Re: Computer architects leaving Intel...

Liste des GroupesRevenir à c arch 
Sujet : Re: Computer architects leaving Intel...
De : ggtgp (at) *nospam* yahoo.com (Brett)
Groupes : comp.arch
Date : 18. Sep 2024, 22:15:55
Autres entêtes
Organisation : A noiseless patient Spider
Message-ID : <vcffub$77jk$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
User-Agent : NewsTap/5.5 (iPad)
EricP <ThatWouldBeTelling@thevillage.com> wrote:
Terje Mathisen wrote:
EricP wrote:
 
Codecs likely have to deal with double-width straddles a lot, whatever
the register word size. So for them it likely happens at 64-bits already.
 
Nothing likely about it: LZ4 is pretty much the only compression
algorithm/lossless codec that never straddles, all the rest tend to
treat the source data as single bitstream of arbitrary length, except
for some built-in chunking mechanism which simplifies faster scanning.
 
The core of the algorithm always starts with knowing the endianness,
then picking up 32 or 64-bit chunks of input data (byte-flipping if
needed) and then extractin the next N bits either from the top of bottom
of the buffer register.
 
AlLmost by definition, this is not code that a compiler is setup to help
you get correct.
 
 
I added a bunch of instructions for dealing with double-width operations.
The main ISA design decision is whether to have register pair specifiers,
R0, R2, R4,... or two separate {r_high,r_low} registers.
In either case the main uArch issue is that now instructions have an
extra
source register and two dest registers, which has a number of
consequences.
But once you bite the bullet on that it simplifies a lot of things,
like how to deal with carry or overflow without flags,
full width multiplies, divide producing both quotient and remainder.
 
Very nice!
 
This means that you can do integer IMAC(), right?
 
(hi, lo) = imac(a, b, c); // == a*b+c
 
The only thing even nicer from the perspective of writing arbitrary
precision library code would be IMAA, i.e. a*b+c+d since that is the
largest combination which is guaranteed to never overflow the double
register target field.
 
Terje
 
 
I thought about IMAC but it was a bit too much.
And unlike FMA there is no precision gain in IMAC, just convenience.
IMAC requires 6 register specifiers, 2 dest and 4 source if you don't
care about overflow/carry on the accumulate.
  2-wide = 2-wide + narrow * narrow
It needs 7 registers, 3 dest and 4 source if you want overflow/carry
on the accumulate.
  3-wide = 2-wide + narrow * narrow
 
I wanted to support checked arithmetic which means full width multiplies.
And I was always bothered by the risc approach of MULL (low part) and
MULH (high part) where they do most of the multiply then toss half away
just because they won't have 2 dest registers.

I always assumed that MULH just grabbed the part that would have been
thrown away. And that is how at least one RISC-V core does it:

https://www.digikey.com/en/blog/how-the-risc-v-multiply-extension-adds-an-efficient-32-bit

They claim 5 cycles, should be six, five for the multiply and one more for
the second result, unless the next instruction does not need a write port,
and does not use the result. You can get a throughput of 5 cycles with
smart coding, but that rarely happens without effort.

So what else I can do with 2 dest registers? Wide add and sub.
Various wide Add,Sub solves the missing carry/overflow flags problems.
 
FMA already requires 3 source registers.
Beside Add,Sub,Mul what else can one do with 3 source and 2 dest registers?
Wide shifts and wide bit-field extract and insert.
 
I went with two (r_hi,r_lo) register specifiers because it gave programmers
more flexibility. I played a bit with even register pairs (R0, R2, R4...)
and found one had to do extra MOVs just form a pair.
(r_hi,r_lo) cost a longer instruction format but I have a variable length
instruction so its mostly a wider fetch and decode pathways to handle
the worst case instruction size.
 
W = Wide = (hi,lo) register pair, N = Narrow = one register.
 
Add forms:
Add   N = N + N        // No carry out
Add3  N = N + N + N    // No carry out
Addw2 W = N + N        // Generate carry
Addw3 W = N + N + N    // Generate + propagate carry
Addw1 W = W + N        // Propagate carry
 
Same for subtract wide.
The three Add forms are chosen to make multi-precision integer
multiply easier. See below.
 
MUluw W = N * N
Mulsw W = N * N
 
Divuw  (quo,rem) = N / N
Divsw  (quo,rem) = N / N
 
Shllw  W = W << size  // Shift left logical
Shlaw  W = W << size  // Shift left arithmetic, fault on signed overflow
Shrlw  W = W >> size  // Shift right logical
Shraw  W = W >> size  // Shift right arithmetic, sign extend
Shrnw  W = W >> size  // Shift right numeric, round -1 to zero
 
Bfextu N = extract (W, size, position)    // Bit-field extract, zero extend
Bfexts N = extract (W, size, position)    // Bit-field extract, sign extend
Bfins  W = insert  (W, N, size, position) // Bit-field insert
 
=====================================
Example unsigned 128 * 128 => 256 multiply:
 
// Unsigned Multiply 128*128 => 256
// (r3,r2)*(r1,r0) => (r3,r2,r1,r0)
// Uses r4,r5,r6,r7,r8 as temp registers
//
muluw r5,r4 = r3*r0
muluw r6,r0 = r2*r0
muluw r8,r7 = r2*r1
muluw r3,r2 = r3*r1
addw3 r4,r1 = r4+r6+r7
addw3 r5,r2 = r5+r8+r2
addw2 r4,r2 = r2+r4
add3     r3 = r3+r5+r4
 
The reason I prefer the separate (r_hi,r_lo) pair specifiers rather
than the even number register pairs R0,R2,R4... is because the above
sequence would require extra moves for form the even numbered pairs.
With separate pairs one can select registers so that everything lands
in the right dest at the right time.


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