[DO NOT MERGE][skip ci] JAVA 17 BWARE COMMIT

[skip ci] revert fed

[skip ci] revert a few files

vvector

bin/systemds

@@ -413,6 +413,7 @@ if [ $WORKER == 1 ]; then
   print_out "#  starting Federated worker on port $PORT"
   CMD=" \
   java $SYSTEMDS_STANDALONE_OPTS \
+  --add-modules=jdk.incubator.vector \
   $LOG4JPROPFULL \
   -jar $SYSTEMDS_JAR_FILE \
   -w $PORT \
@@ -422,6 +423,7 @@ elif [ "$FEDMONITORING" == 1 ]; then
   print_out "#  starting Federated backend monitoring on port $PORT"
   CMD=" \
   java $SYSTEMDS_STANDALONE_OPTS \
+  --add-modules=jdk.incubator.vector \
   $LOG4JPROPFULL \
   -jar $SYSTEMDS_JAR_FILE \
   -fedMonitoring $PORT \
@@ -433,6 +435,7 @@ elif [ $SYSDS_DISTRIBUTED == 0 ]; then
   CMD=" \
   java $SYSTEMDS_STANDALONE_OPTS \
   $LOG4JPROPFULL \
+  --add-modules=jdk.incubator.vector \
   -jar $SYSTEMDS_JAR_FILE \
   -f $SCRIPT_FILE \
   -exec $SYSDS_EXEC_MODE \
@@ -442,6 +445,7 @@ else
   print_out "#  Running script $SCRIPT_FILE distributed with opts: $*"
   CMD=" \
   spark-submit $SYSTEMDS_DISTRIBUTED_OPTS \
+  --add-modules=jdk.incubator.vector \
   $SYSTEMDS_JAR_FILE \
   -f $SCRIPT_FILE \
   -exec $SYSDS_EXEC_MODE \

pom.xml

@@ -67,7 +67,7 @@
 		<!-- aws-java-sdk-bundle version should align with hadoop-aws version -->
 		<!-- aws-java-sdk-bundle.version>1.12.367</aws-java-sdk-bundle.version -->
 		<!-- Set java compile level via argument, ex: 1.8 1.9 10 11-->
-		<java.level>11</java.level>
+		<java.level>17</java.level>
 		<java.version>{java.level}</java.version>
 		<!-->Testing settings<!-->
 		<maven.test.skip>false</maven.test.skip>
@@ -77,6 +77,7 @@
 		<test-forkCount>1C</test-forkCount>
 		<rerun.failing.tests.count>2</rerun.failing.tests.count>
 		<jacoco.skip>false</jacoco.skip>
+		<doc.skip>false</doc.skip>
 		<jacoco.include>**</jacoco.include>
 		<automatedtestbase.outputbuffering>false</automatedtestbase.outputbuffering>
 		<argLine>-Xms3000m -Xmx3000m -Xmn300m</argLine>
@@ -345,6 +346,9 @@
 					<source>${java.level}</source>
 					<target>${java.level}</target>
 					<release>${java.level}</release>
+					<compilerArgs>
+						<arg>--add-modules=jdk.incubator.vector</arg>
+					</compilerArgs>
 				</configuration>
 			</plugin>
 
@@ -367,6 +371,7 @@
 					<systemPropertyVariables>
 						<log4j.configurationFile>file:src/test/resources/log4j.properties</log4j.configurationFile>
 					</systemPropertyVariables>
+					<argLine>--add-modules=jdk.incubator.vector</argLine>
 				</configuration>
 			</plugin>
 
@@ -875,9 +880,10 @@
 						<configuration>
 							<excludePackageNames>*.protobuf</excludePackageNames>
 							<notimestamp>true</notimestamp>
-							<failOnWarnings>true</failOnWarnings>
+							<failOnWarnings>false</failOnWarnings>
 							<quiet>true</quiet>
-							<skip>false</skip>
+							<additionalJOption>--add-modules=jdk.incubator.vector</additionalJOption>
+							<skip>${doc.skip}</skip>
 							<show>public</show>
 							<source>${java.level}</source>
 						</configuration>

src/main/java/org/apache/sysds/hops/AggBinaryOp.java

@@ -439,8 +439,7 @@ private boolean isApplicableForTransitiveSparkExecType(boolean left)
 				|| (left && !isLeftTransposeRewriteApplicable(true)))
 			&& getInput(index).getParent().size()==1 //bagg is only parent
 			&& !getInput(index).areDimsBelowThreshold() 
-			&& (getInput(index).optFindExecType() == ExecType.SPARK
-				|| (getInput(index) instanceof DataOp && ((DataOp)getInput(index)).hasOnlyRDD()))
+			&& getInput(index).hasSparkOutput()
 			&& getInput(index).getOutputMemEstimate()>getOutputMemEstimate();
 	}
 

src/main/java/org/apache/sysds/hops/BinaryOp.java

@@ -747,8 +747,8 @@ protected ExecType optFindExecType(boolean transitive) {
 
 		checkAndSetForcedPlatform();
 
-		DataType dt1 = getInput().get(0).getDataType();
-		DataType dt2 = getInput().get(1).getDataType();
+		final DataType dt1 = getInput(0).getDataType();
+		final DataType dt2 = getInput(1).getDataType();
 
 		if( _etypeForced != null ) {
 			setExecType(_etypeForced);
@@ -796,18 +796,28 @@ else if ( dt1 == DataType.SCALAR && dt2 == DataType.MATRIX ) {
 			checkAndSetInvalidCPDimsAndSize();
 		}
 
-		//spark-specific decision refinement (execute unary scalar w/ spark input and
+		// spark-specific decision refinement (execute unary scalar w/ spark input and
 		// single parent also in spark because it's likely cheap and reduces intermediates)
-		if(transitive && _etype == ExecType.CP && _etypeForced != ExecType.CP && _etypeForced != ExecType.FED &&
-			getDataType().isMatrix() // output should be a matrix
-			&& (dt1.isScalar() || dt2.isScalar()) // one side should be scalar
-			&& supportsMatrixScalarOperations() // scalar operations
-			&& !(getInput().get(dt1.isScalar() ? 1 : 0) instanceof DataOp) // input is not checkpoint
-			&& getInput().get(dt1.isScalar() ? 1 : 0).getParent().size() == 1 // unary scalar is only parent
-			&& !HopRewriteUtils.isSingleBlock(getInput().get(dt1.isScalar() ? 1 : 0)) // single block triggered exec
-			&& getInput().get(dt1.isScalar() ? 1 : 0).optFindExecType() == ExecType.SPARK) {
-			// pull unary scalar operation into spark
-			_etype = ExecType.SPARK;
+		if(transitive // we allow transitive Spark operations. continue sequences of spark operations
+			&& _etype == ExecType.CP // The instruction is currently in CP
+			&& _etypeForced != ExecType.CP // not forced CP
+			&& _etypeForced != ExecType.FED // not federated
+			&& (getDataType().isMatrix() || getDataType().isFrame()) // output should be a matrix or frame
+		) {
+			final boolean v1 = getInput(0).isScalarOrVectorBellowBlockSize();
+			final boolean v2 = getInput(1).isScalarOrVectorBellowBlockSize();
+			final boolean left = v1 == true; // left side is the vector or scalar
+			final Hop sparkIn = getInput(left ? 1 : 0);
+			if((v1 ^ v2) // XOR only one side is allowed to be a vector or a scalar.
+				&& (supportsMatrixScalarOperations() || op == OpOp2.APPLY_SCHEMA)  // supported operation
+				&& sparkIn.getParent().size() == 1 // only one parent
+				&& !HopRewriteUtils.isSingleBlock(sparkIn) // single block triggered exec
+				&& sparkIn.optFindExecType() == ExecType.SPARK // input was spark op.
+				&& !(sparkIn instanceof DataOp) // input is not checkpoint
+			) {
+				// pull operation into spark
+				_etype = ExecType.SPARK;
+			}
 		}
 
 		if( OptimizerUtils.ALLOW_BINARY_UPDATE_IN_PLACE &&
@@ -837,7 +847,7 @@ else if( (op == OpOp2.CBIND && getDataType().isList())
 				|| (op == OpOp2.RBIND && getDataType().isList())) {
 			_etype = ExecType.CP;
 		}
-		
+
 		//mark for recompile (forever)
 		setRequiresRecompileIfNecessary();
 

src/main/java/org/apache/sysds/hops/Hop.java

@@ -1040,6 +1040,12 @@ public final String toString() {
 	// ========================================================================================
 
 
+	protected boolean isScalarOrVectorBellowBlockSize(){
+		return getDataType().isScalar() || (dimsKnown() && 
+			(( _dc.getRows() == 1 && _dc.getCols() < ConfigurationManager.getBlocksize())
+			|| _dc.getCols() == 1 && _dc.getRows() < ConfigurationManager.getBlocksize()));
+	}
+
 	protected boolean isVector() {
 		return (dimsKnown() && (_dc.getRows() == 1 || _dc.getCols() == 1) );
 	}
@@ -1624,6 +1630,11 @@ protected void setMemoryAndComputeEstimates(Lop lop) {
 		lop.setComputeEstimate(ComputeCost.getHOPComputeCost(this));
 	}
 
+	protected boolean hasSparkOutput(){
+		return (this.optFindExecType() == ExecType.SPARK
+			|| (this instanceof DataOp && ((DataOp)this).hasOnlyRDD()));
+	}
+
 	/**
 	 * Set parse information.
 	 *

src/main/java/org/apache/sysds/hops/UnaryOp.java

@@ -366,7 +366,11 @@ protected double computeOutputMemEstimate( long dim1, long dim2, long nnz )
 		} else {
 			sparsity = OptimizerUtils.getSparsity(dim1, dim2, nnz);
 		}
-		return OptimizerUtils.estimateSizeExactSparsity(dim1, dim2, sparsity);
+
+		if(getDataType() == DataType.FRAME)
+			return OptimizerUtils.estimateSizeExactFrame(dim1, dim2);
+		else
+			return OptimizerUtils.estimateSizeExactSparsity(dim1, dim2, sparsity);
 	}
 
 	@Override
@@ -463,6 +467,13 @@ public boolean isMetadataOperation() {
 			|| _op == OpOp1.CAST_AS_LIST;
 	}
 
+	private boolean isDisallowedSparkOps(){
+		return isCumulativeUnaryOperation() 
+			|| isCastUnaryOperation()
+			|| _op==OpOp1.MEDIAN
+			|| _op==OpOp1.IQM;
+	}
+
 	@Override
 	protected ExecType optFindExecType(boolean transitive) 
 	{
@@ -493,19 +504,22 @@ else if ( getInput().get(0).areDimsBelowThreshold() || getInput().get(0).isVecto
 			checkAndSetInvalidCPDimsAndSize();
 		}
 
+
 		//spark-specific decision refinement (execute unary w/ spark input and 
 		//single parent also in spark because it's likely cheap and reduces intermediates)
-		if( _etype == ExecType.CP && _etypeForced != ExecType.CP
-			&& getInput().get(0).optFindExecType() == ExecType.SPARK 
-			&& getDataType().isMatrix() 
-			&& !isCumulativeUnaryOperation() && !isCastUnaryOperation()
-			&& _op!=OpOp1.MEDIAN && _op!=OpOp1.IQM
-			&& !(getInput().get(0) instanceof DataOp)    //input is not checkpoint
-			&& getInput().get(0).getParent().size()==1 ) //unary is only parent
-		{
+		if(_etype == ExecType.CP // currently CP instruction
+			&& _etype != ExecType.SPARK /// currently not SP.
+			&& _etypeForced != ExecType.CP // not forced as CP instruction
+			&& getInput(0).hasSparkOutput() // input is a spark instruction
+			&& (getDataType().isMatrix() || getDataType().isFrame()) // output is a matrix or frame
+			&& !isDisallowedSparkOps() // is invalid spark instruction
+			// && !(getInput().get(0) instanceof DataOp) // input is not checkpoint
+			// && getInput(0).getParent().size() <= 1// unary is only parent
+		) {
 			//pull unary operation into spark 
 			_etype = ExecType.SPARK;
 		}
+
 
 		//mark for recompile (forever)
 		setRequiresRecompileIfNecessary();
@@ -520,7 +534,7 @@ && getInput().get(0).getParent().size()==1 ) //unary is only parent
 		} else {
 			setRequiresRecompileIfNecessary();
 		}
-		
+
 		return _etype;
 	}
 

src/main/java/org/apache/sysds/runtime/compress/CompressedMatrixBlock.java

@@ -58,6 +58,7 @@
 import org.apache.sysds.runtime.compress.lib.CLALibMMChain;
 import org.apache.sysds.runtime.compress.lib.CLALibMatrixMult;
 import org.apache.sysds.runtime.compress.lib.CLALibMerge;
+import org.apache.sysds.runtime.compress.lib.CLALibReorg;
 import org.apache.sysds.runtime.compress.lib.CLALibReplace;
 import org.apache.sysds.runtime.compress.lib.CLALibReshape;
 import org.apache.sysds.runtime.compress.lib.CLALibRexpand;
@@ -72,7 +73,6 @@
 import org.apache.sysds.runtime.data.DenseBlock;
 import org.apache.sysds.runtime.data.SparseBlock;
 import org.apache.sysds.runtime.data.SparseRow;
-import org.apache.sysds.runtime.functionobjects.SwapIndex;
 import org.apache.sysds.runtime.instructions.InstructionUtils;
 import org.apache.sysds.runtime.instructions.cp.CM_COV_Object;
 import org.apache.sysds.runtime.instructions.cp.ScalarObject;
@@ -226,6 +226,7 @@ public void allocateColGroup(AColGroup cg) {
 	 * @param colGroups new ColGroups in the MatrixBlock
 	 */
 	public void allocateColGroupList(List<AColGroup> colGroups) {
+		cachedMemorySize = -1;
 		_colGroups = colGroups;
 	}
 
@@ -351,7 +352,6 @@ public long recomputeNonZeros(int k) {
 			List<Future<Long>> tasks = new ArrayList<>();
 			for(AColGroup g : _colGroups)
 				tasks.add(pool.submit(() -> g.getNumberNonZeros(rlen)));
-
 			long nnz = 0;
 			for(Future<Long> t : tasks)
 				nnz += t.get();
@@ -398,7 +398,6 @@ public long estimateSizeInMemory() {
 	public long estimateCompressedSizeInMemory() {
 
 		if(cachedMemorySize <= -1L) {
-
 			long total = baseSizeInMemory();
 			// take into consideration duplicate dictionaries
 			Set<IDictionary> dicts = new HashSet<>();
@@ -413,7 +412,6 @@ public long estimateCompressedSizeInMemory() {
 			}
 			cachedMemorySize = total;
 			return total;
-
 		}
 		else
 			return cachedMemorySize;
@@ -635,21 +633,7 @@ public MatrixBlock replaceOperations(MatrixValue result, double pattern, double
 
 	@Override
 	public MatrixBlock reorgOperations(ReorgOperator op, MatrixValue ret, int startRow, int startColumn, int length) {
-		if(op.fn instanceof SwapIndex && this.getNumColumns() == 1) {
-			MatrixBlock tmp = decompress(op.getNumThreads());
-			long nz = tmp.setNonZeros(tmp.getNonZeros());
-			tmp = new MatrixBlock(tmp.getNumColumns(), tmp.getNumRows(), tmp.getDenseBlockValues());
-			tmp.setNonZeros(nz);
-			return tmp;
-		}
-		else {
-			// Allow transpose to be compressed output. In general we need to have a transposed flag on
-			// the compressed matrix. https://issues.apache.org/jira/browse/SYSTEMDS-3025
-			String message = op.getClass().getSimpleName() + " -- " + op.fn.getClass().getSimpleName();
-			MatrixBlock tmp = getUncompressed(message, op.getNumThreads());
-			return tmp.reorgOperations(op, ret, startRow, startColumn, length);
-		}
-
+		return CLALibReorg.reorg(this, op, (MatrixBlock) ret, startRow, startColumn, length);
 	}
 
 	public boolean isOverlapping() {
@@ -1002,6 +986,10 @@ public MatrixBlock getUncompressed() {
 		return getUncompressed((String) null);
 	}
 
+	public MatrixBlock getUncompressed(int k){
+		return getUncompressed((String) null, k);
+	}
+
 	public MatrixBlock getUncompressed(String operation) {
 		return getUncompressed(operation,
 			ConfigurationManager.isParallelMatrixOperations() ? InfrastructureAnalyzer.getLocalParallelism() : 1);
@@ -1213,8 +1201,8 @@ public void examSparsity(boolean allowCSR, int k) {
 	}
 
 	@Override
-	public void sparseToDense(int k) {
-		// do nothing
+	public MatrixBlock sparseToDense(int k) {
+		return this; // do nothing
 	}
 
 	@Override