Grcuda 85 support cusparse (#25)

* initial commit, folder creation for cusparse

* added function to CUSPARSERegistry, missing: nfi functions' input settings

* nfi signatures completed, missing: Desc types mgmt

* final commit from my machine :(

* enum addition and casting

* completed enum revision

* added first test cusparseCoo

* added SpMV test, not working

* Removed useless function instantiations in CUSPARSERegistry. For the moment being we are exposing to the user the functions to create various matrix descripts as well as creating/destroying the handle. This needs to be changed

* Almost functioning version of testSpMV. The error that i'm getting now is related to the enums, but at least no more polyglot exceptions

* initial support for cusparse

* cleaned code, more tests added

* minor additions to tests

* minor fixes to context and options for enabling cusparse

* formatting and cleaning complete

* minor fixes to context options in cusparse tests

* changelog updated

* Removed unused imports

* Modified libcusparse.so.11 -> libcusparse.so

* updated copyright for new files

* added suport for async functions, implementation of non-exposed functions

* removed useless initialization

* added sparseSgemvi

* begun implementation of proxiesn [commit della svolta]

* added basic functions to proxyspmv, to be testes

* added proxy for Sgemvi

* Completed functions for SpMV and Sgemvi, context creation missing

* tests ready for proxies, not working (context issues)

* working on contexts

* context creation fails

* minor additions to sparse proxy

* createCoo now works

* proxies all right, non valid handle in buffersize function

* IT WORKS

* initial steps for tests implementation for Sgemvi and SpMV with CSR format

* working tests for coo and csr with spmv, sgemvi does not work (does not update the vector passed as input)

* minor fixes to tests

* finished testing sgemvi and spmv

* finished testing sgemvi and spmv

* added test for libraries integration

* minor fixes, all good, streams' functioning for libraries interoperability checked with profiler

* changelog updated

* tests

* Added breaks to switch statement

* partially working tests for TGemvi and SpMV

* Fixed spmv Tests for coo and csr, gemvi still needs to be fixed

* TGemvi now works with data types C and S

* added streams syncing to tests

* added syncing, sometimes (after mx clean) crs/coo do not work, tgemvi does not work with double types, despite syncing

* Removed double and double complex from tests

* GrCUDAOptions updated for cuSPARSE

* fixed context

* removed ternary expressions

* small cleanups; fixed tracking of array dependencies not working in cusparse

* updated changelog

Co-authored-by: Francesco Sgherzi <[email protected]>
Co-authored-by: Francesco Sgherzi <[email protected]>
Co-authored-by: Alberto Parravicini <[email protected]>

Author: 3 people

CHANGELOG.md

@@ -1,3 +1,13 @@
+# 2021-11-21
+
+* Enabled support for cuSPARSE
+  * Operations with sparse matrices from cuSPARSE library are now supported
+  * **Known limitation:** Not all data types are supported: in particular Tgemvi does not support double data types 
+  (both complex and not)
+  * Concurrent operations on parallel streams were analyzed using Nvidia Profiler
+
+=======
+
 # 2021-11-17
 
 * Added the support of precise timing of kernels, for debugging and complex scheduling policies
@@ -22,6 +32,7 @@
 * Changed all the print in the source code in log events, with different logging levels
 * Added documentation about logging in docs
 
+
 # 2021-10-13
 
 * Enabled support for cuBLAS and cuML in the async scheduler

grcuda-data

@@ -40,13 +40,14 @@
 import java.util.Collection;
 import java.util.function.Function;
 
-import com.nvidia.grcuda.runtime.executioncontext.ExecutionPolicyEnum;
-import com.nvidia.grcuda.test.util.GrCUDATestUtil;
+import org.graalvm.polyglot.Context;
+import org.graalvm.polyglot.Value;
 import org.junit.Test;
 import org.junit.runner.RunWith;
 import org.junit.runners.Parameterized;
-import org.graalvm.polyglot.Context;
-import org.graalvm.polyglot.Value;
+
+import com.nvidia.grcuda.runtime.executioncontext.ExecutionPolicyEnum;
+import com.nvidia.grcuda.test.util.GrCUDATestUtil;
 
 @RunWith(Parameterized.class)
 public class CUBLASTest {
@@ -243,8 +244,8 @@ public static void assertOutputVectorIsCorrect(int len, Value deviceArray,
         }
     }
 
-    private void assertOutputVectorIsCorrect(int len, Value deviceArray,
-                    Function<Integer, Integer> outFunc) {
+    void assertOutputVectorIsCorrect(int len, Value deviceArray,
+                                     Function<Integer, Integer> outFunc) {
         CUBLASTest.assertOutputVectorIsCorrect(len, deviceArray, outFunc, this.typeChar);
     }
 

projects/com.nvidia.grcuda.test/src/com/nvidia/grcuda/test/cudalibraries/CUBLASTest.java

@@ -127,5 +127,3 @@ private static boolean isOptionRedundantForSync(GrCUDATestOptionsStruct options)
         return false;
     }
 }
-
-

projects/com.nvidia.grcuda.test/src/com/nvidia/grcuda/test/cudalibraries/CUSPARSETest.java

@@ -37,6 +37,7 @@
 
 import com.nvidia.grcuda.cudalibraries.cublas.CUBLASRegistry;
 import com.nvidia.grcuda.cudalibraries.cuml.CUMLRegistry;
+import com.nvidia.grcuda.cudalibraries.cusparse.CUSPARSERegistry;
 import com.nvidia.grcuda.cudalibraries.tensorrt.TensorRTRegistry;
 import com.nvidia.grcuda.functions.BindAllFunction;
 import com.nvidia.grcuda.functions.BindFunction;
@@ -151,6 +152,11 @@ public GrCUDAContext(Env env) {
             namespace.addNamespace(trt);
             new TensorRTRegistry(this).registerTensorRTFunctions(trt);
         }
+        if (grCUDAOptionMap.isCuSPARSEEnabled()) {
+            Namespace sparse = new Namespace(CUSPARSERegistry.NAMESPACE);
+            namespace.addNamespace(sparse);
+            new CUSPARSERegistry(this).registerCUSPARSEFunctions(sparse);
+        }
         this.rootNamespace = namespace;
     }
 
@@ -196,6 +202,7 @@ public ConcurrentHashMap<Class<?>, CallTarget> getMapCallTargets() {
         return uncachedMapCallTargets;
     }
 
+
     /**
      * Compute the maximum number of concurrent threads that can be spawned by GrCUDA.
      * This value is usually smaller or equal than the number of logical CPU threads available on the machine.
@@ -215,4 +222,4 @@ public GrCUDAOptionMap getOptions() {
     public void cleanup() {
         this.grCUDAExecutionContext.cleanup();
     }
-}
+}

projects/com.nvidia.grcuda.test/src/com/nvidia/grcuda/test/util/GrCUDATestUtil.java

@@ -166,6 +166,14 @@ public String getCuMLLibrary(){
         return (String) getOptionValueFromOptionKey(GrCUDAOptions.CuMLLibrary);
     }
 
+    public Boolean isCuSPARSEEnabled(){
+        return (Boolean) getOptionValueFromOptionKey(GrCUDAOptions.CuSPARSEEnabled);
+    }
+
+    public String getCuSPARSELibrary(){
+        return (String) getOptionValueFromOptionKey(GrCUDAOptions.CuSPARSELibrary);
+    }
+
     public ExecutionPolicyEnum getExecutionPolicy(){
         return (ExecutionPolicyEnum) getOptionValueFromOptionKey(GrCUDAOptions.ExecutionPolicy);
     }

projects/com.nvidia.grcuda/src/com/nvidia/grcuda/GrCUDAContext.java

@@ -42,6 +42,7 @@
 import com.nvidia.grcuda.cudalibraries.cublas.CUBLASRegistry;
 import com.nvidia.grcuda.cudalibraries.cuml.CUMLRegistry;
 import com.nvidia.grcuda.cudalibraries.tensorrt.TensorRTRegistry;
+import com.nvidia.grcuda.cudalibraries.cusparse.CUSPARSERegistry;
 import com.oracle.truffle.api.Option;
 
 @Option.Group(GrCUDALanguage.ID)
@@ -50,13 +51,13 @@ public final class GrCUDAOptions {
     @Option(category = OptionCategory.USER, help = "Enable cuBLAS support.", stability = OptionStability.STABLE) //
     public static final OptionKey<Boolean> CuBLASEnabled = new OptionKey<>(true);
 
-    @Option(category = OptionCategory.USER, help = "Set the location of the cublas library.", stability = OptionStability.STABLE) //
+    @Option(category = OptionCategory.USER, help = "Set the location of the cuBLAS library.", stability = OptionStability.STABLE) //
     public static final OptionKey<String> CuBLASLibrary = new OptionKey<>(CUBLASRegistry.DEFAULT_LIBRARY);
 
     @Option(category = OptionCategory.USER, help = "Enable cuML support.", stability = OptionStability.STABLE) //
     public static final OptionKey<Boolean> CuMLEnabled = new OptionKey<>(true);
 
-    @Option(category = OptionCategory.USER, help = "Set the location of the cuml library.", stability = OptionStability.STABLE) //
+    @Option(category = OptionCategory.USER, help = "Set the location of the cuML library.", stability = OptionStability.STABLE) //
     public static final OptionKey<String> CuMLLibrary = new OptionKey<>(CUMLRegistry.DEFAULT_LIBRARY);
 
     @Option(category = OptionCategory.USER, help = "Choose the scheduling policy of GrCUDA computations", stability = OptionStability.EXPERIMENTAL) //
@@ -86,6 +87,13 @@ public final class GrCUDAOptions {
     @Option(category = OptionCategory.USER, help = "Set the location of the TensorRT library.", stability = OptionStability.STABLE) //
     public static final OptionKey<String> TensorRTLibrary = new OptionKey<>(TensorRTRegistry.DEFAULT_LIBRARY);
 
+    @Option(category = OptionCategory.USER, help = "Enable cuSPARSE support.", stability = OptionStability.STABLE) //
+    public static final OptionKey<Boolean> CuSPARSEEnabled = new OptionKey<>(true);
+
+    @Option(category = OptionCategory.USER, help = "Set the location of the cuSPARSE library.", stability = OptionStability.EXPERIMENTAL) //
+    public static final OptionKey<String> CuSPARSELibrary = new OptionKey<>(CUSPARSERegistry.DEFAULT_LIBRARY);
+
     @Option(category = OptionCategory.USER, help = "Log the kernels execution time.", stability = OptionStability.STABLE) //
     public static final OptionKey<Boolean> TimeComputation = new OptionKey<>(GrCUDAOptionMap.DEFAULT_TIME_COMPUTATION);
 }
+

projects/com.nvidia.grcuda/src/com/nvidia/grcuda/GrCUDAOptionMap.java

@@ -46,7 +46,7 @@
  */
 public abstract class CUDALibraryFunction extends Function {
 
-    private final List<ComputationArgument> computationArguments;
+    protected final List<ComputationArgument> computationArguments;
 
     /**
      * Constructor, it takes the name of the wrapped function and its NFI signature,

projects/com.nvidia.grcuda/src/com/nvidia/grcuda/GrCUDAOptions.java

@@ -116,7 +116,7 @@ public Object call(Object[] arguments) throws ArityException {
                 }
             };
 
-            // create wrapper for cublasDestroy: cublasError_t cublasDestroy(long handle) -> void
+            // create wrapper for cublasDestroy: cublasError_t cublasDestroy(long handle)
             // cublasDestroy(long handle)
             cublasDestroyFunction = new Function(CUBLAS_CUBLASDESTROY.getName()) {
                 @Override
@@ -190,7 +190,6 @@ public void registerCUBLASFunctions(Namespace namespace) {
                 @TruffleBoundary
                 protected Object call(Object[] arguments) {
                     ensureInitialized();
-
                     try {
                         if (nfiFunction == null) {
                             CompilerDirectives.transferToInterpreterAndInvalidate();
@@ -265,4 +264,5 @@ private static String cublasReturnCodeToString(int returnCode) {
                             "(sint64, sint32, sint32, sint32, sint32, sint32, pointer, pointer, sint32, pointer, sint32, pointer, pointer, sint32): sint32"));
         }
     }
+
 }