Open CL working like a beauty

  • 36 Replies
  • 1217 Views

0 Members and 1 Guest are viewing this topic.

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #15 on: December 29, 2019, 11:58:53 PM »
r = sqrt(z2.x - z2.y - z2.z + (z2.y * z2.z) / (z2.x)) option has been added to scator power 2 formula.  (not in openCL yet, but will implement it in next few days)

I have revisted analytic DE but have been unable to get it to work any better. DeltaDE is still the preferred method.


Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #16 on: December 30, 2019, 12:10:23 AM »
r = sqrt(z2.x - z2.y - z2.z + (z2.y * z2.z) / (z2.x)) option has been added to scator power 2 formula.  (not in openCL yet, but will implement it in next few days)

I have revisted analytic DE but have been unable to get it to work any better. DeltaDE is still the preferred method.

Great! I just want to confirm that this r is correct only for real scator algebra (not imaginary).
By the way, what is used in the programme is r² (r squared) since there is no point in evaluating the sqrt, i.e. bailout is 4 (not 2) in usual formulae (in scators is 100 which is fine). This is important, because if r² is negative, we are still in the reals and no problem should arise. However, if the root is evaluated, problems arise, since inequalities are defined for the reals, not for complex numbers.
Thank you once again.
Please ask Budhi (I am spelling it incorrectly!?) how you would like to be cited.
mfg

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #17 on: December 30, 2019, 12:39:09 AM »
at the moment i am reworking  it all,  you can currently choose what r to use for the  analyticDE. Bailout is something i need to figure out, as how to implement it in the formula code oppoesd to compute_fractal.cpp

for buddhi  use   Krzysztof Marczak

For correct color work you need to use Color-By-Numbers

e.g

sc radial is coloring based on distance from origin, the center of each color is in 0.1 units from the origin
sc minR again the colors are in 0.1 intervals (this is like the standard orbit trap coloring but numerically correct)


Code: [Select]
# Mandelbulber settings file
# version 2.21
# only modified parameters
[main_parameters]
aux_light_intensity_1 1;
aux_light_position_1 1 -3 -2;
bailout 1000;
camera -0.7 -5 0;
camera_distance_to_target 5;
camera_rotation 0 0 0.0004009397527962243;
camera_top 6.997718788369144e-06 0 0.9999999999755159;
DE_thresh 0.0003;
fill_light_color 6c00 5300 6c00;
flight_last_to_render 0;
formula_1 152;
fov 20.96;
image_height 540;
image_proportion 4;
image_width 960;
iteration_threshold_mode true;
keyframe_last_to_render 0;
limit_min -10 0 -10;
limits_enabled true;
main_light_alpha -20;
main_light_beta 15;
main_light_intensity 0.5;
main_light_visibility_size 2;
mat1_coloring_speed 0.5;
mat1_fractal_coloring_extra_color_enabled_false true;
mat1_fractal_coloring_ic_rad_enabled_false true;
mat1_fractal_coloring_ic_rad_weight 10;
mat1_fractal_coloring_orbit_trap_weight 10;
mat1_fractal_coloring_xyzC_111 0 0 0;
mat1_is_defined true;
mat1_surface_color_gradient 0 aa0000 500 005500 1000 030a84 1500 ffff00 2000 55ffff 2500 ff55ff 3000 f5bd22 3500 808d54 4000 0b5e87 4500 68737e 5000 000000 5500 b5526d 6000 a51c64 6500 6f5da9 7000 3b9fee 7500 87cfe1 8000 d4ffd4 8500 bfd187 9000 aba43b 9500 d48232;
shadows_cone_angle 3;
target -0.7 0 0;
[fractal_1]
IFS_rotation_enabled true;
[fractal_2]
IFS_rotation_enabled true;
[fractal_3]
IFS_rotation_enabled true;
[fractal_4]
IFS_rotation_enabled true;

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #18 on: December 30, 2019, 01:28:25 AM »
yep the r2 for bailout has been on the TODO list for sometime but it involves adding some backwards compatibility code, and  a lot of boring work :), but in theory it will speed up the main loop for many linearDE type fractals that only need r calculalted in the final calculation.

with color there are two existing transforms where we can add any additional color calculations that you might need.

Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #19 on: December 30, 2019, 01:29:35 AM »
at the moment i am reworking  it all,  you can currently choose what r to use for the  analyticDE. Bailout is something i need to figure out, as how to implement it in the formula code oppoesd to compute_fractal.cpp

for buddhi  use   Krzysztof Marczak

For correct color work you need to use Color-By-Numbers

e.g

sc radial is coloring based on distance from origin, the center of each color is in 0.1 units from the origin
sc minR again the colors are in 0.1 intervals (this is like the standard orbit trap coloring but numerically correct)



I went to 'color by numbers' in the program but where do I see 'sc radial' and 'sc minR'
How do I choose one or the other?
(I a sorry for such elementary questions ...)

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #20 on: December 30, 2019, 01:36:51 AM »
they were just names for my jpgs.  The settings i provided are the  sc minR =   Color by number orbittrap component (standard), . Disable that  and enable colorvalue initial conditions components and you will get the color based on the points initial radial distance from the origin

Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #21 on: January 07, 2020, 01:52:04 AM »
I will have a look at bug zip, but  the difference may be due to single versus double precision floating point'

the standard coloring function is based on  the minimum distance the point gets to the origin during the iterations

The bug when opencl colours dark seems to be when x > -1 and there is a bounding box.
I can confirm that it happens consistently with various views and conditions.

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #22 on: January 07, 2020, 08:33:26 AM »
it happens with other fractal too,  e.g quaternion in both openCL and c++, but the extent of the effect varies.
at the moment i am guessing it has to do with light.  Set light horiz angle to 90 and vert angle to 0, this will reduce the effect. This is an issue for Krzysztof.


On a separate topic, download latest build, the analyicDE Scator Pow 2 has been updated and can now be found in the Custom DE section. (The formula Scator Pow2 StdR will be removed as it can be replicated in Scator Pow 2 UI)

We now have  three r options, but you will see that Real formula using analyticDE with Real r  does not exactly replicate  the version using DeltaDE  Logarthmic
Imaginary works well.

Force bailout (default 2.0) is a hack. This does not effect the sliced limits view, but normal view you will get different surface texture depending on this setting.

Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #23 on: January 07, 2020, 05:21:52 PM »
it happens with other fractal too,  e.g quaternion in both openCL and c++, but the extent of the effect varies.
at the moment i am guessing it has to do with light.  Set light horiz angle to 90 and vert angle to 0, this will reduce the effect. This is an issue for Krzysztof.


On a separate topic, download latest build, the analyicDE Scator Pow 2 has been updated and can now be found in the Custom DE section. (The formula Scator Pow2 StdR will be removed as it can be replicated in Scator Pow 2 UI)

We now have  three r options, but you will see that Real formula using analyticDE with Real r  does not exactly replicate  the version using DeltaDE  Logarthmic
Imaginary works well.

Force bailout (default 2.0) is a hack. This does not effect the sliced limits view, but normal view you will get different surface texture depending on this setting.

Thank you for both of you for looking into the bug.

I will download latest build to try the three options. Thanks. Will be back to you once I try them.

Offline R-TEAM

  • *
  • Fractal Friend
  • **
  • Posts: 12
« Reply #24 on: January 13, 2020, 11:43:11 PM »
Hi,

have just updated to latest version - and it works nice with OpenCL .... (FirePro w8100)
The only thing i miss is doubleprecision to go deeper with OpenCL.
Suggestion -> place the switch (single or double precision) in the same actual OpenCL render option (fast/medium/full) as it is faster to zoom in with Single precision and switch to Double if needed as open the menu and set the (actual na) option in the OpenCL settings to double if needed ...

Regards

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #25 on: January 15, 2020, 09:58:00 AM »
R-Team, I think the answer is yes, but not high on the TODO list. There are few users who have GPU cards that can handle double precision

Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #26 on: January 31, 2020, 05:07:24 AM »
Dear friends,

I git cloned the last mandelbulber version today 30-jan-2020. And to my dismay, opencl is no longer working with the scator fractals (neither imaginary nor real). I shall be most grateful if you remedy this problem.

yours,

Offline mclarekin

  • *
  • Fractal Freak
  • **
  • Posts: 777
« Reply #27 on: February 01, 2020, 05:17:52 AM »
it is working for me in win10, i will try ubuntu later
Code: [Select]
# Mandelbulber settings file
# version 2.21
# only modified parameters
[main_parameters]
aux_light_intensity_1 1;
aux_light_position_1 1 -3 -2;
bailout 1000;
camera -0.7 -5 0;
camera_distance_to_target 5;
camera_rotation 0 0 0.0004009397527962243;
camera_top 6.997718788369144e-06 0 0.9999999999755159;
DE_thresh 0.0003;
fill_light_color 6c00 5300 6c00;
flight_last_to_render 0;
formula_1 152;
fov 20.96;
image_height 540;
image_proportion 4;
image_width 960;
iteration_threshold_mode true;
keyframe_last_to_render 0;
limit_min -10 0 -10;
limits_enabled true;
main_light_alpha 84;
main_light_beta 0;
main_light_intensity 1.4;
main_light_visibility_size 2;
mat1_is_defined true;
mat1_surface_color_gradient 0 fd6029 999 698403 1999 fff59b 2999 f5bd22 3999 0b5e87 5000 c68876 5999 a51c64 7000 3b9fee 7999 d4ffd4 9000 aba53c;
shadows_cone_angle 3;
target -0.7 0 0;

Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #28 on: February 01, 2020, 07:29:11 AM »
Dear Mclarekin,

The system where scator algebra with opencl is not working is: (mandelbulb renders fine with opencl, for example)
Operating System: Kubuntu 19.10
KDE Plasma Version: 5.16.5
KDE Frameworks Version: 5.62.0
Qt Version: 5.12.4
Kernel Version: 5.3.0-29-generic
OS Type: 64-bit
Processors: 12 × Intel® Xeon® CPU E5-2620 v2 @ 2.10

GP106 [GeForce GTX 1060 6GB]



Offline mfg

  • *
  • Fractal Fanatic
  • ***
  • Posts: 20
    • servidor de luz
« Reply #29 on: February 01, 2020, 07:39:33 AM »
I confirm that with your settings no fractal is produced with opencl but renders fine without it.


xx
Open CL not working with AMD pro drivers/ cards

Started by Dievolve on Mandelbulber

3 Replies
425 Views
Last post October 08, 2019, 02:01:46 PM
by R-TEAM
xx
version 2.17: open cl working gtx 1080 with hitches (linux)

Started by JerryS on Mandelbulber

1 Replies
298 Views
Last post April 27, 2019, 06:03:42 AM
by mclarekin
xx
The beauty of self-similarity

Started by Nintendokater on Fractal Image Gallery

2 Replies
223 Views
Last post August 19, 2020, 07:45:46 PM
by Caleidoscope
xx
Another way to see the inner beauty of fractals

Started by kohlenstoff on Fractal movie gallery

0 Replies
223 Views
Last post August 28, 2018, 11:12:37 PM
by kohlenstoff
xx
Into The Beauty - Mandelbrot Set Zoom vs Julia Set

Started by mfcc64 on Fractal movie gallery

8 Replies
241 Views
Last post October 15, 2020, 08:48:36 PM
by Adam Majewski