Merge pull request #340 from jjfumero/feat/api/info

New calls to profile the Execution Plan: withThreadInfo and withPrintKernel

tornado-api/src/main/java/uk/ac/manchester/tornado/api/ImmutableTaskGraph.java

@@ -177,12 +177,27 @@ void disableProfiler(ProfilerMode profilerMode) {
         taskGraph.disableProfiler(profilerMode);
     }
 
-    public void withConcurrentDevices() {
+    void withConcurrentDevices() {
         taskGraph.withConcurrentDevices();
     }
 
-    public void withoutConcurrentDevices() {
+    void withoutConcurrentDevices() {
         taskGraph.withoutConcurrentDevices();
     }
 
+    void withThreadInfo() {
+        taskGraph.withThreadInfo();
+    }
+
+    void withoutThreadInfo() {
+        taskGraph.withoutThreadInfo();
+    }
+
+    void withPrintKernel() {
+        taskGraph.withPrintKernel();
+    }
+
+    void withoutPrintKernel() {
+        taskGraph.withoutPrintKernel();
+    }
 }

tornado-api/src/main/java/uk/ac/manchester/tornado/api/TaskGraph.java

@@ -897,12 +897,27 @@ void disableProfiler(ProfilerMode profilerMode) {
         taskGraphImpl.disableProfiler(profilerMode);
     }
 
-    public void withConcurrentDevices() {
+    void withConcurrentDevices() {
         taskGraphImpl.withConcurrentDevices();
     }
 
-    public void withoutConcurrentDevices() {
+    void withoutConcurrentDevices() {
         taskGraphImpl.withoutConcurrentDevices();
     }
 
+    void withThreadInfo() {
+        taskGraphImpl.withThreadInfo();
+    }
+
+    void withoutThreadInfo() {
+        taskGraphImpl.withoutThreadInfo();
+    }
+
+    void withPrintKernel() {
+        taskGraphImpl.withPrintKernel();
+    }
+
+    void withoutPrintKernel() {
+        taskGraphImpl.withoutPrintKernel();
+    }
 }

tornado-api/src/main/java/uk/ac/manchester/tornado/api/TornadoExecutionPlan.java

@@ -336,6 +336,26 @@ public TornadoExecutionPlan clearProfiles() {
         return this;
     }
 
+    public TornadoExecutionPlan withThreadInfo() {
+        tornadoExecutor.withThreadInfo();
+        return this;
+    }
+
+    public TornadoExecutionPlan withoutThreadInfo() {
+        tornadoExecutor.withoutThreadInfo();
+        return this;
+    }
+
+    public TornadoExecutionPlan withPrintKernel() {
+        tornadoExecutor.withPrintKernel();
+        return this;
+    }
+
+    public TornadoExecutionPlan withoutPrintKernel() {
+        tornadoExecutor.withoutPrintKernel();
+        return this;
+    }
+
     static class TornadoExecutor {
 
         private List<ImmutableTaskGraph> immutableTaskGraphList;
@@ -497,5 +517,21 @@ void enableProfiler(ProfilerMode profilerMode) {
         void disableProfiler(ProfilerMode profilerMode) {
             immutableTaskGraphList.forEach(immutableTaskGraph -> immutableTaskGraph.disableProfiler(profilerMode));
         }
+
+        void withThreadInfo() {
+            immutableTaskGraphList.forEach(ImmutableTaskGraph::withThreadInfo);
+        }
+
+        void withoutThreadInfo() {
+            immutableTaskGraphList.forEach(ImmutableTaskGraph::withoutThreadInfo);
+        }
+
+        void withPrintKernel() {
+            immutableTaskGraphList.forEach(ImmutableTaskGraph::withPrintKernel);
+        }
+
+        void withoutPrintKernel() {
+            immutableTaskGraphList.forEach(ImmutableTaskGraph::withoutPrintKernel);
+        }
     }
 }

tornado-api/src/main/java/uk/ac/manchester/tornado/api/TornadoTaskGraphInterface.java

@@ -115,4 +115,12 @@ public interface TornadoTaskGraphInterface extends ProfileInterface {
     void withConcurrentDevices();
 
     void withoutConcurrentDevices();
+
+    void withThreadInfo();
+
+    void withoutThreadInfo();
+
+    void withPrintKernel();
+
+    void withoutPrintKernel();
 }

tornado-api/src/main/java/uk/ac/manchester/tornado/api/memory/TaskMetaDataInterface.java

@@ -51,4 +51,8 @@ public interface TaskMetaDataInterface {
     int getDeviceIndex();
 
     void setDevice(TornadoDevice device);
+
+    boolean isPrintKernelEnabled();
+
+    void setPrintKernelFlag(boolean printKernelEnabled);
 }

tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/OCLCodeCache.java

@@ -62,7 +62,6 @@
 public class OCLCodeCache {
 
     private static final String FALSE = "False";
-    private static final String TRUE = "True";
     private static final int SPIRV_MAGIC_NUMBER = 119734787;
     private static final String OPENCL_SOURCE_SUFFIX = ".cl";
     private final boolean OPENCL_CACHE_ENABLE = Boolean.parseBoolean(getProperty("tornado.opencl.codecache.enable", FALSE));
@@ -478,15 +477,17 @@ private String getDeviceVendor() {
         return deviceContext.getPlatformContext().getPlatform().getVendor().toLowerCase().split("\\(")[0];
     }
 
-    OCLInstalledCode installFPGASource(String id, String entryPoint, byte[] source, boolean shouldCompile) { // TODO Override this method for each FPGA backend
+    OCLInstalledCode installFPGASource(String id, String entryPoint, byte[] source, boolean shouldCompile, boolean printKernel) { // TODO Override this method for each FPGA backend
         String[] compilationCommand;
         final String inputFile = fpgaSourceDir + entryPoint + OPENCL_SOURCE_SUFFIX;
         final String outputFile = fpgaSourceDir + entryPoint;
         File fpgaBitStreamFile = new File(outputFile);
 
         appendSourceToFile(source, entryPoint);
 
-        RuntimeUtilities.maybePrintSource(source);
+        if (printKernel) {
+            RuntimeUtilities.dumpKernel(source);
+        }
 
         String[] commandRename;
         String[] linkCommand = null;
@@ -583,7 +584,9 @@ public OCLInstalledCode installSource(TaskMetaData meta, String id, String entry
             appendSourceToFile(source, entryPoint);
         }
 
-        RuntimeUtilities.maybePrintSource(source);
+        if (meta.isPrintKernelEnabled()) {
+            RuntimeUtilities.dumpKernel(source);
+        }
 
         final long t0 = System.nanoTime();
         program.build(meta.getCompilerFlags());
@@ -599,16 +602,16 @@ public OCLInstalledCode installSource(TaskMetaData meta, String id, String entry
                 debug(log);
             }
             final Path outDir = resolveLogDirectory();
-            final String identifier = id + "-" + entryPoint;
+            final String identifier = STR."\{id}-\{entryPoint}";
             error("Unable to compile task %s: check logs at %s/%s.log", identifier, outDir.toAbsolutePath(), identifier);
 
-            File file = new File(outDir + "/" + identifier + ".log");
+            File file = new File(STR."\{outDir}/\{identifier}.log");
             try (FileOutputStream fos = new FileOutputStream(file)) {
                 fos.write(log.getBytes());
             } catch (IOException e) {
                 error("unable to write error log: ", e.getMessage());
             }
-            file = new File(outDir + "/" + identifier + OPENCL_SOURCE_SUFFIX);
+            file = new File(STR."\{outDir}/\{identifier}\{OPENCL_SOURCE_SUFFIX}");
             try (FileOutputStream fos = new FileOutputStream(file)) {
                 fos.write(source);
             } catch (IOException e) {
@@ -633,13 +636,13 @@ public OCLInstalledCode installSource(TaskMetaData meta, String id, String entry
             if (meta.shouldPrintCompileTimes()) {
                 debug("compile: kernel %s opencl %.9f\n", entryPoint, (t1 - t0) * 1e-9f);
             }
-            cache.put(id + "-" + entryPoint, code);
+            cache.put(STR."\{id}-\{entryPoint}", code);
 
             // BUG Apple does not seem to like implementing the OpenCL spec
-            // properly, this causes a sigfault.
+            // properly, this causes a SIGFAULT.
             if ((OPENCL_CACHE_ENABLE || OPENCL_DUMP_BINS) && !deviceContext.getPlatformContext().getPlatform().getVendor().equalsIgnoreCase("Apple")) {
                 final Path outDir = resolveCacheDirectory();
-                program.dumpBinaries(outDir.toAbsolutePath() + "/" + entryPoint);
+                program.dumpBinaries(STR."\{outDir.toAbsolutePath()}/\{entryPoint}");
             }
         } else {
             warn("\tunable to compile %s", entryPoint);
@@ -672,11 +675,11 @@ private OCLInstalledCode installBinary(String id, String entryPoint, byte[] bina
             long afterLoad = (TornadoOptions.TIME_IN_NANOSECONDS) ? System.nanoTime() : System.currentTimeMillis();
 
             if (PRINT_LOAD_TIME) {
-                System.out.println("Binary load time: " + (afterLoad - beforeLoad) + (TornadoOptions.TIME_IN_NANOSECONDS ? " ns" : " ms") + " \n");
+                System.out.println(STR."Binary load time: \{afterLoad - beforeLoad}\{TornadoOptions.TIME_IN_NANOSECONDS ? " ns" : " ms"} \n");
             }
 
             if (program == null) {
-                throw new OCLException("unable to load binary for " + entryPoint);
+                throw new OCLException(STR."unable to load binary for \{entryPoint}");
             }
 
             program.build("");

tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/OCLDeviceContext.java

@@ -394,16 +394,19 @@ public int readBuffer(long bufferId, long offset, long bytes, long hostPointer,
                 : null), EventDescriptor.DESC_READ_SEGMENT, queue);
     }
 
+    @Override
     public int enqueueBarrier(int[] events) {
         long oclEvent = queue.enqueueBarrier(oclEventPool.serialiseEvents(events, queue) ? oclEventPool.waitEventsBuffer : null);
         return queue.getOpenclVersion() < 120 ? -1 : oclEventPool.registerEvent(oclEvent, EventDescriptor.DESC_SYNC_BARRIER, queue);
     }
 
+    @Override
     public int enqueueMarker(int[] events) {
         long oclEvent = queue.enqueueMarker(oclEventPool.serialiseEvents(events, queue) ? oclEventPool.waitEventsBuffer : null);
         return queue.getOpenclVersion() < 120 ? -1 : oclEventPool.registerEvent(oclEvent, EventDescriptor.DESC_SYNC_MARKER, queue);
     }
 
+    @Override
     public void reset() {
         oclEventPool.reset();
         codeCache.reset();
@@ -483,13 +486,15 @@ public void retainEvent(int localEventId) {
         oclEventPool.retainEvent(localEventId);
     }
 
+    @Override
     public Event resolveEvent(int event) {
         if (event == -1) {
             return EMPTY_EVENT;
         }
         return new OCLEvent(oclEventPool.getDescriptor(event).getNameDescription(), queue, event, oclEventPool.getOCLEvent(event));
     }
 
+    @Override
     public void flush() {
         queue.flush();
     }
@@ -498,10 +503,12 @@ public void finish() {
         queue.finish();
     }
 
+    @Override
     public void flushEvents() {
         queue.flushEvents();
     }
 
+    @Override
     public boolean isKernelAvailable() {
         return codeCache.isKernelAvailable();
     }
@@ -510,15 +517,18 @@ public OCLInstalledCode installCode(OCLCompilationResult result) {
         return installCode(result.getMeta(), result.getId(), result.getName(), result.getTargetCode());
     }
 
+    @Override
     public OCLInstalledCode installCode(TaskMetaData meta, String id, String entryPoint, byte[] code) {
         entryPoint = checkKernelName(entryPoint);
         return codeCache.installSource(meta, id, entryPoint, code);
     }
 
-    public OCLInstalledCode installCode(String id, String entryPoint, byte[] code, boolean shouldCompile) {
-        return codeCache.installFPGASource(id, entryPoint, code, shouldCompile);
+    @Override
+    public OCLInstalledCode installCode(String id, String entryPoint, byte[] code, boolean shouldCompile, boolean printKernel) {
+        return codeCache.installFPGASource(id, entryPoint, code, shouldCompile, printKernel);
     }
 
+    @Override
     public boolean isCached(String id, String entryPoint) {
         entryPoint = checkKernelName(entryPoint);
         return codeCache.isCached(id + "-" + entryPoint);
@@ -535,6 +545,7 @@ public OCLInstalledCode getInstalledCode(String id, String entryPoint) {
         return codeCache.getInstalledCode(id, entryPoint);
     }
 
+    @Override
     public OCLCodeCache getCodeCache() {
         return this.codeCache;
     }

tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/OCLDeviceContextInterface.java

@@ -12,7 +12,7 @@
  *
  * This code is distributed in the hope that it will be useful, but WITHOUT
  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
  * version 2 for more details (a copy is included in the LICENSE file that
  * accompanied this code).
  *
@@ -42,7 +42,7 @@ public interface OCLDeviceContextInterface extends TornadoDeviceContext {
 
     OCLInstalledCode getInstalledCode(String id, String entryPoint);
 
-    OCLInstalledCode installCode(String id, String entryPoint, byte[] code, boolean shouldCompile);
+    OCLInstalledCode installCode(String id, String entryPoint, byte[] code, boolean shouldCompile, boolean printKernel);
 
     OCLInstalledCode installCode(OCLCompilationResult result);
 

tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/runtime/OCLTornadoDevice.java

@@ -283,7 +283,7 @@ private TornadoInstalledCode compileTask(SchedulableTask task) {
             OCLInstalledCode installedCode;
             if (OCLBackend.isDeviceAnFPGAAccelerator(deviceContext)) {
                 // A) for FPGA
-                installedCode = deviceContext.installCode(result.getId(), result.getName(), result.getTargetCode(), task.shouldCompile());
+                installedCode = deviceContext.installCode(result.getId(), result.getName(), result.getTargetCode(), task.shouldCompile(), task.meta().isPrintKernelEnabled());
             } else {
                 // B) for CPU multi-core or GPU
                 installedCode = deviceContext.installCode(result);
@@ -318,7 +318,7 @@ private TornadoInstalledCode compilePreBuiltTask(SchedulableTask task) {
             OCLInstalledCode installedCode;
             if (OCLBackend.isDeviceAnFPGAAccelerator(deviceContext)) {
                 // A) for FPGA
-                installedCode = deviceContext.installCode(task.getId(), executable.getEntryPoint(), source, task.shouldCompile());
+                installedCode = deviceContext.installCode(task.getId(), executable.getEntryPoint(), source, task.shouldCompile(), task.meta().isPrintKernelEnabled());
             } else {
                 // B) for CPU multi-core or GPU
                 installedCode = deviceContext.installCode(executable.meta(), task.getId(), executable.getEntryPoint(), source);

tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/tests/TestOpenCLTornadoCompiler.java

@@ -1,5 +1,5 @@
 /*
- * This file is part of Tornado: A heterogeneous programming framework: 
+ * This file is part of Tornado: A heterogeneous programming framework:
  * https://github.com/beehive-lab/tornadovm
  *
  * Copyright (c) 2013-2020, APT Group, Department of Computer Science,
@@ -12,7 +12,7 @@
  *
  * This code is distributed in the hope that it will be useful, but WITHOUT
  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
  * version 2 for more details (a copy is included in the LICENSE file that
  * accompanied this code).
  *
@@ -71,7 +71,7 @@ public static void main(String[] args) {
         OCLInstalledCode code = codeCache.installSource(meta, "saxpy", "saxpy", source);
 
         String generatedSourceCode = code.getGeneratedSourceCode();
-        if (TornadoOptions.PRINT_SOURCE) {
+        if (meta.isPrintKernelEnabled()) {
             System.out.println("Compiled code: " + generatedSourceCode);
         }
     }