Skip to content
Snippets Groups Projects

AparapiFractals

Closed AparapiFractals
stefanettim/aparapi-examples:b7 into master
Closed Marco Stefanetti requested to merge stefanettim/aparapi-examples:b7 into master

AparapiFractals

  • Mandelbrot explorer : mouse controls and zoom

af

  • benchmark : no graphics, tests on different devices
Benchmark - Soft 
  canvas        : 500 x 500 
  maxIterations : 100,000 
  region        : -2.0000000000000000d,-2.0000000000000000d 2.0000000000000000d,2.0000000000000000d 
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
| KEY                                      | DEVICE                         | LOCAL SIZES  | NAME                           | Elapsed(ms)|
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
| AUTO                                     | NVIDIA<GPU>                    | 10 x 25      | GeForce GTX 1650 SUPER         |        363 |
| NVIDIA<GPU> (11681328)                   | NVIDIA<GPU>                    | 8 x 8        | GeForce GTX 1650 SUPER         |        330 |
| Java Alternative Algorithm (-2)          | Java Alternative Algorithm     | 8 x 8        | Java Alternative Algorithm     |        427 |
| Java Thread Pool (-3)                    | Java Thread Pool               | 8 x 8        | Java Thread Pool               |        390 |
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
============================================================================================================================================
Benchmark - Hard 
  canvas        : 500 x 500 
  maxIterations : 1,000,000 
  region        : -2.0000000000000000d,-2.0000000000000000d 2.0000000000000000d,2.0000000000000000d 
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
| KEY                                      | DEVICE                         | LOCAL SIZES  | NAME                           | Elapsed(ms)|
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
| AUTO                                     | NVIDIA<GPU>                    | 10 x 25      | GeForce GTX 1650 SUPER         |       3222 |
| NVIDIA<GPU> (20593056)                   | NVIDIA<GPU>                    | 8 x 8        | GeForce GTX 1650 SUPER         |       3042 |
| Java Alternative Algorithm (-2)          | Java Alternative Algorithm     | 8 x 8        | Java Alternative Algorithm     |       3986 |
| Java Thread Pool (-3)                    | Java Thread Pool               | 8 x 8        | Java Thread Pool               |       4786 |
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+

Merge request reports

Loading
Loading

Activity

Filter activity
  • Approvals
  • Assignees & reviewers
  • Comments (from bots)
  • Comments (from users)
  • Commits & branches
  • Edits
  • Labels
  • Lock status
  • Mentions
  • Merge request status
  • Tracking
  • Jeffrey Phillips Freeman
    Jeffrey Phillips Freeman @freemo ·
    Owner

    Just ran this, it mostly works great but with a few issues...

    1. the GUI freezes when the background operation is working. In other words if i run a benchmark the GUI iis frozen until the benchmark finishes. If i scroll in and the scrolling is slow enough the GUI freezes as well. GUI code should never run code that isnt responsive without sticking it into a seperate thread.

    2. It doesnt seem to handle multiple graphics cards directly. While at first it was doing a fine job of running off my gpu, when i tried to switch to one of the other GPUs in the list it didnt actually switch which GPU it was utilizing and continued as normal on the first GPU

    3. This may be related, probably is, but when i try to run the hard benchmark from the command line it errors out for me (error pasted at the bottom).

    4. related closesly to #3 if i run the benchmark from the GUI I see similar errors and can not resume operation of the GUI.

     38) AparapiFractals - soft benchmark 
 39) AparapiFractals - hard benchmark 

Enter your selection, or q/Q to quit: 39
#pragma OPENCL EXTENSION cl_khr_fp64 : enable

typedef struct This_s{
   int wmax;
   int hmax;
   double cx1;
   double wx;
   double cy1;
   double hy;
   int max_iterations;
   __global int *result;
   int result__javaArrayLength0;
   int result__javaArrayDimension0;
   int result__javaArrayLength1;
   int result__javaArrayDimension1;
   int passid;
}This;
int get_pass_id(This *this){
   return this->passid;
}
__kernel void run(
   int wmax, 
   int hmax, 
   double cx1, 
   double wx, 
   double cy1, 
   double hy, 
   int max_iterations, 
   __global int *result, 
   int result__javaArrayLength0, 
   int result__javaArrayDimension0, 
   int result__javaArrayLength1, 
   int result__javaArrayDimension1, 
   int passid
){
   This thisStruct;
   This* this=&thisStruct;
   this->wmax = wmax;
   this->hmax = hmax;
   this->cx1 = cx1;
   this->wx = wx;
   this->cy1 = cy1;
   this->hy = hy;
   this->max_iterations = max_iterations;
   this->result = result;
   this->result__javaArrayLength0 = result__javaArrayLength0;
   this->result__javaArrayDimension0 = result__javaArrayDimension0;
   this->result__javaArrayLength1 = result__javaArrayLength1;
   this->result__javaArrayDimension1 = result__javaArrayDimension1;
   this->passid = passid;
   {
      int w = get_global_id(0);
      int h = get_global_id(1);
      if (w<this->wmax && h<this->hmax){
         double cx = this->cx1 + ((double)w * this->wx);
         double cy = this->cy1 + ((double)h * this->hy);
         double xn = cx;
         double yn = cy;
         double y2 = cy * cy;
         int t = 0;
         t = 0;
         for (; t<this->max_iterations && ((xn * xn) + y2)<4.0; t++){
            yn = ((2.0 * xn) * yn) + cy;
            xn = ((xn * xn) - y2) + cx;
            y2 = yn * yn;
         }
         (&this->result[w * this->result__javaArrayDimension0])[h]  = t;
      }
      return;
   }
}


Starting benchmark...
Example : 'GeForce GTX 1650 SUPER' soft 348ms, hard 3058ms 
Example : 'Java Alternative Algorithm', AMD 3700X, soft 390ms, hard 3993ms 

OperatingSystem: Linux CPU:32 amd64
============================================================================================================================================
Benchmark - Hard 
  canvas        : 500 x 500 
  maxIterations : 1,000,000 
  region        : -2.0000000000000000d,-2.0000000000000000d 2.0000000000000000d,2.0000000000000000d 

+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
| KEY                                      | DEVICE                         | LOCAL SIZES  | NAME                           | Elapsed(ms)|
+------------------------------------------+--------------------------------+--------------+--------------------------------+------------+
after clEnqueueNDRangeKernel, globalSize[0] = 500, localSize[0] = 20
after clEnqueueNDRangeKernel, globalSize[1] = 500, localSize[1] = 25
!!!!!!! clEnqueueNDRangeKernel() failed invalid work group size
Nov 18, 2020 5:44:28 PM com.aparapi.internal.kernel.KernelRunner fallBackToNextDevice
WARNING: Device failed for AfKernel, devices={AMD<GPU>|AMD<GPU>|AMD<GPU>|AMD<GPU>|Java Alternative Algorithm|Java Thread Pool}: OpenCL execution seems to have failed (runKernelJNI returned -54)
com.aparapi.internal.exception.AparapiException: OpenCL execution seems to have failed (runKernelJNI returned -54)
	at com.aparapi.internal.kernel.KernelRunner.executeOpenCL(KernelRunner.java:1263)
	at com.aparapi.internal.kernel.KernelRunner.executeInternalInner(KernelRunner.java:1722)
	at com.aparapi.internal.kernel.KernelRunner.executeInternalOuter(KernelRunner.java:1383)
	at com.aparapi.internal.kernel.KernelRunner.execute(KernelRunner.java:1374)
	at com.aparapi.Kernel.execute(Kernel.java:2897)
	at com.aparapi.Kernel.execute(Kernel.java:2854)
	at com.aparapi.Kernel.execute(Kernel.java:2794)
	at com.aparapi.examples.afmandelbrot.AfAparapiUtils.execute(AfAparapiUtils.java:156)
	at com.aparapi.examples.afmandelbrot.AfAparapiUtils.benchmark(AfAparapiUtils.java:201)
	at com.aparapi.examples.afmandelbrot.AfMain.benchmarkHard(AfMain.java:331)
	at com.aparapi.examples.afmandelbrot.AfBenchmark.main(AfBenchmark.java:42)
	at com.aparapi.examples.All.selected(All.java:234)
	at com.aparapi.examples.All.main(All.java:99)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
	at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
	at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
	at java.base/java.lang.reflect.Method.invoke(Method.java:564)
	at org.codehaus.mojo.exec.ExecJavaMojo$1.run(ExecJavaMojo.java:282)
	at java.base/java.lang.Thread.run(Thread.java:832)
Nov 18, 2020 5:44:28 PM com.aparapi.internal.kernel.KernelRunner fallBackToNextDevice
WARNING: Trying next device: AMD<GPU>
| AUTO                                     | AMD<GPU>                       | 20 x 25      | gfx900                         |       1340 |
[WARNING] 
java.lang.AssertionError: user supplied Device incompatible with current EXECUTION_MODE or getTargetDevice(); device = AMD<GPU>; kernel = AfKernel, devices={AMD<GPU>|AMD<GPU>|AMD<GPU>|Java Alternative Algorithm|Java Thread Pool}
    at com.aparapi.internal.kernel.KernelRunner.executeInternalInner (KernelRunner.java:1425)
    at com.aparapi.internal.kernel.KernelRunner.executeInternalOuter (KernelRunner.java:1383)
    at com.aparapi.internal.kernel.KernelRunner.execute (KernelRunner.java:1374)
    at com.aparapi.Kernel.execute (Kernel.java:2897)
    at com.aparapi.Kernel.execute (Kernel.java:2854)
    at com.aparapi.Kernel.execute (Kernel.java:2794)
    at com.aparapi.examples.afmandelbrot.AfAparapiUtils.execute (AfAparapiUtils.java:156)
    at com.aparapi.examples.afmandelbrot.AfAparapiUtils.benchmark (AfAparapiUtils.java:201)
    at com.aparapi.examples.afmandelbrot.AfMain.benchmarkHard (AfMain.java:331)
    at com.aparapi.examples.afmandelbrot.AfBenchmark.main (AfBenchmark.java:42)
    at com.aparapi.examples.All.selected (All.java:234)
    at com.aparapi.examples.All.main (All.java:99)
    at jdk.internal.reflect.NativeMethodAccessorImpl.invoke0 (Native Method)
    at jdk.internal.reflect.NativeMethodAccessorImpl.invoke (NativeMethodAccessorImpl.java:62)
    at jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke (DelegatingMethodAccessorImpl.java:43)
    at java.lang.reflect.Method.invoke (Method.java:564)
    at org.codehaus.mojo.exec.ExecJavaMojo$1.run (ExecJavaMojo.java:282)
    at java.lang.Thread.run (Thread.java:832)
  • Marco Stefanetti added 1 commit

    added 1 commit

    • e1be5b34 - improved GUI and multi GPU testing

    Compare with previous version

  • Marco Stefanetti
    Marco Stefanetti @stefanettim ·
    Author Contributor

    1. Improved GUI, It's almost all on separate threads.
      Zoom scrolling at deeper zoom level is not fluent and auto-disables realtime refresh.

    2. I bought a second GPU and tested GPU switching.

    3. I have no idea, I never had this error.
      clEnqueueNDRangeKernel() failed invalid work group size may be related to global and local width.
      Let's see what happens with this new version.

    4. GUI and Benchmark use the same kernel.

    Benchmark with 2 GPUs :

    AparapiFractals - Mandelbrot Benchmark - Soft 
  image size     : 500 x 500 
  maxIterations  : 100.000 
  complex region : -2,0000000000000000d,-2,0000000000000000d 2,0000000000000000d,2,0000000000000000d 

+-----+--------------------------------+----------------+--------------------------------------------+----------+--------+------------+
|Type | shortDescription               | deviceId       | Name                                       | LSizes   | ExMode | Elapsed(ms)|
+-----+--------------------------------+----------------+--------------------------------------------+----------+--------+------------+
| GPU | NVIDIA<GPU>                    | 2116615257056  | GeForce GTX 1650 SUPER                     | 10 x 25  |   AUTO |        358 |
| GPU | NVIDIA<GPU>                    | 2116615257056  | GeForce GTX 1650 SUPER                     | 8 x 8    |    GPU |        335 |
| GPU | AMD<GPU>                       | 2116618425296  | Oland                                      | 8 x 8    |    GPU |       1259 |
| ALT | Java Alternative Algorithm     | -2             | Java Alternative Algorithm                 | 8 x 8    |    JTP |        427 |
| JTP | Java Thread Pool               | -3             | Java Thread Pool                           | 8 x 8    |    JTP |        399 |
+-----+--------------------------------+----------------+--------------------------------------------+----------+--------+------------+

=======================================================================================================================================

Profiles by Kernel Subclass (mean elapsed times in milliseconds)

Device               Count    CLASS_MODEL_BUILT          INIT_JNI  OPENCL_GENERATED   OPENCL_COMPILED   PREPARE_EXECUTE          EXECUTED  Total
----------------- [[ AfKernel ]] ---------------------------------------------------------------------------------------------------
NVIDIA<GPU>          5                   28,393            31,012             0,836             0,388             0,044           139,218  199,892
AMD<GPU>             2                    0,006             0,015             0,000            37,955             0,029           630,363  668,368
Java Thread Pool     4                    0,000             0,000             0,000             0,000             0,009           207,219  207,227
  • Marco Stefanetti added 1 commit

    added 1 commit

    • 674fb3d5 - fixed GUI refresh while dragging

    Compare with previous version

  • Marco Stefanetti closed

    closed

Please register or sign in to reply