400mhz Processor In Gpx2, Please!

[Prophet]

But Jeff, if some company gave us what we actually want, we'd be completely happy with it. And we'd stop buying crap we don't need.

Nice to see you around BTW.

 

[skeezix]

*g* True, why give us what we want, when they can milk us on that road? Damn

Around? /me is always around

CaSTaway/PSP coming soon

jeff

 

[Sonic-NKT]

i hope a castaway/gpx2 follows

 

[Talyz]

Even the open project Ninja isn't far as powerful as we would want it to be...

 

[skeezix]

CaSTaway will likely come to GPX2, but so far GPX2 doesn't exist so I'm not worried about it

jeff

 

[LDChen]

I think that current MHz (233 or 250) is fast enought, althought it isn't even an ARM9E core.
Don't forget that the device can count on a second CPU which can be programmed independently, at least this is the theory.
This feature should be really useful (for example, it can execute OpenGL commands) and if you think on the resolution of LCD, well, we have surely more than the minimal power for handling his screen size.
It's also true that some tasks may require lot of processing power and the absence of the hardware support for floating point may cause troubles, but hey, I hope you wouldn't use a similar device just for emulating an old console!
However, into my wishlist I still have a small flash with a bios instead of a bigger one with Linux...

Sincerely,

LDChen

 

[Squidge]

It's been said that the 940 supports hardware floating point, and so the 920 can use it as a hw fpu coproc. How much truth there's in that is another thing however.

 

[x68000]

I think that current MHz (233 or 250) is fast enought, althought it isn't even an ARM9E core.
Don't forget that the device can count on a second CPU which can be programmed independently, at least this is the theory.
This feature should be really useful (for example, it can execute OpenGL commands) and if you think on the resolution of LCD, well, we have surely more than the minimal power for handling his screen size.
It's also true that some tasks may require lot of processing power and the absence of the hardware support for floating point may cause troubles, but hey, I hope you wouldn't use a similar device just for emulating an old console!
However, into my wishlist I still have a small flash with a bios instead of a bigger one with Linux...

Sincerely,

LDChen

​

you seem quite optimistic about the gpx2, so can we assume that you may dev for the system ld?

 

[LDChen]

It's been said that the 940 supports hardware floating point, and so the 920 can use it as a hw fpu coproc. How much truth there's in that is another thing however.

Well, those ARM920T and ARM940T are ARMV4 cores.
They can't have the so called "Vector Floating Point coprocessor" because they are an old generation.
You can usually get hardware VFP on ARM10 or ARM11, not less.
Althought it's true that the ARM cores can't be touched, a company (like the MesDigital, they produce the MP2520F used into GPX2) could syntetize an external device for handling floating point operations.
Don't get me wrong when I talk about the "external device"... it would be always into the same chip, like for example the UARTs.
In my opinion this is the only way for having some hardware support for floating points actually, but it doesn't seem there is something like that (please correct me if I'm wrong).
This device still looks very promising because his dual core architecture.
The best, like you said, is using the slave processor in a smart way.
This means that the 940T will execute floating point in *software* or more complex calculations like matrix algebra and these actions will free the master 920T for other tasks.
I'm not sure that you will be able to use it exacly like a true math coprocessor.
It will surely help a lot, but I doubt it can do miracles.
The fact is that a program is a sequential flow of intruction.
If you write:
a=sin(b);
you need to wait that "sin(b)" is finished before assign the result to "a".
This problem could be partially bypassed with a FIFO (you first push all calculations and then you read the results).
Or it's possible to provide some functions that are opcodes for the software coprocessor implemented into the 940T.
And you will use them instead of usual sin(), cos(), etc...
For example, let's suppose that into the 940T is emulated an FP processor with some registers and:

float a;

fp_load_immediate(0, 1); // load into FP register 0 the value "1"
fp_load_immediate(1, 2); // load into FP register 1 the value "2"
fp_add_registers(2,0,1); // execute sum of register 0 and 1
// and save result into register 2
/* do other tasks here */
....
/* read result of previous calc. */
a = fp_read_register(2); // let's hope that a=3!

Of course these are just my impressions.
Nobody said it will be like that.
We will know better when somebody will use/review the GPX2.
Things will completely different if a VFP would be available, but I still believe it's impossible.
But afterall, why putting the VFP into the slave processor? Besides the fact it can't exist with ARM9x0T cores, it would be a nonsense!

you seem quite optimistic about the gpx2, so can we assume that you may dev for the system ld?

I think it's really interesting.
I was very sceptic about the cores and the politics around this device ("play games with emulators"? just emulators???).
I'm really attracted by the dual core programming, althought we don't know if we can write programs for both processors.
I initially put my interests on the PSP, but it really seems that Sony hates legal free software on his device.
It's simply unacceptable.
So, it's surely better to support a platform that makes things easier for his users, right?

Sincerely,

LDChen

 

[x68000]

It's been said that the 940 supports hardware floating point, and so the 920 can use it as a hw fpu coproc. How much truth there's in that is another thing however.

Well, those ARM920T and ARM940T are ARMV4 cores.
They can't have the so called "Vector Floating Point coprocessor" because they are an old generation.
You can usually get hardware VFP on ARM10 or ARM11, not less.
Althought it's true that the ARM cores can't be touched, a company (like the MesDigital, they produce the MP2520F used into GPX2) could syntetize an external device for handling floating point operations.
Don't get me wrong when I talk about the "external device"... it would be always into the same chip, like for example the UARTs.
In my opinion this is the only way for having some hardware support for floating points actually, but it doesn't seem there is something like that (please correct me if I'm wrong).
This device still looks very promising because his dual core architecture.
The best, like you said, is using the slave processor in a smart way.
This means that the 940T will execute floating point in *software* or more complex calculations like matrix algebra and these actions will free the master 920T for other tasks.
I'm not sure that you will be able to use it exacly like a true math coprocessor.
It will surely help a lot, but I doubt it can do miracles.
The fact is that a program is a sequential flow of intruction.
If you write:
a=sin(B);
you need to wait that "sin(B)" is finished before assign the result to "a".
This problem could be partially bypassed with a FIFO (you first push all calculations and then you read the results).
Or it's possible to provide some functions that are opcodes for the software coprocessor implemented into the 940T.
And you will use them instead of usual sin(), cos(), etc...
For example, let's suppose that into the 940T is emulated an FP processor with some registers and:

float a;

fp_load_immediate(0, 1); // load into FP register 0 the value "1"
fp_load_immediate(1, 2); // load into FP register 1 the value "2"
fp_add_registers(2,0,1); // execute sum of register 0 and 1
// and save result into register 2
/* do other tasks here */
....
/* read result of previous calc. */
a = fp_read_register(2); // let's hope that a=3!

Of course these are just my impressions.
Nobody said it will be like that.
We will know better when somebody will use/review the GPX2.
Things will completely different if a VFP would be available, but I still believe it's impossible.
But afterall, why putting the VFP into the slave processor? Besides the fact it can't exist with ARM9x0T cores, it would be a nonsense!

you seem quite optimistic about the gpx2, so can we assume that you may dev for the system ld?

I think it's really interesting.
I was very sceptic about the cores and the politics around this device ("play games with emulators"? just emulators???).
I'm really attracted by the dual core programming, althought we don't know if we can write programs for both processors.
I initially put my interests on the PSP, but it really seems that Sony hates legal free software on his device.
It's simply unacceptable.
So, it's surely better to support a platform that makes things easier for his users, right?

Sincerely,

LDChen

​

exactly! what would you consider porting to the gpx2?

 

[Vimacs]

well gph sayd they support hardware float, but could fo course e miss information, we will see.

x68000 do you realy have to quote endless long posts when you just want to write one line? dosnt realy make the therds better readable.

 

[Jarska333]

Hey, just incase nobody noticed, Craig has updated his opening post.

(Just incase nobody noticed... This is in the wrong thread... :huh: )

 

[x68000]

well gph sayd they support hardware float, but could fo course e miss information, we will see.

x68000 do you realy have to quote endless long posts when you just want to write one line? dosnt realy make the therds better readable.

​

it was a one off - i'm very sorry :rolleyes: