Re: Computer architects leaving Intel...

Liste des GroupesRevenir à c arch 
Sujet : Re: Computer architects leaving Intel...
De : paaronclayton (at) *nospam* gmail.com (Paul A. Clayton)
Groupes : comp.arch
Date : 22. Sep 2024, 21:43:38
Autres entêtes
Organisation : A noiseless patient Spider
Message-ID : <vcpvhs$2bgj0$1@dont-email.me>
References : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24
User-Agent : Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101 Thunderbird/91.0
On 9/19/24 11:07 AM, EricP wrote:
[snip]
If the multiplier is pipelined with a latency of 5 and throughput of 1,
then MULL takes 5 cycles and MULL,MULH takes 6.
 But those two multiplies still are tossing away 50% of their work.
I do not remember how multipliers are actually implemented — and
am not motivated to refresh my memory at the moment — but I
thought a multiply low would not need to generate the upper bits,
so I do not understand where your "50% of their work" is coming
from.
The high result needs the low result carry-out but not the rest of
the result. (An approximate multiply high for multiply by
reciprocal might be useful, avoiding the low result work. There
might also be ways that a multiplier could be configured to also
provide bit mixing similar to middle result for generating a
hash?)
I seem to recall a PowerPC implementation did semi-pipelined 32-
bit multiplication 16-bits at a time. This presumably saved area
and power while also facilitating early out for small
multiplicands, at the cost of some latency and substantial
throughput compared to a fully pipelined multiplication. If I
remember correctly, this produced a result for 16-bit by 32-bit
multiplication, which is different from generating a low or high
result.

And if it does fuse them then the internal uArch cost is the same as if
you had designed it optimally from the start, except now you have
to pay for a fuser.
 <sound of soap box being dragged out>
This idea that macro-op fusion is some magic solution is bullshit.
1) It's not free.
Neither is increasing the number of opcodes or providing extender
prefixes. If one wants binary compatibility, non-fusing
implementations would work.
(I tend to favor providing a translation layer between software
distribution format and instruction cache format, which reduces
the binary compatibility constraint.)

2) It only works where Decode can see *all* the required lookahead
    instructions, which means you have to pay for an N-lane decoder
    but only get 1 lane.
Most fusion is for two adjacent instructions, which significantly
limits the complexity. The fusable patterns are also a subset of
all pairs of two instructions, so complete two-way decoding may
not be needed.
There may also be optimization opportunities from looking ahead.
Mitch Alsup proposed such for branch handling in a scalar
implementation. Apart from fusion, there might be advantages for
avoiding bank conflicts in a banked register file. I.e., the cost
of lookahead might be shared by multiple techniques/optimizations.
I tend to agree that fusion tends to be a workaround for sub-
optimal instruction encoding, but it seems that encoding involves
a lot of tradeoffs.

3) It's probabilistic as it depends on how the fetch buffers get loaded.
    Eg if the fetch buffer contains a valid instruction but does not have
    a next instruction, do you stall Decode to see if a fuser might arrive
    or dispatch it anyway.
This is also somewhat true for variable length encodings that cross fetch boundaries. In general a boundary-crossing instruction
would probably stall even if such was not strictly necessary
(e.g., if the missing information is opcode refinement — not
related to instruction routing — or an immediate or even a
register source identifier specifying a value that can have
delayed use (e.g., value of a store, addend of a FMADD).
This does seem a weakness, but fusion is not entirely negative
factors.

4) It gets exponentially expensive if you start doing multiple instruction
    lanes because decode has to deal with all the permutations of
    fusion possibilities.
This is also a factor in mere superscalar decode/execute.
Detecting that an instruction is dependent on another would
normally stall the execution of that instruction.
(I feel that encoding some of the dependency information could
be useful to avoid some of this work. In theory, common
dependency detection could also be more broadly useful; e.g.,
operand availability detection and execution/operand routing.)

5) Any fused instructions leave (multiple) bubbles that should be
    compacted out or there wasn't much point to doing the fusion.
Even with reduced operations per cycle, fusion could still provide
a net energy benefit.

In my opinion it is better to have an ISA that is optimal by design
rather than being patched up by fusion later.
Fusion is mostly presented for "patching up", but there are also
considerations of diverse microarchitectures. With pre-fused
instructions, an implementation might need to crack some of those
instructions. Software optimized for such an implementation might
also prefer more flexible compile-time scheduling of pre-cracked
operations.
A load-op instruction is perhaps particularly difficult because
one needs frequent stalls, a skewed (or second chance) pipeline to
hide the load latency, out-of-order execution, or some other stall
avoidance mechanism.
There are also constraints in encoding granularity.

Some of this inefficiency is caused by clinging to now 40 year old
risc design *guidelines* (ie not even rules) that:
- instructions have at most 1 dest and 2 source registers
FMADD seems to have mostly killed the 2-source limit. AArch64's
paired load removes the 2 destination limit. (Paired destinations
were common for early double precision implementations.)

- register specifier fields are either source or dest, never both
This seems mostly a code density consideration. I think using a
single name for both a source and a destination is not so
horrible, but I am not a hardware guy.

- instructions should take at most 1 clock (they never did)
That was clearly overconstraining.

These self imposed design restrictions cause ISA designers to miss
some possible more optimal solutions. The result is things like
RISC-V's memory reference linkage structures taking 6 instructions
to build a 64-bit PC-relative address. And I'm pretty sure we won't
see any 6 instruction fusers for quite some time.
I very much doubt a compiler would generate such outside of some
real-time application where the time constancy might justify the
code bloat.

<sound of soap box being dragged back to cupboard>
I do not mean my response to be heckling. Your points are very
true. However, I think fusion is a technique — like cracking —
that is a natural part of an architect's toolbox.

Date Sujet#  Auteur
27 Aug 24 * Computer architects leaving Intel...539Thomas Koenig
27 Aug 24 +- Re: Computer architects leaving Intel...1Michael S
27 Aug 24 +- Re: Computer architects leaving Intel...1Stephen Fuld
27 Aug 24 `* Re: Computer architects leaving Intel...536John Dallman
28 Aug 24  +* Re: Computer architects leaving Intel...529BGB
28 Aug 24  i`* Re: Computer architects leaving Intel...528MitchAlsup1
28 Aug 24  i `* Re: Computer architects leaving Intel...527BGB
28 Aug 24  i  +* Re: Computer architects leaving Intel...2Robert Finch
28 Aug 24  i  i`- Re: Computer architects leaving Intel...1BGB
28 Aug 24  i  `* Re: Computer architects leaving Intel...524MitchAlsup1
29 Aug 24  i   `* Re: Computer architects leaving Intel...523BGB
29 Aug 24  i    +* Re: Computer architects leaving Intel...511MitchAlsup1
29 Aug 24  i    i`* Re: Computer architects leaving Intel...510BGB
30 Aug 24  i    i +* Re: Computer architects leaving Intel...499John Dallman
30 Aug 24  i    i i+* Re: Computer architects leaving Intel...11Thomas Koenig
30 Aug 24  i    i ii+- Re: Computer architects leaving Intel...1Michael S
30 Aug 24  i    i ii+* Re: Computer architects leaving Intel...8Anton Ertl
30 Aug 24  i    i iii+* Re: Computer architects leaving Intel...2Michael S
30 Aug 24  i    i iiii`- Re: Computer architects leaving Intel...1Anton Ertl
30 Aug 24  i    i iii`* Re: Computer architects leaving Intel...5John Dallman
30 Aug 24  i    i iii `* Re: Computer architects leaving Intel...4Brett
30 Aug 24  i    i iii  +- Re: Computer architects leaving Intel...1John Dallman
2 Sep 24  i    i iii  `* Re: Computer architects leaving Intel...2Terje Mathisen
2 Sep 24  i    i iii   `- Re: Computer architects leaving Intel...1Thomas Koenig
30 Aug 24  i    i ii`- Re: Computer architects leaving Intel...1BGB
30 Aug 24  i    i i`* Re: Computer architects leaving Intel...487Anton Ertl
30 Aug 24  i    i i +* Re: Computer architects leaving Intel...302John Dallman
30 Aug 24  i    i i i`* Re: Computer architects leaving Intel...301David Brown
30 Aug 24  i    i i i +* Re: Computer architects leaving Intel...293Anton Ertl
30 Aug 24  i    i i i i`* Re: Computer architects leaving Intel...292Bernd Linsel
31 Aug 24  i    i i i i +- Re: Computer architects leaving Intel...1Thomas Koenig
31 Aug 24  i    i i i i `* Re: Computer architects leaving Intel...290Thomas Koenig
31 Aug 24  i    i i i i  +- Re: Computer architects leaving Intel...1Thomas Koenig
31 Aug 24  i    i i i i  `* Re: Computer architects leaving Intel...288Bernd Linsel
31 Aug 24  i    i i i i   +- Re: Computer architects leaving Intel...1Thomas Koenig
31 Aug 24  i    i i i i   +* Re: Computer architects leaving Intel...2Thomas Koenig
31 Aug 24  i    i i i i   i`- Re: Computer architects leaving Intel...1Bernd Linsel
31 Aug 24  i    i i i i   `* Re: Computer architects leaving Intel...284Anton Ertl
31 Aug 24  i    i i i i    +* Re: Computer architects leaving Intel...279Thomas Koenig
31 Aug 24  i    i i i i    i+* Re: Computer architects leaving Intel...157Bernd Linsel
31 Aug 24  i    i i i i    ii+* Re: Computer architects leaving Intel...153MitchAlsup1
1 Sep 24  i    i i i i    iii`* Re: Computer architects leaving Intel...152Stephen Fuld
2 Sep 24  i    i i i i    iii `* Re: Computer architects leaving Intel...151Terje Mathisen
2 Sep 24  i    i i i i    iii  `* Re: Computer architects leaving Intel...150Stephen Fuld
3 Sep 24  i    i i i i    iii   +* Re: Computer architects leaving Intel...139David Brown
3 Sep 24  i    i i i i    iii   i+* Re: Computer architects leaving Intel...108Stephen Fuld
4 Sep 24  i    i i i i    iii   ii`* Re: Computer architects leaving Intel...107David Brown
4 Sep 24  i    i i i i    iii   ii +* Re: Computer architects leaving Intel...103Terje Mathisen
4 Sep 24  i    i i i i    iii   ii i+* Re: Computer architects leaving Intel...101David Brown
4 Sep 24  i    i i i i    iii   ii ii+* Re: Computer architects leaving Intel...97jseigh
4 Sep 24  i    i i i i    iii   ii iii`* Re: Computer architects leaving Intel...96David Brown
4 Sep 24  i    i i i i    iii   ii iii `* Re: Computer architects leaving Intel...95Brett
4 Sep 24  i    i i i i    iii   ii iii  +- Re: Computer architects leaving Intel...1Thomas Koenig
4 Sep 24  i    i i i i    iii   ii iii  +- Re: Computer architects leaving Intel...1MitchAlsup1
5 Sep 24  i    i i i i    iii   ii iii  +* Re: Computer architects leaving Intel...8BGB
5 Sep 24  i    i i i i    iii   ii iii  i`* Re: Computer architects leaving Intel...7MitchAlsup1
5 Sep 24  i    i i i i    iii   ii iii  i `* Re: Computer architects leaving Intel...6David Brown
5 Sep 24  i    i i i i    iii   ii iii  i  `* Re: Computer architects leaving Intel...5Niklas Holsti
5 Sep 24  i    i i i i    iii   ii iii  i   `* Re: Computer architects leaving Intel...4David Brown
6 Sep 24  i    i i i i    iii   ii iii  i    `* Re: Computer architects leaving Intel...3BGB
6 Sep 24  i    i i i i    iii   ii iii  i     `* Re: Computer architects leaving Intel...2David Brown
9 Sep 24  i    i i i i    iii   ii iii  i      `- Re: Computer architects leaving Intel...1BGB
5 Sep 24  i    i i i i    iii   ii iii  +* Re: Computer architects leaving Intel...83David Brown
5 Sep 24  i    i i i i    iii   ii iii  i`* Re: Computer architects leaving Intel...82Terje Mathisen
5 Sep 24  i    i i i i    iii   ii iii  i +* Re: Computer architects leaving Intel...79David Brown
5 Sep 24  i    i i i i    iii   ii iii  i i+* Re: Computer architects leaving Intel...2Thomas Koenig
7 Sep 24  i    i i i i    iii   ii iii  i ii`- Re: Computer architects leaving Intel...1Tim Rentsch
5 Sep 24  i    i i i i    iii   ii iii  i i+* Re: Computer architects leaving Intel...74Terje Mathisen
5 Sep 24  i    i i i i    iii   ii iii  i ii+* Re: Computer architects leaving Intel...16David Brown
9 Sep 24  i    i i i i    iii   ii iii  i iii`* Re: Computer architects leaving Intel...15Terje Mathisen
9 Sep 24  i    i i i i    iii   ii iii  i iii +* Re: Computer architects leaving Intel...12David Brown
9 Sep 24  i    i i i i    iii   ii iii  i iii i`* Re: Computer architects leaving Intel...11Brett
10 Sep 24  i    i i i i    iii   ii iii  i iii i +* Re: Computer architects leaving Intel...5Terje Mathisen
10 Sep 24  i    i i i i    iii   ii iii  i iii i i`* Re: Computer architects leaving Intel...4Brett
10 Sep 24  i    i i i i    iii   ii iii  i iii i i +* Re: Computer architects leaving Intel...2Michael S
11 Sep 24  i    i i i i    iii   ii iii  i iii i i i`- Re: Computer architects leaving Intel...1Brett
11 Sep 24  i    i i i i    iii   ii iii  i iii i i `- Re: Computer architects leaving Intel...1Terje Mathisen
10 Sep 24  i    i i i i    iii   ii iii  i iii i `* Re: Computer architects leaving Intel...5David Brown
10 Sep 24  i    i i i i    iii   ii iii  i iii i  +* Re: Computer architects leaving Intel...3Anton Ertl
10 Sep 24  i    i i i i    iii   ii iii  i iii i  i`* Re: Computer architects leaving Intel...2David Brown
10 Sep 24  i    i i i i    iii   ii iii  i iii i  i `- Re: Computer architects leaving Intel...1Stefan Monnier
10 Sep 24  i    i i i i    iii   ii iii  i iii i  `- Re: Computer architects leaving Intel...1BGB
9 Sep 24  i    i i i i    iii   ii iii  i iii `* Re: Computer architects leaving Intel...2Michael S
10 Sep 24  i    i i i i    iii   ii iii  i iii  `- Re: Computer architects leaving Intel...1Michael S
5 Sep 24  i    i i i i    iii   ii iii  i ii+* Re: Computer architects leaving Intel...45Bernd Linsel
6 Sep 24  i    i i i i    iii   ii iii  i iii+- Re: Computer architects leaving Intel...1David Brown
9 Sep 24  i    i i i i    iii   ii iii  i iii+* Re: Computer architects leaving Intel...2Terje Mathisen
9 Sep 24  i    i i i i    iii   ii iii  i iiii`- Re: Computer architects leaving Intel...1Tim Rentsch
14 Sep 24  i    i i i i    iii   ii iii  i iii`* Re: Computer architects leaving Intel...41Kent Dickey
14 Sep 24  i    i i i i    iii   ii iii  i iii +* Re: Computer architects leaving Intel...32Anton Ertl
14 Sep 24  i    i i i i    iii   ii iii  i iii i+* Re: Computer architects leaving Intel...29MitchAlsup1
14 Sep 24  i    i i i i    iii   ii iii  i iii ii`* Re: Computer architects leaving Intel...28Thomas Koenig
15 Sep 24  i    i i i i    iii   ii iii  i iii ii `* Re: Computer architects leaving Intel...27David Brown
16 Sep 24  i    i i i i    iii   ii iii  i iii ii  +* Re: Computer architects leaving Intel...5Thomas Koenig
16 Sep 24  i    i i i i    iii   ii iii  i iii ii  i`* Re: Computer architects leaving Intel...4David Brown
16 Sep 24  i    i i i i    iii   ii iii  i iii ii  i `* Re: Computer architects leaving Intel...3Thomas Koenig
17 Sep 24  i    i i i i    iii   ii iii  i iii ii  i  +- Re: Upwards and downwards compatible, Computer architects leaving Intel...1John Levine
17 Sep 24  i    i i i i    iii   ii iii  i iii ii  i  `- Re: Computer architects leaving Intel...1David Brown
16 Sep 24  i    i i i i    iii   ii iii  i iii ii  `* Re: Computer architects leaving Intel...21Terje Mathisen
16 Sep 24  i    i i i i    iii   ii iii  i iii ii   `* Re: Computer architects leaving Intel...20David Brown
16 Sep 24  i    i i i i    iii   ii iii  i iii ii    +* Re: Computer architects leaving Intel...14Michael S
17 Sep 24  i    i i i i    iii   ii iii  i iii ii    `* Re: Computer architects leaving Intel...5Terje Mathisen
15 Sep 24  i    i i i i    iii   ii iii  i iii i`* Re: Computer architects leaving Intel...2BGB
14 Sep 24  i    i i i i    iii   ii iii  i iii +* Re: Computer architects leaving Intel...3Thomas Koenig
16 Sep 24  i    i i i i    iii   ii iii  i iii `* Re: Computer architects leaving Intel...5Tim Rentsch
6 Sep 24  i    i i i i    iii   ii iii  i ii+* Re: Computer architects leaving Intel...3Tim Rentsch
7 Sep 24  i    i i i i    iii   ii iii  i ii`* Re: Computer architects leaving Intel...9Chris M. Thomasson
5 Sep 24  i    i i i i    iii   ii iii  i i`* Re: Computer architects leaving Intel...2MitchAlsup1
5 Sep 24  i    i i i i    iii   ii iii  i `* Re: Computer architects leaving Intel...2MitchAlsup1
7 Sep 24  i    i i i i    iii   ii iii  `- Re: Computer architects leaving Intel...1Tim Rentsch
4 Sep 24  i    i i i i    iii   ii ii`* Re: Computer architects leaving Intel...3Thomas Koenig
6 Sep 24  i    i i i i    iii   ii i`- Re: Computer architects leaving Intel...1Chris M. Thomasson
4 Sep 24  i    i i i i    iii   ii +- Re: Computer architects leaving Intel...1jseigh
13 Sep 24  i    i i i i    iii   ii `* Re: Computer architects leaving Intel...2Stephen Fuld
3 Sep 24  i    i i i i    iii   i`* Re: Computer architects leaving Intel...30Stefan Monnier
3 Sep 24  i    i i i i    iii   `* Re: Computer architects leaving Intel...10Terje Mathisen
31 Aug 24  i    i i i i    ii`* Re: Computer architects leaving Intel...3Thomas Koenig
1 Sep 24  i    i i i i    i`* Re: Computer architects leaving Intel...121David Brown
1 Sep 24  i    i i i i    +* Re: Computer architects leaving Intel...3John Dallman
3 Sep 24  i    i i i i    `- Re: Computer architects leaving Intel...1Stefan Monnier
30 Aug 24  i    i i i +- Re: Computer architects leaving Intel...1MitchAlsup1
30 Aug 24  i    i i i +* Re: Computer architects leaving Intel...4Stefan Monnier
30 Aug 24  i    i i i `* Re: Computer architects leaving Intel...2John Dallman
8 Sep 24  i    i i `* Re: Computer architects leaving Intel...184Tim Rentsch
30 Aug 24  i    i `* Re: Computer architects leaving Intel...10MitchAlsup1
31 Aug 24  i    `* Re: Computer architects leaving Intel...11Paul A. Clayton
29 Aug 24  `* Re: Computer architects leaving Intel...6Anton Ertl

Haut de la page

Les messages affichés proviennent d'usenet.

NewsPortal