6 Uops Equals One Clock Cycle???
Bryan Parkoff
clock cycle. It is the most important to find each instruction that it has
2, 3, or 4 uops before adds second and third instruction that second and
third instruction only have 1 uops. It sounds like that three instructions
can be executed at the same time under only one clock cycle.
Do you think that it can be very difficult by trying to locate
instruction with a number of uops from the table?
Do 411 Pattern still useful for Pentium 4 and Pentium 5 (Presc)?
--
Bryan Parkoff
Matt Taylor
news:6doac.2543$uV6...@fe2.texas.rr.com...
> I understand that 411 Pattern uses to convert 4, 5, or 6 uops to one
> clock cycle. It is the most important to find each instruction that it
has
> 2, 3, or 4 uops before adds second and third instruction that second and
> third instruction only have 1 uops. It sounds like that three
instructions
> can be executed at the same time under only one clock cycle.
Decoding is not the same thing as executing. Newer processors like the
Pentium-III, Pentium-IV, and Athlon don't execute x86 instructions.
Internally they are RISC-like processors. They convert x86 instructions into
internal RISC opcodes, and then they execute those opcodes.
The Pentium-III can convert up to 3 x86 instructions into up to 6 micro-ops
(its internal instruction set). The 4-1-1 pattern means that the first
instruction can decode into 4 micro-ops. The other 2 instructions will only
decode if they decode into 1 micro-op. Most micro-ops will take 1 cycle to
complete. If I am not mistaken, the Pentium-III can only execute 3 micro-ops
per cycle.
> Do you think that it can be very difficult by trying to locate
> instruction with a number of uops from the table?
Most instructions that perform computation on registers only decode into 1
u-op. If you have the tables memorized, it probably isn't too hard.
> Do 411 Pattern still useful for Pentium 4 and Pentium 5 (Presc)?
Of course not. Pentium-IV can decode only 1 x86 instruction per cycle.
By the way, Prescott is Pentium-IV E, not Pentium-V. The Pentium-IV was
radically different from the Pentium-III. Prescott is just a tweaked
Pentium-IV. It sounds like the Pentium-V will be a redesigned Pentium-M
(mobile): http://www.theinquirer.net/?article=14257
-Matt
Grumble
I am amazed that you still attempt to decrypt Parkoff's ravings.
Matt Taylor wrote:
> Newer processors like the Pentium-III, Pentium-IV [...]
Intel's marketing department insisted on calling it the Pentium 4,
because Pentium IV sounded like a CPU with a drug addiction ;-)
> By the way, Prescott is Pentium-IV E, not Pentium-V. The Pentium-IV
> was radically different from the Pentium-III. Prescott is just a
> tweaked Pentium-IV.
I respectfully disagree :-)
It seems to me that there are more differences between Northwood
and Prescott than between Barton and Hammer...
--
Regards, Grumble
hutch--
If you are working on anything like late model hardware, it wold be a
good idea to have a wade through the intel design manual on a PIV or
similar as the days of simple clock cycles are over. Instruction
timings are influenced by instruction order to some extent, pairing
and avoidance of a number of well known traps.
Pipelining came in on a 486 DX 33 and the design has changed much
sinse that time and to get the swing of what current hardware will do,
you need to design some decent testing methods to test specific code
in real time. For all the virtues of a design theory, if it is not
faster than what you had problems with, its a waste of time.
Regards,
hutch at movsd dot com
Matt Taylor
news:c4e0e9$jdc$1...@news-rocq.inria.fr...
> Matt,
>
> I am amazed that you still attempt to decrypt Parkoff's ravings.
There is method in the madness. While his posts are a butchery of English,
it is still evident to me that he understands how the processor works better
than he did 6 months ago.
> Matt Taylor wrote:
>
> > Newer processors like the Pentium-III, Pentium-IV [...]
>
> Intel's marketing department insisted on calling it the Pentium 4,
> because Pentium IV sounded like a CPU with a drug addiction ;-)
I started referring to the marketting names with Roman numerals to avoid
confusion. At various times in the past, I have called them P6 and P7, and
at least one person misinterpreted P6 and P7 as Pentium-VI and Pentium-VII
respectively. If I call them Pentium-3 and Pentium-4, then when Pentium-V
finally comes around that nomenclature will break down. (I have not seen the
386 and 486 referred to as P3 and P4, though this is conceivable too.)
> > By the way, Prescott is Pentium-IV E, not Pentium-V. The Pentium-IV
> > was radically different from the Pentium-III. Prescott is just a
> > tweaked Pentium-IV.
>
> I respectfully disagree :-)
>
> It seems to me that there are more differences between Northwood
> and Prescott than between Barton and Hammer...
Barton and Hammer have the same relationship. K8 is just a tweaked K7 and is
not deserving of being called the 8th generation; there was no real
innovation. AMD widened the ALUs to 64-bits, added cache, made numerous
improvements to the decoders, and made the multiply unit faster. Otherwise
the design is still roughly the same.
Both Pentium-IV and Prescott use the same NetBurst design with a trace cache
and a lot of pipeline stages. Prescott is still a P7 core processor; the
design is a tweaked version of Northwood. The choice to call it Pentium-IV E
instead of Pentium-V was more political: how do you explain to customers
that your latest processor runs hotter and slower? Still, Prescott is a
Pentium-IV underneath the hood. There is no fundamental difference.
-Matt
Ivan Korotkov
> confusion. At various times in the past, I have called them P6 and P7, and
> at least one person misinterpreted P6 and P7 as Pentium-VI and Pentium-VII
> respectively. If I call them Pentium-3 and Pentium-4, then when Pentium-V
> finally comes around that nomenclature will break down. (I have not seen
the
> 386 and 486 referred to as P3 and P4, though this is conceivable too.)
And when Intel produces Pentium-V (5) it will be messed with P5!
Ivan
Bo Persson
"Matt Taylor" <pa...@tampabay.rr.com> skrev i meddelandet news:JaGac.366066>
> Both Pentium-IV and Prescott use the same NetBurst design with a trace
cache
> and a lot of pipeline stages. Prescott is still a P7 core processor; the
> design is a tweaked version of Northwood. The choice to call it Pentium-IV
E
> instead of Pentium-V was more political: how do you explain to customers
> that your latest processor runs hotter and slower? Still, Prescott is a
> Pentium-IV underneath the hood. There is no fundamental difference.
That hasn't stopped Intel before. :-)
Remember when they changed the name from Pentium II to Pentium III. Same
chip, but a new year.
Later they did a redesign with an integrated cache and different packaging,
but kept the name.
You shouldn't expect any logic involved, as this is all done by the
marketing department.
Bo Persson
Grumble
> Grumble wrote:
>
>> Intel's marketing department insisted on calling it the Pentium 4,
>> because Pentium IV sounded like a CPU with a drug addiction ;-)
>
> I started referring to the marketing names with Roman numerals to
> avoid confusion. At various times in the past, I have called them
> P6 and P7, and at least one person misinterpreted P6 and P7 as
> Pentium-VI and Pentium-VII respectively. If I call them Pentium-3
> and Pentium-4, then when Pentium-V finally comes around that
> nomenclature will break down. (I have not seen the 386 and 486
> referred to as P3 and P4, though this is conceivable too.)
I don't think Intel ever used P7 to describe NetBurst. And they used
Roman numerals for Pentium II and Pentium III.
P5 = Pentium, Pentium MMX
P6 = Pentium Pro, Pentium II, Pentium III
NetBurst = Pentium 4
> Both Pentium-IV and Prescott use the same NetBurst design with a
> trace cache and a lot of pipeline stages. Prescott is still a P7
> core processor; the design is a tweaked version of Northwood. The
> choice to call it Pentium-IV E instead of Pentium-V was more
> political: how do you explain to customers that your latest
> processor runs hotter and slower? Still, Prescott is a Pentium-IV
> underneath the hood. There is no fundamental difference.
Intel updated the L1 data cache, the pipeline, and several execution
units. I'm no hardware engineer, but I'd think that's a lot of work.
Matt Taylor
> Matt Taylor wrote:
>
> > Grumble wrote:
> >
> >> Intel's marketing department insisted on calling it the Pentium 4,
> >> because Pentium IV sounded like a CPU with a drug addiction ;-)
> >
> > I started referring to the marketing names with Roman numerals to
> > avoid confusion. At various times in the past, I have called them
> > P6 and P7, and at least one person misinterpreted P6 and P7 as
> > Pentium-VI and Pentium-VII respectively. If I call them Pentium-3
> > and Pentium-4, then when Pentium-V finally comes around that
> > nomenclature will break down. (I have not seen the 386 and 486
> > referred to as P3 and P4, though this is conceivable too.)
>
> I don't think Intel ever used P7 to describe NetBurst. And they used
> Roman numerals for Pentium II and Pentium III.
>
> P5 = Pentium, Pentium MMX
> P6 = Pentium Pro, Pentium II, Pentium III
> NetBurst = Pentium 4
I don't know if Intel has or not; I have seen others use it occasionally,
but offhand I cannot think of any specifics. P7 means "7th generation", and
certainly the Pentium-IV is that.
> > Both Pentium-IV and Prescott use the same NetBurst design with a
> > trace cache and a lot of pipeline stages. Prescott is still a P7
> > core processor; the design is a tweaked version of Northwood. The
> > choice to call it Pentium-IV E instead of Pentium-V was more
> > political: how do you explain to customers that your latest
> > processor runs hotter and slower? Still, Prescott is a Pentium-IV
> > underneath the hood. There is no fundamental difference.
>
> Intel updated the L1 data cache, the pipeline, and several execution
> units. I'm no hardware engineer, but I'd think that's a lot of work.
Intel did not completely redesign the processor as they did from the Pentium
(P54C) to Pentium Pro (P6) or from the Pentium-III (P6) to the Williamette
(P7). Many things remain: double-speed ALUs, trace cache, and most of the
pipeline stages. No other x86 processor has a trace cache or double-speed
ALUs, and no processor even approaches the raw number of pipeline stages of
either processor.
Also, the execution unit changes are tweaks, not fundamental changes. They
added a shifter to one of the ALUs and a multiply unit. Otherwise they
operate the same. Note that the TC is still limited to 6 u-ops/2 cycles
despite earlier hints that it would be increased to 8. Intel took the
existing design and tweaked it.
-Matt
arjan de lumens
>
> I don't know if Intel has or not; I have seen others use it occasionally,
> but offhand I cannot think of any specifics. P7 means "7th generation", and
> certainly the Pentium-IV is that.
>
IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4; the
Netburst architecture was "P68".
Grumble
> Matt Taylor wrote:
>
>> P7 means "7th generation", and certainly the Pentium-IV is that.
>
> IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4;
> the Netburst architecture was "P68".
It seems you are correct indeed.
http://www.theregister.co.uk/1999/07/16/whats_the_bet_on_willamette/
Published Friday 16th July 1999 10:35 GMT
Willamette, as well as Merced, have existed as design concepts for
seven years so far. Merced was then called P7 and Willamette P68.
Grumble
> Matt Taylor wrote:
>
>> P7 means "7th generation", and certainly the Pentium-IV is that.
>
> IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4;
> the Netburst architecture was "P68".
It seems you are correct indeed!
Grumble
> Matt Taylor wrote:
>
>> P7 means "7th generation", and certainly the Pentium-IV is that.
>
> IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4;
> the Netburst architecture was "P68".
It seems you are correct indeed!
Grumble
> Matt Taylor wrote:
>
>> P7 means "7th generation", and certainly the Pentium-IV is that.
>
> IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4;
> the Netburst architecture was "P68".
It seems you are correct indeed!
Grumble
> Matt Taylor wrote:
>
>> P7 means "7th generation", and certainly the Pentium-IV is that.
>
> IIRC, Intel used P7 long ago to refer to Itanium, not Pentium4;
> the Netburst architecture was "P68".
It seems you are correct indeed.