Mailing List Archive: 49091 messages
  • Home
  • Script library
  • AltME Archive
  • Mailing list
  • Articles Index
  • Site search
 

Amiga View text rendering (was rebol/view on a 486?)

 [1/11] from: carl:cybercraft at: 21-Jun-2001 20:35


On 21-Jun-01, Holzammer, Jean wrote:
> CPU: > I'm using /view on a 68060/50 (comparable to a Pentium I with same
<<quoted lines omitted: 5>>
> When typing text into an input field it takes almost 1 sec. for one > character to be drawn !
May I take a wild guess and assume you're using an Amiga? But even so, there's something pretty odd in that 1 second a character comment. I'm using (abusing?:) View on a 50Mhz 68030 Amiga 1200 through an Ateo Pixel64 GFX card and when in 24bit mode text-entry in View 'fields keeps up with what my key-repeat rate is set to. It's horrible to the point of unusability in a View 'area though, but I'd guess it's still managing about 3 characters a second. So, given a decent GFX-card, View text-entry should be quite pleasent on a 50Mhz 060 Amiga. (Though through AGA I guess not.) Any other 060 users like to comment? That said, View's text-rendering should be a lot faster than it is on the Amiga. One other thing to note: View1.0 for the Amiga was released with some error-checking (or whatever) still turned on. View1.1 and 1.2 are noticably faster, so if your View is prior to 1.1 an upgrade would speed it up a bit.
> So I think, using view on a 486 or a Pentium I/100 for the GUI > doesn't make much sense.
<<quoted lines omitted: 4>>
> X. > Jean
-- Carl Read [carl--cybercraft--co--nz]

 [2/11] from: jean:holzammer:faedv-n:bayern at: 21-Jun-2001 11:13


>May I take a wild guess and assume you're using an Amiga?
yep.
>So, given a decent GFX-card, View text-entry should be quite
pleasent
>on a 50Mhz 060 Amiga. (Though through AGA I guess not.) Any other >060 users like to comment?
AGA only :(
>That said, View's text-rendering should be a lot faster than it is
on
>the Amiga.
If RT would compile CPU specific executables, view might run a lot faster , especially on a 060 as it has a different architecture than the other 68K processors. Modern C-Compilers can optimize (pipeline !) quite good. Many 68060-optimized programs run 1.5 to 2.0 times faster on a 060 than a 020 version on the same cpu.
>One other thing to note: View1.0 for the Amiga was released with
some
>error-checking (or whatever) still turned on. View1.1 and 1.2 are >noticably faster, so if your View is prior to 1.1 an upgrade would >speed it up a bit.
I use View1.2. Jean

 [3/11] from: carl:cybercraft at: 21-Jun-2001 21:36


On 21-Jun-01, Holzammer, Jean wrote:
>> May I take a wild guess and assume you're using an Amiga? > yep.
<<quoted lines omitted: 3>>
>> 060 users like to comment? > AGA only :(
Oh well... Other 060 users? Is it just his lack of a gfx-card that's slowing his text-entry down so much, or is it still sluggish with a decent card?
>> That said, View's text-rendering should be a lot faster than it is >> on the Amiga. > If RT would compile CPU specific executables, view might run a
lot
> faster , especially on a 060 as it has a different architecture than > the other 68K processors. Modern C-Compilers can optimize (pipeline
<<quoted lines omitted: 7>>
> I use View1.2. > Jean
-- Carl Read [carl--cybercraft--co--nz]

 [4/11] from: chris:starforge:demon at: 21-Jun-2001 11:00


Carl Read wrote:
> Oh well... > > Other 060 users? Is it just his lack of a gfx-card that's slowing his > text-entry down so much, or is it still sluggish with a decent card?
Without knowing how /View works (in other words - this could be utter tripe, so don't read too much into it), I'd guess that /view does all its work in a chunky buffer which is then shown on screen. With a graphics card you're the limited by the size of your app and the fill rate of your graphics chipset, usually many times faster than AGA. If you aren't working in graphics memory then you're also limited by the transfer rate to the graphics memory, but compared to the core bus on the Amiga, this is trivial. But with AGA you need to do a c2p conversion. This is Not Fast at the best of times and even areas as small as 320x256 can bog an Amiga down regardless of processor: the bottleneck is the chipram bus, not the processor, as the planar information has to be written into a chipram buffer for the blitter to get at it. Go up to normal application sizes like 640x480/800x600 and you're looking at a LOT of work per frame and the bus bottleneck has a serious impact on performance. Getting a graphics card will increase the speed of all graphical operations - the amount /view will benefit depends largely on how wide of the mark I am with these guesses. Chris

 [5/11] from: ochnygosch::web::de at: 21-Jun-2001 16:56


On 21-Jun-01, Carl wrote:
>>> May I take a wild guess and assume you're using an Amiga? > >> yep.
Hey, I'm not the only one. :-)
> Other 060 users? Is it just his lack of a gfx-card that's slowing his > text-entry down so much, or is it still sluggish with a decent card?
I'm using a Blizzard 1260 IV with 64 MB RAM and a Voodoo3 3000. The only thing I can say is that Rebol/View works well. Even the Textinput-Field is pretty fast. Regards -- Jens Köcke Keep the momentum going Amiga

 [6/11] from: holger:rebol at: 21-Jun-2001 9:23


On Thu, Jun 21, 2001 at 11:13:20AM +0200, Holzammer, Jean wrote:
> If RT would compile CPU specific executables, view might run a lot > faster , especially on a 060 as it has a different architecture than the > other 68K processors. Modern C-Compilers can optimize (pipeline !) quite > good. Many 68060-optimized programs run 1.5 to 2.0 times faster on a 060 > than a 020 version on the same cpu.
Only for code that uses the FPU a lot, which REBOL usually does not. Other than that there is virtually no difference in optimization goals between 020 and 060. The 060 scheduler in SAS/C makes no measurable difference (in our tests). Pipelining is done automatically by the CPU, not by the C compiler. -- Holger Kruse [holger--rebol--com]

 [7/11] from: holger:rebol at: 21-Jun-2001 9:25


On Thu, Jun 21, 2001 at 11:13:20AM +0200, Holzammer, Jean wrote:
> AGA only :(
No surprise then. For AGA REBOL needs to quantize its 24-bit screen buffer AND c2p-convert it in real-time, and then copy everything through chip mem buffers. That is bound to be slow. A graphics card makes a huge difference. -- Holger Kruse [holger--rebol--com]

 [8/11] from: arolls:bigpond:au at: 22-Jun-2001 3:08


> But with AGA you need to do a c2p conversion. This is Not Fast at the > best of times and even areas as small as 320x256 can bog an Amiga down
<<quoted lines omitted: 4>>
> the bus bottleneck has a serious impact on performance. > Chris
This "chunky2planar conversion is so cpu-expensive" argument was debunked and proven by the guy who ported Doom to the amiga. He wrote that the c2p conversion ended up taking less than 5% (or was it 1%?) of cpu-time. I was getting a decent frame-rate on an '040 aga amiga, something like 15 fps 320x256. Anton.

 [9/11] from: chris:starforge at: 21-Jun-2001 18:52


#21-Jun-01# Message from *Anton*: Hi Anton,
> This "chunky2planar conversion is so cpu-expensive" > argument was debunked and proven by the guy who > ported Doom to the amiga.
No argument. Now tell me where I said it was CPU expensive please? A quote from my email: can bog an Amiga down regardless of processor: the bottleneck is the *CHIPRAM BUS*, *not* the *processor*, and the bus bottleneck has a serious impact on performance. I've gone through this argument 10,000 times and people just don't seem to get it. AGA is slow, old, massively underpowered, has a fill rate that makes a toddler look like a speed artist and it is tied to a system bus which runs ar a fixed 17Mhz (IIRC). You can have as speedy a processor as you want, but it can only shove those bits over the motherboard at a fraction of its speed, and *that* is what kills performance. Chris -- New sig in the works Explorer 2260, Designer and Coder http://www.starforge.co.uk -- One good reason why computers can do more work than people is that they never have to stop and answer the phone.

 [10/11] from: jean:holzammer:faedv-n:bayern at: 22-Jun-2001 8:19


>> Modern C-Compilers can optimize (pipeline !) quite > >> good. Many 68060-optimized programs run 1.5 to 2.0 times faster on a > 060 > >> than a 020 version on the same cpu. >
I.
>Pipelining is done automatically by the CPU, not by the C compiler. >Holger Kruse
Of course it is done by the CPU. What I tried to say is: The compiler can arrange the order of (assembler) instructions in a way to better utilize the possibilities of the pipeline. Example: a:=1+2 b:=b+3 c:=3+4 d:=b+c the second command cannot be processed until the first one is computed. a compiler can optimize the code like this: a:=1+2 c:=3+4 b:=b+3 d:=b+c So while a is computed the cpu can not only fetch and decode the instruction c, but already pass it to the second integer unit of the 060, so it is processsed simultaniously to instruction a. II. If I got you right, /view always redraws the complete graphics buffer and passes it to the GPU even if you just change one pixel in a window ? This would explain why /view is slowed down by AGA a lot more than other applications. III. to the list: I hope this was not to off-topic. Jean

 [11/11] from: holger:rebol at: 22-Jun-2001 8:49


On Fri, Jun 22, 2001 at 08:19:24AM +0200, Holzammer, Jean wrote:
> The compiler can arrange the order of (assembler) instructions in a > way to better utilize the possibilities of the pipeline.
Yes, the SAS/C scheduler does that, but it has no noticable effect with REBOL.
> If I got you right, /view always redraws the complete graphics > buffer and passes it to the GPU even if you just change one pixel in a > window ?
No. REBOL maintains damage regions and refreshes faces during 'show. -- Holger Kruse [holger--rebol--com]

Notes
  • Quoted lines have been omitted from some messages.
    View the message alone to see the lines that have been omitted