|
Kalles Fraktaler
|
 |
« Reply #240 on: March 21, 2014, 07:25:19 PM » |
|
Thank you, Karl!!! Now we can zoom to infinity!!!
P.s. I don't understand how your classes works... For SIMD I learn 'Raw' code with standart types...
SIMD is not so difficult after all http://www.codeproject.com/Articles/4522/Introduction-to-SSE-ProgrammingBut it would require a re-arrangement of everything in Kalles Fraktaler and I don't feel that it is worth it. Because there is no SIMD/SSE available for the 80-bit long double datatype, that KF use when zooming deeper than e600. And if you use pixel grouping 1 in Mandel Machine, i.e. no SIMD/SSE, it is as "slow" as Kalles Fraktaler, so I don't think my program is so bad after all.  But still, Kalles Fraktaler is and will always be an experimental prototype and will never be a commercial product. I shared it at the first place only so that people like Dinkydau could run it on their super hardware |
|
|
|
|
Logged
|
|
|
|
|
lycium
|
|
« Reply #241 on: March 21, 2014, 08:37:30 PM » |
|
double precision hasn't been 80bit for a long time, it's 64bit. long double isn't 80 bit either anymore, iirc.
|
|
|
|
|
Logged
|
|
|
|
claude
Fractal Bachius
Posts: 563
|
|
« Reply #242 on: March 21, 2014, 08:51:13 PM » |
|
double precision hasn't been 80bit for a long time, it's 64bit. long double isn't 80 bit either anymore, iirc.
depends on compiler / architecture - if in doubt, test it! $ uname -a
Linux frappuccino 3.2.0-4-amd64 #1 SMP Debian 3.2.54-2 x86_64 GNU/Linux
$ cat numeric-limits.c
#include <stdio.h>
int main() {
#define TEST(T) \
do { T x = 1, y = 1; int n = 0; \
while (x + y != x) { y /= 2; n += 1; } \
printf("%s\t%d\n", #T, n); \
} while(0)
printf("# type\tbits\n");
TEST(float);
TEST(double);
TEST(long double);
return 0;
}
$ gcc -std=c99 -Wall -pedantic -Wextra -o numeric-limits numeric-limits.c
$ ./numeric-limits
# type bits
float 24
double 53
long double 64 |
|
|
|
|
Logged
|
|
|
|
|
lycium
|
|
« Reply #243 on: March 21, 2014, 08:52:50 PM » |
|
Yup, and in your example you could have mentioned the architecture  What I do know is that you don't have 80bit FP in 64bit mode, i.e. with SSE. |
|
|
|
|
Logged
|
|
|
|
claude
Fractal Bachius
Posts: 563
|
|
« Reply #244 on: March 21, 2014, 09:00:30 PM » |
|
Yup, and in your example you could have mentioned the architecture sorry, here it is $ file numeric-limits
numeric-limits: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked (uses shared libs), for GNU/Linux 2.6.26, BuildID[sha1]=0xc3f88ef11acdff68ec51a9c5a0e2f185e75c8518, not stripped
|
|
|
|
|
Logged
|
|
|
|
|
lycium
|
|
« Reply #245 on: March 21, 2014, 09:14:07 PM » |
|
Hmmmmm, that's very interesting! I was under the impression that FP maths is always 64bit on modern archs, though I guess this could still be true (and the 80bit is being emulated).
|
|
|
|
|
Logged
|
|
|
|
|
Botond Kósa
|
|
« Reply #246 on: March 21, 2014, 09:27:04 PM » |
|
Hmmmmm, that's very interesting! I was under the impression that FP maths is always 64bit on modern archs, though I guess this could still be true (and the 80bit is being emulated).
There is no restriction on machine code using 80bit FP maths is x64 mode. The precision of the generated machine code depends on the compiler used. Microsoft's C/C++ compiler generates code that uses the SSE registers and instructions, therefore it is limited to 64bit precision. GCC is still able to generate code that uses the old x87 register stack with 80bit precision. This is also possible in assembly code, of course. BTW, Kalle, what compiler do you use for Kalles Fraktaler 64bit? |
|
|
|
« Last Edit: March 21, 2014, 09:29:12 PM by Botond Kósa »
|
Logged
|
|
|
|
|
|
|
Dinkydau
|
|
« Reply #248 on: April 03, 2014, 06:44:20 AM » |
|
What is this?? It looks like a julia set, but kalles fraktaler can't even render julia sets. There's supposed to be a minibrot here.  Re = -1.76867872112652352344459411386653663775287705156707967520571895215849441304559320913172522013960055361761841257031034787731503477
Im = 0.00177119601904983495003899071878584173417083134810882664974540521061624750516761580029876916300817101854808397887294224198269995 Magnification: 1.83475743775E106 |
|
|
|
|
Logged
|
|
|
|
|
Kalles Fraktaler
|
|
« Reply #249 on: April 03, 2014, 10:31:03 AM » |
|
What is this?? It looks like a julia set, but kalles fraktaler can't even render julia sets. There's supposed to be a minibrot here. http://i538.photobucket.com/albums/ff342/formule/julia_set.pngRe = -1.76867872112652352344459411386653663775287705156707967520571895215849441304559320913172522013960055361761841257031034787731503477
Im = 0.00177119601904983495003899071878584173417083134810882664974540521061624750516761580029876916300817101854808397887294224198269995 Magnification: 1.83475743775E106 Strange... Are you using version 2.3.3? I think I had a bug that distorted minibrots in an earlier version (2.3 - 2.3.2) I tried the location in 2.3.3 but I did get a minibrot in the center, 10 million max iterations required. |
|
|
|
|
Logged
|
|
|
|
SeryZone
Strange Attractor
Posts: 253
Contemplate...
|
|
« Reply #250 on: April 03, 2014, 01:40:43 PM » |
|
|
|
|
|
|
Logged
|
|
|
|
|
ellarien
|
|
« Reply #251 on: April 03, 2014, 01:57:02 PM » |
|
I came across something similar once, but unfortunately didn't save the evidence. There was a slightly distorted minibrot, I tried to glitch-fix it, and it turned into something Julia-shaped. Another attempt at fixing sorted it out. (It wasn't so deep that it was a big problem -- I hardly ever go past about 10^6 iterations.) It was a few weeks back, but I think it was after the version that mentioned fixing the distorted-minibrot problem.
|
|
|
|
|
Logged
|
|
|
|
|
Dinkydau
|
|
« Reply #252 on: April 03, 2014, 06:15:26 PM » |
|
Strange... Are you using version 2.3.3? I think I had a bug that distorted minibrots in an earlier version (2.3 - 2.3.2)
I tried the location in 2.3.3 but I did get a minibrot in the center, 10 million max iterations required.
No, I'm using 2.3.2 actually. I'll update. On your website there is no 2.3.3 and in the program the version 2.3.3 is mentioned, so I think I do have the latest version. Maybe it matters that the resolution was 3840×2160? I just got the correct minibrot in low-resolution right from the start. In high resolution I placed the reference point inside the weird julia set, then there was what seemed to be some circle-like julia set, and when I placed the main reference in the middle of that, the correct minibrot appeared after all. |
|
|
|
« Last Edit: April 03, 2014, 06:32:01 PM by Dinkydau »
|
Logged
|
|
|
|
laser blaster
Iterator
Posts: 178
|
|
« Reply #253 on: April 03, 2014, 06:56:39 PM » |
|
I encountered a black colored Julia set while zooming in Kalles Fraktaler, and all I had to do was increase the iterations, and it was fixed. I thought that was normal behavior.
But I did find a glitch that you might already know about. If you keep zooming into the very tip of the needle of the M-set, the auto glitch fixing stops working and the screen gets half-covered in a solid color. In general, the needle of the set is prone to glitches.
|
|
|
|
|
Logged
|
|
|
|
|
Kalles Fraktaler
|
|
« Reply #254 on: April 03, 2014, 08:49:50 PM » |
|
No, I'm using 2.3.2 actually. I'll update. On your website there is no 2.3.3 and in the program the version 2.3.3 is mentioned, so I think I do have the latest version.
Maybe it matters that the resolution was 3840×2160? I just got the correct minibrot in low-resolution right from the start.
In high resolution I placed the reference point inside the weird julia set, then there was what seemed to be some circle-like julia set, and when I placed the main reference in the middle of that, the correct minibrot appeared after all.
There is this optimization, 'ignore small addends' or what Botond (Mandel machine) called them, that causes this, I think. It causes distortions of minibrots. KF detects the minibrot if the reference does not bail out, which maybe is not always enough as it seems. But I don't know a better way to detect it, currently... My biphome site hasn't worked at all for some week now, and I haven't been able to upload anything for several months. The page on Chillheimer's domain is the official site now I encountered a black colored Julia set while zooming in Kalles Fraktaler, and all I had to do was increase the iterations, and it was fixed. I thought that was normal behavior.
But I did find a glitch that you might already know about. If you keep zooming into the very tip of the needle of the M-set, the auto glitch fixing stops working and the screen gets half-covered in a solid color. In general, the needle of the set is prone to glitches.
It is probably the addends that cause this thing in the needle too. I will make an update this weekend with some other small things as well. |
|
|
|
|
Logged
|
|
|
|
|
January 27, 2014, 12:08:08 PM
Dilber MC Theme by HarzeM