diff options
| -rw-r--r-- | src/com/jogamp/opencl/CLContext.java | 13 | ||||
| -rw-r--r-- | src/com/jogamp/opencl/CLDevice.java | 11 | ||||
| -rw-r--r-- | src/com/jogamp/opencl/CLProgram.java | 14 | ||||
| -rw-r--r-- | src/com/jogamp/opencl/CLProgramBuilder.java | 20 | ||||
| -rw-r--r-- | test/com/jogamp/opencl/CLCommandQueueTest.java | 6 |
5 files changed, 43 insertions, 21 deletions
diff --git a/src/com/jogamp/opencl/CLContext.java b/src/com/jogamp/opencl/CLContext.java index c3bd0e3f..163375eb 100644 --- a/src/com/jogamp/opencl/CLContext.java +++ b/src/com/jogamp/opencl/CLContext.java @@ -287,6 +287,7 @@ public class CLContext extends CLObjectResource { */ public CLProgram createProgram(Map<CLDevice, byte[]> binaries) { CLProgram program = CLProgram.create(this, binaries); + program.setNoSource(); programs.add(program); return program; } @@ -486,11 +487,8 @@ public class CLContext extends CLObjectResource { private void release(Collection<? extends CLResource> resources) { // resources remove themselves when released, see above - if(!resources.isEmpty()) { - CLResource[] array = resources.toArray(new CLResource[resources.size()]); - for (CLResource resource : array) { - resource.release(); - } + while(!resources.isEmpty()) { + resources.iterator().next().release(); } } @@ -509,9 +507,8 @@ public class CLContext extends CLObjectResource { synchronized(queuesMap) { final Collection<List<CLCommandQueue>> queuesList = queuesMap.values(); - for( Iterator<List<CLCommandQueue>> queuesI = queuesList.iterator(); queuesI.hasNext(); ) { - release(queuesI.next()); - } + while(!queuesList.isEmpty()) + release(queuesList.iterator().next()); } } finally { diff --git a/src/com/jogamp/opencl/CLDevice.java b/src/com/jogamp/opencl/CLDevice.java index b47deb2b..25fb009f 100644 --- a/src/com/jogamp/opencl/CLDevice.java +++ b/src/com/jogamp/opencl/CLDevice.java @@ -635,6 +635,15 @@ public class CLDevice extends CLObject { } /** + * Returns {@link #isExtensionAvailable}("cl_khr_icd"). + * @see #getExtensions() + */ + @CLProperty("cl_khr_icd") + public boolean isICDAvailable() { + return isExtensionAvailable("cl_khr_icd"); + } + + /** * Returns {@link #isExtensionAvailable}("cl_khr_gl_sharing") || {@link #isExtensionAvailable}("cl_APPLE_gl_sharing"). * @see #getExtensions() */ @@ -850,7 +859,7 @@ public class CLDevice extends CLObject { ROUND_TO_NEAREST(CL_FP_ROUND_TO_NEAREST), /** - * round to +ve and –ve infinity rounding modes supported. + * round to positive and negative infinity rounding modes supported. */ ROUND_TO_INF(CL_FP_ROUND_TO_INF), diff --git a/src/com/jogamp/opencl/CLProgram.java b/src/com/jogamp/opencl/CLProgram.java index 3919d374..9dd3db82 100644 --- a/src/com/jogamp/opencl/CLProgram.java +++ b/src/com/jogamp/opencl/CLProgram.java @@ -70,6 +70,8 @@ public class CLProgram extends CLObjectResource { private boolean executable; private boolean released; + /** Set if program created from binary, or else getting source can crash the driver on Macs. */ + private boolean noSource; private CLProgram(CLContext context, long id) { super(context, id); @@ -592,10 +594,20 @@ public class CLProgram extends CLObjectResource { } /** + * Must set this if the program is created from binary so we know not to call getSource(), + * which can SIGSEGV on Macs if there is no source. + */ + public void setNoSource() { + noSource = true; + } + + /** * Returns the source code of this program. Note: sources are not cached, * each call of this method calls into Open */ public String getSource() { + if(noSource) + return ""; // some drivers return IVE codes if the program haven't been built from source. try{ return getProgramInfoString(CL_PROGRAM_SOURCE); @@ -810,7 +822,7 @@ public class CLProgram extends CLObjectResource { public final static String UNSAFE_MATH = "-cl-unsafe-math-optimizations"; /** - * Allow optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or ±∞. + * Allow optimizations for floating-point arithmetic that assume that arguments and results are not NaNs or plus/minus infinity. * This option may violate the OpenCL numerical compliance requirements defined in in section 7.4 for * single-precision floating-point, section 9.3.9 for double-precision floating-point, and edge case behavior in section 7.5. */ diff --git a/src/com/jogamp/opencl/CLProgramBuilder.java b/src/com/jogamp/opencl/CLProgramBuilder.java index e80b9923..dd0c64f9 100644 --- a/src/com/jogamp/opencl/CLProgramBuilder.java +++ b/src/com/jogamp/opencl/CLProgramBuilder.java @@ -276,14 +276,16 @@ public final class CLProgramBuilder implements CLProgramConfiguration, Serializa private void writeObject(ObjectOutputStream out) throws IOException { out.defaultWriteObject(); - String suffix = null; + String suffix = ""; if(!binariesMap.isEmpty()) { - CLPlatform platform = binariesMap.keySet().iterator().next().getPlatform(); - suffix = platform.getICDSuffix(); + CLDevice device = binariesMap.keySet().iterator().next(); + if(device.isICDAvailable()) + suffix = device.getPlatform().getICDSuffix(); } - out.writeUTF(suffix); // null if we have no binaries or no devices specified + // empty string if we have no binaries or no devices specified, or if cl_khr_icd isn't supported + out.writeUTF(suffix); out.writeInt(binariesMap.size()); // may be 0 for (Map.Entry<CLDevice, byte[]> entry : binariesMap.entrySet()) { @@ -299,22 +301,22 @@ public final class CLProgramBuilder implements CLProgramConfiguration, Serializa private void readObject(ObjectInputStream in) throws IOException, ClassNotFoundException { in.defaultReadObject(); - String suffix = in.readUTF(); + String suffix = in.readUTF(); // empty string means no suffix was written; just picks first platform CLPlatform platform = null; for (CLPlatform p : CLPlatform.listCLPlatforms()) { - if(p.getICDSuffix().equals(suffix)) { + if(suffix.isEmpty() || p.getICDSuffix().equals(suffix)) { platform = p; break; } } - + this.binariesMap = new LinkedHashMap<CLDevice, byte[]>(); List<CLDevice> devices; if(platform != null) { - devices = new ArrayList(Arrays.asList(platform.listCLDevices())); + devices = new ArrayList<CLDevice>(Arrays.asList(platform.listCLDevices())); }else{ - devices = Collections.EMPTY_LIST; + devices = Collections.emptyList(); } int mapSize = in.readInt(); diff --git a/test/com/jogamp/opencl/CLCommandQueueTest.java b/test/com/jogamp/opencl/CLCommandQueueTest.java index c9b1b567..672cc8b0 100644 --- a/test/com/jogamp/opencl/CLCommandQueueTest.java +++ b/test/com/jogamp/opencl/CLCommandQueueTest.java @@ -474,7 +474,9 @@ public class CLCommandQueueTest extends UITestCase { @Override public void run() { - int groupSize = queue2.getDevice().getMaxWorkItemSizes()[0]; + int maxWorkItemSize = queue2.getDevice().getMaxWorkItemSizes()[0]; + int kernelWorkGroupSize = (int)vectorAddKernel2.getWorkGroupSize( queue2.getDevice() ); + int localWorkSize = Math.min( maxWorkItemSize, kernelWorkGroupSize ); fillBuffer(clBufferA2.buffer, 12345); fillBuffer(clBufferB2.buffer, 67890); @@ -488,7 +490,7 @@ public class CLCommandQueueTest extends UITestCase { // System.out.println("D kernels"); CLEventList events2 = new CLEventList(2); - queue2.put1DRangeKernel(vectorAddKernel2, 0, elements, groupSize, events2); + queue2.put1DRangeKernel(vectorAddKernel2, 0, elements, localWorkSize, events2); queue2.putReadBuffer(clBufferD, false, events2); barrier.waitFor(queue2, events2); |