From google, iphone 4 uses cortex-a8 and that cpu uses vfp-lite which
is 10 times slower than a regular vfp used in cortex-a9.
Сбт, 08 Сен 2012, Joey K Tuttle писал(а):
> That's a nice speedup! It will be interesting to try the iPhone 5.
>
> I'm curious if anyone has a timing from a generation 3
Yes, I have done that from the start and share as much as possible by defining
in every profilex.ijs (from the
profilex_template.ijs)
user=. '\j64-701-user'
temp=. '...\J\user'
This I did for \j602\bin\profilex.ijs ; \j701\bin\profilex.ijs ;
\j64-602\bin\profilex.ijs and; \j64-701\bin\p
Right... I'd like to report back.
Kind people have tried to engage seriously with my problem -- which I
did feel was general enough and important enough to concern others. I
thank them and I'm deeply grateful. I feel I owe it to them to explain
my change of thinking, and how it has come about.
--
Following your moral, I NEVER make my own profile and instead use what comes
with the package of J. If I want to change things I must do it each time
manually so I know I won't forget at some point that I changed it myself and
it won't work the same way if someone else uses it.
This probably work
default=. ([: *@#@". [) >@{ ] ; [: ". [
MODE=. 'MODE' default '-t -4 -a -p -s -f'
For one-time initialization I use
butifundef =: 2 : 'if. 0 > 4!:0 Not quite the same as yours - it works only for global names, and you
would want 'butifundeforempty' - but it's something I can keep in my
too
Yes, my stock version of "default", loaded at startup, uses 4!:0 and
is explicit.
I like your use of a conjunction instead of a verb here. Or I may
replace my explicit verb with the new version of "default" I gave, and
have only recently discovered -- provided the latter does not
misbehave on me.
It's also a counsel of perfection when you're frenziedly debugging,
restarting J each time. You want as fast a turnaround as you can get.
Soon you're hacking your startup.ijs in a Notepad session, left
open...
...Not for the faint-hearted.
But my point was not that it tripped **me** up -- but ima
Very impressive.. thanks
I got all the unknowns.
I was going to include more things in the "sovleeqn" but I get "limit error"
After getting unknowns (a-h) I have to transform them back to the image with
some formula.
I followed your outline to create "solveeqn"
solveeqn=: 4 : 0
...
...
limit error typically means you are doing something different from
what you think you are doing. Here's an example:
i.1+i.1+i.9
|limit error
I would probably start by examining the values used in the expression
which is getting this error. They are probably overly large, or
something.
>.
yes I guess they're overly larger
The size of the result should be 50 x (2 x 1090*1376)
normal right?
maybe I have to change the strategy
Raul Miller-4 wrote:
>
> limit error typically means you are doing something different from
> what you think you are doing. Here's an example:
>
>i.
yes I guess they're overly larger
The size of the result should be 50 x (2 x 1090*1376)
normal right?
maybe I have to change the strategy
Raul Miller-4 wrote:
>
> limit error typically means you are doing something different from
> what you think you are doing. Here's an example:
>
>i.
Assuming 8 bytes per number, representing this array would need about:
2 ^.*/8 50 2 1090 1376
30.1602
30 bits of address space. Or, slightly over 1 gigabyte.
Add another factor of 5 for intermediate results... you probably need
a 64 bit implementation of J to run this code. (If that is what
timing update.
Paul tested the best result (so far) for android is about 3.75 seconds.
Сбт, 08 Сен 2012, bill lam писал(а):
> I am curious to know what are the timings for iphone and ipad.
>
> It is around 6 to 11 seconds on android depending on CPU.
>
> Срд, 05 Сен 2012, Paul Jackson писал(а)
Yes, I did move from J6 on Windows to J7 on Mac
I browsed the links you sent. Linking J7 to lbfgs on a Mac looks a
formidable task or at least a task requiring much expertise that I
don't have. Maybe someone with such expertise can
do this. I, for one, would be very grateful.
Have also check
14 matches
Mail list logo
