Re: Continuations

Liste des GroupesRevenir à c arch 
Sujet : Re: Continuations
De : mitchalsup (at) *nospam* aol.com (MitchAlsup1)
Groupes : comp.arch
Date : 18. Jul 2024, 00:17:37
Autres entêtes
Organisation : Rocksolid Light
Message-ID : <99f80e5c5452ec87cf6f5a70dcb33863@www.novabbs.org>
References : 1 2 3 4 5 6 7 8 9 10 11 12
User-Agent : Rocksolid Light
On Wed, 17 Jul 2024 20:56:06 +0000, Stephen Fuld wrote:

MitchAlsup1 wrote:
>
On Wed, 17 Jul 2024 18:30:47 +0000, Stephen Fuld wrote:
>
MitchAlsup1 wrote:
>
On Wed, 17 Jul 2024 16:50:27 +0000, Thomas Koenig wrote:
>
MitchAlsup1 <mitchalsup@aol.com> schrieb:
>
What I am talking about is to improve their performance until a
sin() takes about the same number of cycles of FDIV, not 10×
more.
>
Maybe time for a little story.
>
Some unspecified time ago, a colleague did CFD calculations
which included fluid flow (including turbulence modelling and
diffusion) and quite a few chemical reactions together.  So, he
evaluated a huge number of Arrhenius equations,
>
k = A * exp(-E_a/(R*T))
>
and because some of the reactions he looked at were highly
exothermic or endothermic, he needed tiny relaxation factors
(aka small steps).  His calculaiton spent most of the time
evaluating the Arrhenius equation above many, many, many, many
times.
>
A single calculation took months, and he didn't use weak
hardware.
>
A fully pipelined evaluation of, let's say, four parallel exp
and four parallel fdiv instructions would have reduced his
calculation time by orders of magnitude, and allowed him to
explore the design space instead of just scratching the surface.
>
(By the way, if I had found a reasonable way to incorporate the
Arrhenius equation into your ISA, I would have done so already
:-)
>
   FMUL     Rt,RR,RT
   FDIV     Rt,-RE,Rt
   EXP      Rt,Rt
   FMUL     Rk,RA,Rt
>
Does not look "all that bad" to me.
>
So for your GbOoO CPU, how many of the various FP operations, and
the EXP instruction can be done in parallel?
>
FMUL is   4 cycles of latency fully pipelined
FDIV is ~20 cycles of latency not   pipelined
EXP  is ~16 cycles of latency not   pipelined
>
They are all performed in the FMAC unit and here the instructions are
serially dependent.
>
So, 44 cycles of latency, a 1-wide machine and a 6-wide machine would
see the same latency; that is, GBOoO is not a differentiator.
>
>
Good, I get that.  But Thomas' original discussion of the problem
indicated that it was very parallel, so the question is, in your
design, how many of those calculations can go in in parallel?
The FDIV and EXP instructions consume all the FMAC cycles, so even if
you completely unrolled the loop, you are not going to get more than
6-cycles less in performing the repeated calculations.
A really BIG implementation with 4 FMAC units per core could unroll
the loop (by reservation stations) such that each iteration would
still be 44-cycles, but you could run 4 in parallel and achieve 4
results every 44-cycles--which to most people smells like 1 result
every 11-cycles.
{Would be an interesting reservation station design, though}
>
>
>

Date Sujet#  Auteur
13 Jul 24 * Continuations138Lawrence D'Oliveiro
13 Jul 24 +* Re: Continuations4BGB
14 Jul 24 i+* Re: Continuations2aph
15 Jul 24 ii`- Re: Continuations1Lawrence D'Oliveiro
14 Jul 24 i`- Re: Continuations1Anton Ertl
13 Jul 24 +* Re: Continuations23John Dallman
14 Jul 24 i+* Re: Continuations21Lawrence D'Oliveiro
14 Jul 24 ii`* Re: Continuations20George Neuner
14 Jul 24 ii `* Re: Continuations19John Levine
14 Jul 24 ii  `* Re: Continuations18Niklas Holsti
15 Jul 24 ii   +* Re: Continuations16John Levine
15 Jul 24 ii   i+- Re: Continuations1Terje Mathisen
15 Jul 24 ii   i+- Re: Continuations1John Levine
15 Jul 24 ii   i+* Re: Continuations9Niklas Holsti
16 Jul 24 ii   ii`* Re: Continuations8Lawrence D'Oliveiro
16 Jul 24 ii   ii `* Re: Continuations7John Levine
16 Jul 24 ii   ii  +- Re: Continuations1Chris M. Thomasson
16 Jul 24 ii   ii  `* Re: Continuations5Lawrence D'Oliveiro
16 Jul 24 ii   ii   `* Re: Continuations4John Levine
16 Jul 24 ii   ii    `* Re: Continuations3Lawrence D'Oliveiro
16 Jul 24 ii   ii     `* Re: Continuations2MitchAlsup1
17 Jul 24 ii   ii      `- Re: Continuations1Lawrence D'Oliveiro
16 Jul 24 ii   i+* Re: Continuations3Lawrence D'Oliveiro
16 Jul 24 ii   ii`* Re: Continuations2MitchAlsup1
16 Jul 24 ii   ii `- Re: Continuations1Lawrence D'Oliveiro
16 Jul 24 ii   i`- Re: Continuations1MitchAlsup1
16 Jul 24 ii   `- Re: Continuations1Lawrence D'Oliveiro
14 Jul 24 i`- Re: Continuations1BGB
13 Jul 24 +- Re: Continuations1BGB
14 Jul 24 +* Re: Continuations10Lawrence D'Oliveiro
15 Jul 24 i+* Re: Continuations7Thomas Koenig
15 Jul 24 ii`* Re: Continuations6Thomas Koenig
16 Jul 24 ii +* Re: Continuations4Thomas Koenig
16 Jul 24 ii i+* Re: Continuations2MitchAlsup1
17 Jul 24 ii ii`- Re: Continuations1Lawrence D'Oliveiro
17 Jul 24 ii i`- Re: Continuations1Lawrence D'Oliveiro
17 Jul 24 ii `- Re: Continuations1John Dallman
16 Jul 24 i+- Re: Continuations1Lawrence D'Oliveiro
16 Jul 24 i`- Re: Continuations1John Levine
14 Jul 24 +- Re: Continuations1George Neuner
14 Jul 24 +* Re: Continuations92John Savard
14 Jul 24 i+- Re: Continuations1BGB
15 Jul 24 i`* Re: Continuations90Lawrence D'Oliveiro
16 Jul 24 i `* Re: Continuations89John Savard
16 Jul 24 i  +* Re: Continuations2MitchAlsup1
17 Jul 24 i  i`- Re: Continuations1Lawrence D'Oliveiro
16 Jul 24 i  `* Re: Continuations86MitchAlsup1
17 Jul 24 i   +* Re: Continuations69John Savard
17 Jul 24 i   i`* Re: Continuations68MitchAlsup1
17 Jul 24 i   i `* Re: Continuations67Thomas Koenig
17 Jul 24 i   i  +- Re: Continuations1Thomas Koenig
17 Jul 24 i   i  +- Re: Continuations1Michael S
17 Jul 24 i   i  +* Re: Continuations37MitchAlsup1
17 Jul 24 i   i  i`* Re: Continuations36Stephen Fuld
17 Jul 24 i   i  i `* Re: Continuations35MitchAlsup1
17 Jul 24 i   i  i  +* Re: Continuations22Stephen Fuld
18 Jul 24 i   i  i  i+* Re: Continuations8MitchAlsup1
18 Jul 24 i   i  i  ii+- Re: Continuations1Michael S
18 Jul 24 i   i  i  ii`* Re: Continuations6MitchAlsup1
19 Jul 24 i   i  i  ii +- Re: Continuations1Stephen Fuld
21 Jul 24 i   i  i  ii +* Re: Reservation stations [was Continuations]2Anton Ertl
21 Jul 24 i   i  i  ii i`- Re: Reservation stations [was Continuations]1MitchAlsup1
21 Jul 24 i   i  i  ii `* Re: Reservation stations [was Continuations]2MitchAlsup1
22 Jul 24 i   i  i  ii  `- IPC (was: Reservation stations)1Anton Ertl
18 Jul 24 i   i  i  i+* Re: Continuations11Thomas Koenig
18 Jul 24 i   i  i  ii`* Re: Continuations10Michael S
18 Jul 24 i   i  i  ii `* Re: Continuations9Thomas Koenig
18 Jul 24 i   i  i  ii  `* Re: Continuations8Michael S
18 Jul 24 i   i  i  ii   +* Re: Continuations6Thomas Koenig
18 Jul 24 i   i  i  ii   i+- Re: Continuations1Michael S
18 Jul 24 i   i  i  ii   i`* Re: Continuations4Michael S
19 Jul 24 i   i  i  ii   i `* Re: Continuations3Thomas Koenig
19 Jul 24 i   i  i  ii   i  `* Re: Continuations2Michael S
20 Jul 24 i   i  i  ii   i   `- Re: Continuations1Thomas Koenig
18 Jul 24 i   i  i  ii   `- Re: Continuations1MitchAlsup1
18 Jul 24 i   i  i  i`* Re: Continuations2John Savard
18 Jul 24 i   i  i  i `- Re: Continuations1Thomas Koenig
18 Jul 24 i   i  i  +* Re: Continuations6Thomas Koenig
18 Jul 24 i   i  i  i`* Re: Continuations5Michael S
18 Jul 24 i   i  i  i `* Re: Continuations4Michael S
18 Jul 24 i   i  i  i  `* Re: Continuations3Thomas Koenig
18 Jul 24 i   i  i  i   `* Re: Continuations2MitchAlsup1
20 Jul 24 i   i  i  i    `- Re: Continuations1Thomas Koenig
18 Jul 24 i   i  i  +* Non-pipelined FDIV/SQRT (was: Continuations)3Stefan Monnier
18 Jul 24 i   i  i  i+- Re: Non-pipelined FDIV/SQRT1MitchAlsup1
28 Jul 24 i   i  i  i`- Re: Non-pipelined FDIV/SQRT1Michael S
18 Jul 24 i   i  i  `* Re: Continuations3MitchAlsup1
28 Jul 24 i   i  i   `* Re: Continuations2Paul A. Clayton
28 Jul 24 i   i  i    `- Re: Continuations1Michael S
19 Jul 24 i   i  `* Re: Continuations27Terje Mathisen
19 Jul 24 i   i   +* Re: Continuations5Thomas Koenig
19 Jul 24 i   i   i+- Re: Continuations1Chris M. Thomasson
19 Jul 24 i   i   i`* Re: Continuations3MitchAlsup1
20 Jul 24 i   i   i +- Re: Continuations1Terje Mathisen
20 Jul 24 i   i   i `- Re: Continuations1Thomas Koenig
19 Jul 24 i   i   `* Re: Continuations21MitchAlsup1
19 Jul 24 i   i    +* Re: Continuations8Terje Mathisen
22 Jul 24 i   i    i`* Re: Continuations7Michael S
22 Jul 24 i   i    i +* Re: Continuations3MitchAlsup1
22 Jul 24 i   i    i i`* Re: Continuations2Michael S
23 Jul 24 i   i    i i `- Re: Continuations1MitchAlsup1
23 Jul 24 i   i    i `* Re: Continuations3Terje Mathisen
19 Jul 24 i   i    `* Faster div or 1/sqrt approximations (was: Continuations)12Thomas Koenig
17 Jul 24 i   +* Re: Continuations3Lawrence D'Oliveiro
17 Jul 24 i   +* Re: Continuations12Stephen Fuld
17 Jul 24 i   `- Re: fancy instructions, Continuations1John Levine
15 Jul 24 +- Re: Continuations1wolfgang kern
15 Jul 24 +* Re: pessimal storage allocation, Continuations3John Levine
15 Jul 24 +- Re: Continuations1MitchAlsup1
16 Jul 24 `- Re: Continuations1Lynn Wheeler

Haut de la page

Les messages affichés proviennent d'usenet.

NewsPortal