Syntopia
|
|
« Reply #15 on: February 06, 2014, 08:18:34 PM » |
|
by cpu opencl implementation do you mean to say using the integrated intel graphics processor?
OpenCL is a parallellization API for both CPU and GPU. AMD and Intel both provide implementations.
|
|
|
Logged
|
|
|
|
quaz0r
Fractal Molossus
Posts: 652
|
|
« Reply #16 on: February 06, 2014, 08:22:21 PM » |
|
yeah, just making sure what you meant. i havent tried that myself.
|
|
|
Logged
|
|
|
|
Buddhi
|
|
« Reply #17 on: February 06, 2014, 08:42:28 PM » |
|
That change to cl_SSAO.cl was supposed to be included in 1.21-1. Either your install failed to update that file or buddhi failed to include it in the installer.
I was too hurry yesterday and I put wrong file into package. Within 15 minutes will be corrected installers on SourceForge
|
|
|
Logged
|
|
|
|
Buddhi
|
|
« Reply #18 on: February 06, 2014, 08:59:00 PM » |
|
Now there are proper files in the installers. I apologize for the mess. Tomorrow I'm flying to Turkey and I had limited time to fix bugs in the code and release the program. Just deleted and reinstalled 1.21-1. It still fails to compile (exe/dlls are dated 05/02, OpenCL code from 02/02). README says 'MANDELBULBER 1.21-1' Another thing: I have an CPU Intel OpenCL 1.2 runtime. I can see there is an option for toggling CPU OpenCL now. I had to toggle it on and off a few times to get Mandelbulber to change to CPU, but when it recognized the CPU platform, it gave the following error OpenCL Build log: Compilation started :323:10: error: cannot combine with previous 'type-name' declaration specifier :323:15: error: expected identifier or '(' :324:8: error: expected identifier or '(' :325:32: error: expected expression Compilation failed
It is possible to run Mandelbulber on CPU OpenCL implementations? About running Mandelbulber with CPU OpenCL I tested in with AMD drivers. It works very well. However with nVidia drivers it doesn't work at all. Which kernel gives this error? "fast" or "full"?
|
|
|
Logged
|
|
|
|
Syntopia
|
|
« Reply #19 on: February 06, 2014, 09:14:06 PM » |
|
Now there are proper files in the installers. I apologize for the mess. Tomorrow I'm flying to Turkey and I had limited time to fix bugs in the code and release the program. About running Mandelbulber with CPU OpenCL I tested in with AMD drivers. It works very well. However with nVidia drivers it doesn't work at all. Which kernel gives this error? "fast" or "full"?
"Fast". The Intel OpenCL CPU runtime is not related to the nVidia drivers - it is a standalone install from http://software.intel.com/en-us/vcsource/tools/opencl-sdk (for Intel Core processors only). I'll try the AMD CPU thing (from the APP SDK) and see if it works on an Intel CPU.
|
|
|
Logged
|
|
|
|
Syntopia
|
|
« Reply #20 on: February 06, 2014, 09:47:24 PM » |
|
I managed to get the Intel OpenCL CPU implementation to work. The problem was the use of 'half' (which is an OpenCL type) as a variable name. Changing this to something else worked, e.g.: float3 halfX = lightVector - viewVector; halfX = fast_normalize(halfX); float specular = dot(normal, halfX);
Weirdly, the image had a few pixel artifacts, not present in the GPU render.
|
|
|
Logged
|
|
|
|
mclarekin
|
|
« Reply #21 on: February 07, 2014, 12:58:03 AM » |
|
Just installed today version of 1.21-1 64bit and got the same issue as yesterday with the float3 thing pretending not to be recognised, but I ignore it and click enable a second time and all is sweet. The very minor point being that with 1.20 64bit this didn't happen.
I run CPU openCL on Intel i3 processor with the AMD SPK driver thingy and so far all is good, no other issues at all.
|
|
|
Logged
|
|
|
|
lukesleftleg
|
|
« Reply #22 on: February 07, 2014, 08:47:07 PM » |
|
Just to say that 1.21-1 seems to work just fine on my nvidia machine, and those OpenCL renders with SSAO just put my previous renders to shame. Anyway, thanks again for another great version of a great fractal renderer.
|
|
|
Logged
|
|
|
|
JohnVV
|
|
« Reply #23 on: February 09, 2014, 11:03:24 AM » |
|
mandelbulber-opencl-1.21.orig on Linux 64bit ( openSUSE 12.3 ) , Nvidia 331 .run driver (libnvidia-opencl.so.331.38 ) OpenCL headers-1.2-8.1 builds but DOSE NOT link headers -- VS. -- libnvidia-opencl.so.331.38 mix up with a bunch of undefined reference to `clReleaseContext' undefined references to `clGetDeviceInfo' undefined reference to `clGetPlatformIDs' --- and so on it starts in render3d.cpp Building target: mandelbulber-opencl Invoking: GCC C++ Linker g++ -o"mandelbulber-opencl" -L/usr/lib/x86_64-linux-gnu/ ./src/Render3D.o ./src/algebra.o ./src/callbacks.o ./src/cimage.o ./src/common_math.o ./src/database.o ./src/files.o ./src/fractal.o ./src/image.o ./src/interface.o ./src/morph.o ./src/primitives.o ./src/settings.o ./src/shaders.o ./src/texture.o ./src/timeline.o ./src/undo.o ./src/cl_support.o ./src/netrender.o -lgthread-2.0 -lglib-2.0 -lgtk-x11-2.0 -lgdk-x11-2.0 -latk-1.0 -lgio-2.0 -lpangoft2-1.0 -lpangocairo-1.0 -lgdk_pixbuf-2.0 -lcairo -lpango-1.0 -lfreetype -lfontconfig -lgobject-2.0 -lgthread-2.0 -lglib-2.0 -lgtk-x11-2.0 -lgdk-x11-2.0 -latk-1.0 -lgio-2.0 -lpangoft2-1.0 -lpangocairo-1.0 -lgdk_pixbuf-2.0 -lcairo -lpango-1.0 -lfreetype -lfontconfig -lgobject-2.0 -lrt -lpng -lOpenCL -ljpeg ./src/Render3D.o: In function `cl::CommandQueue::~CommandQueue()': Render3D.cpp:(.text._ZN2cl12CommandQueueD2Ev[_ZN2cl12CommandQueueD5Ev]+0x9): undefined reference to `clReleaseCommandQueue' ./src/Render3D.o: In function `cl::Context::~Context()': Render3D.cpp:(.text._ZN2cl7ContextD2Ev[_ZN2cl7ContextD5Ev]+0x9): undefined reference to `clReleaseContext' ./src/cl_support.o: In function `CclSupport::Disable()':
---- a lot more ------
using the opencl 1 headers ( from cuda 5.5 )it dose not even get this far it errors out on making the render3d.o --- UPDATE however it WILL link if the AMD openCL lib and headers are used from http://developer.amd.com/tools-and-sdks/heterogeneous-computing/amd-accelerated-parallel-processing-app-sdk/downloads/
|
|
« Last Edit: February 09, 2014, 11:23:38 AM by JohnVV, Reason: update »
|
Logged
|
Running OpenSUSE 42.1-64bit & ScientificLinux-6.7-64bit I don't pitch Linux to my friends,I let Microsoft do it for me!"
|
|
|
quaz0r
Fractal Molossus
Posts: 652
|
|
« Reply #24 on: February 09, 2014, 01:59:10 PM » |
|
ive never installed cuda sdk or anything, just the regular nvidia drivers installs opencl stuff that works. i was using 331 also, just installed 334 and that works also.
|
|
« Last Edit: February 09, 2014, 02:07:55 PM by quaz0r »
|
Logged
|
|
|
|
joe
|
|
« Reply #25 on: February 18, 2014, 03:47:13 PM » |
|
1.21 GIGAWATTS!!
|
|
|
Logged
|
|
|
|
mclarekin
|
|
« Reply #26 on: March 04, 2014, 02:13:05 AM » |
|
Windows openCL GPU timeout problemA while back Syntopia alerted me to TDR issues with Windows OS. So recently I followed his link http://msdn.microsoft.com/en-us/library/windows/hardware/ff569918(v=vs.85).aspxthen without any real knowledge of what I was doing I created the TDR registry keys and increased the defaults as follows: TdrDdiDelay REG_DWORD and set it to 60 seconds TdrDelay REG_DWORD and set it to 30 seconds TdrLimitTime REG_DWORD and set it to 240 seconds and now I can use openCL on a Radeon HD7660D GPU in a WIN7 64 environment without the GPU driver crashing and resetting (this mainly used to happen if I tried to use volumetric light) A big thank you to Syntopia I did wonder if I should use REG_QWORD for seen I was using 64bit OS???
|
|
|
Logged
|
|
|
|
visual.bermarte
|
|
« Reply #27 on: March 18, 2014, 05:37:27 PM » |
|
mandelbulber for mac, openCL CPU rendering tested on Intel(R) Core(TM)2 Duo CPU P7550 @ 2.26GHz VERSION: OpenCL 1.2 DRIVER_VERSION: 1.1 GeForce 9400M DRIVER_VERSION: 8.24.11 310.90.9b01 VERSION: OpenCL 1.0 *************** Intel(R) Core(TM)2 Duo CPU T7700 @ 2.40GHz VERSION: OpenCL 1.2 DRIVER_VERSION: 1.1 GeForce 8600M GT DRIVER_VERSION: CLH 1.0 VERSION: OpenCL 1.0 *************** I still need to do tests on AMD-Intel GPUs; I have used temporary variables on rotations to have kernels that are more digestible and I need to do some minor changes. I think Mavericks handles the whole openCL's affair better than previous OSXs, but, alas, I don't have a fully openCL 1.2 compatible GPU. No openCL 1.2 for Nvidia neither, only for Intel and AMD.
|
|
« Last Edit: March 18, 2014, 05:42:47 PM by visual.bermarte »
|
Logged
|
|
|
|
|
Buddhi
|
|
« Reply #29 on: March 18, 2014, 07:35:15 PM » |
|
Thanks bermarte for update. Many people waits for OSX version. About OpenCL, version 1.1 is enough. The program doesn't use any OpenCL 1.2 extensions.
|
|
|
Logged
|
|
|
|
|