is not PAE.Stubabe2013/12/08 03:24 AM
                    REAL use of PAE.Anon2013/12/02 07:13 PM
            Was providing LPAE that stupid?Robert David Graham2013/11/30 12:48 PM
              Was providing LPAE that stupid?Linus Torvalds2013/11/30 01:34 PM
                Was providing LPAE that stupid?bakaneko2013/11/30 06:45 PM
                Was providing LPAE that stupid?dmcq2013/11/30 06:53 PM
                Was providing LPAE that stupid?mircea2013/12/01 01:35 PM
                  Was providing LPAE that stupid?EduardoS2013/12/01 01:41 PM
                    Was providing LPAE that stupid?Michael S2013/12/01 01:51 PM
                      Was providing LPAE that stupid?NoSpammer2013/12/01 09:03 PM
                        Was providing LPAE that stupid?Klimax2013/12/02 03:14 AM
                          Was providing LPAE that stupid?Michael S2013/12/02 06:10 AM
                            Was providing LPAE that stupid?Klimax2013/12/02 07:03 AM
                              Was providing LPAE that stupid?Klimax2013/12/02 07:08 AM
                                Was providing LPAE that stupid?nksingh2013/12/02 03:59 PM
                              Was providing LPAE that stupid?rwessel2013/12/04 12:24 PM
            Was providing LPAE that stupid?etienne2013/12/02 03:07 AM
        The need for more performanceDoug S2013/11/29 12:49 AM
          The customer is always rightEmil Briggs2013/11/30 06:17 AM
          The need for more performanceAnon2013/12/01 05:03 PM
            The need for more performanceDoug S2013/12/01 05:50 PM
              The need for more performanceDavid Kanter2013/12/01 06:14 PM
                The need for more performanceDoug S2013/12/02 07:01 PM
                  The need for more performanceDavid Kanter2013/12/02 07:05 PM
                    The need for more performancedmcq2013/12/03 02:25 AM
        The need for more performancetmp9902013/11/29 03:32 AM
Reply to this Topic
Name:
Email:
Topic:
Body: No Text
How do you spell purple?